fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include <linux/ratelimit.h>
  41 #include <linux/mount.h>
  42 #include <linux/btrfs.h>
  43 #include <linux/blkdev.h>
  44 #include <linux/posix_acl_xattr.h>
  45 #include <linux/uio.h>
  46 #include "ctree.h"
  47 #include "disk-io.h"
  48 #include "transaction.h"
  49 #include "btrfs_inode.h"
  50 #include "print-tree.h"
  51 #include "ordered-data.h"
  52 #include "xattr.h"
  53 #include "tree-log.h"
  54 #include "volumes.h"
  55 #include "compression.h"
  56 #include "locking.h"
  57 #include "free-space-cache.h"
  58 #include "inode-map.h"
  59 #include "backref.h"
  60 #include "hash.h"
  61 #include "props.h"
  62 #include "qgroup.h"
  63
  64 struct btrfs_iget_args {
  65         struct btrfs_key *location;
  66         struct btrfs_root *root;
  67 };
  68
  69 struct btrfs_dio_data {
  70         u64 outstanding_extents;
  71         u64 reserve;
  72         u64 unsubmitted_oe_range_start;
  73         u64 unsubmitted_oe_range_end;
  74 };
  75
  76 static const struct inode_operations btrfs_dir_inode_operations;
  77 static const struct inode_operations btrfs_symlink_inode_operations;
  78 static const struct inode_operations btrfs_dir_ro_inode_operations;
  79 static const struct inode_operations btrfs_special_inode_operations;
  80 static const struct inode_operations btrfs_file_inode_operations;
  81 static const struct address_space_operations btrfs_aops;
  82 static const struct address_space_operations btrfs_symlink_aops;
  83 static const struct file_operations btrfs_dir_file_operations;
  84 static const struct extent_io_ops btrfs_extent_io_ops;
  85
  86 static struct kmem_cache *btrfs_inode_cachep;
  87 struct kmem_cache *btrfs_trans_handle_cachep;
  88 struct kmem_cache *btrfs_transaction_cachep;
  89 struct kmem_cache *btrfs_path_cachep;
  90 struct kmem_cache *btrfs_free_space_cachep;
  91
  92 #define S_SHIFT 12
  93 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  94         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  95         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  96         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  97         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  98         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  99         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
 100         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
 101 };
 102
 103 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
 104 static int btrfs_truncate(struct inode *inode);
 105 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
 106 static noinline int cow_file_range(struct inode *inode,
 107                                    struct page *locked_page,
 108                                    u64 start, u64 end, int *page_started,
 109                                    unsigned long *nr_written, int unlock);
 110 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
 111                                            u64 len, u64 orig_start,
 112                                            u64 block_start, u64 block_len,
 113                                            u64 orig_block_len, u64 ram_bytes,
 114                                            int type);
 115
 116 static int btrfs_dirty_inode(struct inode *inode);
 117
 118 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 119 void btrfs_test_inode_set_ops(struct inode *inode)
 120 {
 121         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
 122 }
 123 #endif
 124
 125 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 126                                      struct inode *inode,  struct inode *dir,
 127                                      const struct qstr *qstr)
 128 {
 129         int err;
 130
 131         err = btrfs_init_acl(trans, inode, dir);
 132         if (!err)
 133                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 134         return err;
 135 }
 136
 137 /*
 138  * this does all the hard work for inserting an inline extent into
 139  * the btree.  The caller should have done a btrfs_drop_extents so that
 140  * no overlapping inline items exist in the btree
 141  */
 142 static int insert_inline_extent(struct btrfs_trans_handle *trans,
 143                                 struct btrfs_path *path, int extent_inserted,
 144                                 struct btrfs_root *root, struct inode *inode,
 145                                 u64 start, size_t size, size_t compressed_size,
 146                                 int compress_type,
 147                                 struct page **compressed_pages)
 148 {
 149         struct extent_buffer *leaf;
 150         struct page *page = NULL;
 151         char *kaddr;
 152         unsigned long ptr;
 153         struct btrfs_file_extent_item *ei;
 154         int err = 0;
 155         int ret;
 156         size_t cur_size = size;
 157         unsigned long offset;
 158
 159         if (compressed_size && compressed_pages)
 160                 cur_size = compressed_size;
 161
 162         inode_add_bytes(inode, size);
 163
 164         if (!extent_inserted) {
 165                 struct btrfs_key key;
 166                 size_t datasize;
 167
 168                 key.objectid = btrfs_ino(inode);
 169                 key.offset = start;
 170                 key.type = BTRFS_EXTENT_DATA_KEY;
 171
 172                 datasize = btrfs_file_extent_calc_inline_size(cur_size);
 173                 path->leave_spinning = 1;
 174                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 175                                               datasize);
 176                 if (ret) {
 177                         err = ret;
 178                         goto fail;
 179                 }
 180         }
 181         leaf = path->nodes[0];
 182         ei = btrfs_item_ptr(leaf, path->slots[0],
 183                             struct btrfs_file_extent_item);
 184         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 185         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 186         btrfs_set_file_extent_encryption(leaf, ei, 0);
 187         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 188         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 189         ptr = btrfs_file_extent_inline_start(ei);
 190
 191         if (compress_type != BTRFS_COMPRESS_NONE) {
 192                 struct page *cpage;
 193                 int i = 0;
 194                 while (compressed_size > 0) {
 195                         cpage = compressed_pages[i];
 196                         cur_size = min_t(unsigned long, compressed_size,
 197                                        PAGE_SIZE);
 198
 199                         kaddr = kmap_atomic(cpage);
 200                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 201                         kunmap_atomic(kaddr);
 202
 203                         i++;
 204                         ptr += cur_size;
 205                         compressed_size -= cur_size;
 206                 }
 207                 btrfs_set_file_extent_compression(leaf, ei,
 208                                                   compress_type);
 209         } else {
 210                 page = find_get_page(inode->i_mapping,
 211                                      start >> PAGE_SHIFT);
 212                 btrfs_set_file_extent_compression(leaf, ei, 0);
 213                 kaddr = kmap_atomic(page);
 214                 offset = start & (PAGE_SIZE - 1);
 215                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 216                 kunmap_atomic(kaddr);
 217                 put_page(page);
 218         }
 219         btrfs_mark_buffer_dirty(leaf);
 220         btrfs_release_path(path);
 221
 222         /*
 223          * we're an inline extent, so nobody can
 224          * extend the file past i_size without locking
 225          * a page we already have locked.
 226          *
 227          * We must do any isize and inode updates
 228          * before we unlock the pages.  Otherwise we
 229          * could end up racing with unlink.
 230          */
 231         BTRFS_I(inode)->disk_i_size = inode->i_size;
 232         ret = btrfs_update_inode(trans, root, inode);
 233
 234         return ret;
 235 fail:
 236         return err;
 237 }
 238
 239
 240 /*
 241  * conditionally insert an inline extent into the file.  This
 242  * does the checks required to make sure the data is small enough
 243  * to fit as an inline extent.
 244  */
 245 static noinline int cow_file_range_inline(struct btrfs_root *root,
 246                                           struct inode *inode, u64 start,
 247                                           u64 end, size_t compressed_size,
 248                                           int compress_type,
 249                                           struct page **compressed_pages)
 250 {
 251         struct btrfs_trans_handle *trans;
 252         u64 isize = i_size_read(inode);
 253         u64 actual_end = min(end + 1, isize);
 254         u64 inline_len = actual_end - start;
 255         u64 aligned_end = ALIGN(end, root->sectorsize);
 256         u64 data_len = inline_len;
 257         int ret;
 258         struct btrfs_path *path;
 259         int extent_inserted = 0;
 260         u32 extent_item_size;
 261
 262         if (compressed_size)
 263                 data_len = compressed_size;
 264
 265         if (start > 0 ||
 266             actual_end > root->sectorsize ||
 267             data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 268             (!compressed_size &&
 269             (actual_end & (root->sectorsize - 1)) == 0) ||
 270             end + 1 < isize ||
 271             data_len > root->fs_info->max_inline) {
 272                 return 1;
 273         }
 274
 275         path = btrfs_alloc_path();
 276         if (!path)
 277                 return -ENOMEM;
 278
 279         trans = btrfs_join_transaction(root);
 280         if (IS_ERR(trans)) {
 281                 btrfs_free_path(path);
 282                 return PTR_ERR(trans);
 283         }
 284         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 285
 286         if (compressed_size && compressed_pages)
 287                 extent_item_size = btrfs_file_extent_calc_inline_size(
 288                    compressed_size);
 289         else
 290                 extent_item_size = btrfs_file_extent_calc_inline_size(
 291                     inline_len);
 292
 293         ret = __btrfs_drop_extents(trans, root, inode, path,
 294                                    start, aligned_end, NULL,
 295                                    1, 1, extent_item_size, &extent_inserted);
 296         if (ret) {
 297                 btrfs_abort_transaction(trans, root, ret);
 298                 goto out;
 299         }
 300
 301         if (isize > actual_end)
 302                 inline_len = min_t(u64, isize, actual_end);
 303         ret = insert_inline_extent(trans, path, extent_inserted,
 304                                    root, inode, start,
 305                                    inline_len, compressed_size,
 306                                    compress_type, compressed_pages);
 307         if (ret && ret != -ENOSPC) {
 308                 btrfs_abort_transaction(trans, root, ret);
 309                 goto out;
 310         } else if (ret == -ENOSPC) {
 311                 ret = 1;
 312                 goto out;
 313         }
 314
 315         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
 316         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 317         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 318 out:
 319         /*
 320          * Don't forget to free the reserved space, as for inlined extent
 321          * it won't count as data extent, free them directly here.
 322          * And at reserve time, it's always aligned to page size, so
 323          * just free one page here.
 324          */
 325         btrfs_qgroup_free_data(inode, 0, PAGE_SIZE);
 326         btrfs_free_path(path);
 327         btrfs_end_transaction(trans, root);
 328         return ret;
 329 }
 330
 331 struct async_extent {
 332         u64 start;
 333         u64 ram_size;
 334         u64 compressed_size;
 335         struct page **pages;
 336         unsigned long nr_pages;
 337         int compress_type;
 338         struct list_head list;
 339 };
 340
 341 struct async_cow {
 342         struct inode *inode;
 343         struct btrfs_root *root;
 344         struct page *locked_page;
 345         u64 start;
 346         u64 end;
 347         struct list_head extents;
 348         struct btrfs_work work;
 349 };
 350
 351 static noinline int add_async_extent(struct async_cow *cow,
 352                                      u64 start, u64 ram_size,
 353                                      u64 compressed_size,
 354                                      struct page **pages,
 355                                      unsigned long nr_pages,
 356                                      int compress_type)
 357 {
 358         struct async_extent *async_extent;
 359
 360         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 361         BUG_ON(!async_extent); /* -ENOMEM */
 362         async_extent->start = start;
 363         async_extent->ram_size = ram_size;
 364         async_extent->compressed_size = compressed_size;
 365         async_extent->pages = pages;
 366         async_extent->nr_pages = nr_pages;
 367         async_extent->compress_type = compress_type;
 368         list_add_tail(&async_extent->list, &cow->extents);
 369         return 0;
 370 }
 371
 372 static inline int inode_need_compress(struct inode *inode)
 373 {
 374         struct btrfs_root *root = BTRFS_I(inode)->root;
 375
 376         /* force compress */
 377         if (btrfs_test_opt(root, FORCE_COMPRESS))
 378                 return 1;
 379         /* bad compression ratios */
 380         if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 381                 return 0;
 382         if (btrfs_test_opt(root, COMPRESS) ||
 383             BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
 384             BTRFS_I(inode)->force_compress)
 385                 return 1;
 386         return 0;
 387 }
 388
 389 /*
 390  * we create compressed extents in two phases.  The first
 391  * phase compresses a range of pages that have already been
 392  * locked (both pages and state bits are locked).
 393  *
 394  * This is done inside an ordered work queue, and the compression
 395  * is spread across many cpus.  The actual IO submission is step
 396  * two, and the ordered work queue takes care of making sure that
 397  * happens in the same order things were put onto the queue by
 398  * writepages and friends.
 399  *
 400  * If this code finds it can't get good compression, it puts an
 401  * entry onto the work queue to write the uncompressed bytes.  This
 402  * makes sure that both compressed inodes and uncompressed inodes
 403  * are written in the same order that the flusher thread sent them
 404  * down.
 405  */
 406 static noinline void compress_file_range(struct inode *inode,
 407                                         struct page *locked_page,
 408                                         u64 start, u64 end,
 409                                         struct async_cow *async_cow,
 410                                         int *num_added)
 411 {
 412         struct btrfs_root *root = BTRFS_I(inode)->root;
 413         u64 num_bytes;
 414         u64 blocksize = root->sectorsize;
 415         u64 actual_end;
 416         u64 isize = i_size_read(inode);
 417         int ret = 0;
 418         struct page **pages = NULL;
 419         unsigned long nr_pages;
 420         unsigned long nr_pages_ret = 0;
 421         unsigned long total_compressed = 0;
 422         unsigned long total_in = 0;
 423         unsigned long max_compressed = SZ_128K;
 424         unsigned long max_uncompressed = SZ_128K;
 425         int i;
 426         int will_compress;
 427         int compress_type = root->fs_info->compress_type;
 428         int redirty = 0;
 429
 430         /* if this is a small write inside eof, kick off a defrag */
 431         if ((end - start + 1) < SZ_16K &&
 432             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 433                 btrfs_add_inode_defrag(NULL, inode);
 434
 435         actual_end = min_t(u64, isize, end + 1);
 436 again:
 437         will_compress = 0;
 438         nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
 439         nr_pages = min_t(unsigned long, nr_pages, SZ_128K / PAGE_SIZE);
 440
 441         /*
 442          * we don't want to send crud past the end of i_size through
 443          * compression, that's just a waste of CPU time.  So, if the
 444          * end of the file is before the start of our current
 445          * requested range of bytes, we bail out to the uncompressed
 446          * cleanup code that can deal with all of this.
 447          *
 448          * It isn't really the fastest way to fix things, but this is a
 449          * very uncommon corner.
 450          */
 451         if (actual_end <= start)
 452                 goto cleanup_and_bail_uncompressed;
 453
 454         total_compressed = actual_end - start;
 455
 456         /*
 457          * skip compression for a small file range(<=blocksize) that
 458          * isn't an inline extent, since it doesn't save disk space at all.
 459          */
 460         if (total_compressed <= blocksize &&
 461            (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 462                 goto cleanup_and_bail_uncompressed;
 463
 464         /* we want to make sure that amount of ram required to uncompress
 465          * an extent is reasonable, so we limit the total size in ram
 466          * of a compressed extent to 128k.  This is a crucial number
 467          * because it also controls how easily we can spread reads across
 468          * cpus for decompression.
 469          *
 470          * We also want to make sure the amount of IO required to do
 471          * a random read is reasonably small, so we limit the size of
 472          * a compressed extent to 128k.
 473          */
 474         total_compressed = min(total_compressed, max_uncompressed);
 475         num_bytes = ALIGN(end - start + 1, blocksize);
 476         num_bytes = max(blocksize,  num_bytes);
 477         total_in = 0;
 478         ret = 0;
 479
 480         /*
 481          * we do compression for mount -o compress and when the
 482          * inode has not been flagged as nocompress.  This flag can
 483          * change at any time if we discover bad compression ratios.
 484          */
 485         if (inode_need_compress(inode)) {
 486                 WARN_ON(pages);
 487                 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
 488                 if (!pages) {
 489                         /* just bail out to the uncompressed code */
 490                         goto cont;
 491                 }
 492
 493                 if (BTRFS_I(inode)->force_compress)
 494                         compress_type = BTRFS_I(inode)->force_compress;
 495
 496                 /*
 497                  * we need to call clear_page_dirty_for_io on each
 498                  * page in the range.  Otherwise applications with the file
 499                  * mmap'd can wander in and change the page contents while
 500                  * we are compressing them.
 501                  *
 502                  * If the compression fails for any reason, we set the pages
 503                  * dirty again later on.
 504                  */
 505                 extent_range_clear_dirty_for_io(inode, start, end);
 506                 redirty = 1;
 507                 ret = btrfs_compress_pages(compress_type,
 508                                            inode->i_mapping, start,
 509                                            total_compressed, pages,
 510                                            nr_pages, &nr_pages_ret,
 511                                            &total_in,
 512                                            &total_compressed,
 513                                            max_compressed);
 514
 515                 if (!ret) {
 516                         unsigned long offset = total_compressed &
 517                                 (PAGE_SIZE - 1);
 518                         struct page *page = pages[nr_pages_ret - 1];
 519                         char *kaddr;
 520
 521                         /* zero the tail end of the last page, we might be
 522                          * sending it down to disk
 523                          */
 524                         if (offset) {
 525                                 kaddr = kmap_atomic(page);
 526                                 memset(kaddr + offset, 0,
 527                                        PAGE_SIZE - offset);
 528                                 kunmap_atomic(kaddr);
 529                         }
 530                         will_compress = 1;
 531                 }
 532         }
 533 cont:
 534         if (start == 0) {
 535                 /* lets try to make an inline extent */
 536                 if (ret || total_in < (actual_end - start)) {
 537                         /* we didn't compress the entire range, try
 538                          * to make an uncompressed inline extent.
 539                          */
 540                         ret = cow_file_range_inline(root, inode, start, end,
 541                                                     0, 0, NULL);
 542                 } else {
 543                         /* try making a compressed inline extent */
 544                         ret = cow_file_range_inline(root, inode, start, end,
 545                                                     total_compressed,
 546                                                     compress_type, pages);
 547                 }
 548                 if (ret <= 0) {
 549                         unsigned long clear_flags = EXTENT_DELALLOC |
 550                                 EXTENT_DEFRAG;
 551                         unsigned long page_error_op;
 552
 553                         clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
 554                         page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
 555
 556                         /*
 557                          * inline extent creation worked or returned error,
 558                          * we don't need to create any more async work items.
 559                          * Unlock and free up our temp pages.
 560                          */
 561                         extent_clear_unlock_delalloc(inode, start, end, NULL,
 562                                                      clear_flags, PAGE_UNLOCK |
 563                                                      PAGE_CLEAR_DIRTY |
 564                                                      PAGE_SET_WRITEBACK |
 565                                                      page_error_op |
 566                                                      PAGE_END_WRITEBACK);
 567                         goto free_pages_out;
 568                 }
 569         }
 570
 571         if (will_compress) {
 572                 /*
 573                  * we aren't doing an inline extent round the compressed size
 574                  * up to a block size boundary so the allocator does sane
 575                  * things
 576                  */
 577                 total_compressed = ALIGN(total_compressed, blocksize);
 578
 579                 /*
 580                  * one last check to make sure the compression is really a
 581                  * win, compare the page count read with the blocks on disk
 582                  */
 583                 total_in = ALIGN(total_in, PAGE_SIZE);
 584                 if (total_compressed >= total_in) {
 585                         will_compress = 0;
 586                 } else {
 587                         num_bytes = total_in;
 588                 }
 589         }
 590         if (!will_compress && pages) {
 591                 /*
 592                  * the compression code ran but failed to make things smaller,
 593                  * free any pages it allocated and our page pointer array
 594                  */
 595                 for (i = 0; i < nr_pages_ret; i++) {
 596                         WARN_ON(pages[i]->mapping);
 597                         put_page(pages[i]);
 598                 }
 599                 kfree(pages);
 600                 pages = NULL;
 601                 total_compressed = 0;
 602                 nr_pages_ret = 0;
 603
 604                 /* flag the file so we don't compress in the future */
 605                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 606                     !(BTRFS_I(inode)->force_compress)) {
 607                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 608                 }
 609         }
 610         if (will_compress) {
 611                 *num_added += 1;
 612
 613                 /* the async work queues will take care of doing actual
 614                  * allocation on disk for these compressed pages,
 615                  * and will submit them to the elevator.
 616                  */
 617                 add_async_extent(async_cow, start, num_bytes,
 618                                  total_compressed, pages, nr_pages_ret,
 619                                  compress_type);
 620
 621                 if (start + num_bytes < end) {
 622                         start += num_bytes;
 623                         pages = NULL;
 624                         cond_resched();
 625                         goto again;
 626                 }
 627         } else {
 628 cleanup_and_bail_uncompressed:
 629                 /*
 630                  * No compression, but we still need to write the pages in
 631                  * the file we've been given so far.  redirty the locked
 632                  * page if it corresponds to our extent and set things up
 633                  * for the async work queue to run cow_file_range to do
 634                  * the normal delalloc dance
 635                  */
 636                 if (page_offset(locked_page) >= start &&
 637                     page_offset(locked_page) <= end) {
 638                         __set_page_dirty_nobuffers(locked_page);
 639                         /* unlocked later on in the async handlers */
 640                 }
 641                 if (redirty)
 642                         extent_range_redirty_for_io(inode, start, end);
 643                 add_async_extent(async_cow, start, end - start + 1,
 644                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 645                 *num_added += 1;
 646         }
 647
 648         return;
 649
 650 free_pages_out:
 651         for (i = 0; i < nr_pages_ret; i++) {
 652                 WARN_ON(pages[i]->mapping);
 653                 put_page(pages[i]);
 654         }
 655         kfree(pages);
 656 }
 657
 658 static void free_async_extent_pages(struct async_extent *async_extent)
 659 {
 660         int i;
 661
 662         if (!async_extent->pages)
 663                 return;
 664
 665         for (i = 0; i < async_extent->nr_pages; i++) {
 666                 WARN_ON(async_extent->pages[i]->mapping);
 667                 put_page(async_extent->pages[i]);
 668         }
 669         kfree(async_extent->pages);
 670         async_extent->nr_pages = 0;
 671         async_extent->pages = NULL;
 672 }
 673
 674 /*
 675  * phase two of compressed writeback.  This is the ordered portion
 676  * of the code, which only gets called in the order the work was
 677  * queued.  We walk all the async extents created by compress_file_range
 678  * and send them down to the disk.
 679  */
 680 static noinline void submit_compressed_extents(struct inode *inode,
 681                                               struct async_cow *async_cow)
 682 {
 683         struct async_extent *async_extent;
 684         u64 alloc_hint = 0;
 685         struct btrfs_key ins;
 686         struct extent_map *em;
 687         struct btrfs_root *root = BTRFS_I(inode)->root;
 688         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 689         struct extent_io_tree *io_tree;
 690         int ret = 0;
 691
 692 again:
 693         while (!list_empty(&async_cow->extents)) {
 694                 async_extent = list_entry(async_cow->extents.next,
 695                                           struct async_extent, list);
 696                 list_del(&async_extent->list);
 697
 698                 io_tree = &BTRFS_I(inode)->io_tree;
 699
 700 retry:
 701                 /* did the compression code fall back to uncompressed IO? */
 702                 if (!async_extent->pages) {
 703                         int page_started = 0;
 704                         unsigned long nr_written = 0;
 705
 706                         lock_extent(io_tree, async_extent->start,
 707                                          async_extent->start +
 708                                          async_extent->ram_size - 1);
 709
 710                         /* allocate blocks */
 711                         ret = cow_file_range(inode, async_cow->locked_page,
 712                                              async_extent->start,
 713                                              async_extent->start +
 714                                              async_extent->ram_size - 1,
 715                                              &page_started, &nr_written, 0);
 716
 717                         /* JDM XXX */
 718
 719                         /*
 720                          * if page_started, cow_file_range inserted an
 721                          * inline extent and took care of all the unlocking
 722                          * and IO for us.  Otherwise, we need to submit
 723                          * all those pages down to the drive.
 724                          */
 725                         if (!page_started && !ret)
 726                                 extent_write_locked_range(io_tree,
 727                                                   inode, async_extent->start,
 728                                                   async_extent->start +
 729                                                   async_extent->ram_size - 1,
 730                                                   btrfs_get_extent,
 731                                                   WB_SYNC_ALL);
 732                         else if (ret)
 733                                 unlock_page(async_cow->locked_page);
 734                         kfree(async_extent);
 735                         cond_resched();
 736                         continue;
 737                 }
 738
 739                 lock_extent(io_tree, async_extent->start,
 740                             async_extent->start + async_extent->ram_size - 1);
 741
 742                 ret = btrfs_reserve_extent(root,
 743                                            async_extent->compressed_size,
 744                                            async_extent->compressed_size,
 745                                            0, alloc_hint, &ins, 1, 1);
 746                 if (ret) {
 747                         free_async_extent_pages(async_extent);
 748
 749                         if (ret == -ENOSPC) {
 750                                 unlock_extent(io_tree, async_extent->start,
 751                                               async_extent->start +
 752                                               async_extent->ram_size - 1);
 753
 754                                 /*
 755                                  * we need to redirty the pages if we decide to
 756                                  * fallback to uncompressed IO, otherwise we
 757                                  * will not submit these pages down to lower
 758                                  * layers.
 759                                  */
 760                                 extent_range_redirty_for_io(inode,
 761                                                 async_extent->start,
 762                                                 async_extent->start +
 763                                                 async_extent->ram_size - 1);
 764
 765                                 goto retry;
 766                         }
 767                         goto out_free;
 768                 }
 769                 /*
 770                  * here we're doing allocation and writeback of the
 771                  * compressed pages
 772                  */
 773                 btrfs_drop_extent_cache(inode, async_extent->start,
 774                                         async_extent->start +
 775                                         async_extent->ram_size - 1, 0);
 776
 777                 em = alloc_extent_map();
 778                 if (!em) {
 779                         ret = -ENOMEM;
 780                         goto out_free_reserve;
 781                 }
 782                 em->start = async_extent->start;
 783                 em->len = async_extent->ram_size;
 784                 em->orig_start = em->start;
 785                 em->mod_start = em->start;
 786                 em->mod_len = em->len;
 787
 788                 em->block_start = ins.objectid;
 789                 em->block_len = ins.offset;
 790                 em->orig_block_len = ins.offset;
 791                 em->ram_bytes = async_extent->ram_size;
 792                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 793                 em->compress_type = async_extent->compress_type;
 794                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 795                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 796                 em->generation = -1;
 797
 798                 while (1) {
 799                         write_lock(&em_tree->lock);
 800                         ret = add_extent_mapping(em_tree, em, 1);
 801                         write_unlock(&em_tree->lock);
 802                         if (ret != -EEXIST) {
 803                                 free_extent_map(em);
 804                                 break;
 805                         }
 806                         btrfs_drop_extent_cache(inode, async_extent->start,
 807                                                 async_extent->start +
 808                                                 async_extent->ram_size - 1, 0);
 809                 }
 810
 811                 if (ret)
 812                         goto out_free_reserve;
 813
 814                 ret = btrfs_add_ordered_extent_compress(inode,
 815                                                 async_extent->start,
 816                                                 ins.objectid,
 817                                                 async_extent->ram_size,
 818                                                 ins.offset,
 819                                                 BTRFS_ORDERED_COMPRESSED,
 820                                                 async_extent->compress_type);
 821                 if (ret) {
 822                         btrfs_drop_extent_cache(inode, async_extent->start,
 823                                                 async_extent->start +
 824                                                 async_extent->ram_size - 1, 0);
 825                         goto out_free_reserve;
 826                 }
 827
 828                 /*
 829                  * clear dirty, set writeback and unlock the pages.
 830                  */
 831                 extent_clear_unlock_delalloc(inode, async_extent->start,
 832                                 async_extent->start +
 833                                 async_extent->ram_size - 1,
 834                                 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
 835                                 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 836                                 PAGE_SET_WRITEBACK);
 837                 ret = btrfs_submit_compressed_write(inode,
 838                                     async_extent->start,
 839                                     async_extent->ram_size,
 840                                     ins.objectid,
 841                                     ins.offset, async_extent->pages,
 842                                     async_extent->nr_pages);
 843                 if (ret) {
 844                         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
 845                         struct page *p = async_extent->pages[0];
 846                         const u64 start = async_extent->start;
 847                         const u64 end = start + async_extent->ram_size - 1;
 848
 849                         p->mapping = inode->i_mapping;
 850                         tree->ops->writepage_end_io_hook(p, start, end,
 851                                                          NULL, 0);
 852                         p->mapping = NULL;
 853                         extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
 854                                                      PAGE_END_WRITEBACK |
 855                                                      PAGE_SET_ERROR);
 856                         free_async_extent_pages(async_extent);
 857                 }
 858                 alloc_hint = ins.objectid + ins.offset;
 859                 kfree(async_extent);
 860                 cond_resched();
 861         }
 862         return;
 863 out_free_reserve:
 864         btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
 865 out_free:
 866         extent_clear_unlock_delalloc(inode, async_extent->start,
 867                                      async_extent->start +
 868                                      async_extent->ram_size - 1,
 869                                      NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
 870                                      EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
 871                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 872                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
 873                                      PAGE_SET_ERROR);
 874         free_async_extent_pages(async_extent);
 875         kfree(async_extent);
 876         goto again;
 877 }
 878
 879 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 880                                       u64 num_bytes)
 881 {
 882         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 883         struct extent_map *em;
 884         u64 alloc_hint = 0;
 885
 886         read_lock(&em_tree->lock);
 887         em = search_extent_mapping(em_tree, start, num_bytes);
 888         if (em) {
 889                 /*
 890                  * if block start isn't an actual block number then find the
 891                  * first block in this inode and use that as a hint.  If that
 892                  * block is also bogus then just don't worry about it.
 893                  */
 894                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 895                         free_extent_map(em);
 896                         em = search_extent_mapping(em_tree, 0, 0);
 897                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 898                                 alloc_hint = em->block_start;
 899                         if (em)
 900                                 free_extent_map(em);
 901                 } else {
 902                         alloc_hint = em->block_start;
 903                         free_extent_map(em);
 904                 }
 905         }
 906         read_unlock(&em_tree->lock);
 907
 908         return alloc_hint;
 909 }
 910
 911 /*
 912  * when extent_io.c finds a delayed allocation range in the file,
 913  * the call backs end up in this code.  The basic idea is to
 914  * allocate extents on disk for the range, and create ordered data structs
 915  * in ram to track those extents.
 916  *
 917  * locked_page is the page that writepage had locked already.  We use
 918  * it to make sure we don't do extra locks or unlocks.
 919  *
 920  * *page_started is set to one if we unlock locked_page and do everything
 921  * required to start IO on it.  It may be clean and already done with
 922  * IO when we return.
 923  */
 924 static noinline int cow_file_range(struct inode *inode,
 925                                    struct page *locked_page,
 926                                    u64 start, u64 end, int *page_started,
 927                                    unsigned long *nr_written,
 928                                    int unlock)
 929 {
 930         struct btrfs_root *root = BTRFS_I(inode)->root;
 931         u64 alloc_hint = 0;
 932         u64 num_bytes;
 933         unsigned long ram_size;
 934         u64 disk_num_bytes;
 935         u64 cur_alloc_size;
 936         u64 blocksize = root->sectorsize;
 937         struct btrfs_key ins;
 938         struct extent_map *em;
 939         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 940         int ret = 0;
 941
 942         if (btrfs_is_free_space_inode(inode)) {
 943                 WARN_ON_ONCE(1);
 944                 ret = -EINVAL;
 945                 goto out_unlock;
 946         }
 947
 948         num_bytes = ALIGN(end - start + 1, blocksize);
 949         num_bytes = max(blocksize,  num_bytes);
 950         disk_num_bytes = num_bytes;
 951
 952         /* if this is a small write inside eof, kick off defrag */
 953         if (num_bytes < SZ_64K &&
 954             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 955                 btrfs_add_inode_defrag(NULL, inode);
 956
 957         if (start == 0) {
 958                 /* lets try to make an inline extent */
 959                 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
 960                                             NULL);
 961                 if (ret == 0) {
 962                         extent_clear_unlock_delalloc(inode, start, end, NULL,
 963                                      EXTENT_LOCKED | EXTENT_DELALLOC |
 964                                      EXTENT_DEFRAG, PAGE_UNLOCK |
 965                                      PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
 966                                      PAGE_END_WRITEBACK);
 967
 968                         *nr_written = *nr_written +
 969                              (end - start + PAGE_SIZE) / PAGE_SIZE;
 970                         *page_started = 1;
 971                         goto out;
 972                 } else if (ret < 0) {
 973                         goto out_unlock;
 974                 }
 975         }
 976
 977         BUG_ON(disk_num_bytes >
 978                btrfs_super_total_bytes(root->fs_info->super_copy));
 979
 980         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 981         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 982
 983         while (disk_num_bytes > 0) {
 984                 unsigned long op;
 985
 986                 cur_alloc_size = disk_num_bytes;
 987                 ret = btrfs_reserve_extent(root, cur_alloc_size,
 988                                            root->sectorsize, 0, alloc_hint,
 989                                            &ins, 1, 1);
 990                 if (ret < 0)
 991                         goto out_unlock;
 992
 993                 em = alloc_extent_map();
 994                 if (!em) {
 995                         ret = -ENOMEM;
 996                         goto out_reserve;
 997                 }
 998                 em->start = start;
 999                 em->orig_start = em->start;
1000                 ram_size = ins.offset;
1001                 em->len = ins.offset;
1002                 em->mod_start = em->start;
1003                 em->mod_len = em->len;
1004
1005                 em->block_start = ins.objectid;
1006                 em->block_len = ins.offset;
1007                 em->orig_block_len = ins.offset;
1008                 em->ram_bytes = ram_size;
1009                 em->bdev = root->fs_info->fs_devices->latest_bdev;
1010                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1011                 em->generation = -1;
1012
1013                 while (1) {
1014                         write_lock(&em_tree->lock);
1015                         ret = add_extent_mapping(em_tree, em, 1);
1016                         write_unlock(&em_tree->lock);
1017                         if (ret != -EEXIST) {
1018                                 free_extent_map(em);
1019                                 break;
1020                         }
1021                         btrfs_drop_extent_cache(inode, start,
1022                                                 start + ram_size - 1, 0);
1023                 }
1024                 if (ret)
1025                         goto out_reserve;
1026
1027                 cur_alloc_size = ins.offset;
1028                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1029                                                ram_size, cur_alloc_size, 0);
1030                 if (ret)
1031                         goto out_drop_extent_cache;
1032
1033                 if (root->root_key.objectid ==
1034                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1035                         ret = btrfs_reloc_clone_csums(inode, start,
1036                                                       cur_alloc_size);
1037                         if (ret)
1038                                 goto out_drop_extent_cache;
1039                 }
1040
1041                 if (disk_num_bytes < cur_alloc_size)
1042                         break;
1043
1044                 /* we're not doing compressed IO, don't unlock the first
1045                  * page (which the caller expects to stay locked), don't
1046                  * clear any dirty bits and don't set any writeback bits
1047                  *
1048                  * Do set the Private2 bit so we know this page was properly
1049                  * setup for writepage
1050                  */
1051                 op = unlock ? PAGE_UNLOCK : 0;
1052                 op |= PAGE_SET_PRIVATE2;
1053
1054                 extent_clear_unlock_delalloc(inode, start,
1055                                              start + ram_size - 1, locked_page,
1056                                              EXTENT_LOCKED | EXTENT_DELALLOC,
1057                                              op);
1058                 disk_num_bytes -= cur_alloc_size;
1059                 num_bytes -= cur_alloc_size;
1060                 alloc_hint = ins.objectid + ins.offset;
1061                 start += cur_alloc_size;
1062         }
1063 out:
1064         return ret;
1065
1066 out_drop_extent_cache:
1067         btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1068 out_reserve:
1069         btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1070 out_unlock:
1071         extent_clear_unlock_delalloc(inode, start, end, locked_page,
1072                                      EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1073                                      EXTENT_DELALLOC | EXTENT_DEFRAG,
1074                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1075                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1076         goto out;
1077 }
1078
1079 /*
1080  * work queue call back to started compression on a file and pages
1081  */
1082 static noinline void async_cow_start(struct btrfs_work *work)
1083 {
1084         struct async_cow *async_cow;
1085         int num_added = 0;
1086         async_cow = container_of(work, struct async_cow, work);
1087
1088         compress_file_range(async_cow->inode, async_cow->locked_page,
1089                             async_cow->start, async_cow->end, async_cow,
1090                             &num_added);
1091         if (num_added == 0) {
1092                 btrfs_add_delayed_iput(async_cow->inode);
1093                 async_cow->inode = NULL;
1094         }
1095 }
1096
1097 /*
1098  * work queue call back to submit previously compressed pages
1099  */
1100 static noinline void async_cow_submit(struct btrfs_work *work)
1101 {
1102         struct async_cow *async_cow;
1103         struct btrfs_root *root;
1104         unsigned long nr_pages;
1105
1106         async_cow = container_of(work, struct async_cow, work);
1107
1108         root = async_cow->root;
1109         nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1110                 PAGE_SHIFT;
1111
1112         /*
1113          * atomic_sub_return implies a barrier for waitqueue_active
1114          */
1115         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1116             5 * SZ_1M &&
1117             waitqueue_active(&root->fs_info->async_submit_wait))
1118                 wake_up(&root->fs_info->async_submit_wait);
1119
1120         if (async_cow->inode)
1121                 submit_compressed_extents(async_cow->inode, async_cow);
1122 }
1123
1124 static noinline void async_cow_free(struct btrfs_work *work)
1125 {
1126         struct async_cow *async_cow;
1127         async_cow = container_of(work, struct async_cow, work);
1128         if (async_cow->inode)
1129                 btrfs_add_delayed_iput(async_cow->inode);
1130         kfree(async_cow);
1131 }
1132
1133 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1134                                 u64 start, u64 end, int *page_started,
1135                                 unsigned long *nr_written)
1136 {
1137         struct async_cow *async_cow;
1138         struct btrfs_root *root = BTRFS_I(inode)->root;
1139         unsigned long nr_pages;
1140         u64 cur_end;
1141         int limit = 10 * SZ_1M;
1142
1143         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1144                          1, 0, NULL, GFP_NOFS);
1145         while (start < end) {
1146                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1147                 BUG_ON(!async_cow); /* -ENOMEM */
1148                 async_cow->inode = igrab(inode);
1149                 async_cow->root = root;
1150                 async_cow->locked_page = locked_page;
1151                 async_cow->start = start;
1152
1153                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1154                     !btrfs_test_opt(root, FORCE_COMPRESS))
1155                         cur_end = end;
1156                 else
1157                         cur_end = min(end, start + SZ_512K - 1);
1158
1159                 async_cow->end = cur_end;
1160                 INIT_LIST_HEAD(&async_cow->extents);
1161
1162                 btrfs_init_work(&async_cow->work,
1163                                 btrfs_delalloc_helper,
1164                                 async_cow_start, async_cow_submit,
1165                                 async_cow_free);
1166
1167                 nr_pages = (cur_end - start + PAGE_SIZE) >>
1168                         PAGE_SHIFT;
1169                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1170
1171                 btrfs_queue_work(root->fs_info->delalloc_workers,
1172                                  &async_cow->work);
1173
1174                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1175                         wait_event(root->fs_info->async_submit_wait,
1176                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1177                             limit));
1178                 }
1179
1180                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1181                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1182                         wait_event(root->fs_info->async_submit_wait,
1183                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1184                            0));
1185                 }
1186
1187                 *nr_written += nr_pages;
1188                 start = cur_end + 1;
1189         }
1190         *page_started = 1;
1191         return 0;
1192 }
1193
1194 static noinline int csum_exist_in_range(struct btrfs_root *root,
1195                                         u64 bytenr, u64 num_bytes)
1196 {
1197         int ret;
1198         struct btrfs_ordered_sum *sums;
1199         LIST_HEAD(list);
1200
1201         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1202                                        bytenr + num_bytes - 1, &list, 0);
1203         if (ret == 0 && list_empty(&list))
1204                 return 0;
1205
1206         while (!list_empty(&list)) {
1207                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1208                 list_del(&sums->list);
1209                 kfree(sums);
1210         }
1211         return 1;
1212 }
1213
1214 /*
1215  * when nowcow writeback call back.  This checks for snapshots or COW copies
1216  * of the extents that exist in the file, and COWs the file as required.
