fs/btrfs/inode.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/kernel.h>
   7 #include <linux/bio.h>
   8 #include <linux/buffer_head.h>
   9 #include <linux/file.h>
  10 #include <linux/fs.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/highmem.h>
  13 #include <linux/time.h>
  14 #include <linux/init.h>
  15 #include <linux/string.h>
  16 #include <linux/backing-dev.h>
  17 #include <linux/writeback.h>
  18 #include <linux/compat.h>
  19 #include <linux/xattr.h>
  20 #include <linux/posix_acl.h>
  21 #include <linux/falloc.h>
  22 #include <linux/slab.h>
  23 #include <linux/ratelimit.h>
  24 #include <linux/btrfs.h>
  25 #include <linux/blkdev.h>
  26 #include <linux/posix_acl_xattr.h>
  27 #include <linux/uio.h>
  28 #include <linux/magic.h>
  29 #include <linux/iversion.h>
  30 #include <linux/swap.h>
  31 #include <asm/unaligned.h>
  32 #include "ctree.h"
  33 #include "disk-io.h"
  34 #include "transaction.h"
  35 #include "btrfs_inode.h"
  36 #include "print-tree.h"
  37 #include "ordered-data.h"
  38 #include "xattr.h"
  39 #include "tree-log.h"
  40 #include "volumes.h"
  41 #include "compression.h"
  42 #include "locking.h"
  43 #include "free-space-cache.h"
  44 #include "inode-map.h"
  45 #include "backref.h"
  46 #include "props.h"
  47 #include "qgroup.h"
  48 #include "dedupe.h"
  49
  50 struct btrfs_iget_args {
  51         struct btrfs_key *location;
  52         struct btrfs_root *root;
  53 };
  54
  55 struct btrfs_dio_data {
  56         u64 reserve;
  57         u64 unsubmitted_oe_range_start;
  58         u64 unsubmitted_oe_range_end;
  59         int overwrite;
  60 };
  61
  62 static const struct inode_operations btrfs_dir_inode_operations;
  63 static const struct inode_operations btrfs_symlink_inode_operations;
  64 static const struct inode_operations btrfs_dir_ro_inode_operations;
  65 static const struct inode_operations btrfs_special_inode_operations;
  66 static const struct inode_operations btrfs_file_inode_operations;
  67 static const struct address_space_operations btrfs_aops;
  68 static const struct file_operations btrfs_dir_file_operations;
  69 static const struct extent_io_ops btrfs_extent_io_ops;
  70
  71 static struct kmem_cache *btrfs_inode_cachep;
  72 struct kmem_cache *btrfs_trans_handle_cachep;
  73 struct kmem_cache *btrfs_path_cachep;
  74 struct kmem_cache *btrfs_free_space_cachep;
  75
  76 #define S_SHIFT 12
  77 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  78         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  79         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  80         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  81         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  82         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  83         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  84         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  85 };
  86
  87 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
  88 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
  89 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  90 static noinline int cow_file_range(struct inode *inode,
  91                                    struct page *locked_page,
  92                                    u64 start, u64 end, u64 delalloc_end,
  93                                    int *page_started, unsigned long *nr_written,
  94                                    int unlock, struct btrfs_dedupe_hash *hash);
  95 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
  96                                        u64 orig_start, u64 block_start,
  97                                        u64 block_len, u64 orig_block_len,
  98                                        u64 ram_bytes, int compress_type,
  99                                        int type);
 100
 101 static void __endio_write_update_ordered(struct inode *inode,
 102                                          const u64 offset, const u64 bytes,
 103                                          const bool uptodate);
 104
 105 /*
 106  * Cleanup all submitted ordered extents in specified range to handle errors
 107  * from the fill_dellaloc() callback.
 108  *
 109  * NOTE: caller must ensure that when an error happens, it can not call
 110  * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
 111  * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
 112  * to be released, which we want to happen only when finishing the ordered
 113  * extent (btrfs_finish_ordered_io()). Also note that the caller of
 114  * btrfs_run_delalloc_range already does proper cleanup for the first page of
 115  * the range, that is, it invokes the callback writepage_end_io_hook() for the
 116  * range of the first page.
 117  */
 118 static inline void btrfs_cleanup_ordered_extents(struct inode *inode,
 119                                                  const u64 offset,
 120                                                  const u64 bytes)
 121 {
 122         unsigned long index = offset >> PAGE_SHIFT;
 123         unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
 124         struct page *page;
 125
 126         while (index <= end_index) {
 127                 page = find_get_page(inode->i_mapping, index);
 128                 index++;
 129                 if (!page)
 130                         continue;
 131                 ClearPagePrivate2(page);
 132                 put_page(page);
 133         }
 134         return __endio_write_update_ordered(inode, offset + PAGE_SIZE,
 135                                             bytes - PAGE_SIZE, false);
 136 }
 137
 138 static int btrfs_dirty_inode(struct inode *inode);
 139
 140 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 141 void btrfs_test_inode_set_ops(struct inode *inode)
 142 {
 143         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
 144 }
 145 #endif
 146
 147 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 148                                      struct inode *inode,  struct inode *dir,
 149                                      const struct qstr *qstr)
 150 {
 151         int err;
 152
 153         err = btrfs_init_acl(trans, inode, dir);
 154         if (!err)
 155                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 156         return err;
 157 }
 158
 159 /*
 160  * this does all the hard work for inserting an inline extent into
 161  * the btree.  The caller should have done a btrfs_drop_extents so that
 162  * no overlapping inline items exist in the btree
 163  */
 164 static int insert_inline_extent(struct btrfs_trans_handle *trans,
 165                                 struct btrfs_path *path, int extent_inserted,
 166                                 struct btrfs_root *root, struct inode *inode,
 167                                 u64 start, size_t size, size_t compressed_size,
 168                                 int compress_type,
 169                                 struct page **compressed_pages)
 170 {
 171         struct extent_buffer *leaf;
 172         struct page *page = NULL;
 173         char *kaddr;
 174         unsigned long ptr;
 175         struct btrfs_file_extent_item *ei;
 176         int ret;
 177         size_t cur_size = size;
 178         unsigned long offset;
 179
 180         if (compressed_size && compressed_pages)
 181                 cur_size = compressed_size;
 182
 183         inode_add_bytes(inode, size);
 184
 185         if (!extent_inserted) {
 186                 struct btrfs_key key;
 187                 size_t datasize;
 188
 189                 key.objectid = btrfs_ino(BTRFS_I(inode));
 190                 key.offset = start;
 191                 key.type = BTRFS_EXTENT_DATA_KEY;
 192
 193                 datasize = btrfs_file_extent_calc_inline_size(cur_size);
 194                 path->leave_spinning = 1;
 195                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 196                                               datasize);
 197                 if (ret)
 198                         goto fail;
 199         }
 200         leaf = path->nodes[0];
 201         ei = btrfs_item_ptr(leaf, path->slots[0],
 202                             struct btrfs_file_extent_item);
 203         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 204         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 205         btrfs_set_file_extent_encryption(leaf, ei, 0);
 206         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 207         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 208         ptr = btrfs_file_extent_inline_start(ei);
 209
 210         if (compress_type != BTRFS_COMPRESS_NONE) {
 211                 struct page *cpage;
 212                 int i = 0;
 213                 while (compressed_size > 0) {
 214                         cpage = compressed_pages[i];
 215                         cur_size = min_t(unsigned long, compressed_size,
 216                                        PAGE_SIZE);
 217
 218                         kaddr = kmap_atomic(cpage);
 219                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 220                         kunmap_atomic(kaddr);
 221
 222                         i++;
 223                         ptr += cur_size;
 224                         compressed_size -= cur_size;
 225                 }
 226                 btrfs_set_file_extent_compression(leaf, ei,
 227                                                   compress_type);
 228         } else {
 229                 page = find_get_page(inode->i_mapping,
 230                                      start >> PAGE_SHIFT);
 231                 btrfs_set_file_extent_compression(leaf, ei, 0);
 232                 kaddr = kmap_atomic(page);
 233                 offset = start & (PAGE_SIZE - 1);
 234                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 235                 kunmap_atomic(kaddr);
 236                 put_page(page);
 237         }
 238         btrfs_mark_buffer_dirty(leaf);
 239         btrfs_release_path(path);
 240
 241         /*
 242          * we're an inline extent, so nobody can
 243          * extend the file past i_size without locking
 244          * a page we already have locked.
 245          *
 246          * We must do any isize and inode updates
 247          * before we unlock the pages.  Otherwise we
 248          * could end up racing with unlink.
 249          */
 250         BTRFS_I(inode)->disk_i_size = inode->i_size;
 251         ret = btrfs_update_inode(trans, root, inode);
 252
 253 fail:
 254         return ret;
 255 }
 256
 257
 258 /*
 259  * conditionally insert an inline extent into the file.  This
 260  * does the checks required to make sure the data is small enough
 261  * to fit as an inline extent.
 262  */
 263 static noinline int cow_file_range_inline(struct inode *inode, u64 start,
 264                                           u64 end, size_t compressed_size,
 265                                           int compress_type,
 266                                           struct page **compressed_pages)
 267 {
 268         struct btrfs_root *root = BTRFS_I(inode)->root;
 269         struct btrfs_fs_info *fs_info = root->fs_info;
 270         struct btrfs_trans_handle *trans;
 271         u64 isize = i_size_read(inode);
 272         u64 actual_end = min(end + 1, isize);
 273         u64 inline_len = actual_end - start;
 274         u64 aligned_end = ALIGN(end, fs_info->sectorsize);
 275         u64 data_len = inline_len;
 276         int ret;
 277         struct btrfs_path *path;
 278         int extent_inserted = 0;
 279         u32 extent_item_size;
 280
 281         if (compressed_size)
 282                 data_len = compressed_size;
 283
 284         if (start > 0 ||
 285             actual_end > fs_info->sectorsize ||
 286             data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
 287             (!compressed_size &&
 288             (actual_end & (fs_info->sectorsize - 1)) == 0) ||
 289             end + 1 < isize ||
 290             data_len > fs_info->max_inline) {
 291                 return 1;
 292         }
 293
 294         path = btrfs_alloc_path();
 295         if (!path)
 296                 return -ENOMEM;
 297
 298         trans = btrfs_join_transaction(root);
 299         if (IS_ERR(trans)) {
 300                 btrfs_free_path(path);
 301                 return PTR_ERR(trans);
 302         }
 303         trans->block_rsv = &BTRFS_I(inode)->block_rsv;
 304
 305         if (compressed_size && compressed_pages)
 306                 extent_item_size = btrfs_file_extent_calc_inline_size(
 307                    compressed_size);
 308         else
 309                 extent_item_size = btrfs_file_extent_calc_inline_size(
 310                     inline_len);
 311
 312         ret = __btrfs_drop_extents(trans, root, inode, path,
 313                                    start, aligned_end, NULL,
 314                                    1, 1, extent_item_size, &extent_inserted);
 315         if (ret) {
 316                 btrfs_abort_transaction(trans, ret);
 317                 goto out;
 318         }
 319
 320         if (isize > actual_end)
 321                 inline_len = min_t(u64, isize, actual_end);
 322         ret = insert_inline_extent(trans, path, extent_inserted,
 323                                    root, inode, start,
 324                                    inline_len, compressed_size,
 325                                    compress_type, compressed_pages);
 326         if (ret && ret != -ENOSPC) {
 327                 btrfs_abort_transaction(trans, ret);
 328                 goto out;
 329         } else if (ret == -ENOSPC) {
 330                 ret = 1;
 331                 goto out;
 332         }
 333
 334         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
 335         btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0);
 336 out:
 337         /*
 338          * Don't forget to free the reserved space, as for inlined extent
 339          * it won't count as data extent, free them directly here.
 340          * And at reserve time, it's always aligned to page size, so
 341          * just free one page here.
 342          */
 343         btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
 344         btrfs_free_path(path);
 345         btrfs_end_transaction(trans);
 346         return ret;
 347 }
 348
 349 struct async_extent {
 350         u64 start;
 351         u64 ram_size;
 352         u64 compressed_size;
 353         struct page **pages;
 354         unsigned long nr_pages;
 355         int compress_type;
 356         struct list_head list;
 357 };
 358
 359 struct async_cow {
 360         struct inode *inode;
 361         struct btrfs_fs_info *fs_info;
 362         struct page *locked_page;
 363         u64 start;
 364         u64 end;
 365         unsigned int write_flags;
 366         struct list_head extents;
 367         struct btrfs_work work;
 368 };
 369
 370 static noinline int add_async_extent(struct async_cow *cow,
 371                                      u64 start, u64 ram_size,
 372                                      u64 compressed_size,
 373                                      struct page **pages,
 374                                      unsigned long nr_pages,
 375                                      int compress_type)
 376 {
 377         struct async_extent *async_extent;
 378
 379         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 380         BUG_ON(!async_extent); /* -ENOMEM */
 381         async_extent->start = start;
 382         async_extent->ram_size = ram_size;
 383         async_extent->compressed_size = compressed_size;
 384         async_extent->pages = pages;
 385         async_extent->nr_pages = nr_pages;
 386         async_extent->compress_type = compress_type;
 387         list_add_tail(&async_extent->list, &cow->extents);
 388         return 0;
 389 }
 390
 391 static inline int inode_need_compress(struct inode *inode, u64 start, u64 end)
 392 {
 393         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 394
 395         /* force compress */
 396         if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
 397                 return 1;
 398         /* defrag ioctl */
 399         if (BTRFS_I(inode)->defrag_compress)
 400                 return 1;
 401         /* bad compression ratios */
 402         if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 403                 return 0;
 404         if (btrfs_test_opt(fs_info, COMPRESS) ||
 405             BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
 406             BTRFS_I(inode)->prop_compress)
 407                 return btrfs_compress_heuristic(inode, start, end);
 408         return 0;
 409 }
 410
 411 static inline void inode_should_defrag(struct btrfs_inode *inode,
 412                 u64 start, u64 end, u64 num_bytes, u64 small_write)
 413 {
 414         /* If this is a small write inside eof, kick off a defrag */
 415         if (num_bytes < small_write &&
 416             (start > 0 || end + 1 < inode->disk_i_size))
 417                 btrfs_add_inode_defrag(NULL, inode);
 418 }
 419
 420 /*
 421  * we create compressed extents in two phases.  The first
 422  * phase compresses a range of pages that have already been
 423  * locked (both pages and state bits are locked).
 424  *
 425  * This is done inside an ordered work queue, and the compression
 426  * is spread across many cpus.  The actual IO submission is step
 427  * two, and the ordered work queue takes care of making sure that
 428  * happens in the same order things were put onto the queue by
 429  * writepages and friends.
 430  *
 431  * If this code finds it can't get good compression, it puts an
 432  * entry onto the work queue to write the uncompressed bytes.  This
 433  * makes sure that both compressed inodes and uncompressed inodes
 434  * are written in the same order that the flusher thread sent them
 435  * down.
 436  */
 437 static noinline void compress_file_range(struct inode *inode,
 438                                         struct page *locked_page,
 439                                         u64 start, u64 end,
 440                                         struct async_cow *async_cow,
 441                                         int *num_added)
 442 {
 443         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 444         u64 blocksize = fs_info->sectorsize;
 445         u64 actual_end;
 446         u64 isize = i_size_read(inode);
 447         int ret = 0;
 448         struct page **pages = NULL;
 449         unsigned long nr_pages;
 450         unsigned long total_compressed = 0;
 451         unsigned long total_in = 0;
 452         int i;
 453         int will_compress;
 454         int compress_type = fs_info->compress_type;
 455         int redirty = 0;
 456
 457         inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
 458                         SZ_16K);
 459
 460         actual_end = min_t(u64, isize, end + 1);
 461 again:
 462         will_compress = 0;
 463         nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
 464         BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
 465         nr_pages = min_t(unsigned long, nr_pages,
 466                         BTRFS_MAX_COMPRESSED / PAGE_SIZE);
 467
 468         /*
 469          * we don't want to send crud past the end of i_size through
 470          * compression, that's just a waste of CPU time.  So, if the
 471          * end of the file is before the start of our current
 472          * requested range of bytes, we bail out to the uncompressed
 473          * cleanup code that can deal with all of this.
 474          *
 475          * It isn't really the fastest way to fix things, but this is a
 476          * very uncommon corner.
 477          */
 478         if (actual_end <= start)
 479                 goto cleanup_and_bail_uncompressed;
 480
 481         total_compressed = actual_end - start;
 482
 483         /*
 484          * skip compression for a small file range(<=blocksize) that
 485          * isn't an inline extent, since it doesn't save disk space at all.
 486          */
 487         if (total_compressed <= blocksize &&
 488            (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 489                 goto cleanup_and_bail_uncompressed;
 490
 491         total_compressed = min_t(unsigned long, total_compressed,
 492                         BTRFS_MAX_UNCOMPRESSED);
 493         total_in = 0;
 494         ret = 0;
 495
 496         /*
 497          * we do compression for mount -o compress and when the
 498          * inode has not been flagged as nocompress.  This flag can
 499          * change at any time if we discover bad compression ratios.
 500          */
 501         if (inode_need_compress(inode, start, end)) {
 502                 WARN_ON(pages);
 503                 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
 504                 if (!pages) {
 505                         /* just bail out to the uncompressed code */
 506                         nr_pages = 0;
 507                         goto cont;
 508                 }
 509
 510                 if (BTRFS_I(inode)->defrag_compress)
 511                         compress_type = BTRFS_I(inode)->defrag_compress;
 512                 else if (BTRFS_I(inode)->prop_compress)
 513                         compress_type = BTRFS_I(inode)->prop_compress;
 514
 515                 /*
 516                  * we need to call clear_page_dirty_for_io on each
 517                  * page in the range.  Otherwise applications with the file
 518                  * mmap'd can wander in and change the page contents while
 519                  * we are compressing them.
 520                  *
 521                  * If the compression fails for any reason, we set the pages
 522                  * dirty again later on.
 523                  *
 524                  * Note that the remaining part is redirtied, the start pointer
 525                  * has moved, the end is the original one.
 526                  */
 527                 if (!redirty) {
 528                         extent_range_clear_dirty_for_io(inode, start, end);
 529                         redirty = 1;
 530                 }
 531
 532                 /* Compression level is applied here and only here */
 533                 ret = btrfs_compress_pages(
 534                         compress_type | (fs_info->compress_level << 4),
 535                                            inode->i_mapping, start,
 536                                            pages,
 537                                            &nr_pages,
 538                                            &total_in,
 539                                            &total_compressed);
 540
 541                 if (!ret) {
 542                         unsigned long offset = total_compressed &
 543                                 (PAGE_SIZE - 1);
 544                         struct page *page = pages[nr_pages - 1];
 545                         char *kaddr;
 546
 547                         /* zero the tail end of the last page, we might be
 548                          * sending it down to disk
 549                          */
 550                         if (offset) {
 551                                 kaddr = kmap_atomic(page);
 552                                 memset(kaddr + offset, 0,
 553                                        PAGE_SIZE - offset);
 554                                 kunmap_atomic(kaddr);
 555                         }
 556                         will_compress = 1;
 557                 }
 558         }
 559 cont:
 560         if (start == 0) {
 561                 /* lets try to make an inline extent */
 562                 if (ret || total_in < actual_end) {
 563                         /* we didn't compress the entire range, try
 564                          * to make an uncompressed inline extent.
 565                          */
 566                         ret = cow_file_range_inline(inode, start, end, 0,
 567                                                     BTRFS_COMPRESS_NONE, NULL);
 568                 } else {
 569                         /* try making a compressed inline extent */
 570                         ret = cow_file_range_inline(inode, start, end,
 571                                                     total_compressed,
 572                                                     compress_type, pages);
 573                 }
 574                 if (ret <= 0) {
 575                         unsigned long clear_flags = EXTENT_DELALLOC |
 576                                 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
 577                                 EXTENT_DO_ACCOUNTING;
 578                         unsigned long page_error_op;
 579
 580                         page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
 581
 582                         /*
 583                          * inline extent creation worked or returned error,
 584                          * we don't need to create any more async work items.
 585                          * Unlock and free up our temp pages.
 586                          *
 587                          * We use DO_ACCOUNTING here because we need the
 588                          * delalloc_release_metadata to be done _after_ we drop
 589                          * our outstanding extent for clearing delalloc for this
 590                          * range.
 591                          */
 592                         extent_clear_unlock_delalloc(inode, start, end, end,
 593                                                      NULL, clear_flags,
 594                                                      PAGE_UNLOCK |
 595                                                      PAGE_CLEAR_DIRTY |
 596                                                      PAGE_SET_WRITEBACK |
 597                                                      page_error_op |
 598                                                      PAGE_END_WRITEBACK);
 599                         goto free_pages_out;
 600                 }
 601         }
 602
 603         if (will_compress) {
 604                 /*
 605                  * we aren't doing an inline extent round the compressed size
 606                  * up to a block size boundary so the allocator does sane
 607                  * things
 608                  */
 609                 total_compressed = ALIGN(total_compressed, blocksize);
 610
 611                 /*
 612                  * one last check to make sure the compression is really a
 613                  * win, compare the page count read with the blocks on disk,
 614                  * compression must free at least one sector size
 615                  */
 616                 total_in = ALIGN(total_in, PAGE_SIZE);
 617                 if (total_compressed + blocksize <= total_in) {
 618                         *num_added += 1;
 619
 620                         /*
 621                          * The async work queues will take care of doing actual
 622                          * allocation on disk for these compressed pages, and
 623                          * will submit them to the elevator.
 624                          */
 625                         add_async_extent(async_cow, start, total_in,
 626                                         total_compressed, pages, nr_pages,
 627                                         compress_type);
 628
 629                         if (start + total_in < end) {
 630                                 start += total_in;
 631                                 pages = NULL;
 632                                 cond_resched();
 633                                 goto again;
 634                         }
 635                         return;
 636                 }
 637         }
 638         if (pages) {
 639                 /*
 640                  * the compression code ran but failed to make things smaller,
 641                  * free any pages it allocated and our page pointer array
 642                  */
 643                 for (i = 0; i < nr_pages; i++) {
 644                         WARN_ON(pages[i]->mapping);
 645                         put_page(pages[i]);
 646                 }
 647                 kfree(pages);
 648                 pages = NULL;
 649                 total_compressed = 0;
 650                 nr_pages = 0;
 651
 652                 /* flag the file so we don't compress in the future */
 653                 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
 654                     !(BTRFS_I(inode)->prop_compress)) {
 655                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 656                 }
 657         }
 658 cleanup_and_bail_uncompressed:
 659         /*
 660          * No compression, but we still need to write the pages in the file
 661          * we've been given so far.  redirty the locked page if it corresponds
 662          * to our extent and set things up for the async work queue to run
 663          * cow_file_range to do the normal delalloc dance.
 664          */
 665         if (page_offset(locked_page) >= start &&
 666             page_offset(locked_page) <= end)
 667                 __set_page_dirty_nobuffers(locked_page);
 668                 /* unlocked later on in the async handlers */
 669
 670         if (redirty)
 671                 extent_range_redirty_for_io(inode, start, end);
 672         add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
 673                          BTRFS_COMPRESS_NONE);
 674         *num_added += 1;
 675
 676         return;
 677
 678 free_pages_out:
 679         for (i = 0; i < nr_pages; i++) {
 680                 WARN_ON(pages[i]->mapping);
 681                 put_page(pages[i]);
 682         }
 683         kfree(pages);
 684 }
 685
 686 static void free_async_extent_pages(struct async_extent *async_extent)
 687 {
 688         int i;
 689
 690         if (!async_extent->pages)
 691                 return;
 692
 693         for (i = 0; i < async_extent->nr_pages; i++) {
 694                 WARN_ON(async_extent->pages[i]->mapping);
 695                 put_page(async_extent->pages[i]);
 696         }
 697         kfree(async_extent->pages);
 698         async_extent->nr_pages = 0;
 699         async_extent->pages = NULL;
 700 }
 701
 702 /*
 703  * phase two of compressed writeback.  This is the ordered portion
 704  * of the code, which only gets called in the order the work was
 705  * queued.  We walk all the async extents created by compress_file_range
 706  * and send them down to the disk.
 707  */
 708 static noinline void submit_compressed_extents(struct inode *inode,
 709                                               struct async_cow *async_cow)
 710 {
 711         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 712         struct async_extent *async_extent;
 713         u64 alloc_hint = 0;
 714         struct btrfs_key ins;
 715         struct extent_map *em;
 716         struct btrfs_root *root = BTRFS_I(inode)->root;
 717         struct extent_io_tree *io_tree;
 718         int ret = 0;
 719
 720 again:
 721         while (!list_empty(&async_cow->extents)) {
 722                 async_extent = list_entry(async_cow->extents.next,
 723                                           struct async_extent, list);
 724                 list_del(&async_extent->list);
 725
 726                 io_tree = &BTRFS_I(inode)->io_tree;
 727
 728 retry:
 729                 /* did the compression code fall back to uncompressed IO? */
 730                 if (!async_extent->pages) {
 731                         int page_started = 0;
 732                         unsigned long nr_written = 0;
 733
 734                         lock_extent(io_tree, async_extent->start,
 735                                          async_extent->start +
 736                                          async_extent->ram_size - 1);
 737
 738                         /* allocate blocks */
 739                         ret = cow_file_range(inode, async_cow->locked_page,
 740                                              async_extent->start,
 741                                              async_extent->start +
 742                                              async_extent->ram_size - 1,
 743                                              async_extent->start +
 744                                              async_extent->ram_size - 1,
 745                                              &page_started, &nr_written, 0,
 746                                              NULL);
 747
 748                         /* JDM XXX */
 749
 750                         /*
 751                          * if page_started, cow_file_range inserted an
 752                          * inline extent and took care of all the unlocking
 753                          * and IO for us.  Otherwise, we need to submit
 754                          * all those pages down to the drive.
 755                          */
 756                         if (!page_started && !ret)
 757                                 extent_write_locked_range(inode,
 758                                                   async_extent->start,
 759                                                   async_extent->start +
 760                                                   async_extent->ram_size - 1,
 761                                                   WB_SYNC_ALL);
 762                         else if (ret)
 763                                 unlock_page(async_cow->locked_page);
 764                         kfree(async_extent);
 765                         cond_resched();
 766                         continue;
 767                 }
 768
 769                 lock_extent(io_tree, async_extent->start,
 770                             async_extent->start + async_extent->ram_size - 1);
 771
 772                 ret = btrfs_reserve_extent(root, async_extent->ram_size,
 773                                            async_extent->compressed_size,
 774                                            async_extent->compressed_size,
 775                                            0, alloc_hint, &ins, 1, 1);
 776                 if (ret) {
 777                         free_async_extent_pages(async_extent);
 778
 779                         if (ret == -ENOSPC) {
 780                                 unlock_extent(io_tree, async_extent->start,
 781                                               async_extent->start +
 782                                               async_extent->ram_size - 1);
 783
 784                                 /*
 785                                  * we need to redirty the pages if we decide to
 786                                  * fallback to uncompressed IO, otherwise we
 787                                  * will not submit these pages down to lower
 788                                  * layers.
 789                                  */
 790                                 extent_range_redirty_for_io(inode,
 791                                                 async_extent->start,
 792                                                 async_extent->start +
 793                                                 async_extent->ram_size - 1);
 794
 795                                 goto retry;
 796                         }
 797                         goto out_free;
 798                 }
 799                 /*
 800                  * here we're doing allocation and writeback of the
 801                  * compressed pages
 802                  */
 803                 em = create_io_em(inode, async_extent->start,
 804                                   async_extent->ram_size, /* len */
 805                                   async_extent->start, /* orig_start */
 806                                   ins.objectid, /* block_start */
 807                                   ins.offset, /* block_len */
 808                                   ins.offset, /* orig_block_len */
 809                                   async_extent->ram_size, /* ram_bytes */
 810                                   async_extent->compress_type,
 811                                   BTRFS_ORDERED_COMPRESSED);
 812                 if (IS_ERR(em))
 813                         /* ret value is not necessary due to void function */
 814                         goto out_free_reserve;
 815                 free_extent_map(em);
 816
 817                 ret = btrfs_add_ordered_extent_compress(inode,
 818                                                 async_extent->start,
 819                                                 ins.objectid,
 820                                                 async_extent->ram_size,
 821                                                 ins.offset,
 822                                                 BTRFS_ORDERED_COMPRESSED,
 823                                                 async_extent->compress_type);
 824                 if (ret) {
 825                         btrfs_drop_extent_cache(BTRFS_I(inode),
 826                                                 async_extent->start,
 827                                                 async_extent->start +
 828                                                 async_extent->ram_size - 1, 0);
 829                         goto out_free_reserve;
 830                 }
 831                 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
 832
 833                 /*
 834                  * clear dirty, set writeback and unlock the pages.
 835                  */
 836                 extent_clear_unlock_delalloc(inode, async_extent->start,
 837                                 async_extent->start +
 838                                 async_extent->ram_size - 1,
 839                                 async_extent->start +
 840                                 async_extent->ram_size - 1,
 841                                 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
 842                                 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 843                                 PAGE_SET_WRITEBACK);
 844                 if (btrfs_submit_compressed_write(inode,
 845                                     async_extent->start,
 846                                     async_extent->ram_size,
 847                                     ins.objectid,
 848                                     ins.offset, async_extent->pages,
 849                                     async_extent->nr_pages,
 850                                     async_cow->write_flags)) {
 851                         struct page *p = async_extent->pages[0];
 852                         const u64 start = async_extent->start;
 853                         const u64 end = start + async_extent->ram_size - 1;
 854
 855                         p->mapping = inode->i_mapping;
 856                         btrfs_writepage_endio_finish_ordered(p, start, end, 0);
 857
 858                         p->mapping = NULL;
 859                         extent_clear_unlock_delalloc(inode, start, end, end,
 860                                                      NULL, 0,
 861                                                      PAGE_END_WRITEBACK |
 862                                                      PAGE_SET_ERROR);
 863                         free_async_extent_pages(async_extent);
 864                 }
 865                 alloc_hint = ins.objectid + ins.offset;
 866                 kfree(async_extent);
 867                 cond_resched();
 868         }
 869         return;
 870 out_free_reserve:
 871         btrfs_dec_block_group_reservations(fs_info, ins.objectid);
 872         btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
 873 out_free:
 874         extent_clear_unlock_delalloc(inode, async_extent->start,
 875                                      async_extent->start +
 876                                      async_extent->ram_size - 1,
 877                                      async_extent->start +
 878                                      async_extent->ram_size - 1,
 879                                      NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
 880                                      EXTENT_DELALLOC_NEW |
 881                                      EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
 882                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 883                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
 884                                      PAGE_SET_ERROR);
 885         free_async_extent_pages(async_extent);
 886         kfree(async_extent);
 887         goto again;
 888 }
 889
 890 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 891                                       u64 num_bytes)
 892 {
 893         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 894         struct extent_map *em;
 895         u64 alloc_hint = 0;
 896
 897         read_lock(&em_tree->lock);
 898         em = search_extent_mapping(em_tree, start, num_bytes);
 899         if (em) {
 900                 /*
 901                  * if block start isn't an actual block number then find the
 902                  * first block in this inode and use that as a hint.  If that
 903                  * block is also bogus then just don't worry about it.
 904                  */
 905                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 906                         free_extent_map(em);
 907                         em = search_extent_mapping(em_tree, 0, 0);
 908                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 909                                 alloc_hint = em->block_start;
 910                         if (em)
 911                                 free_extent_map(em);
 912                 } else {
 913                         alloc_hint = em->block_start;
 914                         free_extent_map(em);
 915                 }
 916         }
 917         read_unlock(&em_tree->lock);
 918
 919         return alloc_hint;
 920 }
 921
 922 /*
 923  * when extent_io.c finds a delayed allocation range in the file,
 924  * the call backs end up in this code.  The basic idea is to
 925  * allocate extents on disk for the range, and create ordered data structs
 926  * in ram to track those extents.
 927  *
 928  * locked_page is the page that writepage had locked already.  We use
 929  * it to make sure we don't do extra locks or unlocks.
 930  *
 931  * *page_started is set to one if we unlock locked_page and do everything
 932  * required to start IO on it.  It may be clean and already done with
 933  * IO when we return.
 934  */
 935 static noinline int cow_file_range(struct inode *inode,
 936                                    struct page *locked_page,
 937                                    u64 start, u64 end, u64 delalloc_end,
 938                                    int *page_started, unsigned long *nr_written,
 939                                    int unlock, struct btrfs_dedupe_hash *hash)
 940 {
 941         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 942         struct btrfs_root *root = BTRFS_I(inode)->root;
 943         u64 alloc_hint = 0;
 944         u64 num_bytes;
 945         unsigned long ram_size;
 946         u64 cur_alloc_size = 0;
 947         u64 blocksize = fs_info->sectorsize;
 948         struct btrfs_key ins;
 949         struct extent_map *em;
 950         unsigned clear_bits;
 951         unsigned long page_ops;
 952         bool extent_reserved = false;
 953         int ret = 0;
 954
 955         if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
 956                 WARN_ON_ONCE(1);
 957                 ret = -EINVAL;
 958                 goto out_unlock;
 959         }
 960
 961         num_bytes = ALIGN(end - start + 1, blocksize);
 962         num_bytes = max(blocksize,  num_bytes);
 963         ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
 964
 965         inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
 966
 967         if (start == 0) {
 968                 /* lets try to make an inline extent */
 969                 ret = cow_file_range_inline(inode, start, end, 0,
 970                                             BTRFS_COMPRESS_NONE, NULL);
 971                 if (ret == 0) {
 972                         /*
 973                          * We use DO_ACCOUNTING here because we need the
 974                          * delalloc_release_metadata to be run _after_ we drop
 975                          * our outstanding extent for clearing delalloc for this
 976                          * range.
 977                          */
 978                         extent_clear_unlock_delalloc(inode, start, end,
 979                                      delalloc_end, NULL,
 980                                      EXTENT_LOCKED | EXTENT_DELALLOC |
 981                                      EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
 982                                      EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
 983                                      PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
 984                                      PAGE_END_WRITEBACK);
 985                         *nr_written = *nr_written +
 986                              (end - start + PAGE_SIZE) / PAGE_SIZE;
 987                         *page_started = 1;
 988                         goto out;
 989                 } else if (ret < 0) {
 990                         goto out_unlock;
 991                 }
 992         }
 993
 994         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 995         btrfs_drop_extent_cache(BTRFS_I(inode), start,
 996                         start + num_bytes - 1, 0);
 997
 998         while (num_bytes > 0) {
 999                 cur_alloc_size = num_bytes;
1000                 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1001                                            fs_info->sectorsize, 0, alloc_hint,
1002                                            &ins, 1, 1);
1003                 if (ret < 0)
1004                         goto out_unlock;
1005                 cur_alloc_size = ins.offset;
1006                 extent_reserved = true;
1007
1008                 ram_size = ins.offset;
1009                 em = create_io_em(inode, start, ins.offset, /* len */
1010                                   start, /* orig_start */
1011                                   ins.objectid, /* block_start */
1012                                   ins.offset, /* block_len */
1013                                   ins.offset, /* orig_block_len */
1014                                   ram_size, /* ram_bytes */
1015                                   BTRFS_COMPRESS_NONE, /* compress_type */
1016                                   BTRFS_ORDERED_REGULAR /* type */);
1017                 if (IS_ERR(em)) {
1018                         ret = PTR_ERR(em);
1019                         goto out_reserve;
1020                 }
1021                 free_extent_map(em);
1022
1023                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1024                                                ram_size, cur_alloc_size, 0);
1025                 if (ret)
1026                         goto out_drop_extent_cache;
1027
1028                 if (root->root_key.objectid ==
1029                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1030                         ret = btrfs_reloc_clone_csums(inode, start,
1031                                                       cur_alloc_size);
1032                         /*
1033                          * Only drop cache here, and process as normal.
1034                          *
1035                          * We must not allow extent_clear_unlock_delalloc()
1036                          * at out_unlock label to free meta of this ordered
1037                          * extent, as its meta should be freed by
1038                          * btrfs_finish_ordered_io().
1039                          *
1040                          * So we must continue until @start is increased to
1041                          * skip current ordered extent.
1042                          */
1043                         if (ret)
1044                                 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1045                                                 start + ram_size - 1, 0);
1046                 }
1047
1048                 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1049
1050                 /* we're not doing compressed IO, don't unlock the first
1051                  * page (which the caller expects to stay locked), don't
1052                  * clear any dirty bits and don't set any writeback bits
1053                  *
1054                  * Do set the Private2 bit so we know this page was properly
1055                  * setup for writepage
1056                  */
1057                 page_ops = unlock ? PAGE_UNLOCK : 0;
1058                 page_ops |= PAGE_SET_PRIVATE2;
1059
1060                 extent_clear_unlock_delalloc(inode, start,
1061                                              start + ram_size - 1,
1062                                              delalloc_end, locked_page,
1063                                              EXTENT_LOCKED | EXTENT_DELALLOC,
1064                                              page_ops);
1065                 if (num_bytes < cur_alloc_size)
1066                         num_bytes = 0;
1067                 else
1068                         num_bytes -= cur_alloc_size;
1069                 alloc_hint = ins.objectid + ins.offset;
1070                 start += cur_alloc_size;
1071                 extent_reserved = false;
1072
1073                 /*
1074                  * btrfs_reloc_clone_csums() error, since start is increased
1075                  * extent_clear_unlock_delalloc() at out_unlock label won't
1076                  * free metadata of current ordered extent, we're OK to exit.
1077                  */
1078                 if (ret)
1079                         goto out_unlock;
1080         }
1081 out:
1082         return ret;
1083
1084 out_drop_extent_cache:
1085         btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
1086 out_reserve:
1087         btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1088         btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1089 out_unlock:
1090         clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1091                 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1092         page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1093                 PAGE_END_WRITEBACK;
1094         /*
1095          * If we reserved an extent for our delalloc range (or a subrange) and
1096          * failed to create the respective ordered extent, then it means that
1097          * when we reserved the extent we decremented the extent's size from
1098          * the data space_info's bytes_may_use counter and incremented the
1099          * space_info's bytes_reserved counter by the same amount. We must make
1100          * sure extent_clear_unlock_delalloc() does not try to decrement again
1101          * the data space_info's bytes_may_use counter, therefore we do not pass
1102          * it the flag EXTENT_CLEAR_DATA_RESV.
1103          */
1104         if (extent_reserved) {
1105                 extent_clear_unlock_delalloc(inode, start,
1106                                              start + cur_alloc_size,
1107                                              start + cur_alloc_size,
1108                                              locked_page,
1109                                              clear_bits,
1110                                              page_ops);
1111                 start += cur_alloc_size;
1112                 if (start >= end)
1113                         goto out;
1114         }
1115         extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1116                                      locked_page,
1117                                      clear_bits | EXTENT_CLEAR_DATA_RESV,
1118                                      page_ops);
1119         goto out;
1120 }
1121
1122 /*
1123  * work queue call back to started compression on a file and pages
1124  */
1125 static noinline void async_cow_start(struct btrfs_work *work)
1126 {
1127         struct async_cow *async_cow;
1128         int num_added = 0;
1129         async_cow = container_of(work, struct async_cow, work);
1130
1131         compress_file_range(async_cow->inode, async_cow->locked_page,
1132                             async_cow->start, async_cow->end, async_cow,
1133                             &num_added);
1134         if (num_added == 0) {
1135                 btrfs_add_delayed_iput(async_cow->inode);
1136                 async_cow->inode = NULL;
1137         }
1138 }
1139
1140 /*
1141  * work queue call back to submit previously compressed pages
1142  */
1143 static noinline void async_cow_submit(struct btrfs_work *work)
1144 {
1145         struct btrfs_fs_info *fs_info;
1146         struct async_cow *async_cow;
1147         unsigned long nr_pages;
1148
1149         async_cow = container_of(work, struct async_cow, work);
1150
1151         fs_info = async_cow->fs_info;
1152         nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1153                 PAGE_SHIFT;
1154
1155         /* atomic_sub_return implies a barrier */
1156         if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1157             5 * SZ_1M)
1158                 cond_wake_up_nomb(&fs_info->async_submit_wait);
1159
1160         if (async_cow->inode)
1161                 submit_compressed_extents(async_cow->inode, async_cow);
1162 }
1163
1164 static noinline void async_cow_free(struct btrfs_work *work)
1165 {
1166         struct async_cow *async_cow;
1167         async_cow = container_of(work, struct async_cow, work);
1168         if (async_cow->inode)
1169                 btrfs_add_delayed_iput(async_cow->inode);
1170         kfree(async_cow);
1171 }
1172
1173 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1174                                 u64 start, u64 end, int *page_started,
1175                                 unsigned long *nr_written,
1176                                 unsigned int write_flags)
1177 {
1178         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1179         struct async_cow *async_cow;
1180         unsigned long nr_pages;
1181         u64 cur_end;
1182
1183         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1184                          1, 0, NULL);
1185         while (start < end) {
1186                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1187                 BUG_ON(!async_cow); /* -ENOMEM */
1188                 async_cow->inode = igrab(inode);
1189                 async_cow->fs_info = fs_info;
1190                 async_cow->locked_page = locked_page;
1191                 async_cow->start = start;
1192                 async_cow->write_flags = write_flags;
1193
1194                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1195                     !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1196                         cur_end = end;
1197                 else
1198                         cur_end = min(end, start + SZ_512K - 1);
1199
1200                 async_cow->end = cur_end;
1201                 INIT_LIST_HEAD(&async_cow->extents);
1202
1203                 btrfs_init_work(&async_cow->work,
1204                                 btrfs_delalloc_helper,
1205                                 async_cow_start, async_cow_submit,
1206                                 async_cow_free);
1207
1208                 nr_pages = (cur_end - start + PAGE_SIZE) >>
1209                         PAGE_SHIFT;
1210                 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1211
1212                 btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1213
1214                 *nr_written += nr_pages;
1215                 start = cur_end + 1;
1216         }
1217         *page_started = 1;
1218         return 0;
1219 }
1220
1221 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1222                                         u64 bytenr, u64 num_bytes)
1223 {
1224         int ret;
1225         struct btrfs_ordered_sum *sums;
1226         LIST_HEAD(list);
1227
1228         ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1229                                        bytenr + num_bytes - 1, &list, 0);
1230         if (ret == 0 && list_empty(&list))
1231                 return 0;
1232
1233         while (!list_empty(&list)) {
1234                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1235                 list_del(&sums->list);
1236                 kfree(sums);
1237         }
1238         if (ret < 0)
1239                 return ret;
1240         return 1;
1241 }
1242
1243 /*
1244  * when nowcow writeback call back.  This checks for snapshots or COW copies
1245  * of the extents that exist in the file, and COWs the file as required.
