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/aio.h>
  36 #include <linux/bit_spinlock.h>
  37 #include <linux/xattr.h>
  38 #include <linux/posix_acl.h>
  39 #include <linux/falloc.h>
  40 #include <linux/slab.h>
  41 #include <linux/ratelimit.h>
  42 #include <linux/mount.h>
  43 #include <linux/btrfs.h>
  44 #include <linux/blkdev.h>
  45 #include <linux/posix_acl_xattr.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
  63 struct btrfs_iget_args {
  64         struct btrfs_key *location;
  65         struct btrfs_root *root;
  66 };
  67
  68 static const struct inode_operations btrfs_dir_inode_operations;
  69 static const struct inode_operations btrfs_symlink_inode_operations;
  70 static const struct inode_operations btrfs_dir_ro_inode_operations;
  71 static const struct inode_operations btrfs_special_inode_operations;
  72 static const struct inode_operations btrfs_file_inode_operations;
  73 static const struct address_space_operations btrfs_aops;
  74 static const struct address_space_operations btrfs_symlink_aops;
  75 static const struct file_operations btrfs_dir_file_operations;
  76 static struct extent_io_ops btrfs_extent_io_ops;
  77
  78 static struct kmem_cache *btrfs_inode_cachep;
  79 static struct kmem_cache *btrfs_delalloc_work_cachep;
  80 struct kmem_cache *btrfs_trans_handle_cachep;
  81 struct kmem_cache *btrfs_transaction_cachep;
  82 struct kmem_cache *btrfs_path_cachep;
  83 struct kmem_cache *btrfs_free_space_cachep;
  84
  85 #define S_SHIFT 12
  86 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  87         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  88         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  89         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  90         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  91         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  92         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  93         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  94 };
  95
  96 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
  97 static int btrfs_truncate(struct inode *inode);
  98 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  99 static noinline int cow_file_range(struct inode *inode,
 100                                    struct page *locked_page,
 101                                    u64 start, u64 end, int *page_started,
 102                                    unsigned long *nr_written, int unlock);
 103 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
 104                                            u64 len, u64 orig_start,
 105                                            u64 block_start, u64 block_len,
 106                                            u64 orig_block_len, u64 ram_bytes,
 107                                            int type);
 108
 109 static int btrfs_dirty_inode(struct inode *inode);
 110
 111 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 112                                      struct inode *inode,  struct inode *dir,
 113                                      const struct qstr *qstr)
 114 {
 115         int err;
 116
 117         err = btrfs_init_acl(trans, inode, dir);
 118         if (!err)
 119                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 120         return err;
 121 }
 122
 123 /*
 124  * this does all the hard work for inserting an inline extent into
 125  * the btree.  The caller should have done a btrfs_drop_extents so that
 126  * no overlapping inline items exist in the btree
 127  */
 128 static int insert_inline_extent(struct btrfs_trans_handle *trans,
 129                                 struct btrfs_path *path, int extent_inserted,
 130                                 struct btrfs_root *root, struct inode *inode,
 131                                 u64 start, size_t size, size_t compressed_size,
 132                                 int compress_type,
 133                                 struct page **compressed_pages)
 134 {
 135         struct extent_buffer *leaf;
 136         struct page *page = NULL;
 137         char *kaddr;
 138         unsigned long ptr;
 139         struct btrfs_file_extent_item *ei;
 140         int err = 0;
 141         int ret;
 142         size_t cur_size = size;
 143         unsigned long offset;
 144
 145         if (compressed_size && compressed_pages)
 146                 cur_size = compressed_size;
 147
 148         inode_add_bytes(inode, size);
 149
 150         if (!extent_inserted) {
 151                 struct btrfs_key key;
 152                 size_t datasize;
 153
 154                 key.objectid = btrfs_ino(inode);
 155                 key.offset = start;
 156                 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 157
 158                 datasize = btrfs_file_extent_calc_inline_size(cur_size);
 159                 path->leave_spinning = 1;
 160                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 161                                               datasize);
 162                 if (ret) {
 163                         err = ret;
 164                         goto fail;
 165                 }
 166         }
 167         leaf = path->nodes[0];
 168         ei = btrfs_item_ptr(leaf, path->slots[0],
 169                             struct btrfs_file_extent_item);
 170         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 171         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 172         btrfs_set_file_extent_encryption(leaf, ei, 0);
 173         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 174         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 175         ptr = btrfs_file_extent_inline_start(ei);
 176
 177         if (compress_type != BTRFS_COMPRESS_NONE) {
 178                 struct page *cpage;
 179                 int i = 0;
 180                 while (compressed_size > 0) {
 181                         cpage = compressed_pages[i];
 182                         cur_size = min_t(unsigned long, compressed_size,
 183                                        PAGE_CACHE_SIZE);
 184
 185                         kaddr = kmap_atomic(cpage);
 186                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 187                         kunmap_atomic(kaddr);
 188
 189                         i++;
 190                         ptr += cur_size;
 191                         compressed_size -= cur_size;
 192                 }
 193                 btrfs_set_file_extent_compression(leaf, ei,
 194                                                   compress_type);
 195         } else {
 196                 page = find_get_page(inode->i_mapping,
 197                                      start >> PAGE_CACHE_SHIFT);
 198                 btrfs_set_file_extent_compression(leaf, ei, 0);
 199                 kaddr = kmap_atomic(page);
 200                 offset = start & (PAGE_CACHE_SIZE - 1);
 201                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 202                 kunmap_atomic(kaddr);
 203                 page_cache_release(page);
 204         }
 205         btrfs_mark_buffer_dirty(leaf);
 206         btrfs_release_path(path);
 207
 208         /*
 209          * we're an inline extent, so nobody can
 210          * extend the file past i_size without locking
 211          * a page we already have locked.
 212          *
 213          * We must do any isize and inode updates
 214          * before we unlock the pages.  Otherwise we
 215          * could end up racing with unlink.
 216          */
 217         BTRFS_I(inode)->disk_i_size = inode->i_size;
 218         ret = btrfs_update_inode(trans, root, inode);
 219
 220         return ret;
 221 fail:
 222         return err;
 223 }
 224
 225
 226 /*
 227  * conditionally insert an inline extent into the file.  This
 228  * does the checks required to make sure the data is small enough
 229  * to fit as an inline extent.
 230  */
 231 static noinline int cow_file_range_inline(struct btrfs_root *root,
 232                                           struct inode *inode, u64 start,
 233                                           u64 end, size_t compressed_size,
 234                                           int compress_type,
 235                                           struct page **compressed_pages)
 236 {
 237         struct btrfs_trans_handle *trans;
 238         u64 isize = i_size_read(inode);
 239         u64 actual_end = min(end + 1, isize);
 240         u64 inline_len = actual_end - start;
 241         u64 aligned_end = ALIGN(end, root->sectorsize);
 242         u64 data_len = inline_len;
 243         int ret;
 244         struct btrfs_path *path;
 245         int extent_inserted = 0;
 246         u32 extent_item_size;
 247
 248         if (compressed_size)
 249                 data_len = compressed_size;
 250
 251         if (start > 0 ||
 252             actual_end >= PAGE_CACHE_SIZE ||
 253             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 254             (!compressed_size &&
 255             (actual_end & (root->sectorsize - 1)) == 0) ||
 256             end + 1 < isize ||
 257             data_len > root->fs_info->max_inline) {
 258                 return 1;
 259         }
 260
 261         path = btrfs_alloc_path();
 262         if (!path)
 263                 return -ENOMEM;
 264
 265         trans = btrfs_join_transaction(root);
 266         if (IS_ERR(trans)) {
 267                 btrfs_free_path(path);
 268                 return PTR_ERR(trans);
 269         }
 270         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 271
 272         if (compressed_size && compressed_pages)
 273                 extent_item_size = btrfs_file_extent_calc_inline_size(
 274                    compressed_size);
 275         else
 276                 extent_item_size = btrfs_file_extent_calc_inline_size(
 277                     inline_len);
 278
 279         ret = __btrfs_drop_extents(trans, root, inode, path,
 280                                    start, aligned_end, NULL,
 281                                    1, 1, extent_item_size, &extent_inserted);
 282         if (ret) {
 283                 btrfs_abort_transaction(trans, root, ret);
 284                 goto out;
 285         }
 286
 287         if (isize > actual_end)
 288                 inline_len = min_t(u64, isize, actual_end);
 289         ret = insert_inline_extent(trans, path, extent_inserted,
 290                                    root, inode, start,
 291                                    inline_len, compressed_size,
 292                                    compress_type, compressed_pages);
 293         if (ret && ret != -ENOSPC) {
 294                 btrfs_abort_transaction(trans, root, ret);
 295                 goto out;
 296         } else if (ret == -ENOSPC) {
 297                 ret = 1;
 298                 goto out;
 299         }
 300
 301         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
 302         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 303         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 304 out:
 305         btrfs_free_path(path);
 306         btrfs_end_transaction(trans, root);
 307         return ret;
 308 }
 309
 310 struct async_extent {
 311         u64 start;
 312         u64 ram_size;
 313         u64 compressed_size;
 314         struct page **pages;
 315         unsigned long nr_pages;
 316         int compress_type;
 317         struct list_head list;
 318 };
 319
 320 struct async_cow {
 321         struct inode *inode;
 322         struct btrfs_root *root;
 323         struct page *locked_page;
 324         u64 start;
 325         u64 end;
 326         struct list_head extents;
 327         struct btrfs_work work;
 328 };
 329
 330 static noinline int add_async_extent(struct async_cow *cow,
 331                                      u64 start, u64 ram_size,
 332                                      u64 compressed_size,
 333                                      struct page **pages,
 334                                      unsigned long nr_pages,
 335                                      int compress_type)
 336 {
 337         struct async_extent *async_extent;
 338
 339         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 340         BUG_ON(!async_extent); /* -ENOMEM */
 341         async_extent->start = start;
 342         async_extent->ram_size = ram_size;
 343         async_extent->compressed_size = compressed_size;
 344         async_extent->pages = pages;
 345         async_extent->nr_pages = nr_pages;
 346         async_extent->compress_type = compress_type;
 347         list_add_tail(&async_extent->list, &cow->extents);
 348         return 0;
 349 }
 350
 351 /*
 352  * we create compressed extents in two phases.  The first
 353  * phase compresses a range of pages that have already been
 354  * locked (both pages and state bits are locked).
 355  *
 356  * This is done inside an ordered work queue, and the compression
 357  * is spread across many cpus.  The actual IO submission is step
 358  * two, and the ordered work queue takes care of making sure that
 359  * happens in the same order things were put onto the queue by
 360  * writepages and friends.
 361  *
 362  * If this code finds it can't get good compression, it puts an
 363  * entry onto the work queue to write the uncompressed bytes.  This
 364  * makes sure that both compressed inodes and uncompressed inodes
 365  * are written in the same order that the flusher thread sent them
 366  * down.
 367  */
 368 static noinline int compress_file_range(struct inode *inode,
 369                                         struct page *locked_page,
 370                                         u64 start, u64 end,
 371                                         struct async_cow *async_cow,
 372                                         int *num_added)
 373 {
 374         struct btrfs_root *root = BTRFS_I(inode)->root;
 375         u64 num_bytes;
 376         u64 blocksize = root->sectorsize;
 377         u64 actual_end;
 378         u64 isize = i_size_read(inode);
 379         int ret = 0;
 380         struct page **pages = NULL;
 381         unsigned long nr_pages;
 382         unsigned long nr_pages_ret = 0;
 383         unsigned long total_compressed = 0;
 384         unsigned long total_in = 0;
 385         unsigned long max_compressed = 128 * 1024;
 386         unsigned long max_uncompressed = 128 * 1024;
 387         int i;
 388         int will_compress;
 389         int compress_type = root->fs_info->compress_type;
 390         int redirty = 0;
 391
 392         /* if this is a small write inside eof, kick off a defrag */
 393         if ((end - start + 1) < 16 * 1024 &&
 394             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 395                 btrfs_add_inode_defrag(NULL, inode);
 396
 397         /*
 398          * skip compression for a small file range(<=blocksize) that
 399          * isn't an inline extent, since it dosen't save disk space at all.
 400          */
 401         if ((end - start + 1) <= blocksize &&
 402             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 403                 goto cleanup_and_bail_uncompressed;
 404
 405         actual_end = min_t(u64, isize, end + 1);
 406 again:
 407         will_compress = 0;
 408         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 409         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 410
 411         /*
 412          * we don't want to send crud past the end of i_size through
 413          * compression, that's just a waste of CPU time.  So, if the
 414          * end of the file is before the start of our current
 415          * requested range of bytes, we bail out to the uncompressed
 416          * cleanup code that can deal with all of this.
 417          *
 418          * It isn't really the fastest way to fix things, but this is a
 419          * very uncommon corner.
 420          */
 421         if (actual_end <= start)
 422                 goto cleanup_and_bail_uncompressed;
 423
 424         total_compressed = actual_end - start;
 425
 426         /* we want to make sure that amount of ram required to uncompress
 427          * an extent is reasonable, so we limit the total size in ram
 428          * of a compressed extent to 128k.  This is a crucial number
 429          * because it also controls how easily we can spread reads across
 430          * cpus for decompression.
 431          *
 432          * We also want to make sure the amount of IO required to do
 433          * a random read is reasonably small, so we limit the size of
 434          * a compressed extent to 128k.
 435          */
 436         total_compressed = min(total_compressed, max_uncompressed);
 437         num_bytes = ALIGN(end - start + 1, blocksize);
 438         num_bytes = max(blocksize,  num_bytes);
 439         total_in = 0;
 440         ret = 0;
 441
 442         /*
 443          * we do compression for mount -o compress and when the
 444          * inode has not been flagged as nocompress.  This flag can
 445          * change at any time if we discover bad compression ratios.
 446          */
 447         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 448             (btrfs_test_opt(root, COMPRESS) ||
 449              (BTRFS_I(inode)->force_compress) ||
 450              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 451                 WARN_ON(pages);
 452                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 453                 if (!pages) {
 454                         /* just bail out to the uncompressed code */
 455                         goto cont;
 456                 }
 457
 458                 if (BTRFS_I(inode)->force_compress)
 459                         compress_type = BTRFS_I(inode)->force_compress;
 460
 461                 /*
 462                  * we need to call clear_page_dirty_for_io on each
 463                  * page in the range.  Otherwise applications with the file
 464                  * mmap'd can wander in and change the page contents while
 465                  * we are compressing them.
 466                  *
 467                  * If the compression fails for any reason, we set the pages
 468                  * dirty again later on.
 469                  */
 470                 extent_range_clear_dirty_for_io(inode, start, end);
 471                 redirty = 1;
 472                 ret = btrfs_compress_pages(compress_type,
 473                                            inode->i_mapping, start,
 474                                            total_compressed, pages,
 475                                            nr_pages, &nr_pages_ret,
 476                                            &total_in,
 477                                            &total_compressed,
 478                                            max_compressed);
 479
 480                 if (!ret) {
 481                         unsigned long offset = total_compressed &
 482                                 (PAGE_CACHE_SIZE - 1);
 483                         struct page *page = pages[nr_pages_ret - 1];
 484                         char *kaddr;
 485
 486                         /* zero the tail end of the last page, we might be
 487                          * sending it down to disk
 488                          */
 489                         if (offset) {
 490                                 kaddr = kmap_atomic(page);
 491                                 memset(kaddr + offset, 0,
 492                                        PAGE_CACHE_SIZE - offset);
 493                                 kunmap_atomic(kaddr);
 494                         }
 495                         will_compress = 1;
 496                 }
 497         }
 498 cont:
 499         if (start == 0) {
 500                 /* lets try to make an inline extent */
 501                 if (ret || total_in < (actual_end - start)) {
 502                         /* we didn't compress the entire range, try
 503                          * to make an uncompressed inline extent.
 504                          */
 505                         ret = cow_file_range_inline(root, inode, start, end,
 506                                                     0, 0, NULL);
 507                 } else {
 508                         /* try making a compressed inline extent */
 509                         ret = cow_file_range_inline(root, inode, start, end,
 510                                                     total_compressed,
 511                                                     compress_type, pages);
 512                 }
 513                 if (ret <= 0) {
 514                         unsigned long clear_flags = EXTENT_DELALLOC |
 515                                 EXTENT_DEFRAG;
 516                         clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
 517
 518                         /*
 519                          * inline extent creation worked or returned error,
 520                          * we don't need to create any more async work items.
 521                          * Unlock and free up our temp pages.
 522                          */
 523                         extent_clear_unlock_delalloc(inode, start, end, NULL,
 524                                                      clear_flags, PAGE_UNLOCK |
 525                                                      PAGE_CLEAR_DIRTY |
 526                                                      PAGE_SET_WRITEBACK |
 527                                                      PAGE_END_WRITEBACK);
 528                         goto free_pages_out;
 529                 }
 530         }
 531
 532         if (will_compress) {
 533                 /*
 534                  * we aren't doing an inline extent round the compressed size
 535                  * up to a block size boundary so the allocator does sane
 536                  * things
 537                  */
 538                 total_compressed = ALIGN(total_compressed, blocksize);
 539
 540                 /*
 541                  * one last check to make sure the compression is really a
 542                  * win, compare the page count read with the blocks on disk
 543                  */
 544                 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
 545                 if (total_compressed >= total_in) {
 546                         will_compress = 0;
 547                 } else {
 548                         num_bytes = total_in;
 549                 }
 550         }
 551         if (!will_compress && pages) {
 552                 /*
 553                  * the compression code ran but failed to make things smaller,
 554                  * free any pages it allocated and our page pointer array
 555                  */
 556                 for (i = 0; i < nr_pages_ret; i++) {
 557                         WARN_ON(pages[i]->mapping);
 558                         page_cache_release(pages[i]);
 559                 }
 560                 kfree(pages);
 561                 pages = NULL;
 562                 total_compressed = 0;
 563                 nr_pages_ret = 0;
 564
 565                 /* flag the file so we don't compress in the future */
 566                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 567                     !(BTRFS_I(inode)->force_compress)) {
 568                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 569                 }
 570         }
 571         if (will_compress) {
 572                 *num_added += 1;
 573
 574                 /* the async work queues will take care of doing actual
 575                  * allocation on disk for these compressed pages,
 576                  * and will submit them to the elevator.
 577                  */
 578                 add_async_extent(async_cow, start, num_bytes,
 579                                  total_compressed, pages, nr_pages_ret,
 580                                  compress_type);
 581
 582                 if (start + num_bytes < end) {
 583                         start += num_bytes;
 584                         pages = NULL;
 585                         cond_resched();
 586                         goto again;
 587                 }
 588         } else {
 589 cleanup_and_bail_uncompressed:
 590                 /*
 591                  * No compression, but we still need to write the pages in
 592                  * the file we've been given so far.  redirty the locked
 593                  * page if it corresponds to our extent and set things up
 594                  * for the async work queue to run cow_file_range to do
 595                  * the normal delalloc dance
 596                  */
 597                 if (page_offset(locked_page) >= start &&
 598                     page_offset(locked_page) <= end) {
 599                         __set_page_dirty_nobuffers(locked_page);
 600                         /* unlocked later on in the async handlers */
 601                 }
 602                 if (redirty)
 603                         extent_range_redirty_for_io(inode, start, end);
 604                 add_async_extent(async_cow, start, end - start + 1,
 605                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 606                 *num_added += 1;
 607         }
 608
 609 out:
 610         return ret;
 611
 612 free_pages_out:
 613         for (i = 0; i < nr_pages_ret; i++) {
 614                 WARN_ON(pages[i]->mapping);
 615                 page_cache_release(pages[i]);
 616         }
 617         kfree(pages);
 618
 619         goto out;
 620 }
 621
 622 /*
 623  * phase two of compressed writeback.  This is the ordered portion
 624  * of the code, which only gets called in the order the work was
 625  * queued.  We walk all the async extents created by compress_file_range
 626  * and send them down to the disk.
 627  */
 628 static noinline int submit_compressed_extents(struct inode *inode,
 629                                               struct async_cow *async_cow)
 630 {
 631         struct async_extent *async_extent;
 632         u64 alloc_hint = 0;
 633         struct btrfs_key ins;
 634         struct extent_map *em;
 635         struct btrfs_root *root = BTRFS_I(inode)->root;
 636         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 637         struct extent_io_tree *io_tree;
 638         int ret = 0;
 639
 640         if (list_empty(&async_cow->extents))
 641                 return 0;
 642
 643 again:
 644         while (!list_empty(&async_cow->extents)) {
 645                 async_extent = list_entry(async_cow->extents.next,
 646                                           struct async_extent, list);
 647                 list_del(&async_extent->list);
 648
 649                 io_tree = &BTRFS_I(inode)->io_tree;
 650
 651 retry:
 652                 /* did the compression code fall back to uncompressed IO? */
 653                 if (!async_extent->pages) {
 654                         int page_started = 0;
 655                         unsigned long nr_written = 0;
 656
 657                         lock_extent(io_tree, async_extent->start,
 658                                          async_extent->start +
 659                                          async_extent->ram_size - 1);
 660
 661                         /* allocate blocks */
 662                         ret = cow_file_range(inode, async_cow->locked_page,
 663                                              async_extent->start,
 664                                              async_extent->start +
 665                                              async_extent->ram_size - 1,
 666                                              &page_started, &nr_written, 0);
 667
 668                         /* JDM XXX */
 669
 670                         /*
 671                          * if page_started, cow_file_range inserted an
 672                          * inline extent and took care of all the unlocking
 673                          * and IO for us.  Otherwise, we need to submit
 674                          * all those pages down to the drive.
 675                          */
 676                         if (!page_started && !ret)
 677                                 extent_write_locked_range(io_tree,
 678                                                   inode, async_extent->start,
 679                                                   async_extent->start +
 680                                                   async_extent->ram_size - 1,
 681                                                   btrfs_get_extent,
 682                                                   WB_SYNC_ALL);
 683                         else if (ret)
 684                                 unlock_page(async_cow->locked_page);
 685                         kfree(async_extent);
 686                         cond_resched();
 687                         continue;
 688                 }
 689
 690                 lock_extent(io_tree, async_extent->start,
 691                             async_extent->start + async_extent->ram_size - 1);
 692
 693                 ret = btrfs_reserve_extent(root,
 694                                            async_extent->compressed_size,
 695                                            async_extent->compressed_size,
 696                                            0, alloc_hint, &ins, 1, 1);
 697                 if (ret) {
 698                         int i;
 699
 700                         for (i = 0; i < async_extent->nr_pages; i++) {
 701                                 WARN_ON(async_extent->pages[i]->mapping);
 702                                 page_cache_release(async_extent->pages[i]);
 703                         }
 704                         kfree(async_extent->pages);
 705                         async_extent->nr_pages = 0;
 706                         async_extent->pages = NULL;
 707
 708                         if (ret == -ENOSPC) {
 709                                 unlock_extent(io_tree, async_extent->start,
 710                                               async_extent->start +
 711                                               async_extent->ram_size - 1);
 712                                 goto retry;
 713                         }
 714                         goto out_free;
 715                 }
 716
 717                 /*
 718                  * here we're doing allocation and writeback of the
 719                  * compressed pages
 720                  */
 721                 btrfs_drop_extent_cache(inode, async_extent->start,
 722                                         async_extent->start +
 723                                         async_extent->ram_size - 1, 0);
 724
 725                 em = alloc_extent_map();
 726                 if (!em) {
 727                         ret = -ENOMEM;
 728                         goto out_free_reserve;
 729                 }
 730                 em->start = async_extent->start;
 731                 em->len = async_extent->ram_size;
 732                 em->orig_start = em->start;
 733                 em->mod_start = em->start;
 734                 em->mod_len = em->len;
 735
 736                 em->block_start = ins.objectid;
 737                 em->block_len = ins.offset;
 738                 em->orig_block_len = ins.offset;
 739                 em->ram_bytes = async_extent->ram_size;
 740                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 741                 em->compress_type = async_extent->compress_type;
 742                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 743                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 744                 em->generation = -1;
 745
 746                 while (1) {
 747                         write_lock(&em_tree->lock);
 748                         ret = add_extent_mapping(em_tree, em, 1);
 749                         write_unlock(&em_tree->lock);
 750                         if (ret != -EEXIST) {
 751                                 free_extent_map(em);
 752                                 break;
 753                         }
 754                         btrfs_drop_extent_cache(inode, async_extent->start,
 755                                                 async_extent->start +
 756                                                 async_extent->ram_size - 1, 0);
 757                 }
 758
 759                 if (ret)
 760                         goto out_free_reserve;
 761
 762                 ret = btrfs_add_ordered_extent_compress(inode,
 763                                                 async_extent->start,
 764                                                 ins.objectid,
 765                                                 async_extent->ram_size,
 766                                                 ins.offset,
 767                                                 BTRFS_ORDERED_COMPRESSED,
 768                                                 async_extent->compress_type);
 769                 if (ret)
 770                         goto out_free_reserve;
 771
 772                 /*
 773                  * clear dirty, set writeback and unlock the pages.
 774                  */
 775                 extent_clear_unlock_delalloc(inode, async_extent->start,
 776                                 async_extent->start +
 777                                 async_extent->ram_size - 1,
 778                                 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
 779                                 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 780                                 PAGE_SET_WRITEBACK);
 781                 ret = btrfs_submit_compressed_write(inode,
 782                                     async_extent->start,
 783                                     async_extent->ram_size,
 784                                     ins.objectid,
 785                                     ins.offset, async_extent->pages,
 786                                     async_extent->nr_pages);
 787                 alloc_hint = ins.objectid + ins.offset;
 788                 kfree(async_extent);
 789                 if (ret)
 790                         goto out;
 791                 cond_resched();
 792         }
 793         ret = 0;
 794 out:
 795         return ret;
 796 out_free_reserve:
 797         btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
 798 out_free:
 799         extent_clear_unlock_delalloc(inode, async_extent->start,
 800                                      async_extent->start +
 801                                      async_extent->ram_size - 1,
 802                                      NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
 803                                      EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
 804                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 805                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
 806         kfree(async_extent);
 807         goto again;
 808 }
 809
 810 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 811                                       u64 num_bytes)
 812 {
 813         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 814         struct extent_map *em;
 815         u64 alloc_hint = 0;
 816
 817         read_lock(&em_tree->lock);
 818         em = search_extent_mapping(em_tree, start, num_bytes);
 819         if (em) {
 820                 /*
 821                  * if block start isn't an actual block number then find the
 822                  * first block in this inode and use that as a hint.  If that
 823                  * block is also bogus then just don't worry about it.
 824                  */
 825                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 826                         free_extent_map(em);
 827                         em = search_extent_mapping(em_tree, 0, 0);
 828                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 829                                 alloc_hint = em->block_start;
 830                         if (em)
 831                                 free_extent_map(em);
 832                 } else {
 833                         alloc_hint = em->block_start;
 834                         free_extent_map(em);
 835                 }
 836         }
 837         read_unlock(&em_tree->lock);
 838
 839         return alloc_hint;
 840 }
 841
 842 /*
 843  * when extent_io.c finds a delayed allocation range in the file,
 844  * the call backs end up in this code.  The basic idea is to
 845  * allocate extents on disk for the range, and create ordered data structs
 846  * in ram to track those extents.
 847  *
 848  * locked_page is the page that writepage had locked already.  We use
 849  * it to make sure we don't do extra locks or unlocks.
 850  *
 851  * *page_started is set to one if we unlock locked_page and do everything
 852  * required to start IO on it.  It may be clean and already done with
 853  * IO when we return.