1217  *
1218  * If no cow copies or snapshots exist, we write directly to the existing
1219  * blocks on disk
1220  */
1221 static noinline int run_delalloc_nocow(struct inode *inode,
1222                                        struct page *locked_page,
1223                               u64 start, u64 end, int *page_started, int force,
1224                               unsigned long *nr_written)
1225 {
1226         struct btrfs_root *root = BTRFS_I(inode)->root;
1227         struct btrfs_trans_handle *trans;
1228         struct extent_buffer *leaf;
1229         struct btrfs_path *path;
1230         struct btrfs_file_extent_item *fi;
1231         struct btrfs_key found_key;
1232         u64 cow_start;
1233         u64 cur_offset;
1234         u64 extent_end;
1235         u64 extent_offset;
1236         u64 disk_bytenr;
1237         u64 num_bytes;
1238         u64 disk_num_bytes;
1239         u64 ram_bytes;
1240         int extent_type;
1241         int ret, err;
1242         int type;
1243         int nocow;
1244         int check_prev = 1;
1245         bool nolock;
1246         u64 ino = btrfs_ino(inode);
1247
1248         path = btrfs_alloc_path();
1249         if (!path) {
1250                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1251                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1252                                              EXTENT_DO_ACCOUNTING |
1253                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1254                                              PAGE_CLEAR_DIRTY |
1255                                              PAGE_SET_WRITEBACK |
1256                                              PAGE_END_WRITEBACK);
1257                 return -ENOMEM;
1258         }
1259
1260         nolock = btrfs_is_free_space_inode(inode);
1261
1262         if (nolock)
1263                 trans = btrfs_join_transaction_nolock(root);
1264         else
1265                 trans = btrfs_join_transaction(root);
1266
1267         if (IS_ERR(trans)) {
1268                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1269                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1270                                              EXTENT_DO_ACCOUNTING |
1271                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1272                                              PAGE_CLEAR_DIRTY |
1273                                              PAGE_SET_WRITEBACK |
1274                                              PAGE_END_WRITEBACK);
1275                 btrfs_free_path(path);
1276                 return PTR_ERR(trans);
1277         }
1278
1279         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1280
1281         cow_start = (u64)-1;
1282         cur_offset = start;
1283         while (1) {
1284                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1285                                                cur_offset, 0);
1286                 if (ret < 0)
1287                         goto error;
1288                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1289                         leaf = path->nodes[0];
1290                         btrfs_item_key_to_cpu(leaf, &found_key,
1291                                               path->slots[0] - 1);
1292                         if (found_key.objectid == ino &&
1293                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1294                                 path->slots[0]--;
1295                 }
1296                 check_prev = 0;
1297 next_slot:
1298                 leaf = path->nodes[0];
1299                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1300                         ret = btrfs_next_leaf(root, path);
1301                         if (ret < 0)
1302                                 goto error;
1303                         if (ret > 0)
1304                                 break;
1305                         leaf = path->nodes[0];
1306                 }
1307
1308                 nocow = 0;
1309                 disk_bytenr = 0;
1310                 num_bytes = 0;
1311                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1312
1313                 if (found_key.objectid > ino)
1314                         break;
1315                 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1316                     found_key.type < BTRFS_EXTENT_DATA_KEY) {
1317                         path->slots[0]++;
1318                         goto next_slot;
1319                 }
1320                 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1321                     found_key.offset > end)
1322                         break;
1323
1324                 if (found_key.offset > cur_offset) {
1325                         extent_end = found_key.offset;
1326                         extent_type = 0;
1327                         goto out_check;
1328                 }
1329
1330                 fi = btrfs_item_ptr(leaf, path->slots[0],
1331                                     struct btrfs_file_extent_item);
1332                 extent_type = btrfs_file_extent_type(leaf, fi);
1333
1334                 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1335                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1336                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1337                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1338                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1339                         extent_end = found_key.offset +
1340                                 btrfs_file_extent_num_bytes(leaf, fi);
1341                         disk_num_bytes =
1342                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1343                         if (extent_end <= start) {
1344                                 path->slots[0]++;
1345                                 goto next_slot;
1346                         }
1347                         if (disk_bytenr == 0)
1348                                 goto out_check;
1349                         if (btrfs_file_extent_compression(leaf, fi) ||
1350                             btrfs_file_extent_encryption(leaf, fi) ||
1351                             btrfs_file_extent_other_encoding(leaf, fi))
1352                                 goto out_check;
1353                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1354                                 goto out_check;
1355                         if (btrfs_extent_readonly(root, disk_bytenr))
1356                                 goto out_check;
1357                         if (btrfs_cross_ref_exist(trans, root, ino,
1358                                                   found_key.offset -
1359                                                   extent_offset, disk_bytenr))
1360                                 goto out_check;
1361                         disk_bytenr += extent_offset;
1362                         disk_bytenr += cur_offset - found_key.offset;
1363                         num_bytes = min(end + 1, extent_end) - cur_offset;
1364                         /*
1365                          * if there are pending snapshots for this root,
1366                          * we fall into common COW way.
1367                          */
1368                         if (!nolock) {
1369                                 err = btrfs_start_write_no_snapshoting(root);
1370                                 if (!err)
1371                                         goto out_check;
1372                         }
1373                         /*
1374                          * force cow if csum exists in the range.
1375                          * this ensure that csum for a given extent are
1376                          * either valid or do not exist.
1377                          */
1378                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1379                                 goto out_check;
1380                         nocow = 1;
1381                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1382                         extent_end = found_key.offset +
1383                                 btrfs_file_extent_inline_len(leaf,
1384                                                      path->slots[0], fi);
1385                         extent_end = ALIGN(extent_end, root->sectorsize);
1386                 } else {
1387                         BUG_ON(1);
1388                 }
1389 out_check:
1390                 if (extent_end <= start) {
1391                         path->slots[0]++;
1392                         if (!nolock && nocow)
1393                                 btrfs_end_write_no_snapshoting(root);
1394                         goto next_slot;
1395                 }
1396                 if (!nocow) {
1397                         if (cow_start == (u64)-1)
1398                                 cow_start = cur_offset;
1399                         cur_offset = extent_end;
1400                         if (cur_offset > end)
1401                                 break;
1402                         path->slots[0]++;
1403                         goto next_slot;
1404                 }
1405
1406                 btrfs_release_path(path);
1407                 if (cow_start != (u64)-1) {
1408                         ret = cow_file_range(inode, locked_page,
1409                                              cow_start, found_key.offset - 1,
1410                                              page_started, nr_written, 1);
1411                         if (ret) {
1412                                 if (!nolock && nocow)
1413                                         btrfs_end_write_no_snapshoting(root);
1414                                 goto error;
1415                         }
1416                         cow_start = (u64)-1;
1417                 }
1418
1419                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1420                         struct extent_map *em;
1421                         struct extent_map_tree *em_tree;
1422                         em_tree = &BTRFS_I(inode)->extent_tree;
1423                         em = alloc_extent_map();
1424                         BUG_ON(!em); /* -ENOMEM */
1425                         em->start = cur_offset;
1426                         em->orig_start = found_key.offset - extent_offset;
1427                         em->len = num_bytes;
1428                         em->block_len = num_bytes;
1429                         em->block_start = disk_bytenr;
1430                         em->orig_block_len = disk_num_bytes;
1431                         em->ram_bytes = ram_bytes;
1432                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1433                         em->mod_start = em->start;
1434                         em->mod_len = em->len;
1435                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1436                         set_bit(EXTENT_FLAG_FILLING, &em->flags);
1437                         em->generation = -1;
1438                         while (1) {
1439                                 write_lock(&em_tree->lock);
1440                                 ret = add_extent_mapping(em_tree, em, 1);
1441                                 write_unlock(&em_tree->lock);
1442                                 if (ret != -EEXIST) {
1443                                         free_extent_map(em);
1444                                         break;
1445                                 }
1446                                 btrfs_drop_extent_cache(inode, em->start,
1447                                                 em->start + em->len - 1, 0);
1448                         }
1449                         type = BTRFS_ORDERED_PREALLOC;
1450                 } else {
1451                         type = BTRFS_ORDERED_NOCOW;
1452                 }
1453
1454                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1455                                                num_bytes, num_bytes, type);
1456                 BUG_ON(ret); /* -ENOMEM */
1457
1458                 if (root->root_key.objectid ==
1459                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1460                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1461                                                       num_bytes);
1462                         if (ret) {
1463                                 if (!nolock && nocow)
1464                                         btrfs_end_write_no_snapshoting(root);
1465                                 goto error;
1466                         }
1467                 }
1468
1469                 extent_clear_unlock_delalloc(inode, cur_offset,
1470                                              cur_offset + num_bytes - 1,
1471                                              locked_page, EXTENT_LOCKED |
1472                                              EXTENT_DELALLOC, PAGE_UNLOCK |
1473                                              PAGE_SET_PRIVATE2);
1474                 if (!nolock && nocow)
1475                         btrfs_end_write_no_snapshoting(root);
1476                 cur_offset = extent_end;
1477                 if (cur_offset > end)
1478                         break;
1479         }
1480         btrfs_release_path(path);
1481
1482         if (cur_offset <= end && cow_start == (u64)-1) {
1483                 cow_start = cur_offset;
1484                 cur_offset = end;
1485         }
1486
1487         if (cow_start != (u64)-1) {
1488                 ret = cow_file_range(inode, locked_page, cow_start, end,
1489                                      page_started, nr_written, 1);
1490                 if (ret)
1491                         goto error;
1492         }
1493
1494 error:
1495         err = btrfs_end_transaction(trans, root);
1496         if (!ret)
1497                 ret = err;
1498
1499         if (ret && cur_offset < end)
1500                 extent_clear_unlock_delalloc(inode, cur_offset, end,
1501                                              locked_page, EXTENT_LOCKED |
1502                                              EXTENT_DELALLOC | EXTENT_DEFRAG |
1503                                              EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1504                                              PAGE_CLEAR_DIRTY |
1505                                              PAGE_SET_WRITEBACK |
1506                                              PAGE_END_WRITEBACK);
1507         btrfs_free_path(path);
1508         return ret;
1509 }
1510
1511 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1512 {
1513
1514         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1515             !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1516                 return 0;
1517
1518         /*
1519          * @defrag_bytes is a hint value, no spinlock held here,
1520          * if is not zero, it means the file is defragging.
1521          * Force cow if given extent needs to be defragged.
1522          */
1523         if (BTRFS_I(inode)->defrag_bytes &&
1524             test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1525                            EXTENT_DEFRAG, 0, NULL))
1526                 return 1;
1527
1528         return 0;
1529 }
1530
1531 /*
1532  * extent_io.c call back to do delayed allocation processing
1533  */
1534 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1535                               u64 start, u64 end, int *page_started,
1536                               unsigned long *nr_written)
1537 {
1538         int ret;
1539         int force_cow = need_force_cow(inode, start, end);
1540
1541         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1542                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1543                                          page_started, 1, nr_written);
1544         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1545                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1546                                          page_started, 0, nr_written);
1547         } else if (!inode_need_compress(inode)) {
1548                 ret = cow_file_range(inode, locked_page, start, end,
1549                                       page_started, nr_written, 1);
1550         } else {
1551                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1552                         &BTRFS_I(inode)->runtime_flags);
1553                 ret = cow_file_range_async(inode, locked_page, start, end,
1554                                            page_started, nr_written);
1555         }
1556         return ret;
1557 }
1558
1559 static void btrfs_split_extent_hook(struct inode *inode,
1560                                     struct extent_state *orig, u64 split)
1561 {
1562         u64 size;
1563
1564         /* not delalloc, ignore it */
1565         if (!(orig->state & EXTENT_DELALLOC))
1566                 return;
1567
1568         size = orig->end - orig->start + 1;
1569         if (size > BTRFS_MAX_EXTENT_SIZE) {
1570                 u64 num_extents;
1571                 u64 new_size;
1572
1573                 /*
1574                  * See the explanation in btrfs_merge_extent_hook, the same
1575                  * applies here, just in reverse.
1576                  */
1577                 new_size = orig->end - split + 1;
1578                 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1579                                         BTRFS_MAX_EXTENT_SIZE);
1580                 new_size = split - orig->start;
1581                 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1582                                         BTRFS_MAX_EXTENT_SIZE);
1583                 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1584                               BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1585                         return;
1586         }
1587
1588         spin_lock(&BTRFS_I(inode)->lock);
1589         BTRFS_I(inode)->outstanding_extents++;
1590         spin_unlock(&BTRFS_I(inode)->lock);
1591 }
1592
1593 /*
1594  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1595  * extents so we can keep track of new extents that are just merged onto old
1596  * extents, such as when we are doing sequential writes, so we can properly
1597  * account for the metadata space we'll need.
1598  */
1599 static void btrfs_merge_extent_hook(struct inode *inode,
1600                                     struct extent_state *new,
1601                                     struct extent_state *other)
1602 {
1603         u64 new_size, old_size;
1604         u64 num_extents;
1605
1606         /* not delalloc, ignore it */
1607         if (!(other->state & EXTENT_DELALLOC))
1608                 return;
1609
1610         if (new->start > other->start)
1611                 new_size = new->end - other->start + 1;
1612         else
1613                 new_size = other->end - new->start + 1;
1614
1615         /* we're not bigger than the max, unreserve the space and go */
1616         if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1617                 spin_lock(&BTRFS_I(inode)->lock);
1618                 BTRFS_I(inode)->outstanding_extents--;
1619                 spin_unlock(&BTRFS_I(inode)->lock);
1620                 return;
1621         }
1622
1623         /*
1624          * We have to add up either side to figure out how many extents were
1625          * accounted for before we merged into one big extent.  If the number of
1626          * extents we accounted for is <= the amount we need for the new range
1627          * then we can return, otherwise drop.  Think of it like this
1628          *
1629          * [ 4k][MAX_SIZE]
1630          *
1631          * So we've grown the extent by a MAX_SIZE extent, this would mean we
1632          * need 2 outstanding extents, on one side we have 1 and the other side
1633          * we have 1 so they are == and we can return.  But in this case
1634          *
1635          * [MAX_SIZE+4k][MAX_SIZE+4k]
1636          *
1637          * Each range on their own accounts for 2 extents, but merged together
1638          * they are only 3 extents worth of accounting, so we need to drop in
1639          * this case.
1640          */
1641         old_size = other->end - other->start + 1;
1642         num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1643                                 BTRFS_MAX_EXTENT_SIZE);
1644         old_size = new->end - new->start + 1;
1645         num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1646                                  BTRFS_MAX_EXTENT_SIZE);
1647
1648         if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1649                       BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1650                 return;
1651
1652         spin_lock(&BTRFS_I(inode)->lock);
1653         BTRFS_I(inode)->outstanding_extents--;
1654         spin_unlock(&BTRFS_I(inode)->lock);
1655 }
1656
1657 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1658                                       struct inode *inode)
1659 {
1660         spin_lock(&root->delalloc_lock);
1661         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1662                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1663                               &root->delalloc_inodes);
1664                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1665                         &BTRFS_I(inode)->runtime_flags);
1666                 root->nr_delalloc_inodes++;
1667                 if (root->nr_delalloc_inodes == 1) {
1668                         spin_lock(&root->fs_info->delalloc_root_lock);
1669                         BUG_ON(!list_empty(&root->delalloc_root));
1670                         list_add_tail(&root->delalloc_root,
1671                                       &root->fs_info->delalloc_roots);
1672                         spin_unlock(&root->fs_info->delalloc_root_lock);
1673                 }
1674         }
1675         spin_unlock(&root->delalloc_lock);
1676 }
1677
1678 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1679                                      struct inode *inode)
1680 {
1681         spin_lock(&root->delalloc_lock);
1682         if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1683                 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1684                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1685                           &BTRFS_I(inode)->runtime_flags);
1686                 root->nr_delalloc_inodes--;
1687                 if (!root->nr_delalloc_inodes) {
1688                         spin_lock(&root->fs_info->delalloc_root_lock);
1689                         BUG_ON(list_empty(&root->delalloc_root));
1690                         list_del_init(&root->delalloc_root);
1691                         spin_unlock(&root->fs_info->delalloc_root_lock);
1692                 }
1693         }
1694         spin_unlock(&root->delalloc_lock);
1695 }
1696
1697 /*
1698  * extent_io.c set_bit_hook, used to track delayed allocation
1699  * bytes in this file, and to maintain the list of inodes that
1700  * have pending delalloc work to be done.
1701  */
1702 static void btrfs_set_bit_hook(struct inode *inode,
1703                                struct extent_state *state, unsigned *bits)
1704 {
1705
1706         if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1707                 WARN_ON(1);
1708         /*
1709          * set_bit and clear bit hooks normally require _irqsave/restore
1710          * but in this case, we are only testing for the DELALLOC
1711          * bit, which is only set or cleared with irqs on
1712          */
1713         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1714                 struct btrfs_root *root = BTRFS_I(inode)->root;
1715                 u64 len = state->end + 1 - state->start;
1716                 bool do_list = !btrfs_is_free_space_inode(inode);
1717
1718                 if (*bits & EXTENT_FIRST_DELALLOC) {
1719                         *bits &= ~EXTENT_FIRST_DELALLOC;
1720                 } else {
1721                         spin_lock(&BTRFS_I(inode)->lock);
1722                         BTRFS_I(inode)->outstanding_extents++;
1723                         spin_unlock(&BTRFS_I(inode)->lock);
1724                 }
1725
1726                 /* For sanity tests */
1727                 if (btrfs_test_is_dummy_root(root))
1728                         return;
1729
1730                 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1731                                      root->fs_info->delalloc_batch);
1732                 spin_lock(&BTRFS_I(inode)->lock);
1733                 BTRFS_I(inode)->delalloc_bytes += len;
1734                 if (*bits & EXTENT_DEFRAG)
1735                         BTRFS_I(inode)->defrag_bytes += len;
1736                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1737                                          &BTRFS_I(inode)->runtime_flags))
1738                         btrfs_add_delalloc_inodes(root, inode);
1739                 spin_unlock(&BTRFS_I(inode)->lock);
1740         }
1741 }
1742
1743 /*
1744  * extent_io.c clear_bit_hook, see set_bit_hook for why
1745  */
1746 static void btrfs_clear_bit_hook(struct inode *inode,
1747                                  struct extent_state *state,
1748                                  unsigned *bits)
1749 {
1750         u64 len = state->end + 1 - state->start;
1751         u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1752                                     BTRFS_MAX_EXTENT_SIZE);
1753
1754         spin_lock(&BTRFS_I(inode)->lock);
1755         if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1756                 BTRFS_I(inode)->defrag_bytes -= len;
1757         spin_unlock(&BTRFS_I(inode)->lock);
1758
1759         /*
1760          * set_bit and clear bit hooks normally require _irqsave/restore
1761          * but in this case, we are only testing for the DELALLOC
1762          * bit, which is only set or cleared with irqs on
1763          */
1764         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1765                 struct btrfs_root *root = BTRFS_I(inode)->root;
1766                 bool do_list = !btrfs_is_free_space_inode(inode);
1767
1768                 if (*bits & EXTENT_FIRST_DELALLOC) {
1769                         *bits &= ~EXTENT_FIRST_DELALLOC;
1770                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1771                         spin_lock(&BTRFS_I(inode)->lock);
1772                         BTRFS_I(inode)->outstanding_extents -= num_extents;
1773                         spin_unlock(&BTRFS_I(inode)->lock);
1774                 }
1775
1776                 /*
1777                  * We don't reserve metadata space for space cache inodes so we
1778                  * don't need to call dellalloc_release_metadata if there is an
1779                  * error.
1780                  */
1781                 if (*bits & EXTENT_DO_ACCOUNTING &&
1782                     root != root->fs_info->tree_root)
1783                         btrfs_delalloc_release_metadata(inode, len);
1784
1785                 /* For sanity tests. */
1786                 if (btrfs_test_is_dummy_root(root))
1787                         return;
1788
1789                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1790                     && do_list && !(state->state & EXTENT_NORESERVE))
1791                         btrfs_free_reserved_data_space_noquota(inode,
1792                                         state->start, len);
1793
1794                 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1795                                      root->fs_info->delalloc_batch);
1796                 spin_lock(&BTRFS_I(inode)->lock);
1797                 BTRFS_I(inode)->delalloc_bytes -= len;
1798                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1799                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1800                              &BTRFS_I(inode)->runtime_flags))
1801                         btrfs_del_delalloc_inode(root, inode);
1802                 spin_unlock(&BTRFS_I(inode)->lock);
1803         }
1804 }
1805
1806 /*
1807  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1808  * we don't create bios that span stripes or chunks
1809  */
1810 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1811                          size_t size, struct bio *bio,
1812                          unsigned long bio_flags)
1813 {
1814         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1815         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1816         u64 length = 0;
1817         u64 map_length;
1818         int ret;
1819
1820         if (bio_flags & EXTENT_BIO_COMPRESSED)
1821                 return 0;
1822
1823         length = bio->bi_iter.bi_size;
1824         map_length = length;
1825         ret = btrfs_map_block(root->fs_info, rw, logical,
1826                               &map_length, NULL, 0);
1827         /* Will always return 0 with map_multi == NULL */
1828         BUG_ON(ret < 0);
1829         if (map_length < length + size)
1830                 return 1;
1831         return 0;
1832 }
1833
1834 /*
1835  * in order to insert checksums into the metadata in large chunks,
1836  * we wait until bio submission time.   All the pages in the bio are
1837  * checksummed and sums are attached onto the ordered extent record.
1838  *
1839  * At IO completion time the cums attached on the ordered extent record
1840  * are inserted into the btree
1841  */
1842 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1843                                     struct bio *bio, int mirror_num,
1844                                     unsigned long bio_flags,
1845                                     u64 bio_offset)
1846 {
1847         struct btrfs_root *root = BTRFS_I(inode)->root;
1848         int ret = 0;
1849
1850         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1851         BUG_ON(ret); /* -ENOMEM */
1852         return 0;
1853 }
1854
1855 /*
1856  * in order to insert checksums into the metadata in large chunks,
1857  * we wait until bio submission time.   All the pages in the bio are
1858  * checksummed and sums are attached onto the ordered extent record.
1859  *
1860  * At IO completion time the cums attached on the ordered extent record
1861  * are inserted into the btree
1862  */
1863 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1864                           int mirror_num, unsigned long bio_flags,
1865                           u64 bio_offset)
1866 {
1867         struct btrfs_root *root = BTRFS_I(inode)->root;
1868         int ret;
1869
1870         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1871         if (ret) {
1872                 bio->bi_error = ret;
1873                 bio_endio(bio);
1874         }
1875         return ret;
1876 }
1877
1878 /*
1879  * extent_io.c submission hook. This does the right thing for csum calculation
1880  * on write, or reading the csums from the tree before a read
1881  */
1882 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1883                           int mirror_num, unsigned long bio_flags,
1884                           u64 bio_offset)
1885 {
1886         struct btrfs_root *root = BTRFS_I(inode)->root;
1887         enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1888         int ret = 0;
1889         int skip_sum;
1890         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1891
1892         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1893
1894         if (btrfs_is_free_space_inode(inode))
1895                 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1896
1897         if (!(rw & REQ_WRITE)) {
1898                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1899                 if (ret)
1900                         goto out;
1901
1902                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1903                         ret = btrfs_submit_compressed_read(inode, bio,
1904                                                            mirror_num,
1905                                                            bio_flags);
1906                         goto out;
1907                 } else if (!skip_sum) {
1908                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1909                         if (ret)
1910                                 goto out;
1911                 }
1912                 goto mapit;
1913         } else if (async && !skip_sum) {
1914                 /* csum items have already been cloned */
1915                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1916                         goto mapit;
1917                 /* we're doing a write, do the async checksumming */
1918                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1919                                    inode, rw, bio, mirror_num,
1920                                    bio_flags, bio_offset,
1921                                    __btrfs_submit_bio_start,
1922                                    __btrfs_submit_bio_done);
1923                 goto out;
1924         } else if (!skip_sum) {
1925                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1926                 if (ret)
1927                         goto out;
1928         }
1929
1930 mapit:
1931         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1932
1933 out:
1934         if (ret < 0) {
1935                 bio->bi_error = ret;
1936                 bio_endio(bio);
1937         }
1938         return ret;
1939 }
1940
1941 /*
1942  * given a list of ordered sums record them in the inode.  This happens
1943  * at IO completion time based on sums calculated at bio submission time.
1944  */
1945 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1946                              struct inode *inode, u64 file_offset,
1947                              struct list_head *list)
1948 {
1949         struct btrfs_ordered_sum *sum;
1950
1951         list_for_each_entry(sum, list, list) {
1952                 trans->adding_csums = 1;
1953                 btrfs_csum_file_blocks(trans,
1954                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1955                 trans->adding_csums = 0;
1956         }
1957         return 0;
1958 }
1959
1960 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1961                               struct extent_state **cached_state)
1962 {
1963         WARN_ON((end & (PAGE_SIZE - 1)) == 0);
1964         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1965                                    cached_state);
1966 }
1967
1968 /* see btrfs_writepage_start_hook for details on why this is required */
1969 struct btrfs_writepage_fixup {
1970         struct page *page;
1971         struct btrfs_work work;
1972 };
1973
1974 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1975 {
1976         struct btrfs_writepage_fixup *fixup;
1977         struct btrfs_ordered_extent *ordered;
1978         struct extent_state *cached_state = NULL;
1979         struct page *page;
1980         struct inode *inode;
1981         u64 page_start;
1982         u64 page_end;
1983         int ret;
1984
1985         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1986         page = fixup->page;
1987 again:
1988         lock_page(page);
1989         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1990                 ClearPageChecked(page);
1991                 goto out_page;
1992         }
1993
1994         inode = page->mapping->host;
1995         page_start = page_offset(page);
1996         page_end = page_offset(page) + PAGE_SIZE - 1;
1997
1998         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
1999                          &cached_state);
2000
2001         /* already ordered? We're done */
2002         if (PagePrivate2(page))
2003                 goto out;
2004
2005         ordered = btrfs_lookup_ordered_range(inode, page_start,
2006                                         PAGE_SIZE);
2007         if (ordered) {
2008                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2009                                      page_end, &cached_state, GFP_NOFS);
2010                 unlock_page(page);
2011                 btrfs_start_ordered_extent(inode, ordered, 1);
2012                 btrfs_put_ordered_extent(ordered);
2013                 goto again;
2014         }
2015
2016         ret = btrfs_delalloc_reserve_space(inode, page_start,
2017                                            PAGE_SIZE);
2018         if (ret) {
2019                 mapping_set_error(page->mapping, ret);
2020                 end_extent_writepage(page, ret, page_start, page_end);
2021                 ClearPageChecked(page);
2022                 goto out;
2023          }
2024
2025         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
2026         ClearPageChecked(page);
2027         set_page_dirty(page);
2028 out:
2029         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2030                              &cached_state, GFP_NOFS);
2031 out_page:
2032         unlock_page(page);
2033         put_page(page);
2034         kfree(fixup);
2035 }
2036
2037 /*
2038  * There are a few paths in the higher layers of the kernel that directly
2039  * set the page dirty bit without asking the filesystem if it is a
2040  * good idea.  This causes problems because we want to make sure COW
2041  * properly happens and the data=ordered rules are followed.
2042  *
2043  * In our case any range that doesn't have the ORDERED bit set
2044  * hasn't been properly setup for IO.  We kick off an async process
2045  * to fix it up.  The async helper will wait for ordered extents, set
2046  * the delalloc bit and make it safe to write the page.
2047  */
2048 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2049 {
2050         struct inode *inode = page->mapping->host;
2051         struct btrfs_writepage_fixup *fixup;
2052         struct btrfs_root *root = BTRFS_I(inode)->root;
2053
2054         /* this page is properly in the ordered list */
2055         if (TestClearPagePrivate2(page))
2056                 return 0;
2057
2058         if (PageChecked(page))
2059                 return -EAGAIN;
2060
2061         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2062         if (!fixup)
2063                 return -EAGAIN;
2064
2065         SetPageChecked(page);
2066         get_page(page);
2067         btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2068                         btrfs_writepage_fixup_worker, NULL, NULL);
2069         fixup->page = page;
2070         btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
2071         return -EBUSY;
2072 }
2073
2074 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2075                                        struct inode *inode, u64 file_pos,
2076                                        u64 disk_bytenr, u64 disk_num_bytes,
2077                                        u64 num_bytes, u64 ram_bytes,
2078                                        u8 compression, u8 encryption,
2079                                        u16 other_encoding, int extent_type)
2080 {
2081         struct btrfs_root *root = BTRFS_I(inode)->root;
2082         struct btrfs_file_extent_item *fi;
2083         struct btrfs_path *path;
2084         struct extent_buffer *leaf;
2085         struct btrfs_key ins;
2086         int extent_inserted = 0;
2087         int ret;
2088
2089         path = btrfs_alloc_path();
2090         if (!path)
2091                 return -ENOMEM;
2092
2093         /*
2094          * we may be replacing one extent in the tree with another.
2095          * The new extent is pinned in the extent map, and we don't want
2096          * to drop it from the cache until it is completely in the btree.
2097          *
2098          * So, tell btrfs_drop_extents to leave this extent in the cache.
2099          * the caller is expected to unpin it and allow it to be merged
2100          * with the others.
2101          */
2102         ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2103                                    file_pos + num_bytes, NULL, 0,
2104                                    1, sizeof(*fi), &extent_inserted);
2105         if (ret)
2106                 goto out;
2107
2108         if (!extent_inserted) {
2109                 ins.objectid = btrfs_ino(inode);
2110                 ins.offset = file_pos;
2111                 ins.type = BTRFS_EXTENT_DATA_KEY;
2112
2113                 path->leave_spinning = 1;
2114                 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2115                                               sizeof(*fi));
2116                 if (ret)
2117                         goto out;
2118         }
2119         leaf = path->nodes[0];
2120         fi = btrfs_item_ptr(leaf, path->slots[0],
2121                             struct btrfs_file_extent_item);
2122         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2123         btrfs_set_file_extent_type(leaf, fi, extent_type);
2124         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2125         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2126         btrfs_set_file_extent_offset(leaf, fi, 0);
2127         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2128         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2129         btrfs_set_file_extent_compression(leaf, fi, compression);
2130         btrfs_set_file_extent_encryption(leaf, fi, encryption);
2131         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2132
2133         btrfs_mark_buffer_dirty(leaf);
2134         btrfs_release_path(path);
2135
2136         inode_add_bytes(inode, num_bytes);
2137
2138         ins.objectid = disk_bytenr;
2139         ins.offset = disk_num_bytes;
2140         ins.type = BTRFS_EXTENT_ITEM_KEY;
2141         ret = btrfs_alloc_reserved_file_extent(trans, root,
2142                                         root->root_key.objectid,
2143                                         btrfs_ino(inode), file_pos,
2144                                         ram_bytes, &ins);
2145         /*
2146          * Release the reserved range from inode dirty range map, as it is
2147          * already moved into delayed_ref_head
2148          */
2149         btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2150 out:
2151         btrfs_free_path(path);
2152
2153         return ret;
2154 }
2155
2156 /* snapshot-aware defrag */
2157 struct sa_defrag_extent_backref {
2158         struct rb_node node;
2159         struct old_sa_defrag_extent *old;
2160         u64 root_id;
2161         u64 inum;
2162         u64 file_pos;
2163         u64 extent_offset;
2164         u64 num_bytes;
2165         u64 generation;
2166 };
2167
2168 struct old_sa_defrag_extent {
2169         struct list_head list;
2170         struct new_sa_defrag_extent *new;
2171
2172         u64 extent_offset;
2173         u64 bytenr;
2174         u64 offset;
2175         u64 len;
2176         int count;
2177 };
2178
2179 struct new_sa_defrag_extent {
2180         struct rb_root root;
2181         struct list_head head;
2182         struct btrfs_path *path;
2183         struct inode *inode;
2184         u64 file_pos;
2185         u64 len;
2186         u64 bytenr;
2187         u64 disk_len;
2188         u8 compress_type;
2189 };
2190
2191 static int backref_comp(struct sa_defrag_extent_backref *b1,
2192                         struct sa_defrag_extent_backref *b2)
2193 {
2194         if (b1->root_id < b2->root_id)
2195                 return -1;
2196         else if (b1->root_id > b2->root_id)
2197                 return 1;
2198
2199         if (b1->inum < b2->inum)
2200                 return -1;
2201         else if (b1->inum > b2->inum)
2202                 return 1;
2203
2204         if (b1->file_pos < b2->file_pos)
2205                 return -1;
2206         else if (b1->file_pos > b2->file_pos)
2207                 return 1;
2208
2209         /*
2210          * [------------------------------] ===> (a range of space)
2211          *     |<--->|   |<---->| =============> (fs/file tree A)
2212          * |<---------------------------->| ===> (fs/file tree B)
2213          *
2214          * A range of space can refer to two file extents in one tree while
2215          * refer to only one file extent in another tree.
2216          *
2217          * So we may process a disk offset more than one time(two extents in A)
2218          * and locate at the same extent(one extent in B), then insert two same
2219          * backrefs(both refer to the extent in B).
2220          */
2221         return 0;
2222 }
2223
2224 static void backref_insert(struct rb_root *root,
2225                            struct sa_defrag_extent_backref *backref)
2226 {
2227         struct rb_node **p = &root->rb_node;
2228         struct rb_node *parent = NULL;
2229         struct sa_defrag_extent_backref *entry;
2230         int ret;
2231
2232         while (*p) {
2233                 parent = *p;
2234                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2235
2236                 ret = backref_comp(backref, entry);
2237                 if (ret < 0)
2238                         p = &(*p)->rb_left;
2239                 else
2240                         p = &(*p)->rb_right;
2241         }
2242
2243         rb_link_node(&backref->node, parent, p);
2244         rb_insert_color(&backref->node, root);
2245 }
2246
2247 /*
2248  * Note the backref might has changed, and in this case we just return 0.
2249  */
2250 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2251                                        void *ctx)
2252 {
2253         struct btrfs_file_extent_item *extent;
2254         struct btrfs_fs_info *fs_info;
2255         struct old_sa_defrag_extent *old = ctx;
2256         struct new_sa_defrag_extent *new = old->new;
2257         struct btrfs_path *path = new->path;
2258         struct btrfs_key key;
2259         struct btrfs_root *root;
2260         struct sa_defrag_extent_backref *backref;
2261         struct extent_buffer *leaf;
2262         struct inode *inode = new->inode;
2263         int slot;
2264         int ret;
2265         u64 extent_offset;
2266         u64 num_bytes;
2267
2268         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2269             inum == btrfs_ino(inode))
2270                 return 0;
2271
2272         key.objectid = root_id;
2273         key.type = BTRFS_ROOT_ITEM_KEY;
2274         key.offset = (u64)-1;
2275
2276         fs_info = BTRFS_I(inode)->root->fs_info;
2277         root = btrfs_read_fs_root_no_name(fs_info, &key);
2278         if (IS_ERR(root)) {
2279                 if (PTR_ERR(root) == -ENOENT)
2280                         return 0;
2281                 WARN_ON(1);
2282                 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2283                          inum, offset, root_id);
2284                 return PTR_ERR(root);
2285         }
2286
2287         key.objectid = inum;
2288         key.type = BTRFS_EXTENT_DATA_KEY;
2289         if (offset > (u64)-1 << 32)
2290                 key.offset = 0;
2291         else
2292                 key.offset = offset;
2293
2294         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2295         if (WARN_ON(ret < 0))
2296                 return ret;
2297         ret = 0;
2298
2299         while (1) {
2300                 cond_resched();
2301
2302                 leaf = path->nodes[0];
2303                 slot = path->slots[0];
2304
2305                 if (slot >= btrfs_header_nritems(leaf)) {
2306                         ret = btrfs_next_leaf(root, path);
2307                         if (ret < 0) {
2308                                 goto out;
2309                         } else if (ret > 0) {
2310                                 ret = 0;
2311                                 goto out;
2312                         }
2313                         continue;
2314                 }
2315
2316                 path->slots[0]++;
2317
2318                 btrfs_item_key_to_cpu(leaf, &key, slot);
2319
2320                 if (key.objectid > inum)
2321                         goto out;
2322
2323                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2324                         continue;
2325
2326                 extent = btrfs_item_ptr(leaf, slot,
2327                                         struct btrfs_file_extent_item);
2328
2329                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2330                         continue;
2331
2332                 /*
2333                  * 'offset' refers to the exact key.offset,
2334                  * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2335                  * (key.offset - extent_offset).
2336                  */
2337                 if (key.offset != offset)
2338                         continue;
2339
2340                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2341                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2342
2343                 if (extent_offset >= old->extent_offset + old->offset +
2344                     old->len || extent_offset + num_bytes <=
2345                     old->extent_offset + old->offset)
2346                         continue;
2347                 break;
2348         }
2349
2350         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2351         if (!backref) {
2352                 ret = -ENOENT;
2353                 goto out;
2354         }
2355
2356         backref->root_id = root_id;
2357         backref->inum = inum;
2358         backref->file_pos = offset;
2359         backref->num_bytes = num_bytes;
2360         backref->extent_offset = extent_offset;
2361         backref->generation = btrfs_file_extent_generation(leaf, extent);
2362         backref->old = old;
2363         backref_insert(&new->root, backref);
2364         old->count++;
2365 out:
2366         btrfs_release_path(path);
2367         WARN_ON(ret);
2368         return ret;
2369 }
2370
2371 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2372                                    struct new_sa_defrag_extent *new)
2373 {
2374         struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2375         struct old_sa_defrag_extent *old, *tmp;
2376         int ret;
2377
2378         new->path = path;
2379
2380         list_for_each_entry_safe(old, tmp, &new->head, list) {
2381                 ret = iterate_inodes_from_logical(old->bytenr +
2382                                                   old->extent_offset, fs_info,
2383                                                   path, record_one_backref,
2384                                                   old);
2385                 if (ret < 0 && ret != -ENOENT)
2386                         return false;
2387
2388                 /* no backref to be processed for this extent */
2389                 if (!old->count) {
2390                         list_del(&old->list);
2391                         kfree(old);
2392                 }
2393         }
2394
2395         if (list_empty(&new->head))
2396                 return false;
2397
2398         return true;
2399 }
2400
2401 static int relink_is_mergable(struct extent_buffer *leaf,
2402                               struct btrfs_file_extent_item *fi,
2403                               struct new_sa_defrag_extent *new)
2404 {
2405         if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2406                 return 0;
2407
2408         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2409                 return 0;
2410
2411         if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2412                 return 0;
2413
2414         if (btrfs_file_extent_encryption(leaf, fi) ||
2415             btrfs_file_extent_other_encoding(leaf, fi))
2416                 return 0;
2417
2418         return 1;
2419 }
2420
2421 /*
2422  * Note the backref might has changed, and in this case we just return 0.