1246  *
1247  * If no cow copies or snapshots exist, we write directly to the existing
1248  * blocks on disk
1249  */
1250 static noinline int run_delalloc_nocow(struct inode *inode,
1251                                        struct page *locked_page,
1252                               u64 start, u64 end, int *page_started, int force,
1253                               unsigned long *nr_written)
1254 {
1255         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1256         struct btrfs_root *root = BTRFS_I(inode)->root;
1257         struct extent_buffer *leaf;
1258         struct btrfs_path *path;
1259         struct btrfs_file_extent_item *fi;
1260         struct btrfs_key found_key;
1261         struct extent_map *em;
1262         u64 cow_start;
1263         u64 cur_offset;
1264         u64 extent_end;
1265         u64 extent_offset;
1266         u64 disk_bytenr;
1267         u64 num_bytes;
1268         u64 disk_num_bytes;
1269         u64 ram_bytes;
1270         int extent_type;
1271         int ret;
1272         int type;
1273         int nocow;
1274         int check_prev = 1;
1275         bool nolock;
1276         u64 ino = btrfs_ino(BTRFS_I(inode));
1277
1278         path = btrfs_alloc_path();
1279         if (!path) {
1280                 extent_clear_unlock_delalloc(inode, start, end, end,
1281                                              locked_page,
1282                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1283                                              EXTENT_DO_ACCOUNTING |
1284                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1285                                              PAGE_CLEAR_DIRTY |
1286                                              PAGE_SET_WRITEBACK |
1287                                              PAGE_END_WRITEBACK);
1288                 return -ENOMEM;
1289         }
1290
1291         nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
1292
1293         cow_start = (u64)-1;
1294         cur_offset = start;
1295         while (1) {
1296                 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1297                                                cur_offset, 0);
1298                 if (ret < 0)
1299                         goto error;
1300                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1301                         leaf = path->nodes[0];
1302                         btrfs_item_key_to_cpu(leaf, &found_key,
1303                                               path->slots[0] - 1);
1304                         if (found_key.objectid == ino &&
1305                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1306                                 path->slots[0]--;
1307                 }
1308                 check_prev = 0;
1309 next_slot:
1310                 leaf = path->nodes[0];
1311                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1312                         ret = btrfs_next_leaf(root, path);
1313                         if (ret < 0) {
1314                                 if (cow_start != (u64)-1)
1315                                         cur_offset = cow_start;
1316                                 goto error;
1317                         }
1318                         if (ret > 0)
1319                                 break;
1320                         leaf = path->nodes[0];
1321                 }
1322
1323                 nocow = 0;
1324                 disk_bytenr = 0;
1325                 num_bytes = 0;
1326                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1327
1328                 if (found_key.objectid > ino)
1329                         break;
1330                 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1331                     found_key.type < BTRFS_EXTENT_DATA_KEY) {
1332                         path->slots[0]++;
1333                         goto next_slot;
1334                 }
1335                 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1336                     found_key.offset > end)
1337                         break;
1338
1339                 if (found_key.offset > cur_offset) {
1340                         extent_end = found_key.offset;
1341                         extent_type = 0;
1342                         goto out_check;
1343                 }
1344
1345                 fi = btrfs_item_ptr(leaf, path->slots[0],
1346                                     struct btrfs_file_extent_item);
1347                 extent_type = btrfs_file_extent_type(leaf, fi);
1348
1349                 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1350                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1351                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1352                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1353                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1354                         extent_end = found_key.offset +
1355                                 btrfs_file_extent_num_bytes(leaf, fi);
1356                         disk_num_bytes =
1357                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1358                         if (extent_end <= start) {
1359                                 path->slots[0]++;
1360                                 goto next_slot;
1361                         }
1362                         if (disk_bytenr == 0)
1363                                 goto out_check;
1364                         if (btrfs_file_extent_compression(leaf, fi) ||
1365                             btrfs_file_extent_encryption(leaf, fi) ||
1366                             btrfs_file_extent_other_encoding(leaf, fi))
1367                                 goto out_check;
1368                         /*
1369                          * Do the same check as in btrfs_cross_ref_exist but
1370                          * without the unnecessary search.
1371                          */
1372                         if (!nolock &&
1373                             btrfs_file_extent_generation(leaf, fi) <=
1374                             btrfs_root_last_snapshot(&root->root_item))
1375                                 goto out_check;
1376                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1377                                 goto out_check;
1378                         if (btrfs_extent_readonly(fs_info, disk_bytenr))
1379                                 goto out_check;
1380                         ret = btrfs_cross_ref_exist(root, ino,
1381                                                     found_key.offset -
1382                                                     extent_offset, disk_bytenr);
1383                         if (ret) {
1384                                 /*
1385                                  * ret could be -EIO if the above fails to read
1386                                  * metadata.
1387                                  */
1388                                 if (ret < 0) {
1389                                         if (cow_start != (u64)-1)
1390                                                 cur_offset = cow_start;
1391                                         goto error;
1392                                 }
1393
1394                                 WARN_ON_ONCE(nolock);
1395                                 goto out_check;
1396                         }
1397                         disk_bytenr += extent_offset;
1398                         disk_bytenr += cur_offset - found_key.offset;
1399                         num_bytes = min(end + 1, extent_end) - cur_offset;
1400                         /*
1401                          * if there are pending snapshots for this root,
1402                          * we fall into common COW way.
1403                          */
1404                         if (!nolock && atomic_read(&root->snapshot_force_cow))
1405                                 goto out_check;
1406                         /*
1407                          * force cow if csum exists in the range.
1408                          * this ensure that csum for a given extent are
1409                          * either valid or do not exist.
1410                          */
1411                         ret = csum_exist_in_range(fs_info, disk_bytenr,
1412                                                   num_bytes);
1413                         if (ret) {
1414                                 /*
1415                                  * ret could be -EIO if the above fails to read
1416                                  * metadata.
1417                                  */
1418                                 if (ret < 0) {
1419                                         if (cow_start != (u64)-1)
1420                                                 cur_offset = cow_start;
1421                                         goto error;
1422                                 }
1423                                 WARN_ON_ONCE(nolock);
1424                                 goto out_check;
1425                         }
1426                         if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1427                                 goto out_check;
1428                         nocow = 1;
1429                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1430                         extent_end = found_key.offset +
1431                                 btrfs_file_extent_ram_bytes(leaf, fi);
1432                         extent_end = ALIGN(extent_end,
1433                                            fs_info->sectorsize);
1434                 } else {
1435                         BUG_ON(1);
1436                 }
1437 out_check:
1438                 if (extent_end <= start) {
1439                         path->slots[0]++;
1440                         if (nocow)
1441                                 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1442                         goto next_slot;
1443                 }
1444                 if (!nocow) {
1445                         if (cow_start == (u64)-1)
1446                                 cow_start = cur_offset;
1447                         cur_offset = extent_end;
1448                         if (cur_offset > end)
1449                                 break;
1450                         path->slots[0]++;
1451                         goto next_slot;
1452                 }
1453
1454                 btrfs_release_path(path);
1455                 if (cow_start != (u64)-1) {
1456                         ret = cow_file_range(inode, locked_page,
1457                                              cow_start, found_key.offset - 1,
1458                                              end, page_started, nr_written, 1,
1459                                              NULL);
1460                         if (ret) {
1461                                 if (nocow)
1462                                         btrfs_dec_nocow_writers(fs_info,
1463                                                                 disk_bytenr);
1464                                 goto error;
1465                         }
1466                         cow_start = (u64)-1;
1467                 }
1468
1469                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1470                         u64 orig_start = found_key.offset - extent_offset;
1471
1472                         em = create_io_em(inode, cur_offset, num_bytes,
1473                                           orig_start,
1474                                           disk_bytenr, /* block_start */
1475                                           num_bytes, /* block_len */
1476                                           disk_num_bytes, /* orig_block_len */
1477                                           ram_bytes, BTRFS_COMPRESS_NONE,
1478                                           BTRFS_ORDERED_PREALLOC);
1479                         if (IS_ERR(em)) {
1480                                 if (nocow)
1481                                         btrfs_dec_nocow_writers(fs_info,
1482                                                                 disk_bytenr);
1483                                 ret = PTR_ERR(em);
1484                                 goto error;
1485                         }
1486                         free_extent_map(em);
1487                 }
1488
1489                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1490                         type = BTRFS_ORDERED_PREALLOC;
1491                 } else {
1492                         type = BTRFS_ORDERED_NOCOW;
1493                 }
1494
1495                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1496                                                num_bytes, num_bytes, type);
1497                 if (nocow)
1498                         btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1499                 BUG_ON(ret); /* -ENOMEM */
1500
1501                 if (root->root_key.objectid ==
1502                     BTRFS_DATA_RELOC_TREE_OBJECTID)
1503                         /*
1504                          * Error handled later, as we must prevent
1505                          * extent_clear_unlock_delalloc() in error handler
1506                          * from freeing metadata of created ordered extent.
1507                          */
1508                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1509                                                       num_bytes);
1510
1511                 extent_clear_unlock_delalloc(inode, cur_offset,
1512                                              cur_offset + num_bytes - 1, end,
1513                                              locked_page, EXTENT_LOCKED |
1514                                              EXTENT_DELALLOC |
1515                                              EXTENT_CLEAR_DATA_RESV,
1516                                              PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1517
1518                 cur_offset = extent_end;
1519
1520                 /*
1521                  * btrfs_reloc_clone_csums() error, now we're OK to call error
1522                  * handler, as metadata for created ordered extent will only
1523                  * be freed by btrfs_finish_ordered_io().
1524                  */
1525                 if (ret)
1526                         goto error;
1527                 if (cur_offset > end)
1528                         break;
1529         }
1530         btrfs_release_path(path);
1531
1532         if (cur_offset <= end && cow_start == (u64)-1)
1533                 cow_start = cur_offset;
1534
1535         if (cow_start != (u64)-1) {
1536                 cur_offset = end;
1537                 ret = cow_file_range(inode, locked_page, cow_start, end, end,
1538                                      page_started, nr_written, 1, NULL);
1539                 if (ret)
1540                         goto error;
1541         }
1542
1543 error:
1544         if (ret && cur_offset < end)
1545                 extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1546                                              locked_page, EXTENT_LOCKED |
1547                                              EXTENT_DELALLOC | EXTENT_DEFRAG |
1548                                              EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1549                                              PAGE_CLEAR_DIRTY |
1550                                              PAGE_SET_WRITEBACK |
1551                                              PAGE_END_WRITEBACK);
1552         btrfs_free_path(path);
1553         return ret;
1554 }
1555
1556 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1557 {
1558
1559         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1560             !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1561                 return 0;
1562
1563         /*
1564          * @defrag_bytes is a hint value, no spinlock held here,
1565          * if is not zero, it means the file is defragging.
1566          * Force cow if given extent needs to be defragged.
1567          */
1568         if (BTRFS_I(inode)->defrag_bytes &&
1569             test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1570                            EXTENT_DEFRAG, 0, NULL))
1571                 return 1;
1572
1573         return 0;
1574 }
1575
1576 /*
1577  * Function to process delayed allocation (create CoW) for ranges which are
1578  * being touched for the first time.
1579  */
1580 int btrfs_run_delalloc_range(void *private_data, struct page *locked_page,
1581                 u64 start, u64 end, int *page_started, unsigned long *nr_written,
1582                 struct writeback_control *wbc)
1583 {
1584         struct inode *inode = private_data;
1585         int ret;
1586         int force_cow = need_force_cow(inode, start, end);
1587         unsigned int write_flags = wbc_to_write_flags(wbc);
1588
1589         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1590                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1591                                          page_started, 1, nr_written);
1592         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1593                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1594                                          page_started, 0, nr_written);
1595         } else if (!inode_need_compress(inode, start, end)) {
1596                 ret = cow_file_range(inode, locked_page, start, end, end,
1597                                       page_started, nr_written, 1, NULL);
1598         } else {
1599                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1600                         &BTRFS_I(inode)->runtime_flags);
1601                 ret = cow_file_range_async(inode, locked_page, start, end,
1602                                            page_started, nr_written,
1603                                            write_flags);
1604         }
1605         if (ret)
1606                 btrfs_cleanup_ordered_extents(inode, start, end - start + 1);
1607         return ret;
1608 }
1609
1610 void btrfs_split_delalloc_extent(struct inode *inode,
1611                                  struct extent_state *orig, u64 split)
1612 {
1613         u64 size;
1614
1615         /* not delalloc, ignore it */
1616         if (!(orig->state & EXTENT_DELALLOC))
1617                 return;
1618
1619         size = orig->end - orig->start + 1;
1620         if (size > BTRFS_MAX_EXTENT_SIZE) {
1621                 u32 num_extents;
1622                 u64 new_size;
1623
1624                 /*
1625                  * See the explanation in btrfs_merge_delalloc_extent, the same
1626                  * applies here, just in reverse.
1627                  */
1628                 new_size = orig->end - split + 1;
1629                 num_extents = count_max_extents(new_size);
1630                 new_size = split - orig->start;
1631                 num_extents += count_max_extents(new_size);
1632                 if (count_max_extents(size) >= num_extents)
1633                         return;
1634         }
1635
1636         spin_lock(&BTRFS_I(inode)->lock);
1637         btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1638         spin_unlock(&BTRFS_I(inode)->lock);
1639 }
1640
1641 /*
1642  * Handle merged delayed allocation extents so we can keep track of new extents
1643  * that are just merged onto old extents, such as when we are doing sequential
1644  * writes, so we can properly account for the metadata space we'll need.
1645  */
1646 void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new,
1647                                  struct extent_state *other)
1648 {
1649         u64 new_size, old_size;
1650         u32 num_extents;
1651
1652         /* not delalloc, ignore it */
1653         if (!(other->state & EXTENT_DELALLOC))
1654                 return;
1655
1656         if (new->start > other->start)
1657                 new_size = new->end - other->start + 1;
1658         else
1659                 new_size = other->end - new->start + 1;
1660
1661         /* we're not bigger than the max, unreserve the space and go */
1662         if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1663                 spin_lock(&BTRFS_I(inode)->lock);
1664                 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1665                 spin_unlock(&BTRFS_I(inode)->lock);
1666                 return;
1667         }
1668
1669         /*
1670          * We have to add up either side to figure out how many extents were
1671          * accounted for before we merged into one big extent.  If the number of
1672          * extents we accounted for is <= the amount we need for the new range
1673          * then we can return, otherwise drop.  Think of it like this
1674          *
1675          * [ 4k][MAX_SIZE]
1676          *
1677          * So we've grown the extent by a MAX_SIZE extent, this would mean we
1678          * need 2 outstanding extents, on one side we have 1 and the other side
1679          * we have 1 so they are == and we can return.  But in this case
1680          *
1681          * [MAX_SIZE+4k][MAX_SIZE+4k]
1682          *
1683          * Each range on their own accounts for 2 extents, but merged together
1684          * they are only 3 extents worth of accounting, so we need to drop in
1685          * this case.
1686          */
1687         old_size = other->end - other->start + 1;
1688         num_extents = count_max_extents(old_size);
1689         old_size = new->end - new->start + 1;
1690         num_extents += count_max_extents(old_size);
1691         if (count_max_extents(new_size) >= num_extents)
1692                 return;
1693
1694         spin_lock(&BTRFS_I(inode)->lock);
1695         btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1696         spin_unlock(&BTRFS_I(inode)->lock);
1697 }
1698
1699 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1700                                       struct inode *inode)
1701 {
1702         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1703
1704         spin_lock(&root->delalloc_lock);
1705         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1706                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1707                               &root->delalloc_inodes);
1708                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1709                         &BTRFS_I(inode)->runtime_flags);
1710                 root->nr_delalloc_inodes++;
1711                 if (root->nr_delalloc_inodes == 1) {
1712                         spin_lock(&fs_info->delalloc_root_lock);
1713                         BUG_ON(!list_empty(&root->delalloc_root));
1714                         list_add_tail(&root->delalloc_root,
1715                                       &fs_info->delalloc_roots);
1716                         spin_unlock(&fs_info->delalloc_root_lock);
1717                 }
1718         }
1719         spin_unlock(&root->delalloc_lock);
1720 }
1721
1722
1723 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
1724                                 struct btrfs_inode *inode)
1725 {
1726         struct btrfs_fs_info *fs_info = root->fs_info;
1727
1728         if (!list_empty(&inode->delalloc_inodes)) {
1729                 list_del_init(&inode->delalloc_inodes);
1730                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1731                           &inode->runtime_flags);
1732                 root->nr_delalloc_inodes--;
1733                 if (!root->nr_delalloc_inodes) {
1734                         ASSERT(list_empty(&root->delalloc_inodes));
1735                         spin_lock(&fs_info->delalloc_root_lock);
1736                         BUG_ON(list_empty(&root->delalloc_root));
1737                         list_del_init(&root->delalloc_root);
1738                         spin_unlock(&fs_info->delalloc_root_lock);
1739                 }
1740         }
1741 }
1742
1743 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1744                                      struct btrfs_inode *inode)
1745 {
1746         spin_lock(&root->delalloc_lock);
1747         __btrfs_del_delalloc_inode(root, inode);
1748         spin_unlock(&root->delalloc_lock);
1749 }
1750
1751 /*
1752  * Properly track delayed allocation bytes in the inode and to maintain the
1753  * list of inodes that have pending delalloc work to be done.
1754  */
1755 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
1756                                unsigned *bits)
1757 {
1758         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1759
1760         if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1761                 WARN_ON(1);
1762         /*
1763          * set_bit and clear bit hooks normally require _irqsave/restore
1764          * but in this case, we are only testing for the DELALLOC
1765          * bit, which is only set or cleared with irqs on
1766          */
1767         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1768                 struct btrfs_root *root = BTRFS_I(inode)->root;
1769                 u64 len = state->end + 1 - state->start;
1770                 u32 num_extents = count_max_extents(len);
1771                 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1772
1773                 spin_lock(&BTRFS_I(inode)->lock);
1774                 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
1775                 spin_unlock(&BTRFS_I(inode)->lock);
1776
1777                 /* For sanity tests */
1778                 if (btrfs_is_testing(fs_info))
1779                         return;
1780
1781                 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
1782                                          fs_info->delalloc_batch);
1783                 spin_lock(&BTRFS_I(inode)->lock);
1784                 BTRFS_I(inode)->delalloc_bytes += len;
1785                 if (*bits & EXTENT_DEFRAG)
1786                         BTRFS_I(inode)->defrag_bytes += len;
1787                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1788                                          &BTRFS_I(inode)->runtime_flags))
1789                         btrfs_add_delalloc_inodes(root, inode);
1790                 spin_unlock(&BTRFS_I(inode)->lock);
1791         }
1792
1793         if (!(state->state & EXTENT_DELALLOC_NEW) &&
1794             (*bits & EXTENT_DELALLOC_NEW)) {
1795                 spin_lock(&BTRFS_I(inode)->lock);
1796                 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
1797                         state->start;
1798                 spin_unlock(&BTRFS_I(inode)->lock);
1799         }
1800 }
1801
1802 /*
1803  * Once a range is no longer delalloc this function ensures that proper
1804  * accounting happens.
1805  */
1806 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
1807                                  struct extent_state *state, unsigned *bits)
1808 {
1809         struct btrfs_inode *inode = BTRFS_I(vfs_inode);
1810         struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
1811         u64 len = state->end + 1 - state->start;
1812         u32 num_extents = count_max_extents(len);
1813
1814         if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
1815                 spin_lock(&inode->lock);
1816                 inode->defrag_bytes -= len;
1817                 spin_unlock(&inode->lock);
1818         }
1819
1820         /*
1821          * set_bit and clear bit hooks normally require _irqsave/restore
1822          * but in this case, we are only testing for the DELALLOC
1823          * bit, which is only set or cleared with irqs on
1824          */
1825         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1826                 struct btrfs_root *root = inode->root;
1827                 bool do_list = !btrfs_is_free_space_inode(inode);
1828
1829                 spin_lock(&inode->lock);
1830                 btrfs_mod_outstanding_extents(inode, -num_extents);
1831                 spin_unlock(&inode->lock);
1832
1833                 /*
1834                  * We don't reserve metadata space for space cache inodes so we
1835                  * don't need to call dellalloc_release_metadata if there is an
1836                  * error.
1837                  */
1838                 if (*bits & EXTENT_CLEAR_META_RESV &&
1839                     root != fs_info->tree_root)
1840                         btrfs_delalloc_release_metadata(inode, len, false);
1841
1842                 /* For sanity tests. */
1843                 if (btrfs_is_testing(fs_info))
1844                         return;
1845
1846                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
1847                     do_list && !(state->state & EXTENT_NORESERVE) &&
1848                     (*bits & EXTENT_CLEAR_DATA_RESV))
1849                         btrfs_free_reserved_data_space_noquota(
1850                                         &inode->vfs_inode,
1851                                         state->start, len);
1852
1853                 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
1854                                          fs_info->delalloc_batch);
1855                 spin_lock(&inode->lock);
1856                 inode->delalloc_bytes -= len;
1857                 if (do_list && inode->delalloc_bytes == 0 &&
1858                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1859                                         &inode->runtime_flags))
1860                         btrfs_del_delalloc_inode(root, inode);
1861                 spin_unlock(&inode->lock);
1862         }
1863
1864         if ((state->state & EXTENT_DELALLOC_NEW) &&
1865             (*bits & EXTENT_DELALLOC_NEW)) {
1866                 spin_lock(&inode->lock);
1867                 ASSERT(inode->new_delalloc_bytes >= len);
1868                 inode->new_delalloc_bytes -= len;
1869                 spin_unlock(&inode->lock);
1870         }
1871 }
1872
1873 /*
1874  * btrfs_bio_fits_in_stripe - Checks whether the size of the given bio will fit
1875  * in a chunk's stripe. This function ensures that bios do not span a
1876  * stripe/chunk
1877  *
1878  * @page - The page we are about to add to the bio
1879  * @size - size we want to add to the bio
1880  * @bio - bio we want to ensure is smaller than a stripe
1881  * @bio_flags - flags of the bio
1882  *
1883  * return 1 if page cannot be added to the bio
1884  * return 0 if page can be added to the bio
1885  * return error otherwise
1886  */
1887 int btrfs_bio_fits_in_stripe(struct page *page, size_t size, struct bio *bio,
1888                              unsigned long bio_flags)
1889 {
1890         struct inode *inode = page->mapping->host;
1891         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1892         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1893         u64 length = 0;
1894         u64 map_length;
1895         int ret;
1896
1897         if (bio_flags & EXTENT_BIO_COMPRESSED)
1898                 return 0;
1899
1900         length = bio->bi_iter.bi_size;
1901         map_length = length;
1902         ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1903                               NULL, 0);
1904         if (ret < 0)
1905                 return ret;
1906         if (map_length < length + size)
1907                 return 1;
1908         return 0;
1909 }
1910
1911 /*
1912  * in order to insert checksums into the metadata in large chunks,
1913  * we wait until bio submission time.   All the pages in the bio are
1914  * checksummed and sums are attached onto the ordered extent record.
1915  *
1916  * At IO completion time the cums attached on the ordered extent record
1917  * are inserted into the btree
1918  */
1919 static blk_status_t btrfs_submit_bio_start(void *private_data, struct bio *bio,
1920                                     u64 bio_offset)
1921 {
1922         struct inode *inode = private_data;
1923         blk_status_t ret = 0;
1924
1925         ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1926         BUG_ON(ret); /* -ENOMEM */
1927         return 0;
1928 }
1929
1930 /*
1931  * extent_io.c submission hook. This does the right thing for csum calculation
1932  * on write, or reading the csums from the tree before a read.
1933  *
1934  * Rules about async/sync submit,
1935  * a) read:                             sync submit
1936  *
1937  * b) write without checksum:           sync submit
1938  *
1939  * c) write with checksum:
1940  *    c-1) if bio is issued by fsync:   sync submit
1941  *         (sync_writers != 0)
1942  *
1943  *    c-2) if root is reloc root:       sync submit
1944  *         (only in case of buffered IO)
1945  *
1946  *    c-3) otherwise:                   async submit
1947  */
1948 static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio,
1949                                  int mirror_num, unsigned long bio_flags,
1950                                  u64 bio_offset)
1951 {
1952         struct inode *inode = private_data;
1953         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1954         struct btrfs_root *root = BTRFS_I(inode)->root;
1955         enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1956         blk_status_t ret = 0;
1957         int skip_sum;
1958         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1959
1960         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1961
1962         if (btrfs_is_free_space_inode(BTRFS_I(inode)))
1963                 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1964
1965         if (bio_op(bio) != REQ_OP_WRITE) {
1966                 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
1967                 if (ret)
1968                         goto out;
1969
1970                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1971                         ret = btrfs_submit_compressed_read(inode, bio,
1972                                                            mirror_num,
1973                                                            bio_flags);
1974                         goto out;
1975                 } else if (!skip_sum) {
1976                         ret = btrfs_lookup_bio_sums(inode, bio, NULL);
1977                         if (ret)
1978                                 goto out;
1979                 }
1980                 goto mapit;
1981         } else if (async && !skip_sum) {
1982                 /* csum items have already been cloned */
1983                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1984                         goto mapit;
1985                 /* we're doing a write, do the async checksumming */
1986                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
1987                                           bio_offset, inode,
1988                                           btrfs_submit_bio_start);
1989                 goto out;
1990         } else if (!skip_sum) {
1991                 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1992                 if (ret)
1993                         goto out;
1994         }
1995
1996 mapit:
1997         ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
1998
1999 out:
2000         if (ret) {
2001                 bio->bi_status = ret;
2002                 bio_endio(bio);
2003         }
2004         return ret;
2005 }
2006
2007 /*
2008  * given a list of ordered sums record them in the inode.  This happens
2009  * at IO completion time based on sums calculated at bio submission time.
2010  */
2011 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2012                              struct inode *inode, struct list_head *list)
2013 {
2014         struct btrfs_ordered_sum *sum;
2015         int ret;
2016
2017         list_for_each_entry(sum, list, list) {
2018                 trans->adding_csums = true;
2019                 ret = btrfs_csum_file_blocks(trans,
2020                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
2021                 trans->adding_csums = false;
2022                 if (ret)
2023                         return ret;
2024         }
2025         return 0;
2026 }
2027
2028 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2029                               unsigned int extra_bits,
2030                               struct extent_state **cached_state, int dedupe)
2031 {
2032         WARN_ON((end & (PAGE_SIZE - 1)) == 0);
2033         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2034                                    extra_bits, cached_state);
2035 }
2036
2037 /* see btrfs_writepage_start_hook for details on why this is required */
2038 struct btrfs_writepage_fixup {
2039         struct page *page;
2040         struct btrfs_work work;
2041 };
2042
2043 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2044 {
2045         struct btrfs_writepage_fixup *fixup;
2046         struct btrfs_ordered_extent *ordered;
2047         struct extent_state *cached_state = NULL;
2048         struct extent_changeset *data_reserved = NULL;
2049         struct page *page;
2050         struct inode *inode;
2051         u64 page_start;
2052         u64 page_end;
2053         int ret;
2054
2055         fixup = container_of(work, struct btrfs_writepage_fixup, work);
2056         page = fixup->page;
2057 again:
2058         lock_page(page);
2059         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2060                 ClearPageChecked(page);
2061                 goto out_page;
2062         }
2063
2064         inode = page->mapping->host;
2065         page_start = page_offset(page);
2066         page_end = page_offset(page) + PAGE_SIZE - 1;
2067
2068         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2069                          &cached_state);
2070
2071         /* already ordered? We're done */
2072         if (PagePrivate2(page))
2073                 goto out;
2074
2075         ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2076                                         PAGE_SIZE);
2077         if (ordered) {
2078                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2079                                      page_end, &cached_state);
2080                 unlock_page(page);
2081                 btrfs_start_ordered_extent(inode, ordered, 1);
2082                 btrfs_put_ordered_extent(ordered);
2083                 goto again;
2084         }
2085
2086         ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2087                                            PAGE_SIZE);
2088         if (ret) {
2089                 mapping_set_error(page->mapping, ret);
2090                 end_extent_writepage(page, ret, page_start, page_end);
2091                 ClearPageChecked(page);
2092                 goto out;
2093          }
2094
2095         ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2096                                         &cached_state, 0);
2097         if (ret) {
2098                 mapping_set_error(page->mapping, ret);
2099                 end_extent_writepage(page, ret, page_start, page_end);
2100                 ClearPageChecked(page);
2101                 goto out;
2102         }
2103
2104         ClearPageChecked(page);
2105         set_page_dirty(page);
2106         btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, false);
2107 out:
2108         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2109                              &cached_state);
2110 out_page:
2111         unlock_page(page);
2112         put_page(page);
2113         kfree(fixup);
2114         extent_changeset_free(data_reserved);
2115 }
2116
2117 /*
2118  * There are a few paths in the higher layers of the kernel that directly
2119  * set the page dirty bit without asking the filesystem if it is a
2120  * good idea.  This causes problems because we want to make sure COW
2121  * properly happens and the data=ordered rules are followed.
2122  *
2123  * In our case any range that doesn't have the ORDERED bit set
2124  * hasn't been properly setup for IO.  We kick off an async process
2125  * to fix it up.  The async helper will wait for ordered extents, set
2126  * the delalloc bit and make it safe to write the page.
2127  */
2128 int btrfs_writepage_cow_fixup(struct page *page, u64 start, u64 end)
2129 {
2130         struct inode *inode = page->mapping->host;
2131         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2132         struct btrfs_writepage_fixup *fixup;
2133
2134         /* this page is properly in the ordered list */
2135         if (TestClearPagePrivate2(page))
2136                 return 0;
2137
2138         if (PageChecked(page))
2139                 return -EAGAIN;
2140
2141         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2142         if (!fixup)
2143                 return -EAGAIN;
2144
2145         SetPageChecked(page);
2146         get_page(page);
2147         btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2148                         btrfs_writepage_fixup_worker, NULL, NULL);
2149         fixup->page = page;
2150         btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2151         return -EBUSY;
2152 }
2153
2154 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2155                                        struct inode *inode, u64 file_pos,
2156                                        u64 disk_bytenr, u64 disk_num_bytes,
2157                                        u64 num_bytes, u64 ram_bytes,
2158                                        u8 compression, u8 encryption,
2159                                        u16 other_encoding, int extent_type)
2160 {
2161         struct btrfs_root *root = BTRFS_I(inode)->root;
2162         struct btrfs_file_extent_item *fi;
2163         struct btrfs_path *path;
2164         struct extent_buffer *leaf;
2165         struct btrfs_key ins;
2166         u64 qg_released;
2167         int extent_inserted = 0;
2168         int ret;
2169
2170         path = btrfs_alloc_path();
2171         if (!path)
2172                 return -ENOMEM;
2173
2174         /*
2175          * we may be replacing one extent in the tree with another.
2176          * The new extent is pinned in the extent map, and we don't want
2177          * to drop it from the cache until it is completely in the btree.
2178          *
2179          * So, tell btrfs_drop_extents to leave this extent in the cache.
2180          * the caller is expected to unpin it and allow it to be merged
2181          * with the others.
2182          */
2183         ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2184                                    file_pos + num_bytes, NULL, 0,
2185                                    1, sizeof(*fi), &extent_inserted);
2186         if (ret)
2187                 goto out;
2188
2189         if (!extent_inserted) {
2190                 ins.objectid = btrfs_ino(BTRFS_I(inode));
2191                 ins.offset = file_pos;
2192                 ins.type = BTRFS_EXTENT_DATA_KEY;
2193
2194                 path->leave_spinning = 1;
2195                 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2196                                               sizeof(*fi));
2197                 if (ret)
2198                         goto out;
2199         }
2200         leaf = path->nodes[0];
2201         fi = btrfs_item_ptr(leaf, path->slots[0],
2202                             struct btrfs_file_extent_item);
2203         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2204         btrfs_set_file_extent_type(leaf, fi, extent_type);
2205         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2206         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2207         btrfs_set_file_extent_offset(leaf, fi, 0);
2208         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2209         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2210         btrfs_set_file_extent_compression(leaf, fi, compression);
2211         btrfs_set_file_extent_encryption(leaf, fi, encryption);
2212         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2213
2214         btrfs_mark_buffer_dirty(leaf);
2215         btrfs_release_path(path);
2216
2217         inode_add_bytes(inode, num_bytes);
2218
2219         ins.objectid = disk_bytenr;
2220         ins.offset = disk_num_bytes;
2221         ins.type = BTRFS_EXTENT_ITEM_KEY;
2222
2223         /*
2224          * Release the reserved range from inode dirty range map, as it is
2225          * already moved into delayed_ref_head
2226          */
2227         ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2228         if (ret < 0)
2229                 goto out;
2230         qg_released = ret;
2231         ret = btrfs_alloc_reserved_file_extent(trans, root,
2232                                                btrfs_ino(BTRFS_I(inode)),
2233                                                file_pos, qg_released, &ins);
2234 out:
2235         btrfs_free_path(path);
2236
2237         return ret;
2238 }
2239
2240 /* snapshot-aware defrag */
2241 struct sa_defrag_extent_backref {
2242         struct rb_node node;
2243         struct old_sa_defrag_extent *old;
2244         u64 root_id;
2245         u64 inum;
2246         u64 file_pos;
2247         u64 extent_offset;
2248         u64 num_bytes;
2249         u64 generation;
2250 };
2251
2252 struct old_sa_defrag_extent {
2253         struct list_head list;
2254         struct new_sa_defrag_extent *new;
2255
2256         u64 extent_offset;
2257         u64 bytenr;
2258         u64 offset;
2259         u64 len;
2260         int count;
2261 };
2262
2263 struct new_sa_defrag_extent {
2264         struct rb_root root;
2265         struct list_head head;
2266         struct btrfs_path *path;
2267         struct inode *inode;
2268         u64 file_pos;
2269         u64 len;
2270         u64 bytenr;
2271         u64 disk_len;
2272         u8 compress_type;
2273 };
2274
2275 static int backref_comp(struct sa_defrag_extent_backref *b1,
2276                         struct sa_defrag_extent_backref *b2)
2277 {
2278         if (b1->root_id < b2->root_id)
2279                 return -1;
2280         else if (b1->root_id > b2->root_id)
2281                 return 1;
2282
2283         if (b1->inum < b2->inum)
2284                 return -1;
2285         else if (b1->inum > b2->inum)
2286                 return 1;
2287
2288         if (b1->file_pos < b2->file_pos)
2289                 return -1;
2290         else if (b1->file_pos > b2->file_pos)
2291                 return 1;
2292
2293         /*
2294          * [------------------------------] ===> (a range of space)
2295          *     |<--->|   |<---->| =============> (fs/file tree A)
2296          * |<---------------------------->| ===> (fs/file tree B)
2297          *
2298          * A range of space can refer to two file extents in one tree while
2299          * refer to only one file extent in another tree.
2300          *
2301          * So we may process a disk offset more than one time(two extents in A)
2302          * and locate at the same extent(one extent in B), then insert two same
2303          * backrefs(both refer to the extent in B).
2304          */
2305         return 0;
2306 }
2307
2308 static void backref_insert(struct rb_root *root,
2309                            struct sa_defrag_extent_backref *backref)
2310 {
2311         struct rb_node **p = &root->rb_node;
2312         struct rb_node *parent = NULL;
2313         struct sa_defrag_extent_backref *entry;
2314         int ret;
2315
2316         while (*p) {
2317                 parent = *p;
2318                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2319
2320                 ret = backref_comp(backref, entry);
2321                 if (ret < 0)
2322                         p = &(*p)->rb_left;
2323                 else
2324                         p = &(*p)->rb_right;
2325         }
2326
2327         rb_link_node(&backref->node, parent, p);
2328         rb_insert_color(&backref->node, root);
2329 }
2330
2331 /*
2332  * Note the backref might has changed, and in this case we just return 0.
2333  */
2334 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2335                                        void *ctx)
2336 {
2337         struct btrfs_file_extent_item *extent;
2338         struct old_sa_defrag_extent *old = ctx;
2339         struct new_sa_defrag_extent *new = old->new;
2340         struct btrfs_path *path = new->path;
2341         struct btrfs_key key;
2342         struct btrfs_root *root;
2343         struct sa_defrag_extent_backref *backref;
2344         struct extent_buffer *leaf;
2345         struct inode *inode = new->inode;
2346         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2347         int slot;
2348         int ret;
2349         u64 extent_offset;
2350         u64 num_bytes;
2351
2352         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2353             inum == btrfs_ino(BTRFS_I(inode)))
2354                 return 0;
2355
2356         key.objectid = root_id;
2357         key.type = BTRFS_ROOT_ITEM_KEY;
2358         key.offset = (u64)-1;
2359
2360         root = btrfs_read_fs_root_no_name(fs_info, &key);
2361         if (IS_ERR(root)) {
2362                 if (PTR_ERR(root) == -ENOENT)
2363                         return 0;
2364                 WARN_ON(1);
2365                 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2366                          inum, offset, root_id);
2367                 return PTR_ERR(root);
2368         }
2369
2370         key.objectid = inum;
2371         key.type = BTRFS_EXTENT_DATA_KEY;
2372         if (offset > (u64)-1 << 32)
2373                 key.offset = 0;
2374         else
2375                 key.offset = offset;
2376
2377         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2378         if (WARN_ON(ret < 0))
2379                 return ret;
2380         ret = 0;
2381
2382         while (1) {
2383                 cond_resched();
2384
2385                 leaf = path->nodes[0];
2386                 slot = path->slots[0];
2387
2388                 if (slot >= btrfs_header_nritems(leaf)) {
2389                         ret = btrfs_next_leaf(root, path);
2390                         if (ret < 0) {
2391                                 goto out;
2392                         } else if (ret > 0) {
2393                                 ret = 0;
2394                                 goto out;
2395                         }
2396                         continue;
2397                 }
2398
2399                 path->slots[0]++;
2400
2401                 btrfs_item_key_to_cpu(leaf, &key, slot);
2402
2403                 if (key.objectid > inum)
2404                         goto out;
2405
2406                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2407                         continue;
2408
2409                 extent = btrfs_item_ptr(leaf, slot,
2410                                         struct btrfs_file_extent_item);
2411
2412                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2413                         continue;
2414
2415                 /*
2416                  * 'offset' refers to the exact key.offset,
2417                  * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2418                  * (key.offset - extent_offset).
2419                  */
2420                 if (key.offset != offset)
2421                         continue;
2422
2423                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2424                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2425
2426                 if (extent_offset >= old->extent_offset + old->offset +
2427                     old->len || extent_offset + num_bytes <=
2428                     old->extent_offset + old->offset)
2429                         continue;
2430                 break;
2431         }
2432
2433         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2434         if (!backref) {
2435                 ret = -ENOENT;
2436                 goto out;
2437         }
2438
2439         backref->root_id = root_id;
2440         backref->inum = inum;
2441         backref->file_pos = offset;
2442         backref->num_bytes = num_bytes;
2443         backref->extent_offset = extent_offset;
2444         backref->generation = btrfs_file_extent_generation(leaf, extent);
2445         backref->old = old;
2446         backref_insert(&new->root, backref);
2447         old->count++;
2448 out:
2449         btrfs_release_path(path);
2450         WARN_ON(ret);
2451         return ret;
2452 }
2453
2454 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2455                                    struct new_sa_defrag_extent *new)
2456 {
2457         struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2458         struct old_sa_defrag_extent *old, *tmp;
2459         int ret;
2460
2461         new->path = path;
2462
2463         list_for_each_entry_safe(old, tmp, &new->head, list) {
2464                 ret = iterate_inodes_from_logical(old->bytenr +
2465                                                   old->extent_offset, fs_info,
2466                                                   path, record_one_backref,
2467                                                   old, false);
2468                 if (ret < 0 && ret != -ENOENT)
2469                         return false;
2470
2471                 /* no backref to be processed for this extent */
2472                 if (!old->count) {
2473                         list_del(&old->list);
2474                         kfree(old);
2475                 }
2476         }
2477
2478         if (list_empty(&new->head))
2479                 return false;
2480
2481         return true;
2482 }
2483
2484 static int relink_is_mergable(struct extent_buffer *leaf,
2485                               struct btrfs_file_extent_item *fi,
2486                               struct new_sa_defrag_extent *new)
2487 {
2488         if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2489                 return 0;
2490
2491         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2492                 return 0;
2493
2494         if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2495                 return 0;
2496
2497         if (btrfs_file_extent_encryption(leaf, fi) ||
2498             btrfs_file_extent_other_encoding(leaf, fi))
2499                 return 0;
2500
2501         return 1;
2502 }
2503
2504 /*
2505  * Note the backref might has changed, and in this case we just return 0.