 854  */
 855 static noinline int cow_file_range(struct inode *inode,
 856                                    struct page *locked_page,
 857                                    u64 start, u64 end, int *page_started,
 858                                    unsigned long *nr_written,
 859                                    int unlock)
 860 {
 861         struct btrfs_root *root = BTRFS_I(inode)->root;
 862         u64 alloc_hint = 0;
 863         u64 num_bytes;
 864         unsigned long ram_size;
 865         u64 disk_num_bytes;
 866         u64 cur_alloc_size;
 867         u64 blocksize = root->sectorsize;
 868         struct btrfs_key ins;
 869         struct extent_map *em;
 870         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 871         int ret = 0;
 872
 873         if (btrfs_is_free_space_inode(inode)) {
 874                 WARN_ON_ONCE(1);
 875                 ret = -EINVAL;
 876                 goto out_unlock;
 877         }
 878
 879         num_bytes = ALIGN(end - start + 1, blocksize);
 880         num_bytes = max(blocksize,  num_bytes);
 881         disk_num_bytes = num_bytes;
 882
 883         /* if this is a small write inside eof, kick off defrag */
 884         if (num_bytes < 64 * 1024 &&
 885             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 886                 btrfs_add_inode_defrag(NULL, inode);
 887
 888         if (start == 0) {
 889                 /* lets try to make an inline extent */
 890                 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
 891                                             NULL);
 892                 if (ret == 0) {
 893                         extent_clear_unlock_delalloc(inode, start, end, NULL,
 894                                      EXTENT_LOCKED | EXTENT_DELALLOC |
 895                                      EXTENT_DEFRAG, PAGE_UNLOCK |
 896                                      PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
 897                                      PAGE_END_WRITEBACK);
 898
 899                         *nr_written = *nr_written +
 900                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 901                         *page_started = 1;
 902                         goto out;
 903                 } else if (ret < 0) {
 904                         goto out_unlock;
 905                 }
 906         }
 907
 908         BUG_ON(disk_num_bytes >
 909                btrfs_super_total_bytes(root->fs_info->super_copy));
 910
 911         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 912         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 913
 914         while (disk_num_bytes > 0) {
 915                 unsigned long op;
 916
 917                 cur_alloc_size = disk_num_bytes;
 918                 ret = btrfs_reserve_extent(root, cur_alloc_size,
 919                                            root->sectorsize, 0, alloc_hint,
 920                                            &ins, 1, 1);
 921                 if (ret < 0)
 922                         goto out_unlock;
 923
 924                 em = alloc_extent_map();
 925                 if (!em) {
 926                         ret = -ENOMEM;
 927                         goto out_reserve;
 928                 }
 929                 em->start = start;
 930                 em->orig_start = em->start;
 931                 ram_size = ins.offset;
 932                 em->len = ins.offset;
 933                 em->mod_start = em->start;
 934                 em->mod_len = em->len;
 935
 936                 em->block_start = ins.objectid;
 937                 em->block_len = ins.offset;
 938                 em->orig_block_len = ins.offset;
 939                 em->ram_bytes = ram_size;
 940                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 941                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 942                 em->generation = -1;
 943
 944                 while (1) {
 945                         write_lock(&em_tree->lock);
 946                         ret = add_extent_mapping(em_tree, em, 1);
 947                         write_unlock(&em_tree->lock);
 948                         if (ret != -EEXIST) {
 949                                 free_extent_map(em);
 950                                 break;
 951                         }
 952                         btrfs_drop_extent_cache(inode, start,
 953                                                 start + ram_size - 1, 0);
 954                 }
 955                 if (ret)
 956                         goto out_reserve;
 957
 958                 cur_alloc_size = ins.offset;
 959                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 960                                                ram_size, cur_alloc_size, 0);
 961                 if (ret)
 962                         goto out_reserve;
 963
 964                 if (root->root_key.objectid ==
 965                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 966                         ret = btrfs_reloc_clone_csums(inode, start,
 967                                                       cur_alloc_size);
 968                         if (ret)
 969                                 goto out_reserve;
 970                 }
 971
 972                 if (disk_num_bytes < cur_alloc_size)
 973                         break;
 974
 975                 /* we're not doing compressed IO, don't unlock the first
 976                  * page (which the caller expects to stay locked), don't
 977                  * clear any dirty bits and don't set any writeback bits
 978                  *
 979                  * Do set the Private2 bit so we know this page was properly
 980                  * setup for writepage
 981                  */
 982                 op = unlock ? PAGE_UNLOCK : 0;
 983                 op |= PAGE_SET_PRIVATE2;
 984
 985                 extent_clear_unlock_delalloc(inode, start,
 986                                              start + ram_size - 1, locked_page,
 987                                              EXTENT_LOCKED | EXTENT_DELALLOC,
 988                                              op);
 989                 disk_num_bytes -= cur_alloc_size;
 990                 num_bytes -= cur_alloc_size;
 991                 alloc_hint = ins.objectid + ins.offset;
 992                 start += cur_alloc_size;
 993         }
 994 out:
 995         return ret;
 996
 997 out_reserve:
 998         btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
 999 out_unlock:
1000         extent_clear_unlock_delalloc(inode, start, end, locked_page,
1001                                      EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1002                                      EXTENT_DELALLOC | EXTENT_DEFRAG,
1003                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1004                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1005         goto out;
1006 }
1007
1008 /*
1009  * work queue call back to started compression on a file and pages
1010  */
1011 static noinline void async_cow_start(struct btrfs_work *work)
1012 {
1013         struct async_cow *async_cow;
1014         int num_added = 0;
1015         async_cow = container_of(work, struct async_cow, work);
1016
1017         compress_file_range(async_cow->inode, async_cow->locked_page,
1018                             async_cow->start, async_cow->end, async_cow,
1019                             &num_added);
1020         if (num_added == 0) {
1021                 btrfs_add_delayed_iput(async_cow->inode);
1022                 async_cow->inode = NULL;
1023         }
1024 }
1025
1026 /*
1027  * work queue call back to submit previously compressed pages
1028  */
1029 static noinline void async_cow_submit(struct btrfs_work *work)
1030 {
1031         struct async_cow *async_cow;
1032         struct btrfs_root *root;
1033         unsigned long nr_pages;
1034
1035         async_cow = container_of(work, struct async_cow, work);
1036
1037         root = async_cow->root;
1038         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1039                 PAGE_CACHE_SHIFT;
1040
1041         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1042             5 * 1024 * 1024 &&
1043             waitqueue_active(&root->fs_info->async_submit_wait))
1044                 wake_up(&root->fs_info->async_submit_wait);
1045
1046         if (async_cow->inode)
1047                 submit_compressed_extents(async_cow->inode, async_cow);
1048 }
1049
1050 static noinline void async_cow_free(struct btrfs_work *work)
1051 {
1052         struct async_cow *async_cow;
1053         async_cow = container_of(work, struct async_cow, work);
1054         if (async_cow->inode)
1055                 btrfs_add_delayed_iput(async_cow->inode);
1056         kfree(async_cow);
1057 }
1058
1059 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1060                                 u64 start, u64 end, int *page_started,
1061                                 unsigned long *nr_written)
1062 {
1063         struct async_cow *async_cow;
1064         struct btrfs_root *root = BTRFS_I(inode)->root;
1065         unsigned long nr_pages;
1066         u64 cur_end;
1067         int limit = 10 * 1024 * 1024;
1068
1069         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1070                          1, 0, NULL, GFP_NOFS);
1071         while (start < end) {
1072                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1073                 BUG_ON(!async_cow); /* -ENOMEM */
1074                 async_cow->inode = igrab(inode);
1075                 async_cow->root = root;
1076                 async_cow->locked_page = locked_page;
1077                 async_cow->start = start;
1078
1079                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1080                         cur_end = end;
1081                 else
1082                         cur_end = min(end, start + 512 * 1024 - 1);
1083
1084                 async_cow->end = cur_end;
1085                 INIT_LIST_HEAD(&async_cow->extents);
1086
1087                 btrfs_init_work(&async_cow->work, async_cow_start,
1088                                 async_cow_submit, async_cow_free);
1089
1090                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1091                         PAGE_CACHE_SHIFT;
1092                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1093
1094                 btrfs_queue_work(root->fs_info->delalloc_workers,
1095                                  &async_cow->work);
1096
1097                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1098                         wait_event(root->fs_info->async_submit_wait,
1099                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1100                             limit));
1101                 }
1102
1103                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1104                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1105                         wait_event(root->fs_info->async_submit_wait,
1106                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1107                            0));
1108                 }
1109
1110                 *nr_written += nr_pages;
1111                 start = cur_end + 1;
1112         }
1113         *page_started = 1;
1114         return 0;
1115 }
1116
1117 static noinline int csum_exist_in_range(struct btrfs_root *root,
1118                                         u64 bytenr, u64 num_bytes)
1119 {
1120         int ret;
1121         struct btrfs_ordered_sum *sums;
1122         LIST_HEAD(list);
1123
1124         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1125                                        bytenr + num_bytes - 1, &list, 0);
1126         if (ret == 0 && list_empty(&list))
1127                 return 0;
1128
1129         while (!list_empty(&list)) {
1130                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1131                 list_del(&sums->list);
1132                 kfree(sums);
1133         }
1134         return 1;
1135 }
1136
1137 /*
1138  * when nowcow writeback call back.  This checks for snapshots or COW copies
1139  * of the extents that exist in the file, and COWs the file as required.
1140  *
1141  * If no cow copies or snapshots exist, we write directly to the existing
1142  * blocks on disk
1143  */
1144 static noinline int run_delalloc_nocow(struct inode *inode,
1145                                        struct page *locked_page,
1146                               u64 start, u64 end, int *page_started, int force,
1147                               unsigned long *nr_written)
1148 {
1149         struct btrfs_root *root = BTRFS_I(inode)->root;
1150         struct btrfs_trans_handle *trans;
1151         struct extent_buffer *leaf;
1152         struct btrfs_path *path;
1153         struct btrfs_file_extent_item *fi;
1154         struct btrfs_key found_key;
1155         u64 cow_start;
1156         u64 cur_offset;
1157         u64 extent_end;
1158         u64 extent_offset;
1159         u64 disk_bytenr;
1160         u64 num_bytes;
1161         u64 disk_num_bytes;
1162         u64 ram_bytes;
1163         int extent_type;
1164         int ret, err;
1165         int type;
1166         int nocow;
1167         int check_prev = 1;
1168         bool nolock;
1169         u64 ino = btrfs_ino(inode);
1170
1171         path = btrfs_alloc_path();
1172         if (!path) {
1173                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1174                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1175                                              EXTENT_DO_ACCOUNTING |
1176                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1177                                              PAGE_CLEAR_DIRTY |
1178                                              PAGE_SET_WRITEBACK |
1179                                              PAGE_END_WRITEBACK);
1180                 return -ENOMEM;
1181         }
1182
1183         nolock = btrfs_is_free_space_inode(inode);
1184
1185         if (nolock)
1186                 trans = btrfs_join_transaction_nolock(root);
1187         else
1188                 trans = btrfs_join_transaction(root);
1189
1190         if (IS_ERR(trans)) {
1191                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1192                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1193                                              EXTENT_DO_ACCOUNTING |
1194                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1195                                              PAGE_CLEAR_DIRTY |
1196                                              PAGE_SET_WRITEBACK |
1197                                              PAGE_END_WRITEBACK);
1198                 btrfs_free_path(path);
1199                 return PTR_ERR(trans);
1200         }
1201
1202         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1203
1204         cow_start = (u64)-1;
1205         cur_offset = start;
1206         while (1) {
1207                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1208                                                cur_offset, 0);
1209                 if (ret < 0)
1210                         goto error;
1211                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1212                         leaf = path->nodes[0];
1213                         btrfs_item_key_to_cpu(leaf, &found_key,
1214                                               path->slots[0] - 1);
1215                         if (found_key.objectid == ino &&
1216                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1217                                 path->slots[0]--;
1218                 }
1219                 check_prev = 0;
1220 next_slot:
1221                 leaf = path->nodes[0];
1222                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1223                         ret = btrfs_next_leaf(root, path);
1224                         if (ret < 0)
1225                                 goto error;
1226                         if (ret > 0)
1227                                 break;
1228                         leaf = path->nodes[0];
1229                 }
1230
1231                 nocow = 0;
1232                 disk_bytenr = 0;
1233                 num_bytes = 0;
1234                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1235
1236                 if (found_key.objectid > ino ||
1237                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1238                     found_key.offset > end)
1239                         break;
1240
1241                 if (found_key.offset > cur_offset) {
1242                         extent_end = found_key.offset;
1243                         extent_type = 0;
1244                         goto out_check;
1245                 }
1246
1247                 fi = btrfs_item_ptr(leaf, path->slots[0],
1248                                     struct btrfs_file_extent_item);
1249                 extent_type = btrfs_file_extent_type(leaf, fi);
1250
1251                 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1252                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1253                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1254                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1255                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1256                         extent_end = found_key.offset +
1257                                 btrfs_file_extent_num_bytes(leaf, fi);
1258                         disk_num_bytes =
1259                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1260                         if (extent_end <= start) {
1261                                 path->slots[0]++;
1262                                 goto next_slot;
1263                         }
1264                         if (disk_bytenr == 0)
1265                                 goto out_check;
1266                         if (btrfs_file_extent_compression(leaf, fi) ||
1267                             btrfs_file_extent_encryption(leaf, fi) ||
1268                             btrfs_file_extent_other_encoding(leaf, fi))
1269                                 goto out_check;
1270                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1271                                 goto out_check;
1272                         if (btrfs_extent_readonly(root, disk_bytenr))
1273                                 goto out_check;
1274                         if (btrfs_cross_ref_exist(trans, root, ino,
1275                                                   found_key.offset -
1276                                                   extent_offset, disk_bytenr))
1277                                 goto out_check;
1278                         disk_bytenr += extent_offset;
1279                         disk_bytenr += cur_offset - found_key.offset;
1280                         num_bytes = min(end + 1, extent_end) - cur_offset;
1281                         /*
1282                          * if there are pending snapshots for this root,
1283                          * we fall into common COW way.
1284                          */
1285                         if (!nolock) {
1286                                 err = btrfs_start_nocow_write(root);
1287                                 if (!err)
1288                                         goto out_check;
1289                         }
1290                         /*
1291                          * force cow if csum exists in the range.
1292                          * this ensure that csum for a given extent are
1293                          * either valid or do not exist.
1294                          */
1295                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1296                                 goto out_check;
1297                         nocow = 1;
1298                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1299                         extent_end = found_key.offset +
1300                                 btrfs_file_extent_inline_len(leaf,
1301                                                      path->slots[0], fi);
1302                         extent_end = ALIGN(extent_end, root->sectorsize);
1303                 } else {
1304                         BUG_ON(1);
1305                 }
1306 out_check:
1307                 if (extent_end <= start) {
1308                         path->slots[0]++;
1309                         if (!nolock && nocow)
1310                                 btrfs_end_nocow_write(root);
1311                         goto next_slot;
1312                 }
1313                 if (!nocow) {
1314                         if (cow_start == (u64)-1)
1315                                 cow_start = cur_offset;
1316                         cur_offset = extent_end;
1317                         if (cur_offset > end)
1318                                 break;
1319                         path->slots[0]++;
1320                         goto next_slot;
1321                 }
1322
1323                 btrfs_release_path(path);
1324                 if (cow_start != (u64)-1) {
1325                         ret = cow_file_range(inode, locked_page,
1326                                              cow_start, found_key.offset - 1,
1327                                              page_started, nr_written, 1);
1328                         if (ret) {
1329                                 if (!nolock && nocow)
1330                                         btrfs_end_nocow_write(root);
1331                                 goto error;
1332                         }
1333                         cow_start = (u64)-1;
1334                 }
1335
1336                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1337                         struct extent_map *em;
1338                         struct extent_map_tree *em_tree;
1339                         em_tree = &BTRFS_I(inode)->extent_tree;
1340                         em = alloc_extent_map();
1341                         BUG_ON(!em); /* -ENOMEM */
1342                         em->start = cur_offset;
1343                         em->orig_start = found_key.offset - extent_offset;
1344                         em->len = num_bytes;
1345                         em->block_len = num_bytes;
1346                         em->block_start = disk_bytenr;
1347                         em->orig_block_len = disk_num_bytes;
1348                         em->ram_bytes = ram_bytes;
1349                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1350                         em->mod_start = em->start;
1351                         em->mod_len = em->len;
1352                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1353                         set_bit(EXTENT_FLAG_FILLING, &em->flags);
1354                         em->generation = -1;
1355                         while (1) {
1356                                 write_lock(&em_tree->lock);
1357                                 ret = add_extent_mapping(em_tree, em, 1);
1358                                 write_unlock(&em_tree->lock);
1359                                 if (ret != -EEXIST) {
1360                                         free_extent_map(em);
1361                                         break;
1362                                 }
1363                                 btrfs_drop_extent_cache(inode, em->start,
1364                                                 em->start + em->len - 1, 0);
1365                         }
1366                         type = BTRFS_ORDERED_PREALLOC;
1367                 } else {
1368                         type = BTRFS_ORDERED_NOCOW;
1369                 }
1370
1371                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1372                                                num_bytes, num_bytes, type);
1373                 BUG_ON(ret); /* -ENOMEM */
1374
1375                 if (root->root_key.objectid ==
1376                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1377                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1378                                                       num_bytes);
1379                         if (ret) {
1380                                 if (!nolock && nocow)
1381                                         btrfs_end_nocow_write(root);
1382                                 goto error;
1383                         }
1384                 }
1385
1386                 extent_clear_unlock_delalloc(inode, cur_offset,
1387                                              cur_offset + num_bytes - 1,
1388                                              locked_page, EXTENT_LOCKED |
1389                                              EXTENT_DELALLOC, PAGE_UNLOCK |
1390                                              PAGE_SET_PRIVATE2);
1391                 if (!nolock && nocow)
1392                         btrfs_end_nocow_write(root);
1393                 cur_offset = extent_end;
1394                 if (cur_offset > end)
1395                         break;
1396         }
1397         btrfs_release_path(path);
1398
1399         if (cur_offset <= end && cow_start == (u64)-1) {
1400                 cow_start = cur_offset;
1401                 cur_offset = end;
1402         }
1403
1404         if (cow_start != (u64)-1) {
1405                 ret = cow_file_range(inode, locked_page, cow_start, end,
1406                                      page_started, nr_written, 1);
1407                 if (ret)
1408                         goto error;
1409         }
1410
1411 error:
1412         err = btrfs_end_transaction(trans, root);
1413         if (!ret)
1414                 ret = err;
1415
1416         if (ret && cur_offset < end)
1417                 extent_clear_unlock_delalloc(inode, cur_offset, end,
1418                                              locked_page, EXTENT_LOCKED |
1419                                              EXTENT_DELALLOC | EXTENT_DEFRAG |
1420                                              EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1421                                              PAGE_CLEAR_DIRTY |
1422                                              PAGE_SET_WRITEBACK |
1423                                              PAGE_END_WRITEBACK);
1424         btrfs_free_path(path);
1425         return ret;
1426 }
1427
1428 /*
1429  * extent_io.c call back to do delayed allocation processing
1430  */
1431 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1432                               u64 start, u64 end, int *page_started,
1433                               unsigned long *nr_written)
1434 {
1435         int ret;
1436         struct btrfs_root *root = BTRFS_I(inode)->root;
1437
1438         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1439                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1440                                          page_started, 1, nr_written);
1441         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1442                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1443                                          page_started, 0, nr_written);
1444         } else if (!btrfs_test_opt(root, COMPRESS) &&
1445                    !(BTRFS_I(inode)->force_compress) &&
1446                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1447                 ret = cow_file_range(inode, locked_page, start, end,
1448                                       page_started, nr_written, 1);
1449         } else {
1450                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1451                         &BTRFS_I(inode)->runtime_flags);
1452                 ret = cow_file_range_async(inode, locked_page, start, end,
1453                                            page_started, nr_written);
1454         }
1455         return ret;
1456 }
1457
1458 static void btrfs_split_extent_hook(struct inode *inode,
1459                                     struct extent_state *orig, u64 split)
1460 {
1461         /* not delalloc, ignore it */
1462         if (!(orig->state & EXTENT_DELALLOC))
1463                 return;
1464
1465         spin_lock(&BTRFS_I(inode)->lock);
1466         BTRFS_I(inode)->outstanding_extents++;
1467         spin_unlock(&BTRFS_I(inode)->lock);
1468 }
1469
1470 /*
1471  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1472  * extents so we can keep track of new extents that are just merged onto old
1473  * extents, such as when we are doing sequential writes, so we can properly
1474  * account for the metadata space we'll need.
1475  */
1476 static void btrfs_merge_extent_hook(struct inode *inode,
1477                                     struct extent_state *new,
1478                                     struct extent_state *other)
1479 {
1480         /* not delalloc, ignore it */
1481         if (!(other->state & EXTENT_DELALLOC))
1482                 return;
1483
1484         spin_lock(&BTRFS_I(inode)->lock);
1485         BTRFS_I(inode)->outstanding_extents--;
1486         spin_unlock(&BTRFS_I(inode)->lock);
1487 }
1488
1489 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1490                                       struct inode *inode)
1491 {
1492         spin_lock(&root->delalloc_lock);
1493         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1494                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1495                               &root->delalloc_inodes);
1496                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1497                         &BTRFS_I(inode)->runtime_flags);
1498                 root->nr_delalloc_inodes++;
1499                 if (root->nr_delalloc_inodes == 1) {
1500                         spin_lock(&root->fs_info->delalloc_root_lock);
1501                         BUG_ON(!list_empty(&root->delalloc_root));
1502                         list_add_tail(&root->delalloc_root,
1503                                       &root->fs_info->delalloc_roots);
1504                         spin_unlock(&root->fs_info->delalloc_root_lock);
1505                 }
1506         }
1507         spin_unlock(&root->delalloc_lock);
1508 }
1509
1510 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1511                                      struct inode *inode)
1512 {
1513         spin_lock(&root->delalloc_lock);
1514         if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1515                 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1516                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1517                           &BTRFS_I(inode)->runtime_flags);
1518                 root->nr_delalloc_inodes--;
1519                 if (!root->nr_delalloc_inodes) {
1520                         spin_lock(&root->fs_info->delalloc_root_lock);
1521                         BUG_ON(list_empty(&root->delalloc_root));
1522                         list_del_init(&root->delalloc_root);
1523                         spin_unlock(&root->fs_info->delalloc_root_lock);
1524                 }
1525         }
1526         spin_unlock(&root->delalloc_lock);
1527 }
1528
1529 /*
1530  * extent_io.c set_bit_hook, used to track delayed allocation
1531  * bytes in this file, and to maintain the list of inodes that
1532  * have pending delalloc work to be done.
1533  */
1534 static void btrfs_set_bit_hook(struct inode *inode,
1535                                struct extent_state *state, unsigned long *bits)
1536 {
1537
1538         /*
1539          * set_bit and clear bit hooks normally require _irqsave/restore
1540          * but in this case, we are only testing for the DELALLOC
1541          * bit, which is only set or cleared with irqs on
1542          */
1543         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1544                 struct btrfs_root *root = BTRFS_I(inode)->root;
1545                 u64 len = state->end + 1 - state->start;
1546                 bool do_list = !btrfs_is_free_space_inode(inode);
1547
1548                 if (*bits & EXTENT_FIRST_DELALLOC) {
1549                         *bits &= ~EXTENT_FIRST_DELALLOC;
1550                 } else {
1551                         spin_lock(&BTRFS_I(inode)->lock);
1552                         BTRFS_I(inode)->outstanding_extents++;
1553                         spin_unlock(&BTRFS_I(inode)->lock);
1554                 }
1555
1556                 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1557                                      root->fs_info->delalloc_batch);
1558                 spin_lock(&BTRFS_I(inode)->lock);
1559                 BTRFS_I(inode)->delalloc_bytes += len;
1560                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1561                                          &BTRFS_I(inode)->runtime_flags))
1562                         btrfs_add_delalloc_inodes(root, inode);
1563                 spin_unlock(&BTRFS_I(inode)->lock);
1564         }
1565 }
1566
1567 /*
1568  * extent_io.c clear_bit_hook, see set_bit_hook for why
1569  */
1570 static void btrfs_clear_bit_hook(struct inode *inode,
1571                                  struct extent_state *state,
1572                                  unsigned long *bits)
1573 {
1574         /*
1575          * set_bit and clear bit hooks normally require _irqsave/restore
1576          * but in this case, we are only testing for the DELALLOC
1577          * bit, which is only set or cleared with irqs on
1578          */
1579         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1580                 struct btrfs_root *root = BTRFS_I(inode)->root;
1581                 u64 len = state->end + 1 - state->start;
1582                 bool do_list = !btrfs_is_free_space_inode(inode);
1583
1584                 if (*bits & EXTENT_FIRST_DELALLOC) {
1585                         *bits &= ~EXTENT_FIRST_DELALLOC;
1586                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1587                         spin_lock(&BTRFS_I(inode)->lock);
1588                         BTRFS_I(inode)->outstanding_extents--;
1589                         spin_unlock(&BTRFS_I(inode)->lock);
1590                 }
1591
1592                 /*
1593                  * We don't reserve metadata space for space cache inodes so we
1594                  * don't need to call dellalloc_release_metadata if there is an
1595                  * error.
1596                  */
1597                 if (*bits & EXTENT_DO_ACCOUNTING &&
1598                     root != root->fs_info->tree_root)
1599                         btrfs_delalloc_release_metadata(inode, len);
1600
1601                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1602                     && do_list && !(state->state & EXTENT_NORESERVE))
1603                         btrfs_free_reserved_data_space(inode, len);
1604
1605                 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1606                                      root->fs_info->delalloc_batch);
1607                 spin_lock(&BTRFS_I(inode)->lock);
1608                 BTRFS_I(inode)->delalloc_bytes -= len;
1609                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1610                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1611                              &BTRFS_I(inode)->runtime_flags))
1612                         btrfs_del_delalloc_inode(root, inode);
1613                 spin_unlock(&BTRFS_I(inode)->lock);
1614         }
1615 }
1616
1617 /*
1618  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1619  * we don't create bios that span stripes or chunks
1620  */
1621 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1622                          size_t size, struct bio *bio,
1623                          unsigned long bio_flags)
1624 {
1625         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1626         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1627         u64 length = 0;
1628         u64 map_length;
1629         int ret;
1630
1631         if (bio_flags & EXTENT_BIO_COMPRESSED)
1632                 return 0;
1633
1634         length = bio->bi_iter.bi_size;
1635         map_length = length;
1636         ret = btrfs_map_block(root->fs_info, rw, logical,
1637                               &map_length, NULL, 0);
1638         /* Will always return 0 with map_multi == NULL */
1639         BUG_ON(ret < 0);
1640         if (map_length < length + size)
1641                 return 1;
1642         return 0;
1643 }
1644
1645 /*
1646  * in order to insert checksums into the metadata in large chunks,
1647  * we wait until bio submission time.   All the pages in the bio are
1648  * checksummed and sums are attached onto the ordered extent record.
1649  *
1650  * At IO completion time the cums attached on the ordered extent record
1651  * are inserted into the btree
1652  */
1653 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1654                                     struct bio *bio, int mirror_num,
1655                                     unsigned long bio_flags,
1656                                     u64 bio_offset)
1657 {
1658         struct btrfs_root *root = BTRFS_I(inode)->root;
1659         int ret = 0;
1660
1661         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1662         BUG_ON(ret); /* -ENOMEM */
1663         return 0;
1664 }
1665
1666 /*
1667  * in order to insert checksums into the metadata in large chunks,
1668  * we wait until bio submission time.   All the pages in the bio are
1669  * checksummed and sums are attached onto the ordered extent record.
1670  *
1671  * At IO completion time the cums attached on the ordered extent record
1672  * are inserted into the btree
1673  */
1674 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1675                           int mirror_num, unsigned long bio_flags,
1676                           u64 bio_offset)
1677 {
1678         struct btrfs_root *root = BTRFS_I(inode)->root;
1679         int ret;
1680
1681         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1682         if (ret)
1683                 bio_endio(bio, ret);
1684         return ret;
1685 }
1686
1687 /*
1688  * extent_io.c submission hook. This does the right thing for csum calculation
1689  * on write, or reading the csums from the tree before a read
1690  */
1691 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1692                           int mirror_num, unsigned long bio_flags,
1693                           u64 bio_offset)
1694 {
1695         struct btrfs_root *root = BTRFS_I(inode)->root;
1696         int ret = 0;
1697         int skip_sum;
1698         int metadata = 0;
1699         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1700
1701         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1702
1703         if (btrfs_is_free_space_inode(inode))
1704                 metadata = 2;
1705
1706         if (!(rw & REQ_WRITE)) {
1707                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1708                 if (ret)
1709                         goto out;
1710
1711                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1712                         ret = btrfs_submit_compressed_read(inode, bio,
1713                                                            mirror_num,
1714                                                            bio_flags);
1715                         goto out;
1716                 } else if (!skip_sum) {
1717                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1718                         if (ret)
1719                                 goto out;
1720                 }
1721                 goto mapit;
1722         } else if (async && !skip_sum) {
1723                 /* csum items have already been cloned */
1724                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1725                         goto mapit;
1726                 /* we're doing a write, do the async checksumming */
1727                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1728                                    inode, rw, bio, mirror_num,
1729                                    bio_flags, bio_offset,
1730                                    __btrfs_submit_bio_start,
1731                                    __btrfs_submit_bio_done);
1732                 goto out;
1733         } else if (!skip_sum) {
1734                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1735                 if (ret)
1736                         goto out;
1737         }
1738
1739 mapit:
1740         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1741
1742 out:
1743         if (ret < 0)
1744                 bio_endio(bio, ret);
1745         return ret;
1746 }
1747
1748 /*
1749  * given a list of ordered sums record them in the inode.  This happens
1750  * at IO completion time based on sums calculated at bio submission time.
1751  */
1752 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1753                              struct inode *inode, u64 file_offset,
1754                              struct list_head *list)
1755 {
1756         struct btrfs_ordered_sum *sum;
1757
1758         list_for_each_entry(sum, list, list) {
1759                 trans->adding_csums = 1;
1760                 btrfs_csum_file_blocks(trans,
1761                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1762                 trans->adding_csums = 0;
1763         }
1764         return 0;
1765 }
1766
1767 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1768                               struct extent_state **cached_state)
1769 {
1770         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1771         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1772                                    cached_state, GFP_NOFS);
1773 }
1774
1775 /* see btrfs_writepage_start_hook for details on why this is required */
1776 struct btrfs_writepage_fixup {
1777         struct page *page;
1778         struct btrfs_work work;
1779 };
1780
1781 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1782 {
1783         struct btrfs_writepage_fixup *fixup;
1784         struct btrfs_ordered_extent *ordered;
1785         struct extent_state *cached_state = NULL;
1786         struct page *page;
1787         struct inode *inode;
1788         u64 page_start;
1789         u64 page_end;
1790         int ret;
1791
1792         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1793         page = fixup->page;
1794 again:
1795         lock_page(page);
1796         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1797                 ClearPageChecked(page);
1798                 goto out_page;
1799         }
1800
1801         inode = page->mapping->host;
1802         page_start = page_offset(page);
1803         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1804
1805         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1806                          &cached_state);
1807
1808         /* already ordered? We're done */
1809         if (PagePrivate2(page))
1810                 goto out;
1811
1812         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1813         if (ordered) {
1814                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1815                                      page_end, &cached_state, GFP_NOFS);
1816                 unlock_page(page);
1817                 btrfs_start_ordered_extent(inode, ordered, 1);
1818                 btrfs_put_ordered_extent(ordered);
1819                 goto again;
1820         }
1821
1822         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1823         if (ret) {
1824                 mapping_set_error(page->mapping, ret);
1825                 end_extent_writepage(page, ret, page_start, page_end);
1826                 ClearPageChecked(page);
1827                 goto out;
1828          }
1829
1830         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1831         ClearPageChecked(page);
1832         set_page_dirty(page);
1833 out:
1834         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1835                              &cached_state, GFP_NOFS);
1836 out_page:
1837         unlock_page(page);
1838         page_cache_release(page);
1839         kfree(fixup);
1840 }
1841
1842 /*
1843  * There are a few paths in the higher layers of the kernel that directly
1844  * set the page dirty bit without asking the filesystem if it is a
1845  * good idea.  This causes problems because we want to make sure COW
1846  * properly happens and the data=ordered rules are followed.
1847  *
1848  * In our case any range that doesn't have the ORDERED bit set
1849  * hasn't been properly setup for IO.  We kick off an async process
1850  * to fix it up.  The async helper will wait for ordered extents, set
1851  * the delalloc bit and make it safe to write the page.
1852  */
1853 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1854 {
1855         struct inode *inode = page->mapping->host;
1856         struct btrfs_writepage_fixup *fixup;
1857         struct btrfs_root *root = BTRFS_I(inode)->root;
1858
1859         /* this page is properly in the ordered list */
1860         if (TestClearPagePrivate2(page))
1861                 return 0;
1862
1863         if (PageChecked(page))
1864                 return -EAGAIN;
1865
1866         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1867         if (!fixup)
1868                 return -EAGAIN;
1869
1870         SetPageChecked(page);
1871         page_cache_get(page);
1872         btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
1873         fixup->page = page;
1874         btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
1875         return -EBUSY;
1876 }
1877
1878 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1879                                        struct inode *inode, u64 file_pos,
1880                                        u64 disk_bytenr, u64 disk_num_bytes,
1881                                        u64 num_bytes, u64 ram_bytes,
1882                                        u8 compression, u8 encryption,
1883                                        u16 other_encoding, int extent_type)
1884 {
1885         struct btrfs_root *root = BTRFS_I(inode)->root;
1886         struct btrfs_file_extent_item *fi;
1887         struct btrfs_path *path;
1888         struct extent_buffer *leaf;
1889         struct btrfs_key ins;
1890         int extent_inserted = 0;
1891         int ret;
1892
1893         path = btrfs_alloc_path();
1894         if (!path)
1895                 return -ENOMEM;
1896
1897         /*
1898          * we may be replacing one extent in the tree with another.
1899          * The new extent is pinned in the extent map, and we don't want
1900          * to drop it from the cache until it is completely in the btree.
1901          *
1902          * So, tell btrfs_drop_extents to leave this extent in the cache.
1903          * the caller is expected to unpin it and allow it to be merged
1904          * with the others.