2423  */
2424 static noinline int relink_extent_backref(struct btrfs_path *path,
2425                                  struct sa_defrag_extent_backref *prev,
2426                                  struct sa_defrag_extent_backref *backref)
2427 {
2428         struct btrfs_file_extent_item *extent;
2429         struct btrfs_file_extent_item *item;
2430         struct btrfs_ordered_extent *ordered;
2431         struct btrfs_trans_handle *trans;
2432         struct btrfs_fs_info *fs_info;
2433         struct btrfs_root *root;
2434         struct btrfs_key key;
2435         struct extent_buffer *leaf;
2436         struct old_sa_defrag_extent *old = backref->old;
2437         struct new_sa_defrag_extent *new = old->new;
2438         struct inode *src_inode = new->inode;
2439         struct inode *inode;
2440         struct extent_state *cached = NULL;
2441         int ret = 0;
2442         u64 start;
2443         u64 len;
2444         u64 lock_start;
2445         u64 lock_end;
2446         bool merge = false;
2447         int index;
2448
2449         if (prev && prev->root_id == backref->root_id &&
2450             prev->inum == backref->inum &&
2451             prev->file_pos + prev->num_bytes == backref->file_pos)
2452                 merge = true;
2453
2454         /* step 1: get root */
2455         key.objectid = backref->root_id;
2456         key.type = BTRFS_ROOT_ITEM_KEY;
2457         key.offset = (u64)-1;
2458
2459         fs_info = BTRFS_I(src_inode)->root->fs_info;
2460         index = srcu_read_lock(&fs_info->subvol_srcu);
2461
2462         root = btrfs_read_fs_root_no_name(fs_info, &key);
2463         if (IS_ERR(root)) {
2464                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2465                 if (PTR_ERR(root) == -ENOENT)
2466                         return 0;
2467                 return PTR_ERR(root);
2468         }
2469
2470         if (btrfs_root_readonly(root)) {
2471                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2472                 return 0;
2473         }
2474
2475         /* step 2: get inode */
2476         key.objectid = backref->inum;
2477         key.type = BTRFS_INODE_ITEM_KEY;
2478         key.offset = 0;
2479
2480         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2481         if (IS_ERR(inode)) {
2482                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2483                 return 0;
2484         }
2485
2486         srcu_read_unlock(&fs_info->subvol_srcu, index);
2487
2488         /* step 3: relink backref */
2489         lock_start = backref->file_pos;
2490         lock_end = backref->file_pos + backref->num_bytes - 1;
2491         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2492                          &cached);
2493
2494         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2495         if (ordered) {
2496                 btrfs_put_ordered_extent(ordered);
2497                 goto out_unlock;
2498         }
2499
2500         trans = btrfs_join_transaction(root);
2501         if (IS_ERR(trans)) {
2502                 ret = PTR_ERR(trans);
2503                 goto out_unlock;
2504         }
2505
2506         key.objectid = backref->inum;
2507         key.type = BTRFS_EXTENT_DATA_KEY;
2508         key.offset = backref->file_pos;
2509
2510         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2511         if (ret < 0) {
2512                 goto out_free_path;
2513         } else if (ret > 0) {
2514                 ret = 0;
2515                 goto out_free_path;
2516         }
2517
2518         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2519                                 struct btrfs_file_extent_item);
2520
2521         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2522             backref->generation)
2523                 goto out_free_path;
2524
2525         btrfs_release_path(path);
2526
2527         start = backref->file_pos;
2528         if (backref->extent_offset < old->extent_offset + old->offset)
2529                 start += old->extent_offset + old->offset -
2530                          backref->extent_offset;
2531
2532         len = min(backref->extent_offset + backref->num_bytes,
2533                   old->extent_offset + old->offset + old->len);
2534         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2535
2536         ret = btrfs_drop_extents(trans, root, inode, start,
2537                                  start + len, 1);
2538         if (ret)
2539                 goto out_free_path;
2540 again:
2541         key.objectid = btrfs_ino(inode);
2542         key.type = BTRFS_EXTENT_DATA_KEY;
2543         key.offset = start;
2544
2545         path->leave_spinning = 1;
2546         if (merge) {
2547                 struct btrfs_file_extent_item *fi;
2548                 u64 extent_len;
2549                 struct btrfs_key found_key;
2550
2551                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2552                 if (ret < 0)
2553                         goto out_free_path;
2554
2555                 path->slots[0]--;
2556                 leaf = path->nodes[0];
2557                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2558
2559                 fi = btrfs_item_ptr(leaf, path->slots[0],
2560                                     struct btrfs_file_extent_item);
2561                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2562
2563                 if (extent_len + found_key.offset == start &&
2564                     relink_is_mergable(leaf, fi, new)) {
2565                         btrfs_set_file_extent_num_bytes(leaf, fi,
2566                                                         extent_len + len);
2567                         btrfs_mark_buffer_dirty(leaf);
2568                         inode_add_bytes(inode, len);
2569
2570                         ret = 1;
2571                         goto out_free_path;
2572                 } else {
2573                         merge = false;
2574                         btrfs_release_path(path);
2575                         goto again;
2576                 }
2577         }
2578
2579         ret = btrfs_insert_empty_item(trans, root, path, &key,
2580                                         sizeof(*extent));
2581         if (ret) {
2582                 btrfs_abort_transaction(trans, root, ret);
2583                 goto out_free_path;
2584         }
2585
2586         leaf = path->nodes[0];
2587         item = btrfs_item_ptr(leaf, path->slots[0],
2588                                 struct btrfs_file_extent_item);
2589         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2590         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2591         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2592         btrfs_set_file_extent_num_bytes(leaf, item, len);
2593         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2594         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2595         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2596         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2597         btrfs_set_file_extent_encryption(leaf, item, 0);
2598         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2599
2600         btrfs_mark_buffer_dirty(leaf);
2601         inode_add_bytes(inode, len);
2602         btrfs_release_path(path);
2603
2604         ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2605                         new->disk_len, 0,
2606                         backref->root_id, backref->inum,
2607                         new->file_pos); /* start - extent_offset */
2608         if (ret) {
2609                 btrfs_abort_transaction(trans, root, ret);
2610                 goto out_free_path;
2611         }
2612
2613         ret = 1;
2614 out_free_path:
2615         btrfs_release_path(path);
2616         path->leave_spinning = 0;
2617         btrfs_end_transaction(trans, root);
2618 out_unlock:
2619         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2620                              &cached, GFP_NOFS);
2621         iput(inode);
2622         return ret;
2623 }
2624
2625 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2626 {
2627         struct old_sa_defrag_extent *old, *tmp;
2628
2629         if (!new)
2630                 return;
2631
2632         list_for_each_entry_safe(old, tmp, &new->head, list) {
2633                 kfree(old);
2634         }
2635         kfree(new);
2636 }
2637
2638 static void relink_file_extents(struct new_sa_defrag_extent *new)
2639 {
2640         struct btrfs_path *path;
2641         struct sa_defrag_extent_backref *backref;
2642         struct sa_defrag_extent_backref *prev = NULL;
2643         struct inode *inode;
2644         struct btrfs_root *root;
2645         struct rb_node *node;
2646         int ret;
2647
2648         inode = new->inode;
2649         root = BTRFS_I(inode)->root;
2650
2651         path = btrfs_alloc_path();
2652         if (!path)
2653                 return;
2654
2655         if (!record_extent_backrefs(path, new)) {
2656                 btrfs_free_path(path);
2657                 goto out;
2658         }
2659         btrfs_release_path(path);
2660
2661         while (1) {
2662                 node = rb_first(&new->root);
2663                 if (!node)
2664                         break;
2665                 rb_erase(node, &new->root);
2666
2667                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2668
2669                 ret = relink_extent_backref(path, prev, backref);
2670                 WARN_ON(ret < 0);
2671
2672                 kfree(prev);
2673
2674                 if (ret == 1)
2675                         prev = backref;
2676                 else
2677                         prev = NULL;
2678                 cond_resched();
2679         }
2680         kfree(prev);
2681
2682         btrfs_free_path(path);
2683 out:
2684         free_sa_defrag_extent(new);
2685
2686         atomic_dec(&root->fs_info->defrag_running);
2687         wake_up(&root->fs_info->transaction_wait);
2688 }
2689
2690 static struct new_sa_defrag_extent *
2691 record_old_file_extents(struct inode *inode,
2692                         struct btrfs_ordered_extent *ordered)
2693 {
2694         struct btrfs_root *root = BTRFS_I(inode)->root;
2695         struct btrfs_path *path;
2696         struct btrfs_key key;
2697         struct old_sa_defrag_extent *old;
2698         struct new_sa_defrag_extent *new;
2699         int ret;
2700
2701         new = kmalloc(sizeof(*new), GFP_NOFS);
2702         if (!new)
2703                 return NULL;
2704
2705         new->inode = inode;
2706         new->file_pos = ordered->file_offset;
2707         new->len = ordered->len;
2708         new->bytenr = ordered->start;
2709         new->disk_len = ordered->disk_len;
2710         new->compress_type = ordered->compress_type;
2711         new->root = RB_ROOT;
2712         INIT_LIST_HEAD(&new->head);
2713
2714         path = btrfs_alloc_path();
2715         if (!path)
2716                 goto out_kfree;
2717
2718         key.objectid = btrfs_ino(inode);
2719         key.type = BTRFS_EXTENT_DATA_KEY;
2720         key.offset = new->file_pos;
2721
2722         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2723         if (ret < 0)
2724                 goto out_free_path;
2725         if (ret > 0 && path->slots[0] > 0)
2726                 path->slots[0]--;
2727
2728         /* find out all the old extents for the file range */
2729         while (1) {
2730                 struct btrfs_file_extent_item *extent;
2731                 struct extent_buffer *l;
2732                 int slot;
2733                 u64 num_bytes;
2734                 u64 offset;
2735                 u64 end;
2736                 u64 disk_bytenr;
2737                 u64 extent_offset;
2738
2739                 l = path->nodes[0];
2740                 slot = path->slots[0];
2741
2742                 if (slot >= btrfs_header_nritems(l)) {
2743                         ret = btrfs_next_leaf(root, path);
2744                         if (ret < 0)
2745                                 goto out_free_path;
2746                         else if (ret > 0)
2747                                 break;
2748                         continue;
2749                 }
2750
2751                 btrfs_item_key_to_cpu(l, &key, slot);
2752
2753                 if (key.objectid != btrfs_ino(inode))
2754                         break;
2755                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2756                         break;
2757                 if (key.offset >= new->file_pos + new->len)
2758                         break;
2759
2760                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2761
2762                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2763                 if (key.offset + num_bytes < new->file_pos)
2764                         goto next;
2765
2766                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2767                 if (!disk_bytenr)
2768                         goto next;
2769
2770                 extent_offset = btrfs_file_extent_offset(l, extent);
2771
2772                 old = kmalloc(sizeof(*old), GFP_NOFS);
2773                 if (!old)
2774                         goto out_free_path;
2775
2776                 offset = max(new->file_pos, key.offset);
2777                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2778
2779                 old->bytenr = disk_bytenr;
2780                 old->extent_offset = extent_offset;
2781                 old->offset = offset - key.offset;
2782                 old->len = end - offset;
2783                 old->new = new;
2784                 old->count = 0;
2785                 list_add_tail(&old->list, &new->head);
2786 next:
2787                 path->slots[0]++;
2788                 cond_resched();
2789         }
2790
2791         btrfs_free_path(path);
2792         atomic_inc(&root->fs_info->defrag_running);
2793
2794         return new;
2795
2796 out_free_path:
2797         btrfs_free_path(path);
2798 out_kfree:
2799         free_sa_defrag_extent(new);
2800         return NULL;
2801 }
2802
2803 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2804                                          u64 start, u64 len)
2805 {
2806         struct btrfs_block_group_cache *cache;
2807
2808         cache = btrfs_lookup_block_group(root->fs_info, start);
2809         ASSERT(cache);
2810
2811         spin_lock(&cache->lock);
2812         cache->delalloc_bytes -= len;
2813         spin_unlock(&cache->lock);
2814
2815         btrfs_put_block_group(cache);
2816 }
2817
2818 /* as ordered data IO finishes, this gets called so we can finish
2819  * an ordered extent if the range of bytes in the file it covers are
2820  * fully written.
2821  */
2822 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2823 {
2824         struct inode *inode = ordered_extent->inode;
2825         struct btrfs_root *root = BTRFS_I(inode)->root;
2826         struct btrfs_trans_handle *trans = NULL;
2827         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2828         struct extent_state *cached_state = NULL;
2829         struct new_sa_defrag_extent *new = NULL;
2830         int compress_type = 0;
2831         int ret = 0;
2832         u64 logical_len = ordered_extent->len;
2833         bool nolock;
2834         bool truncated = false;
2835
2836         nolock = btrfs_is_free_space_inode(inode);
2837
2838         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2839                 ret = -EIO;
2840                 goto out;
2841         }
2842
2843         btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2844                                      ordered_extent->file_offset +
2845                                      ordered_extent->len - 1);
2846
2847         if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2848                 truncated = true;
2849                 logical_len = ordered_extent->truncated_len;
2850                 /* Truncated the entire extent, don't bother adding */
2851                 if (!logical_len)
2852                         goto out;
2853         }
2854
2855         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2856                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2857
2858                 /*
2859                  * For mwrite(mmap + memset to write) case, we still reserve
2860                  * space for NOCOW range.
2861                  * As NOCOW won't cause a new delayed ref, just free the space
2862                  */
2863                 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2864                                        ordered_extent->len);
2865                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2866                 if (nolock)
2867                         trans = btrfs_join_transaction_nolock(root);
2868                 else
2869                         trans = btrfs_join_transaction(root);
2870                 if (IS_ERR(trans)) {
2871                         ret = PTR_ERR(trans);
2872                         trans = NULL;
2873                         goto out;
2874                 }
2875                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2876                 ret = btrfs_update_inode_fallback(trans, root, inode);
2877                 if (ret) /* -ENOMEM or corruption */
2878                         btrfs_abort_transaction(trans, root, ret);
2879                 goto out;
2880         }
2881
2882         lock_extent_bits(io_tree, ordered_extent->file_offset,
2883                          ordered_extent->file_offset + ordered_extent->len - 1,
2884                          &cached_state);
2885
2886         ret = test_range_bit(io_tree, ordered_extent->file_offset,
2887                         ordered_extent->file_offset + ordered_extent->len - 1,
2888                         EXTENT_DEFRAG, 1, cached_state);
2889         if (ret) {
2890                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2891                 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2892                         /* the inode is shared */
2893                         new = record_old_file_extents(inode, ordered_extent);
2894
2895                 clear_extent_bit(io_tree, ordered_extent->file_offset,
2896                         ordered_extent->file_offset + ordered_extent->len - 1,
2897                         EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2898         }
2899
2900         if (nolock)
2901                 trans = btrfs_join_transaction_nolock(root);
2902         else
2903                 trans = btrfs_join_transaction(root);
2904         if (IS_ERR(trans)) {
2905                 ret = PTR_ERR(trans);
2906                 trans = NULL;
2907                 goto out_unlock;
2908         }
2909
2910         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2911
2912         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2913                 compress_type = ordered_extent->compress_type;
2914         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2915                 BUG_ON(compress_type);
2916                 ret = btrfs_mark_extent_written(trans, inode,
2917                                                 ordered_extent->file_offset,
2918                                                 ordered_extent->file_offset +
2919                                                 logical_len);
2920         } else {
2921                 BUG_ON(root == root->fs_info->tree_root);
2922                 ret = insert_reserved_file_extent(trans, inode,
2923                                                 ordered_extent->file_offset,
2924                                                 ordered_extent->start,
2925                                                 ordered_extent->disk_len,
2926                                                 logical_len, logical_len,
2927                                                 compress_type, 0, 0,
2928                                                 BTRFS_FILE_EXTENT_REG);
2929                 if (!ret)
2930                         btrfs_release_delalloc_bytes(root,
2931                                                      ordered_extent->start,
2932                                                      ordered_extent->disk_len);
2933         }
2934         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2935                            ordered_extent->file_offset, ordered_extent->len,
2936                            trans->transid);
2937         if (ret < 0) {
2938                 btrfs_abort_transaction(trans, root, ret);
2939                 goto out_unlock;
2940         }
2941
2942         add_pending_csums(trans, inode, ordered_extent->file_offset,
2943                           &ordered_extent->list);
2944
2945         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2946         ret = btrfs_update_inode_fallback(trans, root, inode);
2947         if (ret) { /* -ENOMEM or corruption */
2948                 btrfs_abort_transaction(trans, root, ret);
2949                 goto out_unlock;
2950         }
2951         ret = 0;
2952 out_unlock:
2953         unlock_extent_cached(io_tree, ordered_extent->file_offset,
2954                              ordered_extent->file_offset +
2955                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
2956 out:
2957         if (root != root->fs_info->tree_root)
2958                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2959         if (trans)
2960                 btrfs_end_transaction(trans, root);
2961
2962         if (ret || truncated) {
2963                 u64 start, end;
2964
2965                 if (truncated)
2966                         start = ordered_extent->file_offset + logical_len;
2967                 else
2968                         start = ordered_extent->file_offset;
2969                 end = ordered_extent->file_offset + ordered_extent->len - 1;
2970                 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2971
2972                 /* Drop the cache for the part of the extent we didn't write. */
2973                 btrfs_drop_extent_cache(inode, start, end, 0);
2974
2975                 /*
2976                  * If the ordered extent had an IOERR or something else went
2977                  * wrong we need to return the space for this ordered extent
2978                  * back to the allocator.  We only free the extent in the
2979                  * truncated case if we didn't write out the extent at all.
2980                  */
2981                 if ((ret || !logical_len) &&
2982                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2983                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2984                         btrfs_free_reserved_extent(root, ordered_extent->start,
2985                                                    ordered_extent->disk_len, 1);
2986         }
2987
2988
2989         /*
2990          * This needs to be done to make sure anybody waiting knows we are done
2991          * updating everything for this ordered extent.
2992          */
2993         btrfs_remove_ordered_extent(inode, ordered_extent);
2994
2995         /* for snapshot-aware defrag */
2996         if (new) {
2997                 if (ret) {
2998                         free_sa_defrag_extent(new);
2999                         atomic_dec(&root->fs_info->defrag_running);
3000                 } else {
3001                         relink_file_extents(new);
3002                 }
3003         }
3004
3005         /* once for us */
3006         btrfs_put_ordered_extent(ordered_extent);
3007         /* once for the tree */
3008         btrfs_put_ordered_extent(ordered_extent);
3009
3010         return ret;
3011 }
3012
3013 static void finish_ordered_fn(struct btrfs_work *work)
3014 {
3015         struct btrfs_ordered_extent *ordered_extent;
3016         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3017         btrfs_finish_ordered_io(ordered_extent);
3018 }
3019
3020 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3021                                 struct extent_state *state, int uptodate)
3022 {
3023         struct inode *inode = page->mapping->host;
3024         struct btrfs_root *root = BTRFS_I(inode)->root;
3025         struct btrfs_ordered_extent *ordered_extent = NULL;
3026         struct btrfs_workqueue *wq;
3027         btrfs_work_func_t func;
3028
3029         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3030
3031         ClearPagePrivate2(page);
3032         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3033                                             end - start + 1, uptodate))
3034                 return 0;
3035
3036         if (btrfs_is_free_space_inode(inode)) {
3037                 wq = root->fs_info->endio_freespace_worker;
3038                 func = btrfs_freespace_write_helper;
3039         } else {
3040                 wq = root->fs_info->endio_write_workers;
3041                 func = btrfs_endio_write_helper;
3042         }
3043
3044         btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3045                         NULL);
3046         btrfs_queue_work(wq, &ordered_extent->work);
3047
3048         return 0;
3049 }
3050
3051 static int __readpage_endio_check(struct inode *inode,
3052                                   struct btrfs_io_bio *io_bio,
3053                                   int icsum, struct page *page,
3054                                   int pgoff, u64 start, size_t len)
3055 {
3056         char *kaddr;
3057         u32 csum_expected;
3058         u32 csum = ~(u32)0;
3059
3060         csum_expected = *(((u32 *)io_bio->csum) + icsum);
3061
3062         kaddr = kmap_atomic(page);
3063         csum = btrfs_csum_data(kaddr + pgoff, csum,  len);
3064         btrfs_csum_final(csum, (char *)&csum);
3065         if (csum != csum_expected)
3066                 goto zeroit;
3067
3068         kunmap_atomic(kaddr);
3069         return 0;
3070 zeroit:
3071         btrfs_warn_rl(BTRFS_I(inode)->root->fs_info,
3072                 "csum failed ino %llu off %llu csum %u expected csum %u",
3073                            btrfs_ino(inode), start, csum, csum_expected);
3074         memset(kaddr + pgoff, 1, len);
3075         flush_dcache_page(page);
3076         kunmap_atomic(kaddr);
3077         if (csum_expected == 0)
3078                 return 0;
3079         return -EIO;
3080 }
3081
3082 /*
3083  * when reads are done, we need to check csums to verify the data is correct
3084  * if there's a match, we allow the bio to finish.  If not, the code in
3085  * extent_io.c will try to find good copies for us.
3086  */
3087 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3088                                       u64 phy_offset, struct page *page,
3089                                       u64 start, u64 end, int mirror)
3090 {
3091         size_t offset = start - page_offset(page);
3092         struct inode *inode = page->mapping->host;
3093         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3094         struct btrfs_root *root = BTRFS_I(inode)->root;
3095
3096         if (PageChecked(page)) {
3097                 ClearPageChecked(page);
3098                 return 0;
3099         }
3100
3101         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3102                 return 0;
3103
3104         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3105             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3106                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3107                 return 0;
3108         }
3109
3110         phy_offset >>= inode->i_sb->s_blocksize_bits;
3111         return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3112                                       start, (size_t)(end - start + 1));
3113 }
3114
3115 void btrfs_add_delayed_iput(struct inode *inode)
3116 {
3117         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3118         struct btrfs_inode *binode = BTRFS_I(inode);
3119
3120         if (atomic_add_unless(&inode->i_count, -1, 1))
3121                 return;
3122
3123         spin_lock(&fs_info->delayed_iput_lock);
3124         if (binode->delayed_iput_count == 0) {
3125                 ASSERT(list_empty(&binode->delayed_iput));
3126                 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3127         } else {
3128                 binode->delayed_iput_count++;
3129         }
3130         spin_unlock(&fs_info->delayed_iput_lock);
3131 }
3132
3133 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3134 {
3135         struct btrfs_fs_info *fs_info = root->fs_info;
3136
3137         spin_lock(&fs_info->delayed_iput_lock);
3138         while (!list_empty(&fs_info->delayed_iputs)) {
3139                 struct btrfs_inode *inode;
3140
3141                 inode = list_first_entry(&fs_info->delayed_iputs,
3142                                 struct btrfs_inode, delayed_iput);
3143                 if (inode->delayed_iput_count) {
3144                         inode->delayed_iput_count--;
3145                         list_move_tail(&inode->delayed_iput,
3146                                         &fs_info->delayed_iputs);
3147                 } else {
3148                         list_del_init(&inode->delayed_iput);
3149                 }
3150                 spin_unlock(&fs_info->delayed_iput_lock);
3151                 iput(&inode->vfs_inode);
3152                 spin_lock(&fs_info->delayed_iput_lock);
3153         }
3154         spin_unlock(&fs_info->delayed_iput_lock);
3155 }
3156
3157 /*
3158  * This is called in transaction commit time. If there are no orphan
3159  * files in the subvolume, it removes orphan item and frees block_rsv
3160  * structure.
3161  */
3162 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3163                               struct btrfs_root *root)
3164 {
3165         struct btrfs_block_rsv *block_rsv;
3166         int ret;
3167
3168         if (atomic_read(&root->orphan_inodes) ||
3169             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3170                 return;
3171
3172         spin_lock(&root->orphan_lock);
3173         if (atomic_read(&root->orphan_inodes)) {
3174                 spin_unlock(&root->orphan_lock);
3175                 return;
3176         }
3177
3178         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3179                 spin_unlock(&root->orphan_lock);
3180                 return;
3181         }
3182
3183         block_rsv = root->orphan_block_rsv;
3184         root->orphan_block_rsv = NULL;
3185         spin_unlock(&root->orphan_lock);
3186
3187         if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3188             btrfs_root_refs(&root->root_item) > 0) {
3189                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3190                                             root->root_key.objectid);
3191                 if (ret)
3192                         btrfs_abort_transaction(trans, root, ret);
3193                 else
3194                         clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3195                                   &root->state);
3196         }
3197
3198         if (block_rsv) {
3199                 WARN_ON(block_rsv->size > 0);
3200                 btrfs_free_block_rsv(root, block_rsv);
3201         }
3202 }
3203
3204 /*
3205  * This creates an orphan entry for the given inode in case something goes
3206  * wrong in the middle of an unlink/truncate.
3207  *
3208  * NOTE: caller of this function should reserve 5 units of metadata for
3209  *       this function.
3210  */
3211 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3212 {
3213         struct btrfs_root *root = BTRFS_I(inode)->root;
3214         struct btrfs_block_rsv *block_rsv = NULL;
3215         int reserve = 0;
3216         int insert = 0;
3217         int ret;
3218
3219         if (!root->orphan_block_rsv) {
3220                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3221                 if (!block_rsv)
3222                         return -ENOMEM;
3223         }
3224
3225         spin_lock(&root->orphan_lock);
3226         if (!root->orphan_block_rsv) {
3227                 root->orphan_block_rsv = block_rsv;
3228         } else if (block_rsv) {
3229                 btrfs_free_block_rsv(root, block_rsv);
3230                 block_rsv = NULL;
3231         }
3232
3233         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3234                               &BTRFS_I(inode)->runtime_flags)) {
3235 #if 0
3236                 /*
3237                  * For proper ENOSPC handling, we should do orphan
3238                  * cleanup when mounting. But this introduces backward
3239                  * compatibility issue.
3240                  */
3241                 if (!xchg(&root->orphan_item_inserted, 1))
3242                         insert = 2;
3243                 else
3244                         insert = 1;
3245 #endif
3246                 insert = 1;
3247                 atomic_inc(&root->orphan_inodes);
3248         }
3249
3250         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3251                               &BTRFS_I(inode)->runtime_flags))
3252                 reserve = 1;
3253         spin_unlock(&root->orphan_lock);
3254
3255         /* grab metadata reservation from transaction handle */
3256         if (reserve) {
3257                 ret = btrfs_orphan_reserve_metadata(trans, inode);
3258                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3259         }
3260
3261         /* insert an orphan item to track this unlinked/truncated file */
3262         if (insert >= 1) {
3263                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3264                 if (ret) {
3265                         atomic_dec(&root->orphan_inodes);
3266                         if (reserve) {
3267                                 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3268                                           &BTRFS_I(inode)->runtime_flags);
3269                                 btrfs_orphan_release_metadata(inode);
3270                         }
3271                         if (ret != -EEXIST) {
3272                                 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3273                                           &BTRFS_I(inode)->runtime_flags);
3274                                 btrfs_abort_transaction(trans, root, ret);
3275                                 return ret;
3276                         }
3277                 }
3278                 ret = 0;
3279         }
3280
3281         /* insert an orphan item to track subvolume contains orphan files */
3282         if (insert >= 2) {
3283                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3284                                                root->root_key.objectid);
3285                 if (ret && ret != -EEXIST) {
3286                         btrfs_abort_transaction(trans, root, ret);
3287                         return ret;
3288                 }
3289         }
3290         return 0;
3291 }
3292
3293 /*
3294  * We have done the truncate/delete so we can go ahead and remove the orphan
3295  * item for this particular inode.
3296  */
3297 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3298                             struct inode *inode)
3299 {
3300         struct btrfs_root *root = BTRFS_I(inode)->root;
3301         int delete_item = 0;
3302         int release_rsv = 0;
3303         int ret = 0;
3304
3305         spin_lock(&root->orphan_lock);
3306         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3307                                &BTRFS_I(inode)->runtime_flags))
3308                 delete_item = 1;
3309
3310         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3311                                &BTRFS_I(inode)->runtime_flags))
3312                 release_rsv = 1;
3313         spin_unlock(&root->orphan_lock);
3314
3315         if (delete_item) {
3316                 atomic_dec(&root->orphan_inodes);
3317                 if (trans)
3318                         ret = btrfs_del_orphan_item(trans, root,
3319                                                     btrfs_ino(inode));
3320         }
3321
3322         if (release_rsv)
3323                 btrfs_orphan_release_metadata(inode);
3324
3325         return ret;
3326 }
3327
3328 /*
3329  * this cleans up any orphans that may be left on the list from the last use
3330  * of this root.
3331  */
3332 int btrfs_orphan_cleanup(struct btrfs_root *root)
3333 {
3334         struct btrfs_path *path;
3335         struct extent_buffer *leaf;
3336         struct btrfs_key key, found_key;
3337         struct btrfs_trans_handle *trans;
3338         struct inode *inode;
3339         u64 last_objectid = 0;
3340         int ret = 0, nr_unlink = 0, nr_truncate = 0;
3341
3342         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3343                 return 0;
3344
3345         path = btrfs_alloc_path();
3346         if (!path) {
3347                 ret = -ENOMEM;
3348                 goto out;
3349         }
3350         path->reada = READA_BACK;
3351
3352         key.objectid = BTRFS_ORPHAN_OBJECTID;
3353         key.type = BTRFS_ORPHAN_ITEM_KEY;
3354         key.offset = (u64)-1;
3355
3356         while (1) {
3357                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3358                 if (ret < 0)
3359                         goto out;
3360
3361                 /*
3362                  * if ret == 0 means we found what we were searching for, which
3363                  * is weird, but possible, so only screw with path if we didn't
3364                  * find the key and see if we have stuff that matches
3365                  */
3366                 if (ret > 0) {
3367                         ret = 0;
3368                         if (path->slots[0] == 0)
3369                                 break;
3370                         path->slots[0]--;
3371                 }
3372
3373                 /* pull out the item */
3374                 leaf = path->nodes[0];
3375                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3376
3377                 /* make sure the item matches what we want */
3378                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3379                         break;
3380                 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3381                         break;
3382
3383                 /* release the path since we're done with it */
3384                 btrfs_release_path(path);
3385
3386                 /*
3387                  * this is where we are basically btrfs_lookup, without the
3388                  * crossing root thing.  we store the inode number in the
3389                  * offset of the orphan item.
3390                  */
3391
3392                 if (found_key.offset == last_objectid) {
3393                         btrfs_err(root->fs_info,
3394                                 "Error removing orphan entry, stopping orphan cleanup");
3395                         ret = -EINVAL;
3396                         goto out;
3397                 }
3398
3399                 last_objectid = found_key.offset;
3400
3401                 found_key.objectid = found_key.offset;
3402                 found_key.type = BTRFS_INODE_ITEM_KEY;
3403                 found_key.offset = 0;
3404                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3405                 ret = PTR_ERR_OR_ZERO(inode);
3406                 if (ret && ret != -ESTALE)
3407                         goto out;
3408
3409                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3410                         struct btrfs_root *dead_root;
3411                         struct btrfs_fs_info *fs_info = root->fs_info;
3412                         int is_dead_root = 0;
3413
3414                         /*
3415                          * this is an orphan in the tree root. Currently these
3416                          * could come from 2 sources:
3417                          *  a) a snapshot deletion in progress
3418                          *  b) a free space cache inode
3419                          * We need to distinguish those two, as the snapshot
3420                          * orphan must not get deleted.
3421                          * find_dead_roots already ran before us, so if this
3422                          * is a snapshot deletion, we should find the root
3423                          * in the dead_roots list
3424                          */
3425                         spin_lock(&fs_info->trans_lock);
3426                         list_for_each_entry(dead_root, &fs_info->dead_roots,
3427                                             root_list) {
3428                                 if (dead_root->root_key.objectid ==
3429                                     found_key.objectid) {
3430                                         is_dead_root = 1;
3431                                         break;
3432                                 }
3433                         }
3434                         spin_unlock(&fs_info->trans_lock);
3435                         if (is_dead_root) {
3436                                 /* prevent this orphan from being found again */
3437                                 key.offset = found_key.objectid - 1;
3438                                 continue;
3439                         }
3440                 }
3441                 /*
3442                  * Inode is already gone but the orphan item is still there,
3443                  * kill the orphan item.
3444                  */
3445                 if (ret == -ESTALE) {
3446                         trans = btrfs_start_transaction(root, 1);
3447                         if (IS_ERR(trans)) {
3448                                 ret = PTR_ERR(trans);
3449                                 goto out;
3450                         }
3451                         btrfs_debug(root->fs_info, "auto deleting %Lu",
3452                                 found_key.objectid);
3453                         ret = btrfs_del_orphan_item(trans, root,
3454                                                     found_key.objectid);
3455                         btrfs_end_transaction(trans, root);
3456                         if (ret)
3457                                 goto out;
3458                         continue;
3459                 }
3460
3461                 /*
3462                  * add this inode to the orphan list so btrfs_orphan_del does
3463                  * the proper thing when we hit it
3464                  */
3465                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3466                         &BTRFS_I(inode)->runtime_flags);
3467                 atomic_inc(&root->orphan_inodes);
3468
3469                 /* if we have links, this was a truncate, lets do that */
3470                 if (inode->i_nlink) {
3471                         if (WARN_ON(!S_ISREG(inode->i_mode))) {
3472                                 iput(inode);
3473                                 continue;
3474                         }
3475                         nr_truncate++;
3476
3477                         /* 1 for the orphan item deletion. */
3478                         trans = btrfs_start_transaction(root, 1);
3479                         if (IS_ERR(trans)) {
3480                                 iput(inode);
3481                                 ret = PTR_ERR(trans);
3482                                 goto out;
3483                         }
3484                         ret = btrfs_orphan_add(trans, inode);
3485                         btrfs_end_transaction(trans, root);
3486                         if (ret) {
3487                                 iput(inode);
3488                                 goto out;
3489                         }
3490
3491                         ret = btrfs_truncate(inode);
3492                         if (ret)
3493                                 btrfs_orphan_del(NULL, inode);
3494                 } else {
3495                         nr_unlink++;
3496                 }
3497
3498                 /* this will do delete_inode and everything for us */
3499                 iput(inode);
3500                 if (ret)
3501                         goto out;
3502         }
3503         /* release the path since we're done with it */
3504         btrfs_release_path(path);
3505
3506         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3507
3508         if (root->orphan_block_rsv)
3509                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3510                                         (u64)-1);
3511
3512         if (root->orphan_block_rsv ||
3513             test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3514                 trans = btrfs_join_transaction(root);
3515                 if (!IS_ERR(trans))
3516                         btrfs_end_transaction(trans, root);
3517         }
3518
3519         if (nr_unlink)
3520                 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3521         if (nr_truncate)
3522                 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3523
3524 out:
3525         if (ret)
3526                 btrfs_err(root->fs_info,
3527                         "could not do orphan cleanup %d", ret);
3528         btrfs_free_path(path);
3529         return ret;
3530 }
3531
3532 /*
3533  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3534  * don't find any xattrs, we know there can't be any acls.
3535  *
3536  * slot is the slot the inode is in, objectid is the objectid of the inode
3537  */
3538 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3539                                           int slot, u64 objectid,
3540                                           int *first_xattr_slot)
3541 {
3542         u32 nritems = btrfs_header_nritems(leaf);
3543         struct btrfs_key found_key;
3544         static u64 xattr_access = 0;
3545         static u64 xattr_default = 0;
3546         int scanned = 0;
3547
3548         if (!xattr_access) {
3549                 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3550                                         strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3551                 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3552                                         strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3553         }
3554
3555         slot++;
3556         *first_xattr_slot = -1;
3557         while (slot < nritems) {
3558                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3559
3560                 /* we found a different objectid, there must not be acls */
3561                 if (found_key.objectid != objectid)
3562                         return 0;
3563
3564                 /* we found an xattr, assume we've got an acl */
3565                 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3566                         if (*first_xattr_slot == -1)
3567                                 *first_xattr_slot = slot;
3568                         if (found_key.offset == xattr_access ||
3569                             found_key.offset == xattr_default)
3570                                 return 1;
3571                 }
3572
3573                 /*
3574                  * we found a key greater than an xattr key, there can't
3575                  * be any acls later on
3576                  */
3577                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3578                         return 0;
3579
3580                 slot++;
3581                 scanned++;
3582
3583                 /*
3584                  * it goes inode, inode backrefs, xattrs, extents,
3585                  * so if there are a ton of hard links to an inode there can
3586                  * be a lot of backrefs.  Don't waste time searching too hard,
3587                  * this is just an optimization
3588                  */
3589                 if (scanned >= 8)
3590                         break;
3591         }
3592         /* we hit the end of the leaf before we found an xattr or
3593          * something larger than an xattr.  We have to assume the inode
3594          * has acls
3595          */
3596         if (*first_xattr_slot == -1)
3597                 *first_xattr_slot = slot;
3598         return 1;
3599 }
3600
3601 /*
3602  * read an inode from the btree into the in-memory inode
3603  */
3604 static void btrfs_read_locked_inode(struct inode *inode)
3605 {
3606         struct btrfs_path *path;
3607         struct extent_buffer *leaf;
3608         struct btrfs_inode_item *inode_item;
3609         struct btrfs_root *root = BTRFS_I(inode)->root;
3610         struct btrfs_key location;
3611         unsigned long ptr;
3612         int maybe_acls;
3613         u32 rdev;
3614         int ret;
3615         bool filled = false;
3616         int first_xattr_slot;
3617
3618         ret = btrfs_fill_inode(inode, &rdev);
3619         if (!ret)
3620                 filled = true;
3621
3622         path = btrfs_alloc_path();
3623         if (!path)
3624                 goto make_bad;
3625
3626         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3627
3628         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3629         if (ret)
3630                 goto make_bad;
3631
3632         leaf = path->nodes[0];
3633
3634         if (filled)
3635                 goto cache_index;
3636
3637         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3638                                     struct btrfs_inode_item);
3639         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3640         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3641         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3642         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3643         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3644
3645         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3646         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3647
3648         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3649         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3650
3651         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3652         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3653
3654         BTRFS_I(inode)->i_otime.tv_sec =
3655                 btrfs_timespec_sec(leaf, &inode_item->otime);
3656         BTRFS_I(inode)->i_otime.tv_nsec =
3657                 btrfs_timespec_nsec(leaf, &inode_item->otime);
3658
3659         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3660         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3661         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3662
3663         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3664         inode->i_generation = BTRFS_I(inode)->generation;
3665         inode->i_rdev = 0;
3666         rdev = btrfs_inode_rdev(leaf, inode_item);
3667
3668         BTRFS_I(inode)->index_cnt = (u64)-1;
3669         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3670
3671 cache_index:
3672         /*
3673          * If we were modified in the current generation and evicted from memory
3674          * and then re-read we need to do a full sync since we don't have any
3675          * idea about which extents were modified before we were evicted from
3676          * cache.
3677          *
3678          * This is required for both inode re-read from disk and delayed inode
3679          * in delayed_nodes_tree.
3680          */
3681         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3682                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3683                         &BTRFS_I(inode)->runtime_flags);
3684
3685         /*
3686          * We don't persist the id of the transaction where an unlink operation
3687          * against the inode was last made. So here we assume the inode might
3688          * have been evicted, and therefore the exact value of last_unlink_trans
3689          * lost, and set it to last_trans to avoid metadata inconsistencies
3690          * between the inode and its parent if the inode is fsync'ed and the log
3691          * replayed. For example, in the scenario:
3692          *
3693          * touch mydir/foo
3694          * ln mydir/foo mydir/bar
3695          * sync
3696          * unlink mydir/bar
3697          * echo 2 > /proc/sys/vm/drop_caches   # evicts inode
3698          * xfs_io -c fsync mydir/foo
3699          * <power failure>
3700          * mount fs, triggers fsync log replay
3701          *
3702          * We must make sure that when we fsync our inode foo we also log its
3703          * parent inode, otherwise after log replay the parent still has the
3704          * dentry with the "bar" name but our inode foo has a link count of 1
3705          * and doesn't have an inode ref with the name "bar" anymore.
3706          *
3707          * Setting last_unlink_trans to last_trans is a pessimistic approach,
3708          * but it guarantees correctness at the expense of occasional full
3709          * transaction commits on fsync if our inode is a directory, or if our
3710          * inode is not a directory, logging its parent unnecessarily.
3711          */
3712         BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3713
3714         path->slots[0]++;
3715         if (inode->i_nlink != 1 ||
3716             path->slots[0] >= btrfs_header_nritems(leaf))
3717                 goto cache_acl;
3718
3719         btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3720         if (location.objectid != btrfs_ino(inode))
3721                 goto cache_acl;
3722
3723         ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3724         if (location.type == BTRFS_INODE_REF_KEY) {
3725                 struct btrfs_inode_ref *ref;
3726
3727                 ref = (struct btrfs_inode_ref *)ptr;
3728                 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3729         } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3730                 struct btrfs_inode_extref *extref;
3731
3732                 extref = (struct btrfs_inode_extref *)ptr;
3733                 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3734                                                                      extref);
3735         }
3736 cache_acl:
3737         /*
3738          * try to precache a NULL acl entry for files that don't have
3739          * any xattrs or acls
3740          */
3741         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3742                                            btrfs_ino(inode), &first_xattr_slot);
3743         if (first_xattr_slot != -1) {
3744                 path->slots[0] = first_xattr_slot;
3745                 ret = btrfs_load_inode_props(inode, path);
3746                 if (ret)
3747                         btrfs_err(root->fs_info,
3748                                   "error loading props for ino %llu (root %llu): %d",
3749                                   btrfs_ino(inode),
3750                                   root->root_key.objectid, ret);
3751         }
3752         btrfs_free_path(path);
3753
3754         if (!maybe_acls)
3755                 cache_no_acl(inode);
3756
3757         switch (inode->i_mode & S_IFMT) {
3758         case S_IFREG:
3759                 inode->i_mapping->a_ops = &btrfs_aops;
3760                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3761                 inode->i_fop = &btrfs_file_operations;
3762                 inode->i_op = &btrfs_file_inode_operations;
3763                 break;
3764         case S_IFDIR:
3765                 inode->i_fop = &btrfs_dir_file_operations;
3766                 if (root == root->fs_info->tree_root)
3767                         inode->i_op = &btrfs_dir_ro_inode_operations;
3768                 else
3769                         inode->i_op = &btrfs_dir_inode_operations;
3770                 break;
3771         case S_IFLNK:
3772                 inode->i_op = &btrfs_symlink_inode_operations;
3773                 inode_nohighmem(inode);
3774                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3775                 break;
3776         default:
3777                 inode->i_op = &btrfs_special_inode_operations;
3778                 init_special_inode(inode, inode->i_mode, rdev);
3779                 break;
3780         }
3781
3782         btrfs_update_iflags(inode);
3783         return;
3784
3785 make_bad:
3786         btrfs_free_path(path);
3787         make_bad_inode(inode);
3788 }
3789
3790 /*
3791  * given a leaf and an inode, copy the inode fields into the leaf
3792  */
3793 static void fill_inode_item(struct btrfs_trans_handle *trans,
3794                             struct extent_buffer *leaf,
3795                             struct btrfs_inode_item *item,
3796                             struct inode *inode)
3797 {
3798         struct btrfs_map_token token;
3799
3800         btrfs_init_map_token(&token);
3801
3802         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3803         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3804         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3805                                    &token);
3806         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3807         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3808
3809         btrfs_set_token_timespec_sec(leaf, &item->atime,
3810                                      inode->i_atime.tv_sec, &token);
3811         btrfs_set_token_timespec_nsec(leaf, &item->atime,
3812                                       inode->i_atime.tv_nsec, &token);
3813
3814         btrfs_set_token_timespec_sec(leaf, &item->mtime,
3815                                      inode->i_mtime.tv_sec, &token);
3816         btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3817                                       inode->i_mtime.tv_nsec, &token);
3818
3819         btrfs_set_token_timespec_sec(leaf, &item->ctime,
3820                                      inode->i_ctime.tv_sec, &token);
3821         btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3822                                       inode->i_ctime.tv_nsec, &token);
3823
3824         btrfs_set_token_timespec_sec(leaf, &item->otime,
3825                                      BTRFS_I(inode)->i_otime.tv_sec, &token);
3826         btrfs_set_token_timespec_nsec(leaf, &item->otime,
3827                                       BTRFS_I(inode)->i_otime.tv_nsec, &token);
3828
3829         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3830                                      &token);
3831         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3832                                          &token);
3833         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3834         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3835         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3836         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3837         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3838 }
3839
3840 /*
3841  * copy everything in the in-memory inode into the btree.
3842  */
3843 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3844                                 struct btrfs_root *root, struct inode *inode)
3845 {
3846         struct btrfs_inode_item *inode_item;
3847         struct btrfs_path *path;
3848         struct extent_buffer *leaf;
3849         int ret;
3850
3851         path = btrfs_alloc_path();
3852         if (!path)
3853                 return -ENOMEM;
3854
3855         path->leave_spinning = 1;
3856         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3857                                  1);
3858         if (ret) {
3859                 if (ret > 0)
3860                         ret = -ENOENT;
3861                 goto failed;
3862         }
3863
3864         leaf = path->nodes[0];
3865         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3866                                     struct btrfs_inode_item);
3867
3868         fill_inode_item(trans, leaf, inode_item, inode);
3869         btrfs_mark_buffer_dirty(leaf);
3870         btrfs_set_inode_last_trans(trans, inode);
3871         ret = 0;
3872 failed:
3873         btrfs_free_path(path);
3874         return ret;
3875 }
3876
3877 /*
3878  * copy everything in the in-memory inode into the btree.
3879  */
3880 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3881                                 struct btrfs_root *root, struct inode *inode)
3882 {
3883         int ret;
3884
3885         /*
3886          * If the inode is a free space inode, we can deadlock during commit
3887          * if we put it into the delayed code.