2506  */
2507 static noinline int relink_extent_backref(struct btrfs_path *path,
2508                                  struct sa_defrag_extent_backref *prev,
2509                                  struct sa_defrag_extent_backref *backref)
2510 {
2511         struct btrfs_file_extent_item *extent;
2512         struct btrfs_file_extent_item *item;
2513         struct btrfs_ordered_extent *ordered;
2514         struct btrfs_trans_handle *trans;
2515         struct btrfs_root *root;
2516         struct btrfs_key key;
2517         struct extent_buffer *leaf;
2518         struct old_sa_defrag_extent *old = backref->old;
2519         struct new_sa_defrag_extent *new = old->new;
2520         struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2521         struct inode *inode;
2522         struct extent_state *cached = NULL;
2523         int ret = 0;
2524         u64 start;
2525         u64 len;
2526         u64 lock_start;
2527         u64 lock_end;
2528         bool merge = false;
2529         int index;
2530
2531         if (prev && prev->root_id == backref->root_id &&
2532             prev->inum == backref->inum &&
2533             prev->file_pos + prev->num_bytes == backref->file_pos)
2534                 merge = true;
2535
2536         /* step 1: get root */
2537         key.objectid = backref->root_id;
2538         key.type = BTRFS_ROOT_ITEM_KEY;
2539         key.offset = (u64)-1;
2540
2541         index = srcu_read_lock(&fs_info->subvol_srcu);
2542
2543         root = btrfs_read_fs_root_no_name(fs_info, &key);
2544         if (IS_ERR(root)) {
2545                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2546                 if (PTR_ERR(root) == -ENOENT)
2547                         return 0;
2548                 return PTR_ERR(root);
2549         }
2550
2551         if (btrfs_root_readonly(root)) {
2552                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2553                 return 0;
2554         }
2555
2556         /* step 2: get inode */
2557         key.objectid = backref->inum;
2558         key.type = BTRFS_INODE_ITEM_KEY;
2559         key.offset = 0;
2560
2561         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2562         if (IS_ERR(inode)) {
2563                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2564                 return 0;
2565         }
2566
2567         srcu_read_unlock(&fs_info->subvol_srcu, index);
2568
2569         /* step 3: relink backref */
2570         lock_start = backref->file_pos;
2571         lock_end = backref->file_pos + backref->num_bytes - 1;
2572         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2573                          &cached);
2574
2575         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2576         if (ordered) {
2577                 btrfs_put_ordered_extent(ordered);
2578                 goto out_unlock;
2579         }
2580
2581         trans = btrfs_join_transaction(root);
2582         if (IS_ERR(trans)) {
2583                 ret = PTR_ERR(trans);
2584                 goto out_unlock;
2585         }
2586
2587         key.objectid = backref->inum;
2588         key.type = BTRFS_EXTENT_DATA_KEY;
2589         key.offset = backref->file_pos;
2590
2591         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2592         if (ret < 0) {
2593                 goto out_free_path;
2594         } else if (ret > 0) {
2595                 ret = 0;
2596                 goto out_free_path;
2597         }
2598
2599         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2600                                 struct btrfs_file_extent_item);
2601
2602         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2603             backref->generation)
2604                 goto out_free_path;
2605
2606         btrfs_release_path(path);
2607
2608         start = backref->file_pos;
2609         if (backref->extent_offset < old->extent_offset + old->offset)
2610                 start += old->extent_offset + old->offset -
2611                          backref->extent_offset;
2612
2613         len = min(backref->extent_offset + backref->num_bytes,
2614                   old->extent_offset + old->offset + old->len);
2615         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2616
2617         ret = btrfs_drop_extents(trans, root, inode, start,
2618                                  start + len, 1);
2619         if (ret)
2620                 goto out_free_path;
2621 again:
2622         key.objectid = btrfs_ino(BTRFS_I(inode));
2623         key.type = BTRFS_EXTENT_DATA_KEY;
2624         key.offset = start;
2625
2626         path->leave_spinning = 1;
2627         if (merge) {
2628                 struct btrfs_file_extent_item *fi;
2629                 u64 extent_len;
2630                 struct btrfs_key found_key;
2631
2632                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2633                 if (ret < 0)
2634                         goto out_free_path;
2635
2636                 path->slots[0]--;
2637                 leaf = path->nodes[0];
2638                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2639
2640                 fi = btrfs_item_ptr(leaf, path->slots[0],
2641                                     struct btrfs_file_extent_item);
2642                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2643
2644                 if (extent_len + found_key.offset == start &&
2645                     relink_is_mergable(leaf, fi, new)) {
2646                         btrfs_set_file_extent_num_bytes(leaf, fi,
2647                                                         extent_len + len);
2648                         btrfs_mark_buffer_dirty(leaf);
2649                         inode_add_bytes(inode, len);
2650
2651                         ret = 1;
2652                         goto out_free_path;
2653                 } else {
2654                         merge = false;
2655                         btrfs_release_path(path);
2656                         goto again;
2657                 }
2658         }
2659
2660         ret = btrfs_insert_empty_item(trans, root, path, &key,
2661                                         sizeof(*extent));
2662         if (ret) {
2663                 btrfs_abort_transaction(trans, ret);
2664                 goto out_free_path;
2665         }
2666
2667         leaf = path->nodes[0];
2668         item = btrfs_item_ptr(leaf, path->slots[0],
2669                                 struct btrfs_file_extent_item);
2670         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2671         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2672         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2673         btrfs_set_file_extent_num_bytes(leaf, item, len);
2674         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2675         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2676         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2677         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2678         btrfs_set_file_extent_encryption(leaf, item, 0);
2679         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2680
2681         btrfs_mark_buffer_dirty(leaf);
2682         inode_add_bytes(inode, len);
2683         btrfs_release_path(path);
2684
2685         ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2686                         new->disk_len, 0,
2687                         backref->root_id, backref->inum,
2688                         new->file_pos); /* start - extent_offset */
2689         if (ret) {
2690                 btrfs_abort_transaction(trans, ret);
2691                 goto out_free_path;
2692         }
2693
2694         ret = 1;
2695 out_free_path:
2696         btrfs_release_path(path);
2697         path->leave_spinning = 0;
2698         btrfs_end_transaction(trans);
2699 out_unlock:
2700         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2701                              &cached);
2702         iput(inode);
2703         return ret;
2704 }
2705
2706 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2707 {
2708         struct old_sa_defrag_extent *old, *tmp;
2709
2710         if (!new)
2711                 return;
2712
2713         list_for_each_entry_safe(old, tmp, &new->head, list) {
2714                 kfree(old);
2715         }
2716         kfree(new);
2717 }
2718
2719 static void relink_file_extents(struct new_sa_defrag_extent *new)
2720 {
2721         struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2722         struct btrfs_path *path;
2723         struct sa_defrag_extent_backref *backref;
2724         struct sa_defrag_extent_backref *prev = NULL;
2725         struct rb_node *node;
2726         int ret;
2727
2728         path = btrfs_alloc_path();
2729         if (!path)
2730                 return;
2731
2732         if (!record_extent_backrefs(path, new)) {
2733                 btrfs_free_path(path);
2734                 goto out;
2735         }
2736         btrfs_release_path(path);
2737
2738         while (1) {
2739                 node = rb_first(&new->root);
2740                 if (!node)
2741                         break;
2742                 rb_erase(node, &new->root);
2743
2744                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2745
2746                 ret = relink_extent_backref(path, prev, backref);
2747                 WARN_ON(ret < 0);
2748
2749                 kfree(prev);
2750
2751                 if (ret == 1)
2752                         prev = backref;
2753                 else
2754                         prev = NULL;
2755                 cond_resched();
2756         }
2757         kfree(prev);
2758
2759         btrfs_free_path(path);
2760 out:
2761         free_sa_defrag_extent(new);
2762
2763         atomic_dec(&fs_info->defrag_running);
2764         wake_up(&fs_info->transaction_wait);
2765 }
2766
2767 static struct new_sa_defrag_extent *
2768 record_old_file_extents(struct inode *inode,
2769                         struct btrfs_ordered_extent *ordered)
2770 {
2771         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2772         struct btrfs_root *root = BTRFS_I(inode)->root;
2773         struct btrfs_path *path;
2774         struct btrfs_key key;
2775         struct old_sa_defrag_extent *old;
2776         struct new_sa_defrag_extent *new;
2777         int ret;
2778
2779         new = kmalloc(sizeof(*new), GFP_NOFS);
2780         if (!new)
2781                 return NULL;
2782
2783         new->inode = inode;
2784         new->file_pos = ordered->file_offset;
2785         new->len = ordered->len;
2786         new->bytenr = ordered->start;
2787         new->disk_len = ordered->disk_len;
2788         new->compress_type = ordered->compress_type;
2789         new->root = RB_ROOT;
2790         INIT_LIST_HEAD(&new->head);
2791
2792         path = btrfs_alloc_path();
2793         if (!path)
2794                 goto out_kfree;
2795
2796         key.objectid = btrfs_ino(BTRFS_I(inode));
2797         key.type = BTRFS_EXTENT_DATA_KEY;
2798         key.offset = new->file_pos;
2799
2800         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2801         if (ret < 0)
2802                 goto out_free_path;
2803         if (ret > 0 && path->slots[0] > 0)
2804                 path->slots[0]--;
2805
2806         /* find out all the old extents for the file range */
2807         while (1) {
2808                 struct btrfs_file_extent_item *extent;
2809                 struct extent_buffer *l;
2810                 int slot;
2811                 u64 num_bytes;
2812                 u64 offset;
2813                 u64 end;
2814                 u64 disk_bytenr;
2815                 u64 extent_offset;
2816
2817                 l = path->nodes[0];
2818                 slot = path->slots[0];
2819
2820                 if (slot >= btrfs_header_nritems(l)) {
2821                         ret = btrfs_next_leaf(root, path);
2822                         if (ret < 0)
2823                                 goto out_free_path;
2824                         else if (ret > 0)
2825                                 break;
2826                         continue;
2827                 }
2828
2829                 btrfs_item_key_to_cpu(l, &key, slot);
2830
2831                 if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2832                         break;
2833                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2834                         break;
2835                 if (key.offset >= new->file_pos + new->len)
2836                         break;
2837
2838                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2839
2840                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2841                 if (key.offset + num_bytes < new->file_pos)
2842                         goto next;
2843
2844                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2845                 if (!disk_bytenr)
2846                         goto next;
2847
2848                 extent_offset = btrfs_file_extent_offset(l, extent);
2849
2850                 old = kmalloc(sizeof(*old), GFP_NOFS);
2851                 if (!old)
2852                         goto out_free_path;
2853
2854                 offset = max(new->file_pos, key.offset);
2855                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2856
2857                 old->bytenr = disk_bytenr;
2858                 old->extent_offset = extent_offset;
2859                 old->offset = offset - key.offset;
2860                 old->len = end - offset;
2861                 old->new = new;
2862                 old->count = 0;
2863                 list_add_tail(&old->list, &new->head);
2864 next:
2865                 path->slots[0]++;
2866                 cond_resched();
2867         }
2868
2869         btrfs_free_path(path);
2870         atomic_inc(&fs_info->defrag_running);
2871
2872         return new;
2873
2874 out_free_path:
2875         btrfs_free_path(path);
2876 out_kfree:
2877         free_sa_defrag_extent(new);
2878         return NULL;
2879 }
2880
2881 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2882                                          u64 start, u64 len)
2883 {
2884         struct btrfs_block_group_cache *cache;
2885
2886         cache = btrfs_lookup_block_group(fs_info, start);
2887         ASSERT(cache);
2888
2889         spin_lock(&cache->lock);
2890         cache->delalloc_bytes -= len;
2891         spin_unlock(&cache->lock);
2892
2893         btrfs_put_block_group(cache);
2894 }
2895
2896 /* as ordered data IO finishes, this gets called so we can finish
2897  * an ordered extent if the range of bytes in the file it covers are
2898  * fully written.
2899  */
2900 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2901 {
2902         struct inode *inode = ordered_extent->inode;
2903         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2904         struct btrfs_root *root = BTRFS_I(inode)->root;
2905         struct btrfs_trans_handle *trans = NULL;
2906         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2907         struct extent_state *cached_state = NULL;
2908         struct new_sa_defrag_extent *new = NULL;
2909         int compress_type = 0;
2910         int ret = 0;
2911         u64 logical_len = ordered_extent->len;
2912         bool nolock;
2913         bool truncated = false;
2914         bool range_locked = false;
2915         bool clear_new_delalloc_bytes = false;
2916         bool clear_reserved_extent = true;
2917
2918         if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2919             !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2920             !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2921                 clear_new_delalloc_bytes = true;
2922
2923         nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2924
2925         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2926                 ret = -EIO;
2927                 goto out;
2928         }
2929
2930         btrfs_free_io_failure_record(BTRFS_I(inode),
2931                         ordered_extent->file_offset,
2932                         ordered_extent->file_offset +
2933                         ordered_extent->len - 1);
2934
2935         if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2936                 truncated = true;
2937                 logical_len = ordered_extent->truncated_len;
2938                 /* Truncated the entire extent, don't bother adding */
2939                 if (!logical_len)
2940                         goto out;
2941         }
2942
2943         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2944                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2945
2946                 /*
2947                  * For mwrite(mmap + memset to write) case, we still reserve
2948                  * space for NOCOW range.
2949                  * As NOCOW won't cause a new delayed ref, just free the space
2950                  */
2951                 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
2952                                        ordered_extent->len);
2953                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2954                 if (nolock)
2955                         trans = btrfs_join_transaction_nolock(root);
2956                 else
2957                         trans = btrfs_join_transaction(root);
2958                 if (IS_ERR(trans)) {
2959                         ret = PTR_ERR(trans);
2960                         trans = NULL;
2961                         goto out;
2962                 }
2963                 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
2964                 ret = btrfs_update_inode_fallback(trans, root, inode);
2965                 if (ret) /* -ENOMEM or corruption */
2966                         btrfs_abort_transaction(trans, ret);
2967                 goto out;
2968         }
2969
2970         range_locked = true;
2971         lock_extent_bits(io_tree, ordered_extent->file_offset,
2972                          ordered_extent->file_offset + ordered_extent->len - 1,
2973                          &cached_state);
2974
2975         ret = test_range_bit(io_tree, ordered_extent->file_offset,
2976                         ordered_extent->file_offset + ordered_extent->len - 1,
2977                         EXTENT_DEFRAG, 0, cached_state);
2978         if (ret) {
2979                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2980                 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2981                         /* the inode is shared */
2982                         new = record_old_file_extents(inode, ordered_extent);
2983
2984                 clear_extent_bit(io_tree, ordered_extent->file_offset,
2985                         ordered_extent->file_offset + ordered_extent->len - 1,
2986                         EXTENT_DEFRAG, 0, 0, &cached_state);
2987         }
2988
2989         if (nolock)
2990                 trans = btrfs_join_transaction_nolock(root);
2991         else
2992                 trans = btrfs_join_transaction(root);
2993         if (IS_ERR(trans)) {
2994                 ret = PTR_ERR(trans);
2995                 trans = NULL;
2996                 goto out;
2997         }
2998
2999         trans->block_rsv = &BTRFS_I(inode)->block_rsv;
3000
3001         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3002                 compress_type = ordered_extent->compress_type;
3003         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3004                 BUG_ON(compress_type);
3005                 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
3006                                        ordered_extent->len);
3007                 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
3008                                                 ordered_extent->file_offset,
3009                                                 ordered_extent->file_offset +
3010                                                 logical_len);
3011         } else {
3012                 BUG_ON(root == fs_info->tree_root);
3013                 ret = insert_reserved_file_extent(trans, inode,
3014                                                 ordered_extent->file_offset,
3015                                                 ordered_extent->start,
3016                                                 ordered_extent->disk_len,
3017                                                 logical_len, logical_len,
3018                                                 compress_type, 0, 0,
3019                                                 BTRFS_FILE_EXTENT_REG);
3020                 if (!ret) {
3021                         clear_reserved_extent = false;
3022                         btrfs_release_delalloc_bytes(fs_info,
3023                                                      ordered_extent->start,
3024                                                      ordered_extent->disk_len);
3025                 }
3026         }
3027         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
3028                            ordered_extent->file_offset, ordered_extent->len,
3029                            trans->transid);
3030         if (ret < 0) {
3031                 btrfs_abort_transaction(trans, ret);
3032                 goto out;
3033         }
3034
3035         ret = add_pending_csums(trans, inode, &ordered_extent->list);
3036         if (ret) {
3037                 btrfs_abort_transaction(trans, ret);
3038                 goto out;
3039         }
3040
3041         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3042         ret = btrfs_update_inode_fallback(trans, root, inode);
3043         if (ret) { /* -ENOMEM or corruption */
3044                 btrfs_abort_transaction(trans, ret);
3045                 goto out;
3046         }
3047         ret = 0;
3048 out:
3049         if (range_locked || clear_new_delalloc_bytes) {
3050                 unsigned int clear_bits = 0;
3051
3052                 if (range_locked)
3053                         clear_bits |= EXTENT_LOCKED;
3054                 if (clear_new_delalloc_bytes)
3055                         clear_bits |= EXTENT_DELALLOC_NEW;
3056                 clear_extent_bit(&BTRFS_I(inode)->io_tree,
3057                                  ordered_extent->file_offset,
3058                                  ordered_extent->file_offset +
3059                                  ordered_extent->len - 1,
3060                                  clear_bits,
3061                                  (clear_bits & EXTENT_LOCKED) ? 1 : 0,
3062                                  0, &cached_state);
3063         }
3064
3065         if (trans)
3066                 btrfs_end_transaction(trans);
3067
3068         if (ret || truncated) {
3069                 u64 start, end;
3070
3071                 if (truncated)
3072                         start = ordered_extent->file_offset + logical_len;
3073                 else
3074                         start = ordered_extent->file_offset;
3075                 end = ordered_extent->file_offset + ordered_extent->len - 1;
3076                 clear_extent_uptodate(io_tree, start, end, NULL);
3077
3078                 /* Drop the cache for the part of the extent we didn't write. */
3079                 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3080
3081                 /*
3082                  * If the ordered extent had an IOERR or something else went
3083                  * wrong we need to return the space for this ordered extent
3084                  * back to the allocator.  We only free the extent in the
3085                  * truncated case if we didn't write out the extent at all.
3086                  *
3087                  * If we made it past insert_reserved_file_extent before we
3088                  * errored out then we don't need to do this as the accounting
3089                  * has already been done.
3090                  */
3091                 if ((ret || !logical_len) &&
3092                     clear_reserved_extent &&
3093                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3094                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3095                         btrfs_free_reserved_extent(fs_info,
3096                                                    ordered_extent->start,
3097                                                    ordered_extent->disk_len, 1);
3098         }
3099
3100
3101         /*
3102          * This needs to be done to make sure anybody waiting knows we are done
3103          * updating everything for this ordered extent.
3104          */
3105         btrfs_remove_ordered_extent(inode, ordered_extent);
3106
3107         /* for snapshot-aware defrag */
3108         if (new) {
3109                 if (ret) {
3110                         free_sa_defrag_extent(new);
3111                         atomic_dec(&fs_info->defrag_running);
3112                 } else {
3113                         relink_file_extents(new);
3114                 }
3115         }
3116
3117         /* once for us */
3118         btrfs_put_ordered_extent(ordered_extent);
3119         /* once for the tree */
3120         btrfs_put_ordered_extent(ordered_extent);
3121
3122         /* Try to release some metadata so we don't get an OOM but don't wait */
3123         btrfs_btree_balance_dirty_nodelay(fs_info);
3124
3125         return ret;
3126 }
3127
3128 static void finish_ordered_fn(struct btrfs_work *work)
3129 {
3130         struct btrfs_ordered_extent *ordered_extent;
3131         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3132         btrfs_finish_ordered_io(ordered_extent);
3133 }
3134
3135 void btrfs_writepage_endio_finish_ordered(struct page *page, u64 start,
3136                                           u64 end, int uptodate)
3137 {
3138         struct inode *inode = page->mapping->host;
3139         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3140         struct btrfs_ordered_extent *ordered_extent = NULL;
3141         struct btrfs_workqueue *wq;
3142         btrfs_work_func_t func;
3143
3144         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3145
3146         ClearPagePrivate2(page);
3147         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3148                                             end - start + 1, uptodate))
3149                 return;
3150
3151         if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
3152                 wq = fs_info->endio_freespace_worker;
3153                 func = btrfs_freespace_write_helper;
3154         } else {
3155                 wq = fs_info->endio_write_workers;
3156                 func = btrfs_endio_write_helper;
3157         }
3158
3159         btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3160                         NULL);
3161         btrfs_queue_work(wq, &ordered_extent->work);
3162 }
3163
3164 static int __readpage_endio_check(struct inode *inode,
3165                                   struct btrfs_io_bio *io_bio,
3166                                   int icsum, struct page *page,
3167                                   int pgoff, u64 start, size_t len)
3168 {
3169         char *kaddr;
3170         u32 csum_expected;
3171         u32 csum = ~(u32)0;
3172
3173         csum_expected = *(((u32 *)io_bio->csum) + icsum);
3174
3175         kaddr = kmap_atomic(page);
3176         csum = btrfs_csum_data(kaddr + pgoff, csum,  len);
3177         btrfs_csum_final(csum, (u8 *)&csum);
3178         if (csum != csum_expected)
3179                 goto zeroit;
3180
3181         kunmap_atomic(kaddr);
3182         return 0;
3183 zeroit:
3184         btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3185                                     io_bio->mirror_num);
3186         memset(kaddr + pgoff, 1, len);
3187         flush_dcache_page(page);
3188         kunmap_atomic(kaddr);
3189         return -EIO;
3190 }
3191
3192 /*
3193  * when reads are done, we need to check csums to verify the data is correct
3194  * if there's a match, we allow the bio to finish.  If not, the code in
3195  * extent_io.c will try to find good copies for us.
3196  */
3197 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3198                                       u64 phy_offset, struct page *page,
3199                                       u64 start, u64 end, int mirror)
3200 {
3201         size_t offset = start - page_offset(page);
3202         struct inode *inode = page->mapping->host;
3203         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3204         struct btrfs_root *root = BTRFS_I(inode)->root;
3205
3206         if (PageChecked(page)) {
3207                 ClearPageChecked(page);
3208                 return 0;
3209         }
3210
3211         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3212                 return 0;
3213
3214         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3215             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3216                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3217                 return 0;
3218         }
3219
3220         phy_offset >>= inode->i_sb->s_blocksize_bits;
3221         return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3222                                       start, (size_t)(end - start + 1));
3223 }
3224
3225 /*
3226  * btrfs_add_delayed_iput - perform a delayed iput on @inode
3227  *
3228  * @inode: The inode we want to perform iput on
3229  *
3230  * This function uses the generic vfs_inode::i_count to track whether we should
3231  * just decrement it (in case it's > 1) or if this is the last iput then link
3232  * the inode to the delayed iput machinery. Delayed iputs are processed at
3233  * transaction commit time/superblock commit/cleaner kthread.
3234  */
3235 void btrfs_add_delayed_iput(struct inode *inode)
3236 {
3237         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3238         struct btrfs_inode *binode = BTRFS_I(inode);
3239
3240         if (atomic_add_unless(&inode->i_count, -1, 1))
3241                 return;
3242
3243         spin_lock(&fs_info->delayed_iput_lock);
3244         ASSERT(list_empty(&binode->delayed_iput));
3245         list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3246         spin_unlock(&fs_info->delayed_iput_lock);
3247 }
3248
3249 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3250 {
3251
3252         spin_lock(&fs_info->delayed_iput_lock);
3253         while (!list_empty(&fs_info->delayed_iputs)) {
3254                 struct btrfs_inode *inode;
3255
3256                 inode = list_first_entry(&fs_info->delayed_iputs,
3257                                 struct btrfs_inode, delayed_iput);
3258                 list_del_init(&inode->delayed_iput);
3259                 spin_unlock(&fs_info->delayed_iput_lock);
3260                 iput(&inode->vfs_inode);
3261                 spin_lock(&fs_info->delayed_iput_lock);
3262         }
3263         spin_unlock(&fs_info->delayed_iput_lock);
3264 }
3265
3266 /*
3267  * This creates an orphan entry for the given inode in case something goes wrong
3268  * in the middle of an unlink.
3269  */
3270 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3271                      struct btrfs_inode *inode)
3272 {
3273         int ret;
3274
3275         ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3276         if (ret && ret != -EEXIST) {
3277                 btrfs_abort_transaction(trans, ret);
3278                 return ret;
3279         }
3280
3281         return 0;
3282 }
3283
3284 /*
3285  * We have done the delete so we can go ahead and remove the orphan item for
3286  * this particular inode.
3287  */
3288 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3289                             struct btrfs_inode *inode)
3290 {
3291         return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3292 }
3293
3294 /*
3295  * this cleans up any orphans that may be left on the list from the last use
3296  * of this root.
3297  */
3298 int btrfs_orphan_cleanup(struct btrfs_root *root)
3299 {
3300         struct btrfs_fs_info *fs_info = root->fs_info;
3301         struct btrfs_path *path;
3302         struct extent_buffer *leaf;
3303         struct btrfs_key key, found_key;
3304         struct btrfs_trans_handle *trans;
3305         struct inode *inode;
3306         u64 last_objectid = 0;
3307         int ret = 0, nr_unlink = 0;
3308
3309         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3310                 return 0;
3311
3312         path = btrfs_alloc_path();
3313         if (!path) {
3314                 ret = -ENOMEM;
3315                 goto out;
3316         }
3317         path->reada = READA_BACK;
3318
3319         key.objectid = BTRFS_ORPHAN_OBJECTID;
3320         key.type = BTRFS_ORPHAN_ITEM_KEY;
3321         key.offset = (u64)-1;
3322
3323         while (1) {
3324                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3325                 if (ret < 0)
3326                         goto out;
3327
3328                 /*
3329                  * if ret == 0 means we found what we were searching for, which
3330                  * is weird, but possible, so only screw with path if we didn't
3331                  * find the key and see if we have stuff that matches
3332                  */
3333                 if (ret > 0) {
3334                         ret = 0;
3335                         if (path->slots[0] == 0)
3336                                 break;
3337                         path->slots[0]--;
3338                 }
3339
3340                 /* pull out the item */
3341                 leaf = path->nodes[0];
3342                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3343
3344                 /* make sure the item matches what we want */
3345                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3346                         break;
3347                 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3348                         break;
3349
3350                 /* release the path since we're done with it */
3351                 btrfs_release_path(path);
3352
3353                 /*
3354                  * this is where we are basically btrfs_lookup, without the
3355                  * crossing root thing.  we store the inode number in the
3356                  * offset of the orphan item.
3357                  */
3358
3359                 if (found_key.offset == last_objectid) {
3360                         btrfs_err(fs_info,
3361                                   "Error removing orphan entry, stopping orphan cleanup");
3362                         ret = -EINVAL;
3363                         goto out;
3364                 }
3365
3366                 last_objectid = found_key.offset;
3367
3368                 found_key.objectid = found_key.offset;
3369                 found_key.type = BTRFS_INODE_ITEM_KEY;
3370                 found_key.offset = 0;
3371                 inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3372                 ret = PTR_ERR_OR_ZERO(inode);
3373                 if (ret && ret != -ENOENT)
3374                         goto out;
3375
3376                 if (ret == -ENOENT && root == fs_info->tree_root) {
3377                         struct btrfs_root *dead_root;
3378                         struct btrfs_fs_info *fs_info = root->fs_info;
3379                         int is_dead_root = 0;
3380
3381                         /*
3382                          * this is an orphan in the tree root. Currently these
3383                          * could come from 2 sources:
3384                          *  a) a snapshot deletion in progress
3385                          *  b) a free space cache inode
3386                          * We need to distinguish those two, as the snapshot
3387                          * orphan must not get deleted.
3388                          * find_dead_roots already ran before us, so if this
3389                          * is a snapshot deletion, we should find the root
3390                          * in the dead_roots list
3391                          */
3392                         spin_lock(&fs_info->trans_lock);
3393                         list_for_each_entry(dead_root, &fs_info->dead_roots,
3394                                             root_list) {
3395                                 if (dead_root->root_key.objectid ==
3396                                     found_key.objectid) {
3397                                         is_dead_root = 1;
3398                                         break;
3399                                 }
3400                         }
3401                         spin_unlock(&fs_info->trans_lock);
3402                         if (is_dead_root) {
3403                                 /* prevent this orphan from being found again */
3404                                 key.offset = found_key.objectid - 1;
3405                                 continue;
3406                         }
3407
3408                 }
3409
3410                 /*
3411                  * If we have an inode with links, there are a couple of
3412                  * possibilities. Old kernels (before v3.12) used to create an
3413                  * orphan item for truncate indicating that there were possibly
3414                  * extent items past i_size that needed to be deleted. In v3.12,
3415                  * truncate was changed to update i_size in sync with the extent
3416                  * items, but the (useless) orphan item was still created. Since
3417                  * v4.18, we don't create the orphan item for truncate at all.
3418                  *
3419                  * So, this item could mean that we need to do a truncate, but
3420                  * only if this filesystem was last used on a pre-v3.12 kernel
3421                  * and was not cleanly unmounted. The odds of that are quite
3422                  * slim, and it's a pain to do the truncate now, so just delete
3423                  * the orphan item.
3424                  *
3425                  * It's also possible that this orphan item was supposed to be
3426                  * deleted but wasn't. The inode number may have been reused,
3427                  * but either way, we can delete the orphan item.
3428                  */
3429                 if (ret == -ENOENT || inode->i_nlink) {
3430                         if (!ret)
3431                                 iput(inode);
3432                         trans = btrfs_start_transaction(root, 1);
3433                         if (IS_ERR(trans)) {
3434                                 ret = PTR_ERR(trans);
3435                                 goto out;
3436                         }
3437                         btrfs_debug(fs_info, "auto deleting %Lu",
3438                                     found_key.objectid);
3439                         ret = btrfs_del_orphan_item(trans, root,
3440                                                     found_key.objectid);
3441                         btrfs_end_transaction(trans);
3442                         if (ret)
3443                                 goto out;
3444                         continue;
3445                 }
3446
3447                 nr_unlink++;
3448
3449                 /* this will do delete_inode and everything for us */
3450                 iput(inode);
3451         }
3452         /* release the path since we're done with it */
3453         btrfs_release_path(path);
3454
3455         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3456
3457         if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3458                 trans = btrfs_join_transaction(root);
3459                 if (!IS_ERR(trans))
3460                         btrfs_end_transaction(trans);
3461         }
3462
3463         if (nr_unlink)
3464                 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3465
3466 out:
3467         if (ret)
3468                 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3469         btrfs_free_path(path);
3470         return ret;
3471 }
3472
3473 /*
3474  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3475  * don't find any xattrs, we know there can't be any acls.
3476  *
3477  * slot is the slot the inode is in, objectid is the objectid of the inode
3478  */
3479 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3480                                           int slot, u64 objectid,
3481                                           int *first_xattr_slot)
3482 {
3483         u32 nritems = btrfs_header_nritems(leaf);
3484         struct btrfs_key found_key;
3485         static u64 xattr_access = 0;
3486         static u64 xattr_default = 0;
3487         int scanned = 0;
3488
3489         if (!xattr_access) {
3490                 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3491                                         strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3492                 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3493                                         strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3494         }
3495
3496         slot++;
3497         *first_xattr_slot = -1;
3498         while (slot < nritems) {
3499                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3500
3501                 /* we found a different objectid, there must not be acls */
3502                 if (found_key.objectid != objectid)
3503                         return 0;
3504
3505                 /* we found an xattr, assume we've got an acl */
3506                 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3507                         if (*first_xattr_slot == -1)
3508                                 *first_xattr_slot = slot;
3509                         if (found_key.offset == xattr_access ||
3510                             found_key.offset == xattr_default)
3511                                 return 1;
3512                 }
3513
3514                 /*
3515                  * we found a key greater than an xattr key, there can't
3516                  * be any acls later on
3517                  */
3518                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3519                         return 0;
3520
3521                 slot++;
3522                 scanned++;
3523
3524                 /*
3525                  * it goes inode, inode backrefs, xattrs, extents,
3526                  * so if there are a ton of hard links to an inode there can
3527                  * be a lot of backrefs.  Don't waste time searching too hard,
3528                  * this is just an optimization
3529                  */
3530                 if (scanned >= 8)
3531                         break;
3532         }
3533         /* we hit the end of the leaf before we found an xattr or
3534          * something larger than an xattr.  We have to assume the inode
3535          * has acls
3536          */
3537         if (*first_xattr_slot == -1)
3538                 *first_xattr_slot = slot;
3539         return 1;
3540 }
3541
3542 /*
3543  * read an inode from the btree into the in-memory inode
3544  */
3545 static int btrfs_read_locked_inode(struct inode *inode,
3546                                    struct btrfs_path *in_path)
3547 {
3548         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3549         struct btrfs_path *path = in_path;
3550         struct extent_buffer *leaf;
3551         struct btrfs_inode_item *inode_item;
3552         struct btrfs_root *root = BTRFS_I(inode)->root;
3553         struct btrfs_key location;
3554         unsigned long ptr;
3555         int maybe_acls;
3556         u32 rdev;
3557         int ret;
3558         bool filled = false;
3559         int first_xattr_slot;
3560
3561         ret = btrfs_fill_inode(inode, &rdev);
3562         if (!ret)
3563                 filled = true;
3564
3565         if (!path) {
3566                 path = btrfs_alloc_path();
3567                 if (!path)
3568                         return -ENOMEM;
3569         }
3570
3571         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3572
3573         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3574         if (ret) {
3575                 if (path != in_path)
3576                         btrfs_free_path(path);
3577                 return ret;
3578         }
3579
3580         leaf = path->nodes[0];
3581
3582         if (filled)
3583                 goto cache_index;
3584
3585         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3586                                     struct btrfs_inode_item);
3587         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3588         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3589         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3590         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3591         btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3592
3593         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3594         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3595
3596         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3597         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3598
3599         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3600         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3601
3602         BTRFS_I(inode)->i_otime.tv_sec =
3603                 btrfs_timespec_sec(leaf, &inode_item->otime);
3604         BTRFS_I(inode)->i_otime.tv_nsec =
3605                 btrfs_timespec_nsec(leaf, &inode_item->otime);
3606
3607         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3608         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3609         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3610
3611         inode_set_iversion_queried(inode,
3612                                    btrfs_inode_sequence(leaf, inode_item));
3613         inode->i_generation = BTRFS_I(inode)->generation;
3614         inode->i_rdev = 0;
3615         rdev = btrfs_inode_rdev(leaf, inode_item);
3616
3617         BTRFS_I(inode)->index_cnt = (u64)-1;
3618         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3619
3620 cache_index:
3621         /*
3622          * If we were modified in the current generation and evicted from memory
3623          * and then re-read we need to do a full sync since we don't have any
3624          * idea about which extents were modified before we were evicted from
3625          * cache.
3626          *
3627          * This is required for both inode re-read from disk and delayed inode
3628          * in delayed_nodes_tree.
3629          */
3630         if (BTRFS_I(inode)->last_trans == fs_info->generation)
3631                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3632                         &BTRFS_I(inode)->runtime_flags);
3633
3634         /*
3635          * We don't persist the id of the transaction where an unlink operation
3636          * against the inode was last made. So here we assume the inode might
3637          * have been evicted, and therefore the exact value of last_unlink_trans
3638          * lost, and set it to last_trans to avoid metadata inconsistencies
3639          * between the inode and its parent if the inode is fsync'ed and the log
3640          * replayed. For example, in the scenario:
3641          *
3642          * touch mydir/foo
3643          * ln mydir/foo mydir/bar
3644          * sync
3645          * unlink mydir/bar
3646          * echo 2 > /proc/sys/vm/drop_caches   # evicts inode
3647          * xfs_io -c fsync mydir/foo
3648          * <power failure>
3649          * mount fs, triggers fsync log replay
3650          *
3651          * We must make sure that when we fsync our inode foo we also log its
3652          * parent inode, otherwise after log replay the parent still has the
3653          * dentry with the "bar" name but our inode foo has a link count of 1
3654          * and doesn't have an inode ref with the name "bar" anymore.
3655          *
3656          * Setting last_unlink_trans to last_trans is a pessimistic approach,
3657          * but it guarantees correctness at the expense of occasional full
3658          * transaction commits on fsync if our inode is a directory, or if our
3659          * inode is not a directory, logging its parent unnecessarily.
3660          */
3661         BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3662         /*
3663          * Similar reasoning for last_link_trans, needs to be set otherwise
3664          * for a case like the following:
3665          *
3666          * mkdir A
3667          * touch foo
3668          * ln foo A/bar
3669          * echo 2 > /proc/sys/vm/drop_caches
3670          * fsync foo
3671          * <power failure>
3672          *
3673          * Would result in link bar and directory A not existing after the power
3674          * failure.
3675          */
3676         BTRFS_I(inode)->last_link_trans = BTRFS_I(inode)->last_trans;
3677
3678         path->slots[0]++;
3679         if (inode->i_nlink != 1 ||
3680             path->slots[0] >= btrfs_header_nritems(leaf))
3681                 goto cache_acl;
3682
3683         btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3684         if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3685                 goto cache_acl;
3686
3687         ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3688         if (location.type == BTRFS_INODE_REF_KEY) {
3689                 struct btrfs_inode_ref *ref;
3690
3691                 ref = (struct btrfs_inode_ref *)ptr;
3692                 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3693         } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3694                 struct btrfs_inode_extref *extref;
3695
3696                 extref = (struct btrfs_inode_extref *)ptr;
3697                 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3698                                                                      extref);
3699         }
3700 cache_acl:
3701         /*
3702          * try to precache a NULL acl entry for files that don't have
3703          * any xattrs or acls
3704          */
3705         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3706                         btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3707         if (first_xattr_slot != -1) {
3708                 path->slots[0] = first_xattr_slot;
3709                 ret = btrfs_load_inode_props(inode, path);
3710                 if (ret)
3711                         btrfs_err(fs_info,
3712                                   "error loading props for ino %llu (root %llu): %d",
3713                                   btrfs_ino(BTRFS_I(inode)),
3714                                   root->root_key.objectid, ret);
3715         }
3716         if (path != in_path)
3717                 btrfs_free_path(path);
3718
3719         if (!maybe_acls)
3720                 cache_no_acl(inode);
3721
3722         switch (inode->i_mode & S_IFMT) {
3723         case S_IFREG:
3724                 inode->i_mapping->a_ops = &btrfs_aops;
3725                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3726                 inode->i_fop = &btrfs_file_operations;
3727                 inode->i_op = &btrfs_file_inode_operations;
3728                 break;
3729         case S_IFDIR:
3730                 inode->i_fop = &btrfs_dir_file_operations;
3731                 inode->i_op = &btrfs_dir_inode_operations;
3732                 break;
3733         case S_IFLNK:
3734                 inode->i_op = &btrfs_symlink_inode_operations;
3735                 inode_nohighmem(inode);
3736                 inode->i_mapping->a_ops = &btrfs_aops;
3737                 break;
3738         default:
3739                 inode->i_op = &btrfs_special_inode_operations;
3740                 init_special_inode(inode, inode->i_mode, rdev);
3741                 break;
3742         }
3743
3744         btrfs_sync_inode_flags_to_i_flags(inode);
3745         return 0;
3746 }
3747
3748 /*
3749  * given a leaf and an inode, copy the inode fields into the leaf
3750  */
3751 static void fill_inode_item(struct btrfs_trans_handle *trans,
3752                             struct extent_buffer *leaf,
3753                             struct btrfs_inode_item *item,
3754                             struct inode *inode)
3755 {
3756         struct btrfs_map_token token;
3757
3758         btrfs_init_map_token(&token);
3759
3760         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3761         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3762         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3763                                    &token);
3764         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3765         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3766
3767         btrfs_set_token_timespec_sec(leaf, &item->atime,
3768                                      inode->i_atime.tv_sec, &token);
3769         btrfs_set_token_timespec_nsec(leaf, &item->atime,
3770                                       inode->i_atime.tv_nsec, &token);
3771
3772         btrfs_set_token_timespec_sec(leaf, &item->mtime,
3773                                      inode->i_mtime.tv_sec, &token);
3774         btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3775                                       inode->i_mtime.tv_nsec, &token);
3776
3777         btrfs_set_token_timespec_sec(leaf, &item->ctime,
3778                                      inode->i_ctime.tv_sec, &token);
3779         btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3780                                       inode->i_ctime.tv_nsec, &token);
3781
3782         btrfs_set_token_timespec_sec(leaf, &item->otime,
3783                                      BTRFS_I(inode)->i_otime.tv_sec, &token);
3784         btrfs_set_token_timespec_nsec(leaf, &item->otime,
3785                                       BTRFS_I(inode)->i_otime.tv_nsec, &token);
3786
3787         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3788                                      &token);
3789         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3790                                          &token);
3791         btrfs_set_token_inode_sequence(leaf, item, inode_peek_iversion(inode),
3792                                        &token);
3793         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3794         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3795         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3796         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3797 }
3798
3799 /*
3800  * copy everything in the in-memory inode into the btree.
3801  */
3802 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3803                                 struct btrfs_root *root, struct inode *inode)
3804 {
3805         struct btrfs_inode_item *inode_item;
3806         struct btrfs_path *path;
3807         struct extent_buffer *leaf;
3808         int ret;
3809
3810         path = btrfs_alloc_path();
3811         if (!path)
3812                 return -ENOMEM;
3813
3814         path->leave_spinning = 1;
3815         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3816                                  1);
3817         if (ret) {
3818                 if (ret > 0)
3819                         ret = -ENOENT;
3820                 goto failed;
3821         }
3822
3823         leaf = path->nodes[0];
3824         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3825                                     struct btrfs_inode_item);
3826
3827         fill_inode_item(trans, leaf, inode_item, inode);
3828         btrfs_mark_buffer_dirty(leaf);
3829         btrfs_set_inode_last_trans(trans, inode);
3830         ret = 0;
3831 failed:
3832         btrfs_free_path(path);
3833         return ret;
3834 }
3835
3836 /*
3837  * copy everything in the in-memory inode into the btree.
3838  */
3839 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3840                                 struct btrfs_root *root, struct inode *inode)
3841 {
3842         struct btrfs_fs_info *fs_info = root->fs_info;
3843         int ret;
3844
3845         /*
3846          * If the inode is a free space inode, we can deadlock during commit
3847          * if we put it into the delayed code.
3848          *
3849          * The data relocation inode should also be directly updated
3850          * without delay
3851          */
3852         if (!btrfs_is_free_space_inode(BTRFS_I(inode))
3853             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3854             && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3855                 btrfs_update_root_times(trans, root);
3856
3857                 ret = btrfs_delayed_update_inode(trans, root, inode);
3858                 if (!ret)
3859                         btrfs_set_inode_last_trans(trans, inode);
3860                 return ret;
3861         }
3862
3863         return btrfs_update_inode_item(trans, root, inode);
3864 }
3865
3866 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3867                                          struct btrfs_root *root,
3868                                          struct inode *inode)
3869 {
3870         int ret;
3871
3872         ret = btrfs_update_inode(trans, root, inode);
3873         if (ret == -ENOSPC)
3874                 return btrfs_update_inode_item(trans, root, inode);
3875         return ret;
3876 }
3877
3878 /*
3879  * unlink helper that gets used here in inode.c and in the tree logging
3880  * recovery code.  It remove a link in a directory with a given name, and
3881  * also drops the back refs in the inode to the directory
3882  */
3883 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3884                                 struct btrfs_root *root,
3885                                 struct btrfs_inode *dir,
3886                                 struct btrfs_inode *inode,
3887                                 const char *name, int name_len)
3888 {
3889         struct btrfs_fs_info *fs_info = root->fs_info;
3890         struct btrfs_path *path;
3891         int ret = 0;
3892         struct extent_buffer *leaf;
3893         struct btrfs_dir_item *di;
3894         struct btrfs_key key;
3895         u64 index;
3896         u64 ino = btrfs_ino(inode);
3897         u64 dir_ino = btrfs_ino(dir);
3898
3899         path = btrfs_alloc_path();
3900         if (!path) {
3901                 ret = -ENOMEM;
3902                 goto out;
3903         }
3904
3905         path->leave_spinning = 1;
3906         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3907                                     name, name_len, -1);
3908         if (IS_ERR_OR_NULL(di)) {
3909                 ret = di ? PTR_ERR(di) : -ENOENT;
3910                 goto err;
3911         }
3912         leaf = path->nodes[0];
3913         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3914         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3915         if (ret)
3916                 goto err;
3917         btrfs_release_path(path);
3918
3919         /*
3920          * If we don't have dir index, we have to get it by looking up
3921          * the inode ref, since we get the inode ref, remove it directly,
3922          * it is unnecessary to do delayed deletion.
3923          *
3924          * But if we have dir index, needn't search inode ref to get it.
3925          * Since the inode ref is close to the inode item, it is better
3926          * that we delay to delete it, and just do this deletion when
3927          * we update the inode item.
3928          */
3929         if (inode->dir_index) {
3930                 ret = btrfs_delayed_delete_inode_ref(inode);
3931                 if (!ret) {
3932                         index = inode->dir_index;
3933                         goto skip_backref;
3934                 }
3935         }
3936
3937         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3938                                   dir_ino, &index);
3939         if (ret) {
3940                 btrfs_info(fs_info,
3941                         "failed to delete reference to %.*s, inode %llu parent %llu",
3942                         name_len, name, ino, dir_ino);
3943                 btrfs_abort_transaction(trans, ret);
3944                 goto err;
3945         }
3946 skip_backref:
3947         ret = btrfs_delete_delayed_dir_index(trans, dir, index);
3948         if (ret) {
3949                 btrfs_abort_transaction(trans, ret);
3950                 goto err;
3951         }
3952
3953         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
3954                         dir_ino);
3955         if (ret != 0 && ret != -ENOENT) {
3956                 btrfs_abort_transaction(trans, ret);
3957                 goto err;
3958         }
3959
3960         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
3961                         index);
3962         if (ret == -ENOENT)
3963                 ret = 0;
3964         else if (ret)
3965                 btrfs_abort_transaction(trans, ret);
3966 err:
3967         btrfs_free_path(path);
3968         if (ret)
3969                 goto out;
3970
3971         btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
3972         inode_inc_iversion(&inode->vfs_inode);
3973         inode_inc_iversion(&dir->vfs_inode);
3974         inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
3975                 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
3976         ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
3977 out:
3978         return ret;
3979 }
3980
3981 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3982                        struct btrfs_root *root,
3983                        struct btrfs_inode *dir, struct btrfs_inode *inode,
3984                        const char *name, int name_len)
3985 {
3986         int ret;
3987         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3988         if (!ret) {
3989                 drop_nlink(&inode->vfs_inode);
3990                 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
3991         }
3992         return ret;
3993 }
3994
3995 /*
3996  * helper to start transaction for unlink and rmdir.
3997  *
3998  * unlink and rmdir are special in btrfs, they do not always free space, so
3999  * if we cannot make our reservations the normal way try and see if there is
4000  * plenty of slack room in the global reserve to migrate, otherwise we cannot
4001  * allow the unlink to occur.
4002  */
4003 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4004 {
4005         struct btrfs_root *root = BTRFS_I(dir)->root;
4006
4007         /*
4008          * 1 for the possible orphan item
4009          * 1 for the dir item
4010          * 1 for the dir index
4011          * 1 for the inode ref
4012          * 1 for the inode
4013          */
4014         return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4015 }
4016
4017 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4018 {
4019         struct btrfs_root *root = BTRFS_I(dir)->root;
4020         struct btrfs_trans_handle *trans;
4021         struct inode *inode = d_inode(dentry);
4022         int ret;
4023
4024         trans = __unlink_start_trans(dir);
4025         if (IS_ERR(trans))
4026                 return PTR_ERR(trans);
4027
4028         btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4029                         0);
4030
4031         ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4032                         BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4033                         dentry->d_name.len);
4034         if (ret)
4035                 goto out;
4036
4037         if (inode->i_nlink == 0) {
4038                 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4039                 if (ret)
4040                         goto out;
4041         }
4042
4043 out:
4044         btrfs_end_transaction(trans);
4045         btrfs_btree_balance_dirty(root->fs_info);
4046         return ret;
4047 }
4048
4049 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4050                                struct inode *dir, u64 objectid,
4051                                const char *name, int name_len)
4052 {
4053         struct btrfs_root *root = BTRFS_I(dir)->root;
4054         struct btrfs_path *path;
4055         struct extent_buffer *leaf;
4056         struct btrfs_dir_item *di;
4057         struct btrfs_key key;
4058         u64 index;
4059         int ret;
4060         u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4061
4062         path = btrfs_alloc_path();
4063         if (!path)
4064                 return -ENOMEM;
4065
4066         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4067                                    name, name_len, -1);
4068         if (IS_ERR_OR_NULL(di)) {
4069                 ret = di ? PTR_ERR(di) : -ENOENT;
4070                 goto out;
4071         }
4072
4073         leaf = path->nodes[0];
4074         btrfs_dir_item_key_to_cpu(leaf, di, &key);
4075         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4076         ret = btrfs_delete_one_dir_name(trans, root, path, di);
4077         if (ret) {
4078                 btrfs_abort_transaction(trans, ret);
4079                 goto out;
4080         }
4081         btrfs_release_path(path);
4082
4083         ret = btrfs_del_root_ref(trans, objectid, root->root_key.objectid,
4084                                  dir_ino, &index, name, name_len);
4085         if (ret < 0) {
4086                 if (ret != -ENOENT) {
4087                         btrfs_abort_transaction(trans, ret);
4088                         goto out;
4089                 }
4090                 di = btrfs_search_dir_index_item(root, path, dir_ino,
4091                                                  name, name_len);
4092                 if (IS_ERR_OR_NULL(di)) {
4093                         if (!di)
4094                                 ret = -ENOENT;
4095                         else
4096                                 ret = PTR_ERR(di);
4097                         btrfs_abort_transaction(trans, ret);
4098                         goto out;
4099                 }
4100
4101                 leaf = path->nodes[0];
4102                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4103                 index = key.offset;
4104         }
4105         btrfs_release_path(path);
4106
4107         ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
4108         if (ret) {
4109                 btrfs_abort_transaction(trans, ret);
4110                 goto out;
4111         }
4112
4113         btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4114         inode_inc_iversion(dir);
4115         dir->i_mtime = dir->i_ctime = current_time(dir);
4116         ret = btrfs_update_inode_fallback(trans, root, dir);
4117         if (ret)
4118                 btrfs_abort_transaction(trans, ret);
4119 out:
4120         btrfs_free_path(path);
4121         return ret;
4122 }
4123
4124 /*
4125  * Helper to check if the subvolume references other subvolumes or if it's
4126  * default.