1905          */
1906         ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
1907                                    file_pos + num_bytes, NULL, 0,
1908                                    1, sizeof(*fi), &extent_inserted);
1909         if (ret)
1910                 goto out;
1911
1912         if (!extent_inserted) {
1913                 ins.objectid = btrfs_ino(inode);
1914                 ins.offset = file_pos;
1915                 ins.type = BTRFS_EXTENT_DATA_KEY;
1916
1917                 path->leave_spinning = 1;
1918                 ret = btrfs_insert_empty_item(trans, root, path, &ins,
1919                                               sizeof(*fi));
1920                 if (ret)
1921                         goto out;
1922         }
1923         leaf = path->nodes[0];
1924         fi = btrfs_item_ptr(leaf, path->slots[0],
1925                             struct btrfs_file_extent_item);
1926         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1927         btrfs_set_file_extent_type(leaf, fi, extent_type);
1928         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1929         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1930         btrfs_set_file_extent_offset(leaf, fi, 0);
1931         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1932         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1933         btrfs_set_file_extent_compression(leaf, fi, compression);
1934         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1935         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1936
1937         btrfs_mark_buffer_dirty(leaf);
1938         btrfs_release_path(path);
1939
1940         inode_add_bytes(inode, num_bytes);
1941
1942         ins.objectid = disk_bytenr;
1943         ins.offset = disk_num_bytes;
1944         ins.type = BTRFS_EXTENT_ITEM_KEY;
1945         ret = btrfs_alloc_reserved_file_extent(trans, root,
1946                                         root->root_key.objectid,
1947                                         btrfs_ino(inode), file_pos, &ins);
1948 out:
1949         btrfs_free_path(path);
1950
1951         return ret;
1952 }
1953
1954 /* snapshot-aware defrag */
1955 struct sa_defrag_extent_backref {
1956         struct rb_node node;
1957         struct old_sa_defrag_extent *old;
1958         u64 root_id;
1959         u64 inum;
1960         u64 file_pos;
1961         u64 extent_offset;
1962         u64 num_bytes;
1963         u64 generation;
1964 };
1965
1966 struct old_sa_defrag_extent {
1967         struct list_head list;
1968         struct new_sa_defrag_extent *new;
1969
1970         u64 extent_offset;
1971         u64 bytenr;
1972         u64 offset;
1973         u64 len;
1974         int count;
1975 };
1976
1977 struct new_sa_defrag_extent {
1978         struct rb_root root;
1979         struct list_head head;
1980         struct btrfs_path *path;
1981         struct inode *inode;
1982         u64 file_pos;
1983         u64 len;
1984         u64 bytenr;
1985         u64 disk_len;
1986         u8 compress_type;
1987 };
1988
1989 static int backref_comp(struct sa_defrag_extent_backref *b1,
1990                         struct sa_defrag_extent_backref *b2)
1991 {
1992         if (b1->root_id < b2->root_id)
1993                 return -1;
1994         else if (b1->root_id > b2->root_id)
1995                 return 1;
1996
1997         if (b1->inum < b2->inum)
1998                 return -1;
1999         else if (b1->inum > b2->inum)
2000                 return 1;
2001
2002         if (b1->file_pos < b2->file_pos)
2003                 return -1;
2004         else if (b1->file_pos > b2->file_pos)
2005                 return 1;
2006
2007         /*
2008          * [------------------------------] ===> (a range of space)
2009          *     |<--->|   |<---->| =============> (fs/file tree A)
2010          * |<---------------------------->| ===> (fs/file tree B)
2011          *
2012          * A range of space can refer to two file extents in one tree while
2013          * refer to only one file extent in another tree.
2014          *
2015          * So we may process a disk offset more than one time(two extents in A)
2016          * and locate at the same extent(one extent in B), then insert two same
2017          * backrefs(both refer to the extent in B).
2018          */
2019         return 0;
2020 }
2021
2022 static void backref_insert(struct rb_root *root,
2023                            struct sa_defrag_extent_backref *backref)
2024 {
2025         struct rb_node **p = &root->rb_node;
2026         struct rb_node *parent = NULL;
2027         struct sa_defrag_extent_backref *entry;
2028         int ret;
2029
2030         while (*p) {
2031                 parent = *p;
2032                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2033
2034                 ret = backref_comp(backref, entry);
2035                 if (ret < 0)
2036                         p = &(*p)->rb_left;
2037                 else
2038                         p = &(*p)->rb_right;
2039         }
2040
2041         rb_link_node(&backref->node, parent, p);
2042         rb_insert_color(&backref->node, root);
2043 }
2044
2045 /*
2046  * Note the backref might has changed, and in this case we just return 0.
2047  */
2048 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2049                                        void *ctx)
2050 {
2051         struct btrfs_file_extent_item *extent;
2052         struct btrfs_fs_info *fs_info;
2053         struct old_sa_defrag_extent *old = ctx;
2054         struct new_sa_defrag_extent *new = old->new;
2055         struct btrfs_path *path = new->path;
2056         struct btrfs_key key;
2057         struct btrfs_root *root;
2058         struct sa_defrag_extent_backref *backref;
2059         struct extent_buffer *leaf;
2060         struct inode *inode = new->inode;
2061         int slot;
2062         int ret;
2063         u64 extent_offset;
2064         u64 num_bytes;
2065
2066         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2067             inum == btrfs_ino(inode))
2068                 return 0;
2069
2070         key.objectid = root_id;
2071         key.type = BTRFS_ROOT_ITEM_KEY;
2072         key.offset = (u64)-1;
2073
2074         fs_info = BTRFS_I(inode)->root->fs_info;
2075         root = btrfs_read_fs_root_no_name(fs_info, &key);
2076         if (IS_ERR(root)) {
2077                 if (PTR_ERR(root) == -ENOENT)
2078                         return 0;
2079                 WARN_ON(1);
2080                 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2081                          inum, offset, root_id);
2082                 return PTR_ERR(root);
2083         }
2084
2085         key.objectid = inum;
2086         key.type = BTRFS_EXTENT_DATA_KEY;
2087         if (offset > (u64)-1 << 32)
2088                 key.offset = 0;
2089         else
2090                 key.offset = offset;
2091
2092         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2093         if (WARN_ON(ret < 0))
2094                 return ret;
2095         ret = 0;
2096
2097         while (1) {
2098                 cond_resched();
2099
2100                 leaf = path->nodes[0];
2101                 slot = path->slots[0];
2102
2103                 if (slot >= btrfs_header_nritems(leaf)) {
2104                         ret = btrfs_next_leaf(root, path);
2105                         if (ret < 0) {
2106                                 goto out;
2107                         } else if (ret > 0) {
2108                                 ret = 0;
2109                                 goto out;
2110                         }
2111                         continue;
2112                 }
2113
2114                 path->slots[0]++;
2115
2116                 btrfs_item_key_to_cpu(leaf, &key, slot);
2117
2118                 if (key.objectid > inum)
2119                         goto out;
2120
2121                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2122                         continue;
2123
2124                 extent = btrfs_item_ptr(leaf, slot,
2125                                         struct btrfs_file_extent_item);
2126
2127                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2128                         continue;
2129
2130                 /*
2131                  * 'offset' refers to the exact key.offset,
2132                  * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2133                  * (key.offset - extent_offset).
2134                  */
2135                 if (key.offset != offset)
2136                         continue;
2137
2138                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2139                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2140
2141                 if (extent_offset >= old->extent_offset + old->offset +
2142                     old->len || extent_offset + num_bytes <=
2143                     old->extent_offset + old->offset)
2144                         continue;
2145                 break;
2146         }
2147
2148         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2149         if (!backref) {
2150                 ret = -ENOENT;
2151                 goto out;
2152         }
2153
2154         backref->root_id = root_id;
2155         backref->inum = inum;
2156         backref->file_pos = offset;
2157         backref->num_bytes = num_bytes;
2158         backref->extent_offset = extent_offset;
2159         backref->generation = btrfs_file_extent_generation(leaf, extent);
2160         backref->old = old;
2161         backref_insert(&new->root, backref);
2162         old->count++;
2163 out:
2164         btrfs_release_path(path);
2165         WARN_ON(ret);
2166         return ret;
2167 }
2168
2169 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2170                                    struct new_sa_defrag_extent *new)
2171 {
2172         struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2173         struct old_sa_defrag_extent *old, *tmp;
2174         int ret;
2175
2176         new->path = path;
2177
2178         list_for_each_entry_safe(old, tmp, &new->head, list) {
2179                 ret = iterate_inodes_from_logical(old->bytenr +
2180                                                   old->extent_offset, fs_info,
2181                                                   path, record_one_backref,
2182                                                   old);
2183                 if (ret < 0 && ret != -ENOENT)
2184                         return false;
2185
2186                 /* no backref to be processed for this extent */
2187                 if (!old->count) {
2188                         list_del(&old->list);
2189                         kfree(old);
2190                 }
2191         }
2192
2193         if (list_empty(&new->head))
2194                 return false;
2195
2196         return true;
2197 }
2198
2199 static int relink_is_mergable(struct extent_buffer *leaf,
2200                               struct btrfs_file_extent_item *fi,
2201                               struct new_sa_defrag_extent *new)
2202 {
2203         if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2204                 return 0;
2205
2206         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2207                 return 0;
2208
2209         if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2210                 return 0;
2211
2212         if (btrfs_file_extent_encryption(leaf, fi) ||
2213             btrfs_file_extent_other_encoding(leaf, fi))
2214                 return 0;
2215
2216         return 1;
2217 }
2218
2219 /*
2220  * Note the backref might has changed, and in this case we just return 0.
2221  */
2222 static noinline int relink_extent_backref(struct btrfs_path *path,
2223                                  struct sa_defrag_extent_backref *prev,
2224                                  struct sa_defrag_extent_backref *backref)
2225 {
2226         struct btrfs_file_extent_item *extent;
2227         struct btrfs_file_extent_item *item;
2228         struct btrfs_ordered_extent *ordered;
2229         struct btrfs_trans_handle *trans;
2230         struct btrfs_fs_info *fs_info;
2231         struct btrfs_root *root;
2232         struct btrfs_key key;
2233         struct extent_buffer *leaf;
2234         struct old_sa_defrag_extent *old = backref->old;
2235         struct new_sa_defrag_extent *new = old->new;
2236         struct inode *src_inode = new->inode;
2237         struct inode *inode;
2238         struct extent_state *cached = NULL;
2239         int ret = 0;
2240         u64 start;
2241         u64 len;
2242         u64 lock_start;
2243         u64 lock_end;
2244         bool merge = false;
2245         int index;
2246
2247         if (prev && prev->root_id == backref->root_id &&
2248             prev->inum == backref->inum &&
2249             prev->file_pos + prev->num_bytes == backref->file_pos)
2250                 merge = true;
2251
2252         /* step 1: get root */
2253         key.objectid = backref->root_id;
2254         key.type = BTRFS_ROOT_ITEM_KEY;
2255         key.offset = (u64)-1;
2256
2257         fs_info = BTRFS_I(src_inode)->root->fs_info;
2258         index = srcu_read_lock(&fs_info->subvol_srcu);
2259
2260         root = btrfs_read_fs_root_no_name(fs_info, &key);
2261         if (IS_ERR(root)) {
2262                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2263                 if (PTR_ERR(root) == -ENOENT)
2264                         return 0;
2265                 return PTR_ERR(root);
2266         }
2267
2268         if (btrfs_root_readonly(root)) {
2269                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2270                 return 0;
2271         }
2272
2273         /* step 2: get inode */
2274         key.objectid = backref->inum;
2275         key.type = BTRFS_INODE_ITEM_KEY;
2276         key.offset = 0;
2277
2278         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2279         if (IS_ERR(inode)) {
2280                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2281                 return 0;
2282         }
2283
2284         srcu_read_unlock(&fs_info->subvol_srcu, index);
2285
2286         /* step 3: relink backref */
2287         lock_start = backref->file_pos;
2288         lock_end = backref->file_pos + backref->num_bytes - 1;
2289         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2290                          0, &cached);
2291
2292         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2293         if (ordered) {
2294                 btrfs_put_ordered_extent(ordered);
2295                 goto out_unlock;
2296         }
2297
2298         trans = btrfs_join_transaction(root);
2299         if (IS_ERR(trans)) {
2300                 ret = PTR_ERR(trans);
2301                 goto out_unlock;
2302         }
2303
2304         key.objectid = backref->inum;
2305         key.type = BTRFS_EXTENT_DATA_KEY;
2306         key.offset = backref->file_pos;
2307
2308         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2309         if (ret < 0) {
2310                 goto out_free_path;
2311         } else if (ret > 0) {
2312                 ret = 0;
2313                 goto out_free_path;
2314         }
2315
2316         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2317                                 struct btrfs_file_extent_item);
2318
2319         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2320             backref->generation)
2321                 goto out_free_path;
2322
2323         btrfs_release_path(path);
2324
2325         start = backref->file_pos;
2326         if (backref->extent_offset < old->extent_offset + old->offset)
2327                 start += old->extent_offset + old->offset -
2328                          backref->extent_offset;
2329
2330         len = min(backref->extent_offset + backref->num_bytes,
2331                   old->extent_offset + old->offset + old->len);
2332         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2333
2334         ret = btrfs_drop_extents(trans, root, inode, start,
2335                                  start + len, 1);
2336         if (ret)
2337                 goto out_free_path;
2338 again:
2339         key.objectid = btrfs_ino(inode);
2340         key.type = BTRFS_EXTENT_DATA_KEY;
2341         key.offset = start;
2342
2343         path->leave_spinning = 1;
2344         if (merge) {
2345                 struct btrfs_file_extent_item *fi;
2346                 u64 extent_len;
2347                 struct btrfs_key found_key;
2348
2349                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2350                 if (ret < 0)
2351                         goto out_free_path;
2352
2353                 path->slots[0]--;
2354                 leaf = path->nodes[0];
2355                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2356
2357                 fi = btrfs_item_ptr(leaf, path->slots[0],
2358                                     struct btrfs_file_extent_item);
2359                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2360
2361                 if (extent_len + found_key.offset == start &&
2362                     relink_is_mergable(leaf, fi, new)) {
2363                         btrfs_set_file_extent_num_bytes(leaf, fi,
2364                                                         extent_len + len);
2365                         btrfs_mark_buffer_dirty(leaf);
2366                         inode_add_bytes(inode, len);
2367
2368                         ret = 1;
2369                         goto out_free_path;
2370                 } else {
2371                         merge = false;
2372                         btrfs_release_path(path);
2373                         goto again;
2374                 }
2375         }
2376
2377         ret = btrfs_insert_empty_item(trans, root, path, &key,
2378                                         sizeof(*extent));
2379         if (ret) {
2380                 btrfs_abort_transaction(trans, root, ret);
2381                 goto out_free_path;
2382         }
2383
2384         leaf = path->nodes[0];
2385         item = btrfs_item_ptr(leaf, path->slots[0],
2386                                 struct btrfs_file_extent_item);
2387         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2388         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2389         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2390         btrfs_set_file_extent_num_bytes(leaf, item, len);
2391         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2392         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2393         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2394         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2395         btrfs_set_file_extent_encryption(leaf, item, 0);
2396         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2397
2398         btrfs_mark_buffer_dirty(leaf);
2399         inode_add_bytes(inode, len);
2400         btrfs_release_path(path);
2401
2402         ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2403                         new->disk_len, 0,
2404                         backref->root_id, backref->inum,
2405                         new->file_pos, 0);      /* start - extent_offset */
2406         if (ret) {
2407                 btrfs_abort_transaction(trans, root, ret);
2408                 goto out_free_path;
2409         }
2410
2411         ret = 1;
2412 out_free_path:
2413         btrfs_release_path(path);
2414         path->leave_spinning = 0;
2415         btrfs_end_transaction(trans, root);
2416 out_unlock:
2417         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2418                              &cached, GFP_NOFS);
2419         iput(inode);
2420         return ret;
2421 }
2422
2423 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2424 {
2425         struct old_sa_defrag_extent *old, *tmp;
2426
2427         if (!new)
2428                 return;
2429
2430         list_for_each_entry_safe(old, tmp, &new->head, list) {
2431                 list_del(&old->list);
2432                 kfree(old);
2433         }
2434         kfree(new);
2435 }
2436
2437 static void relink_file_extents(struct new_sa_defrag_extent *new)
2438 {
2439         struct btrfs_path *path;
2440         struct sa_defrag_extent_backref *backref;
2441         struct sa_defrag_extent_backref *prev = NULL;
2442         struct inode *inode;
2443         struct btrfs_root *root;
2444         struct rb_node *node;
2445         int ret;
2446
2447         inode = new->inode;
2448         root = BTRFS_I(inode)->root;
2449
2450         path = btrfs_alloc_path();
2451         if (!path)
2452                 return;
2453
2454         if (!record_extent_backrefs(path, new)) {
2455                 btrfs_free_path(path);
2456                 goto out;
2457         }
2458         btrfs_release_path(path);
2459
2460         while (1) {
2461                 node = rb_first(&new->root);
2462                 if (!node)
2463                         break;
2464                 rb_erase(node, &new->root);
2465
2466                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2467
2468                 ret = relink_extent_backref(path, prev, backref);
2469                 WARN_ON(ret < 0);
2470
2471                 kfree(prev);
2472
2473                 if (ret == 1)
2474                         prev = backref;
2475                 else
2476                         prev = NULL;
2477                 cond_resched();
2478         }
2479         kfree(prev);
2480
2481         btrfs_free_path(path);
2482 out:
2483         free_sa_defrag_extent(new);
2484
2485         atomic_dec(&root->fs_info->defrag_running);
2486         wake_up(&root->fs_info->transaction_wait);
2487 }
2488
2489 static struct new_sa_defrag_extent *
2490 record_old_file_extents(struct inode *inode,
2491                         struct btrfs_ordered_extent *ordered)
2492 {
2493         struct btrfs_root *root = BTRFS_I(inode)->root;
2494         struct btrfs_path *path;
2495         struct btrfs_key key;
2496         struct old_sa_defrag_extent *old;
2497         struct new_sa_defrag_extent *new;
2498         int ret;
2499
2500         new = kmalloc(sizeof(*new), GFP_NOFS);
2501         if (!new)
2502                 return NULL;
2503
2504         new->inode = inode;
2505         new->file_pos = ordered->file_offset;
2506         new->len = ordered->len;
2507         new->bytenr = ordered->start;
2508         new->disk_len = ordered->disk_len;
2509         new->compress_type = ordered->compress_type;
2510         new->root = RB_ROOT;
2511         INIT_LIST_HEAD(&new->head);
2512
2513         path = btrfs_alloc_path();
2514         if (!path)
2515                 goto out_kfree;
2516
2517         key.objectid = btrfs_ino(inode);
2518         key.type = BTRFS_EXTENT_DATA_KEY;
2519         key.offset = new->file_pos;
2520
2521         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2522         if (ret < 0)
2523                 goto out_free_path;
2524         if (ret > 0 && path->slots[0] > 0)
2525                 path->slots[0]--;
2526
2527         /* find out all the old extents for the file range */
2528         while (1) {
2529                 struct btrfs_file_extent_item *extent;
2530                 struct extent_buffer *l;
2531                 int slot;
2532                 u64 num_bytes;
2533                 u64 offset;
2534                 u64 end;
2535                 u64 disk_bytenr;
2536                 u64 extent_offset;
2537
2538                 l = path->nodes[0];
2539                 slot = path->slots[0];
2540
2541                 if (slot >= btrfs_header_nritems(l)) {
2542                         ret = btrfs_next_leaf(root, path);
2543                         if (ret < 0)
2544                                 goto out_free_path;
2545                         else if (ret > 0)
2546                                 break;
2547                         continue;
2548                 }
2549
2550                 btrfs_item_key_to_cpu(l, &key, slot);
2551
2552                 if (key.objectid != btrfs_ino(inode))
2553                         break;
2554                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2555                         break;
2556                 if (key.offset >= new->file_pos + new->len)
2557                         break;
2558
2559                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2560
2561                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2562                 if (key.offset + num_bytes < new->file_pos)
2563                         goto next;
2564
2565                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2566                 if (!disk_bytenr)
2567                         goto next;
2568
2569                 extent_offset = btrfs_file_extent_offset(l, extent);
2570
2571                 old = kmalloc(sizeof(*old), GFP_NOFS);
2572                 if (!old)
2573                         goto out_free_path;
2574
2575                 offset = max(new->file_pos, key.offset);
2576                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2577
2578                 old->bytenr = disk_bytenr;
2579                 old->extent_offset = extent_offset;
2580                 old->offset = offset - key.offset;
2581                 old->len = end - offset;
2582                 old->new = new;
2583                 old->count = 0;
2584                 list_add_tail(&old->list, &new->head);
2585 next:
2586                 path->slots[0]++;
2587                 cond_resched();
2588         }
2589
2590         btrfs_free_path(path);
2591         atomic_inc(&root->fs_info->defrag_running);
2592
2593         return new;
2594
2595 out_free_path:
2596         btrfs_free_path(path);
2597 out_kfree:
2598         free_sa_defrag_extent(new);
2599         return NULL;
2600 }
2601
2602 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2603                                          u64 start, u64 len)
2604 {
2605         struct btrfs_block_group_cache *cache;
2606
2607         cache = btrfs_lookup_block_group(root->fs_info, start);
2608         ASSERT(cache);
2609
2610         spin_lock(&cache->lock);
2611         cache->delalloc_bytes -= len;
2612         spin_unlock(&cache->lock);
2613
2614         btrfs_put_block_group(cache);
2615 }
2616
2617 /* as ordered data IO finishes, this gets called so we can finish
2618  * an ordered extent if the range of bytes in the file it covers are
2619  * fully written.
2620  */
2621 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2622 {
2623         struct inode *inode = ordered_extent->inode;
2624         struct btrfs_root *root = BTRFS_I(inode)->root;
2625         struct btrfs_trans_handle *trans = NULL;
2626         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2627         struct extent_state *cached_state = NULL;
2628         struct new_sa_defrag_extent *new = NULL;
2629         int compress_type = 0;
2630         int ret = 0;
2631         u64 logical_len = ordered_extent->len;
2632         bool nolock;
2633         bool truncated = false;
2634
2635         nolock = btrfs_is_free_space_inode(inode);
2636
2637         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2638                 ret = -EIO;
2639                 goto out;
2640         }
2641
2642         if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2643                 truncated = true;
2644                 logical_len = ordered_extent->truncated_len;
2645                 /* Truncated the entire extent, don't bother adding */
2646                 if (!logical_len)
2647                         goto out;
2648         }
2649
2650         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2651                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2652                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2653                 if (nolock)
2654                         trans = btrfs_join_transaction_nolock(root);
2655                 else
2656                         trans = btrfs_join_transaction(root);
2657                 if (IS_ERR(trans)) {
2658                         ret = PTR_ERR(trans);
2659                         trans = NULL;
2660                         goto out;
2661                 }
2662                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2663                 ret = btrfs_update_inode_fallback(trans, root, inode);
2664                 if (ret) /* -ENOMEM or corruption */
2665                         btrfs_abort_transaction(trans, root, ret);
2666                 goto out;
2667         }
2668
2669         lock_extent_bits(io_tree, ordered_extent->file_offset,
2670                          ordered_extent->file_offset + ordered_extent->len - 1,
2671                          0, &cached_state);
2672
2673         ret = test_range_bit(io_tree, ordered_extent->file_offset,
2674                         ordered_extent->file_offset + ordered_extent->len - 1,
2675                         EXTENT_DEFRAG, 1, cached_state);
2676         if (ret) {
2677                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2678                 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2679                         /* the inode is shared */
2680                         new = record_old_file_extents(inode, ordered_extent);
2681
2682                 clear_extent_bit(io_tree, ordered_extent->file_offset,
2683                         ordered_extent->file_offset + ordered_extent->len - 1,
2684                         EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2685         }
2686
2687         if (nolock)
2688                 trans = btrfs_join_transaction_nolock(root);
2689         else
2690                 trans = btrfs_join_transaction(root);
2691         if (IS_ERR(trans)) {
2692                 ret = PTR_ERR(trans);
2693                 trans = NULL;
2694                 goto out_unlock;
2695         }
2696
2697         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2698
2699         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2700                 compress_type = ordered_extent->compress_type;
2701         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2702                 BUG_ON(compress_type);
2703                 ret = btrfs_mark_extent_written(trans, inode,
2704                                                 ordered_extent->file_offset,
2705                                                 ordered_extent->file_offset +
2706                                                 logical_len);
2707         } else {
2708                 BUG_ON(root == root->fs_info->tree_root);
2709                 ret = insert_reserved_file_extent(trans, inode,
2710                                                 ordered_extent->file_offset,
2711                                                 ordered_extent->start,
2712                                                 ordered_extent->disk_len,
2713                                                 logical_len, logical_len,
2714                                                 compress_type, 0, 0,
2715                                                 BTRFS_FILE_EXTENT_REG);
2716                 if (!ret)
2717                         btrfs_release_delalloc_bytes(root,
2718                                                      ordered_extent->start,
2719                                                      ordered_extent->disk_len);
2720         }
2721         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2722                            ordered_extent->file_offset, ordered_extent->len,
2723                            trans->transid);
2724         if (ret < 0) {
2725                 btrfs_abort_transaction(trans, root, ret);
2726                 goto out_unlock;
2727         }
2728
2729         add_pending_csums(trans, inode, ordered_extent->file_offset,
2730                           &ordered_extent->list);
2731
2732         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2733         ret = btrfs_update_inode_fallback(trans, root, inode);
2734         if (ret) { /* -ENOMEM or corruption */
2735                 btrfs_abort_transaction(trans, root, ret);
2736                 goto out_unlock;
2737         }
2738         ret = 0;
2739 out_unlock:
2740         unlock_extent_cached(io_tree, ordered_extent->file_offset,
2741                              ordered_extent->file_offset +
2742                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
2743 out:
2744         if (root != root->fs_info->tree_root)
2745                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2746         if (trans)
2747                 btrfs_end_transaction(trans, root);
2748
2749         if (ret || truncated) {
2750                 u64 start, end;
2751
2752                 if (truncated)
2753                         start = ordered_extent->file_offset + logical_len;
2754                 else
2755                         start = ordered_extent->file_offset;
2756                 end = ordered_extent->file_offset + ordered_extent->len - 1;
2757                 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2758
2759                 /* Drop the cache for the part of the extent we didn't write. */
2760                 btrfs_drop_extent_cache(inode, start, end, 0);
2761
2762                 /*
2763                  * If the ordered extent had an IOERR or something else went
2764                  * wrong we need to return the space for this ordered extent
2765                  * back to the allocator.  We only free the extent in the
2766                  * truncated case if we didn't write out the extent at all.
2767                  */
2768                 if ((ret || !logical_len) &&
2769                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2770                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2771                         btrfs_free_reserved_extent(root, ordered_extent->start,
2772                                                    ordered_extent->disk_len, 1);
2773         }
2774
2775
2776         /*
2777          * This needs to be done to make sure anybody waiting knows we are done
2778          * updating everything for this ordered extent.
2779          */
2780         btrfs_remove_ordered_extent(inode, ordered_extent);
2781
2782         /* for snapshot-aware defrag */
2783         if (new) {
2784                 if (ret) {
2785                         free_sa_defrag_extent(new);
2786                         atomic_dec(&root->fs_info->defrag_running);
2787                 } else {
2788                         relink_file_extents(new);
2789                 }
2790         }
2791
2792         /* once for us */
2793         btrfs_put_ordered_extent(ordered_extent);
2794         /* once for the tree */
2795         btrfs_put_ordered_extent(ordered_extent);
2796
2797         return ret;
2798 }
2799
2800 static void finish_ordered_fn(struct btrfs_work *work)
2801 {
2802         struct btrfs_ordered_extent *ordered_extent;
2803         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2804         btrfs_finish_ordered_io(ordered_extent);
2805 }
2806
2807 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2808                                 struct extent_state *state, int uptodate)
2809 {
2810         struct inode *inode = page->mapping->host;
2811         struct btrfs_root *root = BTRFS_I(inode)->root;
2812         struct btrfs_ordered_extent *ordered_extent = NULL;
2813         struct btrfs_workqueue *workers;
2814
2815         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2816
2817         ClearPagePrivate2(page);
2818         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2819                                             end - start + 1, uptodate))
2820                 return 0;
2821
2822         btrfs_init_work(&ordered_extent->work, finish_ordered_fn, NULL, NULL);
2823
2824         if (btrfs_is_free_space_inode(inode))
2825                 workers = root->fs_info->endio_freespace_worker;
2826         else
2827                 workers = root->fs_info->endio_write_workers;
2828         btrfs_queue_work(workers, &ordered_extent->work);
2829
2830         return 0;
2831 }
2832
2833 /*
2834  * when reads are done, we need to check csums to verify the data is correct
2835  * if there's a match, we allow the bio to finish.  If not, the code in
2836  * extent_io.c will try to find good copies for us.
2837  */
2838 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2839                                       u64 phy_offset, struct page *page,
2840                                       u64 start, u64 end, int mirror)
2841 {
2842         size_t offset = start - page_offset(page);
2843         struct inode *inode = page->mapping->host;
2844         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2845         char *kaddr;
2846         struct btrfs_root *root = BTRFS_I(inode)->root;
2847         u32 csum_expected;
2848         u32 csum = ~(u32)0;
2849         static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2850                                       DEFAULT_RATELIMIT_BURST);
2851
2852         if (PageChecked(page)) {
2853                 ClearPageChecked(page);
2854                 goto good;
2855         }
2856
2857         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2858                 goto good;
2859
2860         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2861             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2862                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2863                                   GFP_NOFS);
2864                 return 0;
2865         }
2866
2867         phy_offset >>= inode->i_sb->s_blocksize_bits;
2868         csum_expected = *(((u32 *)io_bio->csum) + phy_offset);
2869
2870         kaddr = kmap_atomic(page);
2871         csum = btrfs_csum_data(kaddr + offset, csum,  end - start + 1);
2872         btrfs_csum_final(csum, (char *)&csum);
2873         if (csum != csum_expected)
2874                 goto zeroit;
2875
2876         kunmap_atomic(kaddr);
2877 good:
2878         return 0;
2879
2880 zeroit:
2881         if (__ratelimit(&_rs))
2882                 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
2883                         btrfs_ino(page->mapping->host), start, csum, csum_expected);
2884         memset(kaddr + offset, 1, end - start + 1);
2885         flush_dcache_page(page);
2886         kunmap_atomic(kaddr);
2887         if (csum_expected == 0)
2888                 return 0;
2889         return -EIO;
2890 }
2891
2892 struct delayed_iput {
2893         struct list_head list;
2894         struct inode *inode;
2895 };
2896
2897 /* JDM: If this is fs-wide, why can't we add a pointer to
2898  * btrfs_inode instead and avoid the allocation? */
2899 void btrfs_add_delayed_iput(struct inode *inode)
2900 {
2901         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2902         struct delayed_iput *delayed;
2903
2904         if (atomic_add_unless(&inode->i_count, -1, 1))
2905                 return;
2906
2907         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2908         delayed->inode = inode;
2909
2910         spin_lock(&fs_info->delayed_iput_lock);
2911         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2912         spin_unlock(&fs_info->delayed_iput_lock);
2913 }
2914
2915 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2916 {
2917         LIST_HEAD(list);
2918         struct btrfs_fs_info *fs_info = root->fs_info;
2919         struct delayed_iput *delayed;
2920         int empty;
2921
2922         spin_lock(&fs_info->delayed_iput_lock);
2923         empty = list_empty(&fs_info->delayed_iputs);
2924         spin_unlock(&fs_info->delayed_iput_lock);
2925         if (empty)
2926                 return;
2927
2928         spin_lock(&fs_info->delayed_iput_lock);
2929         list_splice_init(&fs_info->delayed_iputs, &list);
2930         spin_unlock(&fs_info->delayed_iput_lock);
2931
2932         while (!list_empty(&list)) {
2933                 delayed = list_entry(list.next, struct delayed_iput, list);
2934                 list_del(&delayed->list);
2935                 iput(delayed->inode);
2936                 kfree(delayed);
2937         }
2938 }
2939
2940 /*
2941  * This is called in transaction commit time. If there are no orphan
2942  * files in the subvolume, it removes orphan item and frees block_rsv
2943  * structure.