3888          *
3889          * The data relocation inode should also be directly updated
3890          * without delay
3891          */
3892         if (!btrfs_is_free_space_inode(inode)
3893             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3894             && !root->fs_info->log_root_recovering) {
3895                 btrfs_update_root_times(trans, root);
3896
3897                 ret = btrfs_delayed_update_inode(trans, root, inode);
3898                 if (!ret)
3899                         btrfs_set_inode_last_trans(trans, inode);
3900                 return ret;
3901         }
3902
3903         return btrfs_update_inode_item(trans, root, inode);
3904 }
3905
3906 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3907                                          struct btrfs_root *root,
3908                                          struct inode *inode)
3909 {
3910         int ret;
3911
3912         ret = btrfs_update_inode(trans, root, inode);
3913         if (ret == -ENOSPC)
3914                 return btrfs_update_inode_item(trans, root, inode);
3915         return ret;
3916 }
3917
3918 /*
3919  * unlink helper that gets used here in inode.c and in the tree logging
3920  * recovery code.  It remove a link in a directory with a given name, and
3921  * also drops the back refs in the inode to the directory
3922  */
3923 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3924                                 struct btrfs_root *root,
3925                                 struct inode *dir, struct inode *inode,
3926                                 const char *name, int name_len)
3927 {
3928         struct btrfs_path *path;
3929         int ret = 0;
3930         struct extent_buffer *leaf;
3931         struct btrfs_dir_item *di;
3932         struct btrfs_key key;
3933         u64 index;
3934         u64 ino = btrfs_ino(inode);
3935         u64 dir_ino = btrfs_ino(dir);
3936
3937         path = btrfs_alloc_path();
3938         if (!path) {
3939                 ret = -ENOMEM;
3940                 goto out;
3941         }
3942
3943         path->leave_spinning = 1;
3944         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3945                                     name, name_len, -1);
3946         if (IS_ERR(di)) {
3947                 ret = PTR_ERR(di);
3948                 goto err;
3949         }
3950         if (!di) {
3951                 ret = -ENOENT;
3952                 goto err;
3953         }
3954         leaf = path->nodes[0];
3955         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3956         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3957         if (ret)
3958                 goto err;
3959         btrfs_release_path(path);
3960
3961         /*
3962          * If we don't have dir index, we have to get it by looking up
3963          * the inode ref, since we get the inode ref, remove it directly,
3964          * it is unnecessary to do delayed deletion.
3965          *
3966          * But if we have dir index, needn't search inode ref to get it.
3967          * Since the inode ref is close to the inode item, it is better
3968          * that we delay to delete it, and just do this deletion when
3969          * we update the inode item.
3970          */
3971         if (BTRFS_I(inode)->dir_index) {
3972                 ret = btrfs_delayed_delete_inode_ref(inode);
3973                 if (!ret) {
3974                         index = BTRFS_I(inode)->dir_index;
3975                         goto skip_backref;
3976                 }
3977         }
3978
3979         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3980                                   dir_ino, &index);
3981         if (ret) {
3982                 btrfs_info(root->fs_info,
3983                         "failed to delete reference to %.*s, inode %llu parent %llu",
3984                         name_len, name, ino, dir_ino);
3985                 btrfs_abort_transaction(trans, root, ret);
3986                 goto err;
3987         }
3988 skip_backref:
3989         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3990         if (ret) {
3991                 btrfs_abort_transaction(trans, root, ret);
3992                 goto err;
3993         }
3994
3995         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3996                                          inode, dir_ino);
3997         if (ret != 0 && ret != -ENOENT) {
3998                 btrfs_abort_transaction(trans, root, ret);
3999                 goto err;
4000         }
4001
4002         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
4003                                            dir, index);
4004         if (ret == -ENOENT)
4005                 ret = 0;
4006         else if (ret)
4007                 btrfs_abort_transaction(trans, root, ret);
4008 err:
4009         btrfs_free_path(path);
4010         if (ret)
4011                 goto out;
4012
4013         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4014         inode_inc_iversion(inode);
4015         inode_inc_iversion(dir);
4016         inode->i_ctime = dir->i_mtime =
4017                 dir->i_ctime = current_fs_time(inode->i_sb);
4018         ret = btrfs_update_inode(trans, root, dir);
4019 out:
4020         return ret;
4021 }
4022
4023 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4024                        struct btrfs_root *root,
4025                        struct inode *dir, struct inode *inode,
4026                        const char *name, int name_len)
4027 {
4028         int ret;
4029         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4030         if (!ret) {
4031                 drop_nlink(inode);
4032                 ret = btrfs_update_inode(trans, root, inode);
4033         }
4034         return ret;
4035 }
4036
4037 /*
4038  * helper to start transaction for unlink and rmdir.
4039  *
4040  * unlink and rmdir are special in btrfs, they do not always free space, so
4041  * if we cannot make our reservations the normal way try and see if there is
4042  * plenty of slack room in the global reserve to migrate, otherwise we cannot
4043  * allow the unlink to occur.
4044  */
4045 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4046 {
4047         struct btrfs_root *root = BTRFS_I(dir)->root;
4048
4049         /*
4050          * 1 for the possible orphan item
4051          * 1 for the dir item
4052          * 1 for the dir index
4053          * 1 for the inode ref
4054          * 1 for the inode
4055          */
4056         return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4057 }
4058
4059 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4060 {
4061         struct btrfs_root *root = BTRFS_I(dir)->root;
4062         struct btrfs_trans_handle *trans;
4063         struct inode *inode = d_inode(dentry);
4064         int ret;
4065
4066         trans = __unlink_start_trans(dir);
4067         if (IS_ERR(trans))
4068                 return PTR_ERR(trans);
4069
4070         btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
4071
4072         ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4073                                  dentry->d_name.name, dentry->d_name.len);
4074         if (ret)
4075                 goto out;
4076
4077         if (inode->i_nlink == 0) {
4078                 ret = btrfs_orphan_add(trans, inode);
4079                 if (ret)
4080                         goto out;
4081         }
4082
4083 out:
4084         btrfs_end_transaction(trans, root);
4085         btrfs_btree_balance_dirty(root);
4086         return ret;
4087 }
4088
4089 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4090                         struct btrfs_root *root,
4091                         struct inode *dir, u64 objectid,
4092                         const char *name, int name_len)
4093 {
4094         struct btrfs_path *path;
4095         struct extent_buffer *leaf;
4096         struct btrfs_dir_item *di;
4097         struct btrfs_key key;
4098         u64 index;
4099         int ret;
4100         u64 dir_ino = btrfs_ino(dir);
4101
4102         path = btrfs_alloc_path();
4103         if (!path)
4104                 return -ENOMEM;
4105
4106         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4107                                    name, name_len, -1);
4108         if (IS_ERR_OR_NULL(di)) {
4109                 if (!di)
4110                         ret = -ENOENT;
4111                 else
4112                         ret = PTR_ERR(di);
4113                 goto out;
4114         }
4115
4116         leaf = path->nodes[0];
4117         btrfs_dir_item_key_to_cpu(leaf, di, &key);
4118         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4119         ret = btrfs_delete_one_dir_name(trans, root, path, di);
4120         if (ret) {
4121                 btrfs_abort_transaction(trans, root, ret);
4122                 goto out;
4123         }
4124         btrfs_release_path(path);
4125
4126         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4127                                  objectid, root->root_key.objectid,
4128                                  dir_ino, &index, name, name_len);
4129         if (ret < 0) {
4130                 if (ret != -ENOENT) {
4131                         btrfs_abort_transaction(trans, root, ret);
4132                         goto out;
4133                 }
4134                 di = btrfs_search_dir_index_item(root, path, dir_ino,
4135                                                  name, name_len);
4136                 if (IS_ERR_OR_NULL(di)) {
4137                         if (!di)
4138                                 ret = -ENOENT;
4139                         else
4140                                 ret = PTR_ERR(di);
4141                         btrfs_abort_transaction(trans, root, ret);
4142                         goto out;
4143                 }
4144
4145                 leaf = path->nodes[0];
4146                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4147                 btrfs_release_path(path);
4148                 index = key.offset;
4149         }
4150         btrfs_release_path(path);
4151
4152         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4153         if (ret) {
4154                 btrfs_abort_transaction(trans, root, ret);
4155                 goto out;
4156         }
4157
4158         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4159         inode_inc_iversion(dir);
4160         dir->i_mtime = dir->i_ctime = current_fs_time(dir->i_sb);
4161         ret = btrfs_update_inode_fallback(trans, root, dir);
4162         if (ret)
4163                 btrfs_abort_transaction(trans, root, ret);
4164 out:
4165         btrfs_free_path(path);
4166         return ret;
4167 }
4168
4169 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4170 {
4171         struct inode *inode = d_inode(dentry);
4172         int err = 0;
4173         struct btrfs_root *root = BTRFS_I(dir)->root;
4174         struct btrfs_trans_handle *trans;
4175
4176         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4177                 return -ENOTEMPTY;
4178         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4179                 return -EPERM;
4180
4181         trans = __unlink_start_trans(dir);
4182         if (IS_ERR(trans))
4183                 return PTR_ERR(trans);
4184
4185         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4186                 err = btrfs_unlink_subvol(trans, root, dir,
4187                                           BTRFS_I(inode)->location.objectid,
4188                                           dentry->d_name.name,
4189                                           dentry->d_name.len);
4190                 goto out;
4191         }
4192
4193         err = btrfs_orphan_add(trans, inode);
4194         if (err)
4195                 goto out;
4196
4197         /* now the directory is empty */
4198         err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4199                                  dentry->d_name.name, dentry->d_name.len);
4200         if (!err)
4201                 btrfs_i_size_write(inode, 0);
4202 out:
4203         btrfs_end_transaction(trans, root);
4204         btrfs_btree_balance_dirty(root);
4205
4206         return err;
4207 }
4208
4209 static int truncate_space_check(struct btrfs_trans_handle *trans,
4210                                 struct btrfs_root *root,
4211                                 u64 bytes_deleted)
4212 {
4213         int ret;
4214
4215         /*
4216          * This is only used to apply pressure to the enospc system, we don't
4217          * intend to use this reservation at all.
4218          */
4219         bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4220         bytes_deleted *= root->nodesize;
4221         ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4222                                   bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4223         if (!ret) {
4224                 trace_btrfs_space_reservation(root->fs_info, "transaction",
4225                                               trans->transid,
4226                                               bytes_deleted, 1);
4227                 trans->bytes_reserved += bytes_deleted;
4228         }
4229         return ret;
4230
4231 }
4232
4233 static int truncate_inline_extent(struct inode *inode,
4234                                   struct btrfs_path *path,
4235                                   struct btrfs_key *found_key,
4236                                   const u64 item_end,
4237                                   const u64 new_size)
4238 {
4239         struct extent_buffer *leaf = path->nodes[0];
4240         int slot = path->slots[0];
4241         struct btrfs_file_extent_item *fi;
4242         u32 size = (u32)(new_size - found_key->offset);
4243         struct btrfs_root *root = BTRFS_I(inode)->root;
4244
4245         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4246
4247         if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4248                 loff_t offset = new_size;
4249                 loff_t page_end = ALIGN(offset, PAGE_SIZE);
4250
4251                 /*
4252                  * Zero out the remaining of the last page of our inline extent,
4253                  * instead of directly truncating our inline extent here - that
4254                  * would be much more complex (decompressing all the data, then
4255                  * compressing the truncated data, which might be bigger than
4256                  * the size of the inline extent, resize the extent, etc).
4257                  * We release the path because to get the page we might need to
4258                  * read the extent item from disk (data not in the page cache).
4259                  */
4260                 btrfs_release_path(path);
4261                 return btrfs_truncate_block(inode, offset, page_end - offset,
4262                                         0);
4263         }
4264
4265         btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4266         size = btrfs_file_extent_calc_inline_size(size);
4267         btrfs_truncate_item(root, path, size, 1);
4268
4269         if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4270                 inode_sub_bytes(inode, item_end + 1 - new_size);
4271
4272         return 0;
4273 }
4274
4275 /*
4276  * this can truncate away extent items, csum items and directory items.
4277  * It starts at a high offset and removes keys until it can't find
4278  * any higher than new_size
4279  *
4280  * csum items that cross the new i_size are truncated to the new size
4281  * as well.
4282  *
4283  * min_type is the minimum key type to truncate down to.  If set to 0, this
4284  * will kill all the items on this inode, including the INODE_ITEM_KEY.
4285  */
4286 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4287                                struct btrfs_root *root,
4288                                struct inode *inode,
4289                                u64 new_size, u32 min_type)
4290 {
4291         struct btrfs_path *path;
4292         struct extent_buffer *leaf;
4293         struct btrfs_file_extent_item *fi;
4294         struct btrfs_key key;
4295         struct btrfs_key found_key;
4296         u64 extent_start = 0;
4297         u64 extent_num_bytes = 0;
4298         u64 extent_offset = 0;
4299         u64 item_end = 0;
4300         u64 last_size = new_size;
4301         u32 found_type = (u8)-1;
4302         int found_extent;
4303         int del_item;
4304         int pending_del_nr = 0;
4305         int pending_del_slot = 0;
4306         int extent_type = -1;
4307         int ret;
4308         int err = 0;
4309         u64 ino = btrfs_ino(inode);
4310         u64 bytes_deleted = 0;
4311         bool be_nice = 0;
4312         bool should_throttle = 0;
4313         bool should_end = 0;
4314
4315         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4316
4317         /*
4318          * for non-free space inodes and ref cows, we want to back off from
4319          * time to time
4320          */
4321         if (!btrfs_is_free_space_inode(inode) &&
4322             test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4323                 be_nice = 1;
4324
4325         path = btrfs_alloc_path();
4326         if (!path)
4327                 return -ENOMEM;
4328         path->reada = READA_BACK;
4329
4330         /*
4331          * We want to drop from the next block forward in case this new size is
4332          * not block aligned since we will be keeping the last block of the
4333          * extent just the way it is.
4334          */
4335         if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4336             root == root->fs_info->tree_root)
4337                 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4338                                         root->sectorsize), (u64)-1, 0);
4339
4340         /*
4341          * This function is also used to drop the items in the log tree before
4342          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4343          * it is used to drop the loged items. So we shouldn't kill the delayed
4344          * items.
4345          */
4346         if (min_type == 0 && root == BTRFS_I(inode)->root)
4347                 btrfs_kill_delayed_inode_items(inode);
4348
4349         key.objectid = ino;
4350         key.offset = (u64)-1;
4351         key.type = (u8)-1;
4352
4353 search_again:
4354         /*
4355          * with a 16K leaf size and 128MB extents, you can actually queue
4356          * up a huge file in a single leaf.  Most of the time that
4357          * bytes_deleted is > 0, it will be huge by the time we get here
4358          */
4359         if (be_nice && bytes_deleted > SZ_32M) {
4360                 if (btrfs_should_end_transaction(trans, root)) {
4361                         err = -EAGAIN;
4362                         goto error;
4363                 }
4364         }
4365
4366
4367         path->leave_spinning = 1;
4368         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4369         if (ret < 0) {
4370                 err = ret;
4371                 goto out;
4372         }
4373
4374         if (ret > 0) {
4375                 /* there are no items in the tree for us to truncate, we're
4376                  * done
4377                  */
4378                 if (path->slots[0] == 0)
4379                         goto out;
4380                 path->slots[0]--;
4381         }
4382
4383         while (1) {
4384                 fi = NULL;
4385                 leaf = path->nodes[0];
4386                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4387                 found_type = found_key.type;
4388
4389                 if (found_key.objectid != ino)
4390                         break;
4391
4392                 if (found_type < min_type)
4393                         break;
4394
4395                 item_end = found_key.offset;
4396                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4397                         fi = btrfs_item_ptr(leaf, path->slots[0],
4398                                             struct btrfs_file_extent_item);
4399                         extent_type = btrfs_file_extent_type(leaf, fi);
4400                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4401                                 item_end +=
4402                                     btrfs_file_extent_num_bytes(leaf, fi);
4403                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4404                                 item_end += btrfs_file_extent_inline_len(leaf,
4405                                                          path->slots[0], fi);
4406                         }
4407                         item_end--;
4408                 }
4409                 if (found_type > min_type) {
4410                         del_item = 1;
4411                 } else {
4412                         if (item_end < new_size)
4413                                 break;
4414                         if (found_key.offset >= new_size)
4415                                 del_item = 1;
4416                         else
4417                                 del_item = 0;
4418                 }
4419                 found_extent = 0;
4420                 /* FIXME, shrink the extent if the ref count is only 1 */
4421                 if (found_type != BTRFS_EXTENT_DATA_KEY)
4422                         goto delete;
4423
4424                 if (del_item)
4425                         last_size = found_key.offset;
4426                 else
4427                         last_size = new_size;
4428
4429                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4430                         u64 num_dec;
4431                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4432                         if (!del_item) {
4433                                 u64 orig_num_bytes =
4434                                         btrfs_file_extent_num_bytes(leaf, fi);
4435                                 extent_num_bytes = ALIGN(new_size -
4436                                                 found_key.offset,
4437                                                 root->sectorsize);
4438                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4439                                                          extent_num_bytes);
4440                                 num_dec = (orig_num_bytes -
4441                                            extent_num_bytes);
4442                                 if (test_bit(BTRFS_ROOT_REF_COWS,
4443                                              &root->state) &&
4444                                     extent_start != 0)
4445                                         inode_sub_bytes(inode, num_dec);
4446                                 btrfs_mark_buffer_dirty(leaf);
4447                         } else {
4448                                 extent_num_bytes =
4449                                         btrfs_file_extent_disk_num_bytes(leaf,
4450                                                                          fi);
4451                                 extent_offset = found_key.offset -
4452                                         btrfs_file_extent_offset(leaf, fi);
4453
4454                                 /* FIXME blocksize != 4096 */
4455                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4456                                 if (extent_start != 0) {
4457                                         found_extent = 1;
4458                                         if (test_bit(BTRFS_ROOT_REF_COWS,
4459                                                      &root->state))
4460                                                 inode_sub_bytes(inode, num_dec);
4461                                 }
4462                         }
4463                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4464                         /*
4465                          * we can't truncate inline items that have had
4466                          * special encodings
4467                          */
4468                         if (!del_item &&
4469                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
4470                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4471
4472                                 /*
4473                                  * Need to release path in order to truncate a
4474                                  * compressed extent. So delete any accumulated
4475                                  * extent items so far.
4476                                  */
4477                                 if (btrfs_file_extent_compression(leaf, fi) !=
4478                                     BTRFS_COMPRESS_NONE && pending_del_nr) {
4479                                         err = btrfs_del_items(trans, root, path,
4480                                                               pending_del_slot,
4481                                                               pending_del_nr);
4482                                         if (err) {
4483                                                 btrfs_abort_transaction(trans,
4484                                                                         root,
4485                                                                         err);
4486                                                 goto error;
4487                                         }
4488                                         pending_del_nr = 0;
4489                                 }
4490
4491                                 err = truncate_inline_extent(inode, path,
4492                                                              &found_key,
4493                                                              item_end,
4494                                                              new_size);
4495                                 if (err) {
4496                                         btrfs_abort_transaction(trans,
4497                                                                 root, err);
4498                                         goto error;
4499                                 }
4500                         } else if (test_bit(BTRFS_ROOT_REF_COWS,
4501                                             &root->state)) {
4502                                 inode_sub_bytes(inode, item_end + 1 - new_size);
4503                         }
4504                 }
4505 delete:
4506                 if (del_item) {
4507                         if (!pending_del_nr) {
4508                                 /* no pending yet, add ourselves */
4509                                 pending_del_slot = path->slots[0];
4510                                 pending_del_nr = 1;
4511                         } else if (pending_del_nr &&
4512                                    path->slots[0] + 1 == pending_del_slot) {
4513                                 /* hop on the pending chunk */
4514                                 pending_del_nr++;
4515                                 pending_del_slot = path->slots[0];
4516                         } else {
4517                                 BUG();
4518                         }
4519                 } else {
4520                         break;
4521                 }
4522                 should_throttle = 0;
4523
4524                 if (found_extent &&
4525                     (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4526                      root == root->fs_info->tree_root)) {
4527                         btrfs_set_path_blocking(path);
4528                         bytes_deleted += extent_num_bytes;
4529                         ret = btrfs_free_extent(trans, root, extent_start,
4530                                                 extent_num_bytes, 0,
4531                                                 btrfs_header_owner(leaf),
4532                                                 ino, extent_offset);
4533                         BUG_ON(ret);
4534                         if (btrfs_should_throttle_delayed_refs(trans, root))
4535                                 btrfs_async_run_delayed_refs(root,
4536                                         trans->delayed_ref_updates * 2, 0);
4537                         if (be_nice) {
4538                                 if (truncate_space_check(trans, root,
4539                                                          extent_num_bytes)) {
4540                                         should_end = 1;
4541                                 }
4542                                 if (btrfs_should_throttle_delayed_refs(trans,
4543                                                                        root)) {
4544                                         should_throttle = 1;
4545                                 }
4546                         }
4547                 }
4548
4549                 if (found_type == BTRFS_INODE_ITEM_KEY)
4550                         break;
4551
4552                 if (path->slots[0] == 0 ||
4553                     path->slots[0] != pending_del_slot ||
4554                     should_throttle || should_end) {
4555                         if (pending_del_nr) {
4556                                 ret = btrfs_del_items(trans, root, path,
4557                                                 pending_del_slot,
4558                                                 pending_del_nr);
4559                                 if (ret) {
4560                                         btrfs_abort_transaction(trans,
4561                                                                 root, ret);
4562                                         goto error;
4563                                 }
4564                                 pending_del_nr = 0;
4565                         }
4566                         btrfs_release_path(path);
4567                         if (should_throttle) {
4568                                 unsigned long updates = trans->delayed_ref_updates;
4569                                 if (updates) {
4570                                         trans->delayed_ref_updates = 0;
4571                                         ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4572                                         if (ret && !err)
4573                                                 err = ret;
4574                                 }
4575                         }
4576                         /*
4577                          * if we failed to refill our space rsv, bail out
4578                          * and let the transaction restart
4579                          */
4580                         if (should_end) {
4581                                 err = -EAGAIN;
4582                                 goto error;
4583                         }
4584                         goto search_again;
4585                 } else {
4586                         path->slots[0]--;
4587                 }
4588         }
4589 out:
4590         if (pending_del_nr) {
4591                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4592                                       pending_del_nr);
4593                 if (ret)
4594                         btrfs_abort_transaction(trans, root, ret);
4595         }
4596 error:
4597         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4598                 btrfs_ordered_update_i_size(inode, last_size, NULL);
4599
4600         btrfs_free_path(path);
4601
4602         if (be_nice && bytes_deleted > SZ_32M) {
4603                 unsigned long updates = trans->delayed_ref_updates;
4604                 if (updates) {
4605                         trans->delayed_ref_updates = 0;
4606                         ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4607                         if (ret && !err)
4608                                 err = ret;
4609                 }
4610         }
4611         return err;
4612 }
4613
4614 /*
4615  * btrfs_truncate_block - read, zero a chunk and write a block
4616  * @inode - inode that we're zeroing
4617  * @from - the offset to start zeroing
4618  * @len - the length to zero, 0 to zero the entire range respective to the
4619  *      offset
4620  * @front - zero up to the offset instead of from the offset on
4621  *
4622  * This will find the block for the "from" offset and cow the block and zero the
4623  * part we want to zero.  This is used with truncate and hole punching.
4624  */
4625 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4626                         int front)
4627 {
4628         struct address_space *mapping = inode->i_mapping;
4629         struct btrfs_root *root = BTRFS_I(inode)->root;
4630         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4631         struct btrfs_ordered_extent *ordered;
4632         struct extent_state *cached_state = NULL;
4633         char *kaddr;
4634         u32 blocksize = root->sectorsize;
4635         pgoff_t index = from >> PAGE_SHIFT;
4636         unsigned offset = from & (blocksize - 1);
4637         struct page *page;
4638         gfp_t mask = btrfs_alloc_write_mask(mapping);
4639         int ret = 0;
4640         u64 block_start;
4641         u64 block_end;
4642
4643         if ((offset & (blocksize - 1)) == 0 &&
4644             (!len || ((len & (blocksize - 1)) == 0)))
4645                 goto out;
4646
4647         ret = btrfs_delalloc_reserve_space(inode,
4648                         round_down(from, blocksize), blocksize);
4649         if (ret)
4650                 goto out;
4651
4652 again:
4653         page = find_or_create_page(mapping, index, mask);
4654         if (!page) {
4655                 btrfs_delalloc_release_space(inode,
4656                                 round_down(from, blocksize),
4657                                 blocksize);
4658                 ret = -ENOMEM;
4659                 goto out;
4660         }
4661
4662         block_start = round_down(from, blocksize);
4663         block_end = block_start + blocksize - 1;
4664
4665         if (!PageUptodate(page)) {
4666                 ret = btrfs_readpage(NULL, page);
4667                 lock_page(page);
4668                 if (page->mapping != mapping) {
4669                         unlock_page(page);
4670                         put_page(page);
4671                         goto again;
4672                 }
4673                 if (!PageUptodate(page)) {
4674                         ret = -EIO;
4675                         goto out_unlock;
4676                 }
4677         }
4678         wait_on_page_writeback(page);
4679
4680         lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4681         set_page_extent_mapped(page);
4682
4683         ordered = btrfs_lookup_ordered_extent(inode, block_start);
4684         if (ordered) {
4685                 unlock_extent_cached(io_tree, block_start, block_end,
4686                                      &cached_state, GFP_NOFS);
4687                 unlock_page(page);
4688                 put_page(page);
4689                 btrfs_start_ordered_extent(inode, ordered, 1);
4690                 btrfs_put_ordered_extent(ordered);
4691                 goto again;
4692         }
4693
4694         clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4695                           EXTENT_DIRTY | EXTENT_DELALLOC |
4696                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4697                           0, 0, &cached_state, GFP_NOFS);
4698
4699         ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
4700                                         &cached_state);
4701         if (ret) {
4702                 unlock_extent_cached(io_tree, block_start, block_end,
4703                                      &cached_state, GFP_NOFS);
4704                 goto out_unlock;
4705         }
4706
4707         if (offset != blocksize) {
4708                 if (!len)
4709                         len = blocksize - offset;
4710                 kaddr = kmap(page);
4711                 if (front)
4712                         memset(kaddr + (block_start - page_offset(page)),
4713                                 0, offset);
4714                 else
4715                         memset(kaddr + (block_start - page_offset(page)) +  offset,
4716                                 0, len);
4717                 flush_dcache_page(page);
4718                 kunmap(page);
4719         }
4720         ClearPageChecked(page);
4721         set_page_dirty(page);
4722         unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
4723                              GFP_NOFS);
4724
4725 out_unlock:
4726         if (ret)
4727                 btrfs_delalloc_release_space(inode, block_start,
4728                                              blocksize);
4729         unlock_page(page);
4730         put_page(page);
4731 out:
4732         return ret;
4733 }
4734
4735 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4736                              u64 offset, u64 len)
4737 {
4738         struct btrfs_trans_handle *trans;
4739         int ret;
4740
4741         /*
4742          * Still need to make sure the inode looks like it's been updated so
4743          * that any holes get logged if we fsync.
4744          */
4745         if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4746                 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4747                 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4748                 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4749                 return 0;
4750         }
4751
4752         /*
4753          * 1 - for the one we're dropping
4754          * 1 - for the one we're adding
4755          * 1 - for updating the inode.
4756          */
4757         trans = btrfs_start_transaction(root, 3);
4758         if (IS_ERR(trans))
4759                 return PTR_ERR(trans);
4760
4761         ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4762         if (ret) {
4763                 btrfs_abort_transaction(trans, root, ret);
4764                 btrfs_end_transaction(trans, root);
4765                 return ret;
4766         }
4767
4768         ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4769                                        0, 0, len, 0, len, 0, 0, 0);
4770         if (ret)
4771                 btrfs_abort_transaction(trans, root, ret);
4772         else
4773                 btrfs_update_inode(trans, root, inode);
4774         btrfs_end_transaction(trans, root);
4775         return ret;
4776 }
4777
4778 /*
4779  * This function puts in dummy file extents for the area we're creating a hole
4780  * for.  So if we are truncating this file to a larger size we need to insert
4781  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4782  * the range between oldsize and size
4783  */
4784 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4785 {
4786         struct btrfs_root *root = BTRFS_I(inode)->root;
4787         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4788         struct extent_map *em = NULL;
4789         struct extent_state *cached_state = NULL;
4790         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4791         u64 hole_start = ALIGN(oldsize, root->sectorsize);
4792         u64 block_end = ALIGN(size, root->sectorsize);
4793         u64 last_byte;
4794         u64 cur_offset;
4795         u64 hole_size;
4796         int err = 0;
4797
4798         /*
4799          * If our size started in the middle of a block we need to zero out the
4800          * rest of the block before we expand the i_size, otherwise we could
4801          * expose stale data.
4802          */
4803         err = btrfs_truncate_block(inode, oldsize, 0, 0);
4804         if (err)
4805                 return err;
4806
4807         if (size <= hole_start)
4808                 return 0;
4809
4810         while (1) {
4811                 struct btrfs_ordered_extent *ordered;
4812
4813                 lock_extent_bits(io_tree, hole_start, block_end - 1,
4814                                  &cached_state);
4815                 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4816                                                      block_end - hole_start);
4817                 if (!ordered)
4818                         break;
4819                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4820                                      &cached_state, GFP_NOFS);
4821                 btrfs_start_ordered_extent(inode, ordered, 1);
4822                 btrfs_put_ordered_extent(ordered);
4823         }
4824
4825         cur_offset = hole_start;
4826         while (1) {
4827                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4828                                 block_end - cur_offset, 0);
4829                 if (IS_ERR(em)) {
4830                         err = PTR_ERR(em);
4831                         em = NULL;
4832                         break;
4833                 }
4834                 last_byte = min(extent_map_end(em), block_end);
4835                 last_byte = ALIGN(last_byte , root->sectorsize);
4836                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4837                         struct extent_map *hole_em;
4838                         hole_size = last_byte - cur_offset;
4839
4840                         err = maybe_insert_hole(root, inode, cur_offset,
4841                                                 hole_size);
4842                         if (err)
4843                                 break;
4844                         btrfs_drop_extent_cache(inode, cur_offset,
4845                                                 cur_offset + hole_size - 1, 0);
4846                         hole_em = alloc_extent_map();
4847                         if (!hole_em) {
4848                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4849                                         &BTRFS_I(inode)->runtime_flags);
4850                                 goto next;
4851                         }
4852                         hole_em->start = cur_offset;
4853                         hole_em->len = hole_size;
4854                         hole_em->orig_start = cur_offset;
4855
4856                         hole_em->block_start = EXTENT_MAP_HOLE;
4857                         hole_em->block_len = 0;
4858                         hole_em->orig_block_len = 0;
4859                         hole_em->ram_bytes = hole_size;
4860                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4861                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
4862                         hole_em->generation = root->fs_info->generation;
4863
4864                         while (1) {
4865                                 write_lock(&em_tree->lock);
4866                                 err = add_extent_mapping(em_tree, hole_em, 1);
4867                                 write_unlock(&em_tree->lock);
4868                                 if (err != -EEXIST)
4869                                         break;
4870                                 btrfs_drop_extent_cache(inode, cur_offset,
4871                                                         cur_offset +
4872                                                         hole_size - 1, 0);
4873                         }
4874                         free_extent_map(hole_em);
4875                 }
4876 next:
4877                 free_extent_map(em);
4878                 em = NULL;
4879                 cur_offset = last_byte;
4880                 if (cur_offset >= block_end)
4881                         break;
4882         }
4883         free_extent_map(em);
4884         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4885                              GFP_NOFS);
4886         return err;
4887 }
4888
4889 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4890 {
4891         struct btrfs_root *root = BTRFS_I(inode)->root;
4892         struct btrfs_trans_handle *trans;
4893         loff_t oldsize = i_size_read(inode);
4894         loff_t newsize = attr->ia_size;
4895         int mask = attr->ia_valid;
4896         int ret;
4897
4898         /*
4899          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4900          * special case where we need to update the times despite not having
4901          * these flags set.  For all other operations the VFS set these flags
4902          * explicitly if it wants a timestamp update.
4903          */
4904         if (newsize != oldsize) {
4905                 inode_inc_iversion(inode);
4906                 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4907                         inode->i_ctime = inode->i_mtime =
4908                                 current_fs_time(inode->i_sb);
4909         }
4910
4911         if (newsize > oldsize) {
4912                 /*
4913                  * Don't do an expanding truncate while snapshoting is ongoing.
4914                  * This is to ensure the snapshot captures a fully consistent
4915                  * state of this file - if the snapshot captures this expanding
4916                  * truncation, it must capture all writes that happened before
4917                  * this truncation.
4918                  */
4919                 btrfs_wait_for_snapshot_creation(root);
4920                 ret = btrfs_cont_expand(inode, oldsize, newsize);
4921                 if (ret) {
4922                         btrfs_end_write_no_snapshoting(root);
4923                         return ret;
4924                 }
4925
4926                 trans = btrfs_start_transaction(root, 1);
4927                 if (IS_ERR(trans)) {
4928                         btrfs_end_write_no_snapshoting(root);
4929                         return PTR_ERR(trans);
4930                 }
4931
4932                 i_size_write(inode, newsize);
4933                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4934                 pagecache_isize_extended(inode, oldsize, newsize);
4935                 ret = btrfs_update_inode(trans, root, inode);
4936                 btrfs_end_write_no_snapshoting(root);
4937                 btrfs_end_transaction(trans, root);
4938         } else {
4939
4940                 /*
4941                  * We're truncating a file that used to have good data down to
4942                  * zero. Make sure it gets into the ordered flush list so that
4943                  * any new writes get down to disk quickly.
4944                  */
4945                 if (newsize == 0)
4946                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4947                                 &BTRFS_I(inode)->runtime_flags);
4948
4949                 /*
4950                  * 1 for the orphan item we're going to add
4951                  * 1 for the orphan item deletion.
4952                  */
4953                 trans = btrfs_start_transaction(root, 2);
4954                 if (IS_ERR(trans))
4955                         return PTR_ERR(trans);
4956
4957                 /*
4958                  * We need to do this in case we fail at _any_ point during the
4959                  * actual truncate.  Once we do the truncate_setsize we could
4960                  * invalidate pages which forces any outstanding ordered io to
4961                  * be instantly completed which will give us extents that need
4962                  * to be truncated.  If we fail to get an orphan inode down we
4963                  * could have left over extents that were never meant to live,
4964                  * so we need to guarantee from this point on that everything
4965                  * will be consistent.
4966                  */
4967                 ret = btrfs_orphan_add(trans, inode);
4968                 btrfs_end_transaction(trans, root);
4969                 if (ret)
4970                         return ret;
4971
4972                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4973                 truncate_setsize(inode, newsize);
4974
4975                 /* Disable nonlocked read DIO to avoid the end less truncate */
4976                 btrfs_inode_block_unlocked_dio(inode);
4977                 inode_dio_wait(inode);
4978                 btrfs_inode_resume_unlocked_dio(inode);
4979
4980                 ret = btrfs_truncate(inode);
4981                 if (ret && inode->i_nlink) {
4982                         int err;
4983
4984                         /*
4985                          * failed to truncate, disk_i_size is only adjusted down
4986                          * as we remove extents, so it should represent the true
4987                          * size of the inode, so reset the in memory size and
4988                          * delete our orphan entry.
4989                          */
4990                         trans = btrfs_join_transaction(root);
4991                         if (IS_ERR(trans)) {
4992                                 btrfs_orphan_del(NULL, inode);
4993                                 return ret;
4994                         }
4995                         i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4996                         err = btrfs_orphan_del(trans, inode);
4997                         if (err)
4998                                 btrfs_abort_transaction(trans, root, err);
4999                         btrfs_end_transaction(trans, root);
5000                 }
5001         }
5002
5003         return ret;
5004 }
5005
5006 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5007 {
5008         struct inode *inode = d_inode(dentry);
5009         struct btrfs_root *root = BTRFS_I(inode)->root;
5010         int err;
5011
5012         if (btrfs_root_readonly(root))
5013                 return -EROFS;
5014
5015         err = inode_change_ok(inode, attr);
5016         if (err)
5017                 return err;
5018
5019         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5020                 err = btrfs_setsize(inode, attr);
5021                 if (err)
5022                         return err;
5023         }
5024
5025         if (attr->ia_valid) {
5026                 setattr_copy(inode, attr);
5027                 inode_inc_iversion(inode);
5028                 err = btrfs_dirty_inode(inode);
5029
5030                 if (!err && attr->ia_valid & ATTR_MODE)
5031                         err = posix_acl_chmod(inode, inode->i_mode);
5032         }
5033
5034         return err;
5035 }
5036
5037 /*
5038  * While truncating the inode pages during eviction, we get the VFS calling
5039  * btrfs_invalidatepage() against each page of the inode. This is slow because
5040  * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5041  * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5042  * extent_state structures over and over, wasting lots of time.
5043  *
5044  * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5045  * those expensive operations on a per page basis and do only the ordered io
5046  * finishing, while we release here the extent_map and extent_state structures,
5047  * without the excessive merging and splitting.
5048  */
5049 static void evict_inode_truncate_pages(struct inode *inode)
5050 {
5051         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5052         struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5053         struct rb_node *node;
5054
5055         ASSERT(inode->i_state & I_FREEING);
5056         truncate_inode_pages_final(&inode->i_data);
5057
5058         write_lock(&map_tree->lock);
5059         while (!RB_EMPTY_ROOT(&map_tree->map)) {
5060                 struct extent_map *em;
5061
5062                 node = rb_first(&map_tree->map);
5063                 em = rb_entry(node, struct extent_map, rb_node);
5064                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5065                 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5066                 remove_extent_mapping(map_tree, em);
5067                 free_extent_map(em);
5068                 if (need_resched()) {
5069                         write_unlock(&map_tree->lock);
5070                         cond_resched();
5071                         write_lock(&map_tree->lock);
5072                 }
5073         }
5074         write_unlock(&map_tree->lock);
5075
5076         /*
5077          * Keep looping until we have no more ranges in the io tree.
5078          * We can have ongoing bios started by readpages (called from readahead)
5079          * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5080          * still in progress (unlocked the pages in the bio but did not yet
5081          * unlocked the ranges in the io tree). Therefore this means some
5082          * ranges can still be locked and eviction started because before
5083          * submitting those bios, which are executed by a separate task (work
5084          * queue kthread), inode references (inode->i_count) were not taken
5085          * (which would be dropped in the end io callback of each bio).
5086          * Therefore here we effectively end up waiting for those bios and
5087          * anyone else holding locked ranges without having bumped the inode's
5088          * reference count - if we don't do it, when they access the inode's
5089          * io_tree to unlock a range it may be too late, leading to an
5090          * use-after-free issue.
5091          */
5092         spin_lock(&io_tree->lock);
5093         while (!RB_EMPTY_ROOT(&io_tree->state)) {
5094                 struct extent_state *state;
5095                 struct extent_state *cached_state = NULL;
5096                 u64 start;
5097                 u64 end;
5098
5099                 node = rb_first(&io_tree->state);
5100                 state = rb_entry(node, struct extent_state, rb_node);
5101                 start = state->start;
5102                 end = state->end;
5103                 spin_unlock(&io_tree->lock);
5104
5105                 lock_extent_bits(io_tree, start, end, &cached_state);
5106
5107                 /*
5108                  * If still has DELALLOC flag, the extent didn't reach disk,
5109                  * and its reserved space won't be freed by delayed_ref.
5110                  * So we need to free its reserved space here.
5111                  * (Refer to comment in btrfs_invalidatepage, case 2)
5112                  *
5113                  * Note, end is the bytenr of last byte, so we need + 1 here.
5114                  */
5115                 if (state->state & EXTENT_DELALLOC)
5116                         btrfs_qgroup_free_data(inode, start, end - start + 1);
5117
5118                 clear_extent_bit(io_tree, start, end,
5119                                  EXTENT_LOCKED | EXTENT_DIRTY |
5120                                  EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5121                                  EXTENT_DEFRAG, 1, 1,
5122                                  &cached_state, GFP_NOFS);
5123
5124                 cond_resched();
5125                 spin_lock(&io_tree->lock);
5126         }
5127         spin_unlock(&io_tree->lock);
5128 }
5129
5130 void btrfs_evict_inode(struct inode *inode)
5131 {
5132         struct btrfs_trans_handle *trans;
5133         struct btrfs_root *root = BTRFS_I(inode)->root;
5134         struct btrfs_block_rsv *rsv, *global_rsv;
5135         int steal_from_global = 0;
5136         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
5137         int ret;
5138
5139         trace_btrfs_inode_evict(inode);
5140
5141         evict_inode_truncate_pages(inode);
5142
5143         if (inode->i_nlink &&
5144             ((btrfs_root_refs(&root->root_item) != 0 &&
5145               root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5146              btrfs_is_free_space_inode(inode)))
5147                 goto no_delete;
5148
5149         if (is_bad_inode(inode)) {
5150                 btrfs_orphan_del(NULL, inode);
5151                 goto no_delete;
5152         }
5153         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5154         if (!special_file(inode->i_mode))
5155                 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5156
5157         btrfs_free_io_failure_record(inode, 0, (u64)-1);
5158
5159         if (root->fs_info->log_root_recovering) {
5160                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5161                                  &BTRFS_I(inode)->runtime_flags));
5162                 goto no_delete;
5163         }
5164
5165         if (inode->i_nlink > 0) {
5166                 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5167                        root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5168                 goto no_delete;
5169         }
5170
5171         ret = btrfs_commit_inode_delayed_inode(inode);
5172         if (ret) {
5173                 btrfs_orphan_del(NULL, inode);
5174                 goto no_delete;
5175         }
5176
5177         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
5178         if (!rsv) {
5179                 btrfs_orphan_del(NULL, inode);
5180                 goto no_delete;
5181         }
5182         rsv->size = min_size;
5183         rsv->failfast = 1;
5184         global_rsv = &root->fs_info->global_block_rsv;
5185
5186         btrfs_i_size_write(inode, 0);
5187
5188         /*
5189          * This is a bit simpler than btrfs_truncate since we've already
5190          * reserved our space for our orphan item in the unlink, so we just
5191          * need to reserve some slack space in case we add bytes and update
5192          * inode item when doing the truncate.