4127  */
4128 static noinline int may_destroy_subvol(struct btrfs_root *root)
4129 {
4130         struct btrfs_fs_info *fs_info = root->fs_info;
4131         struct btrfs_path *path;
4132         struct btrfs_dir_item *di;
4133         struct btrfs_key key;
4134         u64 dir_id;
4135         int ret;
4136
4137         path = btrfs_alloc_path();
4138         if (!path)
4139                 return -ENOMEM;
4140
4141         /* Make sure this root isn't set as the default subvol */
4142         dir_id = btrfs_super_root_dir(fs_info->super_copy);
4143         di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4144                                    dir_id, "default", 7, 0);
4145         if (di && !IS_ERR(di)) {
4146                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4147                 if (key.objectid == root->root_key.objectid) {
4148                         ret = -EPERM;
4149                         btrfs_err(fs_info,
4150                                   "deleting default subvolume %llu is not allowed",
4151                                   key.objectid);
4152                         goto out;
4153                 }
4154                 btrfs_release_path(path);
4155         }
4156
4157         key.objectid = root->root_key.objectid;
4158         key.type = BTRFS_ROOT_REF_KEY;
4159         key.offset = (u64)-1;
4160
4161         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4162         if (ret < 0)
4163                 goto out;
4164         BUG_ON(ret == 0);
4165
4166         ret = 0;
4167         if (path->slots[0] > 0) {
4168                 path->slots[0]--;
4169                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4170                 if (key.objectid == root->root_key.objectid &&
4171                     key.type == BTRFS_ROOT_REF_KEY)
4172                         ret = -ENOTEMPTY;
4173         }
4174 out:
4175         btrfs_free_path(path);
4176         return ret;
4177 }
4178
4179 /* Delete all dentries for inodes belonging to the root */
4180 static void btrfs_prune_dentries(struct btrfs_root *root)
4181 {
4182         struct btrfs_fs_info *fs_info = root->fs_info;
4183         struct rb_node *node;
4184         struct rb_node *prev;
4185         struct btrfs_inode *entry;
4186         struct inode *inode;
4187         u64 objectid = 0;
4188
4189         if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
4190                 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4191
4192         spin_lock(&root->inode_lock);
4193 again:
4194         node = root->inode_tree.rb_node;
4195         prev = NULL;
4196         while (node) {
4197                 prev = node;
4198                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4199
4200                 if (objectid < btrfs_ino(entry))
4201                         node = node->rb_left;
4202                 else if (objectid > btrfs_ino(entry))
4203                         node = node->rb_right;
4204                 else
4205                         break;
4206         }
4207         if (!node) {
4208                 while (prev) {
4209                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4210                         if (objectid <= btrfs_ino(entry)) {
4211                                 node = prev;
4212                                 break;
4213                         }
4214                         prev = rb_next(prev);
4215                 }
4216         }
4217         while (node) {
4218                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4219                 objectid = btrfs_ino(entry) + 1;
4220                 inode = igrab(&entry->vfs_inode);
4221                 if (inode) {
4222                         spin_unlock(&root->inode_lock);
4223                         if (atomic_read(&inode->i_count) > 1)
4224                                 d_prune_aliases(inode);
4225                         /*
4226                          * btrfs_drop_inode will have it removed from the inode
4227                          * cache when its usage count hits zero.
4228                          */
4229                         iput(inode);
4230                         cond_resched();
4231                         spin_lock(&root->inode_lock);
4232                         goto again;
4233                 }
4234
4235                 if (cond_resched_lock(&root->inode_lock))
4236                         goto again;
4237
4238                 node = rb_next(node);
4239         }
4240         spin_unlock(&root->inode_lock);
4241 }
4242
4243 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4244 {
4245         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4246         struct btrfs_root *root = BTRFS_I(dir)->root;
4247         struct inode *inode = d_inode(dentry);
4248         struct btrfs_root *dest = BTRFS_I(inode)->root;
4249         struct btrfs_trans_handle *trans;
4250         struct btrfs_block_rsv block_rsv;
4251         u64 root_flags;
4252         int ret;
4253         int err;
4254
4255         /*
4256          * Don't allow to delete a subvolume with send in progress. This is
4257          * inside the inode lock so the error handling that has to drop the bit
4258          * again is not run concurrently.
4259          */
4260         spin_lock(&dest->root_item_lock);
4261         if (dest->send_in_progress) {
4262                 spin_unlock(&dest->root_item_lock);
4263                 btrfs_warn(fs_info,
4264                            "attempt to delete subvolume %llu during send",
4265                            dest->root_key.objectid);
4266                 return -EPERM;
4267         }
4268         root_flags = btrfs_root_flags(&dest->root_item);
4269         btrfs_set_root_flags(&dest->root_item,
4270                              root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4271         spin_unlock(&dest->root_item_lock);
4272
4273         down_write(&fs_info->subvol_sem);
4274
4275         err = may_destroy_subvol(dest);
4276         if (err)
4277                 goto out_up_write;
4278
4279         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4280         /*
4281          * One for dir inode,
4282          * two for dir entries,
4283          * two for root ref/backref.
4284          */
4285         err = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4286         if (err)
4287                 goto out_up_write;
4288
4289         trans = btrfs_start_transaction(root, 0);
4290         if (IS_ERR(trans)) {
4291                 err = PTR_ERR(trans);
4292                 goto out_release;
4293         }
4294         trans->block_rsv = &block_rsv;
4295         trans->bytes_reserved = block_rsv.size;
4296
4297         btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4298
4299         ret = btrfs_unlink_subvol(trans, dir, dest->root_key.objectid,
4300                                   dentry->d_name.name, dentry->d_name.len);
4301         if (ret) {
4302                 err = ret;
4303                 btrfs_abort_transaction(trans, ret);
4304                 goto out_end_trans;
4305         }
4306
4307         btrfs_record_root_in_trans(trans, dest);
4308
4309         memset(&dest->root_item.drop_progress, 0,
4310                 sizeof(dest->root_item.drop_progress));
4311         dest->root_item.drop_level = 0;
4312         btrfs_set_root_refs(&dest->root_item, 0);
4313
4314         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4315                 ret = btrfs_insert_orphan_item(trans,
4316                                         fs_info->tree_root,
4317                                         dest->root_key.objectid);
4318                 if (ret) {
4319                         btrfs_abort_transaction(trans, ret);
4320                         err = ret;
4321                         goto out_end_trans;
4322                 }
4323         }
4324
4325         ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4326                                   BTRFS_UUID_KEY_SUBVOL,
4327                                   dest->root_key.objectid);
4328         if (ret && ret != -ENOENT) {
4329                 btrfs_abort_transaction(trans, ret);
4330                 err = ret;
4331                 goto out_end_trans;
4332         }
4333         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4334                 ret = btrfs_uuid_tree_remove(trans,
4335                                           dest->root_item.received_uuid,
4336                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4337                                           dest->root_key.objectid);
4338                 if (ret && ret != -ENOENT) {
4339                         btrfs_abort_transaction(trans, ret);
4340                         err = ret;
4341                         goto out_end_trans;
4342                 }
4343         }
4344
4345 out_end_trans:
4346         trans->block_rsv = NULL;
4347         trans->bytes_reserved = 0;
4348         ret = btrfs_end_transaction(trans);
4349         if (ret && !err)
4350                 err = ret;
4351         inode->i_flags |= S_DEAD;
4352 out_release:
4353         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
4354 out_up_write:
4355         up_write(&fs_info->subvol_sem);
4356         if (err) {
4357                 spin_lock(&dest->root_item_lock);
4358                 root_flags = btrfs_root_flags(&dest->root_item);
4359                 btrfs_set_root_flags(&dest->root_item,
4360                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4361                 spin_unlock(&dest->root_item_lock);
4362         } else {
4363                 d_invalidate(dentry);
4364                 btrfs_prune_dentries(dest);
4365                 ASSERT(dest->send_in_progress == 0);
4366
4367                 /* the last ref */
4368                 if (dest->ino_cache_inode) {
4369                         iput(dest->ino_cache_inode);
4370                         dest->ino_cache_inode = NULL;
4371                 }
4372         }
4373
4374         return err;
4375 }
4376
4377 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4378 {
4379         struct inode *inode = d_inode(dentry);
4380         int err = 0;
4381         struct btrfs_root *root = BTRFS_I(dir)->root;
4382         struct btrfs_trans_handle *trans;
4383         u64 last_unlink_trans;
4384
4385         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4386                 return -ENOTEMPTY;
4387         if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4388                 return btrfs_delete_subvolume(dir, dentry);
4389
4390         trans = __unlink_start_trans(dir);
4391         if (IS_ERR(trans))
4392                 return PTR_ERR(trans);
4393
4394         if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4395                 err = btrfs_unlink_subvol(trans, dir,
4396                                           BTRFS_I(inode)->location.objectid,
4397                                           dentry->d_name.name,
4398                                           dentry->d_name.len);
4399                 goto out;
4400         }
4401
4402         err = btrfs_orphan_add(trans, BTRFS_I(inode));
4403         if (err)
4404                 goto out;
4405
4406         last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4407
4408         /* now the directory is empty */
4409         err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4410                         BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4411                         dentry->d_name.len);
4412         if (!err) {
4413                 btrfs_i_size_write(BTRFS_I(inode), 0);
4414                 /*
4415                  * Propagate the last_unlink_trans value of the deleted dir to
4416                  * its parent directory. This is to prevent an unrecoverable
4417                  * log tree in the case we do something like this:
4418                  * 1) create dir foo
4419                  * 2) create snapshot under dir foo
4420                  * 3) delete the snapshot
4421                  * 4) rmdir foo
4422                  * 5) mkdir foo
4423                  * 6) fsync foo or some file inside foo
4424                  */
4425                 if (last_unlink_trans >= trans->transid)
4426                         BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4427         }
4428 out:
4429         btrfs_end_transaction(trans);
4430         btrfs_btree_balance_dirty(root->fs_info);
4431
4432         return err;
4433 }
4434
4435 static int truncate_space_check(struct btrfs_trans_handle *trans,
4436                                 struct btrfs_root *root,
4437                                 u64 bytes_deleted)
4438 {
4439         struct btrfs_fs_info *fs_info = root->fs_info;
4440         int ret;
4441
4442         /*
4443          * This is only used to apply pressure to the enospc system, we don't
4444          * intend to use this reservation at all.
4445          */
4446         bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4447         bytes_deleted *= fs_info->nodesize;
4448         ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4449                                   bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4450         if (!ret) {
4451                 trace_btrfs_space_reservation(fs_info, "transaction",
4452                                               trans->transid,
4453                                               bytes_deleted, 1);
4454                 trans->bytes_reserved += bytes_deleted;
4455         }
4456         return ret;
4457
4458 }
4459
4460 /*
4461  * Return this if we need to call truncate_block for the last bit of the
4462  * truncate.
4463  */
4464 #define NEED_TRUNCATE_BLOCK 1
4465
4466 /*
4467  * this can truncate away extent items, csum items and directory items.
4468  * It starts at a high offset and removes keys until it can't find
4469  * any higher than new_size
4470  *
4471  * csum items that cross the new i_size are truncated to the new size
4472  * as well.
4473  *
4474  * min_type is the minimum key type to truncate down to.  If set to 0, this
4475  * will kill all the items on this inode, including the INODE_ITEM_KEY.
4476  */
4477 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4478                                struct btrfs_root *root,
4479                                struct inode *inode,
4480                                u64 new_size, u32 min_type)
4481 {
4482         struct btrfs_fs_info *fs_info = root->fs_info;
4483         struct btrfs_path *path;
4484         struct extent_buffer *leaf;
4485         struct btrfs_file_extent_item *fi;
4486         struct btrfs_key key;
4487         struct btrfs_key found_key;
4488         u64 extent_start = 0;
4489         u64 extent_num_bytes = 0;
4490         u64 extent_offset = 0;
4491         u64 item_end = 0;
4492         u64 last_size = new_size;
4493         u32 found_type = (u8)-1;
4494         int found_extent;
4495         int del_item;
4496         int pending_del_nr = 0;
4497         int pending_del_slot = 0;
4498         int extent_type = -1;
4499         int ret;
4500         u64 ino = btrfs_ino(BTRFS_I(inode));
4501         u64 bytes_deleted = 0;
4502         bool be_nice = false;
4503         bool should_throttle = false;
4504         bool should_end = false;
4505
4506         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4507
4508         /*
4509          * for non-free space inodes and ref cows, we want to back off from
4510          * time to time
4511          */
4512         if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4513             test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4514                 be_nice = true;
4515
4516         path = btrfs_alloc_path();
4517         if (!path)
4518                 return -ENOMEM;
4519         path->reada = READA_BACK;
4520
4521         /*
4522          * We want to drop from the next block forward in case this new size is
4523          * not block aligned since we will be keeping the last block of the
4524          * extent just the way it is.
4525          */
4526         if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4527             root == fs_info->tree_root)
4528                 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4529                                         fs_info->sectorsize),
4530                                         (u64)-1, 0);
4531
4532         /*
4533          * This function is also used to drop the items in the log tree before
4534          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4535          * it is used to drop the loged items. So we shouldn't kill the delayed
4536          * items.
4537          */
4538         if (min_type == 0 && root == BTRFS_I(inode)->root)
4539                 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4540
4541         key.objectid = ino;
4542         key.offset = (u64)-1;
4543         key.type = (u8)-1;
4544
4545 search_again:
4546         /*
4547          * with a 16K leaf size and 128MB extents, you can actually queue
4548          * up a huge file in a single leaf.  Most of the time that
4549          * bytes_deleted is > 0, it will be huge by the time we get here
4550          */
4551         if (be_nice && bytes_deleted > SZ_32M &&
4552             btrfs_should_end_transaction(trans)) {
4553                 ret = -EAGAIN;
4554                 goto out;
4555         }
4556
4557         path->leave_spinning = 1;
4558         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4559         if (ret < 0)
4560                 goto out;
4561
4562         if (ret > 0) {
4563                 ret = 0;
4564                 /* there are no items in the tree for us to truncate, we're
4565                  * done
4566                  */
4567                 if (path->slots[0] == 0)
4568                         goto out;
4569                 path->slots[0]--;
4570         }
4571
4572         while (1) {
4573                 fi = NULL;
4574                 leaf = path->nodes[0];
4575                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4576                 found_type = found_key.type;
4577
4578                 if (found_key.objectid != ino)
4579                         break;
4580
4581                 if (found_type < min_type)
4582                         break;
4583
4584                 item_end = found_key.offset;
4585                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4586                         fi = btrfs_item_ptr(leaf, path->slots[0],
4587                                             struct btrfs_file_extent_item);
4588                         extent_type = btrfs_file_extent_type(leaf, fi);
4589                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4590                                 item_end +=
4591                                     btrfs_file_extent_num_bytes(leaf, fi);
4592
4593                                 trace_btrfs_truncate_show_fi_regular(
4594                                         BTRFS_I(inode), leaf, fi,
4595                                         found_key.offset);
4596                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4597                                 item_end += btrfs_file_extent_ram_bytes(leaf,
4598                                                                         fi);
4599
4600                                 trace_btrfs_truncate_show_fi_inline(
4601                                         BTRFS_I(inode), leaf, fi, path->slots[0],
4602                                         found_key.offset);
4603                         }
4604                         item_end--;
4605                 }
4606                 if (found_type > min_type) {
4607                         del_item = 1;
4608                 } else {
4609                         if (item_end < new_size)
4610                                 break;
4611                         if (found_key.offset >= new_size)
4612                                 del_item = 1;
4613                         else
4614                                 del_item = 0;
4615                 }
4616                 found_extent = 0;
4617                 /* FIXME, shrink the extent if the ref count is only 1 */
4618                 if (found_type != BTRFS_EXTENT_DATA_KEY)
4619                         goto delete;
4620
4621                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4622                         u64 num_dec;
4623                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4624                         if (!del_item) {
4625                                 u64 orig_num_bytes =
4626                                         btrfs_file_extent_num_bytes(leaf, fi);
4627                                 extent_num_bytes = ALIGN(new_size -
4628                                                 found_key.offset,
4629                                                 fs_info->sectorsize);
4630                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4631                                                          extent_num_bytes);
4632                                 num_dec = (orig_num_bytes -
4633                                            extent_num_bytes);
4634                                 if (test_bit(BTRFS_ROOT_REF_COWS,
4635                                              &root->state) &&
4636                                     extent_start != 0)
4637                                         inode_sub_bytes(inode, num_dec);
4638                                 btrfs_mark_buffer_dirty(leaf);
4639                         } else {
4640                                 extent_num_bytes =
4641                                         btrfs_file_extent_disk_num_bytes(leaf,
4642                                                                          fi);
4643                                 extent_offset = found_key.offset -
4644                                         btrfs_file_extent_offset(leaf, fi);
4645
4646                                 /* FIXME blocksize != 4096 */
4647                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4648                                 if (extent_start != 0) {
4649                                         found_extent = 1;
4650                                         if (test_bit(BTRFS_ROOT_REF_COWS,
4651                                                      &root->state))
4652                                                 inode_sub_bytes(inode, num_dec);
4653                                 }
4654                         }
4655                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4656                         /*
4657                          * we can't truncate inline items that have had
4658                          * special encodings
4659                          */
4660                         if (!del_item &&
4661                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
4662                             btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4663                             btrfs_file_extent_compression(leaf, fi) == 0) {
4664                                 u32 size = (u32)(new_size - found_key.offset);
4665
4666                                 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4667                                 size = btrfs_file_extent_calc_inline_size(size);
4668                                 btrfs_truncate_item(root->fs_info, path, size, 1);
4669                         } else if (!del_item) {
4670                                 /*
4671                                  * We have to bail so the last_size is set to
4672                                  * just before this extent.
4673                                  */
4674                                 ret = NEED_TRUNCATE_BLOCK;
4675                                 break;
4676                         }
4677
4678                         if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4679                                 inode_sub_bytes(inode, item_end + 1 - new_size);
4680                 }
4681 delete:
4682                 if (del_item)
4683                         last_size = found_key.offset;
4684                 else
4685                         last_size = new_size;
4686                 if (del_item) {
4687                         if (!pending_del_nr) {
4688                                 /* no pending yet, add ourselves */
4689                                 pending_del_slot = path->slots[0];
4690                                 pending_del_nr = 1;
4691                         } else if (pending_del_nr &&
4692                                    path->slots[0] + 1 == pending_del_slot) {
4693                                 /* hop on the pending chunk */
4694                                 pending_del_nr++;
4695                                 pending_del_slot = path->slots[0];
4696                         } else {
4697                                 BUG();
4698                         }
4699                 } else {
4700                         break;
4701                 }
4702                 should_throttle = false;
4703
4704                 if (found_extent &&
4705                     (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4706                      root == fs_info->tree_root)) {
4707                         btrfs_set_path_blocking(path);
4708                         bytes_deleted += extent_num_bytes;
4709                         ret = btrfs_free_extent(trans, root, extent_start,
4710                                                 extent_num_bytes, 0,
4711                                                 btrfs_header_owner(leaf),
4712                                                 ino, extent_offset);
4713                         if (ret) {
4714                                 btrfs_abort_transaction(trans, ret);
4715                                 break;
4716                         }
4717                         if (btrfs_should_throttle_delayed_refs(trans))
4718                                 btrfs_async_run_delayed_refs(fs_info,
4719                                         trans->delayed_ref_updates * 2,
4720                                         trans->transid, 0);
4721                         if (be_nice) {
4722                                 if (truncate_space_check(trans, root,
4723                                                          extent_num_bytes)) {
4724                                         should_end = true;
4725                                 }
4726                                 if (btrfs_should_throttle_delayed_refs(trans))
4727                                         should_throttle = true;
4728                         }
4729                 }
4730
4731                 if (found_type == BTRFS_INODE_ITEM_KEY)
4732                         break;
4733
4734                 if (path->slots[0] == 0 ||
4735                     path->slots[0] != pending_del_slot ||
4736                     should_throttle || should_end) {
4737                         if (pending_del_nr) {
4738                                 ret = btrfs_del_items(trans, root, path,
4739                                                 pending_del_slot,
4740                                                 pending_del_nr);
4741                                 if (ret) {
4742                                         btrfs_abort_transaction(trans, ret);
4743                                         break;
4744                                 }
4745                                 pending_del_nr = 0;
4746                         }
4747                         btrfs_release_path(path);
4748                         if (should_throttle) {
4749                                 unsigned long updates = trans->delayed_ref_updates;
4750                                 if (updates) {
4751                                         trans->delayed_ref_updates = 0;
4752                                         ret = btrfs_run_delayed_refs(trans,
4753                                                                    updates * 2);
4754                                         if (ret)
4755                                                 break;
4756                                 }
4757                         }
4758                         /*
4759                          * if we failed to refill our space rsv, bail out
4760                          * and let the transaction restart
4761                          */
4762                         if (should_end) {
4763                                 ret = -EAGAIN;
4764                                 break;
4765                         }
4766                         goto search_again;
4767                 } else {
4768                         path->slots[0]--;
4769                 }
4770         }
4771 out:
4772         if (ret >= 0 && pending_del_nr) {
4773                 int err;
4774
4775                 err = btrfs_del_items(trans, root, path, pending_del_slot,
4776                                       pending_del_nr);
4777                 if (err) {
4778                         btrfs_abort_transaction(trans, err);
4779                         ret = err;
4780                 }
4781         }
4782         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4783                 ASSERT(last_size >= new_size);
4784                 if (!ret && last_size > new_size)
4785                         last_size = new_size;
4786                 btrfs_ordered_update_i_size(inode, last_size, NULL);
4787         }
4788
4789         btrfs_free_path(path);
4790
4791         if (be_nice && bytes_deleted > SZ_32M && (ret >= 0 || ret == -EAGAIN)) {
4792                 unsigned long updates = trans->delayed_ref_updates;
4793                 int err;
4794
4795                 if (updates) {
4796                         trans->delayed_ref_updates = 0;
4797                         err = btrfs_run_delayed_refs(trans, updates * 2);
4798                         if (err)
4799                                 ret = err;
4800                 }
4801         }
4802         return ret;
4803 }
4804
4805 /*
4806  * btrfs_truncate_block - read, zero a chunk and write a block
4807  * @inode - inode that we're zeroing
4808  * @from - the offset to start zeroing
4809  * @len - the length to zero, 0 to zero the entire range respective to the
4810  *      offset
4811  * @front - zero up to the offset instead of from the offset on
4812  *
4813  * This will find the block for the "from" offset and cow the block and zero the
4814  * part we want to zero.  This is used with truncate and hole punching.
4815  */
4816 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4817                         int front)
4818 {
4819         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4820         struct address_space *mapping = inode->i_mapping;
4821         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4822         struct btrfs_ordered_extent *ordered;
4823         struct extent_state *cached_state = NULL;
4824         struct extent_changeset *data_reserved = NULL;
4825         char *kaddr;
4826         u32 blocksize = fs_info->sectorsize;
4827         pgoff_t index = from >> PAGE_SHIFT;
4828         unsigned offset = from & (blocksize - 1);
4829         struct page *page;
4830         gfp_t mask = btrfs_alloc_write_mask(mapping);
4831         int ret = 0;
4832         u64 block_start;
4833         u64 block_end;
4834
4835         if (IS_ALIGNED(offset, blocksize) &&
4836             (!len || IS_ALIGNED(len, blocksize)))
4837                 goto out;
4838
4839         block_start = round_down(from, blocksize);
4840         block_end = block_start + blocksize - 1;
4841
4842         ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
4843                                            block_start, blocksize);
4844         if (ret)
4845                 goto out;
4846
4847 again:
4848         page = find_or_create_page(mapping, index, mask);
4849         if (!page) {
4850                 btrfs_delalloc_release_space(inode, data_reserved,
4851                                              block_start, blocksize, true);
4852                 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, true);
4853                 ret = -ENOMEM;
4854                 goto out;
4855         }
4856
4857         if (!PageUptodate(page)) {
4858                 ret = btrfs_readpage(NULL, page);
4859                 lock_page(page);
4860                 if (page->mapping != mapping) {
4861                         unlock_page(page);
4862                         put_page(page);
4863                         goto again;
4864                 }
4865                 if (!PageUptodate(page)) {
4866                         ret = -EIO;
4867                         goto out_unlock;
4868                 }
4869         }
4870         wait_on_page_writeback(page);
4871
4872         lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4873         set_page_extent_mapped(page);
4874
4875         ordered = btrfs_lookup_ordered_extent(inode, block_start);
4876         if (ordered) {
4877                 unlock_extent_cached(io_tree, block_start, block_end,
4878                                      &cached_state);
4879                 unlock_page(page);
4880                 put_page(page);
4881                 btrfs_start_ordered_extent(inode, ordered, 1);
4882                 btrfs_put_ordered_extent(ordered);
4883                 goto again;
4884         }
4885
4886         clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4887                           EXTENT_DIRTY | EXTENT_DELALLOC |
4888                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4889                           0, 0, &cached_state);
4890
4891         ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4892                                         &cached_state, 0);
4893         if (ret) {
4894                 unlock_extent_cached(io_tree, block_start, block_end,
4895                                      &cached_state);
4896                 goto out_unlock;
4897         }
4898
4899         if (offset != blocksize) {
4900                 if (!len)
4901                         len = blocksize - offset;
4902                 kaddr = kmap(page);
4903                 if (front)
4904                         memset(kaddr + (block_start - page_offset(page)),
4905                                 0, offset);
4906                 else
4907                         memset(kaddr + (block_start - page_offset(page)) +  offset,
4908                                 0, len);
4909                 flush_dcache_page(page);
4910                 kunmap(page);
4911         }
4912         ClearPageChecked(page);
4913         set_page_dirty(page);
4914         unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4915
4916 out_unlock:
4917         if (ret)
4918                 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4919                                              blocksize, true);
4920         btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, (ret != 0));
4921         unlock_page(page);
4922         put_page(page);
4923 out:
4924         extent_changeset_free(data_reserved);
4925         return ret;
4926 }
4927
4928 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4929                              u64 offset, u64 len)
4930 {
4931         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4932         struct btrfs_trans_handle *trans;
4933         int ret;
4934
4935         /*
4936          * Still need to make sure the inode looks like it's been updated so
4937          * that any holes get logged if we fsync.
4938          */
4939         if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4940                 BTRFS_I(inode)->last_trans = fs_info->generation;
4941                 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4942                 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4943                 return 0;
4944         }
4945
4946         /*
4947          * 1 - for the one we're dropping
4948          * 1 - for the one we're adding
4949          * 1 - for updating the inode.
4950          */
4951         trans = btrfs_start_transaction(root, 3);
4952         if (IS_ERR(trans))
4953                 return PTR_ERR(trans);
4954
4955         ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4956         if (ret) {
4957                 btrfs_abort_transaction(trans, ret);
4958                 btrfs_end_transaction(trans);
4959                 return ret;
4960         }
4961
4962         ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4963                         offset, 0, 0, len, 0, len, 0, 0, 0);
4964         if (ret)
4965                 btrfs_abort_transaction(trans, ret);
4966         else
4967                 btrfs_update_inode(trans, root, inode);
4968         btrfs_end_transaction(trans);
4969         return ret;
4970 }
4971
4972 /*
4973  * This function puts in dummy file extents for the area we're creating a hole
4974  * for.  So if we are truncating this file to a larger size we need to insert
4975  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4976  * the range between oldsize and size
4977  */
4978 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4979 {
4980         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4981         struct btrfs_root *root = BTRFS_I(inode)->root;
4982         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4983         struct extent_map *em = NULL;
4984         struct extent_state *cached_state = NULL;
4985         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4986         u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4987         u64 block_end = ALIGN(size, fs_info->sectorsize);
4988         u64 last_byte;
4989         u64 cur_offset;
4990         u64 hole_size;
4991         int err = 0;
4992
4993         /*
4994          * If our size started in the middle of a block we need to zero out the
4995          * rest of the block before we expand the i_size, otherwise we could
4996          * expose stale data.
4997          */
4998         err = btrfs_truncate_block(inode, oldsize, 0, 0);
4999         if (err)
5000                 return err;
5001
5002         if (size <= hole_start)
5003                 return 0;
5004
5005         while (1) {
5006                 struct btrfs_ordered_extent *ordered;
5007
5008                 lock_extent_bits(io_tree, hole_start, block_end - 1,
5009                                  &cached_state);
5010                 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
5011                                                      block_end - hole_start);
5012                 if (!ordered)
5013                         break;
5014                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
5015                                      &cached_state);
5016                 btrfs_start_ordered_extent(inode, ordered, 1);
5017                 btrfs_put_ordered_extent(ordered);
5018         }
5019
5020         cur_offset = hole_start;
5021         while (1) {
5022                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
5023                                 block_end - cur_offset, 0);
5024                 if (IS_ERR(em)) {
5025                         err = PTR_ERR(em);
5026                         em = NULL;
5027                         break;
5028                 }
5029                 last_byte = min(extent_map_end(em), block_end);
5030                 last_byte = ALIGN(last_byte, fs_info->sectorsize);
5031                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5032                         struct extent_map *hole_em;
5033                         hole_size = last_byte - cur_offset;
5034
5035                         err = maybe_insert_hole(root, inode, cur_offset,
5036                                                 hole_size);
5037                         if (err)
5038                                 break;
5039                         btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
5040                                                 cur_offset + hole_size - 1, 0);
5041                         hole_em = alloc_extent_map();
5042                         if (!hole_em) {
5043                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5044                                         &BTRFS_I(inode)->runtime_flags);
5045                                 goto next;
5046                         }
5047                         hole_em->start = cur_offset;
5048                         hole_em->len = hole_size;
5049                         hole_em->orig_start = cur_offset;
5050
5051                         hole_em->block_start = EXTENT_MAP_HOLE;
5052                         hole_em->block_len = 0;
5053                         hole_em->orig_block_len = 0;
5054                         hole_em->ram_bytes = hole_size;
5055                         hole_em->bdev = fs_info->fs_devices->latest_bdev;
5056                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
5057                         hole_em->generation = fs_info->generation;
5058
5059                         while (1) {
5060                                 write_lock(&em_tree->lock);
5061                                 err = add_extent_mapping(em_tree, hole_em, 1);
5062                                 write_unlock(&em_tree->lock);
5063                                 if (err != -EEXIST)
5064                                         break;
5065                                 btrfs_drop_extent_cache(BTRFS_I(inode),
5066                                                         cur_offset,
5067                                                         cur_offset +
5068                                                         hole_size - 1, 0);
5069                         }
5070                         free_extent_map(hole_em);
5071                 }
5072 next:
5073                 free_extent_map(em);
5074                 em = NULL;
5075                 cur_offset = last_byte;
5076                 if (cur_offset >= block_end)
5077                         break;
5078         }
5079         free_extent_map(em);
5080         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
5081         return err;
5082 }
5083
5084 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5085 {
5086         struct btrfs_root *root = BTRFS_I(inode)->root;
5087         struct btrfs_trans_handle *trans;
5088         loff_t oldsize = i_size_read(inode);
5089         loff_t newsize = attr->ia_size;
5090         int mask = attr->ia_valid;
5091         int ret;
5092
5093         /*
5094          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5095          * special case where we need to update the times despite not having
5096          * these flags set.  For all other operations the VFS set these flags
5097          * explicitly if it wants a timestamp update.
5098          */
5099         if (newsize != oldsize) {
5100                 inode_inc_iversion(inode);
5101                 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5102                         inode->i_ctime = inode->i_mtime =
5103                                 current_time(inode);
5104         }
5105
5106         if (newsize > oldsize) {
5107                 /*
5108                  * Don't do an expanding truncate while snapshotting is ongoing.
5109                  * This is to ensure the snapshot captures a fully consistent
5110                  * state of this file - if the snapshot captures this expanding
5111                  * truncation, it must capture all writes that happened before
5112                  * this truncation.
5113                  */
5114                 btrfs_wait_for_snapshot_creation(root);
5115                 ret = btrfs_cont_expand(inode, oldsize, newsize);
5116                 if (ret) {
5117                         btrfs_end_write_no_snapshotting(root);
5118                         return ret;
5119                 }
5120
5121                 trans = btrfs_start_transaction(root, 1);
5122                 if (IS_ERR(trans)) {
5123                         btrfs_end_write_no_snapshotting(root);
5124                         return PTR_ERR(trans);
5125                 }
5126
5127                 i_size_write(inode, newsize);
5128                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
5129                 pagecache_isize_extended(inode, oldsize, newsize);
5130                 ret = btrfs_update_inode(trans, root, inode);
5131                 btrfs_end_write_no_snapshotting(root);
5132                 btrfs_end_transaction(trans);
5133         } else {
5134
5135                 /*
5136                  * We're truncating a file that used to have good data down to
5137                  * zero. Make sure it gets into the ordered flush list so that
5138                  * any new writes get down to disk quickly.
5139                  */
5140                 if (newsize == 0)
5141                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5142                                 &BTRFS_I(inode)->runtime_flags);
5143
5144                 truncate_setsize(inode, newsize);
5145
5146                 /* Disable nonlocked read DIO to avoid the end less truncate */
5147                 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
5148                 inode_dio_wait(inode);
5149                 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
5150
5151                 ret = btrfs_truncate(inode, newsize == oldsize);
5152                 if (ret && inode->i_nlink) {
5153                         int err;
5154
5155                         /*
5156                          * Truncate failed, so fix up the in-memory size. We
5157                          * adjusted disk_i_size down as we removed extents, so
5158                          * wait for disk_i_size to be stable and then update the
5159                          * in-memory size to match.
5160                          */
5161                         err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5162                         if (err)
5163                                 return err;
5164                         i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5165                 }
5166         }
5167
5168         return ret;
5169 }
5170
5171 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5172 {
5173         struct inode *inode = d_inode(dentry);
5174         struct btrfs_root *root = BTRFS_I(inode)->root;
5175         int err;
5176
5177         if (btrfs_root_readonly(root))
5178                 return -EROFS;
5179
5180         err = setattr_prepare(dentry, attr);
5181         if (err)
5182                 return err;
5183
5184         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5185                 err = btrfs_setsize(inode, attr);
5186                 if (err)
5187                         return err;
5188         }
5189
5190         if (attr->ia_valid) {
5191                 setattr_copy(inode, attr);
5192                 inode_inc_iversion(inode);
5193                 err = btrfs_dirty_inode(inode);
5194
5195                 if (!err && attr->ia_valid & ATTR_MODE)
5196                         err = posix_acl_chmod(inode, inode->i_mode);
5197         }
5198
5199         return err;
5200 }
5201
5202 /*
5203  * While truncating the inode pages during eviction, we get the VFS calling
5204  * btrfs_invalidatepage() against each page of the inode. This is slow because
5205  * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5206  * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5207  * extent_state structures over and over, wasting lots of time.
5208  *
5209  * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5210  * those expensive operations on a per page basis and do only the ordered io
5211  * finishing, while we release here the extent_map and extent_state structures,
5212  * without the excessive merging and splitting.
5213  */
5214 static void evict_inode_truncate_pages(struct inode *inode)
5215 {
5216         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5217         struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5218         struct rb_node *node;
5219
5220         ASSERT(inode->i_state & I_FREEING);
5221         truncate_inode_pages_final(&inode->i_data);
5222
5223         write_lock(&map_tree->lock);
5224         while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
5225                 struct extent_map *em;
5226
5227                 node = rb_first_cached(&map_tree->map);
5228                 em = rb_entry(node, struct extent_map, rb_node);
5229                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5230                 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5231                 remove_extent_mapping(map_tree, em);
5232                 free_extent_map(em);
5233                 if (need_resched()) {
5234                         write_unlock(&map_tree->lock);
5235                         cond_resched();
5236                         write_lock(&map_tree->lock);
5237                 }
5238         }
5239         write_unlock(&map_tree->lock);
5240
5241         /*
5242          * Keep looping until we have no more ranges in the io tree.
5243          * We can have ongoing bios started by readpages (called from readahead)
5244          * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5245          * still in progress (unlocked the pages in the bio but did not yet
5246          * unlocked the ranges in the io tree). Therefore this means some
5247          * ranges can still be locked and eviction started because before
5248          * submitting those bios, which are executed by a separate task (work
5249          * queue kthread), inode references (inode->i_count) were not taken
5250          * (which would be dropped in the end io callback of each bio).
5251          * Therefore here we effectively end up waiting for those bios and
5252          * anyone else holding locked ranges without having bumped the inode's
5253          * reference count - if we don't do it, when they access the inode's
5254          * io_tree to unlock a range it may be too late, leading to an
5255          * use-after-free issue.
5256          */
5257         spin_lock(&io_tree->lock);
5258         while (!RB_EMPTY_ROOT(&io_tree->state)) {
5259                 struct extent_state *state;
5260                 struct extent_state *cached_state = NULL;
5261                 u64 start;
5262                 u64 end;
5263                 unsigned state_flags;
5264
5265                 node = rb_first(&io_tree->state);
5266                 state = rb_entry(node, struct extent_state, rb_node);
5267                 start = state->start;
5268                 end = state->end;
5269                 state_flags = state->state;
5270                 spin_unlock(&io_tree->lock);
5271
5272                 lock_extent_bits(io_tree, start, end, &cached_state);
5273
5274                 /*
5275                  * If still has DELALLOC flag, the extent didn't reach disk,
5276                  * and its reserved space won't be freed by delayed_ref.
5277                  * So we need to free its reserved space here.
5278                  * (Refer to comment in btrfs_invalidatepage, case 2)
5279                  *
5280                  * Note, end is the bytenr of last byte, so we need + 1 here.
5281                  */
5282                 if (state_flags & EXTENT_DELALLOC)
5283                         btrfs_qgroup_free_data(inode, NULL, start, end - start + 1);
5284
5285                 clear_extent_bit(io_tree, start, end,
5286                                  EXTENT_LOCKED | EXTENT_DIRTY |
5287                                  EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5288                                  EXTENT_DEFRAG, 1, 1, &cached_state);
5289
5290                 cond_resched();
5291                 spin_lock(&io_tree->lock);
5292         }
5293         spin_unlock(&io_tree->lock);
5294 }
5295
5296 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5297                                                         struct btrfs_block_rsv *rsv)
5298 {
5299         struct btrfs_fs_info *fs_info = root->fs_info;
5300         struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5301         int failures = 0;
5302
5303         for (;;) {
5304                 struct btrfs_trans_handle *trans;
5305                 int ret;
5306
5307                 ret = btrfs_block_rsv_refill(root, rsv, rsv->size,
5308                                              BTRFS_RESERVE_FLUSH_LIMIT);
5309
5310                 if (ret && ++failures > 2) {
5311                         btrfs_warn(fs_info,
5312                                    "could not allocate space for a delete; will truncate on mount");
5313                         return ERR_PTR(-ENOSPC);
5314                 }
5315
5316                 trans = btrfs_join_transaction(root);
5317                 if (IS_ERR(trans) || !ret)
5318                         return trans;
5319
5320                 /*
5321                  * Try to steal from the global reserve if there is space for
5322                  * it.
5323                  */
5324                 if (!btrfs_check_space_for_delayed_refs(trans) &&
5325                     !btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, false))
5326                         return trans;
5327
5328                 /* If not, commit and try again. */
5329                 ret = btrfs_commit_transaction(trans);
5330                 if (ret)
5331                         return ERR_PTR(ret);
5332         }
5333 }
5334
5335 void btrfs_evict_inode(struct inode *inode)
5336 {
5337         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5338         struct btrfs_trans_handle *trans;
5339         struct btrfs_root *root = BTRFS_I(inode)->root;
5340         struct btrfs_block_rsv *rsv;
5341         int ret;
5342
5343         trace_btrfs_inode_evict(inode);
5344
5345         if (!root) {
5346                 clear_inode(inode);
5347                 return;
5348         }
5349
5350         evict_inode_truncate_pages(inode);
5351
5352         if (inode->i_nlink &&
5353             ((btrfs_root_refs(&root->root_item) != 0 &&
5354               root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5355              btrfs_is_free_space_inode(BTRFS_I(inode))))
5356                 goto no_delete;
5357
5358         if (is_bad_inode(inode))
5359                 goto no_delete;
5360
5361         btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5362
5363         if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5364                 goto no_delete;
5365
5366         if (inode->i_nlink > 0) {
5367                 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5368                        root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5369                 goto no_delete;
5370         }
5371
5372         ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5373         if (ret)
5374                 goto no_delete;
5375
5376         rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5377         if (!rsv)
5378                 goto no_delete;
5379         rsv->size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5380         rsv->failfast = 1;
5381
5382         btrfs_i_size_write(BTRFS_I(inode), 0);
5383
5384         while (1) {
5385                 trans = evict_refill_and_join(root, rsv);
5386                 if (IS_ERR(trans))
5387                         goto free_rsv;
5388
5389                 trans->block_rsv = rsv;
5390
5391                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5392                 trans->block_rsv = &fs_info->trans_block_rsv;
5393                 btrfs_end_transaction(trans);
5394                 btrfs_btree_balance_dirty(fs_info);
5395                 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5396                         goto free_rsv;
5397                 else if (!ret)
5398                         break;
5399         }
5400
5401         /*
5402          * Errors here aren't a big deal, it just means we leave orphan items in
5403          * the tree. They will be cleaned up on the next mount. If the inode
5404          * number gets reused, cleanup deletes the orphan item without doing
5405          * anything, and unlink reuses the existing orphan item.
5406          *
5407          * If it turns out that we are dropping too many of these, we might want
5408          * to add a mechanism for retrying these after a commit.
5409          */
5410         trans = evict_refill_and_join(root, rsv);
5411         if (!IS_ERR(trans)) {
5412                 trans->block_rsv = rsv;
5413                 btrfs_orphan_del(trans, BTRFS_I(inode));
5414                 trans->block_rsv = &fs_info->trans_block_rsv;
5415                 btrfs_end_transaction(trans);
5416         }
5417
5418         if (!(root == fs_info->tree_root ||
5419               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5420                 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5421
5422 free_rsv:
5423         btrfs_free_block_rsv(fs_info, rsv);
5424 no_delete:
5425         /*
5426          * If we didn't successfully delete, the orphan item will still be in
5427          * the tree and we'll retry on the next mount. Again, we might also want
5428          * to retry these periodically in the future.
5429          */
5430         btrfs_remove_delayed_node(BTRFS_I(inode));
5431         clear_inode(inode);
5432 }
5433
5434 /*
5435  * this returns the key found in the dir entry in the location pointer.
5436  * If no dir entries were found, returns -ENOENT.
5437  * If found a corrupted location in dir entry, returns -EUCLEAN.