2944  */
2945 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2946                               struct btrfs_root *root)
2947 {
2948         struct btrfs_block_rsv *block_rsv;
2949         int ret;
2950
2951         if (atomic_read(&root->orphan_inodes) ||
2952             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2953                 return;
2954
2955         spin_lock(&root->orphan_lock);
2956         if (atomic_read(&root->orphan_inodes)) {
2957                 spin_unlock(&root->orphan_lock);
2958                 return;
2959         }
2960
2961         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2962                 spin_unlock(&root->orphan_lock);
2963                 return;
2964         }
2965
2966         block_rsv = root->orphan_block_rsv;
2967         root->orphan_block_rsv = NULL;
2968         spin_unlock(&root->orphan_lock);
2969
2970         if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
2971             btrfs_root_refs(&root->root_item) > 0) {
2972                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2973                                             root->root_key.objectid);
2974                 if (ret)
2975                         btrfs_abort_transaction(trans, root, ret);
2976                 else
2977                         clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
2978                                   &root->state);
2979         }
2980
2981         if (block_rsv) {
2982                 WARN_ON(block_rsv->size > 0);
2983                 btrfs_free_block_rsv(root, block_rsv);
2984         }
2985 }
2986
2987 /*
2988  * This creates an orphan entry for the given inode in case something goes
2989  * wrong in the middle of an unlink/truncate.
2990  *
2991  * NOTE: caller of this function should reserve 5 units of metadata for
2992  *       this function.
2993  */
2994 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2995 {
2996         struct btrfs_root *root = BTRFS_I(inode)->root;
2997         struct btrfs_block_rsv *block_rsv = NULL;
2998         int reserve = 0;
2999         int insert = 0;
3000         int ret;
3001
3002         if (!root->orphan_block_rsv) {
3003                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3004                 if (!block_rsv)
3005                         return -ENOMEM;
3006         }
3007
3008         spin_lock(&root->orphan_lock);
3009         if (!root->orphan_block_rsv) {
3010                 root->orphan_block_rsv = block_rsv;
3011         } else if (block_rsv) {
3012                 btrfs_free_block_rsv(root, block_rsv);
3013                 block_rsv = NULL;
3014         }
3015
3016         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3017                               &BTRFS_I(inode)->runtime_flags)) {
3018 #if 0
3019                 /*
3020                  * For proper ENOSPC handling, we should do orphan
3021                  * cleanup when mounting. But this introduces backward
3022                  * compatibility issue.
3023                  */
3024                 if (!xchg(&root->orphan_item_inserted, 1))
3025                         insert = 2;
3026                 else
3027                         insert = 1;
3028 #endif
3029                 insert = 1;
3030                 atomic_inc(&root->orphan_inodes);
3031         }
3032
3033         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3034                               &BTRFS_I(inode)->runtime_flags))
3035                 reserve = 1;
3036         spin_unlock(&root->orphan_lock);
3037
3038         /* grab metadata reservation from transaction handle */
3039         if (reserve) {
3040                 ret = btrfs_orphan_reserve_metadata(trans, inode);
3041                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3042         }
3043
3044         /* insert an orphan item to track this unlinked/truncated file */
3045         if (insert >= 1) {
3046                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3047                 if (ret) {
3048                         atomic_dec(&root->orphan_inodes);
3049                         if (reserve) {
3050                                 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3051                                           &BTRFS_I(inode)->runtime_flags);
3052                                 btrfs_orphan_release_metadata(inode);
3053                         }
3054                         if (ret != -EEXIST) {
3055                                 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3056                                           &BTRFS_I(inode)->runtime_flags);
3057                                 btrfs_abort_transaction(trans, root, ret);
3058                                 return ret;
3059                         }
3060                 }
3061                 ret = 0;
3062         }
3063
3064         /* insert an orphan item to track subvolume contains orphan files */
3065         if (insert >= 2) {
3066                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3067                                                root->root_key.objectid);
3068                 if (ret && ret != -EEXIST) {
3069                         btrfs_abort_transaction(trans, root, ret);
3070                         return ret;
3071                 }
3072         }
3073         return 0;
3074 }
3075
3076 /*
3077  * We have done the truncate/delete so we can go ahead and remove the orphan
3078  * item for this particular inode.
3079  */
3080 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3081                             struct inode *inode)
3082 {
3083         struct btrfs_root *root = BTRFS_I(inode)->root;
3084         int delete_item = 0;
3085         int release_rsv = 0;
3086         int ret = 0;
3087
3088         spin_lock(&root->orphan_lock);
3089         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3090                                &BTRFS_I(inode)->runtime_flags))
3091                 delete_item = 1;
3092
3093         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3094                                &BTRFS_I(inode)->runtime_flags))
3095                 release_rsv = 1;
3096         spin_unlock(&root->orphan_lock);
3097
3098         if (delete_item) {
3099                 atomic_dec(&root->orphan_inodes);
3100                 if (trans)
3101                         ret = btrfs_del_orphan_item(trans, root,
3102                                                     btrfs_ino(inode));
3103         }
3104
3105         if (release_rsv)
3106                 btrfs_orphan_release_metadata(inode);
3107
3108         return ret;
3109 }
3110
3111 /*
3112  * this cleans up any orphans that may be left on the list from the last use
3113  * of this root.
3114  */
3115 int btrfs_orphan_cleanup(struct btrfs_root *root)
3116 {
3117         struct btrfs_path *path;
3118         struct extent_buffer *leaf;
3119         struct btrfs_key key, found_key;
3120         struct btrfs_trans_handle *trans;
3121         struct inode *inode;
3122         u64 last_objectid = 0;
3123         int ret = 0, nr_unlink = 0, nr_truncate = 0;
3124
3125         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3126                 return 0;
3127
3128         path = btrfs_alloc_path();
3129         if (!path) {
3130                 ret = -ENOMEM;
3131                 goto out;
3132         }
3133         path->reada = -1;
3134
3135         key.objectid = BTRFS_ORPHAN_OBJECTID;
3136         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3137         key.offset = (u64)-1;
3138
3139         while (1) {
3140                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3141                 if (ret < 0)
3142                         goto out;
3143
3144                 /*
3145                  * if ret == 0 means we found what we were searching for, which
3146                  * is weird, but possible, so only screw with path if we didn't
3147                  * find the key and see if we have stuff that matches
3148                  */
3149                 if (ret > 0) {
3150                         ret = 0;
3151                         if (path->slots[0] == 0)
3152                                 break;
3153                         path->slots[0]--;
3154                 }
3155
3156                 /* pull out the item */
3157                 leaf = path->nodes[0];
3158                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3159
3160                 /* make sure the item matches what we want */
3161                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3162                         break;
3163                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3164                         break;
3165
3166                 /* release the path since we're done with it */
3167                 btrfs_release_path(path);
3168
3169                 /*
3170                  * this is where we are basically btrfs_lookup, without the
3171                  * crossing root thing.  we store the inode number in the
3172                  * offset of the orphan item.
3173                  */
3174
3175                 if (found_key.offset == last_objectid) {
3176                         btrfs_err(root->fs_info,
3177                                 "Error removing orphan entry, stopping orphan cleanup");
3178                         ret = -EINVAL;
3179                         goto out;
3180                 }
3181
3182                 last_objectid = found_key.offset;
3183
3184                 found_key.objectid = found_key.offset;
3185                 found_key.type = BTRFS_INODE_ITEM_KEY;
3186                 found_key.offset = 0;
3187                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3188                 ret = PTR_ERR_OR_ZERO(inode);
3189                 if (ret && ret != -ESTALE)
3190                         goto out;
3191
3192                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3193                         struct btrfs_root *dead_root;
3194                         struct btrfs_fs_info *fs_info = root->fs_info;
3195                         int is_dead_root = 0;
3196
3197                         /*
3198                          * this is an orphan in the tree root. Currently these
3199                          * could come from 2 sources:
3200                          *  a) a snapshot deletion in progress
3201                          *  b) a free space cache inode
3202                          * We need to distinguish those two, as the snapshot
3203                          * orphan must not get deleted.
3204                          * find_dead_roots already ran before us, so if this
3205                          * is a snapshot deletion, we should find the root
3206                          * in the dead_roots list
3207                          */
3208                         spin_lock(&fs_info->trans_lock);
3209                         list_for_each_entry(dead_root, &fs_info->dead_roots,
3210                                             root_list) {
3211                                 if (dead_root->root_key.objectid ==
3212                                     found_key.objectid) {
3213                                         is_dead_root = 1;
3214                                         break;
3215                                 }
3216                         }
3217                         spin_unlock(&fs_info->trans_lock);
3218                         if (is_dead_root) {
3219                                 /* prevent this orphan from being found again */
3220                                 key.offset = found_key.objectid - 1;
3221                                 continue;
3222                         }
3223                 }
3224                 /*
3225                  * Inode is already gone but the orphan item is still there,
3226                  * kill the orphan item.
3227                  */
3228                 if (ret == -ESTALE) {
3229                         trans = btrfs_start_transaction(root, 1);
3230                         if (IS_ERR(trans)) {
3231                                 ret = PTR_ERR(trans);
3232                                 goto out;
3233                         }
3234                         btrfs_debug(root->fs_info, "auto deleting %Lu",
3235                                 found_key.objectid);
3236                         ret = btrfs_del_orphan_item(trans, root,
3237                                                     found_key.objectid);
3238                         btrfs_end_transaction(trans, root);
3239                         if (ret)
3240                                 goto out;
3241                         continue;
3242                 }
3243
3244                 /*
3245                  * add this inode to the orphan list so btrfs_orphan_del does
3246                  * the proper thing when we hit it
3247                  */
3248                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3249                         &BTRFS_I(inode)->runtime_flags);
3250                 atomic_inc(&root->orphan_inodes);
3251
3252                 /* if we have links, this was a truncate, lets do that */
3253                 if (inode->i_nlink) {
3254                         if (WARN_ON(!S_ISREG(inode->i_mode))) {
3255                                 iput(inode);
3256                                 continue;
3257                         }
3258                         nr_truncate++;
3259
3260                         /* 1 for the orphan item deletion. */
3261                         trans = btrfs_start_transaction(root, 1);
3262                         if (IS_ERR(trans)) {
3263                                 iput(inode);
3264                                 ret = PTR_ERR(trans);
3265                                 goto out;
3266                         }
3267                         ret = btrfs_orphan_add(trans, inode);
3268                         btrfs_end_transaction(trans, root);
3269                         if (ret) {
3270                                 iput(inode);
3271                                 goto out;
3272                         }
3273
3274                         ret = btrfs_truncate(inode);
3275                         if (ret)
3276                                 btrfs_orphan_del(NULL, inode);
3277                 } else {
3278                         nr_unlink++;
3279                 }
3280
3281                 /* this will do delete_inode and everything for us */
3282                 iput(inode);
3283                 if (ret)
3284                         goto out;
3285         }
3286         /* release the path since we're done with it */
3287         btrfs_release_path(path);
3288
3289         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3290
3291         if (root->orphan_block_rsv)
3292                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3293                                         (u64)-1);
3294
3295         if (root->orphan_block_rsv ||
3296             test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3297                 trans = btrfs_join_transaction(root);
3298                 if (!IS_ERR(trans))
3299                         btrfs_end_transaction(trans, root);
3300         }
3301
3302         if (nr_unlink)
3303                 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3304         if (nr_truncate)
3305                 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3306
3307 out:
3308         if (ret)
3309                 btrfs_crit(root->fs_info,
3310                         "could not do orphan cleanup %d", ret);
3311         btrfs_free_path(path);
3312         return ret;
3313 }
3314
3315 /*
3316  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3317  * don't find any xattrs, we know there can't be any acls.
3318  *
3319  * slot is the slot the inode is in, objectid is the objectid of the inode
3320  */
3321 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3322                                           int slot, u64 objectid,
3323                                           int *first_xattr_slot)
3324 {
3325         u32 nritems = btrfs_header_nritems(leaf);
3326         struct btrfs_key found_key;
3327         static u64 xattr_access = 0;
3328         static u64 xattr_default = 0;
3329         int scanned = 0;
3330
3331         if (!xattr_access) {
3332                 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3333                                         strlen(POSIX_ACL_XATTR_ACCESS));
3334                 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3335                                         strlen(POSIX_ACL_XATTR_DEFAULT));
3336         }
3337
3338         slot++;
3339         *first_xattr_slot = -1;
3340         while (slot < nritems) {
3341                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3342
3343                 /* we found a different objectid, there must not be acls */
3344                 if (found_key.objectid != objectid)
3345                         return 0;
3346
3347                 /* we found an xattr, assume we've got an acl */
3348                 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3349                         if (*first_xattr_slot == -1)
3350                                 *first_xattr_slot = slot;
3351                         if (found_key.offset == xattr_access ||
3352                             found_key.offset == xattr_default)
3353                                 return 1;
3354                 }
3355
3356                 /*
3357                  * we found a key greater than an xattr key, there can't
3358                  * be any acls later on
3359                  */
3360                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3361                         return 0;
3362
3363                 slot++;
3364                 scanned++;
3365
3366                 /*
3367                  * it goes inode, inode backrefs, xattrs, extents,
3368                  * so if there are a ton of hard links to an inode there can
3369                  * be a lot of backrefs.  Don't waste time searching too hard,
3370                  * this is just an optimization
3371                  */
3372                 if (scanned >= 8)
3373                         break;
3374         }
3375         /* we hit the end of the leaf before we found an xattr or
3376          * something larger than an xattr.  We have to assume the inode
3377          * has acls
3378          */
3379         if (*first_xattr_slot == -1)
3380                 *first_xattr_slot = slot;
3381         return 1;
3382 }
3383
3384 /*
3385  * read an inode from the btree into the in-memory inode
3386  */
3387 static void btrfs_read_locked_inode(struct inode *inode)
3388 {
3389         struct btrfs_path *path;
3390         struct extent_buffer *leaf;
3391         struct btrfs_inode_item *inode_item;
3392         struct btrfs_timespec *tspec;
3393         struct btrfs_root *root = BTRFS_I(inode)->root;
3394         struct btrfs_key location;
3395         unsigned long ptr;
3396         int maybe_acls;
3397         u32 rdev;
3398         int ret;
3399         bool filled = false;
3400         int first_xattr_slot;
3401
3402         ret = btrfs_fill_inode(inode, &rdev);
3403         if (!ret)
3404                 filled = true;
3405
3406         path = btrfs_alloc_path();
3407         if (!path)
3408                 goto make_bad;
3409
3410         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3411
3412         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3413         if (ret)
3414                 goto make_bad;
3415
3416         leaf = path->nodes[0];
3417
3418         if (filled)
3419                 goto cache_index;
3420
3421         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3422                                     struct btrfs_inode_item);
3423         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3424         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3425         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3426         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3427         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3428
3429         tspec = btrfs_inode_atime(inode_item);
3430         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3431         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3432
3433         tspec = btrfs_inode_mtime(inode_item);
3434         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3435         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3436
3437         tspec = btrfs_inode_ctime(inode_item);
3438         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3439         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3440
3441         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3442         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3443         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3444
3445         /*
3446          * If we were modified in the current generation and evicted from memory
3447          * and then re-read we need to do a full sync since we don't have any
3448          * idea about which extents were modified before we were evicted from
3449          * cache.
3450          */
3451         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3452                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3453                         &BTRFS_I(inode)->runtime_flags);
3454
3455         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3456         inode->i_generation = BTRFS_I(inode)->generation;
3457         inode->i_rdev = 0;
3458         rdev = btrfs_inode_rdev(leaf, inode_item);
3459
3460         BTRFS_I(inode)->index_cnt = (u64)-1;
3461         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3462
3463 cache_index:
3464         path->slots[0]++;
3465         if (inode->i_nlink != 1 ||
3466             path->slots[0] >= btrfs_header_nritems(leaf))
3467                 goto cache_acl;
3468
3469         btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3470         if (location.objectid != btrfs_ino(inode))
3471                 goto cache_acl;
3472
3473         ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3474         if (location.type == BTRFS_INODE_REF_KEY) {
3475                 struct btrfs_inode_ref *ref;
3476
3477                 ref = (struct btrfs_inode_ref *)ptr;
3478                 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3479         } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3480                 struct btrfs_inode_extref *extref;
3481
3482                 extref = (struct btrfs_inode_extref *)ptr;
3483                 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3484                                                                      extref);
3485         }
3486 cache_acl:
3487         /*
3488          * try to precache a NULL acl entry for files that don't have
3489          * any xattrs or acls
3490          */
3491         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3492                                            btrfs_ino(inode), &first_xattr_slot);
3493         if (first_xattr_slot != -1) {
3494                 path->slots[0] = first_xattr_slot;
3495                 ret = btrfs_load_inode_props(inode, path);
3496                 if (ret)
3497                         btrfs_err(root->fs_info,
3498                                   "error loading props for ino %llu (root %llu): %d",
3499                                   btrfs_ino(inode),
3500                                   root->root_key.objectid, ret);
3501         }
3502         btrfs_free_path(path);
3503
3504         if (!maybe_acls)
3505                 cache_no_acl(inode);
3506
3507         switch (inode->i_mode & S_IFMT) {
3508         case S_IFREG:
3509                 inode->i_mapping->a_ops = &btrfs_aops;
3510                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3511                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3512                 inode->i_fop = &btrfs_file_operations;
3513                 inode->i_op = &btrfs_file_inode_operations;
3514                 break;
3515         case S_IFDIR:
3516                 inode->i_fop = &btrfs_dir_file_operations;
3517                 if (root == root->fs_info->tree_root)
3518                         inode->i_op = &btrfs_dir_ro_inode_operations;
3519                 else
3520                         inode->i_op = &btrfs_dir_inode_operations;
3521                 break;
3522         case S_IFLNK:
3523                 inode->i_op = &btrfs_symlink_inode_operations;
3524                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3525                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3526                 break;
3527         default:
3528                 inode->i_op = &btrfs_special_inode_operations;
3529                 init_special_inode(inode, inode->i_mode, rdev);
3530                 break;
3531         }
3532
3533         btrfs_update_iflags(inode);
3534         return;
3535
3536 make_bad:
3537         btrfs_free_path(path);
3538         make_bad_inode(inode);
3539 }
3540
3541 /*
3542  * given a leaf and an inode, copy the inode fields into the leaf
3543  */
3544 static void fill_inode_item(struct btrfs_trans_handle *trans,
3545                             struct extent_buffer *leaf,
3546                             struct btrfs_inode_item *item,
3547                             struct inode *inode)
3548 {
3549         struct btrfs_map_token token;
3550
3551         btrfs_init_map_token(&token);
3552
3553         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3554         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3555         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3556                                    &token);
3557         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3558         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3559
3560         btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3561                                      inode->i_atime.tv_sec, &token);
3562         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3563                                       inode->i_atime.tv_nsec, &token);
3564
3565         btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3566                                      inode->i_mtime.tv_sec, &token);
3567         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3568                                       inode->i_mtime.tv_nsec, &token);
3569
3570         btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3571                                      inode->i_ctime.tv_sec, &token);
3572         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3573                                       inode->i_ctime.tv_nsec, &token);
3574
3575         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3576                                      &token);
3577         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3578                                          &token);
3579         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3580         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3581         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3582         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3583         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3584 }
3585
3586 /*
3587  * copy everything in the in-memory inode into the btree.
3588  */
3589 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3590                                 struct btrfs_root *root, struct inode *inode)
3591 {
3592         struct btrfs_inode_item *inode_item;
3593         struct btrfs_path *path;
3594         struct extent_buffer *leaf;
3595         int ret;
3596
3597         path = btrfs_alloc_path();
3598         if (!path)
3599                 return -ENOMEM;
3600
3601         path->leave_spinning = 1;
3602         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3603                                  1);
3604         if (ret) {
3605                 if (ret > 0)
3606                         ret = -ENOENT;
3607                 goto failed;
3608         }
3609
3610         leaf = path->nodes[0];
3611         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3612                                     struct btrfs_inode_item);
3613
3614         fill_inode_item(trans, leaf, inode_item, inode);
3615         btrfs_mark_buffer_dirty(leaf);
3616         btrfs_set_inode_last_trans(trans, inode);
3617         ret = 0;
3618 failed:
3619         btrfs_free_path(path);
3620         return ret;
3621 }
3622
3623 /*
3624  * copy everything in the in-memory inode into the btree.
3625  */
3626 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3627                                 struct btrfs_root *root, struct inode *inode)
3628 {
3629         int ret;
3630
3631         /*
3632          * If the inode is a free space inode, we can deadlock during commit
3633          * if we put it into the delayed code.
3634          *
3635          * The data relocation inode should also be directly updated
3636          * without delay
3637          */
3638         if (!btrfs_is_free_space_inode(inode)
3639             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3640                 btrfs_update_root_times(trans, root);
3641
3642                 ret = btrfs_delayed_update_inode(trans, root, inode);
3643                 if (!ret)
3644                         btrfs_set_inode_last_trans(trans, inode);
3645                 return ret;
3646         }
3647
3648         return btrfs_update_inode_item(trans, root, inode);
3649 }
3650
3651 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3652                                          struct btrfs_root *root,
3653                                          struct inode *inode)
3654 {
3655         int ret;
3656
3657         ret = btrfs_update_inode(trans, root, inode);
3658         if (ret == -ENOSPC)
3659                 return btrfs_update_inode_item(trans, root, inode);
3660         return ret;
3661 }
3662
3663 /*
3664  * unlink helper that gets used here in inode.c and in the tree logging
3665  * recovery code.  It remove a link in a directory with a given name, and
3666  * also drops the back refs in the inode to the directory
3667  */
3668 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3669                                 struct btrfs_root *root,
3670                                 struct inode *dir, struct inode *inode,
3671                                 const char *name, int name_len)
3672 {
3673         struct btrfs_path *path;
3674         int ret = 0;
3675         struct extent_buffer *leaf;
3676         struct btrfs_dir_item *di;
3677         struct btrfs_key key;
3678         u64 index;
3679         u64 ino = btrfs_ino(inode);
3680         u64 dir_ino = btrfs_ino(dir);
3681
3682         path = btrfs_alloc_path();
3683         if (!path) {
3684                 ret = -ENOMEM;
3685                 goto out;
3686         }
3687
3688         path->leave_spinning = 1;
3689         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3690                                     name, name_len, -1);
3691         if (IS_ERR(di)) {
3692                 ret = PTR_ERR(di);
3693                 goto err;
3694         }
3695         if (!di) {
3696                 ret = -ENOENT;
3697                 goto err;
3698         }
3699         leaf = path->nodes[0];
3700         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3701         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3702         if (ret)
3703                 goto err;
3704         btrfs_release_path(path);
3705
3706         /*
3707          * If we don't have dir index, we have to get it by looking up
3708          * the inode ref, since we get the inode ref, remove it directly,
3709          * it is unnecessary to do delayed deletion.
3710          *
3711          * But if we have dir index, needn't search inode ref to get it.
3712          * Since the inode ref is close to the inode item, it is better
3713          * that we delay to delete it, and just do this deletion when
3714          * we update the inode item.
3715          */
3716         if (BTRFS_I(inode)->dir_index) {
3717                 ret = btrfs_delayed_delete_inode_ref(inode);
3718                 if (!ret) {
3719                         index = BTRFS_I(inode)->dir_index;
3720                         goto skip_backref;
3721                 }
3722         }
3723
3724         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3725                                   dir_ino, &index);
3726         if (ret) {
3727                 btrfs_info(root->fs_info,
3728                         "failed to delete reference to %.*s, inode %llu parent %llu",
3729                         name_len, name, ino, dir_ino);
3730                 btrfs_abort_transaction(trans, root, ret);
3731                 goto err;
3732         }
3733 skip_backref:
3734         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3735         if (ret) {
3736                 btrfs_abort_transaction(trans, root, ret);
3737                 goto err;
3738         }
3739
3740         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3741                                          inode, dir_ino);
3742         if (ret != 0 && ret != -ENOENT) {
3743                 btrfs_abort_transaction(trans, root, ret);
3744                 goto err;
3745         }
3746
3747         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3748                                            dir, index);
3749         if (ret == -ENOENT)
3750                 ret = 0;
3751         else if (ret)
3752                 btrfs_abort_transaction(trans, root, ret);
3753 err:
3754         btrfs_free_path(path);
3755         if (ret)
3756                 goto out;
3757
3758         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3759         inode_inc_iversion(inode);
3760         inode_inc_iversion(dir);
3761         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3762         ret = btrfs_update_inode(trans, root, dir);
3763 out:
3764         return ret;
3765 }
3766
3767 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3768                        struct btrfs_root *root,
3769                        struct inode *dir, struct inode *inode,
3770                        const char *name, int name_len)
3771 {
3772         int ret;
3773         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3774         if (!ret) {
3775                 drop_nlink(inode);
3776                 ret = btrfs_update_inode(trans, root, inode);
3777         }
3778         return ret;
3779 }
3780
3781 /*
3782  * helper to start transaction for unlink and rmdir.
3783  *
3784  * unlink and rmdir are special in btrfs, they do not always free space, so
3785  * if we cannot make our reservations the normal way try and see if there is
3786  * plenty of slack room in the global reserve to migrate, otherwise we cannot
3787  * allow the unlink to occur.
3788  */
3789 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3790 {
3791         struct btrfs_trans_handle *trans;
3792         struct btrfs_root *root = BTRFS_I(dir)->root;
3793         int ret;
3794
3795         /*
3796          * 1 for the possible orphan item
3797          * 1 for the dir item
3798          * 1 for the dir index
3799          * 1 for the inode ref
3800          * 1 for the inode
3801          */
3802         trans = btrfs_start_transaction(root, 5);
3803         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3804                 return trans;
3805
3806         if (PTR_ERR(trans) == -ENOSPC) {
3807                 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3808
3809                 trans = btrfs_start_transaction(root, 0);
3810                 if (IS_ERR(trans))
3811                         return trans;
3812                 ret = btrfs_cond_migrate_bytes(root->fs_info,
3813                                                &root->fs_info->trans_block_rsv,
3814                                                num_bytes, 5);
3815                 if (ret) {
3816                         btrfs_end_transaction(trans, root);
3817                         return ERR_PTR(ret);
3818                 }
3819                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3820                 trans->bytes_reserved = num_bytes;
3821         }
3822         return trans;
3823 }
3824
3825 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3826 {
3827         struct btrfs_root *root = BTRFS_I(dir)->root;
3828         struct btrfs_trans_handle *trans;
3829         struct inode *inode = dentry->d_inode;
3830         int ret;
3831
3832         trans = __unlink_start_trans(dir);
3833         if (IS_ERR(trans))
3834                 return PTR_ERR(trans);
3835
3836         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3837
3838         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3839                                  dentry->d_name.name, dentry->d_name.len);
3840         if (ret)
3841                 goto out;
3842
3843         if (inode->i_nlink == 0) {
3844                 ret = btrfs_orphan_add(trans, inode);
3845                 if (ret)
3846                         goto out;
3847         }
3848
3849 out:
3850         btrfs_end_transaction(trans, root);
3851         btrfs_btree_balance_dirty(root);
3852         return ret;
3853 }
3854
3855 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3856                         struct btrfs_root *root,
3857                         struct inode *dir, u64 objectid,
3858                         const char *name, int name_len)
3859 {
3860         struct btrfs_path *path;
3861         struct extent_buffer *leaf;
3862         struct btrfs_dir_item *di;
3863         struct btrfs_key key;
3864         u64 index;
3865         int ret;
3866         u64 dir_ino = btrfs_ino(dir);
3867
3868         path = btrfs_alloc_path();
3869         if (!path)
3870                 return -ENOMEM;
3871
3872         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3873                                    name, name_len, -1);
3874         if (IS_ERR_OR_NULL(di)) {
3875                 if (!di)
3876                         ret = -ENOENT;
3877                 else
3878                         ret = PTR_ERR(di);
3879                 goto out;
3880         }
3881
3882         leaf = path->nodes[0];
3883         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3884         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3885         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3886         if (ret) {
3887                 btrfs_abort_transaction(trans, root, ret);
3888                 goto out;
3889         }
3890         btrfs_release_path(path);
3891
3892         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3893                                  objectid, root->root_key.objectid,
3894                                  dir_ino, &index, name, name_len);
3895         if (ret < 0) {
3896                 if (ret != -ENOENT) {
3897                         btrfs_abort_transaction(trans, root, ret);
3898                         goto out;
3899                 }
3900                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3901                                                  name, name_len);
3902                 if (IS_ERR_OR_NULL(di)) {
3903                         if (!di)
3904                                 ret = -ENOENT;
3905                         else
3906                                 ret = PTR_ERR(di);
3907                         btrfs_abort_transaction(trans, root, ret);
3908                         goto out;
3909                 }
3910
3911                 leaf = path->nodes[0];
3912                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3913                 btrfs_release_path(path);
3914                 index = key.offset;
3915         }
3916         btrfs_release_path(path);
3917
3918         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3919         if (ret) {
3920                 btrfs_abort_transaction(trans, root, ret);
3921                 goto out;
3922         }
3923
3924         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3925         inode_inc_iversion(dir);
3926         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3927         ret = btrfs_update_inode_fallback(trans, root, dir);
3928         if (ret)
3929                 btrfs_abort_transaction(trans, root, ret);
3930 out:
3931         btrfs_free_path(path);
3932         return ret;
3933 }
3934
3935 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3936 {
3937         struct inode *inode = dentry->d_inode;
3938         int err = 0;
3939         struct btrfs_root *root = BTRFS_I(dir)->root;
3940         struct btrfs_trans_handle *trans;
3941
3942         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3943                 return -ENOTEMPTY;
3944         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3945                 return -EPERM;
3946
3947         trans = __unlink_start_trans(dir);
3948         if (IS_ERR(trans))
3949                 return PTR_ERR(trans);
3950
3951         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3952                 err = btrfs_unlink_subvol(trans, root, dir,
3953                                           BTRFS_I(inode)->location.objectid,
3954                                           dentry->d_name.name,
3955                                           dentry->d_name.len);
3956                 goto out;
3957         }
3958
3959         err = btrfs_orphan_add(trans, inode);
3960         if (err)
3961                 goto out;
3962
3963         /* now the directory is empty */
3964         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3965                                  dentry->d_name.name, dentry->d_name.len);
3966         if (!err)
3967                 btrfs_i_size_write(inode, 0);
3968 out:
3969         btrfs_end_transaction(trans, root);
3970         btrfs_btree_balance_dirty(root);
3971
3972         return err;
3973 }
3974
3975 /*
3976  * this can truncate away extent items, csum items and directory items.
3977  * It starts at a high offset and removes keys until it can't find
3978  * any higher than new_size
3979  *
3980  * csum items that cross the new i_size are truncated to the new size
3981  * as well.