5193          */
5194         while (1) {
5195                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5196                                              BTRFS_RESERVE_FLUSH_LIMIT);
5197
5198                 /*
5199                  * Try and steal from the global reserve since we will
5200                  * likely not use this space anyway, we want to try as
5201                  * hard as possible to get this to work.
5202                  */
5203                 if (ret)
5204                         steal_from_global++;
5205                 else
5206                         steal_from_global = 0;
5207                 ret = 0;
5208
5209                 /*
5210                  * steal_from_global == 0: we reserved stuff, hooray!
5211                  * steal_from_global == 1: we didn't reserve stuff, boo!
5212                  * steal_from_global == 2: we've committed, still not a lot of
5213                  * room but maybe we'll have room in the global reserve this
5214                  * time.
5215                  * steal_from_global == 3: abandon all hope!
5216                  */
5217                 if (steal_from_global > 2) {
5218                         btrfs_warn(root->fs_info,
5219                                 "Could not get space for a delete, will truncate on mount %d",
5220                                 ret);
5221                         btrfs_orphan_del(NULL, inode);
5222                         btrfs_free_block_rsv(root, rsv);
5223                         goto no_delete;
5224                 }
5225
5226                 trans = btrfs_join_transaction(root);
5227                 if (IS_ERR(trans)) {
5228                         btrfs_orphan_del(NULL, inode);
5229                         btrfs_free_block_rsv(root, rsv);
5230                         goto no_delete;
5231                 }
5232
5233                 /*
5234                  * We can't just steal from the global reserve, we need to make
5235                  * sure there is room to do it, if not we need to commit and try
5236                  * again.
5237                  */
5238                 if (steal_from_global) {
5239                         if (!btrfs_check_space_for_delayed_refs(trans, root))
5240                                 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5241                                                               min_size);
5242                         else
5243                                 ret = -ENOSPC;
5244                 }
5245
5246                 /*
5247                  * Couldn't steal from the global reserve, we have too much
5248                  * pending stuff built up, commit the transaction and try it
5249                  * again.
5250                  */
5251                 if (ret) {
5252                         ret = btrfs_commit_transaction(trans, root);
5253                         if (ret) {
5254                                 btrfs_orphan_del(NULL, inode);
5255                                 btrfs_free_block_rsv(root, rsv);
5256                                 goto no_delete;
5257                         }
5258                         continue;
5259                 } else {
5260                         steal_from_global = 0;
5261                 }
5262
5263                 trans->block_rsv = rsv;
5264
5265                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5266                 if (ret != -ENOSPC && ret != -EAGAIN)
5267                         break;
5268
5269                 trans->block_rsv = &root->fs_info->trans_block_rsv;
5270                 btrfs_end_transaction(trans, root);
5271                 trans = NULL;
5272                 btrfs_btree_balance_dirty(root);
5273         }
5274
5275         btrfs_free_block_rsv(root, rsv);
5276
5277         /*
5278          * Errors here aren't a big deal, it just means we leave orphan items
5279          * in the tree.  They will be cleaned up on the next mount.
5280          */
5281         if (ret == 0) {
5282                 trans->block_rsv = root->orphan_block_rsv;
5283                 btrfs_orphan_del(trans, inode);
5284         } else {
5285                 btrfs_orphan_del(NULL, inode);
5286         }
5287
5288         trans->block_rsv = &root->fs_info->trans_block_rsv;
5289         if (!(root == root->fs_info->tree_root ||
5290               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5291                 btrfs_return_ino(root, btrfs_ino(inode));
5292
5293         btrfs_end_transaction(trans, root);
5294         btrfs_btree_balance_dirty(root);
5295 no_delete:
5296         btrfs_remove_delayed_node(inode);
5297         clear_inode(inode);
5298 }
5299
5300 /*
5301  * this returns the key found in the dir entry in the location pointer.
5302  * If no dir entries were found, location->objectid is 0.
5303  */
5304 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5305                                struct btrfs_key *location)
5306 {
5307         const char *name = dentry->d_name.name;
5308         int namelen = dentry->d_name.len;
5309         struct btrfs_dir_item *di;
5310         struct btrfs_path *path;
5311         struct btrfs_root *root = BTRFS_I(dir)->root;
5312         int ret = 0;
5313
5314         path = btrfs_alloc_path();
5315         if (!path)
5316                 return -ENOMEM;
5317
5318         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
5319                                     namelen, 0);
5320         if (IS_ERR(di))
5321                 ret = PTR_ERR(di);
5322
5323         if (IS_ERR_OR_NULL(di))
5324                 goto out_err;
5325
5326         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5327 out:
5328         btrfs_free_path(path);
5329         return ret;
5330 out_err:
5331         location->objectid = 0;
5332         goto out;
5333 }
5334
5335 /*
5336  * when we hit a tree root in a directory, the btrfs part of the inode
5337  * needs to be changed to reflect the root directory of the tree root.  This
5338  * is kind of like crossing a mount point.
5339  */
5340 static int fixup_tree_root_location(struct btrfs_root *root,
5341                                     struct inode *dir,
5342                                     struct dentry *dentry,
5343                                     struct btrfs_key *location,
5344                                     struct btrfs_root **sub_root)
5345 {
5346         struct btrfs_path *path;
5347         struct btrfs_root *new_root;
5348         struct btrfs_root_ref *ref;
5349         struct extent_buffer *leaf;
5350         struct btrfs_key key;
5351         int ret;
5352         int err = 0;
5353
5354         path = btrfs_alloc_path();
5355         if (!path) {
5356                 err = -ENOMEM;
5357                 goto out;
5358         }
5359
5360         err = -ENOENT;
5361         key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5362         key.type = BTRFS_ROOT_REF_KEY;
5363         key.offset = location->objectid;
5364
5365         ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5366                                 0, 0);
5367         if (ret) {
5368                 if (ret < 0)
5369                         err = ret;
5370                 goto out;
5371         }
5372
5373         leaf = path->nodes[0];
5374         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5375         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5376             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5377                 goto out;
5378
5379         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5380                                    (unsigned long)(ref + 1),
5381                                    dentry->d_name.len);
5382         if (ret)
5383                 goto out;
5384
5385         btrfs_release_path(path);
5386
5387         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5388         if (IS_ERR(new_root)) {
5389                 err = PTR_ERR(new_root);
5390                 goto out;
5391         }
5392
5393         *sub_root = new_root;
5394         location->objectid = btrfs_root_dirid(&new_root->root_item);
5395         location->type = BTRFS_INODE_ITEM_KEY;
5396         location->offset = 0;
5397         err = 0;
5398 out:
5399         btrfs_free_path(path);
5400         return err;
5401 }
5402
5403 static void inode_tree_add(struct inode *inode)
5404 {
5405         struct btrfs_root *root = BTRFS_I(inode)->root;
5406         struct btrfs_inode *entry;
5407         struct rb_node **p;
5408         struct rb_node *parent;
5409         struct rb_node *new = &BTRFS_I(inode)->rb_node;
5410         u64 ino = btrfs_ino(inode);
5411
5412         if (inode_unhashed(inode))
5413                 return;
5414         parent = NULL;
5415         spin_lock(&root->inode_lock);
5416         p = &root->inode_tree.rb_node;
5417         while (*p) {
5418                 parent = *p;
5419                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5420
5421                 if (ino < btrfs_ino(&entry->vfs_inode))
5422                         p = &parent->rb_left;
5423                 else if (ino > btrfs_ino(&entry->vfs_inode))
5424                         p = &parent->rb_right;
5425                 else {
5426                         WARN_ON(!(entry->vfs_inode.i_state &
5427                                   (I_WILL_FREE | I_FREEING)));
5428                         rb_replace_node(parent, new, &root->inode_tree);
5429                         RB_CLEAR_NODE(parent);
5430                         spin_unlock(&root->inode_lock);
5431                         return;
5432                 }
5433         }
5434         rb_link_node(new, parent, p);
5435         rb_insert_color(new, &root->inode_tree);
5436         spin_unlock(&root->inode_lock);
5437 }
5438
5439 static void inode_tree_del(struct inode *inode)
5440 {
5441         struct btrfs_root *root = BTRFS_I(inode)->root;
5442         int empty = 0;
5443
5444         spin_lock(&root->inode_lock);
5445         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5446                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5447                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5448                 empty = RB_EMPTY_ROOT(&root->inode_tree);
5449         }
5450         spin_unlock(&root->inode_lock);
5451
5452         if (empty && btrfs_root_refs(&root->root_item) == 0) {
5453                 synchronize_srcu(&root->fs_info->subvol_srcu);
5454                 spin_lock(&root->inode_lock);
5455                 empty = RB_EMPTY_ROOT(&root->inode_tree);
5456                 spin_unlock(&root->inode_lock);
5457                 if (empty)
5458                         btrfs_add_dead_root(root);
5459         }
5460 }
5461
5462 void btrfs_invalidate_inodes(struct btrfs_root *root)
5463 {
5464         struct rb_node *node;
5465         struct rb_node *prev;
5466         struct btrfs_inode *entry;
5467         struct inode *inode;
5468         u64 objectid = 0;
5469
5470         if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5471                 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5472
5473         spin_lock(&root->inode_lock);
5474 again:
5475         node = root->inode_tree.rb_node;
5476         prev = NULL;
5477         while (node) {
5478                 prev = node;
5479                 entry = rb_entry(node, struct btrfs_inode, rb_node);
5480
5481                 if (objectid < btrfs_ino(&entry->vfs_inode))
5482                         node = node->rb_left;
5483                 else if (objectid > btrfs_ino(&entry->vfs_inode))
5484                         node = node->rb_right;
5485                 else
5486                         break;
5487         }
5488         if (!node) {
5489                 while (prev) {
5490                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
5491                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5492                                 node = prev;
5493                                 break;
5494                         }
5495                         prev = rb_next(prev);
5496                 }
5497         }
5498         while (node) {
5499                 entry = rb_entry(node, struct btrfs_inode, rb_node);
5500                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5501                 inode = igrab(&entry->vfs_inode);
5502                 if (inode) {
5503                         spin_unlock(&root->inode_lock);
5504                         if (atomic_read(&inode->i_count) > 1)
5505                                 d_prune_aliases(inode);
5506                         /*
5507                          * btrfs_drop_inode will have it removed from
5508                          * the inode cache when its usage count
5509                          * hits zero.
5510                          */
5511                         iput(inode);
5512                         cond_resched();
5513                         spin_lock(&root->inode_lock);
5514                         goto again;
5515                 }
5516
5517                 if (cond_resched_lock(&root->inode_lock))
5518                         goto again;
5519
5520                 node = rb_next(node);
5521         }
5522         spin_unlock(&root->inode_lock);
5523 }
5524
5525 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5526 {
5527         struct btrfs_iget_args *args = p;
5528         inode->i_ino = args->location->objectid;
5529         memcpy(&BTRFS_I(inode)->location, args->location,
5530                sizeof(*args->location));
5531         BTRFS_I(inode)->root = args->root;
5532         return 0;
5533 }
5534
5535 static int btrfs_find_actor(struct inode *inode, void *opaque)
5536 {
5537         struct btrfs_iget_args *args = opaque;
5538         return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5539                 args->root == BTRFS_I(inode)->root;
5540 }
5541
5542 static struct inode *btrfs_iget_locked(struct super_block *s,
5543                                        struct btrfs_key *location,
5544                                        struct btrfs_root *root)
5545 {
5546         struct inode *inode;
5547         struct btrfs_iget_args args;
5548         unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5549
5550         args.location = location;
5551         args.root = root;
5552
5553         inode = iget5_locked(s, hashval, btrfs_find_actor,
5554                              btrfs_init_locked_inode,
5555                              (void *)&args);
5556         return inode;
5557 }
5558
5559 /* Get an inode object given its location and corresponding root.
5560  * Returns in *is_new if the inode was read from disk
5561  */
5562 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5563                          struct btrfs_root *root, int *new)
5564 {
5565         struct inode *inode;
5566
5567         inode = btrfs_iget_locked(s, location, root);
5568         if (!inode)
5569                 return ERR_PTR(-ENOMEM);
5570
5571         if (inode->i_state & I_NEW) {
5572                 btrfs_read_locked_inode(inode);
5573                 if (!is_bad_inode(inode)) {
5574                         inode_tree_add(inode);
5575                         unlock_new_inode(inode);
5576                         if (new)
5577                                 *new = 1;
5578                 } else {
5579                         unlock_new_inode(inode);
5580                         iput(inode);
5581                         inode = ERR_PTR(-ESTALE);
5582                 }
5583         }
5584
5585         return inode;
5586 }
5587
5588 static struct inode *new_simple_dir(struct super_block *s,
5589                                     struct btrfs_key *key,
5590                                     struct btrfs_root *root)
5591 {
5592         struct inode *inode = new_inode(s);
5593
5594         if (!inode)
5595                 return ERR_PTR(-ENOMEM);
5596
5597         BTRFS_I(inode)->root = root;
5598         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5599         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5600
5601         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5602         inode->i_op = &btrfs_dir_ro_inode_operations;
5603         inode->i_fop = &simple_dir_operations;
5604         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5605         inode->i_mtime = current_fs_time(inode->i_sb);
5606         inode->i_atime = inode->i_mtime;
5607         inode->i_ctime = inode->i_mtime;
5608         BTRFS_I(inode)->i_otime = inode->i_mtime;
5609
5610         return inode;
5611 }
5612
5613 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5614 {
5615         struct inode *inode;
5616         struct btrfs_root *root = BTRFS_I(dir)->root;
5617         struct btrfs_root *sub_root = root;
5618         struct btrfs_key location;
5619         int index;
5620         int ret = 0;
5621
5622         if (dentry->d_name.len > BTRFS_NAME_LEN)
5623                 return ERR_PTR(-ENAMETOOLONG);
5624
5625         ret = btrfs_inode_by_name(dir, dentry, &location);
5626         if (ret < 0)
5627                 return ERR_PTR(ret);
5628
5629         if (location.objectid == 0)
5630                 return ERR_PTR(-ENOENT);
5631
5632         if (location.type == BTRFS_INODE_ITEM_KEY) {
5633                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5634                 return inode;
5635         }
5636
5637         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5638
5639         index = srcu_read_lock(&root->fs_info->subvol_srcu);
5640         ret = fixup_tree_root_location(root, dir, dentry,
5641                                        &location, &sub_root);
5642         if (ret < 0) {
5643                 if (ret != -ENOENT)
5644                         inode = ERR_PTR(ret);
5645                 else
5646                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
5647         } else {
5648                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5649         }
5650         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5651
5652         if (!IS_ERR(inode) && root != sub_root) {
5653                 down_read(&root->fs_info->cleanup_work_sem);
5654                 if (!(inode->i_sb->s_flags & MS_RDONLY))
5655                         ret = btrfs_orphan_cleanup(sub_root);
5656                 up_read(&root->fs_info->cleanup_work_sem);
5657                 if (ret) {
5658                         iput(inode);
5659                         inode = ERR_PTR(ret);
5660                 }
5661         }
5662
5663         return inode;
5664 }
5665
5666 static int btrfs_dentry_delete(const struct dentry *dentry)
5667 {
5668         struct btrfs_root *root;
5669         struct inode *inode = d_inode(dentry);
5670
5671         if (!inode && !IS_ROOT(dentry))
5672                 inode = d_inode(dentry->d_parent);
5673
5674         if (inode) {
5675                 root = BTRFS_I(inode)->root;
5676                 if (btrfs_root_refs(&root->root_item) == 0)
5677                         return 1;
5678
5679                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5680                         return 1;
5681         }
5682         return 0;
5683 }
5684
5685 static void btrfs_dentry_release(struct dentry *dentry)
5686 {
5687         kfree(dentry->d_fsdata);
5688 }
5689
5690 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5691                                    unsigned int flags)
5692 {
5693         struct inode *inode;
5694
5695         inode = btrfs_lookup_dentry(dir, dentry);
5696         if (IS_ERR(inode)) {
5697                 if (PTR_ERR(inode) == -ENOENT)
5698                         inode = NULL;
5699                 else
5700                         return ERR_CAST(inode);
5701         }
5702
5703         return d_splice_alias(inode, dentry);
5704 }
5705
5706 unsigned char btrfs_filetype_table[] = {
5707         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5708 };
5709
5710 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5711 {
5712         struct inode *inode = file_inode(file);
5713         struct btrfs_root *root = BTRFS_I(inode)->root;
5714         struct btrfs_item *item;
5715         struct btrfs_dir_item *di;
5716         struct btrfs_key key;
5717         struct btrfs_key found_key;
5718         struct btrfs_path *path;
5719         struct list_head ins_list;
5720         struct list_head del_list;
5721         int ret;
5722         struct extent_buffer *leaf;
5723         int slot;
5724         unsigned char d_type;
5725         int over = 0;
5726         u32 di_cur;
5727         u32 di_total;
5728         u32 di_len;
5729         int key_type = BTRFS_DIR_INDEX_KEY;
5730         char tmp_name[32];
5731         char *name_ptr;
5732         int name_len;
5733         int is_curr = 0;        /* ctx->pos points to the current index? */
5734         bool emitted;
5735
5736         /* FIXME, use a real flag for deciding about the key type */
5737         if (root->fs_info->tree_root == root)
5738                 key_type = BTRFS_DIR_ITEM_KEY;
5739
5740         if (!dir_emit_dots(file, ctx))
5741                 return 0;
5742
5743         path = btrfs_alloc_path();
5744         if (!path)
5745                 return -ENOMEM;
5746
5747         path->reada = READA_FORWARD;
5748
5749         if (key_type == BTRFS_DIR_INDEX_KEY) {
5750                 INIT_LIST_HEAD(&ins_list);
5751                 INIT_LIST_HEAD(&del_list);
5752                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5753         }
5754
5755         key.type = key_type;
5756         key.offset = ctx->pos;
5757         key.objectid = btrfs_ino(inode);
5758
5759         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5760         if (ret < 0)
5761                 goto err;
5762
5763         emitted = false;
5764         while (1) {
5765                 leaf = path->nodes[0];
5766                 slot = path->slots[0];
5767                 if (slot >= btrfs_header_nritems(leaf)) {
5768                         ret = btrfs_next_leaf(root, path);
5769                         if (ret < 0)
5770                                 goto err;
5771                         else if (ret > 0)
5772                                 break;
5773                         continue;
5774                 }
5775
5776                 item = btrfs_item_nr(slot);
5777                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5778
5779                 if (found_key.objectid != key.objectid)
5780                         break;
5781                 if (found_key.type != key_type)
5782                         break;
5783                 if (found_key.offset < ctx->pos)
5784                         goto next;
5785                 if (key_type == BTRFS_DIR_INDEX_KEY &&
5786                     btrfs_should_delete_dir_index(&del_list,
5787                                                   found_key.offset))
5788                         goto next;
5789
5790                 ctx->pos = found_key.offset;
5791                 is_curr = 1;
5792
5793                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5794                 di_cur = 0;
5795                 di_total = btrfs_item_size(leaf, item);
5796
5797                 while (di_cur < di_total) {
5798                         struct btrfs_key location;
5799
5800                         if (verify_dir_item(root, leaf, di))
5801                                 break;
5802
5803                         name_len = btrfs_dir_name_len(leaf, di);
5804                         if (name_len <= sizeof(tmp_name)) {
5805                                 name_ptr = tmp_name;
5806                         } else {
5807                                 name_ptr = kmalloc(name_len, GFP_KERNEL);
5808                                 if (!name_ptr) {
5809                                         ret = -ENOMEM;
5810                                         goto err;
5811                                 }
5812                         }
5813                         read_extent_buffer(leaf, name_ptr,
5814                                            (unsigned long)(di + 1), name_len);
5815
5816                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5817                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
5818
5819
5820                         /* is this a reference to our own snapshot? If so
5821                          * skip it.
5822                          *
5823                          * In contrast to old kernels, we insert the snapshot's
5824                          * dir item and dir index after it has been created, so
5825                          * we won't find a reference to our own snapshot. We
5826                          * still keep the following code for backward
5827                          * compatibility.
5828                          */
5829                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
5830                             location.objectid == root->root_key.objectid) {
5831                                 over = 0;
5832                                 goto skip;
5833                         }
5834                         over = !dir_emit(ctx, name_ptr, name_len,
5835                                        location.objectid, d_type);
5836
5837 skip:
5838                         if (name_ptr != tmp_name)
5839                                 kfree(name_ptr);
5840
5841                         if (over)
5842                                 goto nopos;
5843                         emitted = true;
5844                         di_len = btrfs_dir_name_len(leaf, di) +
5845                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
5846                         di_cur += di_len;
5847                         di = (struct btrfs_dir_item *)((char *)di + di_len);
5848                 }
5849 next:
5850                 path->slots[0]++;
5851         }
5852
5853         if (key_type == BTRFS_DIR_INDEX_KEY) {
5854                 if (is_curr)
5855                         ctx->pos++;
5856                 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list, &emitted);
5857                 if (ret)
5858                         goto nopos;
5859         }
5860
5861         /*
5862          * If we haven't emitted any dir entry, we must not touch ctx->pos as
5863          * it was was set to the termination value in previous call. We assume
5864          * that "." and ".." were emitted if we reach this point and set the
5865          * termination value as well for an empty directory.
5866          */
5867         if (ctx->pos > 2 && !emitted)
5868                 goto nopos;
5869
5870         /* Reached end of directory/root. Bump pos past the last item. */
5871         ctx->pos++;
5872
5873         /*
5874          * Stop new entries from being returned after we return the last
5875          * entry.
5876          *
5877          * New directory entries are assigned a strictly increasing
5878          * offset.  This means that new entries created during readdir
5879          * are *guaranteed* to be seen in the future by that readdir.
5880          * This has broken buggy programs which operate on names as
5881          * they're returned by readdir.  Until we re-use freed offsets
5882          * we have this hack to stop new entries from being returned
5883          * under the assumption that they'll never reach this huge
5884          * offset.
5885          *
5886          * This is being careful not to overflow 32bit loff_t unless the
5887          * last entry requires it because doing so has broken 32bit apps
5888          * in the past.
5889          */
5890         if (key_type == BTRFS_DIR_INDEX_KEY) {
5891                 if (ctx->pos >= INT_MAX)
5892                         ctx->pos = LLONG_MAX;
5893                 else
5894                         ctx->pos = INT_MAX;
5895         }
5896 nopos:
5897         ret = 0;
5898 err:
5899         if (key_type == BTRFS_DIR_INDEX_KEY)
5900                 btrfs_put_delayed_items(&ins_list, &del_list);
5901         btrfs_free_path(path);
5902         return ret;
5903 }
5904
5905 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5906 {
5907         struct btrfs_root *root = BTRFS_I(inode)->root;
5908         struct btrfs_trans_handle *trans;
5909         int ret = 0;
5910         bool nolock = false;
5911
5912         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5913                 return 0;
5914
5915         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5916                 nolock = true;
5917
5918         if (wbc->sync_mode == WB_SYNC_ALL) {
5919                 if (nolock)
5920                         trans = btrfs_join_transaction_nolock(root);
5921                 else
5922                         trans = btrfs_join_transaction(root);
5923                 if (IS_ERR(trans))
5924                         return PTR_ERR(trans);
5925                 ret = btrfs_commit_transaction(trans, root);
5926         }
5927         return ret;
5928 }
5929
5930 /*
5931  * This is somewhat expensive, updating the tree every time the
5932  * inode changes.  But, it is most likely to find the inode in cache.
5933  * FIXME, needs more benchmarking...there are no reasons other than performance
5934  * to keep or drop this code.
5935  */
5936 static int btrfs_dirty_inode(struct inode *inode)
5937 {
5938         struct btrfs_root *root = BTRFS_I(inode)->root;
5939         struct btrfs_trans_handle *trans;
5940         int ret;
5941
5942         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5943                 return 0;
5944
5945         trans = btrfs_join_transaction(root);
5946         if (IS_ERR(trans))
5947                 return PTR_ERR(trans);
5948
5949         ret = btrfs_update_inode(trans, root, inode);
5950         if (ret && ret == -ENOSPC) {
5951                 /* whoops, lets try again with the full transaction */
5952                 btrfs_end_transaction(trans, root);
5953                 trans = btrfs_start_transaction(root, 1);
5954                 if (IS_ERR(trans))
5955                         return PTR_ERR(trans);
5956
5957                 ret = btrfs_update_inode(trans, root, inode);
5958         }
5959         btrfs_end_transaction(trans, root);
5960         if (BTRFS_I(inode)->delayed_node)
5961                 btrfs_balance_delayed_items(root);
5962
5963         return ret;
5964 }
5965
5966 /*
5967  * This is a copy of file_update_time.  We need this so we can return error on
5968  * ENOSPC for updating the inode in the case of file write and mmap writes.
5969  */
5970 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5971                              int flags)
5972 {
5973         struct btrfs_root *root = BTRFS_I(inode)->root;
5974
5975         if (btrfs_root_readonly(root))
5976                 return -EROFS;
5977
5978         if (flags & S_VERSION)
5979                 inode_inc_iversion(inode);
5980         if (flags & S_CTIME)
5981                 inode->i_ctime = *now;
5982         if (flags & S_MTIME)
5983                 inode->i_mtime = *now;
5984         if (flags & S_ATIME)
5985                 inode->i_atime = *now;
5986         return btrfs_dirty_inode(inode);
5987 }
5988
5989 /*
5990  * find the highest existing sequence number in a directory
5991  * and then set the in-memory index_cnt variable to reflect
5992  * free sequence numbers
5993  */
5994 static int btrfs_set_inode_index_count(struct inode *inode)
5995 {
5996         struct btrfs_root *root = BTRFS_I(inode)->root;
5997         struct btrfs_key key, found_key;
5998         struct btrfs_path *path;
5999         struct extent_buffer *leaf;
6000         int ret;
6001
6002         key.objectid = btrfs_ino(inode);
6003         key.type = BTRFS_DIR_INDEX_KEY;
6004         key.offset = (u64)-1;
6005
6006         path = btrfs_alloc_path();
6007         if (!path)
6008                 return -ENOMEM;
6009
6010         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6011         if (ret < 0)
6012                 goto out;
6013         /* FIXME: we should be able to handle this */
6014         if (ret == 0)
6015                 goto out;
6016         ret = 0;
6017
6018         /*
6019          * MAGIC NUMBER EXPLANATION:
6020          * since we search a directory based on f_pos we have to start at 2
6021          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6022          * else has to start at 2
6023          */
6024         if (path->slots[0] == 0) {
6025                 BTRFS_I(inode)->index_cnt = 2;
6026                 goto out;
6027         }
6028
6029         path->slots[0]--;
6030
6031         leaf = path->nodes[0];
6032         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6033
6034         if (found_key.objectid != btrfs_ino(inode) ||
6035             found_key.type != BTRFS_DIR_INDEX_KEY) {
6036                 BTRFS_I(inode)->index_cnt = 2;
6037                 goto out;
6038         }
6039
6040         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
6041 out:
6042         btrfs_free_path(path);
6043         return ret;
6044 }
6045
6046 /*
6047  * helper to find a free sequence number in a given directory.  This current
6048  * code is very simple, later versions will do smarter things in the btree
6049  */
6050 int btrfs_set_inode_index(struct inode *dir, u64 *index)
6051 {
6052         int ret = 0;
6053
6054         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
6055                 ret = btrfs_inode_delayed_dir_index_count(dir);
6056                 if (ret) {
6057                         ret = btrfs_set_inode_index_count(dir);
6058                         if (ret)
6059                                 return ret;
6060                 }
6061         }
6062
6063         *index = BTRFS_I(dir)->index_cnt;
6064         BTRFS_I(dir)->index_cnt++;
6065
6066         return ret;
6067 }
6068
6069 static int btrfs_insert_inode_locked(struct inode *inode)
6070 {
6071         struct btrfs_iget_args args;
6072         args.location = &BTRFS_I(inode)->location;
6073         args.root = BTRFS_I(inode)->root;
6074
6075         return insert_inode_locked4(inode,
6076                    btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6077                    btrfs_find_actor, &args);
6078 }
6079
6080 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6081                                      struct btrfs_root *root,
6082                                      struct inode *dir,
6083                                      const char *name, int name_len,
6084                                      u64 ref_objectid, u64 objectid,
6085                                      umode_t mode, u64 *index)
6086 {
6087         struct inode *inode;
6088         struct btrfs_inode_item *inode_item;
6089         struct btrfs_key *location;
6090         struct btrfs_path *path;
6091         struct btrfs_inode_ref *ref;
6092         struct btrfs_key key[2];
6093         u32 sizes[2];
6094         int nitems = name ? 2 : 1;
6095         unsigned long ptr;
6096         int ret;
6097
6098         path = btrfs_alloc_path();
6099         if (!path)
6100                 return ERR_PTR(-ENOMEM);
6101
6102         inode = new_inode(root->fs_info->sb);
6103         if (!inode) {
6104                 btrfs_free_path(path);
6105                 return ERR_PTR(-ENOMEM);
6106         }
6107
6108         /*
6109          * O_TMPFILE, set link count to 0, so that after this point,
6110          * we fill in an inode item with the correct link count.
6111          */
6112         if (!name)
6113                 set_nlink(inode, 0);
6114
6115         /*
6116          * we have to initialize this early, so we can reclaim the inode
6117          * number if we fail afterwards in this function.
6118          */
6119         inode->i_ino = objectid;
6120
6121         if (dir && name) {
6122                 trace_btrfs_inode_request(dir);
6123
6124                 ret = btrfs_set_inode_index(dir, index);
6125                 if (ret) {
6126                         btrfs_free_path(path);
6127                         iput(inode);
6128                         return ERR_PTR(ret);
6129                 }
6130         } else if (dir) {
6131                 *index = 0;
6132         }
6133         /*
6134          * index_cnt is ignored for everything but a dir,
6135          * btrfs_get_inode_index_count has an explanation for the magic
6136          * number
6137          */
6138         BTRFS_I(inode)->index_cnt = 2;
6139         BTRFS_I(inode)->dir_index = *index;
6140         BTRFS_I(inode)->root = root;
6141         BTRFS_I(inode)->generation = trans->transid;
6142         inode->i_generation = BTRFS_I(inode)->generation;
6143
6144         /*
6145          * We could have gotten an inode number from somebody who was fsynced
6146          * and then removed in this same transaction, so let's just set full
6147          * sync since it will be a full sync anyway and this will blow away the
6148          * old info in the log.
6149          */
6150         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6151
6152         key[0].objectid = objectid;
6153         key[0].type = BTRFS_INODE_ITEM_KEY;
6154         key[0].offset = 0;
6155
6156         sizes[0] = sizeof(struct btrfs_inode_item);
6157
6158         if (name) {
6159                 /*
6160                  * Start new inodes with an inode_ref. This is slightly more
6161                  * efficient for small numbers of hard links since they will
6162                  * be packed into one item. Extended refs will kick in if we
6163                  * add more hard links than can fit in the ref item.
6164                  */
6165                 key[1].objectid = objectid;
6166                 key[1].type = BTRFS_INODE_REF_KEY;
6167                 key[1].offset = ref_objectid;
6168
6169                 sizes[1] = name_len + sizeof(*ref);
6170         }
6171
6172         location = &BTRFS_I(inode)->location;
6173         location->objectid = objectid;
6174         location->offset = 0;
6175         location->type = BTRFS_INODE_ITEM_KEY;
6176
6177         ret = btrfs_insert_inode_locked(inode);
6178         if (ret < 0)
6179                 goto fail;
6180
6181         path->leave_spinning = 1;
6182         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6183         if (ret != 0)
6184                 goto fail_unlock;
6185
6186         inode_init_owner(inode, dir, mode);
6187         inode_set_bytes(inode, 0);
6188
6189         inode->i_mtime = current_fs_time(inode->i_sb);
6190         inode->i_atime = inode->i_mtime;
6191         inode->i_ctime = inode->i_mtime;
6192         BTRFS_I(inode)->i_otime = inode->i_mtime;
6193
6194         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6195                                   struct btrfs_inode_item);
6196         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6197                              sizeof(*inode_item));
6198         fill_inode_item(trans, path->nodes[0], inode_item, inode);
6199
6200         if (name) {
6201                 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6202                                      struct btrfs_inode_ref);
6203                 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6204                 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6205                 ptr = (unsigned long)(ref + 1);
6206                 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6207         }
6208
6209         btrfs_mark_buffer_dirty(path->nodes[0]);
6210         btrfs_free_path(path);
6211
6212         btrfs_inherit_iflags(inode, dir);
6213
6214         if (S_ISREG(mode)) {
6215                 if (btrfs_test_opt(root, NODATASUM))
6216                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6217                 if (btrfs_test_opt(root, NODATACOW))
6218                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6219                                 BTRFS_INODE_NODATASUM;
6220         }
6221
6222         inode_tree_add(inode);
6223
6224         trace_btrfs_inode_new(inode);
6225         btrfs_set_inode_last_trans(trans, inode);
6226
6227         btrfs_update_root_times(trans, root);
6228
6229         ret = btrfs_inode_inherit_props(trans, inode, dir);
6230         if (ret)
6231                 btrfs_err(root->fs_info,
6232                           "error inheriting props for ino %llu (root %llu): %d",
6233                           btrfs_ino(inode), root->root_key.objectid, ret);
6234
6235         return inode;
6236
6237 fail_unlock:
6238         unlock_new_inode(inode);
6239 fail:
6240         if (dir && name)
6241                 BTRFS_I(dir)->index_cnt--;
6242         btrfs_free_path(path);
6243         iput(inode);
6244         return ERR_PTR(ret);
6245 }
6246
6247 static inline u8 btrfs_inode_type(struct inode *inode)
6248 {
6249         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6250 }
6251
6252 /*
6253  * utility function to add 'inode' into 'parent_inode' with
6254  * a give name and a given sequence number.
6255  * if 'add_backref' is true, also insert a backref from the
6256  * inode to the parent directory.
6257  */
6258 int btrfs_add_link(struct btrfs_trans_handle *trans,
6259                    struct inode *parent_inode, struct inode *inode,
6260                    const char *name, int name_len, int add_backref, u64 index)
6261 {
6262         int ret = 0;
6263         struct btrfs_key key;
6264         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
6265         u64 ino = btrfs_ino(inode);
6266         u64 parent_ino = btrfs_ino(parent_inode);
6267
6268         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6269                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6270         } else {
6271                 key.objectid = ino;
6272                 key.type = BTRFS_INODE_ITEM_KEY;
6273                 key.offset = 0;
6274         }
6275
6276         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6277                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6278                                          key.objectid, root->root_key.objectid,
6279                                          parent_ino, index, name, name_len);
6280         } else if (add_backref) {
6281                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6282                                              parent_ino, index);
6283         }
6284
6285         /* Nothing to clean up yet */
6286         if (ret)
6287                 return ret;
6288
6289         ret = btrfs_insert_dir_item(trans, root, name, name_len,
6290                                     parent_inode, &key,
6291                                     btrfs_inode_type(inode), index);
6292         if (ret == -EEXIST || ret == -EOVERFLOW)
6293                 goto fail_dir_item;
6294         else if (ret) {
6295                 btrfs_abort_transaction(trans, root, ret);
6296                 return ret;
6297         }
6298
6299         btrfs_i_size_write(parent_inode, parent_inode->i_size +
6300                            name_len * 2);
6301         inode_inc_iversion(parent_inode);
6302         parent_inode->i_mtime = parent_inode->i_ctime =
6303                 current_fs_time(parent_inode->i_sb);
6304         ret = btrfs_update_inode(trans, root, parent_inode);
6305         if (ret)
6306                 btrfs_abort_transaction(trans, root, ret);
6307         return ret;
6308
6309 fail_dir_item:
6310         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6311                 u64 local_index;
6312                 int err;
6313                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6314                                  key.objectid, root->root_key.objectid,
6315                                  parent_ino, &local_index, name, name_len);
6316
6317         } else if (add_backref) {
6318                 u64 local_index;
6319                 int err;
6320
6321                 err = btrfs_del_inode_ref(trans, root, name, name_len,
6322                                           ino, parent_ino, &local_index);
6323         }
6324         return ret;
6325 }
6326
6327 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6328                             struct inode *dir, struct dentry *dentry,
6329                             struct inode *inode, int backref, u64 index)
6330 {
6331         int err = btrfs_add_link(trans, dir, inode,
6332                                  dentry->d_name.name, dentry->d_name.len,
6333                                  backref, index);
6334         if (err > 0)
6335                 err = -EEXIST;
6336         return err;
6337 }
6338
6339 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6340                         umode_t mode, dev_t rdev)
6341 {
6342         struct btrfs_trans_handle *trans;
6343         struct btrfs_root *root = BTRFS_I(dir)->root;
6344         struct inode *inode = NULL;
6345         int err;
6346         int drop_inode = 0;
6347         u64 objectid;
6348         u64 index = 0;
6349
6350         /*
6351          * 2 for inode item and ref
6352          * 2 for dir items
6353          * 1 for xattr if selinux is on
6354          */
6355         trans = btrfs_start_transaction(root, 5);
6356         if (IS_ERR(trans))
6357                 return PTR_ERR(trans);
6358
6359         err = btrfs_find_free_ino(root, &objectid);
6360         if (err)
6361                 goto out_unlock;
6362
6363         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6364                                 dentry->d_name.len, btrfs_ino(dir), objectid,
6365                                 mode, &index);
6366         if (IS_ERR(inode)) {
6367                 err = PTR_ERR(inode);
6368                 goto out_unlock;
6369         }
6370
6371         /*
6372         * If the active LSM wants to access the inode during
6373         * d_instantiate it needs these. Smack checks to see
6374         * if the filesystem supports xattrs by looking at the
6375         * ops vector.
6376         */
6377         inode->i_op = &btrfs_special_inode_operations;
6378         init_special_inode(inode, inode->i_mode, rdev);
6379
6380         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6381         if (err)
6382                 goto out_unlock_inode;
6383
6384         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6385         if (err) {
6386                 goto out_unlock_inode;
6387         } else {
6388                 btrfs_update_inode(trans, root, inode);
6389                 unlock_new_inode(inode);
6390                 d_instantiate(dentry, inode);
6391         }
6392
6393 out_unlock:
6394         btrfs_end_transaction(trans, root);
6395         btrfs_balance_delayed_items(root);
6396         btrfs_btree_balance_dirty(root);
6397         if (drop_inode) {
6398                 inode_dec_link_count(inode);
6399                 iput(inode);
6400         }
6401         return err;
6402
6403 out_unlock_inode:
6404         drop_inode = 1;
6405         unlock_new_inode(inode);
6406         goto out_unlock;
6407
6408 }
6409
6410 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6411                         umode_t mode, bool excl)
6412 {
6413         struct btrfs_trans_handle *trans;
6414         struct btrfs_root *root = BTRFS_I(dir)->root;
6415         struct inode *inode = NULL;
6416         int drop_inode_on_err = 0;
6417         int err;
6418         u64 objectid;
6419         u64 index = 0;
6420
6421         /*
6422          * 2 for inode item and ref
6423          * 2 for dir items
6424          * 1 for xattr if selinux is on
6425          */
6426         trans = btrfs_start_transaction(root, 5);
6427         if (IS_ERR(trans))
6428                 return PTR_ERR(trans);
6429
6430         err = btrfs_find_free_ino(root, &objectid);
6431         if (err)
6432                 goto out_unlock;
6433
6434         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6435                                 dentry->d_name.len, btrfs_ino(dir), objectid,
6436                                 mode, &index);
6437         if (IS_ERR(inode)) {
6438                 err = PTR_ERR(inode);
6439                 goto out_unlock;
6440         }
6441         drop_inode_on_err = 1;
6442         /*
6443         * If the active LSM wants to access the inode during
6444         * d_instantiate it needs these. Smack checks to see
6445         * if the filesystem supports xattrs by looking at the
6446         * ops vector.