5438  */
5439 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5440                                struct btrfs_key *location)
5441 {
5442         const char *name = dentry->d_name.name;
5443         int namelen = dentry->d_name.len;
5444         struct btrfs_dir_item *di;
5445         struct btrfs_path *path;
5446         struct btrfs_root *root = BTRFS_I(dir)->root;
5447         int ret = 0;
5448
5449         path = btrfs_alloc_path();
5450         if (!path)
5451                 return -ENOMEM;
5452
5453         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5454                         name, namelen, 0);
5455         if (IS_ERR_OR_NULL(di)) {
5456                 ret = di ? PTR_ERR(di) : -ENOENT;
5457                 goto out;
5458         }
5459
5460         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5461         if (location->type != BTRFS_INODE_ITEM_KEY &&
5462             location->type != BTRFS_ROOT_ITEM_KEY) {
5463                 ret = -EUCLEAN;
5464                 btrfs_warn(root->fs_info,
5465 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5466                            __func__, name, btrfs_ino(BTRFS_I(dir)),
5467                            location->objectid, location->type, location->offset);
5468         }
5469 out:
5470         btrfs_free_path(path);
5471         return ret;
5472 }
5473
5474 /*
5475  * when we hit a tree root in a directory, the btrfs part of the inode
5476  * needs to be changed to reflect the root directory of the tree root.  This
5477  * is kind of like crossing a mount point.
5478  */
5479 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5480                                     struct inode *dir,
5481                                     struct dentry *dentry,
5482                                     struct btrfs_key *location,
5483                                     struct btrfs_root **sub_root)
5484 {
5485         struct btrfs_path *path;
5486         struct btrfs_root *new_root;
5487         struct btrfs_root_ref *ref;
5488         struct extent_buffer *leaf;
5489         struct btrfs_key key;
5490         int ret;
5491         int err = 0;
5492
5493         path = btrfs_alloc_path();
5494         if (!path) {
5495                 err = -ENOMEM;
5496                 goto out;
5497         }
5498
5499         err = -ENOENT;
5500         key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5501         key.type = BTRFS_ROOT_REF_KEY;
5502         key.offset = location->objectid;
5503
5504         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5505         if (ret) {
5506                 if (ret < 0)
5507                         err = ret;
5508                 goto out;
5509         }
5510
5511         leaf = path->nodes[0];
5512         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5513         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5514             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5515                 goto out;
5516
5517         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5518                                    (unsigned long)(ref + 1),
5519                                    dentry->d_name.len);
5520         if (ret)
5521                 goto out;
5522
5523         btrfs_release_path(path);
5524
5525         new_root = btrfs_read_fs_root_no_name(fs_info, location);
5526         if (IS_ERR(new_root)) {
5527                 err = PTR_ERR(new_root);
5528                 goto out;
5529         }
5530
5531         *sub_root = new_root;
5532         location->objectid = btrfs_root_dirid(&new_root->root_item);
5533         location->type = BTRFS_INODE_ITEM_KEY;
5534         location->offset = 0;
5535         err = 0;
5536 out:
5537         btrfs_free_path(path);
5538         return err;
5539 }
5540
5541 static void inode_tree_add(struct inode *inode)
5542 {
5543         struct btrfs_root *root = BTRFS_I(inode)->root;
5544         struct btrfs_inode *entry;
5545         struct rb_node **p;
5546         struct rb_node *parent;
5547         struct rb_node *new = &BTRFS_I(inode)->rb_node;
5548         u64 ino = btrfs_ino(BTRFS_I(inode));
5549
5550         if (inode_unhashed(inode))
5551                 return;
5552         parent = NULL;
5553         spin_lock(&root->inode_lock);
5554         p = &root->inode_tree.rb_node;
5555         while (*p) {
5556                 parent = *p;
5557                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5558
5559                 if (ino < btrfs_ino(entry))
5560                         p = &parent->rb_left;
5561                 else if (ino > btrfs_ino(entry))
5562                         p = &parent->rb_right;
5563                 else {
5564                         WARN_ON(!(entry->vfs_inode.i_state &
5565                                   (I_WILL_FREE | I_FREEING)));
5566                         rb_replace_node(parent, new, &root->inode_tree);
5567                         RB_CLEAR_NODE(parent);
5568                         spin_unlock(&root->inode_lock);
5569                         return;
5570                 }
5571         }
5572         rb_link_node(new, parent, p);
5573         rb_insert_color(new, &root->inode_tree);
5574         spin_unlock(&root->inode_lock);
5575 }
5576
5577 static void inode_tree_del(struct inode *inode)
5578 {
5579         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5580         struct btrfs_root *root = BTRFS_I(inode)->root;
5581         int empty = 0;
5582
5583         spin_lock(&root->inode_lock);
5584         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5585                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5586                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5587                 empty = RB_EMPTY_ROOT(&root->inode_tree);
5588         }
5589         spin_unlock(&root->inode_lock);
5590
5591         if (empty && btrfs_root_refs(&root->root_item) == 0) {
5592                 synchronize_srcu(&fs_info->subvol_srcu);
5593                 spin_lock(&root->inode_lock);
5594                 empty = RB_EMPTY_ROOT(&root->inode_tree);
5595                 spin_unlock(&root->inode_lock);
5596                 if (empty)
5597                         btrfs_add_dead_root(root);
5598         }
5599 }
5600
5601
5602 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5603 {
5604         struct btrfs_iget_args *args = p;
5605         inode->i_ino = args->location->objectid;
5606         memcpy(&BTRFS_I(inode)->location, args->location,
5607                sizeof(*args->location));
5608         BTRFS_I(inode)->root = args->root;
5609         return 0;
5610 }
5611
5612 static int btrfs_find_actor(struct inode *inode, void *opaque)
5613 {
5614         struct btrfs_iget_args *args = opaque;
5615         return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5616                 args->root == BTRFS_I(inode)->root;
5617 }
5618
5619 static struct inode *btrfs_iget_locked(struct super_block *s,
5620                                        struct btrfs_key *location,
5621                                        struct btrfs_root *root)
5622 {
5623         struct inode *inode;
5624         struct btrfs_iget_args args;
5625         unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5626
5627         args.location = location;
5628         args.root = root;
5629
5630         inode = iget5_locked(s, hashval, btrfs_find_actor,
5631                              btrfs_init_locked_inode,
5632                              (void *)&args);
5633         return inode;
5634 }
5635
5636 /* Get an inode object given its location and corresponding root.
5637  * Returns in *is_new if the inode was read from disk
5638  */
5639 struct inode *btrfs_iget_path(struct super_block *s, struct btrfs_key *location,
5640                               struct btrfs_root *root, int *new,
5641                               struct btrfs_path *path)
5642 {
5643         struct inode *inode;
5644
5645         inode = btrfs_iget_locked(s, location, root);
5646         if (!inode)
5647                 return ERR_PTR(-ENOMEM);
5648
5649         if (inode->i_state & I_NEW) {
5650                 int ret;
5651
5652                 ret = btrfs_read_locked_inode(inode, path);
5653                 if (!ret) {
5654                         inode_tree_add(inode);
5655                         unlock_new_inode(inode);
5656                         if (new)
5657                                 *new = 1;
5658                 } else {
5659                         iget_failed(inode);
5660                         /*
5661                          * ret > 0 can come from btrfs_search_slot called by
5662                          * btrfs_read_locked_inode, this means the inode item
5663                          * was not found.
5664                          */
5665                         if (ret > 0)
5666                                 ret = -ENOENT;
5667                         inode = ERR_PTR(ret);
5668                 }
5669         }
5670
5671         return inode;
5672 }
5673
5674 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5675                          struct btrfs_root *root, int *new)
5676 {
5677         return btrfs_iget_path(s, location, root, new, NULL);
5678 }
5679
5680 static struct inode *new_simple_dir(struct super_block *s,
5681                                     struct btrfs_key *key,
5682                                     struct btrfs_root *root)
5683 {
5684         struct inode *inode = new_inode(s);
5685
5686         if (!inode)
5687                 return ERR_PTR(-ENOMEM);
5688
5689         BTRFS_I(inode)->root = root;
5690         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5691         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5692
5693         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5694         inode->i_op = &btrfs_dir_ro_inode_operations;
5695         inode->i_opflags &= ~IOP_XATTR;
5696         inode->i_fop = &simple_dir_operations;
5697         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5698         inode->i_mtime = current_time(inode);
5699         inode->i_atime = inode->i_mtime;
5700         inode->i_ctime = inode->i_mtime;
5701         BTRFS_I(inode)->i_otime = inode->i_mtime;
5702
5703         return inode;
5704 }
5705
5706 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5707 {
5708         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5709         struct inode *inode;
5710         struct btrfs_root *root = BTRFS_I(dir)->root;
5711         struct btrfs_root *sub_root = root;
5712         struct btrfs_key location;
5713         int index;
5714         int ret = 0;
5715
5716         if (dentry->d_name.len > BTRFS_NAME_LEN)
5717                 return ERR_PTR(-ENAMETOOLONG);
5718
5719         ret = btrfs_inode_by_name(dir, dentry, &location);
5720         if (ret < 0)
5721                 return ERR_PTR(ret);
5722
5723         if (location.type == BTRFS_INODE_ITEM_KEY) {
5724                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5725                 return inode;
5726         }
5727
5728         index = srcu_read_lock(&fs_info->subvol_srcu);
5729         ret = fixup_tree_root_location(fs_info, dir, dentry,
5730                                        &location, &sub_root);
5731         if (ret < 0) {
5732                 if (ret != -ENOENT)
5733                         inode = ERR_PTR(ret);
5734                 else
5735                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
5736         } else {
5737                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5738         }
5739         srcu_read_unlock(&fs_info->subvol_srcu, index);
5740
5741         if (!IS_ERR(inode) && root != sub_root) {
5742                 down_read(&fs_info->cleanup_work_sem);
5743                 if (!sb_rdonly(inode->i_sb))
5744                         ret = btrfs_orphan_cleanup(sub_root);
5745                 up_read(&fs_info->cleanup_work_sem);
5746                 if (ret) {
5747                         iput(inode);
5748                         inode = ERR_PTR(ret);
5749                 }
5750         }
5751
5752         return inode;
5753 }
5754
5755 static int btrfs_dentry_delete(const struct dentry *dentry)
5756 {
5757         struct btrfs_root *root;
5758         struct inode *inode = d_inode(dentry);
5759
5760         if (!inode && !IS_ROOT(dentry))
5761                 inode = d_inode(dentry->d_parent);
5762
5763         if (inode) {
5764                 root = BTRFS_I(inode)->root;
5765                 if (btrfs_root_refs(&root->root_item) == 0)
5766                         return 1;
5767
5768                 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5769                         return 1;
5770         }
5771         return 0;
5772 }
5773
5774 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5775                                    unsigned int flags)
5776 {
5777         struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5778
5779         if (inode == ERR_PTR(-ENOENT))
5780                 inode = NULL;
5781         return d_splice_alias(inode, dentry);
5782 }
5783
5784 unsigned char btrfs_filetype_table[] = {
5785         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5786 };
5787
5788 /*
5789  * All this infrastructure exists because dir_emit can fault, and we are holding
5790  * the tree lock when doing readdir.  For now just allocate a buffer and copy
5791  * our information into that, and then dir_emit from the buffer.  This is
5792  * similar to what NFS does, only we don't keep the buffer around in pagecache
5793  * because I'm afraid I'll mess that up.  Long term we need to make filldir do
5794  * copy_to_user_inatomic so we don't have to worry about page faulting under the
5795  * tree lock.
5796  */
5797 static int btrfs_opendir(struct inode *inode, struct file *file)
5798 {
5799         struct btrfs_file_private *private;
5800
5801         private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5802         if (!private)
5803                 return -ENOMEM;
5804         private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5805         if (!private->filldir_buf) {
5806                 kfree(private);
5807                 return -ENOMEM;
5808         }
5809         file->private_data = private;
5810         return 0;
5811 }
5812
5813 struct dir_entry {
5814         u64 ino;
5815         u64 offset;
5816         unsigned type;
5817         int name_len;
5818 };
5819
5820 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5821 {
5822         while (entries--) {
5823                 struct dir_entry *entry = addr;
5824                 char *name = (char *)(entry + 1);
5825
5826                 ctx->pos = get_unaligned(&entry->offset);
5827                 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5828                                          get_unaligned(&entry->ino),
5829                                          get_unaligned(&entry->type)))
5830                         return 1;
5831                 addr += sizeof(struct dir_entry) +
5832                         get_unaligned(&entry->name_len);
5833                 ctx->pos++;
5834         }
5835         return 0;
5836 }
5837
5838 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5839 {
5840         struct inode *inode = file_inode(file);
5841         struct btrfs_root *root = BTRFS_I(inode)->root;
5842         struct btrfs_file_private *private = file->private_data;
5843         struct btrfs_dir_item *di;
5844         struct btrfs_key key;
5845         struct btrfs_key found_key;
5846         struct btrfs_path *path;
5847         void *addr;
5848         struct list_head ins_list;
5849         struct list_head del_list;
5850         int ret;
5851         struct extent_buffer *leaf;
5852         int slot;
5853         char *name_ptr;
5854         int name_len;
5855         int entries = 0;
5856         int total_len = 0;
5857         bool put = false;
5858         struct btrfs_key location;
5859
5860         if (!dir_emit_dots(file, ctx))
5861                 return 0;
5862
5863         path = btrfs_alloc_path();
5864         if (!path)
5865                 return -ENOMEM;
5866
5867         addr = private->filldir_buf;
5868         path->reada = READA_FORWARD;
5869
5870         INIT_LIST_HEAD(&ins_list);
5871         INIT_LIST_HEAD(&del_list);
5872         put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5873
5874 again:
5875         key.type = BTRFS_DIR_INDEX_KEY;
5876         key.offset = ctx->pos;
5877         key.objectid = btrfs_ino(BTRFS_I(inode));
5878
5879         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5880         if (ret < 0)
5881                 goto err;
5882
5883         while (1) {
5884                 struct dir_entry *entry;
5885
5886                 leaf = path->nodes[0];
5887                 slot = path->slots[0];
5888                 if (slot >= btrfs_header_nritems(leaf)) {
5889                         ret = btrfs_next_leaf(root, path);
5890                         if (ret < 0)
5891                                 goto err;
5892                         else if (ret > 0)
5893                                 break;
5894                         continue;
5895                 }
5896
5897                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5898
5899                 if (found_key.objectid != key.objectid)
5900                         break;
5901                 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5902                         break;
5903                 if (found_key.offset < ctx->pos)
5904                         goto next;
5905                 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5906                         goto next;
5907                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5908                 name_len = btrfs_dir_name_len(leaf, di);
5909                 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5910                     PAGE_SIZE) {
5911                         btrfs_release_path(path);
5912                         ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5913                         if (ret)
5914                                 goto nopos;
5915                         addr = private->filldir_buf;
5916                         entries = 0;
5917                         total_len = 0;
5918                         goto again;
5919                 }
5920
5921                 entry = addr;
5922                 put_unaligned(name_len, &entry->name_len);
5923                 name_ptr = (char *)(entry + 1);
5924                 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5925                                    name_len);
5926                 put_unaligned(btrfs_filetype_table[btrfs_dir_type(leaf, di)],
5927                                 &entry->type);
5928                 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5929                 put_unaligned(location.objectid, &entry->ino);
5930                 put_unaligned(found_key.offset, &entry->offset);
5931                 entries++;
5932                 addr += sizeof(struct dir_entry) + name_len;
5933                 total_len += sizeof(struct dir_entry) + name_len;
5934 next:
5935                 path->slots[0]++;
5936         }
5937         btrfs_release_path(path);
5938
5939         ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5940         if (ret)
5941                 goto nopos;
5942
5943         ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5944         if (ret)
5945                 goto nopos;
5946
5947         /*
5948          * Stop new entries from being returned after we return the last
5949          * entry.
5950          *
5951          * New directory entries are assigned a strictly increasing
5952          * offset.  This means that new entries created during readdir
5953          * are *guaranteed* to be seen in the future by that readdir.
5954          * This has broken buggy programs which operate on names as
5955          * they're returned by readdir.  Until we re-use freed offsets
5956          * we have this hack to stop new entries from being returned
5957          * under the assumption that they'll never reach this huge
5958          * offset.
5959          *
5960          * This is being careful not to overflow 32bit loff_t unless the
5961          * last entry requires it because doing so has broken 32bit apps
5962          * in the past.
5963          */
5964         if (ctx->pos >= INT_MAX)
5965                 ctx->pos = LLONG_MAX;
5966         else
5967                 ctx->pos = INT_MAX;
5968 nopos:
5969         ret = 0;
5970 err:
5971         if (put)
5972                 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5973         btrfs_free_path(path);
5974         return ret;
5975 }
5976
5977 /*
5978  * This is somewhat expensive, updating the tree every time the
5979  * inode changes.  But, it is most likely to find the inode in cache.
5980  * FIXME, needs more benchmarking...there are no reasons other than performance
5981  * to keep or drop this code.
5982  */
5983 static int btrfs_dirty_inode(struct inode *inode)
5984 {
5985         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5986         struct btrfs_root *root = BTRFS_I(inode)->root;
5987         struct btrfs_trans_handle *trans;
5988         int ret;
5989
5990         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5991                 return 0;
5992
5993         trans = btrfs_join_transaction(root);
5994         if (IS_ERR(trans))
5995                 return PTR_ERR(trans);
5996
5997         ret = btrfs_update_inode(trans, root, inode);
5998         if (ret && ret == -ENOSPC) {
5999                 /* whoops, lets try again with the full transaction */
6000                 btrfs_end_transaction(trans);
6001                 trans = btrfs_start_transaction(root, 1);
6002                 if (IS_ERR(trans))
6003                         return PTR_ERR(trans);
6004
6005                 ret = btrfs_update_inode(trans, root, inode);
6006         }
6007         btrfs_end_transaction(trans);
6008         if (BTRFS_I(inode)->delayed_node)
6009                 btrfs_balance_delayed_items(fs_info);
6010
6011         return ret;
6012 }
6013
6014 /*
6015  * This is a copy of file_update_time.  We need this so we can return error on
6016  * ENOSPC for updating the inode in the case of file write and mmap writes.
6017  */
6018 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6019                              int flags)
6020 {
6021         struct btrfs_root *root = BTRFS_I(inode)->root;
6022         bool dirty = flags & ~S_VERSION;
6023
6024         if (btrfs_root_readonly(root))
6025                 return -EROFS;
6026
6027         if (flags & S_VERSION)
6028                 dirty |= inode_maybe_inc_iversion(inode, dirty);
6029         if (flags & S_CTIME)
6030                 inode->i_ctime = *now;
6031         if (flags & S_MTIME)
6032                 inode->i_mtime = *now;
6033         if (flags & S_ATIME)
6034                 inode->i_atime = *now;
6035         return dirty ? btrfs_dirty_inode(inode) : 0;
6036 }
6037
6038 /*
6039  * find the highest existing sequence number in a directory
6040  * and then set the in-memory index_cnt variable to reflect
6041  * free sequence numbers
6042  */
6043 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6044 {
6045         struct btrfs_root *root = inode->root;
6046         struct btrfs_key key, found_key;
6047         struct btrfs_path *path;
6048         struct extent_buffer *leaf;
6049         int ret;
6050
6051         key.objectid = btrfs_ino(inode);
6052         key.type = BTRFS_DIR_INDEX_KEY;
6053         key.offset = (u64)-1;
6054
6055         path = btrfs_alloc_path();
6056         if (!path)
6057                 return -ENOMEM;
6058
6059         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6060         if (ret < 0)
6061                 goto out;
6062         /* FIXME: we should be able to handle this */
6063         if (ret == 0)
6064                 goto out;
6065         ret = 0;
6066
6067         /*
6068          * MAGIC NUMBER EXPLANATION:
6069          * since we search a directory based on f_pos we have to start at 2
6070          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6071          * else has to start at 2
6072          */
6073         if (path->slots[0] == 0) {
6074                 inode->index_cnt = 2;
6075                 goto out;
6076         }
6077
6078         path->slots[0]--;
6079
6080         leaf = path->nodes[0];
6081         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6082
6083         if (found_key.objectid != btrfs_ino(inode) ||
6084             found_key.type != BTRFS_DIR_INDEX_KEY) {
6085                 inode->index_cnt = 2;
6086                 goto out;
6087         }
6088
6089         inode->index_cnt = found_key.offset + 1;
6090 out:
6091         btrfs_free_path(path);
6092         return ret;
6093 }
6094
6095 /*
6096  * helper to find a free sequence number in a given directory.  This current
6097  * code is very simple, later versions will do smarter things in the btree
6098  */
6099 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6100 {
6101         int ret = 0;
6102
6103         if (dir->index_cnt == (u64)-1) {
6104                 ret = btrfs_inode_delayed_dir_index_count(dir);
6105                 if (ret) {
6106                         ret = btrfs_set_inode_index_count(dir);
6107                         if (ret)
6108                                 return ret;
6109                 }
6110         }
6111
6112         *index = dir->index_cnt;
6113         dir->index_cnt++;
6114
6115         return ret;
6116 }
6117
6118 static int btrfs_insert_inode_locked(struct inode *inode)
6119 {
6120         struct btrfs_iget_args args;
6121         args.location = &BTRFS_I(inode)->location;
6122         args.root = BTRFS_I(inode)->root;
6123
6124         return insert_inode_locked4(inode,
6125                    btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6126                    btrfs_find_actor, &args);
6127 }
6128
6129 /*
6130  * Inherit flags from the parent inode.
6131  *
6132  * Currently only the compression flags and the cow flags are inherited.
6133  */
6134 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6135 {
6136         unsigned int flags;
6137
6138         if (!dir)
6139                 return;
6140
6141         flags = BTRFS_I(dir)->flags;
6142
6143         if (flags & BTRFS_INODE_NOCOMPRESS) {
6144                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6145                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6146         } else if (flags & BTRFS_INODE_COMPRESS) {
6147                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6148                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6149         }
6150
6151         if (flags & BTRFS_INODE_NODATACOW) {
6152                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6153                 if (S_ISREG(inode->i_mode))
6154                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6155         }
6156
6157         btrfs_sync_inode_flags_to_i_flags(inode);
6158 }
6159
6160 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6161                                      struct btrfs_root *root,
6162                                      struct inode *dir,
6163                                      const char *name, int name_len,
6164                                      u64 ref_objectid, u64 objectid,
6165                                      umode_t mode, u64 *index)
6166 {
6167         struct btrfs_fs_info *fs_info = root->fs_info;
6168         struct inode *inode;
6169         struct btrfs_inode_item *inode_item;
6170         struct btrfs_key *location;
6171         struct btrfs_path *path;
6172         struct btrfs_inode_ref *ref;
6173         struct btrfs_key key[2];
6174         u32 sizes[2];
6175         int nitems = name ? 2 : 1;
6176         unsigned long ptr;
6177         int ret;
6178
6179         path = btrfs_alloc_path();
6180         if (!path)
6181                 return ERR_PTR(-ENOMEM);
6182
6183         inode = new_inode(fs_info->sb);
6184         if (!inode) {
6185                 btrfs_free_path(path);
6186                 return ERR_PTR(-ENOMEM);
6187         }
6188
6189         /*
6190          * O_TMPFILE, set link count to 0, so that after this point,
6191          * we fill in an inode item with the correct link count.
6192          */
6193         if (!name)
6194                 set_nlink(inode, 0);
6195
6196         /*
6197          * we have to initialize this early, so we can reclaim the inode
6198          * number if we fail afterwards in this function.
6199          */
6200         inode->i_ino = objectid;
6201
6202         if (dir && name) {
6203                 trace_btrfs_inode_request(dir);
6204
6205                 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6206                 if (ret) {
6207                         btrfs_free_path(path);
6208                         iput(inode);
6209                         return ERR_PTR(ret);
6210                 }
6211         } else if (dir) {
6212                 *index = 0;
6213         }
6214         /*
6215          * index_cnt is ignored for everything but a dir,
6216          * btrfs_set_inode_index_count has an explanation for the magic
6217          * number
6218          */
6219         BTRFS_I(inode)->index_cnt = 2;
6220         BTRFS_I(inode)->dir_index = *index;
6221         BTRFS_I(inode)->root = root;
6222         BTRFS_I(inode)->generation = trans->transid;
6223         inode->i_generation = BTRFS_I(inode)->generation;
6224
6225         /*
6226          * We could have gotten an inode number from somebody who was fsynced
6227          * and then removed in this same transaction, so let's just set full
6228          * sync since it will be a full sync anyway and this will blow away the
6229          * old info in the log.
6230          */
6231         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6232
6233         key[0].objectid = objectid;
6234         key[0].type = BTRFS_INODE_ITEM_KEY;
6235         key[0].offset = 0;
6236
6237         sizes[0] = sizeof(struct btrfs_inode_item);
6238
6239         if (name) {
6240                 /*
6241                  * Start new inodes with an inode_ref. This is slightly more
6242                  * efficient for small numbers of hard links since they will
6243                  * be packed into one item. Extended refs will kick in if we
6244                  * add more hard links than can fit in the ref item.
6245                  */
6246                 key[1].objectid = objectid;
6247                 key[1].type = BTRFS_INODE_REF_KEY;
6248                 key[1].offset = ref_objectid;
6249
6250                 sizes[1] = name_len + sizeof(*ref);
6251         }
6252
6253         location = &BTRFS_I(inode)->location;
6254         location->objectid = objectid;
6255         location->offset = 0;
6256         location->type = BTRFS_INODE_ITEM_KEY;
6257
6258         ret = btrfs_insert_inode_locked(inode);
6259         if (ret < 0) {
6260                 iput(inode);
6261                 goto fail;
6262         }
6263
6264         path->leave_spinning = 1;
6265         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6266         if (ret != 0)
6267                 goto fail_unlock;
6268
6269         inode_init_owner(inode, dir, mode);
6270         inode_set_bytes(inode, 0);
6271
6272         inode->i_mtime = current_time(inode);
6273         inode->i_atime = inode->i_mtime;
6274         inode->i_ctime = inode->i_mtime;
6275         BTRFS_I(inode)->i_otime = inode->i_mtime;
6276
6277         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6278                                   struct btrfs_inode_item);
6279         memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6280                              sizeof(*inode_item));
6281         fill_inode_item(trans, path->nodes[0], inode_item, inode);
6282
6283         if (name) {
6284                 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6285                                      struct btrfs_inode_ref);
6286                 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6287                 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6288                 ptr = (unsigned long)(ref + 1);
6289                 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6290         }
6291
6292         btrfs_mark_buffer_dirty(path->nodes[0]);
6293         btrfs_free_path(path);
6294
6295         btrfs_inherit_iflags(inode, dir);
6296
6297         if (S_ISREG(mode)) {
6298                 if (btrfs_test_opt(fs_info, NODATASUM))
6299                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6300                 if (btrfs_test_opt(fs_info, NODATACOW))
6301                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6302                                 BTRFS_INODE_NODATASUM;
6303         }
6304
6305         inode_tree_add(inode);
6306
6307         trace_btrfs_inode_new(inode);
6308         btrfs_set_inode_last_trans(trans, inode);
6309
6310         btrfs_update_root_times(trans, root);
6311
6312         ret = btrfs_inode_inherit_props(trans, inode, dir);
6313         if (ret)
6314                 btrfs_err(fs_info,
6315                           "error inheriting props for ino %llu (root %llu): %d",
6316                         btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6317
6318         return inode;
6319
6320 fail_unlock:
6321         discard_new_inode(inode);
6322 fail:
6323         if (dir && name)
6324                 BTRFS_I(dir)->index_cnt--;
6325         btrfs_free_path(path);
6326         return ERR_PTR(ret);
6327 }
6328
6329 static inline u8 btrfs_inode_type(struct inode *inode)
6330 {
6331         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6332 }
6333
6334 /*
6335  * utility function to add 'inode' into 'parent_inode' with
6336  * a give name and a given sequence number.
6337  * if 'add_backref' is true, also insert a backref from the
6338  * inode to the parent directory.
6339  */
6340 int btrfs_add_link(struct btrfs_trans_handle *trans,
6341                    struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6342                    const char *name, int name_len, int add_backref, u64 index)
6343 {
6344         int ret = 0;
6345         struct btrfs_key key;
6346         struct btrfs_root *root = parent_inode->root;
6347         u64 ino = btrfs_ino(inode);
6348         u64 parent_ino = btrfs_ino(parent_inode);
6349
6350         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6351                 memcpy(&key, &inode->root->root_key, sizeof(key));
6352         } else {
6353                 key.objectid = ino;
6354                 key.type = BTRFS_INODE_ITEM_KEY;
6355                 key.offset = 0;
6356         }
6357
6358         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6359                 ret = btrfs_add_root_ref(trans, key.objectid,
6360                                          root->root_key.objectid, parent_ino,
6361                                          index, name, name_len);
6362         } else if (add_backref) {
6363                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6364                                              parent_ino, index);
6365         }
6366
6367         /* Nothing to clean up yet */
6368         if (ret)
6369                 return ret;
6370
6371         ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6372                                     btrfs_inode_type(&inode->vfs_inode), index);
6373         if (ret == -EEXIST || ret == -EOVERFLOW)
6374                 goto fail_dir_item;
6375         else if (ret) {
6376                 btrfs_abort_transaction(trans, ret);
6377                 return ret;
6378         }
6379
6380         btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6381                            name_len * 2);
6382         inode_inc_iversion(&parent_inode->vfs_inode);
6383         parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6384                 current_time(&parent_inode->vfs_inode);
6385         ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6386         if (ret)
6387                 btrfs_abort_transaction(trans, ret);
6388         return ret;
6389
6390 fail_dir_item:
6391         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6392                 u64 local_index;
6393                 int err;
6394                 err = btrfs_del_root_ref(trans, key.objectid,
6395                                          root->root_key.objectid, parent_ino,
6396                                          &local_index, name, name_len);
6397
6398         } else if (add_backref) {
6399                 u64 local_index;
6400                 int err;
6401
6402                 err = btrfs_del_inode_ref(trans, root, name, name_len,
6403                                           ino, parent_ino, &local_index);
6404         }
6405         return ret;
6406 }
6407
6408 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6409                             struct btrfs_inode *dir, struct dentry *dentry,
6410                             struct btrfs_inode *inode, int backref, u64 index)
6411 {
6412         int err = btrfs_add_link(trans, dir, inode,
6413                                  dentry->d_name.name, dentry->d_name.len,
6414                                  backref, index);
6415         if (err > 0)
6416                 err = -EEXIST;
6417         return err;
6418 }
6419
6420 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6421                         umode_t mode, dev_t rdev)
6422 {
6423         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6424         struct btrfs_trans_handle *trans;
6425         struct btrfs_root *root = BTRFS_I(dir)->root;
6426         struct inode *inode = NULL;
6427         int err;
6428         u64 objectid;
6429         u64 index = 0;
6430
6431         /*
6432          * 2 for inode item and ref
6433          * 2 for dir items
6434          * 1 for xattr if selinux is on
6435          */
6436         trans = btrfs_start_transaction(root, 5);
6437         if (IS_ERR(trans))
6438                 return PTR_ERR(trans);
6439
6440         err = btrfs_find_free_ino(root, &objectid);
6441         if (err)
6442                 goto out_unlock;
6443
6444         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6445                         dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6446                         mode, &index);
6447         if (IS_ERR(inode)) {
6448                 err = PTR_ERR(inode);
6449                 inode = NULL;
6450                 goto out_unlock;
6451         }
6452
6453         /*
6454         * If the active LSM wants to access the inode during
6455         * d_instantiate it needs these. Smack checks to see
6456         * if the filesystem supports xattrs by looking at the
6457         * ops vector.
6458         */
6459         inode->i_op = &btrfs_special_inode_operations;
6460         init_special_inode(inode, inode->i_mode, rdev);
6461
6462         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6463         if (err)
6464                 goto out_unlock;
6465
6466         err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6467                         0, index);
6468         if (err)
6469                 goto out_unlock;
6470
6471         btrfs_update_inode(trans, root, inode);
6472         d_instantiate_new(dentry, inode);
6473
6474 out_unlock:
6475         btrfs_end_transaction(trans);
6476         btrfs_btree_balance_dirty(fs_info);
6477         if (err && inode) {
6478                 inode_dec_link_count(inode);
6479                 discard_new_inode(inode);
6480         }
6481         return err;
6482 }
6483
6484 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6485                         umode_t mode, bool excl)
6486 {
6487         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6488         struct btrfs_trans_handle *trans;
6489         struct btrfs_root *root = BTRFS_I(dir)->root;
6490         struct inode *inode = NULL;
6491         int err;
6492         u64 objectid;
6493         u64 index = 0;
6494
6495         /*
6496          * 2 for inode item and ref
6497          * 2 for dir items
6498          * 1 for xattr if selinux is on
6499          */
6500         trans = btrfs_start_transaction(root, 5);
6501         if (IS_ERR(trans))
6502                 return PTR_ERR(trans);
6503
6504         err = btrfs_find_free_ino(root, &objectid);
6505         if (err)
6506                 goto out_unlock;
6507
6508         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6509                         dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6510                         mode, &index);
6511         if (IS_ERR(inode)) {
6512                 err = PTR_ERR(inode);
6513                 inode = NULL;
6514                 goto out_unlock;
6515         }
6516         /*
6517         * If the active LSM wants to access the inode during
6518         * d_instantiate it needs these. Smack checks to see
6519         * if the filesystem supports xattrs by looking at the
6520         * ops vector.
6521         */
6522         inode->i_fop = &btrfs_file_operations;
6523         inode->i_op = &btrfs_file_inode_operations;
6524         inode->i_mapping->a_ops = &btrfs_aops;
6525
6526         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6527         if (err)
6528                 goto out_unlock;
6529
6530         err = btrfs_update_inode(trans, root, inode);
6531         if (err)
6532                 goto out_unlock;
6533
6534         err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6535                         0, index);
6536         if (err)
6537                 goto out_unlock;
6538
6539         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6540         d_instantiate_new(dentry, inode);
6541
6542 out_unlock:
6543         btrfs_end_transaction(trans);
6544         if (err && inode) {
6545                 inode_dec_link_count(inode);
6546                 discard_new_inode(inode);
6547         }
6548         btrfs_btree_balance_dirty(fs_info);
6549         return err;
6550 }
6551
6552 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6553                       struct dentry *dentry)
6554 {
6555         struct btrfs_trans_handle *trans = NULL;
6556         struct btrfs_root *root = BTRFS_I(dir)->root;
6557         struct inode *inode = d_inode(old_dentry);
6558         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6559         u64 index;
6560         int err;
6561         int drop_inode = 0;
6562
6563         /* do not allow sys_link's with other subvols of the same device */
6564         if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6565                 return -EXDEV;
6566
6567         if (inode->i_nlink >= BTRFS_LINK_MAX)
6568                 return -EMLINK;
6569
6570         err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6571         if (err)
6572                 goto fail;
6573
6574         /*
6575          * 2 items for inode and inode ref
6576          * 2 items for dir items
6577          * 1 item for parent inode
6578          * 1 item for orphan item deletion if O_TMPFILE
6579          */
6580         trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6581         if (IS_ERR(trans)) {
6582                 err = PTR_ERR(trans);
6583                 trans = NULL;
6584                 goto fail;
6585         }
6586
6587         /* There are several dir indexes for this inode, clear the cache. */
6588         BTRFS_I(inode)->dir_index = 0ULL;
6589         inc_nlink(inode);
6590         inode_inc_iversion(inode);
6591         inode->i_ctime = current_time(inode);
6592         ihold(inode);
6593         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6594
6595         err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6596                         1, index);
6597
6598         if (err) {
6599                 drop_inode = 1;
6600         } else {
6601                 struct dentry *parent = dentry->d_parent;
6602                 int ret;
6603
6604                 err = btrfs_update_inode(trans, root, inode);
6605                 if (err)
6606                         goto fail;
6607                 if (inode->i_nlink == 1) {
6608                         /*
6609                          * If new hard link count is 1, it's a file created
6610                          * with open(2) O_TMPFILE flag.
6611                          */
6612                         err = btrfs_orphan_del(trans, BTRFS_I(inode));
6613                         if (err)
6614                                 goto fail;
6615                 }
6616                 BTRFS_I(inode)->last_link_trans = trans->transid;
6617                 d_instantiate(dentry, inode);
6618                 ret = btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent,
6619                                          true, NULL);
6620                 if (ret == BTRFS_NEED_TRANS_COMMIT) {
6621                         err = btrfs_commit_transaction(trans);
6622                         trans = NULL;
6623                 }
6624         }
6625
6626 fail:
6627         if (trans)
6628                 btrfs_end_transaction(trans);
6629         if (drop_inode) {
6630                 inode_dec_link_count(inode);
6631                 iput(inode);
6632         }
6633         btrfs_btree_balance_dirty(fs_info);
6634         return err;
6635 }
6636
6637 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6638 {
6639         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6640         struct inode *inode = NULL;
6641         struct btrfs_trans_handle *trans;
6642         struct btrfs_root *root = BTRFS_I(dir)->root;
6643         int err = 0;
6644         u64 objectid = 0;
6645         u64 index = 0;
6646
6647         /*
6648          * 2 items for inode and ref
6649          * 2 items for dir items
6650          * 1 for xattr if selinux is on
6651          */
6652         trans = btrfs_start_transaction(root, 5);
6653         if (IS_ERR(trans))
6654                 return PTR_ERR(trans);
6655
6656         err = btrfs_find_free_ino(root, &objectid);
6657         if (err)
6658                 goto out_fail;
6659
6660         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6661                         dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6662                         S_IFDIR | mode, &index);
6663         if (IS_ERR(inode)) {
6664                 err = PTR_ERR(inode);
6665                 inode = NULL;
6666                 goto out_fail;
6667         }
6668
6669         /* these must be set before we unlock the inode */
6670         inode->i_op = &btrfs_dir_inode_operations;
6671         inode->i_fop = &btrfs_dir_file_operations;
6672
6673         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6674         if (err)
6675                 goto out_fail;
6676
6677         btrfs_i_size_write(BTRFS_I(inode), 0);
6678         err = btrfs_update_inode(trans, root, inode);
6679         if (err)
6680                 goto out_fail;
6681
6682         err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6683                         dentry->d_name.name,
6684                         dentry->d_name.len, 0, index);
6685         if (err)
6686                 goto out_fail;
6687
6688         d_instantiate_new(dentry, inode);
6689
6690 out_fail:
6691         btrfs_end_transaction(trans);
6692         if (err && inode) {
6693                 inode_dec_link_count(inode);
6694                 discard_new_inode(inode);
6695         }
6696         btrfs_btree_balance_dirty(fs_info);
6697         return err;
6698 }
6699
6700 static noinline int uncompress_inline(struct btrfs_path *path,
6701                                       struct page *page,
6702                                       size_t pg_offset, u64 extent_offset,
6703                                       struct btrfs_file_extent_item *item)
6704 {
6705         int ret;
6706         struct extent_buffer *leaf = path->nodes[0];
6707         char *tmp;
6708         size_t max_size;
6709         unsigned long inline_size;
6710         unsigned long ptr;
6711         int compress_type;
6712
6713         WARN_ON(pg_offset != 0);
6714         compress_type = btrfs_file_extent_compression(leaf, item);
6715         max_size = btrfs_file_extent_ram_bytes(leaf, item);
6716         inline_size = btrfs_file_extent_inline_item_len(leaf,
6717                                         btrfs_item_nr(path->slots[0]));
6718         tmp = kmalloc(inline_size, GFP_NOFS);
6719         if (!tmp)
6720                 return -ENOMEM;
6721         ptr = btrfs_file_extent_inline_start(item);
6722
6723         read_extent_buffer(leaf, tmp, ptr, inline_size);
6724
6725         max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6726         ret = btrfs_decompress(compress_type, tmp, page,
6727                                extent_offset, inline_size, max_size);
6728
6729         /*
6730          * decompression code contains a memset to fill in any space between the end
6731          * of the uncompressed data and the end of max_size in case the decompressed
6732          * data ends up shorter than ram_bytes.  That doesn't cover the hole between
6733          * the end of an inline extent and the beginning of the next block, so we
6734          * cover that region here.
6735          */
6736
6737         if (max_size + pg_offset < PAGE_SIZE) {
6738                 char *map = kmap(page);
6739                 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6740                 kunmap(page);
6741         }
6742         kfree(tmp);
6743         return ret;
6744 }
6745
6746 /*
6747  * a bit scary, this does extent mapping from logical file offset to the disk.
6748  * the ugly parts come from merging extents from the disk with the in-ram
6749  * representation.  This gets more complex because of the data=ordered code,
6750  * where the in-ram extents might be locked pending data=ordered completion.
6751  *
6752  * This also copies inline extents directly into the page.
6753  */
6754 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6755                                     struct page *page,
6756                                     size_t pg_offset, u64 start, u64 len,
6757                                     int create)
6758 {
6759         struct btrfs_fs_info *fs_info = inode->root->fs_info;
6760         int ret;
6761         int err = 0;
6762         u64 extent_start = 0;
6763         u64 extent_end = 0;
6764         u64 objectid = btrfs_ino(inode);
6765         u32 found_type;
6766         struct btrfs_path *path = NULL;
6767         struct btrfs_root *root = inode->root;
6768         struct btrfs_file_extent_item *item;
6769         struct extent_buffer *leaf;
6770         struct btrfs_key found_key;
6771         struct extent_map *em = NULL;
6772         struct extent_map_tree *em_tree = &inode->extent_tree;
6773         struct extent_io_tree *io_tree = &inode->io_tree;
6774         const bool new_inline = !page || create;
6775
6776         read_lock(&em_tree->lock);
6777         em = lookup_extent_mapping(em_tree, start, len);
6778         if (em)
6779                 em->bdev = fs_info->fs_devices->latest_bdev;
6780         read_unlock(&em_tree->lock);
6781
6782         if (em) {
6783                 if (em->start > start || em->start + em->len <= start)
6784                         free_extent_map(em);
6785                 else if (em->block_start == EXTENT_MAP_INLINE && page)
6786                         free_extent_map(em);
6787                 else
6788                         goto out;
6789         }
6790         em = alloc_extent_map();
6791         if (!em) {
6792                 err = -ENOMEM;
6793                 goto out;
6794         }
6795         em->bdev = fs_info->fs_devices->latest_bdev;
6796         em->start = EXTENT_MAP_HOLE;
6797         em->orig_start = EXTENT_MAP_HOLE;
6798         em->len = (u64)-1;
6799         em->block_len = (u64)-1;
6800
6801         path = btrfs_alloc_path();
6802         if (!path) {
6803                 err = -ENOMEM;
6804                 goto out;
6805         }
6806
6807         /* Chances are we'll be called again, so go ahead and do readahead */
6808         path->reada = READA_FORWARD;
6809
6810         /*
6811          * Unless we're going to uncompress the inline extent, no sleep would
6812          * happen.
6813          */
6814         path->leave_spinning = 1;
6815
6816         ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6817         if (ret < 0) {
6818                 err = ret;
6819                 goto out;
6820         }
6821
6822         if (ret != 0) {
6823                 if (path->slots[0] == 0)
6824                         goto not_found;
6825                 path->slots[0]--;
6826         }
6827
6828         leaf = path->nodes[0];
6829         item = btrfs_item_ptr(leaf, path->slots[0],
6830                               struct btrfs_file_extent_item);
6831         /* are we inside the extent that was found? */
6832         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6833         found_type = found_key.type;
6834         if (found_key.objectid != objectid ||
6835             found_type != BTRFS_EXTENT_DATA_KEY) {
6836                 /*
6837                  * If we backup past the first extent we want to move forward
6838                  * and see if there is an extent in front of us, otherwise we'll
6839                  * say there is a hole for our whole search range which can
6840                  * cause problems.