3982  *
3983  * min_type is the minimum key type to truncate down to.  If set to 0, this
3984  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3985  */
3986 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3987                                struct btrfs_root *root,
3988                                struct inode *inode,
3989                                u64 new_size, u32 min_type)
3990 {
3991         struct btrfs_path *path;
3992         struct extent_buffer *leaf;
3993         struct btrfs_file_extent_item *fi;
3994         struct btrfs_key key;
3995         struct btrfs_key found_key;
3996         u64 extent_start = 0;
3997         u64 extent_num_bytes = 0;
3998         u64 extent_offset = 0;
3999         u64 item_end = 0;
4000         u64 last_size = (u64)-1;
4001         u32 found_type = (u8)-1;
4002         int found_extent;
4003         int del_item;
4004         int pending_del_nr = 0;
4005         int pending_del_slot = 0;
4006         int extent_type = -1;
4007         int ret;
4008         int err = 0;
4009         u64 ino = btrfs_ino(inode);
4010
4011         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4012
4013         path = btrfs_alloc_path();
4014         if (!path)
4015                 return -ENOMEM;
4016         path->reada = -1;
4017
4018         /*
4019          * We want to drop from the next block forward in case this new size is
4020          * not block aligned since we will be keeping the last block of the
4021          * extent just the way it is.
4022          */
4023         if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4024             root == root->fs_info->tree_root)
4025                 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4026                                         root->sectorsize), (u64)-1, 0);
4027
4028         /*
4029          * This function is also used to drop the items in the log tree before
4030          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4031          * it is used to drop the loged items. So we shouldn't kill the delayed
4032          * items.
4033          */
4034         if (min_type == 0 && root == BTRFS_I(inode)->root)
4035                 btrfs_kill_delayed_inode_items(inode);
4036
4037         key.objectid = ino;
4038         key.offset = (u64)-1;
4039         key.type = (u8)-1;
4040
4041 search_again:
4042         path->leave_spinning = 1;
4043         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4044         if (ret < 0) {
4045                 err = ret;
4046                 goto out;
4047         }
4048
4049         if (ret > 0) {
4050                 /* there are no items in the tree for us to truncate, we're
4051                  * done
4052                  */
4053                 if (path->slots[0] == 0)
4054                         goto out;
4055                 path->slots[0]--;
4056         }
4057
4058         while (1) {
4059                 fi = NULL;
4060                 leaf = path->nodes[0];
4061                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4062                 found_type = btrfs_key_type(&found_key);
4063
4064                 if (found_key.objectid != ino)
4065                         break;
4066
4067                 if (found_type < min_type)
4068                         break;
4069
4070                 item_end = found_key.offset;
4071                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4072                         fi = btrfs_item_ptr(leaf, path->slots[0],
4073                                             struct btrfs_file_extent_item);
4074                         extent_type = btrfs_file_extent_type(leaf, fi);
4075                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4076                                 item_end +=
4077                                     btrfs_file_extent_num_bytes(leaf, fi);
4078                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4079                                 item_end += btrfs_file_extent_inline_len(leaf,
4080                                                          path->slots[0], fi);
4081                         }
4082                         item_end--;
4083                 }
4084                 if (found_type > min_type) {
4085                         del_item = 1;
4086                 } else {
4087                         if (item_end < new_size)
4088                                 break;
4089                         if (found_key.offset >= new_size)
4090                                 del_item = 1;
4091                         else
4092                                 del_item = 0;
4093                 }
4094                 found_extent = 0;
4095                 /* FIXME, shrink the extent if the ref count is only 1 */
4096                 if (found_type != BTRFS_EXTENT_DATA_KEY)
4097                         goto delete;
4098
4099                 if (del_item)
4100                         last_size = found_key.offset;
4101                 else
4102                         last_size = new_size;
4103
4104                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4105                         u64 num_dec;
4106                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4107                         if (!del_item) {
4108                                 u64 orig_num_bytes =
4109                                         btrfs_file_extent_num_bytes(leaf, fi);
4110                                 extent_num_bytes = ALIGN(new_size -
4111                                                 found_key.offset,
4112                                                 root->sectorsize);
4113                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4114                                                          extent_num_bytes);
4115                                 num_dec = (orig_num_bytes -
4116                                            extent_num_bytes);
4117                                 if (test_bit(BTRFS_ROOT_REF_COWS,
4118                                              &root->state) &&
4119                                     extent_start != 0)
4120                                         inode_sub_bytes(inode, num_dec);
4121                                 btrfs_mark_buffer_dirty(leaf);
4122                         } else {
4123                                 extent_num_bytes =
4124                                         btrfs_file_extent_disk_num_bytes(leaf,
4125                                                                          fi);
4126                                 extent_offset = found_key.offset -
4127                                         btrfs_file_extent_offset(leaf, fi);
4128
4129                                 /* FIXME blocksize != 4096 */
4130                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4131                                 if (extent_start != 0) {
4132                                         found_extent = 1;
4133                                         if (test_bit(BTRFS_ROOT_REF_COWS,
4134                                                      &root->state))
4135                                                 inode_sub_bytes(inode, num_dec);
4136                                 }
4137                         }
4138                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4139                         /*
4140                          * we can't truncate inline items that have had
4141                          * special encodings
4142                          */
4143                         if (!del_item &&
4144                             btrfs_file_extent_compression(leaf, fi) == 0 &&
4145                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
4146                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4147                                 u32 size = new_size - found_key.offset;
4148
4149                                 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4150                                         inode_sub_bytes(inode, item_end + 1 -
4151                                                         new_size);
4152
4153                                 /*
4154                                  * update the ram bytes to properly reflect
4155                                  * the new size of our item
4156                                  */
4157                                 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4158                                 size =
4159                                     btrfs_file_extent_calc_inline_size(size);
4160                                 btrfs_truncate_item(root, path, size, 1);
4161                         } else if (test_bit(BTRFS_ROOT_REF_COWS,
4162                                             &root->state)) {
4163                                 inode_sub_bytes(inode, item_end + 1 -
4164                                                 found_key.offset);
4165                         }
4166                 }
4167 delete:
4168                 if (del_item) {
4169                         if (!pending_del_nr) {
4170                                 /* no pending yet, add ourselves */
4171                                 pending_del_slot = path->slots[0];
4172                                 pending_del_nr = 1;
4173                         } else if (pending_del_nr &&
4174                                    path->slots[0] + 1 == pending_del_slot) {
4175                                 /* hop on the pending chunk */
4176                                 pending_del_nr++;
4177                                 pending_del_slot = path->slots[0];
4178                         } else {
4179                                 BUG();
4180                         }
4181                 } else {
4182                         break;
4183                 }
4184                 if (found_extent &&
4185                     (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4186                      root == root->fs_info->tree_root)) {
4187                         btrfs_set_path_blocking(path);
4188                         ret = btrfs_free_extent(trans, root, extent_start,
4189                                                 extent_num_bytes, 0,
4190                                                 btrfs_header_owner(leaf),
4191                                                 ino, extent_offset, 0);
4192                         BUG_ON(ret);
4193                 }
4194
4195                 if (found_type == BTRFS_INODE_ITEM_KEY)
4196                         break;
4197
4198                 if (path->slots[0] == 0 ||
4199                     path->slots[0] != pending_del_slot) {
4200                         if (pending_del_nr) {
4201                                 ret = btrfs_del_items(trans, root, path,
4202                                                 pending_del_slot,
4203                                                 pending_del_nr);
4204                                 if (ret) {
4205                                         btrfs_abort_transaction(trans,
4206                                                                 root, ret);
4207                                         goto error;
4208                                 }
4209                                 pending_del_nr = 0;
4210                         }
4211                         btrfs_release_path(path);
4212                         goto search_again;
4213                 } else {
4214                         path->slots[0]--;
4215                 }
4216         }
4217 out:
4218         if (pending_del_nr) {
4219                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4220                                       pending_del_nr);
4221                 if (ret)
4222                         btrfs_abort_transaction(trans, root, ret);
4223         }
4224 error:
4225         if (last_size != (u64)-1)
4226                 btrfs_ordered_update_i_size(inode, last_size, NULL);
4227         btrfs_free_path(path);
4228         return err;
4229 }
4230
4231 /*
4232  * btrfs_truncate_page - read, zero a chunk and write a page
4233  * @inode - inode that we're zeroing
4234  * @from - the offset to start zeroing
4235  * @len - the length to zero, 0 to zero the entire range respective to the
4236  *      offset
4237  * @front - zero up to the offset instead of from the offset on
4238  *
4239  * This will find the page for the "from" offset and cow the page and zero the
4240  * part we want to zero.  This is used with truncate and hole punching.
4241  */
4242 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4243                         int front)
4244 {
4245         struct address_space *mapping = inode->i_mapping;
4246         struct btrfs_root *root = BTRFS_I(inode)->root;
4247         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4248         struct btrfs_ordered_extent *ordered;
4249         struct extent_state *cached_state = NULL;
4250         char *kaddr;
4251         u32 blocksize = root->sectorsize;
4252         pgoff_t index = from >> PAGE_CACHE_SHIFT;
4253         unsigned offset = from & (PAGE_CACHE_SIZE-1);
4254         struct page *page;
4255         gfp_t mask = btrfs_alloc_write_mask(mapping);
4256         int ret = 0;
4257         u64 page_start;
4258         u64 page_end;
4259
4260         if ((offset & (blocksize - 1)) == 0 &&
4261             (!len || ((len & (blocksize - 1)) == 0)))
4262                 goto out;
4263         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4264         if (ret)
4265                 goto out;
4266
4267 again:
4268         page = find_or_create_page(mapping, index, mask);
4269         if (!page) {
4270                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4271                 ret = -ENOMEM;
4272                 goto out;
4273         }
4274
4275         page_start = page_offset(page);
4276         page_end = page_start + PAGE_CACHE_SIZE - 1;
4277
4278         if (!PageUptodate(page)) {
4279                 ret = btrfs_readpage(NULL, page);
4280                 lock_page(page);
4281                 if (page->mapping != mapping) {
4282                         unlock_page(page);
4283                         page_cache_release(page);
4284                         goto again;
4285                 }
4286                 if (!PageUptodate(page)) {
4287                         ret = -EIO;
4288                         goto out_unlock;
4289                 }
4290         }
4291         wait_on_page_writeback(page);
4292
4293         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4294         set_page_extent_mapped(page);
4295
4296         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4297         if (ordered) {
4298                 unlock_extent_cached(io_tree, page_start, page_end,
4299                                      &cached_state, GFP_NOFS);
4300                 unlock_page(page);
4301                 page_cache_release(page);
4302                 btrfs_start_ordered_extent(inode, ordered, 1);
4303                 btrfs_put_ordered_extent(ordered);
4304                 goto again;
4305         }
4306
4307         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4308                           EXTENT_DIRTY | EXTENT_DELALLOC |
4309                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4310                           0, 0, &cached_state, GFP_NOFS);
4311
4312         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4313                                         &cached_state);
4314         if (ret) {
4315                 unlock_extent_cached(io_tree, page_start, page_end,
4316                                      &cached_state, GFP_NOFS);
4317                 goto out_unlock;
4318         }
4319
4320         if (offset != PAGE_CACHE_SIZE) {
4321                 if (!len)
4322                         len = PAGE_CACHE_SIZE - offset;
4323                 kaddr = kmap(page);
4324                 if (front)
4325                         memset(kaddr, 0, offset);
4326                 else
4327                         memset(kaddr + offset, 0, len);
4328                 flush_dcache_page(page);
4329                 kunmap(page);
4330         }
4331         ClearPageChecked(page);
4332         set_page_dirty(page);
4333         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4334                              GFP_NOFS);
4335
4336 out_unlock:
4337         if (ret)
4338                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4339         unlock_page(page);
4340         page_cache_release(page);
4341 out:
4342         return ret;
4343 }
4344
4345 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4346                              u64 offset, u64 len)
4347 {
4348         struct btrfs_trans_handle *trans;
4349         int ret;
4350
4351         /*
4352          * Still need to make sure the inode looks like it's been updated so
4353          * that any holes get logged if we fsync.
4354          */
4355         if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4356                 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4357                 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4358                 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4359                 return 0;
4360         }
4361
4362         /*
4363          * 1 - for the one we're dropping
4364          * 1 - for the one we're adding
4365          * 1 - for updating the inode.
4366          */
4367         trans = btrfs_start_transaction(root, 3);
4368         if (IS_ERR(trans))
4369                 return PTR_ERR(trans);
4370
4371         ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4372         if (ret) {
4373                 btrfs_abort_transaction(trans, root, ret);
4374                 btrfs_end_transaction(trans, root);
4375                 return ret;
4376         }
4377
4378         ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4379                                        0, 0, len, 0, len, 0, 0, 0);
4380         if (ret)
4381                 btrfs_abort_transaction(trans, root, ret);
4382         else
4383                 btrfs_update_inode(trans, root, inode);
4384         btrfs_end_transaction(trans, root);
4385         return ret;
4386 }
4387
4388 /*
4389  * This function puts in dummy file extents for the area we're creating a hole
4390  * for.  So if we are truncating this file to a larger size we need to insert
4391  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4392  * the range between oldsize and size
4393  */
4394 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4395 {
4396         struct btrfs_root *root = BTRFS_I(inode)->root;
4397         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4398         struct extent_map *em = NULL;
4399         struct extent_state *cached_state = NULL;
4400         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4401         u64 hole_start = ALIGN(oldsize, root->sectorsize);
4402         u64 block_end = ALIGN(size, root->sectorsize);
4403         u64 last_byte;
4404         u64 cur_offset;
4405         u64 hole_size;
4406         int err = 0;
4407
4408         /*
4409          * If our size started in the middle of a page we need to zero out the
4410          * rest of the page before we expand the i_size, otherwise we could
4411          * expose stale data.
4412          */
4413         err = btrfs_truncate_page(inode, oldsize, 0, 0);
4414         if (err)
4415                 return err;
4416
4417         if (size <= hole_start)
4418                 return 0;
4419
4420         while (1) {
4421                 struct btrfs_ordered_extent *ordered;
4422
4423                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4424                                  &cached_state);
4425                 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4426                                                      block_end - hole_start);
4427                 if (!ordered)
4428                         break;
4429                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4430                                      &cached_state, GFP_NOFS);
4431                 btrfs_start_ordered_extent(inode, ordered, 1);
4432                 btrfs_put_ordered_extent(ordered);
4433         }
4434
4435         cur_offset = hole_start;
4436         while (1) {
4437                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4438                                 block_end - cur_offset, 0);
4439                 if (IS_ERR(em)) {
4440                         err = PTR_ERR(em);
4441                         em = NULL;
4442                         break;
4443                 }
4444                 last_byte = min(extent_map_end(em), block_end);
4445                 last_byte = ALIGN(last_byte , root->sectorsize);
4446                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4447                         struct extent_map *hole_em;
4448                         hole_size = last_byte - cur_offset;
4449
4450                         err = maybe_insert_hole(root, inode, cur_offset,
4451                                                 hole_size);
4452                         if (err)
4453                                 break;
4454                         btrfs_drop_extent_cache(inode, cur_offset,
4455                                                 cur_offset + hole_size - 1, 0);
4456                         hole_em = alloc_extent_map();
4457                         if (!hole_em) {
4458                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4459                                         &BTRFS_I(inode)->runtime_flags);
4460                                 goto next;
4461                         }
4462                         hole_em->start = cur_offset;
4463                         hole_em->len = hole_size;
4464                         hole_em->orig_start = cur_offset;
4465
4466                         hole_em->block_start = EXTENT_MAP_HOLE;
4467                         hole_em->block_len = 0;
4468                         hole_em->orig_block_len = 0;
4469                         hole_em->ram_bytes = hole_size;
4470                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4471                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
4472                         hole_em->generation = root->fs_info->generation;
4473
4474                         while (1) {
4475                                 write_lock(&em_tree->lock);
4476                                 err = add_extent_mapping(em_tree, hole_em, 1);
4477                                 write_unlock(&em_tree->lock);
4478                                 if (err != -EEXIST)
4479                                         break;
4480                                 btrfs_drop_extent_cache(inode, cur_offset,
4481                                                         cur_offset +
4482                                                         hole_size - 1, 0);
4483                         }
4484                         free_extent_map(hole_em);
4485                 }
4486 next:
4487                 free_extent_map(em);
4488                 em = NULL;
4489                 cur_offset = last_byte;
4490                 if (cur_offset >= block_end)
4491                         break;
4492         }
4493         free_extent_map(em);
4494         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4495                              GFP_NOFS);
4496         return err;
4497 }
4498
4499 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4500 {
4501         struct btrfs_root *root = BTRFS_I(inode)->root;
4502         struct btrfs_trans_handle *trans;
4503         loff_t oldsize = i_size_read(inode);
4504         loff_t newsize = attr->ia_size;
4505         int mask = attr->ia_valid;
4506         int ret;
4507
4508         /*
4509          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4510          * special case where we need to update the times despite not having
4511          * these flags set.  For all other operations the VFS set these flags
4512          * explicitly if it wants a timestamp update.
4513          */
4514         if (newsize != oldsize) {
4515                 inode_inc_iversion(inode);
4516                 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4517                         inode->i_ctime = inode->i_mtime =
4518                                 current_fs_time(inode->i_sb);
4519         }
4520
4521         if (newsize > oldsize) {
4522                 truncate_pagecache(inode, newsize);
4523                 ret = btrfs_cont_expand(inode, oldsize, newsize);
4524                 if (ret)
4525                         return ret;
4526
4527                 trans = btrfs_start_transaction(root, 1);
4528                 if (IS_ERR(trans))
4529                         return PTR_ERR(trans);
4530
4531                 i_size_write(inode, newsize);
4532                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4533                 ret = btrfs_update_inode(trans, root, inode);
4534                 btrfs_end_transaction(trans, root);
4535         } else {
4536
4537                 /*
4538                  * We're truncating a file that used to have good data down to
4539                  * zero. Make sure it gets into the ordered flush list so that
4540                  * any new writes get down to disk quickly.
4541                  */
4542                 if (newsize == 0)
4543                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4544                                 &BTRFS_I(inode)->runtime_flags);
4545
4546                 /*
4547                  * 1 for the orphan item we're going to add
4548                  * 1 for the orphan item deletion.
4549                  */
4550                 trans = btrfs_start_transaction(root, 2);
4551                 if (IS_ERR(trans))
4552                         return PTR_ERR(trans);
4553
4554                 /*
4555                  * We need to do this in case we fail at _any_ point during the
4556                  * actual truncate.  Once we do the truncate_setsize we could
4557                  * invalidate pages which forces any outstanding ordered io to
4558                  * be instantly completed which will give us extents that need
4559                  * to be truncated.  If we fail to get an orphan inode down we
4560                  * could have left over extents that were never meant to live,
4561                  * so we need to garuntee from this point on that everything
4562                  * will be consistent.
4563                  */
4564                 ret = btrfs_orphan_add(trans, inode);
4565                 btrfs_end_transaction(trans, root);
4566                 if (ret)
4567                         return ret;
4568
4569                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4570                 truncate_setsize(inode, newsize);
4571
4572                 /* Disable nonlocked read DIO to avoid the end less truncate */
4573                 btrfs_inode_block_unlocked_dio(inode);
4574                 inode_dio_wait(inode);
4575                 btrfs_inode_resume_unlocked_dio(inode);
4576
4577                 ret = btrfs_truncate(inode);
4578                 if (ret && inode->i_nlink) {
4579                         int err;
4580
4581                         /*
4582                          * failed to truncate, disk_i_size is only adjusted down
4583                          * as we remove extents, so it should represent the true
4584                          * size of the inode, so reset the in memory size and
4585                          * delete our orphan entry.
4586                          */
4587                         trans = btrfs_join_transaction(root);
4588                         if (IS_ERR(trans)) {
4589                                 btrfs_orphan_del(NULL, inode);
4590                                 return ret;
4591                         }
4592                         i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4593                         err = btrfs_orphan_del(trans, inode);
4594                         if (err)
4595                                 btrfs_abort_transaction(trans, root, err);
4596                         btrfs_end_transaction(trans, root);
4597                 }
4598         }
4599
4600         return ret;
4601 }
4602
4603 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4604 {
4605         struct inode *inode = dentry->d_inode;
4606         struct btrfs_root *root = BTRFS_I(inode)->root;
4607         int err;
4608
4609         if (btrfs_root_readonly(root))
4610                 return -EROFS;
4611
4612         err = inode_change_ok(inode, attr);
4613         if (err)
4614                 return err;
4615
4616         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4617                 err = btrfs_setsize(inode, attr);
4618                 if (err)
4619                         return err;
4620         }
4621
4622         if (attr->ia_valid) {
4623                 setattr_copy(inode, attr);
4624                 inode_inc_iversion(inode);
4625                 err = btrfs_dirty_inode(inode);
4626
4627                 if (!err && attr->ia_valid & ATTR_MODE)
4628                         err = posix_acl_chmod(inode, inode->i_mode);
4629         }
4630
4631         return err;
4632 }
4633
4634 /*
4635  * While truncating the inode pages during eviction, we get the VFS calling
4636  * btrfs_invalidatepage() against each page of the inode. This is slow because
4637  * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4638  * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4639  * extent_state structures over and over, wasting lots of time.
4640  *
4641  * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4642  * those expensive operations on a per page basis and do only the ordered io
4643  * finishing, while we release here the extent_map and extent_state structures,
4644  * without the excessive merging and splitting.
4645  */
4646 static void evict_inode_truncate_pages(struct inode *inode)
4647 {
4648         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4649         struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4650         struct rb_node *node;
4651
4652         ASSERT(inode->i_state & I_FREEING);
4653         truncate_inode_pages_final(&inode->i_data);
4654
4655         write_lock(&map_tree->lock);
4656         while (!RB_EMPTY_ROOT(&map_tree->map)) {
4657                 struct extent_map *em;
4658
4659                 node = rb_first(&map_tree->map);
4660                 em = rb_entry(node, struct extent_map, rb_node);
4661                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4662                 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4663                 remove_extent_mapping(map_tree, em);
4664                 free_extent_map(em);
4665         }
4666         write_unlock(&map_tree->lock);
4667
4668         spin_lock(&io_tree->lock);
4669         while (!RB_EMPTY_ROOT(&io_tree->state)) {
4670                 struct extent_state *state;
4671                 struct extent_state *cached_state = NULL;
4672
4673                 node = rb_first(&io_tree->state);
4674                 state = rb_entry(node, struct extent_state, rb_node);
4675                 atomic_inc(&state->refs);
4676                 spin_unlock(&io_tree->lock);
4677
4678                 lock_extent_bits(io_tree, state->start, state->end,
4679                                  0, &cached_state);
4680                 clear_extent_bit(io_tree, state->start, state->end,
4681                                  EXTENT_LOCKED | EXTENT_DIRTY |
4682                                  EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
4683                                  EXTENT_DEFRAG, 1, 1,
4684                                  &cached_state, GFP_NOFS);
4685                 free_extent_state(state);
4686
4687                 spin_lock(&io_tree->lock);
4688         }
4689         spin_unlock(&io_tree->lock);
4690 }
4691
4692 void btrfs_evict_inode(struct inode *inode)
4693 {
4694         struct btrfs_trans_handle *trans;
4695         struct btrfs_root *root = BTRFS_I(inode)->root;
4696         struct btrfs_block_rsv *rsv, *global_rsv;
4697         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4698         int ret;
4699
4700         trace_btrfs_inode_evict(inode);
4701
4702         evict_inode_truncate_pages(inode);
4703
4704         if (inode->i_nlink &&
4705             ((btrfs_root_refs(&root->root_item) != 0 &&
4706               root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4707              btrfs_is_free_space_inode(inode)))
4708                 goto no_delete;
4709
4710         if (is_bad_inode(inode)) {
4711                 btrfs_orphan_del(NULL, inode);
4712                 goto no_delete;
4713         }
4714         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4715         btrfs_wait_ordered_range(inode, 0, (u64)-1);
4716
4717         if (root->fs_info->log_root_recovering) {
4718                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4719                                  &BTRFS_I(inode)->runtime_flags));
4720                 goto no_delete;
4721         }
4722
4723         if (inode->i_nlink > 0) {
4724                 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4725                        root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
4726                 goto no_delete;
4727         }
4728
4729         ret = btrfs_commit_inode_delayed_inode(inode);
4730         if (ret) {
4731                 btrfs_orphan_del(NULL, inode);
4732                 goto no_delete;
4733         }
4734
4735         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4736         if (!rsv) {
4737                 btrfs_orphan_del(NULL, inode);
4738                 goto no_delete;
4739         }
4740         rsv->size = min_size;
4741         rsv->failfast = 1;
4742         global_rsv = &root->fs_info->global_block_rsv;
4743
4744         btrfs_i_size_write(inode, 0);
4745
4746         /*
4747          * This is a bit simpler than btrfs_truncate since we've already
4748          * reserved our space for our orphan item in the unlink, so we just
4749          * need to reserve some slack space in case we add bytes and update
4750          * inode item when doing the truncate.
4751          */
4752         while (1) {
4753                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4754                                              BTRFS_RESERVE_FLUSH_LIMIT);
4755
4756                 /*
4757                  * Try and steal from the global reserve since we will
4758                  * likely not use this space anyway, we want to try as
4759                  * hard as possible to get this to work.
4760                  */
4761                 if (ret)
4762                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4763
4764                 if (ret) {
4765                         btrfs_warn(root->fs_info,
4766                                 "Could not get space for a delete, will truncate on mount %d",
4767                                 ret);
4768                         btrfs_orphan_del(NULL, inode);
4769                         btrfs_free_block_rsv(root, rsv);
4770                         goto no_delete;
4771                 }
4772
4773                 trans = btrfs_join_transaction(root);
4774                 if (IS_ERR(trans)) {
4775                         btrfs_orphan_del(NULL, inode);
4776                         btrfs_free_block_rsv(root, rsv);
4777                         goto no_delete;
4778                 }
4779
4780                 trans->block_rsv = rsv;
4781
4782                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4783                 if (ret != -ENOSPC)
4784                         break;
4785
4786                 trans->block_rsv = &root->fs_info->trans_block_rsv;
4787                 btrfs_end_transaction(trans, root);
4788                 trans = NULL;
4789                 btrfs_btree_balance_dirty(root);
4790         }
4791
4792         btrfs_free_block_rsv(root, rsv);
4793
4794         /*
4795          * Errors here aren't a big deal, it just means we leave orphan items
4796          * in the tree.  They will be cleaned up on the next mount.
4797          */
4798         if (ret == 0) {
4799                 trans->block_rsv = root->orphan_block_rsv;
4800                 btrfs_orphan_del(trans, inode);
4801         } else {
4802                 btrfs_orphan_del(NULL, inode);
4803         }
4804
4805         trans->block_rsv = &root->fs_info->trans_block_rsv;
4806         if (!(root == root->fs_info->tree_root ||
4807               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4808                 btrfs_return_ino(root, btrfs_ino(inode));
4809
4810         btrfs_end_transaction(trans, root);
4811         btrfs_btree_balance_dirty(root);
4812 no_delete:
4813         btrfs_remove_delayed_node(inode);
4814         clear_inode(inode);
4815         return;
4816 }
4817
4818 /*
4819  * this returns the key found in the dir entry in the location pointer.
4820  * If no dir entries were found, location->objectid is 0.
4821  */
4822 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4823                                struct btrfs_key *location)
4824 {
4825         const char *name = dentry->d_name.name;
4826         int namelen = dentry->d_name.len;
4827         struct btrfs_dir_item *di;
4828         struct btrfs_path *path;
4829         struct btrfs_root *root = BTRFS_I(dir)->root;
4830         int ret = 0;
4831
4832         path = btrfs_alloc_path();
4833         if (!path)
4834                 return -ENOMEM;
4835
4836         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4837                                     namelen, 0);
4838         if (IS_ERR(di))
4839                 ret = PTR_ERR(di);
4840
4841         if (IS_ERR_OR_NULL(di))
4842                 goto out_err;
4843
4844         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4845 out:
4846         btrfs_free_path(path);
4847         return ret;
4848 out_err:
4849         location->objectid = 0;
4850         goto out;
4851 }
4852
4853 /*
4854  * when we hit a tree root in a directory, the btrfs part of the inode
4855  * needs to be changed to reflect the root directory of the tree root.  This
4856  * is kind of like crossing a mount point.
4857  */
4858 static int fixup_tree_root_location(struct btrfs_root *root,
4859                                     struct inode *dir,
4860                                     struct dentry *dentry,
4861                                     struct btrfs_key *location,
4862                                     struct btrfs_root **sub_root)
4863 {
4864         struct btrfs_path *path;
4865         struct btrfs_root *new_root;
4866         struct btrfs_root_ref *ref;
4867         struct extent_buffer *leaf;
4868         int ret;
4869         int err = 0;
4870
4871         path = btrfs_alloc_path();
4872         if (!path) {
4873                 err = -ENOMEM;
4874                 goto out;
4875         }
4876
4877         err = -ENOENT;
4878         ret = btrfs_find_item(root->fs_info->tree_root, path,
4879                                 BTRFS_I(dir)->root->root_key.objectid,
4880                                 location->objectid, BTRFS_ROOT_REF_KEY, NULL);
4881         if (ret) {
4882                 if (ret < 0)
4883                         err = ret;
4884                 goto out;
4885         }
4886
4887         leaf = path->nodes[0];
4888         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4889         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4890             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4891                 goto out;
4892
4893         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4894                                    (unsigned long)(ref + 1),
4895                                    dentry->d_name.len);
4896         if (ret)
4897                 goto out;
4898
4899         btrfs_release_path(path);
4900
4901         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4902         if (IS_ERR(new_root)) {
4903                 err = PTR_ERR(new_root);
4904                 goto out;
4905         }
4906
4907         *sub_root = new_root;
4908         location->objectid = btrfs_root_dirid(&new_root->root_item);
4909         location->type = BTRFS_INODE_ITEM_KEY;
4910         location->offset = 0;
4911         err = 0;
4912 out:
4913         btrfs_free_path(path);
4914         return err;
4915 }
4916
4917 static void inode_tree_add(struct inode *inode)
4918 {
4919         struct btrfs_root *root = BTRFS_I(inode)->root;
4920         struct btrfs_inode *entry;
4921         struct rb_node **p;
4922         struct rb_node *parent;
4923         struct rb_node *new = &BTRFS_I(inode)->rb_node;
4924         u64 ino = btrfs_ino(inode);
4925
4926         if (inode_unhashed(inode))
4927                 return;
4928         parent = NULL;
4929         spin_lock(&root->inode_lock);
4930         p = &root->inode_tree.rb_node;
4931         while (*p) {
4932                 parent = *p;
4933                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4934
4935                 if (ino < btrfs_ino(&entry->vfs_inode))
4936                         p = &parent->rb_left;
4937                 else if (ino > btrfs_ino(&entry->vfs_inode))
4938                         p = &parent->rb_right;
4939                 else {
4940                         WARN_ON(!(entry->vfs_inode.i_state &
4941                                   (I_WILL_FREE | I_FREEING)));
4942                         rb_replace_node(parent, new, &root->inode_tree);
4943                         RB_CLEAR_NODE(parent);
4944                         spin_unlock(&root->inode_lock);
4945                         return;
4946                 }
4947         }
4948         rb_link_node(new, parent, p);
4949         rb_insert_color(new, &root->inode_tree);
4950         spin_unlock(&root->inode_lock);
4951 }
4952
4953 static void inode_tree_del(struct inode *inode)
4954 {
4955         struct btrfs_root *root = BTRFS_I(inode)->root;
4956         int empty = 0;
4957
4958         spin_lock(&root->inode_lock);
4959         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4960                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4961                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4962                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4963         }
4964         spin_unlock(&root->inode_lock);
4965
4966         if (empty && btrfs_root_refs(&root->root_item) == 0) {
4967                 synchronize_srcu(&root->fs_info->subvol_srcu);
4968                 spin_lock(&root->inode_lock);
4969                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4970                 spin_unlock(&root->inode_lock);
4971                 if (empty)
4972                         btrfs_add_dead_root(root);
4973         }
4974 }
4975
4976 void btrfs_invalidate_inodes(struct btrfs_root *root)
4977 {
4978         struct rb_node *node;
4979         struct rb_node *prev;
4980         struct btrfs_inode *entry;
4981         struct inode *inode;
4982         u64 objectid = 0;
4983
4984         if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
4985                 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4986
4987         spin_lock(&root->inode_lock);
4988 again:
4989         node = root->inode_tree.rb_node;
4990         prev = NULL;
4991         while (node) {
4992                 prev = node;
4993                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4994
4995                 if (objectid < btrfs_ino(&entry->vfs_inode))
4996                         node = node->rb_left;
4997                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4998                         node = node->rb_right;
4999                 else
5000                         break;
5001         }
5002         if (!node) {
5003                 while (prev) {
5004                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
5005                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5006                                 node = prev;
5007                                 break;
5008                         }
5009                         prev = rb_next(prev);
5010                 }
5011         }
5012         while (node) {
5013                 entry = rb_entry(node, struct btrfs_inode, rb_node);
5014                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5015                 inode = igrab(&entry->vfs_inode);
5016                 if (inode) {
5017                         spin_unlock(&root->inode_lock);
5018                         if (atomic_read(&inode->i_count) > 1)
5019                                 d_prune_aliases(inode);
5020                         /*
5021                          * btrfs_drop_inode will have it removed from
5022                          * the inode cache when its usage count
5023                          * hits zero.