6447         */
6448         inode->i_fop = &btrfs_file_operations;
6449         inode->i_op = &btrfs_file_inode_operations;
6450         inode->i_mapping->a_ops = &btrfs_aops;
6451
6452         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6453         if (err)
6454                 goto out_unlock_inode;
6455
6456         err = btrfs_update_inode(trans, root, inode);
6457         if (err)
6458                 goto out_unlock_inode;
6459
6460         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6461         if (err)
6462                 goto out_unlock_inode;
6463
6464         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6465         unlock_new_inode(inode);
6466         d_instantiate(dentry, inode);
6467
6468 out_unlock:
6469         btrfs_end_transaction(trans, root);
6470         if (err && drop_inode_on_err) {
6471                 inode_dec_link_count(inode);
6472                 iput(inode);
6473         }
6474         btrfs_balance_delayed_items(root);
6475         btrfs_btree_balance_dirty(root);
6476         return err;
6477
6478 out_unlock_inode:
6479         unlock_new_inode(inode);
6480         goto out_unlock;
6481
6482 }
6483
6484 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6485                       struct dentry *dentry)
6486 {
6487         struct btrfs_trans_handle *trans = NULL;
6488         struct btrfs_root *root = BTRFS_I(dir)->root;
6489         struct inode *inode = d_inode(old_dentry);
6490         u64 index;
6491         int err;
6492         int drop_inode = 0;
6493
6494         /* do not allow sys_link's with other subvols of the same device */
6495         if (root->objectid != BTRFS_I(inode)->root->objectid)
6496                 return -EXDEV;
6497
6498         if (inode->i_nlink >= BTRFS_LINK_MAX)
6499                 return -EMLINK;
6500
6501         err = btrfs_set_inode_index(dir, &index);
6502         if (err)
6503                 goto fail;
6504
6505         /*
6506          * 2 items for inode and inode ref
6507          * 2 items for dir items
6508          * 1 item for parent inode
6509          */
6510         trans = btrfs_start_transaction(root, 5);
6511         if (IS_ERR(trans)) {
6512                 err = PTR_ERR(trans);
6513                 trans = NULL;
6514                 goto fail;
6515         }
6516
6517         /* There are several dir indexes for this inode, clear the cache. */
6518         BTRFS_I(inode)->dir_index = 0ULL;
6519         inc_nlink(inode);
6520         inode_inc_iversion(inode);
6521         inode->i_ctime = current_fs_time(inode->i_sb);
6522         ihold(inode);
6523         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6524
6525         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6526
6527         if (err) {
6528                 drop_inode = 1;
6529         } else {
6530                 struct dentry *parent = dentry->d_parent;
6531                 err = btrfs_update_inode(trans, root, inode);
6532                 if (err)
6533                         goto fail;
6534                 if (inode->i_nlink == 1) {
6535                         /*
6536                          * If new hard link count is 1, it's a file created
6537                          * with open(2) O_TMPFILE flag.
6538                          */
6539                         err = btrfs_orphan_del(trans, inode);
6540                         if (err)
6541                                 goto fail;
6542                 }
6543                 d_instantiate(dentry, inode);
6544                 btrfs_log_new_name(trans, inode, NULL, parent);
6545         }
6546
6547         btrfs_balance_delayed_items(root);
6548 fail:
6549         if (trans)
6550                 btrfs_end_transaction(trans, root);
6551         if (drop_inode) {
6552                 inode_dec_link_count(inode);
6553                 iput(inode);
6554         }
6555         btrfs_btree_balance_dirty(root);
6556         return err;
6557 }
6558
6559 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6560 {
6561         struct inode *inode = NULL;
6562         struct btrfs_trans_handle *trans;
6563         struct btrfs_root *root = BTRFS_I(dir)->root;
6564         int err = 0;
6565         int drop_on_err = 0;
6566         u64 objectid = 0;
6567         u64 index = 0;
6568
6569         /*
6570          * 2 items for inode and ref
6571          * 2 items for dir items
6572          * 1 for xattr if selinux is on
6573          */
6574         trans = btrfs_start_transaction(root, 5);
6575         if (IS_ERR(trans))
6576                 return PTR_ERR(trans);
6577
6578         err = btrfs_find_free_ino(root, &objectid);
6579         if (err)
6580                 goto out_fail;
6581
6582         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6583                                 dentry->d_name.len, btrfs_ino(dir), objectid,
6584                                 S_IFDIR | mode, &index);
6585         if (IS_ERR(inode)) {
6586                 err = PTR_ERR(inode);
6587                 goto out_fail;
6588         }
6589
6590         drop_on_err = 1;
6591         /* these must be set before we unlock the inode */
6592         inode->i_op = &btrfs_dir_inode_operations;
6593         inode->i_fop = &btrfs_dir_file_operations;
6594
6595         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6596         if (err)
6597                 goto out_fail_inode;
6598
6599         btrfs_i_size_write(inode, 0);
6600         err = btrfs_update_inode(trans, root, inode);
6601         if (err)
6602                 goto out_fail_inode;
6603
6604         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6605                              dentry->d_name.len, 0, index);
6606         if (err)
6607                 goto out_fail_inode;
6608
6609         d_instantiate(dentry, inode);
6610         /*
6611          * mkdir is special.  We're unlocking after we call d_instantiate
6612          * to avoid a race with nfsd calling d_instantiate.
6613          */
6614         unlock_new_inode(inode);
6615         drop_on_err = 0;
6616
6617 out_fail:
6618         btrfs_end_transaction(trans, root);
6619         if (drop_on_err) {
6620                 inode_dec_link_count(inode);
6621                 iput(inode);
6622         }
6623         btrfs_balance_delayed_items(root);
6624         btrfs_btree_balance_dirty(root);
6625         return err;
6626
6627 out_fail_inode:
6628         unlock_new_inode(inode);
6629         goto out_fail;
6630 }
6631
6632 /* Find next extent map of a given extent map, caller needs to ensure locks */
6633 static struct extent_map *next_extent_map(struct extent_map *em)
6634 {
6635         struct rb_node *next;
6636
6637         next = rb_next(&em->rb_node);
6638         if (!next)
6639                 return NULL;
6640         return container_of(next, struct extent_map, rb_node);
6641 }
6642
6643 static struct extent_map *prev_extent_map(struct extent_map *em)
6644 {
6645         struct rb_node *prev;
6646
6647         prev = rb_prev(&em->rb_node);
6648         if (!prev)
6649                 return NULL;
6650         return container_of(prev, struct extent_map, rb_node);
6651 }
6652
6653 /* helper for btfs_get_extent.  Given an existing extent in the tree,
6654  * the existing extent is the nearest extent to map_start,
6655  * and an extent that you want to insert, deal with overlap and insert
6656  * the best fitted new extent into the tree.
6657  */
6658 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6659                                 struct extent_map *existing,
6660                                 struct extent_map *em,
6661                                 u64 map_start)
6662 {
6663         struct extent_map *prev;
6664         struct extent_map *next;
6665         u64 start;
6666         u64 end;
6667         u64 start_diff;
6668
6669         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6670
6671         if (existing->start > map_start) {
6672                 next = existing;
6673                 prev = prev_extent_map(next);
6674         } else {
6675                 prev = existing;
6676                 next = next_extent_map(prev);
6677         }
6678
6679         start = prev ? extent_map_end(prev) : em->start;
6680         start = max_t(u64, start, em->start);
6681         end = next ? next->start : extent_map_end(em);
6682         end = min_t(u64, end, extent_map_end(em));
6683         start_diff = start - em->start;
6684         em->start = start;
6685         em->len = end - start;
6686         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6687             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6688                 em->block_start += start_diff;
6689                 em->block_len -= start_diff;
6690         }
6691         return add_extent_mapping(em_tree, em, 0);
6692 }
6693
6694 static noinline int uncompress_inline(struct btrfs_path *path,
6695                                       struct page *page,
6696                                       size_t pg_offset, u64 extent_offset,
6697                                       struct btrfs_file_extent_item *item)
6698 {
6699         int ret;
6700         struct extent_buffer *leaf = path->nodes[0];
6701         char *tmp;
6702         size_t max_size;
6703         unsigned long inline_size;
6704         unsigned long ptr;
6705         int compress_type;
6706
6707         WARN_ON(pg_offset != 0);
6708         compress_type = btrfs_file_extent_compression(leaf, item);
6709         max_size = btrfs_file_extent_ram_bytes(leaf, item);
6710         inline_size = btrfs_file_extent_inline_item_len(leaf,
6711                                         btrfs_item_nr(path->slots[0]));
6712         tmp = kmalloc(inline_size, GFP_NOFS);
6713         if (!tmp)
6714                 return -ENOMEM;
6715         ptr = btrfs_file_extent_inline_start(item);
6716
6717         read_extent_buffer(leaf, tmp, ptr, inline_size);
6718
6719         max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6720         ret = btrfs_decompress(compress_type, tmp, page,
6721                                extent_offset, inline_size, max_size);
6722         kfree(tmp);
6723         return ret;
6724 }
6725
6726 /*
6727  * a bit scary, this does extent mapping from logical file offset to the disk.
6728  * the ugly parts come from merging extents from the disk with the in-ram
6729  * representation.  This gets more complex because of the data=ordered code,
6730  * where the in-ram extents might be locked pending data=ordered completion.
6731  *
6732  * This also copies inline extents directly into the page.
6733  */
6734
6735 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6736                                     size_t pg_offset, u64 start, u64 len,
6737                                     int create)
6738 {
6739         int ret;
6740         int err = 0;
6741         u64 extent_start = 0;
6742         u64 extent_end = 0;
6743         u64 objectid = btrfs_ino(inode);
6744         u32 found_type;
6745         struct btrfs_path *path = NULL;
6746         struct btrfs_root *root = BTRFS_I(inode)->root;
6747         struct btrfs_file_extent_item *item;
6748         struct extent_buffer *leaf;
6749         struct btrfs_key found_key;
6750         struct extent_map *em = NULL;
6751         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6752         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6753         struct btrfs_trans_handle *trans = NULL;
6754         const bool new_inline = !page || create;
6755
6756 again:
6757         read_lock(&em_tree->lock);
6758         em = lookup_extent_mapping(em_tree, start, len);
6759         if (em)
6760                 em->bdev = root->fs_info->fs_devices->latest_bdev;
6761         read_unlock(&em_tree->lock);
6762
6763         if (em) {
6764                 if (em->start > start || em->start + em->len <= start)
6765                         free_extent_map(em);
6766                 else if (em->block_start == EXTENT_MAP_INLINE && page)
6767                         free_extent_map(em);
6768                 else
6769                         goto out;
6770         }
6771         em = alloc_extent_map();
6772         if (!em) {
6773                 err = -ENOMEM;
6774                 goto out;
6775         }
6776         em->bdev = root->fs_info->fs_devices->latest_bdev;
6777         em->start = EXTENT_MAP_HOLE;
6778         em->orig_start = EXTENT_MAP_HOLE;
6779         em->len = (u64)-1;
6780         em->block_len = (u64)-1;
6781
6782         if (!path) {
6783                 path = btrfs_alloc_path();
6784                 if (!path) {
6785                         err = -ENOMEM;
6786                         goto out;
6787                 }
6788                 /*
6789                  * Chances are we'll be called again, so go ahead and do
6790                  * readahead
6791                  */
6792                 path->reada = READA_FORWARD;
6793         }
6794
6795         ret = btrfs_lookup_file_extent(trans, root, path,
6796                                        objectid, start, trans != NULL);
6797         if (ret < 0) {
6798                 err = ret;
6799                 goto out;
6800         }
6801
6802         if (ret != 0) {
6803                 if (path->slots[0] == 0)
6804                         goto not_found;
6805                 path->slots[0]--;
6806         }
6807
6808         leaf = path->nodes[0];
6809         item = btrfs_item_ptr(leaf, path->slots[0],
6810                               struct btrfs_file_extent_item);
6811         /* are we inside the extent that was found? */
6812         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6813         found_type = found_key.type;
6814         if (found_key.objectid != objectid ||
6815             found_type != BTRFS_EXTENT_DATA_KEY) {
6816                 /*
6817                  * If we backup past the first extent we want to move forward
6818                  * and see if there is an extent in front of us, otherwise we'll
6819                  * say there is a hole for our whole search range which can
6820                  * cause problems.
6821                  */
6822                 extent_end = start;
6823                 goto next;
6824         }
6825
6826         found_type = btrfs_file_extent_type(leaf, item);
6827         extent_start = found_key.offset;
6828         if (found_type == BTRFS_FILE_EXTENT_REG ||
6829             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6830                 extent_end = extent_start +
6831                        btrfs_file_extent_num_bytes(leaf, item);
6832         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6833                 size_t size;
6834                 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6835                 extent_end = ALIGN(extent_start + size, root->sectorsize);
6836         }
6837 next:
6838         if (start >= extent_end) {
6839                 path->slots[0]++;
6840                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6841                         ret = btrfs_next_leaf(root, path);
6842                         if (ret < 0) {
6843                                 err = ret;
6844                                 goto out;
6845                         }
6846                         if (ret > 0)
6847                                 goto not_found;
6848                         leaf = path->nodes[0];
6849                 }
6850                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6851                 if (found_key.objectid != objectid ||
6852                     found_key.type != BTRFS_EXTENT_DATA_KEY)
6853                         goto not_found;
6854                 if (start + len <= found_key.offset)
6855                         goto not_found;
6856                 if (start > found_key.offset)
6857                         goto next;
6858                 em->start = start;
6859                 em->orig_start = start;
6860                 em->len = found_key.offset - start;
6861                 goto not_found_em;
6862         }
6863
6864         btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6865
6866         if (found_type == BTRFS_FILE_EXTENT_REG ||
6867             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6868                 goto insert;
6869         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6870                 unsigned long ptr;
6871                 char *map;
6872                 size_t size;
6873                 size_t extent_offset;
6874                 size_t copy_size;
6875
6876                 if (new_inline)
6877                         goto out;
6878
6879                 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6880                 extent_offset = page_offset(page) + pg_offset - extent_start;
6881                 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6882                                   size - extent_offset);
6883                 em->start = extent_start + extent_offset;
6884                 em->len = ALIGN(copy_size, root->sectorsize);
6885                 em->orig_block_len = em->len;
6886                 em->orig_start = em->start;
6887                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6888                 if (create == 0 && !PageUptodate(page)) {
6889                         if (btrfs_file_extent_compression(leaf, item) !=
6890                             BTRFS_COMPRESS_NONE) {
6891                                 ret = uncompress_inline(path, page, pg_offset,
6892                                                         extent_offset, item);
6893                                 if (ret) {
6894                                         err = ret;
6895                                         goto out;
6896                                 }
6897                         } else {
6898                                 map = kmap(page);
6899                                 read_extent_buffer(leaf, map + pg_offset, ptr,
6900                                                    copy_size);
6901                                 if (pg_offset + copy_size < PAGE_SIZE) {
6902                                         memset(map + pg_offset + copy_size, 0,
6903                                                PAGE_SIZE - pg_offset -
6904                                                copy_size);
6905                                 }
6906                                 kunmap(page);
6907                         }
6908                         flush_dcache_page(page);
6909                 } else if (create && PageUptodate(page)) {
6910                         BUG();
6911                         if (!trans) {
6912                                 kunmap(page);
6913                                 free_extent_map(em);
6914                                 em = NULL;
6915
6916                                 btrfs_release_path(path);
6917                                 trans = btrfs_join_transaction(root);
6918
6919                                 if (IS_ERR(trans))
6920                                         return ERR_CAST(trans);
6921                                 goto again;
6922                         }
6923                         map = kmap(page);
6924                         write_extent_buffer(leaf, map + pg_offset, ptr,
6925                                             copy_size);
6926                         kunmap(page);
6927                         btrfs_mark_buffer_dirty(leaf);
6928                 }
6929                 set_extent_uptodate(io_tree, em->start,
6930                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
6931                 goto insert;
6932         }
6933 not_found:
6934         em->start = start;
6935         em->orig_start = start;
6936         em->len = len;
6937 not_found_em:
6938         em->block_start = EXTENT_MAP_HOLE;
6939         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6940 insert:
6941         btrfs_release_path(path);
6942         if (em->start > start || extent_map_end(em) <= start) {
6943                 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6944                         em->start, em->len, start, len);
6945                 err = -EIO;
6946                 goto out;
6947         }
6948
6949         err = 0;
6950         write_lock(&em_tree->lock);
6951         ret = add_extent_mapping(em_tree, em, 0);
6952         /* it is possible that someone inserted the extent into the tree
6953          * while we had the lock dropped.  It is also possible that
6954          * an overlapping map exists in the tree
6955          */
6956         if (ret == -EEXIST) {
6957                 struct extent_map *existing;
6958
6959                 ret = 0;
6960
6961                 existing = search_extent_mapping(em_tree, start, len);
6962                 /*
6963                  * existing will always be non-NULL, since there must be
6964                  * extent causing the -EEXIST.
6965                  */
6966                 if (start >= extent_map_end(existing) ||
6967                     start <= existing->start) {
6968                         /*
6969                          * The existing extent map is the one nearest to
6970                          * the [start, start + len) range which overlaps
6971                          */
6972                         err = merge_extent_mapping(em_tree, existing,
6973                                                    em, start);
6974                         free_extent_map(existing);
6975                         if (err) {
6976                                 free_extent_map(em);
6977                                 em = NULL;
6978                         }
6979                 } else {
6980                         free_extent_map(em);
6981                         em = existing;
6982                         err = 0;
6983                 }
6984         }
6985         write_unlock(&em_tree->lock);
6986 out:
6987
6988         trace_btrfs_get_extent(root, em);
6989
6990         btrfs_free_path(path);
6991         if (trans) {
6992                 ret = btrfs_end_transaction(trans, root);
6993                 if (!err)
6994                         err = ret;
6995         }
6996         if (err) {
6997                 free_extent_map(em);
6998                 return ERR_PTR(err);
6999         }
7000         BUG_ON(!em); /* Error is always set */
7001         return em;
7002 }
7003
7004 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
7005                                            size_t pg_offset, u64 start, u64 len,
7006                                            int create)
7007 {
7008         struct extent_map *em;
7009         struct extent_map *hole_em = NULL;
7010         u64 range_start = start;
7011         u64 end;
7012         u64 found;
7013         u64 found_end;
7014         int err = 0;
7015
7016         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7017         if (IS_ERR(em))
7018                 return em;
7019         if (em) {
7020                 /*
7021                  * if our em maps to
7022                  * -  a hole or
7023                  * -  a pre-alloc extent,
7024                  * there might actually be delalloc bytes behind it.
7025                  */
7026                 if (em->block_start != EXTENT_MAP_HOLE &&
7027                     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7028                         return em;
7029                 else
7030                         hole_em = em;
7031         }
7032
7033         /* check to see if we've wrapped (len == -1 or similar) */
7034         end = start + len;
7035         if (end < start)
7036                 end = (u64)-1;
7037         else
7038                 end -= 1;
7039
7040         em = NULL;
7041
7042         /* ok, we didn't find anything, lets look for delalloc */
7043         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
7044                                  end, len, EXTENT_DELALLOC, 1);
7045         found_end = range_start + found;
7046         if (found_end < range_start)
7047                 found_end = (u64)-1;
7048
7049         /*
7050          * we didn't find anything useful, return
7051          * the original results from get_extent()
7052          */
7053         if (range_start > end || found_end <= start) {
7054                 em = hole_em;
7055                 hole_em = NULL;
7056                 goto out;
7057         }
7058
7059         /* adjust the range_start to make sure it doesn't
7060          * go backwards from the start they passed in
7061          */
7062         range_start = max(start, range_start);
7063         found = found_end - range_start;
7064
7065         if (found > 0) {
7066                 u64 hole_start = start;
7067                 u64 hole_len = len;
7068
7069                 em = alloc_extent_map();
7070                 if (!em) {
7071                         err = -ENOMEM;
7072                         goto out;
7073                 }
7074                 /*
7075                  * when btrfs_get_extent can't find anything it
7076                  * returns one huge hole
7077                  *
7078                  * make sure what it found really fits our range, and
7079                  * adjust to make sure it is based on the start from
7080                  * the caller
7081                  */
7082                 if (hole_em) {
7083                         u64 calc_end = extent_map_end(hole_em);
7084
7085                         if (calc_end <= start || (hole_em->start > end)) {
7086                                 free_extent_map(hole_em);
7087                                 hole_em = NULL;
7088                         } else {
7089                                 hole_start = max(hole_em->start, start);
7090                                 hole_len = calc_end - hole_start;
7091                         }
7092                 }
7093                 em->bdev = NULL;
7094                 if (hole_em && range_start > hole_start) {
7095                         /* our hole starts before our delalloc, so we
7096                          * have to return just the parts of the hole
7097                          * that go until  the delalloc starts
7098                          */
7099                         em->len = min(hole_len,
7100                                       range_start - hole_start);
7101                         em->start = hole_start;
7102                         em->orig_start = hole_start;
7103                         /*
7104                          * don't adjust block start at all,
7105                          * it is fixed at EXTENT_MAP_HOLE
7106                          */
7107                         em->block_start = hole_em->block_start;
7108                         em->block_len = hole_len;
7109                         if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7110                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7111                 } else {
7112                         em->start = range_start;
7113                         em->len = found;
7114                         em->orig_start = range_start;
7115                         em->block_start = EXTENT_MAP_DELALLOC;
7116                         em->block_len = found;
7117                 }
7118         } else if (hole_em) {
7119                 return hole_em;
7120         }
7121 out:
7122
7123         free_extent_map(hole_em);
7124         if (err) {
7125                 free_extent_map(em);
7126                 return ERR_PTR(err);
7127         }
7128         return em;
7129 }
7130
7131 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7132                                                   u64 start, u64 len)
7133 {
7134         struct btrfs_root *root = BTRFS_I(inode)->root;
7135         struct extent_map *em;
7136         struct btrfs_key ins;
7137         u64 alloc_hint;
7138         int ret;
7139
7140         alloc_hint = get_extent_allocation_hint(inode, start, len);
7141         ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
7142                                    alloc_hint, &ins, 1, 1);
7143         if (ret)
7144                 return ERR_PTR(ret);
7145
7146         /*
7147          * Create the ordered extent before the extent map. This is to avoid
7148          * races with the fast fsync path that would lead to it logging file
7149          * extent items that point to disk extents that were not yet written to.
7150          * The fast fsync path collects ordered extents into a local list and
7151          * then collects all the new extent maps, so we must create the ordered
7152          * extent first and make sure the fast fsync path collects any new
7153          * ordered extents after collecting new extent maps as well.
7154          * The fsync path simply can not rely on inode_dio_wait() because it
7155          * causes deadlock with AIO.
7156          */
7157         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
7158                                            ins.offset, ins.offset, 0);
7159         if (ret) {
7160                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7161                 return ERR_PTR(ret);
7162         }
7163
7164         em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
7165                               ins.offset, ins.offset, ins.offset, 0);
7166         if (IS_ERR(em)) {
7167                 struct btrfs_ordered_extent *oe;
7168
7169                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7170                 oe = btrfs_lookup_ordered_extent(inode, start);
7171                 ASSERT(oe);
7172                 if (WARN_ON(!oe))
7173                         return em;
7174                 set_bit(BTRFS_ORDERED_IOERR, &oe->flags);
7175                 set_bit(BTRFS_ORDERED_IO_DONE, &oe->flags);
7176                 btrfs_remove_ordered_extent(inode, oe);
7177                 /* Once for our lookup and once for the ordered extents tree. */
7178                 btrfs_put_ordered_extent(oe);
7179                 btrfs_put_ordered_extent(oe);
7180         }
7181         return em;
7182 }
7183
7184 /*
7185  * returns 1 when the nocow is safe, < 1 on error, 0 if the
7186  * block must be cow'd
7187  */
7188 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7189                               u64 *orig_start, u64 *orig_block_len,
7190                               u64 *ram_bytes)
7191 {
7192         struct btrfs_trans_handle *trans;
7193         struct btrfs_path *path;
7194         int ret;
7195         struct extent_buffer *leaf;
7196         struct btrfs_root *root = BTRFS_I(inode)->root;
7197         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7198         struct btrfs_file_extent_item *fi;
7199         struct btrfs_key key;
7200         u64 disk_bytenr;
7201         u64 backref_offset;
7202         u64 extent_end;
7203         u64 num_bytes;
7204         int slot;
7205         int found_type;
7206         bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7207
7208         path = btrfs_alloc_path();
7209         if (!path)
7210                 return -ENOMEM;
7211
7212         ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
7213                                        offset, 0);
7214         if (ret < 0)
7215                 goto out;
7216
7217         slot = path->slots[0];
7218         if (ret == 1) {
7219                 if (slot == 0) {
7220                         /* can't find the item, must cow */
7221                         ret = 0;
7222                         goto out;
7223                 }
7224                 slot--;
7225         }
7226         ret = 0;
7227         leaf = path->nodes[0];
7228         btrfs_item_key_to_cpu(leaf, &key, slot);
7229         if (key.objectid != btrfs_ino(inode) ||
7230             key.type != BTRFS_EXTENT_DATA_KEY) {
7231                 /* not our file or wrong item type, must cow */
7232                 goto out;
7233         }
7234
7235         if (key.offset > offset) {
7236                 /* Wrong offset, must cow */
7237                 goto out;
7238         }
7239
7240         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7241         found_type = btrfs_file_extent_type(leaf, fi);
7242         if (found_type != BTRFS_FILE_EXTENT_REG &&
7243             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7244                 /* not a regular extent, must cow */
7245                 goto out;
7246         }
7247
7248         if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7249                 goto out;
7250
7251         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7252         if (extent_end <= offset)
7253                 goto out;
7254
7255         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7256         if (disk_bytenr == 0)
7257                 goto out;
7258
7259         if (btrfs_file_extent_compression(leaf, fi) ||
7260             btrfs_file_extent_encryption(leaf, fi) ||
7261             btrfs_file_extent_other_encoding(leaf, fi))
7262                 goto out;
7263
7264         backref_offset = btrfs_file_extent_offset(leaf, fi);
7265
7266         if (orig_start) {
7267                 *orig_start = key.offset - backref_offset;
7268                 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7269                 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7270         }
7271
7272         if (btrfs_extent_readonly(root, disk_bytenr))
7273                 goto out;
7274
7275         num_bytes = min(offset + *len, extent_end) - offset;
7276         if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7277                 u64 range_end;
7278
7279                 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7280                 ret = test_range_bit(io_tree, offset, range_end,
7281                                      EXTENT_DELALLOC, 0, NULL);
7282                 if (ret) {
7283                         ret = -EAGAIN;
7284                         goto out;
7285                 }
7286         }
7287
7288         btrfs_release_path(path);
7289
7290         /*
7291          * look for other files referencing this extent, if we
7292          * find any we must cow
7293          */
7294         trans = btrfs_join_transaction(root);
7295         if (IS_ERR(trans)) {
7296                 ret = 0;
7297                 goto out;
7298         }
7299
7300         ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7301                                     key.offset - backref_offset, disk_bytenr);
7302         btrfs_end_transaction(trans, root);
7303         if (ret) {
7304                 ret = 0;
7305                 goto out;
7306         }
7307
7308         /*
7309          * adjust disk_bytenr and num_bytes to cover just the bytes
7310          * in this extent we are about to write.  If there
7311          * are any csums in that range we have to cow in order
7312          * to keep the csums correct
7313          */
7314         disk_bytenr += backref_offset;
7315         disk_bytenr += offset - key.offset;
7316         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7317                                 goto out;
7318         /*
7319          * all of the above have passed, it is safe to overwrite this extent
7320          * without cow
7321          */
7322         *len = num_bytes;
7323         ret = 1;
7324 out:
7325         btrfs_free_path(path);
7326         return ret;
7327 }
7328
7329 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7330 {
7331         struct radix_tree_root *root = &inode->i_mapping->page_tree;
7332         int found = false;
7333         void **pagep = NULL;
7334         struct page *page = NULL;
7335         int start_idx;
7336         int end_idx;
7337
7338         start_idx = start >> PAGE_SHIFT;
7339
7340         /*
7341          * end is the last byte in the last page.  end == start is legal
7342          */
7343         end_idx = end >> PAGE_SHIFT;
7344
7345         rcu_read_lock();
7346
7347         /* Most of the code in this while loop is lifted from
7348          * find_get_page.  It's been modified to begin searching from a
7349          * page and return just the first page found in that range.  If the
7350          * found idx is less than or equal to the end idx then we know that
7351          * a page exists.  If no pages are found or if those pages are
7352          * outside of the range then we're fine (yay!) */
7353         while (page == NULL &&
7354                radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7355                 page = radix_tree_deref_slot(pagep);
7356                 if (unlikely(!page))
7357                         break;
7358
7359                 if (radix_tree_exception(page)) {
7360                         if (radix_tree_deref_retry(page)) {
7361                                 page = NULL;
7362                                 continue;
7363                         }
7364                         /*
7365                          * Otherwise, shmem/tmpfs must be storing a swap entry
7366                          * here as an exceptional entry: so return it without
7367                          * attempting to raise page count.
7368                          */
7369                         page = NULL;
7370                         break; /* TODO: Is this relevant for this use case? */
7371                 }
7372
7373                 if (!page_cache_get_speculative(page)) {
7374                         page = NULL;
7375                         continue;
7376                 }
7377
7378                 /*
7379                  * Has the page moved?
7380                  * This is part of the lockless pagecache protocol. See
7381                  * include/linux/pagemap.h for details.
7382                  */
7383                 if (unlikely(page != *pagep)) {
7384                         put_page(page);
7385                         page = NULL;
7386                 }
7387         }
7388
7389         if (page) {
7390                 if (page->index <= end_idx)
7391                         found = true;
7392                 put_page(page);
7393         }
7394
7395         rcu_read_unlock();
7396         return found;
7397 }
7398
7399 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7400                               struct extent_state **cached_state, int writing)
7401 {
7402         struct btrfs_ordered_extent *ordered;
7403         int ret = 0;
7404
7405         while (1) {
7406                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7407                                  cached_state);
7408                 /*
7409                  * We're concerned with the entire range that we're going to be
7410                  * doing DIO to, so we need to make sure there's no ordered
7411                  * extents in this range.
7412                  */
7413                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7414                                                      lockend - lockstart + 1);
7415
7416                 /*
7417                  * We need to make sure there are no buffered pages in this
7418                  * range either, we could have raced between the invalidate in
7419                  * generic_file_direct_write and locking the extent.  The
7420                  * invalidate needs to happen so that reads after a write do not
7421                  * get stale data.
7422                  */
7423                 if (!ordered &&
7424                     (!writing ||
7425                      !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7426                         break;
7427
7428                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7429                                      cached_state, GFP_NOFS);
7430
7431                 if (ordered) {
7432                         /*
7433                          * If we are doing a DIO read and the ordered extent we
7434                          * found is for a buffered write, we can not wait for it
7435                          * to complete and retry, because if we do so we can
7436                          * deadlock with concurrent buffered writes on page
7437                          * locks. This happens only if our DIO read covers more
7438                          * than one extent map, if at this point has already
7439                          * created an ordered extent for a previous extent map
7440                          * and locked its range in the inode's io tree, and a
7441                          * concurrent write against that previous extent map's
7442                          * range and this range started (we unlock the ranges
7443                          * in the io tree only when the bios complete and
7444                          * buffered writes always lock pages before attempting
7445                          * to lock range in the io tree).
7446                          */
7447                         if (writing ||
7448                             test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7449                                 btrfs_start_ordered_extent(inode, ordered, 1);
7450                         else
7451                                 ret = -ENOTBLK;
7452                         btrfs_put_ordered_extent(ordered);
7453                 } else {
7454                         /*
7455                          * We could trigger writeback for this range (and wait
7456                          * for it to complete) and then invalidate the pages for
7457                          * this range (through invalidate_inode_pages2_range()),
7458                          * but that can lead us to a deadlock with a concurrent
7459                          * call to readpages() (a buffered read or a defrag call
7460                          * triggered a readahead) on a page lock due to an
7461                          * ordered dio extent we created before but did not have
7462                          * yet a corresponding bio submitted (whence it can not
7463                          * complete), which makes readpages() wait for that
7464                          * ordered extent to complete while holding a lock on
7465                          * that page.
7466                          */
7467                         ret = -ENOTBLK;
7468                 }
7469
7470                 if (ret)
7471                         break;
7472
7473                 cond_resched();
7474         }
7475
7476         return ret;
7477 }
7478
7479 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7480                                            u64 len, u64 orig_start,
7481                                            u64 block_start, u64 block_len,
7482                                            u64 orig_block_len, u64 ram_bytes,
7483                                            int type)
7484 {
7485         struct extent_map_tree *em_tree;
7486         struct extent_map *em;
7487         struct btrfs_root *root = BTRFS_I(inode)->root;
7488         int ret;
7489
7490         em_tree = &BTRFS_I(inode)->extent_tree;
7491         em = alloc_extent_map();
7492         if (!em)
7493                 return ERR_PTR(-ENOMEM);
7494
7495         em->start = start;
7496         em->orig_start = orig_start;
7497         em->mod_start = start;
7498         em->mod_len = len;
7499         em->len = len;
7500         em->block_len = block_len;
7501         em->block_start = block_start;
7502         em->bdev = root->fs_info->fs_devices->latest_bdev;
7503         em->orig_block_len = orig_block_len;
7504         em->ram_bytes = ram_bytes;
7505         em->generation = -1;
7506         set_bit(EXTENT_FLAG_PINNED, &em->flags);
7507         if (type == BTRFS_ORDERED_PREALLOC)
7508                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7509
7510         do {
7511                 btrfs_drop_extent_cache(inode, em->start,
7512                                 em->start + em->len - 1, 0);
7513                 write_lock(&em_tree->lock);
7514                 ret = add_extent_mapping(em_tree, em, 1);
7515                 write_unlock(&em_tree->lock);
7516         } while (ret == -EEXIST);
7517
7518         if (ret) {
7519                 free_extent_map(em);
7520                 return ERR_PTR(ret);
7521         }
7522
7523         return em;
7524 }
7525
7526 static void adjust_dio_outstanding_extents(struct inode *inode,
7527                                            struct btrfs_dio_data *dio_data,
7528                                            const u64 len)
7529 {
7530         unsigned num_extents;
7531
7532         num_extents = (unsigned) div64_u64(len + BTRFS_MAX_EXTENT_SIZE - 1,
7533                                            BTRFS_MAX_EXTENT_SIZE);
7534         /*
7535          * If we have an outstanding_extents count still set then we're
7536          * within our reservation, otherwise we need to adjust our inode
7537          * counter appropriately.
7538          */
7539         if (dio_data->outstanding_extents) {
7540                 dio_data->outstanding_extents -= num_extents;
7541         } else {
7542                 spin_lock(&BTRFS_I(inode)->lock);
7543                 BTRFS_I(inode)->outstanding_extents += num_extents;
7544                 spin_unlock(&BTRFS_I(inode)->lock);
7545         }
7546 }
7547
7548 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7549                                    struct buffer_head *bh_result, int create)
7550 {
7551         struct extent_map *em;
7552         struct btrfs_root *root = BTRFS_I(inode)->root;
7553         struct extent_state *cached_state = NULL;
7554         struct btrfs_dio_data *dio_data = NULL;
7555         u64 start = iblock << inode->i_blkbits;
7556         u64 lockstart, lockend;
7557         u64 len = bh_result->b_size;
7558         int unlock_bits = EXTENT_LOCKED;
7559         int ret = 0;
7560
7561         if (create)
7562                 unlock_bits |= EXTENT_DIRTY;
7563         else
7564                 len = min_t(u64, len, root->sectorsize);
7565
7566         lockstart = start;
7567         lockend = start + len - 1;
7568
7569         if (current->journal_info) {
7570                 /*
7571                  * Need to pull our outstanding extents and set journal_info to NULL so
7572                  * that anything that needs to check if there's a transaction doesn't get
7573                  * confused.
7574                  */
7575                 dio_data = current->journal_info;
7576                 current->journal_info = NULL;
7577         }
7578
7579         /*
7580          * If this errors out it's because we couldn't invalidate pagecache for
7581          * this range and we need to fallback to buffered.
7582          */
7583         if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7584                                create)) {
7585                 ret = -ENOTBLK;
7586                 goto err;
7587         }
7588
7589         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7590         if (IS_ERR(em)) {
7591                 ret = PTR_ERR(em);
7592                 goto unlock_err;
7593         }
7594
7595         /*
7596          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7597          * io.  INLINE is special, and we could probably kludge it in here, but
7598          * it's still buffered so for safety lets just fall back to the generic
7599          * buffered path.
7600          *
7601          * For COMPRESSED we _have_ to read the entire extent in so we can
7602          * decompress it, so there will be buffering required no matter what we
7603          * do, so go ahead and fallback to buffered.
7604          *
7605          * We return -ENOTBLK because that's what makes DIO go ahead and go back
7606          * to buffered IO.  Don't blame me, this is the price we pay for using
7607          * the generic code.
7608          */
7609         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7610             em->block_start == EXTENT_MAP_INLINE) {
7611                 free_extent_map(em);
7612                 ret = -ENOTBLK;
7613                 goto unlock_err;
7614         }
7615
7616         /* Just a good old fashioned hole, return */
7617         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7618                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7619                 free_extent_map(em);
7620                 goto unlock_err;
7621         }
7622
7623         /*
7624          * We don't allocate a new extent in the following cases
7625          *
7626          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
7627          * existing extent.
7628          * 2) The extent is marked as PREALLOC.  We're good to go here and can
7629          * just use the extent.
7630          *
7631          */
7632         if (!create) {
7633                 len = min(len, em->len - (start - em->start));
7634                 lockstart = start + len;
7635                 goto unlock;
7636         }
7637
7638         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7639             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7640              em->block_start != EXTENT_MAP_HOLE)) {
7641                 int type;
7642                 u64 block_start, orig_start, orig_block_len, ram_bytes;
7643
7644                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7645                         type = BTRFS_ORDERED_PREALLOC;
7646                 else
7647                         type = BTRFS_ORDERED_NOCOW;
7648                 len = min(len, em->len - (start - em->start));
7649                 block_start = em->block_start + (start - em->start);
7650
7651                 if (can_nocow_extent(inode, start, &len, &orig_start,
7652                                      &orig_block_len, &ram_bytes) == 1) {
7653                         if (type == BTRFS_ORDERED_PREALLOC) {
7654                                 free_extent_map(em);
7655                                 em = create_pinned_em(inode, start, len,
7656                                                        orig_start,
7657                                                        block_start, len,
7658                                                        orig_block_len,
7659                                                        ram_bytes, type);
7660                                 if (IS_ERR(em)) {
7661                                         ret = PTR_ERR(em);
7662                                         goto unlock_err;
7663                                 }
7664                         }
7665
7666                         ret = btrfs_add_ordered_extent_dio(inode, start,
7667                                            block_start, len, len, type);
7668                         if (ret) {
7669                                 free_extent_map(em);
7670                                 goto unlock_err;
7671                         }
7672                         goto unlock;
7673                 }
7674         }
7675
7676         /*
7677          * this will cow the extent, reset the len in case we changed
7678          * it above
7679          */
7680         len = bh_result->b_size;
7681         free_extent_map(em);
7682         em = btrfs_new_extent_direct(inode, start, len);
7683         if (IS_ERR(em)) {
7684                 ret = PTR_ERR(em);
7685                 goto unlock_err;
7686         }
7687         len = min(len, em->len - (start - em->start));
7688 unlock:
7689         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7690                 inode->i_blkbits;
7691         bh_result->b_size = len;
7692         bh_result->b_bdev = em->bdev;
7693         set_buffer_mapped(bh_result);
7694         if (create) {
7695                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7696                         set_buffer_new(bh_result);
7697
7698                 /*
7699                  * Need to update the i_size under the extent lock so buffered
7700                  * readers will get the updated i_size when we unlock.
7701                  */
7702                 if (start + len > i_size_read(inode))
7703                         i_size_write(inode, start + len);
7704
7705                 adjust_dio_outstanding_extents(inode, dio_data, len);
7706                 btrfs_free_reserved_data_space(inode, start, len);
7707                 WARN_ON(dio_data->reserve < len);
7708                 dio_data->reserve -= len;
7709                 dio_data->unsubmitted_oe_range_end = start + len;
7710                 current->journal_info = dio_data;
7711         }
7712
7713         /*
7714          * In the case of write we need to clear and unlock the entire range,
7715          * in the case of read we need to unlock only the end area that we
7716          * aren't using if there is any left over space.
7717          */
7718         if (lockstart < lockend) {
7719                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7720                                  lockend, unlock_bits, 1, 0,
7721                                  &cached_state, GFP_NOFS);
7722         } else {
7723                 free_extent_state(cached_state);
7724         }
7725
7726         free_extent_map(em);
7727
7728         return 0;
7729
7730 unlock_err:
7731         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7732                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7733 err:
7734         if (dio_data)
7735                 current->journal_info = dio_data;
7736         /*
7737          * Compensate the delalloc release we do in btrfs_direct_IO() when we
7738          * write less data then expected, so that we don't underflow our inode's
7739          * outstanding extents counter.
7740          */
7741         if (create && dio_data)
7742                 adjust_dio_outstanding_extents(inode, dio_data, len);
7743
7744         return ret;
7745 }
7746
7747 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7748                                         int rw, int mirror_num)
7749 {
7750         struct btrfs_root *root = BTRFS_I(inode)->root;
7751         int ret;
7752
7753         BUG_ON(rw & REQ_WRITE);
7754
7755         bio_get(bio);
7756
7757         ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7758                                   BTRFS_WQ_ENDIO_DIO_REPAIR);
7759         if (ret)
7760                 goto err;
7761
7762         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7763 err:
7764         bio_put(bio);
7765         return ret;
7766 }
7767
7768 static int btrfs_check_dio_repairable(struct inode *inode,
7769                                       struct bio *failed_bio,
7770                                       struct io_failure_record *failrec,
7771                                       int failed_mirror)
7772 {
7773         int num_copies;
7774
7775         num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7776                                       failrec->logical, failrec->len);
7777         if (num_copies == 1) {
7778                 /*
7779                  * we only have a single copy of the data, so don't bother with
7780                  * all the retry and error correction code that follows. no
7781                  * matter what the error is, it is very likely to persist.