6841                  */
6842                 extent_end = start;
6843                 goto next;
6844         }
6845
6846         found_type = btrfs_file_extent_type(leaf, item);
6847         extent_start = found_key.offset;
6848         if (found_type == BTRFS_FILE_EXTENT_REG ||
6849             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6850                 extent_end = extent_start +
6851                        btrfs_file_extent_num_bytes(leaf, item);
6852
6853                 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6854                                                        extent_start);
6855         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6856                 size_t size;
6857
6858                 size = btrfs_file_extent_ram_bytes(leaf, item);
6859                 extent_end = ALIGN(extent_start + size,
6860                                    fs_info->sectorsize);
6861
6862                 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6863                                                       path->slots[0],
6864                                                       extent_start);
6865         }
6866 next:
6867         if (start >= extent_end) {
6868                 path->slots[0]++;
6869                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6870                         ret = btrfs_next_leaf(root, path);
6871                         if (ret < 0) {
6872                                 err = ret;
6873                                 goto out;
6874                         }
6875                         if (ret > 0)
6876                                 goto not_found;
6877                         leaf = path->nodes[0];
6878                 }
6879                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6880                 if (found_key.objectid != objectid ||
6881                     found_key.type != BTRFS_EXTENT_DATA_KEY)
6882                         goto not_found;
6883                 if (start + len <= found_key.offset)
6884                         goto not_found;
6885                 if (start > found_key.offset)
6886                         goto next;
6887                 em->start = start;
6888                 em->orig_start = start;
6889                 em->len = found_key.offset - start;
6890                 goto not_found_em;
6891         }
6892
6893         btrfs_extent_item_to_extent_map(inode, path, item,
6894                         new_inline, em);
6895
6896         if (found_type == BTRFS_FILE_EXTENT_REG ||
6897             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6898                 goto insert;
6899         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6900                 unsigned long ptr;
6901                 char *map;
6902                 size_t size;
6903                 size_t extent_offset;
6904                 size_t copy_size;
6905
6906                 if (new_inline)
6907                         goto out;
6908
6909                 size = btrfs_file_extent_ram_bytes(leaf, item);
6910                 extent_offset = page_offset(page) + pg_offset - extent_start;
6911                 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6912                                   size - extent_offset);
6913                 em->start = extent_start + extent_offset;
6914                 em->len = ALIGN(copy_size, fs_info->sectorsize);
6915                 em->orig_block_len = em->len;
6916                 em->orig_start = em->start;
6917                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6918
6919                 btrfs_set_path_blocking(path);
6920                 if (!PageUptodate(page)) {
6921                         if (btrfs_file_extent_compression(leaf, item) !=
6922                             BTRFS_COMPRESS_NONE) {
6923                                 ret = uncompress_inline(path, page, pg_offset,
6924                                                         extent_offset, item);
6925                                 if (ret) {
6926                                         err = ret;
6927                                         goto out;
6928                                 }
6929                         } else {
6930                                 map = kmap(page);
6931                                 read_extent_buffer(leaf, map + pg_offset, ptr,
6932                                                    copy_size);
6933                                 if (pg_offset + copy_size < PAGE_SIZE) {
6934                                         memset(map + pg_offset + copy_size, 0,
6935                                                PAGE_SIZE - pg_offset -
6936                                                copy_size);
6937                                 }
6938                                 kunmap(page);
6939                         }
6940                         flush_dcache_page(page);
6941                 }
6942                 set_extent_uptodate(io_tree, em->start,
6943                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
6944                 goto insert;
6945         }
6946 not_found:
6947         em->start = start;
6948         em->orig_start = start;
6949         em->len = len;
6950 not_found_em:
6951         em->block_start = EXTENT_MAP_HOLE;
6952 insert:
6953         btrfs_release_path(path);
6954         if (em->start > start || extent_map_end(em) <= start) {
6955                 btrfs_err(fs_info,
6956                           "bad extent! em: [%llu %llu] passed [%llu %llu]",
6957                           em->start, em->len, start, len);
6958                 err = -EIO;
6959                 goto out;
6960         }
6961
6962         err = 0;
6963         write_lock(&em_tree->lock);
6964         err = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6965         write_unlock(&em_tree->lock);
6966 out:
6967         btrfs_free_path(path);
6968
6969         trace_btrfs_get_extent(root, inode, em);
6970
6971         if (err) {
6972                 free_extent_map(em);
6973                 return ERR_PTR(err);
6974         }
6975         BUG_ON(!em); /* Error is always set */
6976         return em;
6977 }
6978
6979 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
6980                 struct page *page,
6981                 size_t pg_offset, u64 start, u64 len,
6982                 int create)
6983 {
6984         struct extent_map *em;
6985         struct extent_map *hole_em = NULL;
6986         u64 range_start = start;
6987         u64 end;
6988         u64 found;
6989         u64 found_end;
6990         int err = 0;
6991
6992         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6993         if (IS_ERR(em))
6994                 return em;
6995         /*
6996          * If our em maps to:
6997          * - a hole or
6998          * - a pre-alloc extent,
6999          * there might actually be delalloc bytes behind it.
7000          */
7001         if (em->block_start != EXTENT_MAP_HOLE &&
7002             !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7003                 return em;
7004         else
7005                 hole_em = em;
7006
7007         /* check to see if we've wrapped (len == -1 or similar) */
7008         end = start + len;
7009         if (end < start)
7010                 end = (u64)-1;
7011         else
7012                 end -= 1;
7013
7014         em = NULL;
7015
7016         /* ok, we didn't find anything, lets look for delalloc */
7017         found = count_range_bits(&inode->io_tree, &range_start,
7018                                  end, len, EXTENT_DELALLOC, 1);
7019         found_end = range_start + found;
7020         if (found_end < range_start)
7021                 found_end = (u64)-1;
7022
7023         /*
7024          * we didn't find anything useful, return
7025          * the original results from get_extent()
7026          */
7027         if (range_start > end || found_end <= start) {
7028                 em = hole_em;
7029                 hole_em = NULL;
7030                 goto out;
7031         }
7032
7033         /* adjust the range_start to make sure it doesn't
7034          * go backwards from the start they passed in
7035          */
7036         range_start = max(start, range_start);
7037         found = found_end - range_start;
7038
7039         if (found > 0) {
7040                 u64 hole_start = start;
7041                 u64 hole_len = len;
7042
7043                 em = alloc_extent_map();
7044                 if (!em) {
7045                         err = -ENOMEM;
7046                         goto out;
7047                 }
7048                 /*
7049                  * when btrfs_get_extent can't find anything it
7050                  * returns one huge hole
7051                  *
7052                  * make sure what it found really fits our range, and
7053                  * adjust to make sure it is based on the start from
7054                  * the caller
7055                  */
7056                 if (hole_em) {
7057                         u64 calc_end = extent_map_end(hole_em);
7058
7059                         if (calc_end <= start || (hole_em->start > end)) {
7060                                 free_extent_map(hole_em);
7061                                 hole_em = NULL;
7062                         } else {
7063                                 hole_start = max(hole_em->start, start);
7064                                 hole_len = calc_end - hole_start;
7065                         }
7066                 }
7067                 em->bdev = NULL;
7068                 if (hole_em && range_start > hole_start) {
7069                         /* our hole starts before our delalloc, so we
7070                          * have to return just the parts of the hole
7071                          * that go until  the delalloc starts
7072                          */
7073                         em->len = min(hole_len,
7074                                       range_start - hole_start);
7075                         em->start = hole_start;
7076                         em->orig_start = hole_start;
7077                         /*
7078                          * don't adjust block start at all,
7079                          * it is fixed at EXTENT_MAP_HOLE
7080                          */
7081                         em->block_start = hole_em->block_start;
7082                         em->block_len = hole_len;
7083                         if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7084                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7085                 } else {
7086                         em->start = range_start;
7087                         em->len = found;
7088                         em->orig_start = range_start;
7089                         em->block_start = EXTENT_MAP_DELALLOC;
7090                         em->block_len = found;
7091                 }
7092         } else {
7093                 return hole_em;
7094         }
7095 out:
7096
7097         free_extent_map(hole_em);
7098         if (err) {
7099                 free_extent_map(em);
7100                 return ERR_PTR(err);
7101         }
7102         return em;
7103 }
7104
7105 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7106                                                   const u64 start,
7107                                                   const u64 len,
7108                                                   const u64 orig_start,
7109                                                   const u64 block_start,
7110                                                   const u64 block_len,
7111                                                   const u64 orig_block_len,
7112                                                   const u64 ram_bytes,
7113                                                   const int type)
7114 {
7115         struct extent_map *em = NULL;
7116         int ret;
7117
7118         if (type != BTRFS_ORDERED_NOCOW) {
7119                 em = create_io_em(inode, start, len, orig_start,
7120                                   block_start, block_len, orig_block_len,
7121                                   ram_bytes,
7122                                   BTRFS_COMPRESS_NONE, /* compress_type */
7123                                   type);
7124                 if (IS_ERR(em))
7125                         goto out;
7126         }
7127         ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7128                                            len, block_len, type);
7129         if (ret) {
7130                 if (em) {
7131                         free_extent_map(em);
7132                         btrfs_drop_extent_cache(BTRFS_I(inode), start,
7133                                                 start + len - 1, 0);
7134                 }
7135                 em = ERR_PTR(ret);
7136         }
7137  out:
7138
7139         return em;
7140 }
7141
7142 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7143                                                   u64 start, u64 len)
7144 {
7145         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7146         struct btrfs_root *root = BTRFS_I(inode)->root;
7147         struct extent_map *em;
7148         struct btrfs_key ins;
7149         u64 alloc_hint;
7150         int ret;
7151
7152         alloc_hint = get_extent_allocation_hint(inode, start, len);
7153         ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7154                                    0, alloc_hint, &ins, 1, 1);
7155         if (ret)
7156                 return ERR_PTR(ret);
7157
7158         em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7159                                      ins.objectid, ins.offset, ins.offset,
7160                                      ins.offset, BTRFS_ORDERED_REGULAR);
7161         btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7162         if (IS_ERR(em))
7163                 btrfs_free_reserved_extent(fs_info, ins.objectid,
7164                                            ins.offset, 1);
7165
7166         return em;
7167 }
7168
7169 /*
7170  * returns 1 when the nocow is safe, < 1 on error, 0 if the
7171  * block must be cow'd
7172  */
7173 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7174                               u64 *orig_start, u64 *orig_block_len,
7175                               u64 *ram_bytes)
7176 {
7177         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7178         struct btrfs_path *path;
7179         int ret;
7180         struct extent_buffer *leaf;
7181         struct btrfs_root *root = BTRFS_I(inode)->root;
7182         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7183         struct btrfs_file_extent_item *fi;
7184         struct btrfs_key key;
7185         u64 disk_bytenr;
7186         u64 backref_offset;
7187         u64 extent_end;
7188         u64 num_bytes;
7189         int slot;
7190         int found_type;
7191         bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7192
7193         path = btrfs_alloc_path();
7194         if (!path)
7195                 return -ENOMEM;
7196
7197         ret = btrfs_lookup_file_extent(NULL, root, path,
7198                         btrfs_ino(BTRFS_I(inode)), offset, 0);
7199         if (ret < 0)
7200                 goto out;
7201
7202         slot = path->slots[0];
7203         if (ret == 1) {
7204                 if (slot == 0) {
7205                         /* can't find the item, must cow */
7206                         ret = 0;
7207                         goto out;
7208                 }
7209                 slot--;
7210         }
7211         ret = 0;
7212         leaf = path->nodes[0];
7213         btrfs_item_key_to_cpu(leaf, &key, slot);
7214         if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7215             key.type != BTRFS_EXTENT_DATA_KEY) {
7216                 /* not our file or wrong item type, must cow */
7217                 goto out;
7218         }
7219
7220         if (key.offset > offset) {
7221                 /* Wrong offset, must cow */
7222                 goto out;
7223         }
7224
7225         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7226         found_type = btrfs_file_extent_type(leaf, fi);
7227         if (found_type != BTRFS_FILE_EXTENT_REG &&
7228             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7229                 /* not a regular extent, must cow */
7230                 goto out;
7231         }
7232
7233         if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7234                 goto out;
7235
7236         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7237         if (extent_end <= offset)
7238                 goto out;
7239
7240         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7241         if (disk_bytenr == 0)
7242                 goto out;
7243
7244         if (btrfs_file_extent_compression(leaf, fi) ||
7245             btrfs_file_extent_encryption(leaf, fi) ||
7246             btrfs_file_extent_other_encoding(leaf, fi))
7247                 goto out;
7248
7249         /*
7250          * Do the same check as in btrfs_cross_ref_exist but without the
7251          * unnecessary search.
7252          */
7253         if (btrfs_file_extent_generation(leaf, fi) <=
7254             btrfs_root_last_snapshot(&root->root_item))
7255                 goto out;
7256
7257         backref_offset = btrfs_file_extent_offset(leaf, fi);
7258
7259         if (orig_start) {
7260                 *orig_start = key.offset - backref_offset;
7261                 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7262                 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7263         }
7264
7265         if (btrfs_extent_readonly(fs_info, disk_bytenr))
7266                 goto out;
7267
7268         num_bytes = min(offset + *len, extent_end) - offset;
7269         if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7270                 u64 range_end;
7271
7272                 range_end = round_up(offset + num_bytes,
7273                                      root->fs_info->sectorsize) - 1;
7274                 ret = test_range_bit(io_tree, offset, range_end,
7275                                      EXTENT_DELALLOC, 0, NULL);
7276                 if (ret) {
7277                         ret = -EAGAIN;
7278                         goto out;
7279                 }
7280         }
7281
7282         btrfs_release_path(path);
7283
7284         /*
7285          * look for other files referencing this extent, if we
7286          * find any we must cow
7287          */
7288
7289         ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7290                                     key.offset - backref_offset, disk_bytenr);
7291         if (ret) {
7292                 ret = 0;
7293                 goto out;
7294         }
7295
7296         /*
7297          * adjust disk_bytenr and num_bytes to cover just the bytes
7298          * in this extent we are about to write.  If there
7299          * are any csums in that range we have to cow in order
7300          * to keep the csums correct
7301          */
7302         disk_bytenr += backref_offset;
7303         disk_bytenr += offset - key.offset;
7304         if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7305                 goto out;
7306         /*
7307          * all of the above have passed, it is safe to overwrite this extent
7308          * without cow
7309          */
7310         *len = num_bytes;
7311         ret = 1;
7312 out:
7313         btrfs_free_path(path);
7314         return ret;
7315 }
7316
7317 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7318                               struct extent_state **cached_state, int writing)
7319 {
7320         struct btrfs_ordered_extent *ordered;
7321         int ret = 0;
7322
7323         while (1) {
7324                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7325                                  cached_state);
7326                 /*
7327                  * We're concerned with the entire range that we're going to be
7328                  * doing DIO to, so we need to make sure there's no ordered
7329                  * extents in this range.
7330                  */
7331                 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7332                                                      lockend - lockstart + 1);
7333
7334                 /*
7335                  * We need to make sure there are no buffered pages in this
7336                  * range either, we could have raced between the invalidate in
7337                  * generic_file_direct_write and locking the extent.  The
7338                  * invalidate needs to happen so that reads after a write do not
7339                  * get stale data.
7340                  */
7341                 if (!ordered &&
7342                     (!writing || !filemap_range_has_page(inode->i_mapping,
7343                                                          lockstart, lockend)))
7344                         break;
7345
7346                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7347                                      cached_state);
7348
7349                 if (ordered) {
7350                         /*
7351                          * If we are doing a DIO read and the ordered extent we
7352                          * found is for a buffered write, we can not wait for it
7353                          * to complete and retry, because if we do so we can
7354                          * deadlock with concurrent buffered writes on page
7355                          * locks. This happens only if our DIO read covers more
7356                          * than one extent map, if at this point has already
7357                          * created an ordered extent for a previous extent map
7358                          * and locked its range in the inode's io tree, and a
7359                          * concurrent write against that previous extent map's
7360                          * range and this range started (we unlock the ranges
7361                          * in the io tree only when the bios complete and
7362                          * buffered writes always lock pages before attempting
7363                          * to lock range in the io tree).
7364                          */
7365                         if (writing ||
7366                             test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7367                                 btrfs_start_ordered_extent(inode, ordered, 1);
7368                         else
7369                                 ret = -ENOTBLK;
7370                         btrfs_put_ordered_extent(ordered);
7371                 } else {
7372                         /*
7373                          * We could trigger writeback for this range (and wait
7374                          * for it to complete) and then invalidate the pages for
7375                          * this range (through invalidate_inode_pages2_range()),
7376                          * but that can lead us to a deadlock with a concurrent
7377                          * call to readpages() (a buffered read or a defrag call
7378                          * triggered a readahead) on a page lock due to an
7379                          * ordered dio extent we created before but did not have
7380                          * yet a corresponding bio submitted (whence it can not
7381                          * complete), which makes readpages() wait for that
7382                          * ordered extent to complete while holding a lock on
7383                          * that page.
7384                          */
7385                         ret = -ENOTBLK;
7386                 }
7387
7388                 if (ret)
7389                         break;
7390
7391                 cond_resched();
7392         }
7393
7394         return ret;
7395 }
7396
7397 /* The callers of this must take lock_extent() */
7398 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7399                                        u64 orig_start, u64 block_start,
7400                                        u64 block_len, u64 orig_block_len,
7401                                        u64 ram_bytes, int compress_type,
7402                                        int type)
7403 {
7404         struct extent_map_tree *em_tree;
7405         struct extent_map *em;
7406         struct btrfs_root *root = BTRFS_I(inode)->root;
7407         int ret;
7408
7409         ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7410                type == BTRFS_ORDERED_COMPRESSED ||
7411                type == BTRFS_ORDERED_NOCOW ||
7412                type == BTRFS_ORDERED_REGULAR);
7413
7414         em_tree = &BTRFS_I(inode)->extent_tree;
7415         em = alloc_extent_map();
7416         if (!em)
7417                 return ERR_PTR(-ENOMEM);
7418
7419         em->start = start;
7420         em->orig_start = orig_start;
7421         em->len = len;
7422         em->block_len = block_len;
7423         em->block_start = block_start;
7424         em->bdev = root->fs_info->fs_devices->latest_bdev;
7425         em->orig_block_len = orig_block_len;
7426         em->ram_bytes = ram_bytes;
7427         em->generation = -1;
7428         set_bit(EXTENT_FLAG_PINNED, &em->flags);
7429         if (type == BTRFS_ORDERED_PREALLOC) {
7430                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7431         } else if (type == BTRFS_ORDERED_COMPRESSED) {
7432                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7433                 em->compress_type = compress_type;
7434         }
7435
7436         do {
7437                 btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7438                                 em->start + em->len - 1, 0);
7439                 write_lock(&em_tree->lock);
7440                 ret = add_extent_mapping(em_tree, em, 1);
7441                 write_unlock(&em_tree->lock);
7442                 /*
7443                  * The caller has taken lock_extent(), who could race with us
7444                  * to add em?
7445                  */
7446         } while (ret == -EEXIST);
7447
7448         if (ret) {
7449                 free_extent_map(em);
7450                 return ERR_PTR(ret);
7451         }
7452
7453         /* em got 2 refs now, callers needs to do free_extent_map once. */
7454         return em;
7455 }
7456
7457
7458 static int btrfs_get_blocks_direct_read(struct extent_map *em,
7459                                         struct buffer_head *bh_result,
7460                                         struct inode *inode,
7461                                         u64 start, u64 len)
7462 {
7463         if (em->block_start == EXTENT_MAP_HOLE ||
7464                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7465                 return -ENOENT;
7466
7467         len = min(len, em->len - (start - em->start));
7468
7469         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7470                 inode->i_blkbits;
7471         bh_result->b_size = len;
7472         bh_result->b_bdev = em->bdev;
7473         set_buffer_mapped(bh_result);
7474
7475         return 0;
7476 }
7477
7478 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7479                                          struct buffer_head *bh_result,
7480                                          struct inode *inode,
7481                                          struct btrfs_dio_data *dio_data,
7482                                          u64 start, u64 len)
7483 {
7484         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7485         struct extent_map *em = *map;
7486         int ret = 0;
7487
7488         /*
7489          * We don't allocate a new extent in the following cases
7490          *
7491          * 1) The inode is marked as NODATACOW. In this case we'll just use the
7492          * existing extent.
7493          * 2) The extent is marked as PREALLOC. We're good to go here and can
7494          * just use the extent.
7495          *
7496          */
7497         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7498             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7499              em->block_start != EXTENT_MAP_HOLE)) {
7500                 int type;
7501                 u64 block_start, orig_start, orig_block_len, ram_bytes;
7502
7503                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7504                         type = BTRFS_ORDERED_PREALLOC;
7505                 else
7506                         type = BTRFS_ORDERED_NOCOW;
7507                 len = min(len, em->len - (start - em->start));
7508                 block_start = em->block_start + (start - em->start);
7509
7510                 if (can_nocow_extent(inode, start, &len, &orig_start,
7511                                      &orig_block_len, &ram_bytes) == 1 &&
7512                     btrfs_inc_nocow_writers(fs_info, block_start)) {
7513                         struct extent_map *em2;
7514
7515                         em2 = btrfs_create_dio_extent(inode, start, len,
7516                                                       orig_start, block_start,
7517                                                       len, orig_block_len,
7518                                                       ram_bytes, type);
7519                         btrfs_dec_nocow_writers(fs_info, block_start);
7520                         if (type == BTRFS_ORDERED_PREALLOC) {
7521                                 free_extent_map(em);
7522                                 *map = em = em2;
7523                         }
7524
7525                         if (em2 && IS_ERR(em2)) {
7526                                 ret = PTR_ERR(em2);
7527                                 goto out;
7528                         }
7529                         /*
7530                          * For inode marked NODATACOW or extent marked PREALLOC,
7531                          * use the existing or preallocated extent, so does not
7532                          * need to adjust btrfs_space_info's bytes_may_use.
7533                          */
7534                         btrfs_free_reserved_data_space_noquota(inode, start,
7535                                                                len);
7536                         goto skip_cow;
7537                 }
7538         }
7539
7540         /* this will cow the extent */
7541         len = bh_result->b_size;
7542         free_extent_map(em);
7543         *map = em = btrfs_new_extent_direct(inode, start, len);
7544         if (IS_ERR(em)) {
7545                 ret = PTR_ERR(em);
7546                 goto out;
7547         }
7548
7549         len = min(len, em->len - (start - em->start));
7550
7551 skip_cow:
7552         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7553                 inode->i_blkbits;
7554         bh_result->b_size = len;
7555         bh_result->b_bdev = em->bdev;
7556         set_buffer_mapped(bh_result);
7557
7558         if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7559                 set_buffer_new(bh_result);
7560
7561         /*
7562          * Need to update the i_size under the extent lock so buffered
7563          * readers will get the updated i_size when we unlock.
7564          */
7565         if (!dio_data->overwrite && start + len > i_size_read(inode))
7566                 i_size_write(inode, start + len);
7567
7568         WARN_ON(dio_data->reserve < len);
7569         dio_data->reserve -= len;
7570         dio_data->unsubmitted_oe_range_end = start + len;
7571         current->journal_info = dio_data;
7572 out:
7573         return ret;
7574 }
7575
7576 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7577                                    struct buffer_head *bh_result, int create)
7578 {
7579         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7580         struct extent_map *em;
7581         struct extent_state *cached_state = NULL;
7582         struct btrfs_dio_data *dio_data = NULL;
7583         u64 start = iblock << inode->i_blkbits;
7584         u64 lockstart, lockend;
7585         u64 len = bh_result->b_size;
7586         int unlock_bits = EXTENT_LOCKED;
7587         int ret = 0;
7588
7589         if (create)
7590                 unlock_bits |= EXTENT_DIRTY;
7591         else
7592                 len = min_t(u64, len, fs_info->sectorsize);
7593
7594         lockstart = start;
7595         lockend = start + len - 1;
7596
7597         if (current->journal_info) {
7598                 /*
7599                  * Need to pull our outstanding extents and set journal_info to NULL so
7600                  * that anything that needs to check if there's a transaction doesn't get
7601                  * confused.
7602                  */
7603                 dio_data = current->journal_info;
7604                 current->journal_info = NULL;
7605         }
7606
7607         /*
7608          * If this errors out it's because we couldn't invalidate pagecache for
7609          * this range and we need to fallback to buffered.
7610          */
7611         if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7612                                create)) {
7613                 ret = -ENOTBLK;
7614                 goto err;
7615         }
7616
7617         em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
7618         if (IS_ERR(em)) {
7619                 ret = PTR_ERR(em);
7620                 goto unlock_err;
7621         }
7622
7623         /*
7624          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7625          * io.  INLINE is special, and we could probably kludge it in here, but
7626          * it's still buffered so for safety lets just fall back to the generic
7627          * buffered path.
7628          *
7629          * For COMPRESSED we _have_ to read the entire extent in so we can
7630          * decompress it, so there will be buffering required no matter what we
7631          * do, so go ahead and fallback to buffered.
7632          *
7633          * We return -ENOTBLK because that's what makes DIO go ahead and go back
7634          * to buffered IO.  Don't blame me, this is the price we pay for using
7635          * the generic code.
7636          */
7637         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7638             em->block_start == EXTENT_MAP_INLINE) {
7639                 free_extent_map(em);
7640                 ret = -ENOTBLK;
7641                 goto unlock_err;
7642         }
7643
7644         if (create) {
7645                 ret = btrfs_get_blocks_direct_write(&em, bh_result, inode,
7646                                                     dio_data, start, len);
7647                 if (ret < 0)
7648                         goto unlock_err;
7649
7650                 /* clear and unlock the entire range */
7651                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7652                                  unlock_bits, 1, 0, &cached_state);
7653         } else {
7654                 ret = btrfs_get_blocks_direct_read(em, bh_result, inode,
7655                                                    start, len);
7656                 /* Can be negative only if we read from a hole */
7657                 if (ret < 0) {
7658                         ret = 0;
7659                         free_extent_map(em);
7660                         goto unlock_err;
7661                 }
7662                 /*
7663                  * We need to unlock only the end area that we aren't using.
7664                  * The rest is going to be unlocked by the endio routine.
7665                  */
7666                 lockstart = start + bh_result->b_size;
7667                 if (lockstart < lockend) {
7668                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7669                                          lockend, unlock_bits, 1, 0,
7670                                          &cached_state);
7671                 } else {
7672                         free_extent_state(cached_state);
7673                 }
7674         }
7675
7676         free_extent_map(em);
7677
7678         return 0;
7679
7680 unlock_err:
7681         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7682                          unlock_bits, 1, 0, &cached_state);
7683 err:
7684         if (dio_data)
7685                 current->journal_info = dio_data;
7686         return ret;
7687 }
7688
7689 static inline blk_status_t submit_dio_repair_bio(struct inode *inode,
7690                                                  struct bio *bio,
7691                                                  int mirror_num)
7692 {
7693         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7694         blk_status_t ret;
7695
7696         BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7697
7698         ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7699         if (ret)
7700                 return ret;
7701
7702         ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
7703
7704         return ret;
7705 }
7706
7707 static int btrfs_check_dio_repairable(struct inode *inode,
7708                                       struct bio *failed_bio,
7709                                       struct io_failure_record *failrec,
7710                                       int failed_mirror)
7711 {
7712         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7713         int num_copies;
7714
7715         num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7716         if (num_copies == 1) {
7717                 /*
7718                  * we only have a single copy of the data, so don't bother with
7719                  * all the retry and error correction code that follows. no
7720                  * matter what the error is, it is very likely to persist.
7721                  */
7722                 btrfs_debug(fs_info,
7723                         "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7724                         num_copies, failrec->this_mirror, failed_mirror);
7725                 return 0;
7726         }
7727
7728         failrec->failed_mirror = failed_mirror;
7729         failrec->this_mirror++;
7730         if (failrec->this_mirror == failed_mirror)
7731                 failrec->this_mirror++;
7732
7733         if (failrec->this_mirror > num_copies) {
7734                 btrfs_debug(fs_info,
7735                         "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7736                         num_copies, failrec->this_mirror, failed_mirror);
7737                 return 0;
7738         }
7739
7740         return 1;
7741 }
7742
7743 static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio,
7744                                    struct page *page, unsigned int pgoff,
7745                                    u64 start, u64 end, int failed_mirror,
7746                                    bio_end_io_t *repair_endio, void *repair_arg)
7747 {
7748         struct io_failure_record *failrec;
7749         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7750         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7751         struct bio *bio;
7752         int isector;
7753         unsigned int read_mode = 0;
7754         int segs;
7755         int ret;
7756         blk_status_t status;
7757         struct bio_vec bvec;
7758
7759         BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
7760
7761         ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7762         if (ret)
7763                 return errno_to_blk_status(ret);
7764
7765         ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7766                                          failed_mirror);
7767         if (!ret) {
7768                 free_io_failure(failure_tree, io_tree, failrec);
7769                 return BLK_STS_IOERR;
7770         }
7771
7772         segs = bio_segments(failed_bio);
7773         bio_get_first_bvec(failed_bio, &bvec);
7774         if (segs > 1 ||
7775             (bvec.bv_len > btrfs_inode_sectorsize(inode)))
7776                 read_mode |= REQ_FAILFAST_DEV;
7777
7778         isector = start - btrfs_io_bio(failed_bio)->logical;
7779         isector >>= inode->i_sb->s_blocksize_bits;
7780         bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7781                                 pgoff, isector, repair_endio, repair_arg);
7782         bio->bi_opf = REQ_OP_READ | read_mode;
7783
7784         btrfs_debug(BTRFS_I(inode)->root->fs_info,
7785                     "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d",
7786                     read_mode, failrec->this_mirror, failrec->in_validation);
7787
7788         status = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
7789         if (status) {
7790                 free_io_failure(failure_tree, io_tree, failrec);
7791                 bio_put(bio);
7792         }
7793
7794         return status;
7795 }
7796
7797 struct btrfs_retry_complete {
7798         struct completion done;
7799         struct inode *inode;
7800         u64 start;
7801         int uptodate;
7802 };
7803
7804 static void btrfs_retry_endio_nocsum(struct bio *bio)
7805 {
7806         struct btrfs_retry_complete *done = bio->bi_private;
7807         struct inode *inode = done->inode;
7808         struct bio_vec *bvec;
7809         struct extent_io_tree *io_tree, *failure_tree;
7810         int i;
7811
7812         if (bio->bi_status)
7813                 goto end;
7814
7815         ASSERT(bio->bi_vcnt == 1);
7816         io_tree = &BTRFS_I(inode)->io_tree;
7817         failure_tree = &BTRFS_I(inode)->io_failure_tree;
7818         ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(inode));
7819
7820         done->uptodate = 1;
7821         ASSERT(!bio_flagged(bio, BIO_CLONED));
7822         bio_for_each_segment_all(bvec, bio, i)
7823                 clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree,
7824                                  io_tree, done->start, bvec->bv_page,
7825                                  btrfs_ino(BTRFS_I(inode)), 0);
7826 end:
7827         complete(&done->done);
7828         bio_put(bio);
7829 }
7830
7831 static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode,
7832                                                 struct btrfs_io_bio *io_bio)
7833 {
7834         struct btrfs_fs_info *fs_info;
7835         struct bio_vec bvec;
7836         struct bvec_iter iter;
7837         struct btrfs_retry_complete done;
7838         u64 start;
7839         unsigned int pgoff;
7840         u32 sectorsize;
7841         int nr_sectors;
7842         blk_status_t ret;
7843         blk_status_t err = BLK_STS_OK;
7844
7845         fs_info = BTRFS_I(inode)->root->fs_info;
7846         sectorsize = fs_info->sectorsize;
7847
7848         start = io_bio->logical;
7849         done.inode = inode;
7850         io_bio->bio.bi_iter = io_bio->iter;
7851
7852         bio_for_each_segment(bvec, &io_bio->bio, iter) {
7853                 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7854                 pgoff = bvec.bv_offset;
7855
7856 next_block_or_try_again:
7857                 done.uptodate = 0;
7858                 done.start = start;
7859                 init_completion(&done.done);
7860
7861                 ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
7862                                 pgoff, start, start + sectorsize - 1,
7863                                 io_bio->mirror_num,
7864                                 btrfs_retry_endio_nocsum, &done);
7865                 if (ret) {
7866                         err = ret;
7867                         goto next;
7868                 }
7869
7870                 wait_for_completion_io(&done.done);
7871
7872                 if (!done.uptodate) {
7873                         /* We might have another mirror, so try again */
7874                         goto next_block_or_try_again;
7875                 }
7876
7877 next:
7878                 start += sectorsize;
7879
7880                 nr_sectors--;
7881                 if (nr_sectors) {
7882                         pgoff += sectorsize;
7883                         ASSERT(pgoff < PAGE_SIZE);
7884                         goto next_block_or_try_again;
7885                 }
7886         }
7887
7888         return err;
7889 }
7890
7891 static void btrfs_retry_endio(struct bio *bio)
7892 {
7893         struct btrfs_retry_complete *done = bio->bi_private;
7894         struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7895         struct extent_io_tree *io_tree, *failure_tree;
7896         struct inode *inode = done->inode;
7897         struct bio_vec *bvec;
7898         int uptodate;
7899         int ret;
7900         int i;
7901
7902         if (bio->bi_status)
7903                 goto end;
7904
7905         uptodate = 1;
7906
7907         ASSERT(bio->bi_vcnt == 1);
7908         ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(done->inode));
7909
7910         io_tree = &BTRFS_I(inode)->io_tree;
7911         failure_tree = &BTRFS_I(inode)->io_failure_tree;
7912
7913         ASSERT(!bio_flagged(bio, BIO_CLONED));
7914         bio_for_each_segment_all(bvec, bio, i) {
7915                 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7916                                              bvec->bv_offset, done->start,
7917                                              bvec->bv_len);
7918                 if (!ret)
7919                         clean_io_failure(BTRFS_I(inode)->root->fs_info,
7920                                          failure_tree, io_tree, done->start,
7921                                          bvec->bv_page,
7922                                          btrfs_ino(BTRFS_I(inode)),
7923                                          bvec->bv_offset);
7924                 else
7925                         uptodate = 0;
7926         }
7927
7928         done->uptodate = uptodate;
7929 end:
7930         complete(&done->done);
7931         bio_put(bio);
7932 }
7933
7934 static blk_status_t __btrfs_subio_endio_read(struct inode *inode,
7935                 struct btrfs_io_bio *io_bio, blk_status_t err)
7936 {
7937         struct btrfs_fs_info *fs_info;
7938         struct bio_vec bvec;
7939         struct bvec_iter iter;
7940         struct btrfs_retry_complete done;
7941         u64 start;
7942         u64 offset = 0;
7943         u32 sectorsize;
7944         int nr_sectors;
7945         unsigned int pgoff;
7946         int csum_pos;
7947         bool uptodate = (err == 0);
7948         int ret;
7949         blk_status_t status;
7950
7951         fs_info = BTRFS_I(inode)->root->fs_info;
7952         sectorsize = fs_info->sectorsize;
7953
7954         err = BLK_STS_OK;
7955         start = io_bio->logical;
7956         done.inode = inode;
7957         io_bio->bio.bi_iter = io_bio->iter;
7958
7959         bio_for_each_segment(bvec, &io_bio->bio, iter) {
7960                 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7961
7962                 pgoff = bvec.bv_offset;
7963 next_block:
7964                 if (uptodate) {
7965                         csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
7966                         ret = __readpage_endio_check(inode, io_bio, csum_pos,
7967                                         bvec.bv_page, pgoff, start, sectorsize);
7968                         if (likely(!ret))
7969                                 goto next;
7970                 }
7971 try_again:
7972                 done.uptodate = 0;
7973                 done.start = start;
7974                 init_completion(&done.done);
7975
7976                 status = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
7977                                         pgoff, start, start + sectorsize - 1,
7978                                         io_bio->mirror_num, btrfs_retry_endio,
7979                                         &done);
7980                 if (status) {
7981                         err = status;
7982                         goto next;
7983                 }
7984
7985                 wait_for_completion_io(&done.done);
7986
7987                 if (!done.uptodate) {
7988                         /* We might have another mirror, so try again */
7989                         goto try_again;
7990                 }
7991 next:
7992                 offset += sectorsize;
7993                 start += sectorsize;
7994
7995                 ASSERT(nr_sectors);
7996
7997                 nr_sectors--;
7998                 if (nr_sectors) {
7999                         pgoff += sectorsize;
8000                         ASSERT(pgoff < PAGE_SIZE);
8001                         goto next_block;
8002                 }
8003         }
8004
8005         return err;
8006 }
8007
8008 static blk_status_t btrfs_subio_endio_read(struct inode *inode,
8009                 struct btrfs_io_bio *io_bio, blk_status_t err)
8010 {
8011         bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8012
8013         if (skip_csum) {
8014                 if (unlikely(err))
8015                         return __btrfs_correct_data_nocsum(inode, io_bio);
8016                 else
8017                         return BLK_STS_OK;
8018         } else {
8019                 return __btrfs_subio_endio_read(inode, io_bio, err);
8020         }
8021 }
8022
8023 static void btrfs_endio_direct_read(struct bio *bio)
8024 {
8025         struct btrfs_dio_private *dip = bio->bi_private;
8026         struct inode *inode = dip->inode;
8027         struct bio *dio_bio;
8028         struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8029         blk_status_t err = bio->bi_status;
8030
8031         if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8032                 err = btrfs_subio_endio_read(inode, io_bio, err);
8033
8034         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8035                       dip->logical_offset + dip->bytes - 1);
8036         dio_bio = dip->dio_bio;
8037
8038         kfree(dip);
8039
8040         dio_bio->bi_status = err;
8041         dio_end_io(dio_bio);
8042         btrfs_io_bio_free_csum(io_bio);
8043         bio_put(bio);
8044 }
8045
8046 static void __endio_write_update_ordered(struct inode *inode,
8047                                          const u64 offset, const u64 bytes,
8048                                          const bool uptodate)
8049 {
8050         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8051         struct btrfs_ordered_extent *ordered = NULL;
8052         struct btrfs_workqueue *wq;
8053         btrfs_work_func_t func;
8054         u64 ordered_offset = offset;
8055         u64 ordered_bytes = bytes;
8056         u64 last_offset;
8057
8058         if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
8059                 wq = fs_info->endio_freespace_worker;
8060                 func = btrfs_freespace_write_helper;
8061         } else {
8062                 wq = fs_info->endio_write_workers;
8063                 func = btrfs_endio_write_helper;
8064         }
8065
8066         while (ordered_offset < offset + bytes) {
8067                 last_offset = ordered_offset;
8068                 if (btrfs_dec_test_first_ordered_pending(inode, &ordered,
8069                                                            &ordered_offset,
8070                                                            ordered_bytes,
8071                                                            uptodate)) {
8072                         btrfs_init_work(&ordered->work, func,
8073                                         finish_ordered_fn,
8074                                         NULL, NULL);
8075                         btrfs_queue_work(wq, &ordered->work);
8076                 }
8077                 /*
8078                  * If btrfs_dec_test_ordered_pending does not find any ordered
8079                  * extent in the range, we can exit.
8080                  */
8081                 if (ordered_offset == last_offset)
8082                         return;
8083                 /*
8084                  * Our bio might span multiple ordered extents. In this case
8085                  * we keep goin until we have accounted the whole dio.
8086                  */
8087                 if (ordered_offset < offset + bytes) {
8088                         ordered_bytes = offset + bytes - ordered_offset;
8089                         ordered = NULL;
8090                 }
8091         }
8092 }
8093
8094 static void btrfs_endio_direct_write(struct bio *bio)
8095 {
8096         struct btrfs_dio_private *dip = bio->bi_private;
8097         struct bio *dio_bio = dip->dio_bio;
8098
8099         __endio_write_update_ordered(dip->inode, dip->logical_offset,
8100                                      dip->bytes, !bio->bi_status);
8101
8102         kfree(dip);
8103
8104         dio_bio->bi_status = bio->bi_status;
8105         dio_end_io(dio_bio);
8106         bio_put(bio);
8107 }
8108
8109 static blk_status_t btrfs_submit_bio_start_direct_io(void *private_data,
8110                                     struct bio *bio, u64 offset)
8111 {
8112         struct inode *inode = private_data;
8113         blk_status_t ret;
8114         ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8115         BUG_ON(ret); /* -ENOMEM */
8116         return 0;
8117 }
8118
8119 static void btrfs_end_dio_bio(struct bio *bio)
8120 {
8121         struct btrfs_dio_private *dip = bio->bi_private;
8122         blk_status_t err = bio->bi_status;
8123
8124         if (err)
8125                 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8126                            "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8127                            btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8128                            bio->bi_opf,
8129                            (unsigned long long)bio->bi_iter.bi_sector,
8130                            bio->bi_iter.bi_size, err);
8131
8132         if (dip->subio_endio)
8133                 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8134
8135         if (err) {
8136                 /*
8137                  * We want to perceive the errors flag being set before
8138                  * decrementing the reference count. We don't need a barrier
8139                  * since atomic operations with a return value are fully
8140                  * ordered as per atomic_t.txt
8141                  */
8142                 dip->errors = 1;
8143         }
8144
8145         /* if there are more bios still pending for this dio, just exit */
8146         if (!atomic_dec_and_test(&dip->pending_bios))
8147                 goto out;
8148
8149         if (dip->errors) {
8150                 bio_io_error(dip->orig_bio);
8151         } else {
8152                 dip->dio_bio->bi_status = BLK_STS_OK;
8153                 bio_endio(dip->orig_bio);
8154         }
8155 out:
8156         bio_put(bio);
8157 }
8158
8159 static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8160                                                  struct btrfs_dio_private *dip,
8161                                                  struct bio *bio,
8162                                                  u64 file_offset)
8163 {
8164         struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8165         struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8166         blk_status_t ret;
8167
8168         /*
8169          * We load all the csum data we need when we submit
8170          * the first bio to reduce the csum tree search and
8171          * contention.
8172          */
8173         if (dip->logical_offset == file_offset) {
8174                 ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8175                                                 file_offset);
8176                 if (ret)
8177                         return ret;
8178         }
8179
8180         if (bio == dip->orig_bio)
8181                 return 0;
8182
8183         file_offset -= dip->logical_offset;
8184         file_offset >>= inode->i_sb->s_blocksize_bits;
8185         io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8186
8187         return 0;
8188 }
8189
8190 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
8191                 struct inode *inode, u64 file_offset, int async_submit)
8192 {
8193         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8194         struct btrfs_dio_private *dip = bio->bi_private;
8195         bool write = bio_op(bio) == REQ_OP_WRITE;
8196         blk_status_t ret;
8197
8198         /* Check btrfs_submit_bio_hook() for rules about async submit. */
8199         if (async_submit)
8200                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8201
8202         if (!write) {
8203                 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8204                 if (ret)
8205                         goto err;
8206         }
8207
8208         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8209                 goto map;
8210
8211         if (write && async_submit) {
8212                 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
8213                                           file_offset, inode,
8214                                           btrfs_submit_bio_start_direct_io);
8215                 goto err;
8216         } else if (write) {
8217                 /*
8218                  * If we aren't doing async submit, calculate the csum of the
8219                  * bio now.
8220                  */
8221                 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8222                 if (ret)
8223                         goto err;
8224         } else {
8225                 ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8226                                                      file_offset);
8227                 if (ret)
8228                         goto err;
8229         }
8230 map:
8231         ret = btrfs_map_bio(fs_info, bio, 0, 0);
8232 err:
8233         return ret;
8234 }
8235
8236 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip)
8237 {
8238         struct inode *inode = dip->inode;
8239         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8240         struct bio *bio;
8241         struct bio *orig_bio = dip->orig_bio;
8242         u64 start_sector = orig_bio->bi_iter.bi_sector;
8243         u64 file_offset = dip->logical_offset;
8244         u64 map_length;
8245         int async_submit = 0;
8246         u64 submit_len;
8247         int clone_offset = 0;
8248         int clone_len;
8249         int ret;
8250         blk_status_t status;
8251
8252         map_length = orig_bio->bi_iter.bi_size;
8253         submit_len = map_length;
8254         ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8255                               &map_length, NULL, 0);
8256         if (ret)
8257                 return -EIO;
8258
8259         if (map_length >= submit_len) {
8260                 bio = orig_bio;
8261                 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8262                 goto submit;
8263         }
8264
8265         /* async crcs make it difficult to collect full stripe writes. */
8266         if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8267                 async_submit = 0;
8268         else
8269                 async_submit = 1;
8270
8271         /* bio split */
8272         ASSERT(map_length <= INT_MAX);
8273         atomic_inc(&dip->pending_bios);
8274         do {
8275                 clone_len = min_t(int, submit_len, map_length);
8276
8277                 /*
8278                  * This will never fail as it's passing GPF_NOFS and
8279                  * the allocation is backed by btrfs_bioset.