5024                          */
5025                         iput(inode);
5026                         cond_resched();
5027                         spin_lock(&root->inode_lock);
5028                         goto again;
5029                 }
5030
5031                 if (cond_resched_lock(&root->inode_lock))
5032                         goto again;
5033
5034                 node = rb_next(node);
5035         }
5036         spin_unlock(&root->inode_lock);
5037 }
5038
5039 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5040 {
5041         struct btrfs_iget_args *args = p;
5042         inode->i_ino = args->location->objectid;
5043         memcpy(&BTRFS_I(inode)->location, args->location,
5044                sizeof(*args->location));
5045         BTRFS_I(inode)->root = args->root;
5046         return 0;
5047 }
5048
5049 static int btrfs_find_actor(struct inode *inode, void *opaque)
5050 {
5051         struct btrfs_iget_args *args = opaque;
5052         return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5053                 args->root == BTRFS_I(inode)->root;
5054 }
5055
5056 static struct inode *btrfs_iget_locked(struct super_block *s,
5057                                        struct btrfs_key *location,
5058                                        struct btrfs_root *root)
5059 {
5060         struct inode *inode;
5061         struct btrfs_iget_args args;
5062         unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5063
5064         args.location = location;
5065         args.root = root;
5066
5067         inode = iget5_locked(s, hashval, btrfs_find_actor,
5068                              btrfs_init_locked_inode,
5069                              (void *)&args);
5070         return inode;
5071 }
5072
5073 /* Get an inode object given its location and corresponding root.
5074  * Returns in *is_new if the inode was read from disk
5075  */
5076 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5077                          struct btrfs_root *root, int *new)
5078 {
5079         struct inode *inode;
5080
5081         inode = btrfs_iget_locked(s, location, root);
5082         if (!inode)
5083                 return ERR_PTR(-ENOMEM);
5084
5085         if (inode->i_state & I_NEW) {
5086                 btrfs_read_locked_inode(inode);
5087                 if (!is_bad_inode(inode)) {
5088                         inode_tree_add(inode);
5089                         unlock_new_inode(inode);
5090                         if (new)
5091                                 *new = 1;
5092                 } else {
5093                         unlock_new_inode(inode);
5094                         iput(inode);
5095                         inode = ERR_PTR(-ESTALE);
5096                 }
5097         }
5098
5099         return inode;
5100 }
5101
5102 static struct inode *new_simple_dir(struct super_block *s,
5103                                     struct btrfs_key *key,
5104                                     struct btrfs_root *root)
5105 {
5106         struct inode *inode = new_inode(s);
5107
5108         if (!inode)
5109                 return ERR_PTR(-ENOMEM);
5110
5111         BTRFS_I(inode)->root = root;
5112         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5113         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5114
5115         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5116         inode->i_op = &btrfs_dir_ro_inode_operations;
5117         inode->i_fop = &simple_dir_operations;
5118         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5119         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5120
5121         return inode;
5122 }
5123
5124 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5125 {
5126         struct inode *inode;
5127         struct btrfs_root *root = BTRFS_I(dir)->root;
5128         struct btrfs_root *sub_root = root;
5129         struct btrfs_key location;
5130         int index;
5131         int ret = 0;
5132
5133         if (dentry->d_name.len > BTRFS_NAME_LEN)
5134                 return ERR_PTR(-ENAMETOOLONG);
5135
5136         ret = btrfs_inode_by_name(dir, dentry, &location);
5137         if (ret < 0)
5138                 return ERR_PTR(ret);
5139
5140         if (location.objectid == 0)
5141                 return ERR_PTR(-ENOENT);
5142
5143         if (location.type == BTRFS_INODE_ITEM_KEY) {
5144                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5145                 return inode;
5146         }
5147
5148         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5149
5150         index = srcu_read_lock(&root->fs_info->subvol_srcu);
5151         ret = fixup_tree_root_location(root, dir, dentry,
5152                                        &location, &sub_root);
5153         if (ret < 0) {
5154                 if (ret != -ENOENT)
5155                         inode = ERR_PTR(ret);
5156                 else
5157                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
5158         } else {
5159                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5160         }
5161         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5162
5163         if (!IS_ERR(inode) && root != sub_root) {
5164                 down_read(&root->fs_info->cleanup_work_sem);
5165                 if (!(inode->i_sb->s_flags & MS_RDONLY))
5166                         ret = btrfs_orphan_cleanup(sub_root);
5167                 up_read(&root->fs_info->cleanup_work_sem);
5168                 if (ret) {
5169                         iput(inode);
5170                         inode = ERR_PTR(ret);
5171                 }
5172         }
5173
5174         return inode;
5175 }
5176
5177 static int btrfs_dentry_delete(const struct dentry *dentry)
5178 {
5179         struct btrfs_root *root;
5180         struct inode *inode = dentry->d_inode;
5181
5182         if (!inode && !IS_ROOT(dentry))
5183                 inode = dentry->d_parent->d_inode;
5184
5185         if (inode) {
5186                 root = BTRFS_I(inode)->root;
5187                 if (btrfs_root_refs(&root->root_item) == 0)
5188                         return 1;
5189
5190                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5191                         return 1;
5192         }
5193         return 0;
5194 }
5195
5196 static void btrfs_dentry_release(struct dentry *dentry)
5197 {
5198         kfree(dentry->d_fsdata);
5199 }
5200
5201 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5202                                    unsigned int flags)
5203 {
5204         struct inode *inode;
5205
5206         inode = btrfs_lookup_dentry(dir, dentry);
5207         if (IS_ERR(inode)) {
5208                 if (PTR_ERR(inode) == -ENOENT)
5209                         inode = NULL;
5210                 else
5211                         return ERR_CAST(inode);
5212         }
5213
5214         return d_materialise_unique(dentry, inode);
5215 }
5216
5217 unsigned char btrfs_filetype_table[] = {
5218         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5219 };
5220
5221 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5222 {
5223         struct inode *inode = file_inode(file);
5224         struct btrfs_root *root = BTRFS_I(inode)->root;
5225         struct btrfs_item *item;
5226         struct btrfs_dir_item *di;
5227         struct btrfs_key key;
5228         struct btrfs_key found_key;
5229         struct btrfs_path *path;
5230         struct list_head ins_list;
5231         struct list_head del_list;
5232         int ret;
5233         struct extent_buffer *leaf;
5234         int slot;
5235         unsigned char d_type;
5236         int over = 0;
5237         u32 di_cur;
5238         u32 di_total;
5239         u32 di_len;
5240         int key_type = BTRFS_DIR_INDEX_KEY;
5241         char tmp_name[32];
5242         char *name_ptr;
5243         int name_len;
5244         int is_curr = 0;        /* ctx->pos points to the current index? */
5245
5246         /* FIXME, use a real flag for deciding about the key type */
5247         if (root->fs_info->tree_root == root)
5248                 key_type = BTRFS_DIR_ITEM_KEY;
5249
5250         if (!dir_emit_dots(file, ctx))
5251                 return 0;
5252
5253         path = btrfs_alloc_path();
5254         if (!path)
5255                 return -ENOMEM;
5256
5257         path->reada = 1;
5258
5259         if (key_type == BTRFS_DIR_INDEX_KEY) {
5260                 INIT_LIST_HEAD(&ins_list);
5261                 INIT_LIST_HEAD(&del_list);
5262                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5263         }
5264
5265         btrfs_set_key_type(&key, key_type);
5266         key.offset = ctx->pos;
5267         key.objectid = btrfs_ino(inode);
5268
5269         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5270         if (ret < 0)
5271                 goto err;
5272
5273         while (1) {
5274                 leaf = path->nodes[0];
5275                 slot = path->slots[0];
5276                 if (slot >= btrfs_header_nritems(leaf)) {
5277                         ret = btrfs_next_leaf(root, path);
5278                         if (ret < 0)
5279                                 goto err;
5280                         else if (ret > 0)
5281                                 break;
5282                         continue;
5283                 }
5284
5285                 item = btrfs_item_nr(slot);
5286                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5287
5288                 if (found_key.objectid != key.objectid)
5289                         break;
5290                 if (btrfs_key_type(&found_key) != key_type)
5291                         break;
5292                 if (found_key.offset < ctx->pos)
5293                         goto next;
5294                 if (key_type == BTRFS_DIR_INDEX_KEY &&
5295                     btrfs_should_delete_dir_index(&del_list,
5296                                                   found_key.offset))
5297                         goto next;
5298
5299                 ctx->pos = found_key.offset;
5300                 is_curr = 1;
5301
5302                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5303                 di_cur = 0;
5304                 di_total = btrfs_item_size(leaf, item);
5305
5306                 while (di_cur < di_total) {
5307                         struct btrfs_key location;
5308
5309                         if (verify_dir_item(root, leaf, di))
5310                                 break;
5311
5312                         name_len = btrfs_dir_name_len(leaf, di);
5313                         if (name_len <= sizeof(tmp_name)) {
5314                                 name_ptr = tmp_name;
5315                         } else {
5316                                 name_ptr = kmalloc(name_len, GFP_NOFS);
5317                                 if (!name_ptr) {
5318                                         ret = -ENOMEM;
5319                                         goto err;
5320                                 }
5321                         }
5322                         read_extent_buffer(leaf, name_ptr,
5323                                            (unsigned long)(di + 1), name_len);
5324
5325                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5326                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
5327
5328
5329                         /* is this a reference to our own snapshot? If so
5330                          * skip it.
5331                          *
5332                          * In contrast to old kernels, we insert the snapshot's
5333                          * dir item and dir index after it has been created, so
5334                          * we won't find a reference to our own snapshot. We
5335                          * still keep the following code for backward
5336                          * compatibility.
5337                          */
5338                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
5339                             location.objectid == root->root_key.objectid) {
5340                                 over = 0;
5341                                 goto skip;
5342                         }
5343                         over = !dir_emit(ctx, name_ptr, name_len,
5344                                        location.objectid, d_type);
5345
5346 skip:
5347                         if (name_ptr != tmp_name)
5348                                 kfree(name_ptr);
5349
5350                         if (over)
5351                                 goto nopos;
5352                         di_len = btrfs_dir_name_len(leaf, di) +
5353                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
5354                         di_cur += di_len;
5355                         di = (struct btrfs_dir_item *)((char *)di + di_len);
5356                 }
5357 next:
5358                 path->slots[0]++;
5359         }
5360
5361         if (key_type == BTRFS_DIR_INDEX_KEY) {
5362                 if (is_curr)
5363                         ctx->pos++;
5364                 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5365                 if (ret)
5366                         goto nopos;
5367         }
5368
5369         /* Reached end of directory/root. Bump pos past the last item. */
5370         ctx->pos++;
5371
5372         /*
5373          * Stop new entries from being returned after we return the last
5374          * entry.
5375          *
5376          * New directory entries are assigned a strictly increasing
5377          * offset.  This means that new entries created during readdir
5378          * are *guaranteed* to be seen in the future by that readdir.
5379          * This has broken buggy programs which operate on names as
5380          * they're returned by readdir.  Until we re-use freed offsets
5381          * we have this hack to stop new entries from being returned
5382          * under the assumption that they'll never reach this huge
5383          * offset.
5384          *
5385          * This is being careful not to overflow 32bit loff_t unless the
5386          * last entry requires it because doing so has broken 32bit apps
5387          * in the past.
5388          */
5389         if (key_type == BTRFS_DIR_INDEX_KEY) {
5390                 if (ctx->pos >= INT_MAX)
5391                         ctx->pos = LLONG_MAX;
5392                 else
5393                         ctx->pos = INT_MAX;
5394         }
5395 nopos:
5396         ret = 0;
5397 err:
5398         if (key_type == BTRFS_DIR_INDEX_KEY)
5399                 btrfs_put_delayed_items(&ins_list, &del_list);
5400         btrfs_free_path(path);
5401         return ret;
5402 }
5403
5404 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5405 {
5406         struct btrfs_root *root = BTRFS_I(inode)->root;
5407         struct btrfs_trans_handle *trans;
5408         int ret = 0;
5409         bool nolock = false;
5410
5411         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5412                 return 0;
5413
5414         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5415                 nolock = true;
5416
5417         if (wbc->sync_mode == WB_SYNC_ALL) {
5418                 if (nolock)
5419                         trans = btrfs_join_transaction_nolock(root);
5420                 else
5421                         trans = btrfs_join_transaction(root);
5422                 if (IS_ERR(trans))
5423                         return PTR_ERR(trans);
5424                 ret = btrfs_commit_transaction(trans, root);
5425         }
5426         return ret;
5427 }
5428
5429 /*
5430  * This is somewhat expensive, updating the tree every time the
5431  * inode changes.  But, it is most likely to find the inode in cache.
5432  * FIXME, needs more benchmarking...there are no reasons other than performance
5433  * to keep or drop this code.
5434  */
5435 static int btrfs_dirty_inode(struct inode *inode)
5436 {
5437         struct btrfs_root *root = BTRFS_I(inode)->root;
5438         struct btrfs_trans_handle *trans;
5439         int ret;
5440
5441         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5442                 return 0;
5443
5444         trans = btrfs_join_transaction(root);
5445         if (IS_ERR(trans))
5446                 return PTR_ERR(trans);
5447
5448         ret = btrfs_update_inode(trans, root, inode);
5449         if (ret && ret == -ENOSPC) {
5450                 /* whoops, lets try again with the full transaction */
5451                 btrfs_end_transaction(trans, root);
5452                 trans = btrfs_start_transaction(root, 1);
5453                 if (IS_ERR(trans))
5454                         return PTR_ERR(trans);
5455
5456                 ret = btrfs_update_inode(trans, root, inode);
5457         }
5458         btrfs_end_transaction(trans, root);
5459         if (BTRFS_I(inode)->delayed_node)
5460                 btrfs_balance_delayed_items(root);
5461
5462         return ret;
5463 }
5464
5465 /*
5466  * This is a copy of file_update_time.  We need this so we can return error on
5467  * ENOSPC for updating the inode in the case of file write and mmap writes.
5468  */
5469 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5470                              int flags)
5471 {
5472         struct btrfs_root *root = BTRFS_I(inode)->root;
5473
5474         if (btrfs_root_readonly(root))
5475                 return -EROFS;
5476
5477         if (flags & S_VERSION)
5478                 inode_inc_iversion(inode);
5479         if (flags & S_CTIME)
5480                 inode->i_ctime = *now;
5481         if (flags & S_MTIME)
5482                 inode->i_mtime = *now;
5483         if (flags & S_ATIME)
5484                 inode->i_atime = *now;
5485         return btrfs_dirty_inode(inode);
5486 }
5487
5488 /*
5489  * find the highest existing sequence number in a directory
5490  * and then set the in-memory index_cnt variable to reflect
5491  * free sequence numbers
5492  */
5493 static int btrfs_set_inode_index_count(struct inode *inode)
5494 {
5495         struct btrfs_root *root = BTRFS_I(inode)->root;
5496         struct btrfs_key key, found_key;
5497         struct btrfs_path *path;
5498         struct extent_buffer *leaf;
5499         int ret;
5500
5501         key.objectid = btrfs_ino(inode);
5502         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5503         key.offset = (u64)-1;
5504
5505         path = btrfs_alloc_path();
5506         if (!path)
5507                 return -ENOMEM;
5508
5509         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5510         if (ret < 0)
5511                 goto out;
5512         /* FIXME: we should be able to handle this */
5513         if (ret == 0)
5514                 goto out;
5515         ret = 0;
5516
5517         /*
5518          * MAGIC NUMBER EXPLANATION:
5519          * since we search a directory based on f_pos we have to start at 2
5520          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5521          * else has to start at 2
5522          */
5523         if (path->slots[0] == 0) {
5524                 BTRFS_I(inode)->index_cnt = 2;
5525                 goto out;
5526         }
5527
5528         path->slots[0]--;
5529
5530         leaf = path->nodes[0];
5531         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5532
5533         if (found_key.objectid != btrfs_ino(inode) ||
5534             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5535                 BTRFS_I(inode)->index_cnt = 2;
5536                 goto out;
5537         }
5538
5539         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5540 out:
5541         btrfs_free_path(path);
5542         return ret;
5543 }
5544
5545 /*
5546  * helper to find a free sequence number in a given directory.  This current
5547  * code is very simple, later versions will do smarter things in the btree
5548  */
5549 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5550 {
5551         int ret = 0;
5552
5553         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5554                 ret = btrfs_inode_delayed_dir_index_count(dir);
5555                 if (ret) {
5556                         ret = btrfs_set_inode_index_count(dir);
5557                         if (ret)
5558                                 return ret;
5559                 }
5560         }
5561
5562         *index = BTRFS_I(dir)->index_cnt;
5563         BTRFS_I(dir)->index_cnt++;
5564
5565         return ret;
5566 }
5567
5568 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5569                                      struct btrfs_root *root,
5570                                      struct inode *dir,
5571                                      const char *name, int name_len,
5572                                      u64 ref_objectid, u64 objectid,
5573                                      umode_t mode, u64 *index)
5574 {
5575         struct inode *inode;
5576         struct btrfs_inode_item *inode_item;
5577         struct btrfs_key *location;
5578         struct btrfs_path *path;
5579         struct btrfs_inode_ref *ref;
5580         struct btrfs_key key[2];
5581         u32 sizes[2];
5582         int nitems = name ? 2 : 1;
5583         unsigned long ptr;
5584         int ret;
5585
5586         path = btrfs_alloc_path();
5587         if (!path)
5588                 return ERR_PTR(-ENOMEM);
5589
5590         inode = new_inode(root->fs_info->sb);
5591         if (!inode) {
5592                 btrfs_free_path(path);
5593                 return ERR_PTR(-ENOMEM);
5594         }
5595
5596         /*
5597          * we have to initialize this early, so we can reclaim the inode
5598          * number if we fail afterwards in this function.
5599          */
5600         inode->i_ino = objectid;
5601
5602         if (dir && name) {
5603                 trace_btrfs_inode_request(dir);
5604
5605                 ret = btrfs_set_inode_index(dir, index);
5606                 if (ret) {
5607                         btrfs_free_path(path);
5608                         iput(inode);
5609                         return ERR_PTR(ret);
5610                 }
5611         } else if (dir) {
5612                 *index = 0;
5613         }
5614         /*
5615          * index_cnt is ignored for everything but a dir,
5616          * btrfs_get_inode_index_count has an explanation for the magic
5617          * number
5618          */
5619         BTRFS_I(inode)->index_cnt = 2;
5620         BTRFS_I(inode)->dir_index = *index;
5621         BTRFS_I(inode)->root = root;
5622         BTRFS_I(inode)->generation = trans->transid;
5623         inode->i_generation = BTRFS_I(inode)->generation;
5624
5625         /*
5626          * We could have gotten an inode number from somebody who was fsynced
5627          * and then removed in this same transaction, so let's just set full
5628          * sync since it will be a full sync anyway and this will blow away the
5629          * old info in the log.
5630          */
5631         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5632
5633         key[0].objectid = objectid;
5634         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5635         key[0].offset = 0;
5636
5637         sizes[0] = sizeof(struct btrfs_inode_item);
5638
5639         if (name) {
5640                 /*
5641                  * Start new inodes with an inode_ref. This is slightly more
5642                  * efficient for small numbers of hard links since they will
5643                  * be packed into one item. Extended refs will kick in if we
5644                  * add more hard links than can fit in the ref item.
5645                  */
5646                 key[1].objectid = objectid;
5647                 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5648                 key[1].offset = ref_objectid;
5649
5650                 sizes[1] = name_len + sizeof(*ref);
5651         }
5652
5653         path->leave_spinning = 1;
5654         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
5655         if (ret != 0)
5656                 goto fail;
5657
5658         inode_init_owner(inode, dir, mode);
5659         inode_set_bytes(inode, 0);
5660         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5661         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5662                                   struct btrfs_inode_item);
5663         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5664                              sizeof(*inode_item));
5665         fill_inode_item(trans, path->nodes[0], inode_item, inode);
5666
5667         if (name) {
5668                 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5669                                      struct btrfs_inode_ref);
5670                 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5671                 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5672                 ptr = (unsigned long)(ref + 1);
5673                 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5674         }
5675
5676         btrfs_mark_buffer_dirty(path->nodes[0]);
5677         btrfs_free_path(path);
5678
5679         location = &BTRFS_I(inode)->location;
5680         location->objectid = objectid;
5681         location->offset = 0;
5682         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5683
5684         btrfs_inherit_iflags(inode, dir);
5685
5686         if (S_ISREG(mode)) {
5687                 if (btrfs_test_opt(root, NODATASUM))
5688                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5689                 if (btrfs_test_opt(root, NODATACOW))
5690                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5691                                 BTRFS_INODE_NODATASUM;
5692         }
5693
5694         btrfs_insert_inode_hash(inode);
5695         inode_tree_add(inode);
5696
5697         trace_btrfs_inode_new(inode);
5698         btrfs_set_inode_last_trans(trans, inode);
5699
5700         btrfs_update_root_times(trans, root);
5701
5702         ret = btrfs_inode_inherit_props(trans, inode, dir);
5703         if (ret)
5704                 btrfs_err(root->fs_info,
5705                           "error inheriting props for ino %llu (root %llu): %d",
5706                           btrfs_ino(inode), root->root_key.objectid, ret);
5707
5708         return inode;
5709 fail:
5710         if (dir && name)
5711                 BTRFS_I(dir)->index_cnt--;
5712         btrfs_free_path(path);
5713         iput(inode);
5714         return ERR_PTR(ret);
5715 }
5716
5717 static inline u8 btrfs_inode_type(struct inode *inode)
5718 {
5719         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5720 }
5721
5722 /*
5723  * utility function to add 'inode' into 'parent_inode' with
5724  * a give name and a given sequence number.
5725  * if 'add_backref' is true, also insert a backref from the
5726  * inode to the parent directory.
5727  */
5728 int btrfs_add_link(struct btrfs_trans_handle *trans,
5729                    struct inode *parent_inode, struct inode *inode,
5730                    const char *name, int name_len, int add_backref, u64 index)
5731 {
5732         int ret = 0;
5733         struct btrfs_key key;
5734         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5735         u64 ino = btrfs_ino(inode);
5736         u64 parent_ino = btrfs_ino(parent_inode);
5737
5738         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5739                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5740         } else {
5741                 key.objectid = ino;
5742                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5743                 key.offset = 0;
5744         }
5745
5746         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5747                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5748                                          key.objectid, root->root_key.objectid,
5749                                          parent_ino, index, name, name_len);
5750         } else if (add_backref) {
5751                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5752                                              parent_ino, index);
5753         }
5754
5755         /* Nothing to clean up yet */
5756         if (ret)
5757                 return ret;
5758
5759         ret = btrfs_insert_dir_item(trans, root, name, name_len,
5760                                     parent_inode, &key,
5761                                     btrfs_inode_type(inode), index);
5762         if (ret == -EEXIST || ret == -EOVERFLOW)
5763                 goto fail_dir_item;
5764         else if (ret) {
5765                 btrfs_abort_transaction(trans, root, ret);
5766                 return ret;
5767         }
5768
5769         btrfs_i_size_write(parent_inode, parent_inode->i_size +
5770                            name_len * 2);
5771         inode_inc_iversion(parent_inode);
5772         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5773         ret = btrfs_update_inode(trans, root, parent_inode);
5774         if (ret)
5775                 btrfs_abort_transaction(trans, root, ret);
5776         return ret;
5777
5778 fail_dir_item:
5779         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5780                 u64 local_index;
5781                 int err;
5782                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5783                                  key.objectid, root->root_key.objectid,
5784                                  parent_ino, &local_index, name, name_len);
5785
5786         } else if (add_backref) {
5787                 u64 local_index;
5788                 int err;
5789
5790                 err = btrfs_del_inode_ref(trans, root, name, name_len,
5791                                           ino, parent_ino, &local_index);
5792         }
5793         return ret;
5794 }
5795
5796 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5797                             struct inode *dir, struct dentry *dentry,
5798                             struct inode *inode, int backref, u64 index)
5799 {
5800         int err = btrfs_add_link(trans, dir, inode,
5801                                  dentry->d_name.name, dentry->d_name.len,
5802                                  backref, index);
5803         if (err > 0)
5804                 err = -EEXIST;
5805         return err;
5806 }
5807
5808 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5809                         umode_t mode, dev_t rdev)
5810 {
5811         struct btrfs_trans_handle *trans;
5812         struct btrfs_root *root = BTRFS_I(dir)->root;
5813         struct inode *inode = NULL;
5814         int err;
5815         int drop_inode = 0;
5816         u64 objectid;
5817         u64 index = 0;
5818
5819         if (!new_valid_dev(rdev))
5820                 return -EINVAL;
5821
5822         /*
5823          * 2 for inode item and ref
5824          * 2 for dir items
5825          * 1 for xattr if selinux is on
5826          */
5827         trans = btrfs_start_transaction(root, 5);
5828         if (IS_ERR(trans))
5829                 return PTR_ERR(trans);
5830
5831         err = btrfs_find_free_ino(root, &objectid);
5832         if (err)
5833                 goto out_unlock;
5834
5835         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5836                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5837                                 mode, &index);
5838         if (IS_ERR(inode)) {
5839                 err = PTR_ERR(inode);
5840                 goto out_unlock;
5841         }
5842
5843         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5844         if (err) {
5845                 drop_inode = 1;
5846                 goto out_unlock;
5847         }
5848
5849         /*
5850         * If the active LSM wants to access the inode during
5851         * d_instantiate it needs these. Smack checks to see
5852         * if the filesystem supports xattrs by looking at the
5853         * ops vector.
5854         */
5855
5856         inode->i_op = &btrfs_special_inode_operations;
5857         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5858         if (err)
5859                 drop_inode = 1;
5860         else {
5861                 init_special_inode(inode, inode->i_mode, rdev);
5862                 btrfs_update_inode(trans, root, inode);
5863                 d_instantiate(dentry, inode);
5864         }
5865 out_unlock:
5866         btrfs_end_transaction(trans, root);
5867         btrfs_balance_delayed_items(root);
5868         btrfs_btree_balance_dirty(root);
5869         if (drop_inode) {
5870                 inode_dec_link_count(inode);
5871                 iput(inode);
5872         }
5873         return err;
5874 }
5875
5876 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5877                         umode_t mode, bool excl)
5878 {
5879         struct btrfs_trans_handle *trans;
5880         struct btrfs_root *root = BTRFS_I(dir)->root;
5881         struct inode *inode = NULL;
5882         int drop_inode_on_err = 0;
5883         int err;
5884         u64 objectid;
5885         u64 index = 0;
5886
5887         /*
5888          * 2 for inode item and ref
5889          * 2 for dir items
5890          * 1 for xattr if selinux is on
5891          */
5892         trans = btrfs_start_transaction(root, 5);
5893         if (IS_ERR(trans))
5894                 return PTR_ERR(trans);
5895
5896         err = btrfs_find_free_ino(root, &objectid);
5897         if (err)
5898                 goto out_unlock;
5899
5900         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5901                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5902                                 mode, &index);
5903         if (IS_ERR(inode)) {
5904                 err = PTR_ERR(inode);
5905                 goto out_unlock;
5906         }
5907         drop_inode_on_err = 1;
5908
5909         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5910         if (err)
5911                 goto out_unlock;
5912
5913         err = btrfs_update_inode(trans, root, inode);
5914         if (err)
5915                 goto out_unlock;
5916
5917         /*
5918         * If the active LSM wants to access the inode during
5919         * d_instantiate it needs these. Smack checks to see
5920         * if the filesystem supports xattrs by looking at the
5921         * ops vector.
5922         */
5923         inode->i_fop = &btrfs_file_operations;
5924         inode->i_op = &btrfs_file_inode_operations;
5925
5926         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5927         if (err)
5928                 goto out_unlock;
5929
5930         inode->i_mapping->a_ops = &btrfs_aops;
5931         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5932         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5933         d_instantiate(dentry, inode);
5934
5935 out_unlock:
5936         btrfs_end_transaction(trans, root);
5937         if (err && drop_inode_on_err) {
5938                 inode_dec_link_count(inode);
5939                 iput(inode);
5940         }
5941         btrfs_balance_delayed_items(root);
5942         btrfs_btree_balance_dirty(root);
5943         return err;
5944 }
5945
5946 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5947                       struct dentry *dentry)
5948 {
5949         struct btrfs_trans_handle *trans;
5950         struct btrfs_root *root = BTRFS_I(dir)->root;
5951         struct inode *inode = old_dentry->d_inode;
5952         u64 index;
5953         int err;
5954         int drop_inode = 0;
5955
5956         /* do not allow sys_link's with other subvols of the same device */
5957         if (root->objectid != BTRFS_I(inode)->root->objectid)
5958                 return -EXDEV;
5959
5960         if (inode->i_nlink >= BTRFS_LINK_MAX)
5961                 return -EMLINK;
5962
5963         err = btrfs_set_inode_index(dir, &index);
5964         if (err)
5965                 goto fail;
5966
5967         /*
5968          * 2 items for inode and inode ref
5969          * 2 items for dir items
5970          * 1 item for parent inode
5971          */
5972         trans = btrfs_start_transaction(root, 5);
5973         if (IS_ERR(trans)) {
5974                 err = PTR_ERR(trans);
5975                 goto fail;
5976         }
5977
5978         /* There are several dir indexes for this inode, clear the cache. */
5979         BTRFS_I(inode)->dir_index = 0ULL;
5980         inc_nlink(inode);
5981         inode_inc_iversion(inode);
5982         inode->i_ctime = CURRENT_TIME;
5983         ihold(inode);
5984         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5985
5986         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5987
5988         if (err) {
5989                 drop_inode = 1;
5990         } else {
5991                 struct dentry *parent = dentry->d_parent;
5992                 err = btrfs_update_inode(trans, root, inode);
5993                 if (err)
5994                         goto fail;
5995                 if (inode->i_nlink == 1) {
5996                         /*
5997                          * If new hard link count is 1, it's a file created
5998                          * with open(2) O_TMPFILE flag.