7782                  */
7783                 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7784                          num_copies, failrec->this_mirror, failed_mirror);
7785                 return 0;
7786         }
7787
7788         failrec->failed_mirror = failed_mirror;
7789         failrec->this_mirror++;
7790         if (failrec->this_mirror == failed_mirror)
7791                 failrec->this_mirror++;
7792
7793         if (failrec->this_mirror > num_copies) {
7794                 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7795                          num_copies, failrec->this_mirror, failed_mirror);
7796                 return 0;
7797         }
7798
7799         return 1;
7800 }
7801
7802 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7803                         struct page *page, unsigned int pgoff,
7804                         u64 start, u64 end, int failed_mirror,
7805                         bio_end_io_t *repair_endio, void *repair_arg)
7806 {
7807         struct io_failure_record *failrec;
7808         struct bio *bio;
7809         int isector;
7810         int read_mode;
7811         int ret;
7812
7813         BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7814
7815         ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7816         if (ret)
7817                 return ret;
7818
7819         ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7820                                          failed_mirror);
7821         if (!ret) {
7822                 free_io_failure(inode, failrec);
7823                 return -EIO;
7824         }
7825
7826         if ((failed_bio->bi_vcnt > 1)
7827                 || (failed_bio->bi_io_vec->bv_len
7828                         > BTRFS_I(inode)->root->sectorsize))
7829                 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7830         else
7831                 read_mode = READ_SYNC;
7832
7833         isector = start - btrfs_io_bio(failed_bio)->logical;
7834         isector >>= inode->i_sb->s_blocksize_bits;
7835         bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7836                                 pgoff, isector, repair_endio, repair_arg);
7837         if (!bio) {
7838                 free_io_failure(inode, failrec);
7839                 return -EIO;
7840         }
7841
7842         btrfs_debug(BTRFS_I(inode)->root->fs_info,
7843                     "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7844                     read_mode, failrec->this_mirror, failrec->in_validation);
7845
7846         ret = submit_dio_repair_bio(inode, bio, read_mode,
7847                                     failrec->this_mirror);
7848         if (ret) {
7849                 free_io_failure(inode, failrec);
7850                 bio_put(bio);
7851         }
7852
7853         return ret;
7854 }
7855
7856 struct btrfs_retry_complete {
7857         struct completion done;
7858         struct inode *inode;
7859         u64 start;
7860         int uptodate;
7861 };
7862
7863 static void btrfs_retry_endio_nocsum(struct bio *bio)
7864 {
7865         struct btrfs_retry_complete *done = bio->bi_private;
7866         struct inode *inode;
7867         struct bio_vec *bvec;
7868         int i;
7869
7870         if (bio->bi_error)
7871                 goto end;
7872
7873         ASSERT(bio->bi_vcnt == 1);
7874         inode = bio->bi_io_vec->bv_page->mapping->host;
7875         ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
7876
7877         done->uptodate = 1;
7878         bio_for_each_segment_all(bvec, bio, i)
7879                 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7880 end:
7881         complete(&done->done);
7882         bio_put(bio);
7883 }
7884
7885 static int __btrfs_correct_data_nocsum(struct inode *inode,
7886                                        struct btrfs_io_bio *io_bio)
7887 {
7888         struct btrfs_fs_info *fs_info;
7889         struct bio_vec *bvec;
7890         struct btrfs_retry_complete done;
7891         u64 start;
7892         unsigned int pgoff;
7893         u32 sectorsize;
7894         int nr_sectors;
7895         int i;
7896         int ret;
7897
7898         fs_info = BTRFS_I(inode)->root->fs_info;
7899         sectorsize = BTRFS_I(inode)->root->sectorsize;
7900
7901         start = io_bio->logical;
7902         done.inode = inode;
7903
7904         bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7905                 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
7906                 pgoff = bvec->bv_offset;
7907
7908 next_block_or_try_again:
7909                 done.uptodate = 0;
7910                 done.start = start;
7911                 init_completion(&done.done);
7912
7913                 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
7914                                 pgoff, start, start + sectorsize - 1,
7915                                 io_bio->mirror_num,
7916                                 btrfs_retry_endio_nocsum, &done);
7917                 if (ret)
7918                         return ret;
7919
7920                 wait_for_completion(&done.done);
7921
7922                 if (!done.uptodate) {
7923                         /* We might have another mirror, so try again */
7924                         goto next_block_or_try_again;
7925                 }
7926
7927                 start += sectorsize;
7928
7929                 if (nr_sectors--) {
7930                         pgoff += sectorsize;
7931                         goto next_block_or_try_again;
7932                 }
7933         }
7934
7935         return 0;
7936 }
7937
7938 static void btrfs_retry_endio(struct bio *bio)
7939 {
7940         struct btrfs_retry_complete *done = bio->bi_private;
7941         struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7942         struct inode *inode;
7943         struct bio_vec *bvec;
7944         u64 start;
7945         int uptodate;
7946         int ret;
7947         int i;
7948
7949         if (bio->bi_error)
7950                 goto end;
7951
7952         uptodate = 1;
7953
7954         start = done->start;
7955
7956         ASSERT(bio->bi_vcnt == 1);
7957         inode = bio->bi_io_vec->bv_page->mapping->host;
7958         ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
7959
7960         bio_for_each_segment_all(bvec, bio, i) {
7961                 ret = __readpage_endio_check(done->inode, io_bio, i,
7962                                         bvec->bv_page, bvec->bv_offset,
7963                                         done->start, bvec->bv_len);
7964                 if (!ret)
7965                         clean_io_failure(done->inode, done->start,
7966                                         bvec->bv_page, bvec->bv_offset);
7967                 else
7968                         uptodate = 0;
7969         }
7970
7971         done->uptodate = uptodate;
7972 end:
7973         complete(&done->done);
7974         bio_put(bio);
7975 }
7976
7977 static int __btrfs_subio_endio_read(struct inode *inode,
7978                                     struct btrfs_io_bio *io_bio, int err)
7979 {
7980         struct btrfs_fs_info *fs_info;
7981         struct bio_vec *bvec;
7982         struct btrfs_retry_complete done;
7983         u64 start;
7984         u64 offset = 0;
7985         u32 sectorsize;
7986         int nr_sectors;
7987         unsigned int pgoff;
7988         int csum_pos;
7989         int i;
7990         int ret;
7991
7992         fs_info = BTRFS_I(inode)->root->fs_info;
7993         sectorsize = BTRFS_I(inode)->root->sectorsize;
7994
7995         err = 0;
7996         start = io_bio->logical;
7997         done.inode = inode;
7998
7999         bio_for_each_segment_all(bvec, &io_bio->bio, i) {
8000                 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
8001
8002                 pgoff = bvec->bv_offset;
8003 next_block:
8004                 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8005                 ret = __readpage_endio_check(inode, io_bio, csum_pos,
8006                                         bvec->bv_page, pgoff, start,
8007                                         sectorsize);
8008                 if (likely(!ret))
8009                         goto next;
8010 try_again:
8011                 done.uptodate = 0;
8012                 done.start = start;
8013                 init_completion(&done.done);
8014
8015                 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
8016                                 pgoff, start, start + sectorsize - 1,
8017                                 io_bio->mirror_num,
8018                                 btrfs_retry_endio, &done);
8019                 if (ret) {
8020                         err = ret;
8021                         goto next;
8022                 }
8023
8024                 wait_for_completion(&done.done);
8025
8026                 if (!done.uptodate) {
8027                         /* We might have another mirror, so try again */
8028                         goto try_again;
8029                 }
8030 next:
8031                 offset += sectorsize;
8032                 start += sectorsize;
8033
8034                 ASSERT(nr_sectors);
8035
8036                 if (--nr_sectors) {
8037                         pgoff += sectorsize;
8038                         goto next_block;
8039                 }
8040         }
8041
8042         return err;
8043 }
8044
8045 static int btrfs_subio_endio_read(struct inode *inode,
8046                                   struct btrfs_io_bio *io_bio, int err)
8047 {
8048         bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8049
8050         if (skip_csum) {
8051                 if (unlikely(err))
8052                         return __btrfs_correct_data_nocsum(inode, io_bio);
8053                 else
8054                         return 0;
8055         } else {
8056                 return __btrfs_subio_endio_read(inode, io_bio, err);
8057         }
8058 }
8059
8060 static void btrfs_endio_direct_read(struct bio *bio)
8061 {
8062         struct btrfs_dio_private *dip = bio->bi_private;
8063         struct inode *inode = dip->inode;
8064         struct bio *dio_bio;
8065         struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8066         int err = bio->bi_error;
8067
8068         if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8069                 err = btrfs_subio_endio_read(inode, io_bio, err);
8070
8071         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8072                       dip->logical_offset + dip->bytes - 1);
8073         dio_bio = dip->dio_bio;
8074
8075         kfree(dip);
8076
8077         dio_bio->bi_error = bio->bi_error;
8078         dio_end_io(dio_bio, bio->bi_error);
8079
8080         if (io_bio->end_io)
8081                 io_bio->end_io(io_bio, err);
8082         bio_put(bio);
8083 }
8084
8085 static void btrfs_endio_direct_write_update_ordered(struct inode *inode,
8086                                                     const u64 offset,
8087                                                     const u64 bytes,
8088                                                     const int uptodate)
8089 {
8090         struct btrfs_root *root = BTRFS_I(inode)->root;
8091         struct btrfs_ordered_extent *ordered = NULL;
8092         u64 ordered_offset = offset;
8093         u64 ordered_bytes = bytes;
8094         int ret;
8095
8096 again:
8097         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8098                                                    &ordered_offset,
8099                                                    ordered_bytes,
8100                                                    uptodate);
8101         if (!ret)
8102                 goto out_test;
8103
8104         btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
8105                         finish_ordered_fn, NULL, NULL);
8106         btrfs_queue_work(root->fs_info->endio_write_workers,
8107                          &ordered->work);
8108 out_test:
8109         /*
8110          * our bio might span multiple ordered extents.  If we haven't
8111          * completed the accounting for the whole dio, go back and try again
8112          */
8113         if (ordered_offset < offset + bytes) {
8114                 ordered_bytes = offset + bytes - ordered_offset;
8115                 ordered = NULL;
8116                 goto again;
8117         }
8118 }
8119
8120 static void btrfs_endio_direct_write(struct bio *bio)
8121 {
8122         struct btrfs_dio_private *dip = bio->bi_private;
8123         struct bio *dio_bio = dip->dio_bio;
8124
8125         btrfs_endio_direct_write_update_ordered(dip->inode,
8126                                                 dip->logical_offset,
8127                                                 dip->bytes,
8128                                                 !bio->bi_error);
8129
8130         kfree(dip);
8131
8132         dio_bio->bi_error = bio->bi_error;
8133         dio_end_io(dio_bio, bio->bi_error);
8134         bio_put(bio);
8135 }
8136
8137 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
8138                                     struct bio *bio, int mirror_num,
8139                                     unsigned long bio_flags, u64 offset)
8140 {
8141         int ret;
8142         struct btrfs_root *root = BTRFS_I(inode)->root;
8143         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
8144         BUG_ON(ret); /* -ENOMEM */
8145         return 0;
8146 }
8147
8148 static void btrfs_end_dio_bio(struct bio *bio)
8149 {
8150         struct btrfs_dio_private *dip = bio->bi_private;
8151         int err = bio->bi_error;
8152
8153         if (err)
8154                 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8155                            "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
8156                            btrfs_ino(dip->inode), bio->bi_rw,
8157                            (unsigned long long)bio->bi_iter.bi_sector,
8158                            bio->bi_iter.bi_size, err);
8159
8160         if (dip->subio_endio)
8161                 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8162
8163         if (err) {
8164                 dip->errors = 1;
8165
8166                 /*
8167                  * before atomic variable goto zero, we must make sure
8168                  * dip->errors is perceived to be set.
8169                  */
8170                 smp_mb__before_atomic();
8171         }
8172
8173         /* if there are more bios still pending for this dio, just exit */
8174         if (!atomic_dec_and_test(&dip->pending_bios))
8175                 goto out;
8176
8177         if (dip->errors) {
8178                 bio_io_error(dip->orig_bio);
8179         } else {
8180                 dip->dio_bio->bi_error = 0;
8181                 bio_endio(dip->orig_bio);
8182         }
8183 out:
8184         bio_put(bio);
8185 }
8186
8187 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8188                                        u64 first_sector, gfp_t gfp_flags)
8189 {
8190         struct bio *bio;
8191         bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
8192         if (bio)
8193                 bio_associate_current(bio);
8194         return bio;
8195 }
8196
8197 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
8198                                                  struct inode *inode,
8199                                                  struct btrfs_dio_private *dip,
8200                                                  struct bio *bio,
8201                                                  u64 file_offset)
8202 {
8203         struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8204         struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8205         int ret;
8206
8207         /*
8208          * We load all the csum data we need when we submit
8209          * the first bio to reduce the csum tree search and
8210          * contention.
8211          */
8212         if (dip->logical_offset == file_offset) {
8213                 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
8214                                                 file_offset);
8215                 if (ret)
8216                         return ret;
8217         }
8218
8219         if (bio == dip->orig_bio)
8220                 return 0;
8221
8222         file_offset -= dip->logical_offset;
8223         file_offset >>= inode->i_sb->s_blocksize_bits;
8224         io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8225
8226         return 0;
8227 }
8228
8229 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8230                                          int rw, u64 file_offset, int skip_sum,
8231                                          int async_submit)
8232 {
8233         struct btrfs_dio_private *dip = bio->bi_private;
8234         int write = rw & REQ_WRITE;
8235         struct btrfs_root *root = BTRFS_I(inode)->root;
8236         int ret;
8237
8238         if (async_submit)
8239                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8240
8241         bio_get(bio);
8242
8243         if (!write) {
8244                 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
8245                                 BTRFS_WQ_ENDIO_DATA);
8246                 if (ret)
8247                         goto err;
8248         }
8249
8250         if (skip_sum)
8251                 goto map;
8252
8253         if (write && async_submit) {
8254                 ret = btrfs_wq_submit_bio(root->fs_info,
8255                                    inode, rw, bio, 0, 0,
8256                                    file_offset,
8257                                    __btrfs_submit_bio_start_direct_io,
8258                                    __btrfs_submit_bio_done);
8259                 goto err;
8260         } else if (write) {
8261                 /*
8262                  * If we aren't doing async submit, calculate the csum of the
8263                  * bio now.
8264                  */
8265                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
8266                 if (ret)
8267                         goto err;
8268         } else {
8269                 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
8270                                                      file_offset);
8271                 if (ret)
8272                         goto err;
8273         }
8274 map:
8275         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
8276 err:
8277         bio_put(bio);
8278         return ret;
8279 }
8280
8281 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
8282                                     int skip_sum)
8283 {
8284         struct inode *inode = dip->inode;
8285         struct btrfs_root *root = BTRFS_I(inode)->root;
8286         struct bio *bio;
8287         struct bio *orig_bio = dip->orig_bio;
8288         struct bio_vec *bvec = orig_bio->bi_io_vec;
8289         u64 start_sector = orig_bio->bi_iter.bi_sector;
8290         u64 file_offset = dip->logical_offset;
8291         u64 submit_len = 0;
8292         u64 map_length;
8293         u32 blocksize = root->sectorsize;
8294         int async_submit = 0;
8295         int nr_sectors;
8296         int ret;
8297         int i;
8298
8299         map_length = orig_bio->bi_iter.bi_size;
8300         ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
8301                               &map_length, NULL, 0);
8302         if (ret)
8303                 return -EIO;
8304
8305         if (map_length >= orig_bio->bi_iter.bi_size) {
8306                 bio = orig_bio;
8307                 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8308                 goto submit;
8309         }
8310
8311         /* async crcs make it difficult to collect full stripe writes. */
8312         if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8313                 async_submit = 0;
8314         else
8315                 async_submit = 1;
8316
8317         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8318         if (!bio)
8319                 return -ENOMEM;
8320
8321         bio->bi_private = dip;
8322         bio->bi_end_io = btrfs_end_dio_bio;
8323         btrfs_io_bio(bio)->logical = file_offset;
8324         atomic_inc(&dip->pending_bios);
8325
8326         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
8327                 nr_sectors = BTRFS_BYTES_TO_BLKS(root->fs_info, bvec->bv_len);
8328                 i = 0;
8329 next_block:
8330                 if (unlikely(map_length < submit_len + blocksize ||
8331                     bio_add_page(bio, bvec->bv_page, blocksize,
8332                             bvec->bv_offset + (i * blocksize)) < blocksize)) {
8333                         /*
8334                          * inc the count before we submit the bio so
8335                          * we know the end IO handler won't happen before
8336                          * we inc the count. Otherwise, the dip might get freed
8337                          * before we're done setting it up
8338                          */
8339                         atomic_inc(&dip->pending_bios);
8340                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
8341                                                      file_offset, skip_sum,
8342                                                      async_submit);
8343                         if (ret) {
8344                                 bio_put(bio);
8345                                 atomic_dec(&dip->pending_bios);
8346                                 goto out_err;
8347                         }
8348
8349                         start_sector += submit_len >> 9;
8350                         file_offset += submit_len;
8351
8352                         submit_len = 0;
8353
8354                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8355                                                   start_sector, GFP_NOFS);
8356                         if (!bio)
8357                                 goto out_err;
8358                         bio->bi_private = dip;
8359                         bio->bi_end_io = btrfs_end_dio_bio;
8360                         btrfs_io_bio(bio)->logical = file_offset;
8361
8362                         map_length = orig_bio->bi_iter.bi_size;
8363                         ret = btrfs_map_block(root->fs_info, rw,
8364                                               start_sector << 9,
8365                                               &map_length, NULL, 0);
8366                         if (ret) {
8367                                 bio_put(bio);
8368                                 goto out_err;
8369                         }
8370
8371                         goto next_block;
8372                 } else {
8373                         submit_len += blocksize;
8374                         if (--nr_sectors) {
8375                                 i++;
8376                                 goto next_block;
8377                         }
8378                         bvec++;
8379                 }
8380         }
8381
8382 submit:
8383         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
8384                                      async_submit);
8385         if (!ret)
8386                 return 0;
8387
8388         bio_put(bio);
8389 out_err:
8390         dip->errors = 1;
8391         /*
8392          * before atomic variable goto zero, we must
8393          * make sure dip->errors is perceived to be set.
8394          */
8395         smp_mb__before_atomic();
8396         if (atomic_dec_and_test(&dip->pending_bios))
8397                 bio_io_error(dip->orig_bio);
8398
8399         /* bio_end_io() will handle error, so we needn't return it */
8400         return 0;
8401 }
8402
8403 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
8404                                 struct inode *inode, loff_t file_offset)
8405 {
8406         struct btrfs_dio_private *dip = NULL;
8407         struct bio *io_bio = NULL;
8408         struct btrfs_io_bio *btrfs_bio;
8409         int skip_sum;
8410         int write = rw & REQ_WRITE;
8411         int ret = 0;
8412
8413         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8414
8415         io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
8416         if (!io_bio) {
8417                 ret = -ENOMEM;
8418                 goto free_ordered;
8419         }
8420
8421         dip = kzalloc(sizeof(*dip), GFP_NOFS);
8422         if (!dip) {
8423                 ret = -ENOMEM;
8424                 goto free_ordered;
8425         }
8426
8427         dip->private = dio_bio->bi_private;
8428         dip->inode = inode;
8429         dip->logical_offset = file_offset;
8430         dip->bytes = dio_bio->bi_iter.bi_size;
8431         dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8432         io_bio->bi_private = dip;
8433         dip->orig_bio = io_bio;
8434         dip->dio_bio = dio_bio;
8435         atomic_set(&dip->pending_bios, 0);
8436         btrfs_bio = btrfs_io_bio(io_bio);
8437         btrfs_bio->logical = file_offset;
8438
8439         if (write) {
8440                 io_bio->bi_end_io = btrfs_endio_direct_write;
8441         } else {
8442                 io_bio->bi_end_io = btrfs_endio_direct_read;
8443                 dip->subio_endio = btrfs_subio_endio_read;
8444         }
8445
8446         /*
8447          * Reset the range for unsubmitted ordered extents (to a 0 length range)
8448          * even if we fail to submit a bio, because in such case we do the
8449          * corresponding error handling below and it must not be done a second
8450          * time by btrfs_direct_IO().
8451          */
8452         if (write) {
8453                 struct btrfs_dio_data *dio_data = current->journal_info;
8454
8455                 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8456                         dip->bytes;
8457                 dio_data->unsubmitted_oe_range_start =
8458                         dio_data->unsubmitted_oe_range_end;
8459         }
8460
8461         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8462         if (!ret)
8463                 return;
8464
8465         if (btrfs_bio->end_io)
8466                 btrfs_bio->end_io(btrfs_bio, ret);
8467
8468 free_ordered:
8469         /*
8470          * If we arrived here it means either we failed to submit the dip
8471          * or we either failed to clone the dio_bio or failed to allocate the
8472          * dip. If we cloned the dio_bio and allocated the dip, we can just
8473          * call bio_endio against our io_bio so that we get proper resource
8474          * cleanup if we fail to submit the dip, otherwise, we must do the
8475          * same as btrfs_endio_direct_[write|read] because we can't call these
8476          * callbacks - they require an allocated dip and a clone of dio_bio.
8477          */
8478         if (io_bio && dip) {
8479                 io_bio->bi_error = -EIO;
8480                 bio_endio(io_bio);
8481                 /*
8482                  * The end io callbacks free our dip, do the final put on io_bio
8483                  * and all the cleanup and final put for dio_bio (through
8484                  * dio_end_io()).
8485                  */
8486                 dip = NULL;
8487                 io_bio = NULL;
8488         } else {
8489                 if (write)
8490                         btrfs_endio_direct_write_update_ordered(inode,
8491                                                 file_offset,
8492                                                 dio_bio->bi_iter.bi_size,
8493                                                 0);
8494                 else
8495                         unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8496                               file_offset + dio_bio->bi_iter.bi_size - 1);
8497
8498                 dio_bio->bi_error = -EIO;
8499                 /*
8500                  * Releases and cleans up our dio_bio, no need to bio_put()
8501                  * nor bio_endio()/bio_io_error() against dio_bio.
8502                  */
8503                 dio_end_io(dio_bio, ret);
8504         }
8505         if (io_bio)
8506                 bio_put(io_bio);
8507         kfree(dip);
8508 }
8509
8510 static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb,
8511                         const struct iov_iter *iter, loff_t offset)
8512 {
8513         int seg;
8514         int i;
8515         unsigned blocksize_mask = root->sectorsize - 1;
8516         ssize_t retval = -EINVAL;
8517
8518         if (offset & blocksize_mask)
8519                 goto out;
8520
8521         if (iov_iter_alignment(iter) & blocksize_mask)
8522                 goto out;
8523
8524         /* If this is a write we don't need to check anymore */
8525         if (iov_iter_rw(iter) == WRITE)
8526                 return 0;
8527         /*
8528          * Check to make sure we don't have duplicate iov_base's in this
8529          * iovec, if so return EINVAL, otherwise we'll get csum errors
8530          * when reading back.
8531          */
8532         for (seg = 0; seg < iter->nr_segs; seg++) {
8533                 for (i = seg + 1; i < iter->nr_segs; i++) {
8534                         if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8535                                 goto out;
8536                 }
8537         }
8538         retval = 0;
8539 out:
8540         return retval;
8541 }
8542
8543 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
8544                                loff_t offset)
8545 {
8546         struct file *file = iocb->ki_filp;
8547         struct inode *inode = file->f_mapping->host;
8548         struct btrfs_root *root = BTRFS_I(inode)->root;
8549         struct btrfs_dio_data dio_data = { 0 };
8550         size_t count = 0;
8551         int flags = 0;
8552         bool wakeup = true;
8553         bool relock = false;
8554         ssize_t ret;
8555
8556         if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset))
8557                 return 0;
8558
8559         inode_dio_begin(inode);
8560         smp_mb__after_atomic();
8561
8562         /*
8563          * The generic stuff only does filemap_write_and_wait_range, which
8564          * isn't enough if we've written compressed pages to this area, so
8565          * we need to flush the dirty pages again to make absolutely sure
8566          * that any outstanding dirty pages are on disk.
8567          */
8568         count = iov_iter_count(iter);
8569         if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8570                      &BTRFS_I(inode)->runtime_flags))
8571                 filemap_fdatawrite_range(inode->i_mapping, offset,
8572                                          offset + count - 1);
8573
8574         if (iov_iter_rw(iter) == WRITE) {
8575                 /*
8576                  * If the write DIO is beyond the EOF, we need update
8577                  * the isize, but it is protected by i_mutex. So we can
8578                  * not unlock the i_mutex at this case.
8579                  */
8580                 if (offset + count <= inode->i_size) {
8581                         inode_unlock(inode);
8582                         relock = true;
8583                 }
8584                 ret = btrfs_delalloc_reserve_space(inode, offset, count);
8585                 if (ret)
8586                         goto out;
8587                 dio_data.outstanding_extents = div64_u64(count +
8588                                                 BTRFS_MAX_EXTENT_SIZE - 1,
8589                                                 BTRFS_MAX_EXTENT_SIZE);
8590
8591                 /*
8592                  * We need to know how many extents we reserved so that we can
8593                  * do the accounting properly if we go over the number we
8594                  * originally calculated.  Abuse current->journal_info for this.
8595                  */
8596                 dio_data.reserve = round_up(count, root->sectorsize);
8597                 dio_data.unsubmitted_oe_range_start = (u64)offset;
8598                 dio_data.unsubmitted_oe_range_end = (u64)offset;
8599                 current->journal_info = &dio_data;
8600         } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8601                                      &BTRFS_I(inode)->runtime_flags)) {
8602                 inode_dio_end(inode);
8603                 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8604                 wakeup = false;
8605         }
8606
8607         ret = __blockdev_direct_IO(iocb, inode,
8608                                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8609                                    iter, offset, btrfs_get_blocks_direct, NULL,
8610                                    btrfs_submit_direct, flags);
8611         if (iov_iter_rw(iter) == WRITE) {
8612                 current->journal_info = NULL;
8613                 if (ret < 0 && ret != -EIOCBQUEUED) {
8614                         if (dio_data.reserve)
8615                                 btrfs_delalloc_release_space(inode, offset,
8616                                                              dio_data.reserve);
8617                         /*
8618                          * On error we might have left some ordered extents
8619                          * without submitting corresponding bios for them, so
8620                          * cleanup them up to avoid other tasks getting them
8621                          * and waiting for them to complete forever.
8622                          */
8623                         if (dio_data.unsubmitted_oe_range_start <
8624                             dio_data.unsubmitted_oe_range_end)
8625                                 btrfs_endio_direct_write_update_ordered(inode,
8626                                         dio_data.unsubmitted_oe_range_start,
8627                                         dio_data.unsubmitted_oe_range_end -
8628                                         dio_data.unsubmitted_oe_range_start,
8629                                         0);
8630                 } else if (ret >= 0 && (size_t)ret < count)
8631                         btrfs_delalloc_release_space(inode, offset,
8632                                                      count - (size_t)ret);
8633         }
8634 out:
8635         if (wakeup)
8636                 inode_dio_end(inode);
8637         if (relock)
8638                 inode_lock(inode);
8639
8640         return ret;
8641 }
8642
8643 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
8644
8645 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8646                 __u64 start, __u64 len)
8647 {
8648         int     ret;
8649
8650         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8651         if (ret)
8652                 return ret;
8653
8654         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8655 }
8656
8657 int btrfs_readpage(struct file *file, struct page *page)
8658 {
8659         struct extent_io_tree *tree;
8660         tree = &BTRFS_I(page->mapping->host)->io_tree;
8661         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8662 }
8663
8664 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8665 {
8666         struct extent_io_tree *tree;
8667         struct inode *inode = page->mapping->host;
8668         int ret;
8669
8670         if (current->flags & PF_MEMALLOC) {
8671                 redirty_page_for_writepage(wbc, page);
8672                 unlock_page(page);
8673                 return 0;
8674         }
8675
8676         /*
8677          * If we are under memory pressure we will call this directly from the
8678          * VM, we need to make sure we have the inode referenced for the ordered
8679          * extent.  If not just return like we didn't do anything.
8680          */
8681         if (!igrab(inode)) {
8682                 redirty_page_for_writepage(wbc, page);
8683                 return AOP_WRITEPAGE_ACTIVATE;
8684         }
8685         tree = &BTRFS_I(page->mapping->host)->io_tree;
8686         ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8687         btrfs_add_delayed_iput(inode);
8688         return ret;
8689 }
8690
8691 static int btrfs_writepages(struct address_space *mapping,
8692                             struct writeback_control *wbc)
8693 {
8694         struct extent_io_tree *tree;
8695
8696         tree = &BTRFS_I(mapping->host)->io_tree;
8697         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8698 }
8699
8700 static int
8701 btrfs_readpages(struct file *file, struct address_space *mapping,
8702                 struct list_head *pages, unsigned nr_pages)
8703 {
8704         struct extent_io_tree *tree;
8705         tree = &BTRFS_I(mapping->host)->io_tree;
8706         return extent_readpages(tree, mapping, pages, nr_pages,
8707                                 btrfs_get_extent);
8708 }
8709 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8710 {
8711         struct extent_io_tree *tree;
8712         struct extent_map_tree *map;
8713         int ret;
8714
8715         tree = &BTRFS_I(page->mapping->host)->io_tree;
8716         map = &BTRFS_I(page->mapping->host)->extent_tree;
8717         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8718         if (ret == 1) {
8719                 ClearPagePrivate(page);
8720                 set_page_private(page, 0);
8721                 put_page(page);
8722         }
8723         return ret;
8724 }
8725
8726 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8727 {
8728         if (PageWriteback(page) || PageDirty(page))
8729                 return 0;
8730         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8731 }
8732
8733 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8734                                  unsigned int length)
8735 {
8736         struct inode *inode = page->mapping->host;
8737         struct extent_io_tree *tree;
8738         struct btrfs_ordered_extent *ordered;
8739         struct extent_state *cached_state = NULL;
8740         u64 page_start = page_offset(page);
8741         u64 page_end = page_start + PAGE_SIZE - 1;
8742         u64 start;
8743         u64 end;
8744         int inode_evicting = inode->i_state & I_FREEING;
8745
8746         /*
8747          * we have the page locked, so new writeback can't start,
8748          * and the dirty bit won't be cleared while we are here.
8749          *
8750          * Wait for IO on this page so that we can safely clear
8751          * the PagePrivate2 bit and do ordered accounting
8752          */
8753         wait_on_page_writeback(page);
8754
8755         tree = &BTRFS_I(inode)->io_tree;
8756         if (offset) {
8757                 btrfs_releasepage(page, GFP_NOFS);
8758                 return;
8759         }
8760
8761         if (!inode_evicting)
8762                 lock_extent_bits(tree, page_start, page_end, &cached_state);
8763 again:
8764         start = page_start;
8765         ordered = btrfs_lookup_ordered_range(inode, start,
8766                                         page_end - start + 1);
8767         if (ordered) {
8768                 end = min(page_end, ordered->file_offset + ordered->len - 1);
8769                 /*
8770                  * IO on this page will never be started, so we need
8771                  * to account for any ordered extents now
8772                  */
8773                 if (!inode_evicting)
8774                         clear_extent_bit(tree, start, end,
8775                                          EXTENT_DIRTY | EXTENT_DELALLOC |
8776                                          EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8777                                          EXTENT_DEFRAG, 1, 0, &cached_state,
8778                                          GFP_NOFS);
8779                 /*
8780                  * whoever cleared the private bit is responsible
8781                  * for the finish_ordered_io
8782                  */
8783                 if (TestClearPagePrivate2(page)) {
8784                         struct btrfs_ordered_inode_tree *tree;
8785                         u64 new_len;
8786
8787                         tree = &BTRFS_I(inode)->ordered_tree;
8788
8789                         spin_lock_irq(&tree->lock);
8790                         set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8791                         new_len = start - ordered->file_offset;
8792                         if (new_len < ordered->truncated_len)
8793                                 ordered->truncated_len = new_len;
8794                         spin_unlock_irq(&tree->lock);
8795
8796                         if (btrfs_dec_test_ordered_pending(inode, &ordered,
8797                                                            start,
8798                                                            end - start + 1, 1))
8799                                 btrfs_finish_ordered_io(ordered);
8800                 }
8801                 btrfs_put_ordered_extent(ordered);
8802                 if (!inode_evicting) {
8803                         cached_state = NULL;
8804                         lock_extent_bits(tree, start, end,
8805                                          &cached_state);
8806                 }
8807
8808                 start = end + 1;
8809                 if (start < page_end)
8810                         goto again;
8811         }
8812
8813         /*
8814          * Qgroup reserved space handler
8815          * Page here will be either
8816          * 1) Already written to disk
8817          *    In this case, its reserved space is released from data rsv map
8818          *    and will be freed by delayed_ref handler finally.
8819          *    So even we call qgroup_free_data(), it won't decrease reserved
8820          *    space.
8821          * 2) Not written to disk
8822          *    This means the reserved space should be freed here.
8823          */
8824         btrfs_qgroup_free_data(inode, page_start, PAGE_SIZE);
8825         if (!inode_evicting) {
8826                 clear_extent_bit(tree, page_start, page_end,
8827                                  EXTENT_LOCKED | EXTENT_DIRTY |
8828                                  EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8829                                  EXTENT_DEFRAG, 1, 1,
8830                                  &cached_state, GFP_NOFS);
8831
8832                 __btrfs_releasepage(page, GFP_NOFS);
8833         }
8834
8835         ClearPageChecked(page);
8836         if (PagePrivate(page)) {
8837                 ClearPagePrivate(page);
8838                 set_page_private(page, 0);
8839                 put_page(page);
8840         }
8841 }
8842
8843 /*
8844  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8845  * called from a page fault handler when a page is first dirtied. Hence we must
8846  * be careful to check for EOF conditions here. We set the page up correctly
8847  * for a written page which means we get ENOSPC checking when writing into
8848  * holes and correct delalloc and unwritten extent mapping on filesystems that
8849  * support these features.
8850  *
8851  * We are not allowed to take the i_mutex here so we have to play games to
8852  * protect against truncate races as the page could now be beyond EOF.  Because
8853  * vmtruncate() writes the inode size before removing pages, once we have the
8854  * page lock we can determine safely if the page is beyond EOF. If it is not
8855  * beyond EOF, then the page is guaranteed safe against truncation until we
8856  * unlock the page.
8857  */
8858 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8859 {
8860         struct page *page = vmf->page;
8861         struct inode *inode = file_inode(vma->vm_file);
8862         struct btrfs_root *root = BTRFS_I(inode)->root;
8863         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8864         struct btrfs_ordered_extent *ordered;
8865         struct extent_state *cached_state = NULL;
8866         char *kaddr;
8867         unsigned long zero_start;
8868         loff_t size;
8869         int ret;
8870         int reserved = 0;
8871         u64 reserved_space;
8872         u64 page_start;
8873         u64 page_end;
8874         u64 end;
8875
8876         reserved_space = PAGE_SIZE;
8877
8878         sb_start_pagefault(inode->i_sb);
8879         page_start = page_offset(page);
8880         page_end = page_start + PAGE_SIZE - 1;
8881         end = page_end;
8882
8883         /*
8884          * Reserving delalloc space after obtaining the page lock can lead to
8885          * deadlock. For example, if a dirty page is locked by this function
8886          * and the call to btrfs_delalloc_reserve_space() ends up triggering
8887          * dirty page write out, then the btrfs_writepage() function could
8888          * end up waiting indefinitely to get a lock on the page currently
8889          * being processed by btrfs_page_mkwrite() function.
8890          */
8891         ret = btrfs_delalloc_reserve_space(inode, page_start,
8892                                            reserved_space);
8893         if (!ret) {
8894                 ret = file_update_time(vma->vm_file);
8895                 reserved = 1;
8896         }
8897         if (ret) {
8898                 if (ret == -ENOMEM)
8899                         ret = VM_FAULT_OOM;
8900                 else /* -ENOSPC, -EIO, etc */
8901                         ret = VM_FAULT_SIGBUS;
8902                 if (reserved)
8903                         goto out;
8904                 goto out_noreserve;
8905         }
8906
8907         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8908 again:
8909         lock_page(page);
8910         size = i_size_read(inode);
8911
8912         if ((page->mapping != inode->i_mapping) ||
8913             (page_start >= size)) {
8914                 /* page got truncated out from underneath us */
8915                 goto out_unlock;
8916         }
8917         wait_on_page_writeback(page);
8918
8919         lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8920         set_page_extent_mapped(page);
8921
8922         /*
8923          * we can't set the delalloc bits if there are pending ordered
8924          * extents.  Drop our locks and wait for them to finish
8925          */
8926         ordered = btrfs_lookup_ordered_range(inode, page_start, page_end);
8927         if (ordered) {
8928                 unlock_extent_cached(io_tree, page_start, page_end,
8929                                      &cached_state, GFP_NOFS);
8930                 unlock_page(page);
8931                 btrfs_start_ordered_extent(inode, ordered, 1);
8932                 btrfs_put_ordered_extent(ordered);
8933                 goto again;
8934         }
8935
8936         if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8937                 reserved_space = round_up(size - page_start, root->sectorsize);
8938                 if (reserved_space < PAGE_SIZE) {
8939                         end = page_start + reserved_space - 1;
8940                         spin_lock(&BTRFS_I(inode)->lock);
8941                         BTRFS_I(inode)->outstanding_extents++;
8942                         spin_unlock(&BTRFS_I(inode)->lock);
8943                         btrfs_delalloc_release_space(inode, page_start,
8944                                                 PAGE_SIZE - reserved_space);
8945                 }
8946         }
8947
8948         /*
8949          * XXX - page_mkwrite gets called every time the page is dirtied, even
8950          * if it was already dirty, so for space accounting reasons we need to
8951          * clear any delalloc bits for the range we are fixing to save.  There
8952          * is probably a better way to do this, but for now keep consistent with
8953          * prepare_pages in the normal write path.
8954          */
8955         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8956                           EXTENT_DIRTY | EXTENT_DELALLOC |
8957                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8958                           0, 0, &cached_state, GFP_NOFS);
8959
8960         ret = btrfs_set_extent_delalloc(inode, page_start, end,
8961                                         &cached_state);
8962         if (ret) {
8963                 unlock_extent_cached(io_tree, page_start, page_end,
8964                                      &cached_state, GFP_NOFS);
8965                 ret = VM_FAULT_SIGBUS;
8966                 goto out_unlock;
8967         }
8968         ret = 0;
8969
8970         /* page is wholly or partially inside EOF */
8971         if (page_start + PAGE_SIZE > size)
8972                 zero_start = size & ~PAGE_MASK;
8973         else
8974                 zero_start = PAGE_SIZE;
8975
8976         if (zero_start != PAGE_SIZE) {
8977                 kaddr = kmap(page);
8978                 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
8979                 flush_dcache_page(page);
8980                 kunmap(page);
8981         }
8982         ClearPageChecked(page);
8983         set_page_dirty(page);
8984         SetPageUptodate(page);
8985
8986         BTRFS_I(inode)->last_trans = root->fs_info->generation;
8987         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8988         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8989
8990         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
8991
8992 out_unlock:
8993         if (!ret) {
8994                 sb_end_pagefault(inode->i_sb);
8995                 return VM_FAULT_LOCKED;
8996         }
8997         unlock_page(page);
8998 out:
8999         btrfs_delalloc_release_space(inode, page_start, reserved_space);
9000 out_noreserve:
9001         sb_end_pagefault(inode->i_sb);
9002         return ret;
9003 }
9004
9005 static int btrfs_truncate(struct inode *inode)
9006 {
9007         struct btrfs_root *root = BTRFS_I(inode)->root;
9008         struct btrfs_block_rsv *rsv;
9009         int ret = 0;
9010         int err = 0;
9011         struct btrfs_trans_handle *trans;
9012         u64 mask = root->sectorsize - 1;
9013         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
9014
9015         ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9016                                        (u64)-1);
9017         if (ret)
9018                 return ret;
9019
9020         /*
9021          * Yes ladies and gentlemen, this is indeed ugly.  The fact is we have
9022          * 3 things going on here
9023          *
9024          * 1) We need to reserve space for our orphan item and the space to
9025          * delete our orphan item.  Lord knows we don't want to have a dangling
9026          * orphan item because we didn't reserve space to remove it.
9027          *
9028          * 2) We need to reserve space to update our inode.
9029          *
9030          * 3) We need to have something to cache all the space that is going to
9031          * be free'd up by the truncate operation, but also have some slack
9032          * space reserved in case it uses space during the truncate (thank you
9033          * very much snapshotting).
9034          *
9035          * And we need these to all be separate.  The fact is we can use a lot of
9036          * space doing the truncate, and we have no earthly idea how much space
9037          * we will use, so we need the truncate reservation to be separate so it
9038          * doesn't end up using space reserved for updating the inode or
9039          * removing the orphan item.  We also need to be able to stop the
9040          * transaction and start a new one, which means we need to be able to
9041          * update the inode several times, and we have no idea of knowing how
9042          * many times that will be, so we can't just reserve 1 item for the
9043          * entirety of the operation, so that has to be done separately as well.
9044          * Then there is the orphan item, which does indeed need to be held on
9045          * to for the whole operation, and we need nobody to touch this reserved
9046          * space except the orphan code.
9047          *
9048          * So that leaves us with
9049          *
9050          * 1) root->orphan_block_rsv - for the orphan deletion.
9051          * 2) rsv - for the truncate reservation, which we will steal from the
9052          * transaction reservation.
9053          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9054          * updating the inode.
9055          */
9056         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
9057         if (!rsv)
9058                 return -ENOMEM;
9059         rsv->size = min_size;
9060         rsv->failfast = 1;
9061
9062         /*
9063          * 1 for the truncate slack space
9064          * 1 for updating the inode.