8280                  */
8281                 bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
8282                                               clone_len);
8283                 bio->bi_private = dip;
8284                 bio->bi_end_io = btrfs_end_dio_bio;
8285                 btrfs_io_bio(bio)->logical = file_offset;
8286
8287                 ASSERT(submit_len >= clone_len);
8288                 submit_len -= clone_len;
8289                 if (submit_len == 0)
8290                         break;
8291
8292                 /*
8293                  * Increase the count before we submit the bio so we know
8294                  * the end IO handler won't happen before we increase the
8295                  * count. Otherwise, the dip might get freed before we're
8296                  * done setting it up.
8297                  */
8298                 atomic_inc(&dip->pending_bios);
8299
8300                 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8301                                                 async_submit);
8302                 if (status) {
8303                         bio_put(bio);
8304                         atomic_dec(&dip->pending_bios);
8305                         goto out_err;
8306                 }
8307
8308                 clone_offset += clone_len;
8309                 start_sector += clone_len >> 9;
8310                 file_offset += clone_len;
8311
8312                 map_length = submit_len;
8313                 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8314                                       start_sector << 9, &map_length, NULL, 0);
8315                 if (ret)
8316                         goto out_err;
8317         } while (submit_len > 0);
8318
8319 submit:
8320         status = btrfs_submit_dio_bio(bio, inode, file_offset, async_submit);
8321         if (!status)
8322                 return 0;
8323
8324         bio_put(bio);
8325 out_err:
8326         dip->errors = 1;
8327         /*
8328          * Before atomic variable goto zero, we must  make sure dip->errors is
8329          * perceived to be set. This ordering is ensured by the fact that an
8330          * atomic operations with a return value are fully ordered as per
8331          * atomic_t.txt
8332          */
8333         if (atomic_dec_and_test(&dip->pending_bios))
8334                 bio_io_error(dip->orig_bio);
8335
8336         /* bio_end_io() will handle error, so we needn't return it */
8337         return 0;
8338 }
8339
8340 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8341                                 loff_t file_offset)
8342 {
8343         struct btrfs_dio_private *dip = NULL;
8344         struct bio *bio = NULL;
8345         struct btrfs_io_bio *io_bio;
8346         bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8347         int ret = 0;
8348
8349         bio = btrfs_bio_clone(dio_bio);
8350
8351         dip = kzalloc(sizeof(*dip), GFP_NOFS);
8352         if (!dip) {
8353                 ret = -ENOMEM;
8354                 goto free_ordered;
8355         }
8356
8357         dip->private = dio_bio->bi_private;
8358         dip->inode = inode;
8359         dip->logical_offset = file_offset;
8360         dip->bytes = dio_bio->bi_iter.bi_size;
8361         dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8362         bio->bi_private = dip;
8363         dip->orig_bio = bio;
8364         dip->dio_bio = dio_bio;
8365         atomic_set(&dip->pending_bios, 0);
8366         io_bio = btrfs_io_bio(bio);
8367         io_bio->logical = file_offset;
8368
8369         if (write) {
8370                 bio->bi_end_io = btrfs_endio_direct_write;
8371         } else {
8372                 bio->bi_end_io = btrfs_endio_direct_read;
8373                 dip->subio_endio = btrfs_subio_endio_read;
8374         }
8375
8376         /*
8377          * Reset the range for unsubmitted ordered extents (to a 0 length range)
8378          * even if we fail to submit a bio, because in such case we do the
8379          * corresponding error handling below and it must not be done a second
8380          * time by btrfs_direct_IO().
8381          */
8382         if (write) {
8383                 struct btrfs_dio_data *dio_data = current->journal_info;
8384
8385                 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8386                         dip->bytes;
8387                 dio_data->unsubmitted_oe_range_start =
8388                         dio_data->unsubmitted_oe_range_end;
8389         }
8390
8391         ret = btrfs_submit_direct_hook(dip);
8392         if (!ret)
8393                 return;
8394
8395         btrfs_io_bio_free_csum(io_bio);
8396
8397 free_ordered:
8398         /*
8399          * If we arrived here it means either we failed to submit the dip
8400          * or we either failed to clone the dio_bio or failed to allocate the
8401          * dip. If we cloned the dio_bio and allocated the dip, we can just
8402          * call bio_endio against our io_bio so that we get proper resource
8403          * cleanup if we fail to submit the dip, otherwise, we must do the
8404          * same as btrfs_endio_direct_[write|read] because we can't call these
8405          * callbacks - they require an allocated dip and a clone of dio_bio.
8406          */
8407         if (bio && dip) {
8408                 bio_io_error(bio);
8409                 /*
8410                  * The end io callbacks free our dip, do the final put on bio
8411                  * and all the cleanup and final put for dio_bio (through
8412                  * dio_end_io()).
8413                  */
8414                 dip = NULL;
8415                 bio = NULL;
8416         } else {
8417                 if (write)
8418                         __endio_write_update_ordered(inode,
8419                                                 file_offset,
8420                                                 dio_bio->bi_iter.bi_size,
8421                                                 false);
8422                 else
8423                         unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8424                               file_offset + dio_bio->bi_iter.bi_size - 1);
8425
8426                 dio_bio->bi_status = BLK_STS_IOERR;
8427                 /*
8428                  * Releases and cleans up our dio_bio, no need to bio_put()
8429                  * nor bio_endio()/bio_io_error() against dio_bio.
8430                  */
8431                 dio_end_io(dio_bio);
8432         }
8433         if (bio)
8434                 bio_put(bio);
8435         kfree(dip);
8436 }
8437
8438 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8439                                const struct iov_iter *iter, loff_t offset)
8440 {
8441         int seg;
8442         int i;
8443         unsigned int blocksize_mask = fs_info->sectorsize - 1;
8444         ssize_t retval = -EINVAL;
8445
8446         if (offset & blocksize_mask)
8447                 goto out;
8448
8449         if (iov_iter_alignment(iter) & blocksize_mask)
8450                 goto out;
8451
8452         /* If this is a write we don't need to check anymore */
8453         if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8454                 return 0;
8455         /*
8456          * Check to make sure we don't have duplicate iov_base's in this
8457          * iovec, if so return EINVAL, otherwise we'll get csum errors
8458          * when reading back.
8459          */
8460         for (seg = 0; seg < iter->nr_segs; seg++) {
8461                 for (i = seg + 1; i < iter->nr_segs; i++) {
8462                         if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8463                                 goto out;
8464                 }
8465         }
8466         retval = 0;
8467 out:
8468         return retval;
8469 }
8470
8471 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8472 {
8473         struct file *file = iocb->ki_filp;
8474         struct inode *inode = file->f_mapping->host;
8475         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8476         struct btrfs_dio_data dio_data = { 0 };
8477         struct extent_changeset *data_reserved = NULL;
8478         loff_t offset = iocb->ki_pos;
8479         size_t count = 0;
8480         int flags = 0;
8481         bool wakeup = true;
8482         bool relock = false;
8483         ssize_t ret;
8484
8485         if (check_direct_IO(fs_info, iter, offset))
8486                 return 0;
8487
8488         inode_dio_begin(inode);
8489
8490         /*
8491          * The generic stuff only does filemap_write_and_wait_range, which
8492          * isn't enough if we've written compressed pages to this area, so
8493          * we need to flush the dirty pages again to make absolutely sure
8494          * that any outstanding dirty pages are on disk.
8495          */
8496         count = iov_iter_count(iter);
8497         if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8498                      &BTRFS_I(inode)->runtime_flags))
8499                 filemap_fdatawrite_range(inode->i_mapping, offset,
8500                                          offset + count - 1);
8501
8502         if (iov_iter_rw(iter) == WRITE) {
8503                 /*
8504                  * If the write DIO is beyond the EOF, we need update
8505                  * the isize, but it is protected by i_mutex. So we can
8506                  * not unlock the i_mutex at this case.
8507                  */
8508                 if (offset + count <= inode->i_size) {
8509                         dio_data.overwrite = 1;
8510                         inode_unlock(inode);
8511                         relock = true;
8512                 } else if (iocb->ki_flags & IOCB_NOWAIT) {
8513                         ret = -EAGAIN;
8514                         goto out;
8515                 }
8516                 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
8517                                                    offset, count);
8518                 if (ret)
8519                         goto out;
8520
8521                 /*
8522                  * We need to know how many extents we reserved so that we can
8523                  * do the accounting properly if we go over the number we
8524                  * originally calculated.  Abuse current->journal_info for this.
8525                  */
8526                 dio_data.reserve = round_up(count,
8527                                             fs_info->sectorsize);
8528                 dio_data.unsubmitted_oe_range_start = (u64)offset;
8529                 dio_data.unsubmitted_oe_range_end = (u64)offset;
8530                 current->journal_info = &dio_data;
8531                 down_read(&BTRFS_I(inode)->dio_sem);
8532         } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8533                                      &BTRFS_I(inode)->runtime_flags)) {
8534                 inode_dio_end(inode);
8535                 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8536                 wakeup = false;
8537         }
8538
8539         ret = __blockdev_direct_IO(iocb, inode,
8540                                    fs_info->fs_devices->latest_bdev,
8541                                    iter, btrfs_get_blocks_direct, NULL,
8542                                    btrfs_submit_direct, flags);
8543         if (iov_iter_rw(iter) == WRITE) {
8544                 up_read(&BTRFS_I(inode)->dio_sem);
8545                 current->journal_info = NULL;
8546                 if (ret < 0 && ret != -EIOCBQUEUED) {
8547                         if (dio_data.reserve)
8548                                 btrfs_delalloc_release_space(inode, data_reserved,
8549                                         offset, dio_data.reserve, true);
8550                         /*
8551                          * On error we might have left some ordered extents
8552                          * without submitting corresponding bios for them, so
8553                          * cleanup them up to avoid other tasks getting them
8554                          * and waiting for them to complete forever.
8555                          */
8556                         if (dio_data.unsubmitted_oe_range_start <
8557                             dio_data.unsubmitted_oe_range_end)
8558                                 __endio_write_update_ordered(inode,
8559                                         dio_data.unsubmitted_oe_range_start,
8560                                         dio_data.unsubmitted_oe_range_end -
8561                                         dio_data.unsubmitted_oe_range_start,
8562                                         false);
8563                 } else if (ret >= 0 && (size_t)ret < count)
8564                         btrfs_delalloc_release_space(inode, data_reserved,
8565                                         offset, count - (size_t)ret, true);
8566                 btrfs_delalloc_release_extents(BTRFS_I(inode), count, false);
8567         }
8568 out:
8569         if (wakeup)
8570                 inode_dio_end(inode);
8571         if (relock)
8572                 inode_lock(inode);
8573
8574         extent_changeset_free(data_reserved);
8575         return ret;
8576 }
8577
8578 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
8579
8580 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8581                 __u64 start, __u64 len)
8582 {
8583         int     ret;
8584
8585         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8586         if (ret)
8587                 return ret;
8588
8589         return extent_fiemap(inode, fieinfo, start, len);
8590 }
8591
8592 int btrfs_readpage(struct file *file, struct page *page)
8593 {
8594         struct extent_io_tree *tree;
8595         tree = &BTRFS_I(page->mapping->host)->io_tree;
8596         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8597 }
8598
8599 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8600 {
8601         struct inode *inode = page->mapping->host;
8602         int ret;
8603
8604         if (current->flags & PF_MEMALLOC) {
8605                 redirty_page_for_writepage(wbc, page);
8606                 unlock_page(page);
8607                 return 0;
8608         }
8609
8610         /*
8611          * If we are under memory pressure we will call this directly from the
8612          * VM, we need to make sure we have the inode referenced for the ordered
8613          * extent.  If not just return like we didn't do anything.
8614          */
8615         if (!igrab(inode)) {
8616                 redirty_page_for_writepage(wbc, page);
8617                 return AOP_WRITEPAGE_ACTIVATE;
8618         }
8619         ret = extent_write_full_page(page, wbc);
8620         btrfs_add_delayed_iput(inode);
8621         return ret;
8622 }
8623
8624 static int btrfs_writepages(struct address_space *mapping,
8625                             struct writeback_control *wbc)
8626 {
8627         return extent_writepages(mapping, wbc);
8628 }
8629
8630 static int
8631 btrfs_readpages(struct file *file, struct address_space *mapping,
8632                 struct list_head *pages, unsigned nr_pages)
8633 {
8634         return extent_readpages(mapping, pages, nr_pages);
8635 }
8636
8637 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8638 {
8639         int ret = try_release_extent_mapping(page, gfp_flags);
8640         if (ret == 1) {
8641                 ClearPagePrivate(page);
8642                 set_page_private(page, 0);
8643                 put_page(page);
8644         }
8645         return ret;
8646 }
8647
8648 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8649 {
8650         if (PageWriteback(page) || PageDirty(page))
8651                 return 0;
8652         return __btrfs_releasepage(page, gfp_flags);
8653 }
8654
8655 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8656                                  unsigned int length)
8657 {
8658         struct inode *inode = page->mapping->host;
8659         struct extent_io_tree *tree;
8660         struct btrfs_ordered_extent *ordered;
8661         struct extent_state *cached_state = NULL;
8662         u64 page_start = page_offset(page);
8663         u64 page_end = page_start + PAGE_SIZE - 1;
8664         u64 start;
8665         u64 end;
8666         int inode_evicting = inode->i_state & I_FREEING;
8667
8668         /*
8669          * we have the page locked, so new writeback can't start,
8670          * and the dirty bit won't be cleared while we are here.
8671          *
8672          * Wait for IO on this page so that we can safely clear
8673          * the PagePrivate2 bit and do ordered accounting
8674          */
8675         wait_on_page_writeback(page);
8676
8677         tree = &BTRFS_I(inode)->io_tree;
8678         if (offset) {
8679                 btrfs_releasepage(page, GFP_NOFS);
8680                 return;
8681         }
8682
8683         if (!inode_evicting)
8684                 lock_extent_bits(tree, page_start, page_end, &cached_state);
8685 again:
8686         start = page_start;
8687         ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8688                                         page_end - start + 1);
8689         if (ordered) {
8690                 end = min(page_end, ordered->file_offset + ordered->len - 1);
8691                 /*
8692                  * IO on this page will never be started, so we need
8693                  * to account for any ordered extents now
8694                  */
8695                 if (!inode_evicting)
8696                         clear_extent_bit(tree, start, end,
8697                                          EXTENT_DIRTY | EXTENT_DELALLOC |
8698                                          EXTENT_DELALLOC_NEW |
8699                                          EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8700                                          EXTENT_DEFRAG, 1, 0, &cached_state);
8701                 /*
8702                  * whoever cleared the private bit is responsible
8703                  * for the finish_ordered_io
8704                  */
8705                 if (TestClearPagePrivate2(page)) {
8706                         struct btrfs_ordered_inode_tree *tree;
8707                         u64 new_len;
8708
8709                         tree = &BTRFS_I(inode)->ordered_tree;
8710
8711                         spin_lock_irq(&tree->lock);
8712                         set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8713                         new_len = start - ordered->file_offset;
8714                         if (new_len < ordered->truncated_len)
8715                                 ordered->truncated_len = new_len;
8716                         spin_unlock_irq(&tree->lock);
8717
8718                         if (btrfs_dec_test_ordered_pending(inode, &ordered,
8719                                                            start,
8720                                                            end - start + 1, 1))
8721                                 btrfs_finish_ordered_io(ordered);
8722                 }
8723                 btrfs_put_ordered_extent(ordered);
8724                 if (!inode_evicting) {
8725                         cached_state = NULL;
8726                         lock_extent_bits(tree, start, end,
8727                                          &cached_state);
8728                 }
8729
8730                 start = end + 1;
8731                 if (start < page_end)
8732                         goto again;
8733         }
8734
8735         /*
8736          * Qgroup reserved space handler
8737          * Page here will be either
8738          * 1) Already written to disk
8739          *    In this case, its reserved space is released from data rsv map
8740          *    and will be freed by delayed_ref handler finally.
8741          *    So even we call qgroup_free_data(), it won't decrease reserved
8742          *    space.
8743          * 2) Not written to disk
8744          *    This means the reserved space should be freed here. However,
8745          *    if a truncate invalidates the page (by clearing PageDirty)
8746          *    and the page is accounted for while allocating extent
8747          *    in btrfs_check_data_free_space() we let delayed_ref to
8748          *    free the entire extent.
8749          */
8750         if (PageDirty(page))
8751                 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
8752         if (!inode_evicting) {
8753                 clear_extent_bit(tree, page_start, page_end,
8754                                  EXTENT_LOCKED | EXTENT_DIRTY |
8755                                  EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
8756                                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
8757                                  &cached_state);
8758
8759                 __btrfs_releasepage(page, GFP_NOFS);
8760         }
8761
8762         ClearPageChecked(page);
8763         if (PagePrivate(page)) {
8764                 ClearPagePrivate(page);
8765                 set_page_private(page, 0);
8766                 put_page(page);
8767         }
8768 }
8769
8770 /*
8771  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8772  * called from a page fault handler when a page is first dirtied. Hence we must
8773  * be careful to check for EOF conditions here. We set the page up correctly
8774  * for a written page which means we get ENOSPC checking when writing into
8775  * holes and correct delalloc and unwritten extent mapping on filesystems that
8776  * support these features.
8777  *
8778  * We are not allowed to take the i_mutex here so we have to play games to
8779  * protect against truncate races as the page could now be beyond EOF.  Because
8780  * truncate_setsize() writes the inode size before removing pages, once we have
8781  * the page lock we can determine safely if the page is beyond EOF. If it is not
8782  * beyond EOF, then the page is guaranteed safe against truncation until we
8783  * unlock the page.
8784  */
8785 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8786 {
8787         struct page *page = vmf->page;
8788         struct inode *inode = file_inode(vmf->vma->vm_file);
8789         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8790         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8791         struct btrfs_ordered_extent *ordered;
8792         struct extent_state *cached_state = NULL;
8793         struct extent_changeset *data_reserved = NULL;
8794         char *kaddr;
8795         unsigned long zero_start;
8796         loff_t size;
8797         vm_fault_t ret;
8798         int ret2;
8799         int reserved = 0;
8800         u64 reserved_space;
8801         u64 page_start;
8802         u64 page_end;
8803         u64 end;
8804
8805         reserved_space = PAGE_SIZE;
8806
8807         sb_start_pagefault(inode->i_sb);
8808         page_start = page_offset(page);
8809         page_end = page_start + PAGE_SIZE - 1;
8810         end = page_end;
8811
8812         /*
8813          * Reserving delalloc space after obtaining the page lock can lead to
8814          * deadlock. For example, if a dirty page is locked by this function
8815          * and the call to btrfs_delalloc_reserve_space() ends up triggering
8816          * dirty page write out, then the btrfs_writepage() function could
8817          * end up waiting indefinitely to get a lock on the page currently
8818          * being processed by btrfs_page_mkwrite() function.
8819          */
8820         ret2 = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
8821                                            reserved_space);
8822         if (!ret2) {
8823                 ret2 = file_update_time(vmf->vma->vm_file);
8824                 reserved = 1;
8825         }
8826         if (ret2) {
8827                 ret = vmf_error(ret2);
8828                 if (reserved)
8829                         goto out;
8830                 goto out_noreserve;
8831         }
8832
8833         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8834 again:
8835         lock_page(page);
8836         size = i_size_read(inode);
8837
8838         if ((page->mapping != inode->i_mapping) ||
8839             (page_start >= size)) {
8840                 /* page got truncated out from underneath us */
8841                 goto out_unlock;
8842         }
8843         wait_on_page_writeback(page);
8844
8845         lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8846         set_page_extent_mapped(page);
8847
8848         /*
8849          * we can't set the delalloc bits if there are pending ordered
8850          * extents.  Drop our locks and wait for them to finish
8851          */
8852         ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8853                         PAGE_SIZE);
8854         if (ordered) {
8855                 unlock_extent_cached(io_tree, page_start, page_end,
8856                                      &cached_state);
8857                 unlock_page(page);
8858                 btrfs_start_ordered_extent(inode, ordered, 1);
8859                 btrfs_put_ordered_extent(ordered);
8860                 goto again;
8861         }
8862
8863         if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8864                 reserved_space = round_up(size - page_start,
8865                                           fs_info->sectorsize);
8866                 if (reserved_space < PAGE_SIZE) {
8867                         end = page_start + reserved_space - 1;
8868                         btrfs_delalloc_release_space(inode, data_reserved,
8869                                         page_start, PAGE_SIZE - reserved_space,
8870                                         true);
8871                 }
8872         }
8873
8874         /*
8875          * page_mkwrite gets called when the page is firstly dirtied after it's
8876          * faulted in, but write(2) could also dirty a page and set delalloc
8877          * bits, thus in this case for space account reason, we still need to
8878          * clear any delalloc bits within this page range since we have to
8879          * reserve data&meta space before lock_page() (see above comments).
8880          */
8881         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8882                           EXTENT_DIRTY | EXTENT_DELALLOC |
8883                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8884                           0, 0, &cached_state);
8885
8886         ret2 = btrfs_set_extent_delalloc(inode, page_start, end, 0,
8887                                         &cached_state, 0);
8888         if (ret2) {
8889                 unlock_extent_cached(io_tree, page_start, page_end,
8890                                      &cached_state);
8891                 ret = VM_FAULT_SIGBUS;
8892                 goto out_unlock;
8893         }
8894         ret2 = 0;
8895
8896         /* page is wholly or partially inside EOF */
8897         if (page_start + PAGE_SIZE > size)
8898                 zero_start = size & ~PAGE_MASK;
8899         else
8900                 zero_start = PAGE_SIZE;
8901
8902         if (zero_start != PAGE_SIZE) {
8903                 kaddr = kmap(page);
8904                 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
8905                 flush_dcache_page(page);
8906                 kunmap(page);
8907         }
8908         ClearPageChecked(page);
8909         set_page_dirty(page);
8910         SetPageUptodate(page);
8911
8912         BTRFS_I(inode)->last_trans = fs_info->generation;
8913         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8914         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8915
8916         unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8917
8918         if (!ret2) {
8919                 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, true);
8920                 sb_end_pagefault(inode->i_sb);
8921                 extent_changeset_free(data_reserved);
8922                 return VM_FAULT_LOCKED;
8923         }
8924
8925 out_unlock:
8926         unlock_page(page);
8927 out:
8928         btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, (ret != 0));
8929         btrfs_delalloc_release_space(inode, data_reserved, page_start,
8930                                      reserved_space, (ret != 0));
8931 out_noreserve:
8932         sb_end_pagefault(inode->i_sb);
8933         extent_changeset_free(data_reserved);
8934         return ret;
8935 }
8936
8937 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8938 {
8939         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8940         struct btrfs_root *root = BTRFS_I(inode)->root;
8941         struct btrfs_block_rsv *rsv;
8942         int ret;
8943         struct btrfs_trans_handle *trans;
8944         u64 mask = fs_info->sectorsize - 1;
8945         u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
8946
8947         if (!skip_writeback) {
8948                 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8949                                                (u64)-1);
8950                 if (ret)
8951                         return ret;
8952         }
8953
8954         /*
8955          * Yes ladies and gentlemen, this is indeed ugly.  We have a couple of
8956          * things going on here:
8957          *
8958          * 1) We need to reserve space to update our inode.
8959          *
8960          * 2) We need to have something to cache all the space that is going to
8961          * be free'd up by the truncate operation, but also have some slack
8962          * space reserved in case it uses space during the truncate (thank you
8963          * very much snapshotting).
8964          *
8965          * And we need these to be separate.  The fact is we can use a lot of
8966          * space doing the truncate, and we have no earthly idea how much space
8967          * we will use, so we need the truncate reservation to be separate so it
8968          * doesn't end up using space reserved for updating the inode.  We also
8969          * need to be able to stop the transaction and start a new one, which
8970          * means we need to be able to update the inode several times, and we
8971          * have no idea of knowing how many times that will be, so we can't just
8972          * reserve 1 item for the entirety of the operation, so that has to be
8973          * done separately as well.
8974          *
8975          * So that leaves us with
8976          *
8977          * 1) rsv - for the truncate reservation, which we will steal from the
8978          * transaction reservation.
8979          * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8980          * updating the inode.
8981          */
8982         rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8983         if (!rsv)
8984                 return -ENOMEM;
8985         rsv->size = min_size;
8986         rsv->failfast = 1;
8987
8988         /*
8989          * 1 for the truncate slack space
8990          * 1 for updating the inode.
8991          */
8992         trans = btrfs_start_transaction(root, 2);
8993         if (IS_ERR(trans)) {
8994                 ret = PTR_ERR(trans);
8995                 goto out;
8996         }
8997
8998         /* Migrate the slack space for the truncate to our reserve */
8999         ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9000                                       min_size, false);
9001         BUG_ON(ret);
9002
9003         /*
9004          * So if we truncate and then write and fsync we normally would just
9005          * write the extents that changed, which is a problem if we need to
9006          * first truncate that entire inode.  So set this flag so we write out
9007          * all of the extents in the inode to the sync log so we're completely
9008          * safe.
9009          */
9010         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9011         trans->block_rsv = rsv;
9012
9013         while (1) {
9014                 ret = btrfs_truncate_inode_items(trans, root, inode,
9015                                                  inode->i_size,
9016                                                  BTRFS_EXTENT_DATA_KEY);
9017                 trans->block_rsv = &fs_info->trans_block_rsv;
9018                 if (ret != -ENOSPC && ret != -EAGAIN)
9019                         break;
9020
9021                 ret = btrfs_update_inode(trans, root, inode);
9022                 if (ret)
9023                         break;
9024
9025                 btrfs_end_transaction(trans);
9026                 btrfs_btree_balance_dirty(fs_info);
9027
9028                 trans = btrfs_start_transaction(root, 2);
9029                 if (IS_ERR(trans)) {
9030                         ret = PTR_ERR(trans);
9031                         trans = NULL;
9032                         break;
9033                 }
9034
9035                 btrfs_block_rsv_release(fs_info, rsv, -1);
9036                 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9037                                               rsv, min_size, false);
9038                 BUG_ON(ret);    /* shouldn't happen */
9039                 trans->block_rsv = rsv;
9040         }
9041
9042         /*
9043          * We can't call btrfs_truncate_block inside a trans handle as we could
9044          * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
9045          * we've truncated everything except the last little bit, and can do
9046          * btrfs_truncate_block and then update the disk_i_size.
9047          */
9048         if (ret == NEED_TRUNCATE_BLOCK) {
9049                 btrfs_end_transaction(trans);
9050                 btrfs_btree_balance_dirty(fs_info);
9051
9052                 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
9053                 if (ret)
9054                         goto out;
9055                 trans = btrfs_start_transaction(root, 1);
9056                 if (IS_ERR(trans)) {
9057                         ret = PTR_ERR(trans);
9058                         goto out;
9059                 }
9060                 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
9061         }
9062
9063         if (trans) {
9064                 int ret2;
9065
9066                 trans->block_rsv = &fs_info->trans_block_rsv;
9067                 ret2 = btrfs_update_inode(trans, root, inode);
9068                 if (ret2 && !ret)
9069                         ret = ret2;
9070
9071                 ret2 = btrfs_end_transaction(trans);
9072                 if (ret2 && !ret)
9073                         ret = ret2;
9074                 btrfs_btree_balance_dirty(fs_info);
9075         }
9076 out:
9077         btrfs_free_block_rsv(fs_info, rsv);
9078
9079         return ret;
9080 }
9081
9082 /*
9083  * create a new subvolume directory/inode (helper for the ioctl).
9084  */
9085 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9086                              struct btrfs_root *new_root,
9087                              struct btrfs_root *parent_root,
9088                              u64 new_dirid)
9089 {
9090         struct inode *inode;
9091         int err;
9092         u64 index = 0;
9093
9094         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9095                                 new_dirid, new_dirid,
9096                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
9097                                 &index);
9098         if (IS_ERR(inode))
9099                 return PTR_ERR(inode);
9100         inode->i_op = &btrfs_dir_inode_operations;
9101         inode->i_fop = &btrfs_dir_file_operations;
9102
9103         set_nlink(inode, 1);
9104         btrfs_i_size_write(BTRFS_I(inode), 0);
9105         unlock_new_inode(inode);
9106
9107         err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9108         if (err)
9109                 btrfs_err(new_root->fs_info,
9110                           "error inheriting subvolume %llu properties: %d",
9111                           new_root->root_key.objectid, err);
9112
9113         err = btrfs_update_inode(trans, new_root, inode);
9114
9115         iput(inode);
9116         return err;
9117 }
9118
9119 struct inode *btrfs_alloc_inode(struct super_block *sb)
9120 {
9121         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
9122         struct btrfs_inode *ei;
9123         struct inode *inode;
9124
9125         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
9126         if (!ei)
9127                 return NULL;
9128
9129         ei->root = NULL;
9130         ei->generation = 0;
9131         ei->last_trans = 0;
9132         ei->last_sub_trans = 0;
9133         ei->logged_trans = 0;
9134         ei->delalloc_bytes = 0;
9135         ei->new_delalloc_bytes = 0;
9136         ei->defrag_bytes = 0;
9137         ei->disk_i_size = 0;
9138         ei->flags = 0;
9139         ei->csum_bytes = 0;
9140         ei->index_cnt = (u64)-1;
9141         ei->dir_index = 0;
9142         ei->last_unlink_trans = 0;
9143         ei->last_link_trans = 0;
9144         ei->last_log_commit = 0;
9145
9146         spin_lock_init(&ei->lock);
9147         ei->outstanding_extents = 0;
9148         if (sb->s_magic != BTRFS_TEST_MAGIC)
9149                 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
9150                                               BTRFS_BLOCK_RSV_DELALLOC);
9151         ei->runtime_flags = 0;
9152         ei->prop_compress = BTRFS_COMPRESS_NONE;
9153         ei->defrag_compress = BTRFS_COMPRESS_NONE;
9154
9155         ei->delayed_node = NULL;
9156
9157         ei->i_otime.tv_sec = 0;
9158         ei->i_otime.tv_nsec = 0;
9159
9160         inode = &ei->vfs_inode;
9161         extent_map_tree_init(&ei->extent_tree);
9162         extent_io_tree_init(&ei->io_tree, inode);
9163         extent_io_tree_init(&ei->io_failure_tree, inode);
9164         ei->io_tree.track_uptodate = 1;
9165         ei->io_failure_tree.track_uptodate = 1;
9166         atomic_set(&ei->sync_writers, 0);
9167         mutex_init(&ei->log_mutex);
9168         mutex_init(&ei->delalloc_mutex);
9169         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9170         INIT_LIST_HEAD(&ei->delalloc_inodes);
9171         INIT_LIST_HEAD(&ei->delayed_iput);
9172         RB_CLEAR_NODE(&ei->rb_node);
9173         init_rwsem(&ei->dio_sem);
9174
9175         return inode;
9176 }
9177
9178 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9179 void btrfs_test_destroy_inode(struct inode *inode)
9180 {
9181         btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9182         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9183 }
9184 #endif
9185
9186 static void btrfs_i_callback(struct rcu_head *head)
9187 {
9188         struct inode *inode = container_of(head, struct inode, i_rcu);
9189         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9190 }
9191
9192 void btrfs_destroy_inode(struct inode *inode)
9193 {
9194         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9195         struct btrfs_ordered_extent *ordered;
9196         struct btrfs_root *root = BTRFS_I(inode)->root;
9197
9198         WARN_ON(!hlist_empty(&inode->i_dentry));
9199         WARN_ON(inode->i_data.nrpages);
9200         WARN_ON(BTRFS_I(inode)->block_rsv.reserved);
9201         WARN_ON(BTRFS_I(inode)->block_rsv.size);
9202         WARN_ON(BTRFS_I(inode)->outstanding_extents);
9203         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9204         WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
9205         WARN_ON(BTRFS_I(inode)->csum_bytes);
9206         WARN_ON(BTRFS_I(inode)->defrag_bytes);
9207
9208         /*
9209          * This can happen where we create an inode, but somebody else also
9210          * created the same inode and we need to destroy the one we already
9211          * created.
9212          */
9213         if (!root)
9214                 goto free;
9215
9216         while (1) {
9217                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9218                 if (!ordered)
9219                         break;
9220                 else {
9221                         btrfs_err(fs_info,
9222                                   "found ordered extent %llu %llu on inode cleanup",
9223                                   ordered->file_offset, ordered->len);
9224                         btrfs_remove_ordered_extent(inode, ordered);
9225                         btrfs_put_ordered_extent(ordered);
9226                         btrfs_put_ordered_extent(ordered);
9227                 }
9228         }
9229         btrfs_qgroup_check_reserved_leak(inode);
9230         inode_tree_del(inode);
9231         btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9232 free:
9233         call_rcu(&inode->i_rcu, btrfs_i_callback);
9234 }
9235
9236 int btrfs_drop_inode(struct inode *inode)
9237 {
9238         struct btrfs_root *root = BTRFS_I(inode)->root;
9239
9240         if (root == NULL)
9241                 return 1;
9242
9243         /* the snap/subvol tree is on deleting */
9244         if (btrfs_root_refs(&root->root_item) == 0)
9245                 return 1;
9246         else
9247                 return generic_drop_inode(inode);
9248 }
9249
9250 static void init_once(void *foo)
9251 {
9252         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9253
9254         inode_init_once(&ei->vfs_inode);
9255 }
9256
9257 void __cold btrfs_destroy_cachep(void)
9258 {
9259         /*
9260          * Make sure all delayed rcu free inodes are flushed before we
9261          * destroy cache.
9262          */
9263         rcu_barrier();
9264         kmem_cache_destroy(btrfs_inode_cachep);
9265         kmem_cache_destroy(btrfs_trans_handle_cachep);
9266         kmem_cache_destroy(btrfs_path_cachep);
9267         kmem_cache_destroy(btrfs_free_space_cachep);
9268 }
9269
9270 int __init btrfs_init_cachep(void)
9271 {
9272         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9273                         sizeof(struct btrfs_inode), 0,
9274                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9275                         init_once);
9276         if (!btrfs_inode_cachep)
9277                 goto fail;
9278
9279         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9280                         sizeof(struct btrfs_trans_handle), 0,
9281                         SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9282         if (!btrfs_trans_handle_cachep)
9283                 goto fail;
9284
9285         btrfs_path_cachep = kmem_cache_create("btrfs_path",
9286                         sizeof(struct btrfs_path), 0,
9287                         SLAB_MEM_SPREAD, NULL);
9288         if (!btrfs_path_cachep)
9289                 goto fail;
9290
9291         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9292                         sizeof(struct btrfs_free_space), 0,
9293                         SLAB_MEM_SPREAD, NULL);
9294         if (!btrfs_free_space_cachep)
9295                 goto fail;
9296
9297         return 0;
9298 fail:
9299         btrfs_destroy_cachep();
9300         return -ENOMEM;
9301 }
9302
9303 static int btrfs_getattr(const struct path *path, struct kstat *stat,
9304                          u32 request_mask, unsigned int flags)
9305 {
9306         u64 delalloc_bytes;
9307         struct inode *inode = d_inode(path->dentry);
9308         u32 blocksize = inode->i_sb->s_blocksize;
9309         u32 bi_flags = BTRFS_I(inode)->flags;
9310
9311         stat->result_mask |= STATX_BTIME;
9312         stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9313         stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9314         if (bi_flags & BTRFS_INODE_APPEND)
9315                 stat->attributes |= STATX_ATTR_APPEND;
9316         if (bi_flags & BTRFS_INODE_COMPRESS)
9317                 stat->attributes |= STATX_ATTR_COMPRESSED;
9318         if (bi_flags & BTRFS_INODE_IMMUTABLE)
9319                 stat->attributes |= STATX_ATTR_IMMUTABLE;
9320         if (bi_flags & BTRFS_INODE_NODUMP)
9321                 stat->attributes |= STATX_ATTR_NODUMP;
9322
9323         stat->attributes_mask |= (STATX_ATTR_APPEND |
9324                                   STATX_ATTR_COMPRESSED |
9325                                   STATX_ATTR_IMMUTABLE |
9326                                   STATX_ATTR_NODUMP);
9327
9328         generic_fillattr(inode, stat);
9329         stat->dev = BTRFS_I(inode)->root->anon_dev;
9330
9331         spin_lock(&BTRFS_I(inode)->lock);
9332         delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9333         spin_unlock(&BTRFS_I(inode)->lock);
9334         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9335                         ALIGN(delalloc_bytes, blocksize)) >> 9;
9336         return 0;
9337 }
9338
9339 static int btrfs_rename_exchange(struct inode *old_dir,
9340                               struct dentry *old_dentry,
9341                               struct inode *new_dir,
9342                               struct dentry *new_dentry)
9343 {
9344         struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9345         struct btrfs_trans_handle *trans;
9346         struct btrfs_root *root = BTRFS_I(old_dir)->root;
9347         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9348         struct inode *new_inode = new_dentry->d_inode;
9349         struct inode *old_inode = old_dentry->d_inode;
9350         struct timespec64 ctime = current_time(old_inode);
9351         struct dentry *parent;
9352         u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9353         u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9354         u64 old_idx = 0;
9355         u64 new_idx = 0;
9356         u64 root_objectid;
9357         int ret;
9358         bool root_log_pinned = false;
9359         bool dest_log_pinned = false;
9360         struct btrfs_log_ctx ctx_root;
9361         struct btrfs_log_ctx ctx_dest;
9362         bool sync_log_root = false;
9363         bool sync_log_dest = false;
9364         bool commit_transaction = false;
9365
9366         /* we only allow rename subvolume link between subvolumes */
9367         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9368                 return -EXDEV;
9369
9370         btrfs_init_log_ctx(&ctx_root, old_inode);
9371         btrfs_init_log_ctx(&ctx_dest, new_inode);
9372
9373         /* close the race window with snapshot create/destroy ioctl */
9374         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9375                 down_read(&fs_info->subvol_sem);
9376         if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9377                 down_read(&fs_info->subvol_sem);
9378
9379         /*
9380          * We want to reserve the absolute worst case amount of items.  So if
9381          * both inodes are subvols and we need to unlink them then that would
9382          * require 4 item modifications, but if they are both normal inodes it
9383          * would require 5 item modifications, so we'll assume their normal
9384          * inodes.  So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9385          * should cover the worst case number of items we'll modify.
9386          */
9387         trans = btrfs_start_transaction(root, 12);
9388         if (IS_ERR(trans)) {
9389                 ret = PTR_ERR(trans);
9390                 goto out_notrans;
9391         }
9392
9393         /*
9394          * We need to find a free sequence number both in the source and
9395          * in the destination directory for the exchange.
9396          */
9397         ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9398         if (ret)
9399                 goto out_fail;
9400         ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9401         if (ret)
9402                 goto out_fail;
9403
9404         BTRFS_I(old_inode)->dir_index = 0ULL;
9405         BTRFS_I(new_inode)->dir_index = 0ULL;
9406
9407         /* Reference for the source. */
9408         if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9409                 /* force full log commit if subvolume involved. */
9410                 btrfs_set_log_full_commit(fs_info, trans);
9411         } else {
9412                 btrfs_pin_log_trans(root);
9413                 root_log_pinned = true;
9414                 ret = btrfs_insert_inode_ref(trans, dest,
9415                                              new_dentry->d_name.name,
9416                                              new_dentry->d_name.len,
9417                                              old_ino,
9418                                              btrfs_ino(BTRFS_I(new_dir)),
9419                                              old_idx);
9420                 if (ret)
9421                         goto out_fail;
9422         }
9423
9424         /* And now for the dest. */
9425         if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9426                 /* force full log commit if subvolume involved. */
9427                 btrfs_set_log_full_commit(fs_info, trans);
9428         } else {
9429                 btrfs_pin_log_trans(dest);
9430                 dest_log_pinned = true;
9431                 ret = btrfs_insert_inode_ref(trans, root,
9432                                              old_dentry->d_name.name,
9433                                              old_dentry->d_name.len,
9434                                              new_ino,
9435                                              btrfs_ino(BTRFS_I(old_dir)),
9436                                              new_idx);
9437                 if (ret)
9438                         goto out_fail;
9439         }
9440
9441         /* Update inode version and ctime/mtime. */
9442         inode_inc_iversion(old_dir);
9443         inode_inc_iversion(new_dir);
9444         inode_inc_iversion(old_inode);
9445         inode_inc_iversion(new_inode);
9446         old_dir->i_ctime = old_dir->i_mtime = ctime;
9447         new_dir->i_ctime = new_dir->i_mtime = ctime;
9448         old_inode->i_ctime = ctime;
9449         new_inode->i_ctime = ctime;
9450
9451         if (old_dentry->d_parent != new_dentry->d_parent) {
9452                 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9453                                 BTRFS_I(old_inode), 1);
9454                 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9455                                 BTRFS_I(new_inode), 1);
9456         }
9457
9458         /* src is a subvolume */
9459         if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9460                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9461                 ret = btrfs_unlink_subvol(trans, old_dir, root_objectid,
9462                                           old_dentry->d_name.name,
9463                                           old_dentry->d_name.len);
9464         } else { /* src is an inode */
9465                 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9466                                            BTRFS_I(old_dentry->d_inode),
9467                                            old_dentry->d_name.name,
9468                                            old_dentry->d_name.len);
9469                 if (!ret)
9470                         ret = btrfs_update_inode(trans, root, old_inode);
9471         }
9472         if (ret) {
9473                 btrfs_abort_transaction(trans, ret);
9474                 goto out_fail;
9475         }
9476
9477         /* dest is a subvolume */
9478         if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9479                 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9480                 ret = btrfs_unlink_subvol(trans, new_dir, root_objectid,
9481                                           new_dentry->d_name.name,
9482                                           new_dentry->d_name.len);
9483         } else { /* dest is an inode */
9484                 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9485                                            BTRFS_I(new_dentry->d_inode),
9486                                            new_dentry->d_name.name,
9487                                            new_dentry->d_name.len);
9488                 if (!ret)
9489                         ret = btrfs_update_inode(trans, dest, new_inode);
9490         }
9491         if (ret) {
9492                 btrfs_abort_transaction(trans, ret);
9493                 goto out_fail;
9494         }
9495
9496         ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9497                              new_dentry->d_name.name,
9498                              new_dentry->d_name.len, 0, old_idx);
9499         if (ret) {
9500                 btrfs_abort_transaction(trans, ret);
9501                 goto out_fail;
9502         }
9503
9504         ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9505                              old_dentry->d_name.name,
9506                              old_dentry->d_name.len, 0, new_idx);
9507         if (ret) {
9508                 btrfs_abort_transaction(trans, ret);
9509                 goto out_fail;
9510         }
9511
9512         if (old_inode->i_nlink == 1)
9513                 BTRFS_I(old_inode)->dir_index = old_idx;
9514         if (new_inode->i_nlink == 1)
9515                 BTRFS_I(new_inode)->dir_index = new_idx;
9516
9517         if (root_log_pinned) {
9518                 parent = new_dentry->d_parent;
9519                 ret = btrfs_log_new_name(trans, BTRFS_I(old_inode),
9520                                          BTRFS_I(old_dir), parent,
9521                                          false, &ctx_root);
9522                 if (ret == BTRFS_NEED_LOG_SYNC)
9523                         sync_log_root = true;
9524                 else if (ret == BTRFS_NEED_TRANS_COMMIT)
9525                         commit_transaction = true;
9526                 ret = 0;
9527                 btrfs_end_log_trans(root);
9528                 root_log_pinned = false;
9529         }
9530         if (dest_log_pinned) {
9531                 if (!commit_transaction) {
9532                         parent = old_dentry->d_parent;
9533                         ret = btrfs_log_new_name(trans, BTRFS_I(new_inode),
9534                                                  BTRFS_I(new_dir), parent,
9535                                                  false, &ctx_dest);
9536                         if (ret == BTRFS_NEED_LOG_SYNC)
9537                                 sync_log_dest = true;
9538                         else if (ret == BTRFS_NEED_TRANS_COMMIT)
9539                                 commit_transaction = true;
9540                         ret = 0;
9541                 }
9542                 btrfs_end_log_trans(dest);
9543                 dest_log_pinned = false;
9544         }
9545 out_fail:
9546         /*
9547          * If we have pinned a log and an error happened, we unpin tasks
9548          * trying to sync the log and force them to fallback to a transaction
9549          * commit if the log currently contains any of the inodes involved in
9550          * this rename operation (to ensure we do not persist a log with an
9551          * inconsistent state for any of these inodes or leading to any
9552          * inconsistencies when replayed). If the transaction was aborted, the
9553          * abortion reason is propagated to userspace when attempting to commit
9554          * the transaction. If the log does not contain any of these inodes, we
9555          * allow the tasks to sync it.