5999                          */
6000                         err = btrfs_orphan_del(trans, inode);
6001                         if (err)
6002                                 goto fail;
6003                 }
6004                 d_instantiate(dentry, inode);
6005                 btrfs_log_new_name(trans, inode, NULL, parent);
6006         }
6007
6008         btrfs_end_transaction(trans, root);
6009         btrfs_balance_delayed_items(root);
6010 fail:
6011         if (drop_inode) {
6012                 inode_dec_link_count(inode);
6013                 iput(inode);
6014         }
6015         btrfs_btree_balance_dirty(root);
6016         return err;
6017 }
6018
6019 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6020 {
6021         struct inode *inode = NULL;
6022         struct btrfs_trans_handle *trans;
6023         struct btrfs_root *root = BTRFS_I(dir)->root;
6024         int err = 0;
6025         int drop_on_err = 0;
6026         u64 objectid = 0;
6027         u64 index = 0;
6028
6029         /*
6030          * 2 items for inode and ref
6031          * 2 items for dir items
6032          * 1 for xattr if selinux is on
6033          */
6034         trans = btrfs_start_transaction(root, 5);
6035         if (IS_ERR(trans))
6036                 return PTR_ERR(trans);
6037
6038         err = btrfs_find_free_ino(root, &objectid);
6039         if (err)
6040                 goto out_fail;
6041
6042         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6043                                 dentry->d_name.len, btrfs_ino(dir), objectid,
6044                                 S_IFDIR | mode, &index);
6045         if (IS_ERR(inode)) {
6046                 err = PTR_ERR(inode);
6047                 goto out_fail;
6048         }
6049
6050         drop_on_err = 1;
6051
6052         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6053         if (err)
6054                 goto out_fail;
6055
6056         inode->i_op = &btrfs_dir_inode_operations;
6057         inode->i_fop = &btrfs_dir_file_operations;
6058
6059         btrfs_i_size_write(inode, 0);
6060         err = btrfs_update_inode(trans, root, inode);
6061         if (err)
6062                 goto out_fail;
6063
6064         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6065                              dentry->d_name.len, 0, index);
6066         if (err)
6067                 goto out_fail;
6068
6069         d_instantiate(dentry, inode);
6070         drop_on_err = 0;
6071
6072 out_fail:
6073         btrfs_end_transaction(trans, root);
6074         if (drop_on_err)
6075                 iput(inode);
6076         btrfs_balance_delayed_items(root);
6077         btrfs_btree_balance_dirty(root);
6078         return err;
6079 }
6080
6081 /* helper for btfs_get_extent.  Given an existing extent in the tree,
6082  * and an extent that you want to insert, deal with overlap and insert
6083  * the new extent into the tree.
6084  */
6085 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6086                                 struct extent_map *existing,
6087                                 struct extent_map *em,
6088                                 u64 map_start, u64 map_len)
6089 {
6090         u64 start_diff;
6091
6092         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6093         start_diff = map_start - em->start;
6094         em->start = map_start;
6095         em->len = map_len;
6096         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6097             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6098                 em->block_start += start_diff;
6099                 em->block_len -= start_diff;
6100         }
6101         return add_extent_mapping(em_tree, em, 0);
6102 }
6103
6104 static noinline int uncompress_inline(struct btrfs_path *path,
6105                                       struct inode *inode, struct page *page,
6106                                       size_t pg_offset, u64 extent_offset,
6107                                       struct btrfs_file_extent_item *item)
6108 {
6109         int ret;
6110         struct extent_buffer *leaf = path->nodes[0];
6111         char *tmp;
6112         size_t max_size;
6113         unsigned long inline_size;
6114         unsigned long ptr;
6115         int compress_type;
6116
6117         WARN_ON(pg_offset != 0);
6118         compress_type = btrfs_file_extent_compression(leaf, item);
6119         max_size = btrfs_file_extent_ram_bytes(leaf, item);
6120         inline_size = btrfs_file_extent_inline_item_len(leaf,
6121                                         btrfs_item_nr(path->slots[0]));
6122         tmp = kmalloc(inline_size, GFP_NOFS);
6123         if (!tmp)
6124                 return -ENOMEM;
6125         ptr = btrfs_file_extent_inline_start(item);
6126
6127         read_extent_buffer(leaf, tmp, ptr, inline_size);
6128
6129         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6130         ret = btrfs_decompress(compress_type, tmp, page,
6131                                extent_offset, inline_size, max_size);
6132         kfree(tmp);
6133         return ret;
6134 }
6135
6136 /*
6137  * a bit scary, this does extent mapping from logical file offset to the disk.
6138  * the ugly parts come from merging extents from the disk with the in-ram
6139  * representation.  This gets more complex because of the data=ordered code,
6140  * where the in-ram extents might be locked pending data=ordered completion.
6141  *
6142  * This also copies inline extents directly into the page.
6143  */
6144
6145 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6146                                     size_t pg_offset, u64 start, u64 len,
6147                                     int create)
6148 {
6149         int ret;
6150         int err = 0;
6151         u64 extent_start = 0;
6152         u64 extent_end = 0;
6153         u64 objectid = btrfs_ino(inode);
6154         u32 found_type;
6155         struct btrfs_path *path = NULL;
6156         struct btrfs_root *root = BTRFS_I(inode)->root;
6157         struct btrfs_file_extent_item *item;
6158         struct extent_buffer *leaf;
6159         struct btrfs_key found_key;
6160         struct extent_map *em = NULL;
6161         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6162         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6163         struct btrfs_trans_handle *trans = NULL;
6164         const bool new_inline = !page || create;
6165
6166 again:
6167         read_lock(&em_tree->lock);
6168         em = lookup_extent_mapping(em_tree, start, len);
6169         if (em)
6170                 em->bdev = root->fs_info->fs_devices->latest_bdev;
6171         read_unlock(&em_tree->lock);
6172
6173         if (em) {
6174                 if (em->start > start || em->start + em->len <= start)
6175                         free_extent_map(em);
6176                 else if (em->block_start == EXTENT_MAP_INLINE && page)
6177                         free_extent_map(em);
6178                 else
6179                         goto out;
6180         }
6181         em = alloc_extent_map();
6182         if (!em) {
6183                 err = -ENOMEM;
6184                 goto out;
6185         }
6186         em->bdev = root->fs_info->fs_devices->latest_bdev;
6187         em->start = EXTENT_MAP_HOLE;
6188         em->orig_start = EXTENT_MAP_HOLE;
6189         em->len = (u64)-1;
6190         em->block_len = (u64)-1;
6191
6192         if (!path) {
6193                 path = btrfs_alloc_path();
6194                 if (!path) {
6195                         err = -ENOMEM;
6196                         goto out;
6197                 }
6198                 /*
6199                  * Chances are we'll be called again, so go ahead and do
6200                  * readahead
6201                  */
6202                 path->reada = 1;
6203         }
6204
6205         ret = btrfs_lookup_file_extent(trans, root, path,
6206                                        objectid, start, trans != NULL);
6207         if (ret < 0) {
6208                 err = ret;
6209                 goto out;
6210         }
6211
6212         if (ret != 0) {
6213                 if (path->slots[0] == 0)
6214                         goto not_found;
6215                 path->slots[0]--;
6216         }
6217
6218         leaf = path->nodes[0];
6219         item = btrfs_item_ptr(leaf, path->slots[0],
6220                               struct btrfs_file_extent_item);
6221         /* are we inside the extent that was found? */
6222         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6223         found_type = btrfs_key_type(&found_key);
6224         if (found_key.objectid != objectid ||
6225             found_type != BTRFS_EXTENT_DATA_KEY) {
6226                 /*
6227                  * If we backup past the first extent we want to move forward
6228                  * and see if there is an extent in front of us, otherwise we'll
6229                  * say there is a hole for our whole search range which can
6230                  * cause problems.
6231                  */
6232                 extent_end = start;
6233                 goto next;
6234         }
6235
6236         found_type = btrfs_file_extent_type(leaf, item);
6237         extent_start = found_key.offset;
6238         if (found_type == BTRFS_FILE_EXTENT_REG ||
6239             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6240                 extent_end = extent_start +
6241                        btrfs_file_extent_num_bytes(leaf, item);
6242         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6243                 size_t size;
6244                 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6245                 extent_end = ALIGN(extent_start + size, root->sectorsize);
6246         }
6247 next:
6248         if (start >= extent_end) {
6249                 path->slots[0]++;
6250                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6251                         ret = btrfs_next_leaf(root, path);
6252                         if (ret < 0) {
6253                                 err = ret;
6254                                 goto out;
6255                         }
6256                         if (ret > 0)
6257                                 goto not_found;
6258                         leaf = path->nodes[0];
6259                 }
6260                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6261                 if (found_key.objectid != objectid ||
6262                     found_key.type != BTRFS_EXTENT_DATA_KEY)
6263                         goto not_found;
6264                 if (start + len <= found_key.offset)
6265                         goto not_found;
6266                 em->start = start;
6267                 em->orig_start = start;
6268                 em->len = found_key.offset - start;
6269                 goto not_found_em;
6270         }
6271
6272         btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6273
6274         if (found_type == BTRFS_FILE_EXTENT_REG ||
6275             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6276                 goto insert;
6277         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6278                 unsigned long ptr;
6279                 char *map;
6280                 size_t size;
6281                 size_t extent_offset;
6282                 size_t copy_size;
6283
6284                 if (new_inline)
6285                         goto out;
6286
6287                 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6288                 extent_offset = page_offset(page) + pg_offset - extent_start;
6289                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6290                                 size - extent_offset);
6291                 em->start = extent_start + extent_offset;
6292                 em->len = ALIGN(copy_size, root->sectorsize);
6293                 em->orig_block_len = em->len;
6294                 em->orig_start = em->start;
6295                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6296                 if (create == 0 && !PageUptodate(page)) {
6297                         if (btrfs_file_extent_compression(leaf, item) !=
6298                             BTRFS_COMPRESS_NONE) {
6299                                 ret = uncompress_inline(path, inode, page,
6300                                                         pg_offset,
6301                                                         extent_offset, item);
6302                                 if (ret) {
6303                                         err = ret;
6304                                         goto out;
6305                                 }
6306                         } else {
6307                                 map = kmap(page);
6308                                 read_extent_buffer(leaf, map + pg_offset, ptr,
6309                                                    copy_size);
6310                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6311                                         memset(map + pg_offset + copy_size, 0,
6312                                                PAGE_CACHE_SIZE - pg_offset -
6313                                                copy_size);
6314                                 }
6315                                 kunmap(page);
6316                         }
6317                         flush_dcache_page(page);
6318                 } else if (create && PageUptodate(page)) {
6319                         BUG();
6320                         if (!trans) {
6321                                 kunmap(page);
6322                                 free_extent_map(em);
6323                                 em = NULL;
6324
6325                                 btrfs_release_path(path);
6326                                 trans = btrfs_join_transaction(root);
6327
6328                                 if (IS_ERR(trans))
6329                                         return ERR_CAST(trans);
6330                                 goto again;
6331                         }
6332                         map = kmap(page);
6333                         write_extent_buffer(leaf, map + pg_offset, ptr,
6334                                             copy_size);
6335                         kunmap(page);
6336                         btrfs_mark_buffer_dirty(leaf);
6337                 }
6338                 set_extent_uptodate(io_tree, em->start,
6339                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
6340                 goto insert;
6341         }
6342 not_found:
6343         em->start = start;
6344         em->orig_start = start;
6345         em->len = len;
6346 not_found_em:
6347         em->block_start = EXTENT_MAP_HOLE;
6348         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6349 insert:
6350         btrfs_release_path(path);
6351         if (em->start > start || extent_map_end(em) <= start) {
6352                 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6353                         em->start, em->len, start, len);
6354                 err = -EIO;
6355                 goto out;
6356         }
6357
6358         err = 0;
6359         write_lock(&em_tree->lock);
6360         ret = add_extent_mapping(em_tree, em, 0);
6361         /* it is possible that someone inserted the extent into the tree
6362          * while we had the lock dropped.  It is also possible that
6363          * an overlapping map exists in the tree
6364          */
6365         if (ret == -EEXIST) {
6366                 struct extent_map *existing;
6367
6368                 ret = 0;
6369
6370                 existing = lookup_extent_mapping(em_tree, start, len);
6371                 if (existing && (existing->start > start ||
6372                     existing->start + existing->len <= start)) {
6373                         free_extent_map(existing);
6374                         existing = NULL;
6375                 }
6376                 if (!existing) {
6377                         existing = lookup_extent_mapping(em_tree, em->start,
6378                                                          em->len);
6379                         if (existing) {
6380                                 err = merge_extent_mapping(em_tree, existing,
6381                                                            em, start,
6382                                                            root->sectorsize);
6383                                 free_extent_map(existing);
6384                                 if (err) {
6385                                         free_extent_map(em);
6386                                         em = NULL;
6387                                 }
6388                         } else {
6389                                 err = -EIO;
6390                                 free_extent_map(em);
6391                                 em = NULL;
6392                         }
6393                 } else {
6394                         free_extent_map(em);
6395                         em = existing;
6396                         err = 0;
6397                 }
6398         }
6399         write_unlock(&em_tree->lock);
6400 out:
6401
6402         trace_btrfs_get_extent(root, em);
6403
6404         if (path)
6405                 btrfs_free_path(path);
6406         if (trans) {
6407                 ret = btrfs_end_transaction(trans, root);
6408                 if (!err)
6409                         err = ret;
6410         }
6411         if (err) {
6412                 free_extent_map(em);
6413                 return ERR_PTR(err);
6414         }
6415         BUG_ON(!em); /* Error is always set */
6416         return em;
6417 }
6418
6419 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6420                                            size_t pg_offset, u64 start, u64 len,
6421                                            int create)
6422 {
6423         struct extent_map *em;
6424         struct extent_map *hole_em = NULL;
6425         u64 range_start = start;
6426         u64 end;
6427         u64 found;
6428         u64 found_end;
6429         int err = 0;
6430
6431         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6432         if (IS_ERR(em))
6433                 return em;
6434         if (em) {
6435                 /*
6436                  * if our em maps to
6437                  * -  a hole or
6438                  * -  a pre-alloc extent,
6439                  * there might actually be delalloc bytes behind it.
6440                  */
6441                 if (em->block_start != EXTENT_MAP_HOLE &&
6442                     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6443                         return em;
6444                 else
6445                         hole_em = em;
6446         }
6447
6448         /* check to see if we've wrapped (len == -1 or similar) */
6449         end = start + len;
6450         if (end < start)
6451                 end = (u64)-1;
6452         else
6453                 end -= 1;
6454
6455         em = NULL;
6456
6457         /* ok, we didn't find anything, lets look for delalloc */
6458         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6459                                  end, len, EXTENT_DELALLOC, 1);
6460         found_end = range_start + found;
6461         if (found_end < range_start)
6462                 found_end = (u64)-1;
6463
6464         /*
6465          * we didn't find anything useful, return
6466          * the original results from get_extent()
6467          */
6468         if (range_start > end || found_end <= start) {
6469                 em = hole_em;
6470                 hole_em = NULL;
6471                 goto out;
6472         }
6473
6474         /* adjust the range_start to make sure it doesn't
6475          * go backwards from the start they passed in
6476          */
6477         range_start = max(start, range_start);
6478         found = found_end - range_start;
6479
6480         if (found > 0) {
6481                 u64 hole_start = start;
6482                 u64 hole_len = len;
6483
6484                 em = alloc_extent_map();
6485                 if (!em) {
6486                         err = -ENOMEM;
6487                         goto out;
6488                 }
6489                 /*
6490                  * when btrfs_get_extent can't find anything it
6491                  * returns one huge hole
6492                  *
6493                  * make sure what it found really fits our range, and
6494                  * adjust to make sure it is based on the start from
6495                  * the caller
6496                  */
6497                 if (hole_em) {
6498                         u64 calc_end = extent_map_end(hole_em);
6499
6500                         if (calc_end <= start || (hole_em->start > end)) {
6501                                 free_extent_map(hole_em);
6502                                 hole_em = NULL;
6503                         } else {
6504                                 hole_start = max(hole_em->start, start);
6505                                 hole_len = calc_end - hole_start;
6506                         }
6507                 }
6508                 em->bdev = NULL;
6509                 if (hole_em && range_start > hole_start) {
6510                         /* our hole starts before our delalloc, so we
6511                          * have to return just the parts of the hole
6512                          * that go until  the delalloc starts
6513                          */
6514                         em->len = min(hole_len,
6515                                       range_start - hole_start);
6516                         em->start = hole_start;
6517                         em->orig_start = hole_start;
6518                         /*
6519                          * don't adjust block start at all,
6520                          * it is fixed at EXTENT_MAP_HOLE
6521                          */
6522                         em->block_start = hole_em->block_start;
6523                         em->block_len = hole_len;
6524                         if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6525                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6526                 } else {
6527                         em->start = range_start;
6528                         em->len = found;
6529                         em->orig_start = range_start;
6530                         em->block_start = EXTENT_MAP_DELALLOC;
6531                         em->block_len = found;
6532                 }
6533         } else if (hole_em) {
6534                 return hole_em;
6535         }
6536 out:
6537
6538         free_extent_map(hole_em);
6539         if (err) {
6540                 free_extent_map(em);
6541                 return ERR_PTR(err);
6542         }
6543         return em;
6544 }
6545
6546 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6547                                                   u64 start, u64 len)
6548 {
6549         struct btrfs_root *root = BTRFS_I(inode)->root;
6550         struct extent_map *em;
6551         struct btrfs_key ins;
6552         u64 alloc_hint;
6553         int ret;
6554
6555         alloc_hint = get_extent_allocation_hint(inode, start, len);
6556         ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
6557                                    alloc_hint, &ins, 1, 1);
6558         if (ret)
6559                 return ERR_PTR(ret);
6560
6561         em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6562                               ins.offset, ins.offset, ins.offset, 0);
6563         if (IS_ERR(em)) {
6564                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
6565                 return em;
6566         }
6567
6568         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6569                                            ins.offset, ins.offset, 0);
6570         if (ret) {
6571                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
6572                 free_extent_map(em);
6573                 return ERR_PTR(ret);
6574         }
6575
6576         return em;
6577 }
6578
6579 /*
6580  * returns 1 when the nocow is safe, < 1 on error, 0 if the
6581  * block must be cow'd
6582  */
6583 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6584                               u64 *orig_start, u64 *orig_block_len,
6585                               u64 *ram_bytes)
6586 {
6587         struct btrfs_trans_handle *trans;
6588         struct btrfs_path *path;
6589         int ret;
6590         struct extent_buffer *leaf;
6591         struct btrfs_root *root = BTRFS_I(inode)->root;
6592         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6593         struct btrfs_file_extent_item *fi;
6594         struct btrfs_key key;
6595         u64 disk_bytenr;
6596         u64 backref_offset;
6597         u64 extent_end;
6598         u64 num_bytes;
6599         int slot;
6600         int found_type;
6601         bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6602
6603         path = btrfs_alloc_path();
6604         if (!path)
6605                 return -ENOMEM;
6606
6607         ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
6608                                        offset, 0);
6609         if (ret < 0)
6610                 goto out;
6611
6612         slot = path->slots[0];
6613         if (ret == 1) {
6614                 if (slot == 0) {
6615                         /* can't find the item, must cow */
6616                         ret = 0;
6617                         goto out;
6618                 }
6619                 slot--;
6620         }
6621         ret = 0;
6622         leaf = path->nodes[0];
6623         btrfs_item_key_to_cpu(leaf, &key, slot);
6624         if (key.objectid != btrfs_ino(inode) ||
6625             key.type != BTRFS_EXTENT_DATA_KEY) {
6626                 /* not our file or wrong item type, must cow */
6627                 goto out;
6628         }
6629
6630         if (key.offset > offset) {
6631                 /* Wrong offset, must cow */
6632                 goto out;
6633         }
6634
6635         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6636         found_type = btrfs_file_extent_type(leaf, fi);
6637         if (found_type != BTRFS_FILE_EXTENT_REG &&
6638             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6639                 /* not a regular extent, must cow */
6640                 goto out;
6641         }
6642
6643         if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6644                 goto out;
6645
6646         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6647         if (extent_end <= offset)
6648                 goto out;
6649
6650         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6651         if (disk_bytenr == 0)
6652                 goto out;
6653
6654         if (btrfs_file_extent_compression(leaf, fi) ||
6655             btrfs_file_extent_encryption(leaf, fi) ||
6656             btrfs_file_extent_other_encoding(leaf, fi))
6657                 goto out;
6658
6659         backref_offset = btrfs_file_extent_offset(leaf, fi);
6660
6661         if (orig_start) {
6662                 *orig_start = key.offset - backref_offset;
6663                 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6664                 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6665         }
6666
6667         if (btrfs_extent_readonly(root, disk_bytenr))
6668                 goto out;
6669
6670         num_bytes = min(offset + *len, extent_end) - offset;
6671         if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6672                 u64 range_end;
6673
6674                 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
6675                 ret = test_range_bit(io_tree, offset, range_end,
6676                                      EXTENT_DELALLOC, 0, NULL);
6677                 if (ret) {
6678                         ret = -EAGAIN;
6679                         goto out;
6680                 }
6681         }
6682
6683         btrfs_release_path(path);
6684
6685         /*
6686          * look for other files referencing this extent, if we
6687          * find any we must cow
6688          */
6689         trans = btrfs_join_transaction(root);
6690         if (IS_ERR(trans)) {
6691                 ret = 0;
6692                 goto out;
6693         }
6694
6695         ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6696                                     key.offset - backref_offset, disk_bytenr);
6697         btrfs_end_transaction(trans, root);
6698         if (ret) {
6699                 ret = 0;
6700                 goto out;
6701         }
6702
6703         /*
6704          * adjust disk_bytenr and num_bytes to cover just the bytes
6705          * in this extent we are about to write.  If there
6706          * are any csums in that range we have to cow in order
6707          * to keep the csums correct
6708          */
6709         disk_bytenr += backref_offset;
6710         disk_bytenr += offset - key.offset;
6711         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6712                                 goto out;
6713         /*
6714          * all of the above have passed, it is safe to overwrite this extent
6715          * without cow
6716          */
6717         *len = num_bytes;
6718         ret = 1;
6719 out:
6720         btrfs_free_path(path);
6721         return ret;
6722 }
6723
6724 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
6725 {
6726         struct radix_tree_root *root = &inode->i_mapping->page_tree;
6727         int found = false;
6728         void **pagep = NULL;
6729         struct page *page = NULL;
6730         int start_idx;
6731         int end_idx;
6732
6733         start_idx = start >> PAGE_CACHE_SHIFT;
6734
6735         /*
6736          * end is the last byte in the last page.  end == start is legal
6737          */
6738         end_idx = end >> PAGE_CACHE_SHIFT;
6739
6740         rcu_read_lock();
6741
6742         /* Most of the code in this while loop is lifted from
6743          * find_get_page.  It's been modified to begin searching from a
6744          * page and return just the first page found in that range.  If the
6745          * found idx is less than or equal to the end idx then we know that
6746          * a page exists.  If no pages are found or if those pages are
6747          * outside of the range then we're fine (yay!) */
6748         while (page == NULL &&
6749                radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
6750                 page = radix_tree_deref_slot(pagep);
6751                 if (unlikely(!page))
6752                         break;
6753
6754                 if (radix_tree_exception(page)) {
6755                         if (radix_tree_deref_retry(page)) {
6756                                 page = NULL;
6757                                 continue;
6758                         }
6759                         /*
6760                          * Otherwise, shmem/tmpfs must be storing a swap entry
6761                          * here as an exceptional entry: so return it without
6762                          * attempting to raise page count.
6763                          */
6764                         page = NULL;
6765                         break; /* TODO: Is this relevant for this use case? */
6766                 }
6767
6768                 if (!page_cache_get_speculative(page)) {
6769                         page = NULL;
6770                         continue;
6771                 }
6772
6773                 /*
6774                  * Has the page moved?
6775                  * This is part of the lockless pagecache protocol. See
6776                  * include/linux/pagemap.h for details.
6777                  */
6778                 if (unlikely(page != *pagep)) {
6779                         page_cache_release(page);
6780                         page = NULL;
6781                 }
6782         }
6783
6784         if (page) {
6785                 if (page->index <= end_idx)
6786                         found = true;
6787                 page_cache_release(page);
6788         }
6789
6790         rcu_read_unlock();
6791         return found;
6792 }
6793
6794 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6795                               struct extent_state **cached_state, int writing)
6796 {
6797         struct btrfs_ordered_extent *ordered;
6798         int ret = 0;
6799
6800         while (1) {
6801                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6802                                  0, cached_state);
6803                 /*
6804                  * We're concerned with the entire range that we're going to be
6805                  * doing DIO to, so we need to make sure theres no ordered
6806                  * extents in this range.
6807                  */
6808                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6809                                                      lockend - lockstart + 1);
6810
6811                 /*
6812                  * We need to make sure there are no buffered pages in this
6813                  * range either, we could have raced between the invalidate in
6814                  * generic_file_direct_write and locking the extent.  The
6815                  * invalidate needs to happen so that reads after a write do not
6816                  * get stale data.
6817                  */
6818                 if (!ordered &&
6819                     (!writing ||
6820                      !btrfs_page_exists_in_range(inode, lockstart, lockend)))
6821                         break;
6822
6823                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6824                                      cached_state, GFP_NOFS);
6825
6826                 if (ordered) {
6827                         btrfs_start_ordered_extent(inode, ordered, 1);
6828                         btrfs_put_ordered_extent(ordered);
6829                 } else {
6830                         /* Screw you mmap */
6831                         ret = filemap_write_and_wait_range(inode->i_mapping,
6832                                                            lockstart,
6833                                                            lockend);
6834                         if (ret)
6835                                 break;
6836
6837                         /*
6838                          * If we found a page that couldn't be invalidated just
6839                          * fall back to buffered.
6840                          */
6841                         ret = invalidate_inode_pages2_range(inode->i_mapping,
6842                                         lockstart >> PAGE_CACHE_SHIFT,
6843                                         lockend >> PAGE_CACHE_SHIFT);
6844                         if (ret)
6845                                 break;
6846                 }
6847
6848                 cond_resched();
6849         }
6850
6851         return ret;
6852 }
6853
6854 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6855                                            u64 len, u64 orig_start,
6856                                            u64 block_start, u64 block_len,
6857                                            u64 orig_block_len, u64 ram_bytes,
6858                                            int type)
6859 {
6860         struct extent_map_tree *em_tree;
6861         struct extent_map *em;
6862         struct btrfs_root *root = BTRFS_I(inode)->root;
6863         int ret;
6864
6865         em_tree = &BTRFS_I(inode)->extent_tree;
6866         em = alloc_extent_map();
6867         if (!em)
6868                 return ERR_PTR(-ENOMEM);
6869
6870         em->start = start;
6871         em->orig_start = orig_start;
6872         em->mod_start = start;
6873         em->mod_len = len;
6874         em->len = len;
6875         em->block_len = block_len;
6876         em->block_start = block_start;
6877         em->bdev = root->fs_info->fs_devices->latest_bdev;
6878         em->orig_block_len = orig_block_len;
6879         em->ram_bytes = ram_bytes;
6880         em->generation = -1;
6881         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6882         if (type == BTRFS_ORDERED_PREALLOC)
6883                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6884
6885         do {
6886                 btrfs_drop_extent_cache(inode, em->start,
6887                                 em->start + em->len - 1, 0);
6888                 write_lock(&em_tree->lock);
6889                 ret = add_extent_mapping(em_tree, em, 1);
6890                 write_unlock(&em_tree->lock);
6891         } while (ret == -EEXIST);
6892
6893         if (ret) {
6894                 free_extent_map(em);
6895                 return ERR_PTR(ret);
6896         }
6897
6898         return em;
6899 }
6900
6901
6902 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6903                                    struct buffer_head *bh_result, int create)
6904 {
6905         struct extent_map *em;
6906         struct btrfs_root *root = BTRFS_I(inode)->root;
6907         struct extent_state *cached_state = NULL;
6908         u64 start = iblock << inode->i_blkbits;
6909         u64 lockstart, lockend;
6910         u64 len = bh_result->b_size;
6911         int unlock_bits = EXTENT_LOCKED;
6912         int ret = 0;
6913
6914         if (create)
6915                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6916         else
6917                 len = min_t(u64, len, root->sectorsize);
6918
6919         lockstart = start;
6920         lockend = start + len - 1;
6921
6922         /*
6923          * If this errors out it's because we couldn't invalidate pagecache for
6924          * this range and we need to fallback to buffered.
6925          */
6926         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6927                 return -ENOTBLK;
6928
6929         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6930         if (IS_ERR(em)) {
6931                 ret = PTR_ERR(em);
6932                 goto unlock_err;
6933         }
6934
6935         /*
6936          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6937          * io.  INLINE is special, and we could probably kludge it in here, but
6938          * it's still buffered so for safety lets just fall back to the generic
6939          * buffered path.
6940          *
6941          * For COMPRESSED we _have_ to read the entire extent in so we can
6942          * decompress it, so there will be buffering required no matter what we
6943          * do, so go ahead and fallback to buffered.
6944          *
6945          * We return -ENOTBLK because thats what makes DIO go ahead and go back
6946          * to buffered IO.  Don't blame me, this is the price we pay for using
6947          * the generic code.
6948          */
6949         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6950             em->block_start == EXTENT_MAP_INLINE) {
6951                 free_extent_map(em);
6952                 ret = -ENOTBLK;
6953                 goto unlock_err;
6954         }
6955
6956         /* Just a good old fashioned hole, return */
6957         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6958                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6959                 free_extent_map(em);
6960                 goto unlock_err;
6961         }
6962
6963         /*
6964          * We don't allocate a new extent in the following cases
6965          *
6966          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6967          * existing extent.
6968          * 2) The extent is marked as PREALLOC.  We're good to go here and can
6969          * just use the extent.