9065          */
9066         trans = btrfs_start_transaction(root, 2);
9067         if (IS_ERR(trans)) {
9068                 err = PTR_ERR(trans);
9069                 goto out;
9070         }
9071
9072         /* Migrate the slack space for the truncate to our reserve */
9073         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
9074                                       min_size);
9075         BUG_ON(ret);
9076
9077         /*
9078          * So if we truncate and then write and fsync we normally would just
9079          * write the extents that changed, which is a problem if we need to
9080          * first truncate that entire inode.  So set this flag so we write out
9081          * all of the extents in the inode to the sync log so we're completely
9082          * safe.
9083          */
9084         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9085         trans->block_rsv = rsv;
9086
9087         while (1) {
9088                 ret = btrfs_truncate_inode_items(trans, root, inode,
9089                                                  inode->i_size,
9090                                                  BTRFS_EXTENT_DATA_KEY);
9091                 if (ret != -ENOSPC && ret != -EAGAIN) {
9092                         err = ret;
9093                         break;
9094                 }
9095
9096                 trans->block_rsv = &root->fs_info->trans_block_rsv;
9097                 ret = btrfs_update_inode(trans, root, inode);
9098                 if (ret) {
9099                         err = ret;
9100                         break;
9101                 }
9102
9103                 btrfs_end_transaction(trans, root);
9104                 btrfs_btree_balance_dirty(root);
9105
9106                 trans = btrfs_start_transaction(root, 2);
9107                 if (IS_ERR(trans)) {
9108                         ret = err = PTR_ERR(trans);
9109                         trans = NULL;
9110                         break;
9111                 }
9112
9113                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
9114                                               rsv, min_size);
9115                 BUG_ON(ret);    /* shouldn't happen */
9116                 trans->block_rsv = rsv;
9117         }
9118
9119         if (ret == 0 && inode->i_nlink > 0) {
9120                 trans->block_rsv = root->orphan_block_rsv;
9121                 ret = btrfs_orphan_del(trans, inode);
9122                 if (ret)
9123                         err = ret;
9124         }
9125
9126         if (trans) {
9127                 trans->block_rsv = &root->fs_info->trans_block_rsv;
9128                 ret = btrfs_update_inode(trans, root, inode);
9129                 if (ret && !err)
9130                         err = ret;
9131
9132                 ret = btrfs_end_transaction(trans, root);
9133                 btrfs_btree_balance_dirty(root);
9134         }
9135
9136 out:
9137         btrfs_free_block_rsv(root, rsv);
9138
9139         if (ret && !err)
9140                 err = ret;
9141
9142         return err;
9143 }
9144
9145 /*
9146  * create a new subvolume directory/inode (helper for the ioctl).
9147  */
9148 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9149                              struct btrfs_root *new_root,
9150                              struct btrfs_root *parent_root,
9151                              u64 new_dirid)
9152 {
9153         struct inode *inode;
9154         int err;
9155         u64 index = 0;
9156
9157         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9158                                 new_dirid, new_dirid,
9159                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
9160                                 &index);
9161         if (IS_ERR(inode))
9162                 return PTR_ERR(inode);
9163         inode->i_op = &btrfs_dir_inode_operations;
9164         inode->i_fop = &btrfs_dir_file_operations;
9165
9166         set_nlink(inode, 1);
9167         btrfs_i_size_write(inode, 0);
9168         unlock_new_inode(inode);
9169
9170         err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9171         if (err)
9172                 btrfs_err(new_root->fs_info,
9173                           "error inheriting subvolume %llu properties: %d",
9174                           new_root->root_key.objectid, err);
9175
9176         err = btrfs_update_inode(trans, new_root, inode);
9177
9178         iput(inode);
9179         return err;
9180 }
9181
9182 struct inode *btrfs_alloc_inode(struct super_block *sb)
9183 {
9184         struct btrfs_inode *ei;
9185         struct inode *inode;
9186
9187         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9188         if (!ei)
9189                 return NULL;
9190
9191         ei->root = NULL;
9192         ei->generation = 0;
9193         ei->last_trans = 0;
9194         ei->last_sub_trans = 0;
9195         ei->logged_trans = 0;
9196         ei->delalloc_bytes = 0;
9197         ei->defrag_bytes = 0;
9198         ei->disk_i_size = 0;
9199         ei->flags = 0;
9200         ei->csum_bytes = 0;
9201         ei->index_cnt = (u64)-1;
9202         ei->dir_index = 0;
9203         ei->last_unlink_trans = 0;
9204         ei->last_log_commit = 0;
9205         ei->delayed_iput_count = 0;
9206
9207         spin_lock_init(&ei->lock);
9208         ei->outstanding_extents = 0;
9209         ei->reserved_extents = 0;
9210
9211         ei->runtime_flags = 0;
9212         ei->force_compress = BTRFS_COMPRESS_NONE;
9213
9214         ei->delayed_node = NULL;
9215
9216         ei->i_otime.tv_sec = 0;
9217         ei->i_otime.tv_nsec = 0;
9218
9219         inode = &ei->vfs_inode;
9220         extent_map_tree_init(&ei->extent_tree);
9221         extent_io_tree_init(&ei->io_tree, &inode->i_data);
9222         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
9223         ei->io_tree.track_uptodate = 1;
9224         ei->io_failure_tree.track_uptodate = 1;
9225         atomic_set(&ei->sync_writers, 0);
9226         mutex_init(&ei->log_mutex);
9227         mutex_init(&ei->delalloc_mutex);
9228         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9229         INIT_LIST_HEAD(&ei->delalloc_inodes);
9230         INIT_LIST_HEAD(&ei->delayed_iput);
9231         RB_CLEAR_NODE(&ei->rb_node);
9232
9233         return inode;
9234 }
9235
9236 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9237 void btrfs_test_destroy_inode(struct inode *inode)
9238 {
9239         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9240         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9241 }
9242 #endif
9243
9244 static void btrfs_i_callback(struct rcu_head *head)
9245 {
9246         struct inode *inode = container_of(head, struct inode, i_rcu);
9247         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9248 }
9249
9250 void btrfs_destroy_inode(struct inode *inode)
9251 {
9252         struct btrfs_ordered_extent *ordered;
9253         struct btrfs_root *root = BTRFS_I(inode)->root;
9254
9255         WARN_ON(!hlist_empty(&inode->i_dentry));
9256         WARN_ON(inode->i_data.nrpages);
9257         WARN_ON(BTRFS_I(inode)->outstanding_extents);
9258         WARN_ON(BTRFS_I(inode)->reserved_extents);
9259         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9260         WARN_ON(BTRFS_I(inode)->csum_bytes);
9261         WARN_ON(BTRFS_I(inode)->defrag_bytes);
9262
9263         /*
9264          * This can happen where we create an inode, but somebody else also
9265          * created the same inode and we need to destroy the one we already
9266          * created.
9267          */
9268         if (!root)
9269                 goto free;
9270
9271         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9272                      &BTRFS_I(inode)->runtime_flags)) {
9273                 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
9274                         btrfs_ino(inode));
9275                 atomic_dec(&root->orphan_inodes);
9276         }
9277
9278         while (1) {
9279                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9280                 if (!ordered)
9281                         break;
9282                 else {
9283                         btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
9284                                 ordered->file_offset, ordered->len);
9285                         btrfs_remove_ordered_extent(inode, ordered);
9286                         btrfs_put_ordered_extent(ordered);
9287                         btrfs_put_ordered_extent(ordered);
9288                 }
9289         }
9290         btrfs_qgroup_check_reserved_leak(inode);
9291         inode_tree_del(inode);
9292         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9293 free:
9294         call_rcu(&inode->i_rcu, btrfs_i_callback);
9295 }
9296
9297 int btrfs_drop_inode(struct inode *inode)
9298 {
9299         struct btrfs_root *root = BTRFS_I(inode)->root;
9300
9301         if (root == NULL)
9302                 return 1;
9303
9304         /* the snap/subvol tree is on deleting */
9305         if (btrfs_root_refs(&root->root_item) == 0)
9306                 return 1;
9307         else
9308                 return generic_drop_inode(inode);
9309 }
9310
9311 static void init_once(void *foo)
9312 {
9313         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9314
9315         inode_init_once(&ei->vfs_inode);
9316 }
9317
9318 void btrfs_destroy_cachep(void)
9319 {
9320         /*
9321          * Make sure all delayed rcu free inodes are flushed before we
9322          * destroy cache.
9323          */
9324         rcu_barrier();
9325         kmem_cache_destroy(btrfs_inode_cachep);
9326         kmem_cache_destroy(btrfs_trans_handle_cachep);
9327         kmem_cache_destroy(btrfs_transaction_cachep);
9328         kmem_cache_destroy(btrfs_path_cachep);
9329         kmem_cache_destroy(btrfs_free_space_cachep);
9330 }
9331
9332 int btrfs_init_cachep(void)
9333 {
9334         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9335                         sizeof(struct btrfs_inode), 0,
9336                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9337                         init_once);
9338         if (!btrfs_inode_cachep)
9339                 goto fail;
9340
9341         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9342                         sizeof(struct btrfs_trans_handle), 0,
9343                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9344         if (!btrfs_trans_handle_cachep)
9345                 goto fail;
9346
9347         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9348                         sizeof(struct btrfs_transaction), 0,
9349                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9350         if (!btrfs_transaction_cachep)
9351                 goto fail;
9352
9353         btrfs_path_cachep = kmem_cache_create("btrfs_path",
9354                         sizeof(struct btrfs_path), 0,
9355                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9356         if (!btrfs_path_cachep)
9357                 goto fail;
9358
9359         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9360                         sizeof(struct btrfs_free_space), 0,
9361                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9362         if (!btrfs_free_space_cachep)
9363                 goto fail;
9364
9365         return 0;
9366 fail:
9367         btrfs_destroy_cachep();
9368         return -ENOMEM;
9369 }
9370
9371 static int btrfs_getattr(struct vfsmount *mnt,
9372                          struct dentry *dentry, struct kstat *stat)
9373 {
9374         u64 delalloc_bytes;
9375         struct inode *inode = d_inode(dentry);
9376         u32 blocksize = inode->i_sb->s_blocksize;
9377
9378         generic_fillattr(inode, stat);
9379         stat->dev = BTRFS_I(inode)->root->anon_dev;
9380
9381         spin_lock(&BTRFS_I(inode)->lock);
9382         delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9383         spin_unlock(&BTRFS_I(inode)->lock);
9384         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9385                         ALIGN(delalloc_bytes, blocksize)) >> 9;
9386         return 0;
9387 }
9388
9389 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9390                            struct inode *new_dir, struct dentry *new_dentry)
9391 {
9392         struct btrfs_trans_handle *trans;
9393         struct btrfs_root *root = BTRFS_I(old_dir)->root;
9394         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9395         struct inode *new_inode = d_inode(new_dentry);
9396         struct inode *old_inode = d_inode(old_dentry);
9397         u64 index = 0;
9398         u64 root_objectid;
9399         int ret;
9400         u64 old_ino = btrfs_ino(old_inode);
9401
9402         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9403                 return -EPERM;
9404
9405         /* we only allow rename subvolume link between subvolumes */
9406         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9407                 return -EXDEV;
9408
9409         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9410             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
9411                 return -ENOTEMPTY;
9412
9413         if (S_ISDIR(old_inode->i_mode) && new_inode &&
9414             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9415                 return -ENOTEMPTY;
9416
9417
9418         /* check for collisions, even if the  name isn't there */
9419         ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9420                              new_dentry->d_name.name,
9421                              new_dentry->d_name.len);
9422
9423         if (ret) {
9424                 if (ret == -EEXIST) {
9425                         /* we shouldn't get
9426                          * eexist without a new_inode */
9427                         if (WARN_ON(!new_inode)) {
9428                                 return ret;
9429                         }
9430                 } else {
9431                         /* maybe -EOVERFLOW */
9432                         return ret;
9433                 }
9434         }
9435         ret = 0;
9436
9437         /*
9438          * we're using rename to replace one file with another.  Start IO on it
9439          * now so  we don't add too much work to the end of the transaction
9440          */
9441         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9442                 filemap_flush(old_inode->i_mapping);
9443
9444         /* close the racy window with snapshot create/destroy ioctl */
9445         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9446                 down_read(&root->fs_info->subvol_sem);
9447         /*
9448          * We want to reserve the absolute worst case amount of items.  So if
9449          * both inodes are subvols and we need to unlink them then that would
9450          * require 4 item modifications, but if they are both normal inodes it
9451          * would require 5 item modifications, so we'll assume their normal
9452          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9453          * should cover the worst case number of items we'll modify.
9454          */
9455         trans = btrfs_start_transaction(root, 11);
9456         if (IS_ERR(trans)) {
9457                 ret = PTR_ERR(trans);
9458                 goto out_notrans;
9459         }
9460
9461         if (dest != root)
9462                 btrfs_record_root_in_trans(trans, dest);
9463
9464         ret = btrfs_set_inode_index(new_dir, &index);
9465         if (ret)
9466                 goto out_fail;
9467
9468         BTRFS_I(old_inode)->dir_index = 0ULL;
9469         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9470                 /* force full log commit if subvolume involved. */
9471                 btrfs_set_log_full_commit(root->fs_info, trans);
9472         } else {
9473                 ret = btrfs_insert_inode_ref(trans, dest,
9474                                              new_dentry->d_name.name,
9475                                              new_dentry->d_name.len,
9476                                              old_ino,
9477                                              btrfs_ino(new_dir), index);
9478                 if (ret)
9479                         goto out_fail;
9480                 /*
9481                  * this is an ugly little race, but the rename is required
9482                  * to make sure that if we crash, the inode is either at the
9483                  * old name or the new one.  pinning the log transaction lets
9484                  * us make sure we don't allow a log commit to come in after
9485                  * we unlink the name but before we add the new name back in.
9486                  */
9487                 btrfs_pin_log_trans(root);
9488         }
9489
9490         inode_inc_iversion(old_dir);
9491         inode_inc_iversion(new_dir);
9492         inode_inc_iversion(old_inode);
9493         old_dir->i_ctime = old_dir->i_mtime =
9494         new_dir->i_ctime = new_dir->i_mtime =
9495         old_inode->i_ctime = current_fs_time(old_dir->i_sb);
9496
9497         if (old_dentry->d_parent != new_dentry->d_parent)
9498                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9499
9500         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9501                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9502                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9503                                         old_dentry->d_name.name,
9504                                         old_dentry->d_name.len);
9505         } else {
9506                 ret = __btrfs_unlink_inode(trans, root, old_dir,
9507                                         d_inode(old_dentry),
9508                                         old_dentry->d_name.name,
9509                                         old_dentry->d_name.len);
9510                 if (!ret)
9511                         ret = btrfs_update_inode(trans, root, old_inode);
9512         }
9513         if (ret) {
9514                 btrfs_abort_transaction(trans, root, ret);
9515                 goto out_fail;
9516         }
9517
9518         if (new_inode) {
9519                 inode_inc_iversion(new_inode);
9520                 new_inode->i_ctime = current_fs_time(new_inode->i_sb);
9521                 if (unlikely(btrfs_ino(new_inode) ==
9522                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9523                         root_objectid = BTRFS_I(new_inode)->location.objectid;
9524                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
9525                                                 root_objectid,
9526                                                 new_dentry->d_name.name,
9527                                                 new_dentry->d_name.len);
9528                         BUG_ON(new_inode->i_nlink == 0);
9529                 } else {
9530                         ret = btrfs_unlink_inode(trans, dest, new_dir,
9531                                                  d_inode(new_dentry),
9532                                                  new_dentry->d_name.name,
9533                                                  new_dentry->d_name.len);
9534                 }
9535                 if (!ret && new_inode->i_nlink == 0)
9536                         ret = btrfs_orphan_add(trans, d_inode(new_dentry));
9537                 if (ret) {
9538                         btrfs_abort_transaction(trans, root, ret);
9539                         goto out_fail;
9540                 }
9541         }
9542
9543         ret = btrfs_add_link(trans, new_dir, old_inode,
9544                              new_dentry->d_name.name,
9545                              new_dentry->d_name.len, 0, index);
9546         if (ret) {
9547                 btrfs_abort_transaction(trans, root, ret);
9548                 goto out_fail;
9549         }
9550
9551         if (old_inode->i_nlink == 1)
9552                 BTRFS_I(old_inode)->dir_index = index;
9553
9554         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
9555                 struct dentry *parent = new_dentry->d_parent;
9556                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9557                 btrfs_end_log_trans(root);
9558         }
9559 out_fail:
9560         btrfs_end_transaction(trans, root);
9561 out_notrans:
9562         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9563                 up_read(&root->fs_info->subvol_sem);
9564
9565         return ret;
9566 }
9567
9568 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9569                          struct inode *new_dir, struct dentry *new_dentry,
9570                          unsigned int flags)
9571 {
9572         if (flags & ~RENAME_NOREPLACE)
9573                 return -EINVAL;
9574
9575         return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
9576 }
9577
9578 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9579 {
9580         struct btrfs_delalloc_work *delalloc_work;
9581         struct inode *inode;
9582
9583         delalloc_work = container_of(work, struct btrfs_delalloc_work,
9584                                      work);
9585         inode = delalloc_work->inode;
9586         filemap_flush(inode->i_mapping);
9587         if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9588                                 &BTRFS_I(inode)->runtime_flags))
9589                 filemap_flush(inode->i_mapping);
9590
9591         if (delalloc_work->delay_iput)
9592                 btrfs_add_delayed_iput(inode);
9593         else
9594                 iput(inode);
9595         complete(&delalloc_work->completion);
9596 }
9597
9598 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9599                                                     int delay_iput)
9600 {
9601         struct btrfs_delalloc_work *work;
9602
9603         work = kmalloc(sizeof(*work), GFP_NOFS);
9604         if (!work)
9605                 return NULL;
9606
9607         init_completion(&work->completion);
9608         INIT_LIST_HEAD(&work->list);
9609         work->inode = inode;
9610         work->delay_iput = delay_iput;
9611         WARN_ON_ONCE(!inode);
9612         btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9613                         btrfs_run_delalloc_work, NULL, NULL);
9614
9615         return work;
9616 }
9617
9618 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9619 {
9620         wait_for_completion(&work->completion);
9621         kfree(work);
9622 }
9623
9624 /*
9625  * some fairly slow code that needs optimization. This walks the list
9626  * of all the inodes with pending delalloc and forces them to disk.
9627  */
9628 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9629                                    int nr)
9630 {
9631         struct btrfs_inode *binode;
9632         struct inode *inode;
9633         struct btrfs_delalloc_work *work, *next;
9634         struct list_head works;
9635         struct list_head splice;
9636         int ret = 0;
9637
9638         INIT_LIST_HEAD(&works);
9639         INIT_LIST_HEAD(&splice);
9640
9641         mutex_lock(&root->delalloc_mutex);
9642         spin_lock(&root->delalloc_lock);
9643         list_splice_init(&root->delalloc_inodes, &splice);
9644         while (!list_empty(&splice)) {
9645                 binode = list_entry(splice.next, struct btrfs_inode,
9646                                     delalloc_inodes);
9647
9648                 list_move_tail(&binode->delalloc_inodes,
9649                                &root->delalloc_inodes);
9650                 inode = igrab(&binode->vfs_inode);
9651                 if (!inode) {
9652                         cond_resched_lock(&root->delalloc_lock);
9653                         continue;
9654                 }
9655                 spin_unlock(&root->delalloc_lock);
9656
9657                 work = btrfs_alloc_delalloc_work(inode, delay_iput);
9658                 if (!work) {
9659                         if (delay_iput)
9660                                 btrfs_add_delayed_iput(inode);
9661                         else
9662                                 iput(inode);
9663                         ret = -ENOMEM;
9664                         goto out;
9665                 }
9666                 list_add_tail(&work->list, &works);
9667                 btrfs_queue_work(root->fs_info->flush_workers,
9668                                  &work->work);
9669                 ret++;
9670                 if (nr != -1 && ret >= nr)
9671                         goto out;
9672                 cond_resched();
9673                 spin_lock(&root->delalloc_lock);
9674         }
9675         spin_unlock(&root->delalloc_lock);
9676
9677 out:
9678         list_for_each_entry_safe(work, next, &works, list) {
9679                 list_del_init(&work->list);
9680                 btrfs_wait_and_free_delalloc_work(work);
9681         }
9682
9683         if (!list_empty_careful(&splice)) {
9684                 spin_lock(&root->delalloc_lock);
9685                 list_splice_tail(&splice, &root->delalloc_inodes);
9686                 spin_unlock(&root->delalloc_lock);
9687         }
9688         mutex_unlock(&root->delalloc_mutex);
9689         return ret;
9690 }
9691
9692 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9693 {
9694         int ret;
9695
9696         if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
9697                 return -EROFS;
9698
9699         ret = __start_delalloc_inodes(root, delay_iput, -1);
9700         if (ret > 0)
9701                 ret = 0;
9702         /*
9703          * the filemap_flush will queue IO into the worker threads, but
9704          * we have to make sure the IO is actually started and that
9705          * ordered extents get created before we return
9706          */
9707         atomic_inc(&root->fs_info->async_submit_draining);
9708         while (atomic_read(&root->fs_info->nr_async_submits) ||
9709               atomic_read(&root->fs_info->async_delalloc_pages)) {
9710                 wait_event(root->fs_info->async_submit_wait,
9711                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9712                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
9713         }
9714         atomic_dec(&root->fs_info->async_submit_draining);
9715         return ret;
9716 }
9717
9718 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9719                                int nr)
9720 {
9721         struct btrfs_root *root;
9722         struct list_head splice;
9723         int ret;
9724
9725         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9726                 return -EROFS;
9727
9728         INIT_LIST_HEAD(&splice);
9729
9730         mutex_lock(&fs_info->delalloc_root_mutex);
9731         spin_lock(&fs_info->delalloc_root_lock);
9732         list_splice_init(&fs_info->delalloc_roots, &splice);
9733         while (!list_empty(&splice) && nr) {
9734                 root = list_first_entry(&splice, struct btrfs_root,
9735                                         delalloc_root);
9736                 root = btrfs_grab_fs_root(root);
9737                 BUG_ON(!root);
9738                 list_move_tail(&root->delalloc_root,
9739                                &fs_info->delalloc_roots);
9740                 spin_unlock(&fs_info->delalloc_root_lock);
9741
9742                 ret = __start_delalloc_inodes(root, delay_iput, nr);
9743                 btrfs_put_fs_root(root);
9744                 if (ret < 0)
9745                         goto out;
9746
9747                 if (nr != -1) {
9748                         nr -= ret;
9749                         WARN_ON(nr < 0);
9750                 }
9751                 spin_lock(&fs_info->delalloc_root_lock);
9752         }
9753         spin_unlock(&fs_info->delalloc_root_lock);
9754
9755         ret = 0;
9756         atomic_inc(&fs_info->async_submit_draining);
9757         while (atomic_read(&fs_info->nr_async_submits) ||
9758               atomic_read(&fs_info->async_delalloc_pages)) {
9759                 wait_event(fs_info->async_submit_wait,
9760                    (atomic_read(&fs_info->nr_async_submits) == 0 &&
9761                     atomic_read(&fs_info->async_delalloc_pages) == 0));
9762         }
9763         atomic_dec(&fs_info->async_submit_draining);
9764 out:
9765         if (!list_empty_careful(&splice)) {
9766                 spin_lock(&fs_info->delalloc_root_lock);
9767                 list_splice_tail(&splice, &fs_info->delalloc_roots);
9768                 spin_unlock(&fs_info->delalloc_root_lock);
9769         }
9770         mutex_unlock(&fs_info->delalloc_root_mutex);
9771         return ret;
9772 }
9773
9774 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9775                          const char *symname)
9776 {
9777         struct btrfs_trans_handle *trans;
9778         struct btrfs_root *root = BTRFS_I(dir)->root;
9779         struct btrfs_path *path;
9780         struct btrfs_key key;
9781         struct inode *inode = NULL;
9782         int err;
9783         int drop_inode = 0;
9784         u64 objectid;
9785         u64 index = 0;
9786         int name_len;
9787         int datasize;
9788         unsigned long ptr;
9789         struct btrfs_file_extent_item *ei;
9790         struct extent_buffer *leaf;
9791
9792         name_len = strlen(symname);
9793         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9794                 return -ENAMETOOLONG;
9795
9796         /*
9797          * 2 items for inode item and ref
9798          * 2 items for dir items
9799          * 1 item for updating parent inode item
9800          * 1 item for the inline extent item
9801          * 1 item for xattr if selinux is on
9802          */
9803         trans = btrfs_start_transaction(root, 7);
9804         if (IS_ERR(trans))
9805                 return PTR_ERR(trans);
9806
9807         err = btrfs_find_free_ino(root, &objectid);
9808         if (err)
9809                 goto out_unlock;
9810
9811         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9812                                 dentry->d_name.len, btrfs_ino(dir), objectid,
9813                                 S_IFLNK|S_IRWXUGO, &index);
9814         if (IS_ERR(inode)) {
9815                 err = PTR_ERR(inode);
9816                 goto out_unlock;
9817         }
9818
9819         /*
9820         * If the active LSM wants to access the inode during
9821         * d_instantiate it needs these. Smack checks to see
9822         * if the filesystem supports xattrs by looking at the
9823         * ops vector.
9824         */
9825         inode->i_fop = &btrfs_file_operations;
9826         inode->i_op = &btrfs_file_inode_operations;
9827         inode->i_mapping->a_ops = &btrfs_aops;
9828         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9829
9830         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9831         if (err)
9832                 goto out_unlock_inode;
9833
9834         path = btrfs_alloc_path();
9835         if (!path) {
9836                 err = -ENOMEM;
9837                 goto out_unlock_inode;
9838         }
9839         key.objectid = btrfs_ino(inode);
9840         key.offset = 0;
9841         key.type = BTRFS_EXTENT_DATA_KEY;
9842         datasize = btrfs_file_extent_calc_inline_size(name_len);
9843         err = btrfs_insert_empty_item(trans, root, path, &key,
9844                                       datasize);
9845         if (err) {
9846                 btrfs_free_path(path);
9847                 goto out_unlock_inode;
9848         }
9849         leaf = path->nodes[0];
9850         ei = btrfs_item_ptr(leaf, path->slots[0],
9851                             struct btrfs_file_extent_item);
9852         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9853         btrfs_set_file_extent_type(leaf, ei,
9854                                    BTRFS_FILE_EXTENT_INLINE);
9855         btrfs_set_file_extent_encryption(leaf, ei, 0);
9856         btrfs_set_file_extent_compression(leaf, ei, 0);
9857         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9858         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9859
9860         ptr = btrfs_file_extent_inline_start(ei);
9861         write_extent_buffer(leaf, symname, ptr, name_len);
9862         btrfs_mark_buffer_dirty(leaf);
9863         btrfs_free_path(path);
9864
9865         inode->i_op = &btrfs_symlink_inode_operations;
9866         inode_nohighmem(inode);
9867         inode->i_mapping->a_ops = &btrfs_symlink_aops;
9868         inode_set_bytes(inode, name_len);
9869         btrfs_i_size_write(inode, name_len);
9870         err = btrfs_update_inode(trans, root, inode);
9871         /*
9872          * Last step, add directory indexes for our symlink inode. This is the
9873          * last step to avoid extra cleanup of these indexes if an error happens
9874          * elsewhere above.
9875          */
9876         if (!err)
9877                 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
9878         if (err) {
9879                 drop_inode = 1;
9880                 goto out_unlock_inode;
9881         }
9882
9883         unlock_new_inode(inode);
9884         d_instantiate(dentry, inode);
9885
9886 out_unlock:
9887         btrfs_end_transaction(trans, root);
9888         if (drop_inode) {
9889                 inode_dec_link_count(inode);
9890                 iput(inode);
9891         }
9892         btrfs_btree_balance_dirty(root);
9893         return err;
9894
9895 out_unlock_inode:
9896         drop_inode = 1;
9897         unlock_new_inode(inode);
9898         goto out_unlock;
9899 }
9900
9901 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9902                                        u64 start, u64 num_bytes, u64 min_size,
9903                                        loff_t actual_len, u64 *alloc_hint,
9904                                        struct btrfs_trans_handle *trans)
9905 {
9906         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9907         struct extent_map *em;
9908         struct btrfs_root *root = BTRFS_I(inode)->root;
9909         struct btrfs_key ins;
9910         u64 cur_offset = start;
9911         u64 i_size;
9912         u64 cur_bytes;
9913         u64 last_alloc = (u64)-1;
9914         int ret = 0;
9915         bool own_trans = true;
9916
9917         if (trans)
9918                 own_trans = false;
9919         while (num_bytes > 0) {
9920                 if (own_trans) {
9921                         trans = btrfs_start_transaction(root, 3);
9922                         if (IS_ERR(trans)) {
9923                                 ret = PTR_ERR(trans);
9924                                 break;
9925                         }
9926                 }
9927
9928                 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9929                 cur_bytes = max(cur_bytes, min_size);
9930                 /*
9931                  * If we are severely fragmented we could end up with really
9932                  * small allocations, so if the allocator is returning small
9933                  * chunks lets make its job easier by only searching for those
9934                  * sized chunks.
9935                  */
9936                 cur_bytes = min(cur_bytes, last_alloc);
9937                 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
9938                                            *alloc_hint, &ins, 1, 0);
9939                 if (ret) {
9940                         if (own_trans)
9941                                 btrfs_end_transaction(trans, root);
9942                         break;
9943                 }
9944
9945                 last_alloc = ins.offset;
9946                 ret = insert_reserved_file_extent(trans, inode,
9947                                                   cur_offset, ins.objectid,
9948                                                   ins.offset, ins.offset,
9949                                                   ins.offset, 0, 0, 0,
9950                                                   BTRFS_FILE_EXTENT_PREALLOC);
9951                 if (ret) {
9952                         btrfs_free_reserved_extent(root, ins.objectid,
9953                                                    ins.offset, 0);
9954                         btrfs_abort_transaction(trans, root, ret);
9955                         if (own_trans)
9956                                 btrfs_end_transaction(trans, root);
9957                         break;
9958                 }
9959
9960                 btrfs_drop_extent_cache(inode, cur_offset,
9961                                         cur_offset + ins.offset -1, 0);
9962
9963                 em = alloc_extent_map();
9964                 if (!em) {
9965                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9966                                 &BTRFS_I(inode)->runtime_flags);
9967                         goto next;
9968                 }
9969
9970                 em->start = cur_offset;
9971                 em->orig_start = cur_offset;
9972                 em->len = ins.offset;
9973                 em->block_start = ins.objectid;
9974                 em->block_len = ins.offset;
9975                 em->orig_block_len = ins.offset;
9976                 em->ram_bytes = ins.offset;
9977                 em->bdev = root->fs_info->fs_devices->latest_bdev;
9978                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9979                 em->generation = trans->transid;
9980
9981                 while (1) {
9982                         write_lock(&em_tree->lock);
9983                         ret = add_extent_mapping(em_tree, em, 1);
9984                         write_unlock(&em_tree->lock);
9985                         if (ret != -EEXIST)
9986                                 break;
9987                         btrfs_drop_extent_cache(inode, cur_offset,
9988                                                 cur_offset + ins.offset - 1,
9989                                                 0);
9990                 }
9991                 free_extent_map(em);
9992 next:
9993                 num_bytes -= ins.offset;
9994                 cur_offset += ins.offset;
9995                 *alloc_hint = ins.objectid + ins.offset;
9996
9997                 inode_inc_iversion(inode);
9998                 inode->i_ctime = current_fs_time(inode->i_sb);
9999                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10000                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10001                     (actual_len > inode->i_size) &&
10002                     (cur_offset > inode->i_size)) {
10003                         if (cur_offset > actual_len)
10004                                 i_size = actual_len;
10005                         else
10006                                 i_size = cur_offset;
10007                         i_size_write(inode, i_size);
10008                         btrfs_ordered_update_i_size(inode, i_size, NULL);
10009                 }
10010
10011                 ret = btrfs_update_inode(trans, root, inode);
10012
10013                 if (ret) {
10014                         btrfs_abort_transaction(trans, root, ret);
10015                         if (own_trans)
10016                                 btrfs_end_transaction(trans, root);
10017                         break;
10018                 }
10019
10020                 if (own_trans)
10021                         btrfs_end_transaction(trans, root);
10022         }
10023         return ret;
10024 }
10025
10026 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10027                               u64 start, u64 num_bytes, u64 min_size,
10028                               loff_t actual_len, u64 *alloc_hint)
10029 {
10030         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10031                                            min_size, actual_len, alloc_hint,
10032                                            NULL);
10033 }
10034
10035 int btrfs_prealloc_file_range_trans(struct inode *inode,
10036                                     struct btrfs_trans_handle *trans, int mode,
10037                                     u64 start, u64 num_bytes, u64 min_size,
10038                                     loff_t actual_len, u64 *alloc_hint)
10039 {
10040         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10041                                            min_size, actual_len, alloc_hint, trans);
10042 }
10043
10044 static int btrfs_set_page_dirty(struct page *page)
10045 {
10046         return __set_page_dirty_nobuffers(page);
10047 }
10048
10049 static int btrfs_permission(struct inode *inode, int mask)
10050 {
10051         struct btrfs_root *root = BTRFS_I(inode)->root;
10052         umode_t mode = inode->i_mode;
10053
10054         if (mask & MAY_WRITE &&
10055             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10056                 if (btrfs_root_readonly(root))
10057                         return -EROFS;
10058                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10059                         return -EACCES;
10060         }
10061         return generic_permission(inode, mask);
10062 }
10063
10064 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10065 {
10066         struct btrfs_trans_handle *trans;
10067         struct btrfs_root *root = BTRFS_I(dir)->root;
10068         struct inode *inode = NULL;
10069         u64 objectid;
10070         u64 index;
10071         int ret = 0;
10072
10073         /*
10074          * 5 units required for adding orphan entry
10075          */
10076         trans = btrfs_start_transaction(root, 5);
10077         if (IS_ERR(trans))
10078                 return PTR_ERR(trans);
10079
10080         ret = btrfs_find_free_ino(root, &objectid);
10081         if (ret)
10082                 goto out;
10083
10084         inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10085                                 btrfs_ino(dir), objectid, mode, &index);
10086         if (IS_ERR(inode)) {
10087                 ret = PTR_ERR(inode);
10088                 inode = NULL;
10089                 goto out;
10090         }
10091
10092         inode->i_fop = &btrfs_file_operations;
10093         inode->i_op = &btrfs_file_inode_operations;
10094
10095         inode->i_mapping->a_ops = &btrfs_aops;
10096         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10097
10098         ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10099         if (ret)
10100                 goto out_inode;
10101
10102         ret = btrfs_update_inode(trans, root, inode);
10103         if (ret)
10104                 goto out_inode;
10105         ret = btrfs_orphan_add(trans, inode);
10106         if (ret)
10107                 goto out_inode;
10108
10109         /*
10110          * We set number of links to 0 in btrfs_new_inode(), and here we set
10111          * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10112          * through:
10113          *
10114          *    d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10115          */
10116         set_nlink(inode, 1);
10117         unlock_new_inode(inode);
10118         d_tmpfile(dentry, inode);
10119         mark_inode_dirty(inode);
10120
10121 out:
10122         btrfs_end_transaction(trans, root);
10123         if (ret)
10124                 iput(inode);
10125         btrfs_balance_delayed_items(root);
10126         btrfs_btree_balance_dirty(root);
10127         return ret;
10128
10129 out_inode:
10130         unlock_new_inode(inode);
10131         goto out;
10132
10133 }
10134
10135 /* Inspired by filemap_check_errors() */
10136 int btrfs_inode_check_errors(struct inode *inode)
10137 {
10138         int ret = 0;
10139
10140         if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
10141             test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
10142                 ret = -ENOSPC;
10143         if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
10144             test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
10145                 ret = -EIO;
10146
10147         return ret;
10148 }
10149
10150 static const struct inode_operations btrfs_dir_inode_operations = {
10151         .getattr        = btrfs_getattr,
10152         .lookup         = btrfs_lookup,
10153         .create         = btrfs_create,
10154         .unlink         = btrfs_unlink,
10155         .link           = btrfs_link,
10156         .mkdir          = btrfs_mkdir,
10157         .rmdir          = btrfs_rmdir,
10158         .rename2        = btrfs_rename2,
10159         .symlink        = btrfs_symlink,
10160         .setattr        = btrfs_setattr,
10161         .mknod          = btrfs_mknod,
10162         .setxattr       = btrfs_setxattr,
10163         .getxattr       = generic_getxattr,
10164         .listxattr      = btrfs_listxattr,
10165         .removexattr    = btrfs_removexattr,
10166         .permission     = btrfs_permission,
10167         .get_acl        = btrfs_get_acl,
10168         .set_acl        = btrfs_set_acl,
10169         .update_time    = btrfs_update_time,
10170         .tmpfile        = btrfs_tmpfile,
10171 };
10172 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10173         .lookup         = btrfs_lookup,
10174         .permission     = btrfs_permission,
10175         .get_acl        = btrfs_get_acl,
10176         .set_acl        = btrfs_set_acl,
10177         .update_time    = btrfs_update_time,
10178 };
10179
10180 static const struct file_operations btrfs_dir_file_operations = {
10181         .llseek         = generic_file_llseek,
10182         .read           = generic_read_dir,
10183         .iterate        = btrfs_real_readdir,
10184         .unlocked_ioctl = btrfs_ioctl,
10185 #ifdef CONFIG_COMPAT
10186         .compat_ioctl   = btrfs_ioctl,
10187 #endif
10188         .release        = btrfs_release_file,
10189         .fsync          = btrfs_sync_file,
10190 };
10191
10192 static const struct extent_io_ops btrfs_extent_io_ops = {
10193         .fill_delalloc = run_delalloc_range,
10194         .submit_bio_hook = btrfs_submit_bio_hook,
10195         .merge_bio_hook = btrfs_merge_bio_hook,
10196         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10197         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10198         .writepage_start_hook = btrfs_writepage_start_hook,
10199         .set_bit_hook = btrfs_set_bit_hook,
10200         .clear_bit_hook = btrfs_clear_bit_hook,
10201         .merge_extent_hook = btrfs_merge_extent_hook,
10202         .split_extent_hook = btrfs_split_extent_hook,
10203 };
10204
10205 /*
10206  * btrfs doesn't support the bmap operation because swapfiles
10207  * use bmap to make a mapping of extents in the file.  They assume
10208  * these extents won't change over the life of the file and they
10209  * use the bmap result to do IO directly to the drive.
10210  *
10211  * the btrfs bmap call would return logical addresses that aren't
10212  * suitable for IO and they also will change frequently as COW
10213  * operations happen.  So, swapfile + btrfs == corruption.
10214  *
10215  * For now we're avoiding this by dropping bmap.
10216  */
10217 static const struct address_space_operations btrfs_aops = {
10218         .readpage       = btrfs_readpage,
10219         .writepage      = btrfs_writepage,
10220         .writepages     = btrfs_writepages,
10221         .readpages      = btrfs_readpages,
10222         .direct_IO      = btrfs_direct_IO,
10223         .invalidatepage = btrfs_invalidatepage,
10224         .releasepage    = btrfs_releasepage,
10225         .set_page_dirty = btrfs_set_page_dirty,
10226         .error_remove_page = generic_error_remove_page,
10227 };
10228
10229 static const struct address_space_operations btrfs_symlink_aops = {
10230         .readpage       = btrfs_readpage,
10231         .writepage      = btrfs_writepage,
10232         .invalidatepage = btrfs_invalidatepage,
10233         .releasepage    = btrfs_releasepage,
10234 };
10235
10236 static const struct inode_operations btrfs_file_inode_operations = {
10237         .getattr        = btrfs_getattr,
10238         .setattr        = btrfs_setattr,
10239         .setxattr       = btrfs_setxattr,
10240         .getxattr       = generic_getxattr,
10241         .listxattr      = btrfs_listxattr,
10242         .removexattr    = btrfs_removexattr,
10243         .permission     = btrfs_permission,
10244         .fiemap         = btrfs_fiemap,
10245         .get_acl        = btrfs_get_acl,
10246         .set_acl        = btrfs_set_acl,
10247         .update_time    = btrfs_update_time,
10248 };
10249 static const struct inode_operations btrfs_special_inode_operations = {
10250         .getattr        = btrfs_getattr,
10251         .setattr        = btrfs_setattr,
10252         .permission     = btrfs_permission,
10253         .setxattr       = btrfs_setxattr,
10254         .getxattr       = generic_getxattr,
10255         .listxattr      = btrfs_listxattr,
10256         .removexattr    = btrfs_removexattr,
10257         .get_acl        = btrfs_get_acl,
10258         .set_acl        = btrfs_set_acl,
10259         .update_time    = btrfs_update_time,
10260 };
10261 static const struct inode_operations btrfs_symlink_inode_operations = {
10262         .readlink       = generic_readlink,
10263         .get_link       = page_get_link,
10264         .getattr        = btrfs_getattr,
10265         .setattr        = btrfs_setattr,
10266         .permission     = btrfs_permission,
10267         .setxattr       = btrfs_setxattr,
10268         .getxattr       = generic_getxattr,
10269         .listxattr      = btrfs_listxattr,
10270         .removexattr    = btrfs_removexattr,
10271         .update_time    = btrfs_update_time,
10272 };
10273
10274 const struct dentry_operations btrfs_dentry_operations = {
10275         .d_delete       = btrfs_dentry_delete,
10276         .d_release      = btrfs_dentry_release,
10277 };