9556          */
9557         if (ret && (root_log_pinned || dest_log_pinned)) {
9558                 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9559                     btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9560                     btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9561                     (new_inode &&
9562                      btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9563                         btrfs_set_log_full_commit(fs_info, trans);
9564
9565                 if (root_log_pinned) {
9566                         btrfs_end_log_trans(root);
9567                         root_log_pinned = false;
9568                 }
9569                 if (dest_log_pinned) {
9570                         btrfs_end_log_trans(dest);
9571                         dest_log_pinned = false;
9572                 }
9573         }
9574         if (!ret && sync_log_root && !commit_transaction) {
9575                 ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root,
9576                                      &ctx_root);
9577                 if (ret)
9578                         commit_transaction = true;
9579         }
9580         if (!ret && sync_log_dest && !commit_transaction) {
9581                 ret = btrfs_sync_log(trans, BTRFS_I(new_inode)->root,
9582                                      &ctx_dest);
9583                 if (ret)
9584                         commit_transaction = true;
9585         }
9586         if (commit_transaction) {
9587                 ret = btrfs_commit_transaction(trans);
9588         } else {
9589                 int ret2;
9590
9591                 ret2 = btrfs_end_transaction(trans);
9592                 ret = ret ? ret : ret2;
9593         }
9594 out_notrans:
9595         if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9596                 up_read(&fs_info->subvol_sem);
9597         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9598                 up_read(&fs_info->subvol_sem);
9599
9600         return ret;
9601 }
9602
9603 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9604                                      struct btrfs_root *root,
9605                                      struct inode *dir,
9606                                      struct dentry *dentry)
9607 {
9608         int ret;
9609         struct inode *inode;
9610         u64 objectid;
9611         u64 index;
9612
9613         ret = btrfs_find_free_ino(root, &objectid);
9614         if (ret)
9615                 return ret;
9616
9617         inode = btrfs_new_inode(trans, root, dir,
9618                                 dentry->d_name.name,
9619                                 dentry->d_name.len,
9620                                 btrfs_ino(BTRFS_I(dir)),
9621                                 objectid,
9622                                 S_IFCHR | WHITEOUT_MODE,
9623                                 &index);
9624
9625         if (IS_ERR(inode)) {
9626                 ret = PTR_ERR(inode);
9627                 return ret;
9628         }
9629
9630         inode->i_op = &btrfs_special_inode_operations;
9631         init_special_inode(inode, inode->i_mode,
9632                 WHITEOUT_DEV);
9633
9634         ret = btrfs_init_inode_security(trans, inode, dir,
9635                                 &dentry->d_name);
9636         if (ret)
9637                 goto out;
9638
9639         ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9640                                 BTRFS_I(inode), 0, index);
9641         if (ret)
9642                 goto out;
9643
9644         ret = btrfs_update_inode(trans, root, inode);
9645 out:
9646         unlock_new_inode(inode);
9647         if (ret)
9648                 inode_dec_link_count(inode);
9649         iput(inode);
9650
9651         return ret;
9652 }
9653
9654 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9655                            struct inode *new_dir, struct dentry *new_dentry,
9656                            unsigned int flags)
9657 {
9658         struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9659         struct btrfs_trans_handle *trans;
9660         unsigned int trans_num_items;
9661         struct btrfs_root *root = BTRFS_I(old_dir)->root;
9662         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9663         struct inode *new_inode = d_inode(new_dentry);
9664         struct inode *old_inode = d_inode(old_dentry);
9665         u64 index = 0;
9666         u64 root_objectid;
9667         int ret;
9668         u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9669         bool log_pinned = false;
9670         struct btrfs_log_ctx ctx;
9671         bool sync_log = false;
9672         bool commit_transaction = false;
9673
9674         if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9675                 return -EPERM;
9676
9677         /* we only allow rename subvolume link between subvolumes */
9678         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9679                 return -EXDEV;
9680
9681         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9682             (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9683                 return -ENOTEMPTY;
9684
9685         if (S_ISDIR(old_inode->i_mode) && new_inode &&
9686             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9687                 return -ENOTEMPTY;
9688
9689
9690         /* check for collisions, even if the  name isn't there */
9691         ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9692                              new_dentry->d_name.name,
9693                              new_dentry->d_name.len);
9694
9695         if (ret) {
9696                 if (ret == -EEXIST) {
9697                         /* we shouldn't get
9698                          * eexist without a new_inode */
9699                         if (WARN_ON(!new_inode)) {
9700                                 return ret;
9701                         }
9702                 } else {
9703                         /* maybe -EOVERFLOW */
9704                         return ret;
9705                 }
9706         }
9707         ret = 0;
9708
9709         /*
9710          * we're using rename to replace one file with another.  Start IO on it
9711          * now so  we don't add too much work to the end of the transaction
9712          */
9713         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9714                 filemap_flush(old_inode->i_mapping);
9715
9716         /* close the racy window with snapshot create/destroy ioctl */
9717         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9718                 down_read(&fs_info->subvol_sem);
9719         /*
9720          * We want to reserve the absolute worst case amount of items.  So if
9721          * both inodes are subvols and we need to unlink them then that would
9722          * require 4 item modifications, but if they are both normal inodes it
9723          * would require 5 item modifications, so we'll assume they are normal
9724          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9725          * should cover the worst case number of items we'll modify.
9726          * If our rename has the whiteout flag, we need more 5 units for the
9727          * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9728          * when selinux is enabled).
9729          */
9730         trans_num_items = 11;
9731         if (flags & RENAME_WHITEOUT)
9732                 trans_num_items += 5;
9733         trans = btrfs_start_transaction(root, trans_num_items);
9734         if (IS_ERR(trans)) {
9735                 ret = PTR_ERR(trans);
9736                 goto out_notrans;
9737         }
9738
9739         if (dest != root)
9740                 btrfs_record_root_in_trans(trans, dest);
9741
9742         ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9743         if (ret)
9744                 goto out_fail;
9745
9746         BTRFS_I(old_inode)->dir_index = 0ULL;
9747         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9748                 /* force full log commit if subvolume involved. */
9749                 btrfs_set_log_full_commit(fs_info, trans);
9750         } else {
9751                 btrfs_pin_log_trans(root);
9752                 log_pinned = true;
9753                 ret = btrfs_insert_inode_ref(trans, dest,
9754                                              new_dentry->d_name.name,
9755                                              new_dentry->d_name.len,
9756                                              old_ino,
9757                                              btrfs_ino(BTRFS_I(new_dir)), index);
9758                 if (ret)
9759                         goto out_fail;
9760         }
9761
9762         inode_inc_iversion(old_dir);
9763         inode_inc_iversion(new_dir);
9764         inode_inc_iversion(old_inode);
9765         old_dir->i_ctime = old_dir->i_mtime =
9766         new_dir->i_ctime = new_dir->i_mtime =
9767         old_inode->i_ctime = current_time(old_dir);
9768
9769         if (old_dentry->d_parent != new_dentry->d_parent)
9770                 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9771                                 BTRFS_I(old_inode), 1);
9772
9773         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9774                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9775                 ret = btrfs_unlink_subvol(trans, old_dir, root_objectid,
9776                                         old_dentry->d_name.name,
9777                                         old_dentry->d_name.len);
9778         } else {
9779                 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9780                                         BTRFS_I(d_inode(old_dentry)),
9781                                         old_dentry->d_name.name,
9782                                         old_dentry->d_name.len);
9783                 if (!ret)
9784                         ret = btrfs_update_inode(trans, root, old_inode);
9785         }
9786         if (ret) {
9787                 btrfs_abort_transaction(trans, ret);
9788                 goto out_fail;
9789         }
9790
9791         if (new_inode) {
9792                 inode_inc_iversion(new_inode);
9793                 new_inode->i_ctime = current_time(new_inode);
9794                 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9795                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9796                         root_objectid = BTRFS_I(new_inode)->location.objectid;
9797                         ret = btrfs_unlink_subvol(trans, new_dir, root_objectid,
9798                                                 new_dentry->d_name.name,
9799                                                 new_dentry->d_name.len);
9800                         BUG_ON(new_inode->i_nlink == 0);
9801                 } else {
9802                         ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9803                                                  BTRFS_I(d_inode(new_dentry)),
9804                                                  new_dentry->d_name.name,
9805                                                  new_dentry->d_name.len);
9806                 }
9807                 if (!ret && new_inode->i_nlink == 0)
9808                         ret = btrfs_orphan_add(trans,
9809                                         BTRFS_I(d_inode(new_dentry)));
9810                 if (ret) {
9811                         btrfs_abort_transaction(trans, ret);
9812                         goto out_fail;
9813                 }
9814         }
9815
9816         ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9817                              new_dentry->d_name.name,
9818                              new_dentry->d_name.len, 0, index);
9819         if (ret) {
9820                 btrfs_abort_transaction(trans, ret);
9821                 goto out_fail;
9822         }
9823
9824         if (old_inode->i_nlink == 1)
9825                 BTRFS_I(old_inode)->dir_index = index;
9826
9827         if (log_pinned) {
9828                 struct dentry *parent = new_dentry->d_parent;
9829
9830                 btrfs_init_log_ctx(&ctx, old_inode);
9831                 ret = btrfs_log_new_name(trans, BTRFS_I(old_inode),
9832                                          BTRFS_I(old_dir), parent,
9833                                          false, &ctx);
9834                 if (ret == BTRFS_NEED_LOG_SYNC)
9835                         sync_log = true;
9836                 else if (ret == BTRFS_NEED_TRANS_COMMIT)
9837                         commit_transaction = true;
9838                 ret = 0;
9839                 btrfs_end_log_trans(root);
9840                 log_pinned = false;
9841         }
9842
9843         if (flags & RENAME_WHITEOUT) {
9844                 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9845                                                 old_dentry);
9846
9847                 if (ret) {
9848                         btrfs_abort_transaction(trans, ret);
9849                         goto out_fail;
9850                 }
9851         }
9852 out_fail:
9853         /*
9854          * If we have pinned the log and an error happened, we unpin tasks
9855          * trying to sync the log and force them to fallback to a transaction
9856          * commit if the log currently contains any of the inodes involved in
9857          * this rename operation (to ensure we do not persist a log with an
9858          * inconsistent state for any of these inodes or leading to any
9859          * inconsistencies when replayed). If the transaction was aborted, the
9860          * abortion reason is propagated to userspace when attempting to commit
9861          * the transaction. If the log does not contain any of these inodes, we
9862          * allow the tasks to sync it.
9863          */
9864         if (ret && log_pinned) {
9865                 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9866                     btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9867                     btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9868                     (new_inode &&
9869                      btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9870                         btrfs_set_log_full_commit(fs_info, trans);
9871
9872                 btrfs_end_log_trans(root);
9873                 log_pinned = false;
9874         }
9875         if (!ret && sync_log) {
9876                 ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root, &ctx);
9877                 if (ret)
9878                         commit_transaction = true;
9879         }
9880         if (commit_transaction) {
9881                 ret = btrfs_commit_transaction(trans);
9882         } else {
9883                 int ret2;
9884
9885                 ret2 = btrfs_end_transaction(trans);
9886                 ret = ret ? ret : ret2;
9887         }
9888 out_notrans:
9889         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9890                 up_read(&fs_info->subvol_sem);
9891
9892         return ret;
9893 }
9894
9895 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9896                          struct inode *new_dir, struct dentry *new_dentry,
9897                          unsigned int flags)
9898 {
9899         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9900                 return -EINVAL;
9901
9902         if (flags & RENAME_EXCHANGE)
9903                 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9904                                           new_dentry);
9905
9906         return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9907 }
9908
9909 struct btrfs_delalloc_work {
9910         struct inode *inode;
9911         struct completion completion;
9912         struct list_head list;
9913         struct btrfs_work work;
9914 };
9915
9916 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9917 {
9918         struct btrfs_delalloc_work *delalloc_work;
9919         struct inode *inode;
9920
9921         delalloc_work = container_of(work, struct btrfs_delalloc_work,
9922                                      work);
9923         inode = delalloc_work->inode;
9924         filemap_flush(inode->i_mapping);
9925         if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9926                                 &BTRFS_I(inode)->runtime_flags))
9927                 filemap_flush(inode->i_mapping);
9928
9929         iput(inode);
9930         complete(&delalloc_work->completion);
9931 }
9932
9933 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9934 {
9935         struct btrfs_delalloc_work *work;
9936
9937         work = kmalloc(sizeof(*work), GFP_NOFS);
9938         if (!work)
9939                 return NULL;
9940
9941         init_completion(&work->completion);
9942         INIT_LIST_HEAD(&work->list);
9943         work->inode = inode;
9944         WARN_ON_ONCE(!inode);
9945         btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9946                         btrfs_run_delalloc_work, NULL, NULL);
9947
9948         return work;
9949 }
9950
9951 /*
9952  * some fairly slow code that needs optimization. This walks the list
9953  * of all the inodes with pending delalloc and forces them to disk.
9954  */
9955 static int start_delalloc_inodes(struct btrfs_root *root, int nr, bool snapshot)
9956 {
9957         struct btrfs_inode *binode;
9958         struct inode *inode;
9959         struct btrfs_delalloc_work *work, *next;
9960         struct list_head works;
9961         struct list_head splice;
9962         int ret = 0;
9963
9964         INIT_LIST_HEAD(&works);
9965         INIT_LIST_HEAD(&splice);
9966
9967         mutex_lock(&root->delalloc_mutex);
9968         spin_lock(&root->delalloc_lock);
9969         list_splice_init(&root->delalloc_inodes, &splice);
9970         while (!list_empty(&splice)) {
9971                 binode = list_entry(splice.next, struct btrfs_inode,
9972                                     delalloc_inodes);
9973
9974                 list_move_tail(&binode->delalloc_inodes,
9975                                &root->delalloc_inodes);
9976                 inode = igrab(&binode->vfs_inode);
9977                 if (!inode) {
9978                         cond_resched_lock(&root->delalloc_lock);
9979                         continue;
9980                 }
9981                 spin_unlock(&root->delalloc_lock);
9982
9983                 if (snapshot)
9984                         set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9985                                 &binode->runtime_flags);
9986                 work = btrfs_alloc_delalloc_work(inode);
9987                 if (!work) {
9988                         iput(inode);
9989                         ret = -ENOMEM;
9990                         goto out;
9991                 }
9992                 list_add_tail(&work->list, &works);
9993                 btrfs_queue_work(root->fs_info->flush_workers,
9994                                  &work->work);
9995                 ret++;
9996                 if (nr != -1 && ret >= nr)
9997                         goto out;
9998                 cond_resched();
9999                 spin_lock(&root->delalloc_lock);
10000         }
10001         spin_unlock(&root->delalloc_lock);
10002
10003 out:
10004         list_for_each_entry_safe(work, next, &works, list) {
10005                 list_del_init(&work->list);
10006                 wait_for_completion(&work->completion);
10007                 kfree(work);
10008         }
10009
10010         if (!list_empty(&splice)) {
10011                 spin_lock(&root->delalloc_lock);
10012                 list_splice_tail(&splice, &root->delalloc_inodes);
10013                 spin_unlock(&root->delalloc_lock);
10014         }
10015         mutex_unlock(&root->delalloc_mutex);
10016         return ret;
10017 }
10018
10019 int btrfs_start_delalloc_snapshot(struct btrfs_root *root)
10020 {
10021         struct btrfs_fs_info *fs_info = root->fs_info;
10022         int ret;
10023
10024         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10025                 return -EROFS;
10026
10027         ret = start_delalloc_inodes(root, -1, true);
10028         if (ret > 0)
10029                 ret = 0;
10030         return ret;
10031 }
10032
10033 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int nr)
10034 {
10035         struct btrfs_root *root;
10036         struct list_head splice;
10037         int ret;
10038
10039         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10040                 return -EROFS;
10041
10042         INIT_LIST_HEAD(&splice);
10043
10044         mutex_lock(&fs_info->delalloc_root_mutex);
10045         spin_lock(&fs_info->delalloc_root_lock);
10046         list_splice_init(&fs_info->delalloc_roots, &splice);
10047         while (!list_empty(&splice) && nr) {
10048                 root = list_first_entry(&splice, struct btrfs_root,
10049                                         delalloc_root);
10050                 root = btrfs_grab_fs_root(root);
10051                 BUG_ON(!root);
10052                 list_move_tail(&root->delalloc_root,
10053                                &fs_info->delalloc_roots);
10054                 spin_unlock(&fs_info->delalloc_root_lock);
10055
10056                 ret = start_delalloc_inodes(root, nr, false);
10057                 btrfs_put_fs_root(root);
10058                 if (ret < 0)
10059                         goto out;
10060
10061                 if (nr != -1) {
10062                         nr -= ret;
10063                         WARN_ON(nr < 0);
10064                 }
10065                 spin_lock(&fs_info->delalloc_root_lock);
10066         }
10067         spin_unlock(&fs_info->delalloc_root_lock);
10068
10069         ret = 0;
10070 out:
10071         if (!list_empty(&splice)) {
10072                 spin_lock(&fs_info->delalloc_root_lock);
10073                 list_splice_tail(&splice, &fs_info->delalloc_roots);
10074                 spin_unlock(&fs_info->delalloc_root_lock);
10075         }
10076         mutex_unlock(&fs_info->delalloc_root_mutex);
10077         return ret;
10078 }
10079
10080 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10081                          const char *symname)
10082 {
10083         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10084         struct btrfs_trans_handle *trans;
10085         struct btrfs_root *root = BTRFS_I(dir)->root;
10086         struct btrfs_path *path;
10087         struct btrfs_key key;
10088         struct inode *inode = NULL;
10089         int err;
10090         u64 objectid;
10091         u64 index = 0;
10092         int name_len;
10093         int datasize;
10094         unsigned long ptr;
10095         struct btrfs_file_extent_item *ei;
10096         struct extent_buffer *leaf;
10097
10098         name_len = strlen(symname);
10099         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10100                 return -ENAMETOOLONG;
10101
10102         /*
10103          * 2 items for inode item and ref
10104          * 2 items for dir items
10105          * 1 item for updating parent inode item
10106          * 1 item for the inline extent item
10107          * 1 item for xattr if selinux is on
10108          */
10109         trans = btrfs_start_transaction(root, 7);
10110         if (IS_ERR(trans))
10111                 return PTR_ERR(trans);
10112
10113         err = btrfs_find_free_ino(root, &objectid);
10114         if (err)
10115                 goto out_unlock;
10116
10117         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10118                                 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10119                                 objectid, S_IFLNK|S_IRWXUGO, &index);
10120         if (IS_ERR(inode)) {
10121                 err = PTR_ERR(inode);
10122                 inode = NULL;
10123                 goto out_unlock;
10124         }
10125
10126         /*
10127         * If the active LSM wants to access the inode during
10128         * d_instantiate it needs these. Smack checks to see
10129         * if the filesystem supports xattrs by looking at the
10130         * ops vector.
10131         */
10132         inode->i_fop = &btrfs_file_operations;
10133         inode->i_op = &btrfs_file_inode_operations;
10134         inode->i_mapping->a_ops = &btrfs_aops;
10135         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10136
10137         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10138         if (err)
10139                 goto out_unlock;
10140
10141         path = btrfs_alloc_path();
10142         if (!path) {
10143                 err = -ENOMEM;
10144                 goto out_unlock;
10145         }
10146         key.objectid = btrfs_ino(BTRFS_I(inode));
10147         key.offset = 0;
10148         key.type = BTRFS_EXTENT_DATA_KEY;
10149         datasize = btrfs_file_extent_calc_inline_size(name_len);
10150         err = btrfs_insert_empty_item(trans, root, path, &key,
10151                                       datasize);
10152         if (err) {
10153                 btrfs_free_path(path);
10154                 goto out_unlock;
10155         }
10156         leaf = path->nodes[0];
10157         ei = btrfs_item_ptr(leaf, path->slots[0],
10158                             struct btrfs_file_extent_item);
10159         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10160         btrfs_set_file_extent_type(leaf, ei,
10161                                    BTRFS_FILE_EXTENT_INLINE);
10162         btrfs_set_file_extent_encryption(leaf, ei, 0);
10163         btrfs_set_file_extent_compression(leaf, ei, 0);
10164         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10165         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10166
10167         ptr = btrfs_file_extent_inline_start(ei);
10168         write_extent_buffer(leaf, symname, ptr, name_len);
10169         btrfs_mark_buffer_dirty(leaf);
10170         btrfs_free_path(path);
10171
10172         inode->i_op = &btrfs_symlink_inode_operations;
10173         inode_nohighmem(inode);
10174         inode->i_mapping->a_ops = &btrfs_aops;
10175         inode_set_bytes(inode, name_len);
10176         btrfs_i_size_write(BTRFS_I(inode), name_len);
10177         err = btrfs_update_inode(trans, root, inode);
10178         /*
10179          * Last step, add directory indexes for our symlink inode. This is the
10180          * last step to avoid extra cleanup of these indexes if an error happens
10181          * elsewhere above.
10182          */
10183         if (!err)
10184                 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10185                                 BTRFS_I(inode), 0, index);
10186         if (err)
10187                 goto out_unlock;
10188
10189         d_instantiate_new(dentry, inode);
10190
10191 out_unlock:
10192         btrfs_end_transaction(trans);
10193         if (err && inode) {
10194                 inode_dec_link_count(inode);
10195                 discard_new_inode(inode);
10196         }
10197         btrfs_btree_balance_dirty(fs_info);
10198         return err;
10199 }
10200
10201 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10202                                        u64 start, u64 num_bytes, u64 min_size,
10203                                        loff_t actual_len, u64 *alloc_hint,
10204                                        struct btrfs_trans_handle *trans)
10205 {
10206         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10207         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10208         struct extent_map *em;
10209         struct btrfs_root *root = BTRFS_I(inode)->root;
10210         struct btrfs_key ins;
10211         u64 cur_offset = start;
10212         u64 i_size;
10213         u64 cur_bytes;
10214         u64 last_alloc = (u64)-1;
10215         int ret = 0;
10216         bool own_trans = true;
10217         u64 end = start + num_bytes - 1;
10218
10219         if (trans)
10220                 own_trans = false;
10221         while (num_bytes > 0) {
10222                 if (own_trans) {
10223                         trans = btrfs_start_transaction(root, 3);
10224                         if (IS_ERR(trans)) {
10225                                 ret = PTR_ERR(trans);
10226                                 break;
10227                         }
10228                 }
10229
10230                 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10231                 cur_bytes = max(cur_bytes, min_size);
10232                 /*
10233                  * If we are severely fragmented we could end up with really
10234                  * small allocations, so if the allocator is returning small
10235                  * chunks lets make its job easier by only searching for those
10236                  * sized chunks.
10237                  */
10238                 cur_bytes = min(cur_bytes, last_alloc);
10239                 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10240                                 min_size, 0, *alloc_hint, &ins, 1, 0);
10241                 if (ret) {
10242                         if (own_trans)
10243                                 btrfs_end_transaction(trans);
10244                         break;
10245                 }
10246                 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10247
10248                 last_alloc = ins.offset;
10249                 ret = insert_reserved_file_extent(trans, inode,
10250                                                   cur_offset, ins.objectid,
10251                                                   ins.offset, ins.offset,
10252                                                   ins.offset, 0, 0, 0,
10253                                                   BTRFS_FILE_EXTENT_PREALLOC);
10254                 if (ret) {
10255                         btrfs_free_reserved_extent(fs_info, ins.objectid,
10256                                                    ins.offset, 0);
10257                         btrfs_abort_transaction(trans, ret);
10258                         if (own_trans)
10259                                 btrfs_end_transaction(trans);
10260                         break;
10261                 }
10262
10263                 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10264                                         cur_offset + ins.offset -1, 0);
10265
10266                 em = alloc_extent_map();
10267                 if (!em) {
10268                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10269                                 &BTRFS_I(inode)->runtime_flags);
10270                         goto next;
10271                 }
10272
10273                 em->start = cur_offset;
10274                 em->orig_start = cur_offset;
10275                 em->len = ins.offset;
10276                 em->block_start = ins.objectid;
10277                 em->block_len = ins.offset;
10278                 em->orig_block_len = ins.offset;
10279                 em->ram_bytes = ins.offset;
10280                 em->bdev = fs_info->fs_devices->latest_bdev;
10281                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10282                 em->generation = trans->transid;
10283
10284                 while (1) {
10285                         write_lock(&em_tree->lock);
10286                         ret = add_extent_mapping(em_tree, em, 1);
10287                         write_unlock(&em_tree->lock);
10288                         if (ret != -EEXIST)
10289                                 break;
10290                         btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10291                                                 cur_offset + ins.offset - 1,
10292                                                 0);
10293                 }
10294                 free_extent_map(em);
10295 next:
10296                 num_bytes -= ins.offset;
10297                 cur_offset += ins.offset;
10298                 *alloc_hint = ins.objectid + ins.offset;
10299
10300                 inode_inc_iversion(inode);
10301                 inode->i_ctime = current_time(inode);
10302                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10303                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10304                     (actual_len > inode->i_size) &&
10305                     (cur_offset > inode->i_size)) {
10306                         if (cur_offset > actual_len)
10307                                 i_size = actual_len;
10308                         else
10309                                 i_size = cur_offset;
10310                         i_size_write(inode, i_size);
10311                         btrfs_ordered_update_i_size(inode, i_size, NULL);
10312                 }
10313
10314                 ret = btrfs_update_inode(trans, root, inode);
10315
10316                 if (ret) {
10317                         btrfs_abort_transaction(trans, ret);
10318                         if (own_trans)
10319                                 btrfs_end_transaction(trans);
10320                         break;
10321                 }
10322
10323                 if (own_trans)
10324                         btrfs_end_transaction(trans);
10325         }
10326         if (cur_offset < end)
10327                 btrfs_free_reserved_data_space(inode, NULL, cur_offset,
10328                         end - cur_offset + 1);
10329         return ret;
10330 }
10331
10332 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10333                               u64 start, u64 num_bytes, u64 min_size,
10334                               loff_t actual_len, u64 *alloc_hint)
10335 {
10336         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10337                                            min_size, actual_len, alloc_hint,
10338                                            NULL);
10339 }
10340
10341 int btrfs_prealloc_file_range_trans(struct inode *inode,
10342                                     struct btrfs_trans_handle *trans, int mode,
10343                                     u64 start, u64 num_bytes, u64 min_size,
10344                                     loff_t actual_len, u64 *alloc_hint)
10345 {
10346         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10347                                            min_size, actual_len, alloc_hint, trans);
10348 }
10349
10350 static int btrfs_set_page_dirty(struct page *page)
10351 {
10352         return __set_page_dirty_nobuffers(page);
10353 }
10354
10355 static int btrfs_permission(struct inode *inode, int mask)
10356 {
10357         struct btrfs_root *root = BTRFS_I(inode)->root;
10358         umode_t mode = inode->i_mode;
10359
10360         if (mask & MAY_WRITE &&
10361             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10362                 if (btrfs_root_readonly(root))
10363                         return -EROFS;
10364                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10365                         return -EACCES;
10366         }
10367         return generic_permission(inode, mask);
10368 }
10369
10370 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10371 {
10372         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10373         struct btrfs_trans_handle *trans;
10374         struct btrfs_root *root = BTRFS_I(dir)->root;
10375         struct inode *inode = NULL;
10376         u64 objectid;
10377         u64 index;
10378         int ret = 0;
10379
10380         /*
10381          * 5 units required for adding orphan entry
10382          */
10383         trans = btrfs_start_transaction(root, 5);
10384         if (IS_ERR(trans))
10385                 return PTR_ERR(trans);
10386
10387         ret = btrfs_find_free_ino(root, &objectid);
10388         if (ret)
10389                 goto out;
10390
10391         inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10392                         btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10393         if (IS_ERR(inode)) {
10394                 ret = PTR_ERR(inode);
10395                 inode = NULL;
10396                 goto out;
10397         }
10398
10399         inode->i_fop = &btrfs_file_operations;
10400         inode->i_op = &btrfs_file_inode_operations;
10401
10402         inode->i_mapping->a_ops = &btrfs_aops;
10403         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10404
10405         ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10406         if (ret)
10407                 goto out;
10408
10409         ret = btrfs_update_inode(trans, root, inode);
10410         if (ret)
10411                 goto out;
10412         ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10413         if (ret)
10414                 goto out;
10415
10416         /*
10417          * We set number of links to 0 in btrfs_new_inode(), and here we set
10418          * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10419          * through:
10420          *
10421          *    d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10422          */
10423         set_nlink(inode, 1);
10424         d_tmpfile(dentry, inode);
10425         unlock_new_inode(inode);
10426         mark_inode_dirty(inode);
10427 out:
10428         btrfs_end_transaction(trans);
10429         if (ret && inode)
10430                 discard_new_inode(inode);
10431         btrfs_btree_balance_dirty(fs_info);
10432         return ret;
10433 }
10434
10435 void btrfs_set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
10436 {
10437         struct inode *inode = tree->private_data;
10438         unsigned long index = start >> PAGE_SHIFT;
10439         unsigned long end_index = end >> PAGE_SHIFT;
10440         struct page *page;
10441
10442         while (index <= end_index) {
10443                 page = find_get_page(inode->i_mapping, index);
10444                 ASSERT(page); /* Pages should be in the extent_io_tree */
10445                 set_page_writeback(page);
10446                 put_page(page);
10447                 index++;
10448         }
10449 }
10450
10451 #ifdef CONFIG_SWAP
10452 /*
10453  * Add an entry indicating a block group or device which is pinned by a
10454  * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10455  * negative errno on failure.
10456  */
10457 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10458                                   bool is_block_group)
10459 {
10460         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10461         struct btrfs_swapfile_pin *sp, *entry;
10462         struct rb_node **p;
10463         struct rb_node *parent = NULL;
10464
10465         sp = kmalloc(sizeof(*sp), GFP_NOFS);
10466         if (!sp)
10467                 return -ENOMEM;
10468         sp->ptr = ptr;
10469         sp->inode = inode;
10470         sp->is_block_group = is_block_group;
10471
10472         spin_lock(&fs_info->swapfile_pins_lock);
10473         p = &fs_info->swapfile_pins.rb_node;
10474         while (*p) {
10475                 parent = *p;
10476                 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10477                 if (sp->ptr < entry->ptr ||
10478                     (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10479                         p = &(*p)->rb_left;
10480                 } else if (sp->ptr > entry->ptr ||
10481                            (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10482                         p = &(*p)->rb_right;
10483                 } else {
10484                         spin_unlock(&fs_info->swapfile_pins_lock);
10485                         kfree(sp);
10486                         return 1;
10487                 }
10488         }
10489         rb_link_node(&sp->node, parent, p);
10490         rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10491         spin_unlock(&fs_info->swapfile_pins_lock);
10492         return 0;
10493 }
10494
10495 /* Free all of the entries pinned by this swapfile. */
10496 static void btrfs_free_swapfile_pins(struct inode *inode)
10497 {
10498         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10499         struct btrfs_swapfile_pin *sp;
10500         struct rb_node *node, *next;
10501
10502         spin_lock(&fs_info->swapfile_pins_lock);
10503         node = rb_first(&fs_info->swapfile_pins);
10504         while (node) {
10505                 next = rb_next(node);
10506                 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10507                 if (sp->inode == inode) {
10508                         rb_erase(&sp->node, &fs_info->swapfile_pins);
10509                         if (sp->is_block_group)
10510                                 btrfs_put_block_group(sp->ptr);
10511                         kfree(sp);
10512                 }
10513                 node = next;
10514         }
10515         spin_unlock(&fs_info->swapfile_pins_lock);
10516 }
10517
10518 struct btrfs_swap_info {
10519         u64 start;
10520         u64 block_start;
10521         u64 block_len;
10522         u64 lowest_ppage;
10523         u64 highest_ppage;
10524         unsigned long nr_pages;
10525         int nr_extents;
10526 };
10527
10528 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10529                                  struct btrfs_swap_info *bsi)
10530 {
10531         unsigned long nr_pages;
10532         u64 first_ppage, first_ppage_reported, next_ppage;
10533         int ret;
10534
10535         first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT;
10536         next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len,
10537                                 PAGE_SIZE) >> PAGE_SHIFT;
10538
10539         if (first_ppage >= next_ppage)
10540                 return 0;
10541         nr_pages = next_ppage - first_ppage;
10542
10543         first_ppage_reported = first_ppage;
10544         if (bsi->start == 0)
10545                 first_ppage_reported++;
10546         if (bsi->lowest_ppage > first_ppage_reported)
10547                 bsi->lowest_ppage = first_ppage_reported;
10548         if (bsi->highest_ppage < (next_ppage - 1))
10549                 bsi->highest_ppage = next_ppage - 1;
10550
10551         ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10552         if (ret < 0)
10553                 return ret;
10554         bsi->nr_extents += ret;
10555         bsi->nr_pages += nr_pages;
10556         return 0;
10557 }
10558
10559 static void btrfs_swap_deactivate(struct file *file)
10560 {
10561         struct inode *inode = file_inode(file);
10562
10563         btrfs_free_swapfile_pins(inode);
10564         atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10565 }
10566
10567 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10568                                sector_t *span)
10569 {
10570         struct inode *inode = file_inode(file);
10571         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10572         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10573         struct extent_state *cached_state = NULL;
10574         struct extent_map *em = NULL;
10575         struct btrfs_device *device = NULL;
10576         struct btrfs_swap_info bsi = {
10577                 .lowest_ppage = (sector_t)-1ULL,
10578         };
10579         int ret = 0;
10580         u64 isize;
10581         u64 start;
10582
10583         /*
10584          * If the swap file was just created, make sure delalloc is done. If the
10585          * file changes again after this, the user is doing something stupid and
10586          * we don't really care.
10587          */
10588         ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10589         if (ret)
10590                 return ret;
10591
10592         /*
10593          * The inode is locked, so these flags won't change after we check them.
10594          */
10595         if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10596                 btrfs_warn(fs_info, "swapfile must not be compressed");
10597                 return -EINVAL;
10598         }
10599         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10600                 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10601                 return -EINVAL;
10602         }
10603         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10604                 btrfs_warn(fs_info, "swapfile must not be checksummed");
10605                 return -EINVAL;
10606         }
10607
10608         /*
10609          * Balance or device remove/replace/resize can move stuff around from
10610          * under us. The EXCL_OP flag makes sure they aren't running/won't run
10611          * concurrently while we are mapping the swap extents, and
10612          * fs_info->swapfile_pins prevents them from running while the swap file
10613          * is active and moving the extents. Note that this also prevents a
10614          * concurrent device add which isn't actually necessary, but it's not
10615          * really worth the trouble to allow it.
10616          */
10617         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
10618                 btrfs_warn(fs_info,
10619            "cannot activate swapfile while exclusive operation is running");
10620                 return -EBUSY;
10621         }
10622         /*
10623          * Snapshots can create extents which require COW even if NODATACOW is
10624          * set. We use this counter to prevent snapshots. We must increment it
10625          * before walking the extents because we don't want a concurrent
10626          * snapshot to run after we've already checked the extents.
10627          */
10628         atomic_inc(&BTRFS_I(inode)->root->nr_swapfiles);
10629
10630         isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10631
10632         lock_extent_bits(io_tree, 0, isize - 1, &cached_state);
10633         start = 0;
10634         while (start < isize) {
10635                 u64 logical_block_start, physical_block_start;
10636                 struct btrfs_block_group_cache *bg;
10637                 u64 len = isize - start;
10638
10639                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
10640                 if (IS_ERR(em)) {
10641                         ret = PTR_ERR(em);
10642                         goto out;
10643                 }
10644
10645                 if (em->block_start == EXTENT_MAP_HOLE) {
10646                         btrfs_warn(fs_info, "swapfile must not have holes");
10647                         ret = -EINVAL;
10648                         goto out;
10649                 }
10650                 if (em->block_start == EXTENT_MAP_INLINE) {
10651                         /*
10652                          * It's unlikely we'll ever actually find ourselves
10653                          * here, as a file small enough to fit inline won't be
10654                          * big enough to store more than the swap header, but in
10655                          * case something changes in the future, let's catch it
10656                          * here rather than later.
10657                          */
10658                         btrfs_warn(fs_info, "swapfile must not be inline");
10659                         ret = -EINVAL;
10660                         goto out;
10661                 }
10662                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10663                         btrfs_warn(fs_info, "swapfile must not be compressed");
10664                         ret = -EINVAL;
10665                         goto out;
10666                 }
10667
10668                 logical_block_start = em->block_start + (start - em->start);
10669                 len = min(len, em->len - (start - em->start));
10670                 free_extent_map(em);
10671                 em = NULL;
10672
10673                 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL);
10674                 if (ret < 0) {
10675                         goto out;
10676                 } else if (ret) {
10677                         ret = 0;
10678                 } else {
10679                         btrfs_warn(fs_info,
10680                                    "swapfile must not be copy-on-write");
10681                         ret = -EINVAL;
10682                         goto out;
10683                 }
10684
10685                 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10686                 if (IS_ERR(em)) {
10687                         ret = PTR_ERR(em);
10688                         goto out;
10689                 }
10690
10691                 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10692                         btrfs_warn(fs_info,
10693                                    "swapfile must have single data profile");
10694                         ret = -EINVAL;
10695                         goto out;
10696                 }
10697
10698                 if (device == NULL) {
10699                         device = em->map_lookup->stripes[0].dev;
10700                         ret = btrfs_add_swapfile_pin(inode, device, false);
10701                         if (ret == 1)
10702                                 ret = 0;
10703                         else if (ret)
10704                                 goto out;
10705                 } else if (device != em->map_lookup->stripes[0].dev) {
10706                         btrfs_warn(fs_info, "swapfile must be on one device");
10707                         ret = -EINVAL;
10708                         goto out;
10709                 }
10710
10711                 physical_block_start = (em->map_lookup->stripes[0].physical +
10712                                         (logical_block_start - em->start));
10713                 len = min(len, em->len - (logical_block_start - em->start));
10714                 free_extent_map(em);
10715                 em = NULL;
10716
10717                 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10718                 if (!bg) {
10719                         btrfs_warn(fs_info,
10720                            "could not find block group containing swapfile");
10721                         ret = -EINVAL;
10722                         goto out;
10723                 }
10724
10725                 ret = btrfs_add_swapfile_pin(inode, bg, true);
10726                 if (ret) {
10727                         btrfs_put_block_group(bg);
10728                         if (ret == 1)
10729                                 ret = 0;
10730                         else
10731                                 goto out;
10732                 }
10733
10734                 if (bsi.block_len &&
10735                     bsi.block_start + bsi.block_len == physical_block_start) {
10736                         bsi.block_len += len;
10737                 } else {
10738                         if (bsi.block_len) {
10739                                 ret = btrfs_add_swap_extent(sis, &bsi);
10740                                 if (ret)
10741                                         goto out;
10742                         }
10743                         bsi.start = start;
10744                         bsi.block_start = physical_block_start;
10745                         bsi.block_len = len;
10746                 }
10747
10748                 start += len;
10749         }
10750
10751         if (bsi.block_len)
10752                 ret = btrfs_add_swap_extent(sis, &bsi);
10753
10754 out:
10755         if (!IS_ERR_OR_NULL(em))
10756                 free_extent_map(em);
10757
10758         unlock_extent_cached(io_tree, 0, isize - 1, &cached_state);
10759
10760         if (ret)
10761                 btrfs_swap_deactivate(file);
10762
10763         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
10764
10765         if (ret)
10766                 return ret;
10767
10768         if (device)
10769                 sis->bdev = device->bdev;
10770         *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10771         sis->max = bsi.nr_pages;
10772         sis->pages = bsi.nr_pages - 1;
10773         sis->highest_bit = bsi.nr_pages - 1;
10774         return bsi.nr_extents;
10775 }
10776 #else
10777 static void btrfs_swap_deactivate(struct file *file)
10778 {
10779 }
10780
10781 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10782                                sector_t *span)
10783 {
10784         return -EOPNOTSUPP;
10785 }
10786 #endif
10787
10788 static const struct inode_operations btrfs_dir_inode_operations = {
10789         .getattr        = btrfs_getattr,
10790         .lookup         = btrfs_lookup,
10791         .create         = btrfs_create,
10792         .unlink         = btrfs_unlink,
10793         .link           = btrfs_link,
10794         .mkdir          = btrfs_mkdir,
10795         .rmdir          = btrfs_rmdir,
10796         .rename         = btrfs_rename2,
10797         .symlink        = btrfs_symlink,
10798         .setattr        = btrfs_setattr,
10799         .mknod          = btrfs_mknod,
10800         .listxattr      = btrfs_listxattr,
10801         .permission     = btrfs_permission,
10802         .get_acl        = btrfs_get_acl,
10803         .set_acl        = btrfs_set_acl,
10804         .update_time    = btrfs_update_time,
10805         .tmpfile        = btrfs_tmpfile,
10806 };
10807 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10808         .lookup         = btrfs_lookup,
10809         .permission     = btrfs_permission,
10810         .update_time    = btrfs_update_time,
10811 };
10812
10813 static const struct file_operations btrfs_dir_file_operations = {
10814         .llseek         = generic_file_llseek,
10815         .read           = generic_read_dir,
10816         .iterate_shared = btrfs_real_readdir,
10817         .open           = btrfs_opendir,
10818         .unlocked_ioctl = btrfs_ioctl,
10819 #ifdef CONFIG_COMPAT
10820         .compat_ioctl   = btrfs_compat_ioctl,
10821 #endif
10822         .release        = btrfs_release_file,
10823         .fsync          = btrfs_sync_file,
10824 };
10825
10826 static const struct extent_io_ops btrfs_extent_io_ops = {
10827         /* mandatory callbacks */
10828         .submit_bio_hook = btrfs_submit_bio_hook,
10829         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10830 };
10831
10832 /*
10833  * btrfs doesn't support the bmap operation because swapfiles
10834  * use bmap to make a mapping of extents in the file.  They assume
10835  * these extents won't change over the life of the file and they
10836  * use the bmap result to do IO directly to the drive.
10837  *
10838  * the btrfs bmap call would return logical addresses that aren't
10839  * suitable for IO and they also will change frequently as COW
10840  * operations happen.  So, swapfile + btrfs == corruption.
10841  *
10842  * For now we're avoiding this by dropping bmap.
10843  */
10844 static const struct address_space_operations btrfs_aops = {
10845         .readpage       = btrfs_readpage,
10846         .writepage      = btrfs_writepage,
10847         .writepages     = btrfs_writepages,
10848         .readpages      = btrfs_readpages,
10849         .direct_IO      = btrfs_direct_IO,
10850         .invalidatepage = btrfs_invalidatepage,
10851         .releasepage    = btrfs_releasepage,
10852         .set_page_dirty = btrfs_set_page_dirty,
10853         .error_remove_page = generic_error_remove_page,
10854         .swap_activate  = btrfs_swap_activate,
10855         .swap_deactivate = btrfs_swap_deactivate,
10856 };
10857
10858 static const struct inode_operations btrfs_file_inode_operations = {
10859         .getattr        = btrfs_getattr,
10860         .setattr        = btrfs_setattr,
10861         .listxattr      = btrfs_listxattr,
10862         .permission     = btrfs_permission,
10863         .fiemap         = btrfs_fiemap,
10864         .get_acl        = btrfs_get_acl,
10865         .set_acl        = btrfs_set_acl,
10866         .update_time    = btrfs_update_time,
10867 };
10868 static const struct inode_operations btrfs_special_inode_operations = {
10869         .getattr        = btrfs_getattr,
10870         .setattr        = btrfs_setattr,
10871         .permission     = btrfs_permission,
10872         .listxattr      = btrfs_listxattr,
10873         .get_acl        = btrfs_get_acl,
10874         .set_acl        = btrfs_set_acl,
10875         .update_time    = btrfs_update_time,
10876 };
10877 static const struct inode_operations btrfs_symlink_inode_operations = {
10878         .get_link       = page_get_link,
10879         .getattr        = btrfs_getattr,
10880         .setattr        = btrfs_setattr,
10881         .permission     = btrfs_permission,
10882         .listxattr      = btrfs_listxattr,
10883         .update_time    = btrfs_update_time,
10884 };
10885
10886 const struct dentry_operations btrfs_dentry_operations = {
10887         .d_delete       = btrfs_dentry_delete,
10888 };