6970          *
6971          */
6972         if (!create) {
6973                 len = min(len, em->len - (start - em->start));
6974                 lockstart = start + len;
6975                 goto unlock;
6976         }
6977
6978         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6979             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6980              em->block_start != EXTENT_MAP_HOLE)) {
6981                 int type;
6982                 int ret;
6983                 u64 block_start, orig_start, orig_block_len, ram_bytes;
6984
6985                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6986                         type = BTRFS_ORDERED_PREALLOC;
6987                 else
6988                         type = BTRFS_ORDERED_NOCOW;
6989                 len = min(len, em->len - (start - em->start));
6990                 block_start = em->block_start + (start - em->start);
6991
6992                 if (can_nocow_extent(inode, start, &len, &orig_start,
6993                                      &orig_block_len, &ram_bytes) == 1) {
6994                         if (type == BTRFS_ORDERED_PREALLOC) {
6995                                 free_extent_map(em);
6996                                 em = create_pinned_em(inode, start, len,
6997                                                        orig_start,
6998                                                        block_start, len,
6999                                                        orig_block_len,
7000                                                        ram_bytes, type);
7001                                 if (IS_ERR(em))
7002                                         goto unlock_err;
7003                         }
7004
7005                         ret = btrfs_add_ordered_extent_dio(inode, start,
7006                                            block_start, len, len, type);
7007                         if (ret) {
7008                                 free_extent_map(em);
7009                                 goto unlock_err;
7010                         }
7011                         goto unlock;
7012                 }
7013         }
7014
7015         /*
7016          * this will cow the extent, reset the len in case we changed
7017          * it above
7018          */
7019         len = bh_result->b_size;
7020         free_extent_map(em);
7021         em = btrfs_new_extent_direct(inode, start, len);
7022         if (IS_ERR(em)) {
7023                 ret = PTR_ERR(em);
7024                 goto unlock_err;
7025         }
7026         len = min(len, em->len - (start - em->start));
7027 unlock:
7028         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7029                 inode->i_blkbits;
7030         bh_result->b_size = len;
7031         bh_result->b_bdev = em->bdev;
7032         set_buffer_mapped(bh_result);
7033         if (create) {
7034                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7035                         set_buffer_new(bh_result);
7036
7037                 /*
7038                  * Need to update the i_size under the extent lock so buffered
7039                  * readers will get the updated i_size when we unlock.
7040                  */
7041                 if (start + len > i_size_read(inode))
7042                         i_size_write(inode, start + len);
7043
7044                 spin_lock(&BTRFS_I(inode)->lock);
7045                 BTRFS_I(inode)->outstanding_extents++;
7046                 spin_unlock(&BTRFS_I(inode)->lock);
7047
7048                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7049                                      lockstart + len - 1, EXTENT_DELALLOC, NULL,
7050                                      &cached_state, GFP_NOFS);
7051                 BUG_ON(ret);
7052         }
7053
7054         /*
7055          * In the case of write we need to clear and unlock the entire range,
7056          * in the case of read we need to unlock only the end area that we
7057          * aren't using if there is any left over space.
7058          */
7059         if (lockstart < lockend) {
7060                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7061                                  lockend, unlock_bits, 1, 0,
7062                                  &cached_state, GFP_NOFS);
7063         } else {
7064                 free_extent_state(cached_state);
7065         }
7066
7067         free_extent_map(em);
7068
7069         return 0;
7070
7071 unlock_err:
7072         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7073                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7074         return ret;
7075 }
7076
7077 static void btrfs_endio_direct_read(struct bio *bio, int err)
7078 {
7079         struct btrfs_dio_private *dip = bio->bi_private;
7080         struct bio_vec *bvec;
7081         struct inode *inode = dip->inode;
7082         struct btrfs_root *root = BTRFS_I(inode)->root;
7083         struct bio *dio_bio;
7084         u32 *csums = (u32 *)dip->csum;
7085         u64 start;
7086         int i;
7087
7088         start = dip->logical_offset;
7089         bio_for_each_segment_all(bvec, bio, i) {
7090                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
7091                         struct page *page = bvec->bv_page;
7092                         char *kaddr;
7093                         u32 csum = ~(u32)0;
7094                         unsigned long flags;
7095
7096                         local_irq_save(flags);
7097                         kaddr = kmap_atomic(page);
7098                         csum = btrfs_csum_data(kaddr + bvec->bv_offset,
7099                                                csum, bvec->bv_len);
7100                         btrfs_csum_final(csum, (char *)&csum);
7101                         kunmap_atomic(kaddr);
7102                         local_irq_restore(flags);
7103
7104                         flush_dcache_page(bvec->bv_page);
7105                         if (csum != csums[i]) {
7106                                 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
7107                                           btrfs_ino(inode), start, csum,
7108                                           csums[i]);
7109                                 err = -EIO;
7110                         }
7111                 }
7112
7113                 start += bvec->bv_len;
7114         }
7115
7116         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
7117                       dip->logical_offset + dip->bytes - 1);
7118         dio_bio = dip->dio_bio;
7119
7120         kfree(dip);
7121
7122         /* If we had a csum failure make sure to clear the uptodate flag */
7123         if (err)
7124                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7125         dio_end_io(dio_bio, err);
7126         bio_put(bio);
7127 }
7128
7129 static void btrfs_endio_direct_write(struct bio *bio, int err)
7130 {
7131         struct btrfs_dio_private *dip = bio->bi_private;
7132         struct inode *inode = dip->inode;
7133         struct btrfs_root *root = BTRFS_I(inode)->root;
7134         struct btrfs_ordered_extent *ordered = NULL;
7135         u64 ordered_offset = dip->logical_offset;
7136         u64 ordered_bytes = dip->bytes;
7137         struct bio *dio_bio;
7138         int ret;
7139
7140         if (err)
7141                 goto out_done;
7142 again:
7143         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7144                                                    &ordered_offset,
7145                                                    ordered_bytes, !err);
7146         if (!ret)
7147                 goto out_test;
7148
7149         btrfs_init_work(&ordered->work, finish_ordered_fn, NULL, NULL);
7150         btrfs_queue_work(root->fs_info->endio_write_workers,
7151                          &ordered->work);
7152 out_test:
7153         /*
7154          * our bio might span multiple ordered extents.  If we haven't
7155          * completed the accounting for the whole dio, go back and try again
7156          */
7157         if (ordered_offset < dip->logical_offset + dip->bytes) {
7158                 ordered_bytes = dip->logical_offset + dip->bytes -
7159                         ordered_offset;
7160                 ordered = NULL;
7161                 goto again;
7162         }
7163 out_done:
7164         dio_bio = dip->dio_bio;
7165
7166         kfree(dip);
7167
7168         /* If we had an error make sure to clear the uptodate flag */
7169         if (err)
7170                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7171         dio_end_io(dio_bio, err);
7172         bio_put(bio);
7173 }
7174
7175 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7176                                     struct bio *bio, int mirror_num,
7177                                     unsigned long bio_flags, u64 offset)
7178 {
7179         int ret;
7180         struct btrfs_root *root = BTRFS_I(inode)->root;
7181         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7182         BUG_ON(ret); /* -ENOMEM */
7183         return 0;
7184 }
7185
7186 static void btrfs_end_dio_bio(struct bio *bio, int err)
7187 {
7188         struct btrfs_dio_private *dip = bio->bi_private;
7189
7190         if (err) {
7191                 btrfs_err(BTRFS_I(dip->inode)->root->fs_info,
7192                           "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7193                       btrfs_ino(dip->inode), bio->bi_rw,
7194                       (unsigned long long)bio->bi_iter.bi_sector,
7195                       bio->bi_iter.bi_size, err);
7196                 dip->errors = 1;
7197
7198                 /*
7199                  * before atomic variable goto zero, we must make sure
7200                  * dip->errors is perceived to be set.
7201                  */
7202                 smp_mb__before_atomic();
7203         }
7204
7205         /* if there are more bios still pending for this dio, just exit */
7206         if (!atomic_dec_and_test(&dip->pending_bios))
7207                 goto out;
7208
7209         if (dip->errors) {
7210                 bio_io_error(dip->orig_bio);
7211         } else {
7212                 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7213                 bio_endio(dip->orig_bio, 0);
7214         }
7215 out:
7216         bio_put(bio);
7217 }
7218
7219 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7220                                        u64 first_sector, gfp_t gfp_flags)
7221 {
7222         int nr_vecs = bio_get_nr_vecs(bdev);
7223         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7224 }
7225
7226 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7227                                          int rw, u64 file_offset, int skip_sum,
7228                                          int async_submit)
7229 {
7230         struct btrfs_dio_private *dip = bio->bi_private;
7231         int write = rw & REQ_WRITE;
7232         struct btrfs_root *root = BTRFS_I(inode)->root;
7233         int ret;
7234
7235         if (async_submit)
7236                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7237
7238         bio_get(bio);
7239
7240         if (!write) {
7241                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7242                 if (ret)
7243                         goto err;
7244         }
7245
7246         if (skip_sum)
7247                 goto map;
7248
7249         if (write && async_submit) {
7250                 ret = btrfs_wq_submit_bio(root->fs_info,
7251                                    inode, rw, bio, 0, 0,
7252                                    file_offset,
7253                                    __btrfs_submit_bio_start_direct_io,
7254                                    __btrfs_submit_bio_done);
7255                 goto err;
7256         } else if (write) {
7257                 /*
7258                  * If we aren't doing async submit, calculate the csum of the
7259                  * bio now.
7260                  */
7261                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7262                 if (ret)
7263                         goto err;
7264         } else if (!skip_sum) {
7265                 ret = btrfs_lookup_bio_sums_dio(root, inode, dip, bio,
7266                                                 file_offset);
7267                 if (ret)
7268                         goto err;
7269         }
7270
7271 map:
7272         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7273 err:
7274         bio_put(bio);
7275         return ret;
7276 }
7277
7278 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7279                                     int skip_sum)
7280 {
7281         struct inode *inode = dip->inode;
7282         struct btrfs_root *root = BTRFS_I(inode)->root;
7283         struct bio *bio;
7284         struct bio *orig_bio = dip->orig_bio;
7285         struct bio_vec *bvec = orig_bio->bi_io_vec;
7286         u64 start_sector = orig_bio->bi_iter.bi_sector;
7287         u64 file_offset = dip->logical_offset;
7288         u64 submit_len = 0;
7289         u64 map_length;
7290         int nr_pages = 0;
7291         int ret = 0;
7292         int async_submit = 0;
7293
7294         map_length = orig_bio->bi_iter.bi_size;
7295         ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7296                               &map_length, NULL, 0);
7297         if (ret) {
7298                 bio_put(orig_bio);
7299                 return -EIO;
7300         }
7301
7302         if (map_length >= orig_bio->bi_iter.bi_size) {
7303                 bio = orig_bio;
7304                 goto submit;
7305         }
7306
7307         /* async crcs make it difficult to collect full stripe writes. */
7308         if (btrfs_get_alloc_profile(root, 1) &
7309             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7310                 async_submit = 0;
7311         else
7312                 async_submit = 1;
7313
7314         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7315         if (!bio)
7316                 return -ENOMEM;
7317         bio->bi_private = dip;
7318         bio->bi_end_io = btrfs_end_dio_bio;
7319         atomic_inc(&dip->pending_bios);
7320
7321         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7322                 if (unlikely(map_length < submit_len + bvec->bv_len ||
7323                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7324                                  bvec->bv_offset) < bvec->bv_len)) {
7325                         /*
7326                          * inc the count before we submit the bio so
7327                          * we know the end IO handler won't happen before
7328                          * we inc the count. Otherwise, the dip might get freed
7329                          * before we're done setting it up
7330                          */
7331                         atomic_inc(&dip->pending_bios);
7332                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
7333                                                      file_offset, skip_sum,
7334                                                      async_submit);
7335                         if (ret) {
7336                                 bio_put(bio);
7337                                 atomic_dec(&dip->pending_bios);
7338                                 goto out_err;
7339                         }
7340
7341                         start_sector += submit_len >> 9;
7342                         file_offset += submit_len;
7343
7344                         submit_len = 0;
7345                         nr_pages = 0;
7346
7347                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7348                                                   start_sector, GFP_NOFS);
7349                         if (!bio)
7350                                 goto out_err;
7351                         bio->bi_private = dip;
7352                         bio->bi_end_io = btrfs_end_dio_bio;
7353
7354                         map_length = orig_bio->bi_iter.bi_size;
7355                         ret = btrfs_map_block(root->fs_info, rw,
7356                                               start_sector << 9,
7357                                               &map_length, NULL, 0);
7358                         if (ret) {
7359                                 bio_put(bio);
7360                                 goto out_err;
7361                         }
7362                 } else {
7363                         submit_len += bvec->bv_len;
7364                         nr_pages++;
7365                         bvec++;
7366                 }
7367         }
7368
7369 submit:
7370         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7371                                      async_submit);
7372         if (!ret)
7373                 return 0;
7374
7375         bio_put(bio);
7376 out_err:
7377         dip->errors = 1;
7378         /*
7379          * before atomic variable goto zero, we must
7380          * make sure dip->errors is perceived to be set.
7381          */
7382         smp_mb__before_atomic();
7383         if (atomic_dec_and_test(&dip->pending_bios))
7384                 bio_io_error(dip->orig_bio);
7385
7386         /* bio_end_io() will handle error, so we needn't return it */
7387         return 0;
7388 }
7389
7390 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7391                                 struct inode *inode, loff_t file_offset)
7392 {
7393         struct btrfs_root *root = BTRFS_I(inode)->root;
7394         struct btrfs_dio_private *dip;
7395         struct bio *io_bio;
7396         int skip_sum;
7397         int sum_len;
7398         int write = rw & REQ_WRITE;
7399         int ret = 0;
7400         u16 csum_size;
7401
7402         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7403
7404         io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7405         if (!io_bio) {
7406                 ret = -ENOMEM;
7407                 goto free_ordered;
7408         }
7409
7410         if (!skip_sum && !write) {
7411                 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
7412                 sum_len = dio_bio->bi_iter.bi_size >>
7413                         inode->i_sb->s_blocksize_bits;
7414                 sum_len *= csum_size;
7415         } else {
7416                 sum_len = 0;
7417         }
7418
7419         dip = kmalloc(sizeof(*dip) + sum_len, GFP_NOFS);
7420         if (!dip) {
7421                 ret = -ENOMEM;
7422                 goto free_io_bio;
7423         }
7424
7425         dip->private = dio_bio->bi_private;
7426         dip->inode = inode;
7427         dip->logical_offset = file_offset;
7428         dip->bytes = dio_bio->bi_iter.bi_size;
7429         dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
7430         io_bio->bi_private = dip;
7431         dip->errors = 0;
7432         dip->orig_bio = io_bio;
7433         dip->dio_bio = dio_bio;
7434         atomic_set(&dip->pending_bios, 0);
7435
7436         if (write)
7437                 io_bio->bi_end_io = btrfs_endio_direct_write;
7438         else
7439                 io_bio->bi_end_io = btrfs_endio_direct_read;
7440
7441         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7442         if (!ret)
7443                 return;
7444
7445 free_io_bio:
7446         bio_put(io_bio);
7447
7448 free_ordered:
7449         /*
7450          * If this is a write, we need to clean up the reserved space and kill
7451          * the ordered extent.
7452          */
7453         if (write) {
7454                 struct btrfs_ordered_extent *ordered;
7455                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7456                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7457                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7458                         btrfs_free_reserved_extent(root, ordered->start,
7459                                                    ordered->disk_len, 1);
7460                 btrfs_put_ordered_extent(ordered);
7461                 btrfs_put_ordered_extent(ordered);
7462         }
7463         bio_endio(dio_bio, ret);
7464 }
7465
7466 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7467                         const struct iov_iter *iter, loff_t offset)
7468 {
7469         int seg;
7470         int i;
7471         unsigned blocksize_mask = root->sectorsize - 1;
7472         ssize_t retval = -EINVAL;
7473
7474         if (offset & blocksize_mask)
7475                 goto out;
7476
7477         if (iov_iter_alignment(iter) & blocksize_mask)
7478                 goto out;
7479
7480         /* If this is a write we don't need to check anymore */
7481         if (rw & WRITE)
7482                 return 0;
7483         /*
7484          * Check to make sure we don't have duplicate iov_base's in this
7485          * iovec, if so return EINVAL, otherwise we'll get csum errors
7486          * when reading back.
7487          */
7488         for (seg = 0; seg < iter->nr_segs; seg++) {
7489                 for (i = seg + 1; i < iter->nr_segs; i++) {
7490                         if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
7491                                 goto out;
7492                 }
7493         }
7494         retval = 0;
7495 out:
7496         return retval;
7497 }
7498
7499 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7500                         struct iov_iter *iter, loff_t offset)
7501 {
7502         struct file *file = iocb->ki_filp;
7503         struct inode *inode = file->f_mapping->host;
7504         size_t count = 0;
7505         int flags = 0;
7506         bool wakeup = true;
7507         bool relock = false;
7508         ssize_t ret;
7509
7510         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iter, offset))
7511                 return 0;
7512
7513         atomic_inc(&inode->i_dio_count);
7514         smp_mb__after_atomic();
7515
7516         /*
7517          * The generic stuff only does filemap_write_and_wait_range, which
7518          * isn't enough if we've written compressed pages to this area, so
7519          * we need to flush the dirty pages again to make absolutely sure
7520          * that any outstanding dirty pages are on disk.
7521          */
7522         count = iov_iter_count(iter);
7523         if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7524                      &BTRFS_I(inode)->runtime_flags))
7525                 filemap_fdatawrite_range(inode->i_mapping, offset, count);
7526
7527         if (rw & WRITE) {
7528                 /*
7529                  * If the write DIO is beyond the EOF, we need update
7530                  * the isize, but it is protected by i_mutex. So we can
7531                  * not unlock the i_mutex at this case.
7532                  */
7533                 if (offset + count <= inode->i_size) {
7534                         mutex_unlock(&inode->i_mutex);
7535                         relock = true;
7536                 }
7537                 ret = btrfs_delalloc_reserve_space(inode, count);
7538                 if (ret)
7539                         goto out;
7540         } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7541                                      &BTRFS_I(inode)->runtime_flags))) {
7542                 inode_dio_done(inode);
7543                 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7544                 wakeup = false;
7545         }
7546
7547         ret = __blockdev_direct_IO(rw, iocb, inode,
7548                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7549                         iter, offset, btrfs_get_blocks_direct, NULL,
7550                         btrfs_submit_direct, flags);
7551         if (rw & WRITE) {
7552                 if (ret < 0 && ret != -EIOCBQUEUED)
7553                         btrfs_delalloc_release_space(inode, count);
7554                 else if (ret >= 0 && (size_t)ret < count)
7555                         btrfs_delalloc_release_space(inode,
7556                                                      count - (size_t)ret);
7557                 else
7558                         btrfs_delalloc_release_metadata(inode, 0);
7559         }
7560 out:
7561         if (wakeup)
7562                 inode_dio_done(inode);
7563         if (relock)
7564                 mutex_lock(&inode->i_mutex);
7565
7566         return ret;
7567 }
7568
7569 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
7570
7571 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7572                 __u64 start, __u64 len)
7573 {
7574         int     ret;
7575
7576         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7577         if (ret)
7578                 return ret;
7579
7580         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7581 }
7582
7583 int btrfs_readpage(struct file *file, struct page *page)
7584 {
7585         struct extent_io_tree *tree;
7586         tree = &BTRFS_I(page->mapping->host)->io_tree;
7587         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7588 }
7589
7590 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7591 {
7592         struct extent_io_tree *tree;
7593
7594
7595         if (current->flags & PF_MEMALLOC) {
7596                 redirty_page_for_writepage(wbc, page);
7597                 unlock_page(page);
7598                 return 0;
7599         }
7600         tree = &BTRFS_I(page->mapping->host)->io_tree;
7601         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7602 }
7603
7604 static int btrfs_writepages(struct address_space *mapping,
7605                             struct writeback_control *wbc)
7606 {
7607         struct extent_io_tree *tree;
7608
7609         tree = &BTRFS_I(mapping->host)->io_tree;
7610         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7611 }
7612
7613 static int
7614 btrfs_readpages(struct file *file, struct address_space *mapping,
7615                 struct list_head *pages, unsigned nr_pages)
7616 {
7617         struct extent_io_tree *tree;
7618         tree = &BTRFS_I(mapping->host)->io_tree;
7619         return extent_readpages(tree, mapping, pages, nr_pages,
7620                                 btrfs_get_extent);
7621 }
7622 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7623 {
7624         struct extent_io_tree *tree;
7625         struct extent_map_tree *map;
7626         int ret;
7627
7628         tree = &BTRFS_I(page->mapping->host)->io_tree;
7629         map = &BTRFS_I(page->mapping->host)->extent_tree;
7630         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7631         if (ret == 1) {
7632                 ClearPagePrivate(page);
7633                 set_page_private(page, 0);
7634                 page_cache_release(page);
7635         }
7636         return ret;
7637 }
7638
7639 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7640 {
7641         if (PageWriteback(page) || PageDirty(page))
7642                 return 0;
7643         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7644 }
7645
7646 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
7647                                  unsigned int length)
7648 {
7649         struct inode *inode = page->mapping->host;
7650         struct extent_io_tree *tree;
7651         struct btrfs_ordered_extent *ordered;
7652         struct extent_state *cached_state = NULL;
7653         u64 page_start = page_offset(page);
7654         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7655         int inode_evicting = inode->i_state & I_FREEING;
7656
7657         /*
7658          * we have the page locked, so new writeback can't start,
7659          * and the dirty bit won't be cleared while we are here.
7660          *
7661          * Wait for IO on this page so that we can safely clear
7662          * the PagePrivate2 bit and do ordered accounting
7663          */
7664         wait_on_page_writeback(page);
7665
7666         tree = &BTRFS_I(inode)->io_tree;
7667         if (offset) {
7668                 btrfs_releasepage(page, GFP_NOFS);
7669                 return;
7670         }
7671
7672         if (!inode_evicting)
7673                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7674         ordered = btrfs_lookup_ordered_extent(inode, page_start);
7675         if (ordered) {
7676                 /*
7677                  * IO on this page will never be started, so we need
7678                  * to account for any ordered extents now
7679                  */
7680                 if (!inode_evicting)
7681                         clear_extent_bit(tree, page_start, page_end,
7682                                          EXTENT_DIRTY | EXTENT_DELALLOC |
7683                                          EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7684                                          EXTENT_DEFRAG, 1, 0, &cached_state,
7685                                          GFP_NOFS);
7686                 /*
7687                  * whoever cleared the private bit is responsible
7688                  * for the finish_ordered_io
7689                  */
7690                 if (TestClearPagePrivate2(page)) {
7691                         struct btrfs_ordered_inode_tree *tree;
7692                         u64 new_len;
7693
7694                         tree = &BTRFS_I(inode)->ordered_tree;
7695
7696                         spin_lock_irq(&tree->lock);
7697                         set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
7698                         new_len = page_start - ordered->file_offset;
7699                         if (new_len < ordered->truncated_len)
7700                                 ordered->truncated_len = new_len;
7701                         spin_unlock_irq(&tree->lock);
7702
7703                         if (btrfs_dec_test_ordered_pending(inode, &ordered,
7704                                                            page_start,
7705                                                            PAGE_CACHE_SIZE, 1))
7706                                 btrfs_finish_ordered_io(ordered);
7707                 }
7708                 btrfs_put_ordered_extent(ordered);
7709                 if (!inode_evicting) {
7710                         cached_state = NULL;
7711                         lock_extent_bits(tree, page_start, page_end, 0,
7712                                          &cached_state);
7713                 }
7714         }
7715
7716         if (!inode_evicting) {
7717                 clear_extent_bit(tree, page_start, page_end,
7718                                  EXTENT_LOCKED | EXTENT_DIRTY |
7719                                  EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
7720                                  EXTENT_DEFRAG, 1, 1,
7721                                  &cached_state, GFP_NOFS);
7722
7723                 __btrfs_releasepage(page, GFP_NOFS);
7724         }
7725
7726         ClearPageChecked(page);
7727         if (PagePrivate(page)) {
7728                 ClearPagePrivate(page);
7729                 set_page_private(page, 0);
7730                 page_cache_release(page);
7731         }
7732 }
7733
7734 /*
7735  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7736  * called from a page fault handler when a page is first dirtied. Hence we must
7737  * be careful to check for EOF conditions here. We set the page up correctly
7738  * for a written page which means we get ENOSPC checking when writing into
7739  * holes and correct delalloc and unwritten extent mapping on filesystems that
7740  * support these features.
7741  *
7742  * We are not allowed to take the i_mutex here so we have to play games to
7743  * protect against truncate races as the page could now be beyond EOF.  Because
7744  * vmtruncate() writes the inode size before removing pages, once we have the
7745  * page lock we can determine safely if the page is beyond EOF. If it is not
7746  * beyond EOF, then the page is guaranteed safe against truncation until we
7747  * unlock the page.
7748  */
7749 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7750 {
7751         struct page *page = vmf->page;
7752         struct inode *inode = file_inode(vma->vm_file);
7753         struct btrfs_root *root = BTRFS_I(inode)->root;
7754         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7755         struct btrfs_ordered_extent *ordered;
7756         struct extent_state *cached_state = NULL;
7757         char *kaddr;
7758         unsigned long zero_start;
7759         loff_t size;
7760         int ret;
7761         int reserved = 0;
7762         u64 page_start;
7763         u64 page_end;
7764
7765         sb_start_pagefault(inode->i_sb);
7766         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7767         if (!ret) {
7768                 ret = file_update_time(vma->vm_file);
7769                 reserved = 1;
7770         }
7771         if (ret) {
7772                 if (ret == -ENOMEM)
7773                         ret = VM_FAULT_OOM;
7774                 else /* -ENOSPC, -EIO, etc */
7775                         ret = VM_FAULT_SIGBUS;
7776                 if (reserved)
7777                         goto out;
7778                 goto out_noreserve;
7779         }
7780
7781         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7782 again:
7783         lock_page(page);
7784         size = i_size_read(inode);
7785         page_start = page_offset(page);
7786         page_end = page_start + PAGE_CACHE_SIZE - 1;
7787
7788         if ((page->mapping != inode->i_mapping) ||
7789             (page_start >= size)) {
7790                 /* page got truncated out from underneath us */
7791                 goto out_unlock;
7792         }
7793         wait_on_page_writeback(page);
7794
7795         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7796         set_page_extent_mapped(page);
7797
7798         /*
7799          * we can't set the delalloc bits if there are pending ordered
7800          * extents.  Drop our locks and wait for them to finish
7801          */
7802         ordered = btrfs_lookup_ordered_extent(inode, page_start);
7803         if (ordered) {
7804                 unlock_extent_cached(io_tree, page_start, page_end,
7805                                      &cached_state, GFP_NOFS);
7806                 unlock_page(page);
7807                 btrfs_start_ordered_extent(inode, ordered, 1);
7808                 btrfs_put_ordered_extent(ordered);
7809                 goto again;
7810         }
7811
7812         /*
7813          * XXX - page_mkwrite gets called every time the page is dirtied, even
7814          * if it was already dirty, so for space accounting reasons we need to
7815          * clear any delalloc bits for the range we are fixing to save.  There
7816          * is probably a better way to do this, but for now keep consistent with
7817          * prepare_pages in the normal write path.
7818          */
7819         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7820                           EXTENT_DIRTY | EXTENT_DELALLOC |
7821                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7822                           0, 0, &cached_state, GFP_NOFS);
7823
7824         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7825                                         &cached_state);
7826         if (ret) {
7827                 unlock_extent_cached(io_tree, page_start, page_end,
7828                                      &cached_state, GFP_NOFS);
7829                 ret = VM_FAULT_SIGBUS;
7830                 goto out_unlock;
7831         }
7832         ret = 0;
7833
7834         /* page is wholly or partially inside EOF */
7835         if (page_start + PAGE_CACHE_SIZE > size)
7836                 zero_start = size & ~PAGE_CACHE_MASK;
7837         else
7838                 zero_start = PAGE_CACHE_SIZE;
7839
7840         if (zero_start != PAGE_CACHE_SIZE) {
7841                 kaddr = kmap(page);
7842                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7843                 flush_dcache_page(page);
7844                 kunmap(page);
7845         }
7846         ClearPageChecked(page);
7847         set_page_dirty(page);
7848         SetPageUptodate(page);
7849
7850         BTRFS_I(inode)->last_trans = root->fs_info->generation;
7851         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7852         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7853
7854         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7855
7856 out_unlock:
7857         if (!ret) {
7858                 sb_end_pagefault(inode->i_sb);
7859                 return VM_FAULT_LOCKED;
7860         }
7861         unlock_page(page);
7862 out:
7863         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7864 out_noreserve:
7865         sb_end_pagefault(inode->i_sb);
7866         return ret;
7867 }
7868
7869 static int btrfs_truncate(struct inode *inode)
7870 {
7871         struct btrfs_root *root = BTRFS_I(inode)->root;
7872         struct btrfs_block_rsv *rsv;
7873         int ret = 0;
7874         int err = 0;
7875         struct btrfs_trans_handle *trans;
7876         u64 mask = root->sectorsize - 1;
7877         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7878
7879         ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
7880                                        (u64)-1);
7881         if (ret)
7882                 return ret;
7883
7884         /*
7885          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
7886          * 3 things going on here
7887          *
7888          * 1) We need to reserve space for our orphan item and the space to
7889          * delete our orphan item.  Lord knows we don't want to have a dangling
7890          * orphan item because we didn't reserve space to remove it.
7891          *
7892          * 2) We need to reserve space to update our inode.
7893          *
7894          * 3) We need to have something to cache all the space that is going to
7895          * be free'd up by the truncate operation, but also have some slack
7896          * space reserved in case it uses space during the truncate (thank you
7897          * very much snapshotting).
7898          *
7899          * And we need these to all be seperate.  The fact is we can use alot of
7900          * space doing the truncate, and we have no earthly idea how much space
7901          * we will use, so we need the truncate reservation to be seperate so it
7902          * doesn't end up using space reserved for updating the inode or
7903          * removing the orphan item.  We also need to be able to stop the
7904          * transaction and start a new one, which means we need to be able to
7905          * update the inode several times, and we have no idea of knowing how
7906          * many times that will be, so we can't just reserve 1 item for the
7907          * entirety of the opration, so that has to be done seperately as well.
7908          * Then there is the orphan item, which does indeed need to be held on
7909          * to for the whole operation, and we need nobody to touch this reserved
7910          * space except the orphan code.
7911          *
7912          * So that leaves us with
7913          *
7914          * 1) root->orphan_block_rsv - for the orphan deletion.
7915          * 2) rsv - for the truncate reservation, which we will steal from the
7916          * transaction reservation.
7917          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7918          * updating the inode.
7919          */
7920         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7921         if (!rsv)
7922                 return -ENOMEM;
7923         rsv->size = min_size;
7924         rsv->failfast = 1;
7925
7926         /*
7927          * 1 for the truncate slack space
7928          * 1 for updating the inode.
7929          */
7930         trans = btrfs_start_transaction(root, 2);
7931         if (IS_ERR(trans)) {
7932                 err = PTR_ERR(trans);
7933                 goto out;
7934         }
7935
7936         /* Migrate the slack space for the truncate to our reserve */
7937         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7938                                       min_size);
7939         BUG_ON(ret);
7940
7941         /*
7942          * setattr is responsible for setting the ordered_data_close flag,
7943          * but that is only tested during the last file release.  That
7944          * could happen well after the next commit, leaving a great big
7945          * window where new writes may get lost if someone chooses to write
7946          * to this file after truncating to zero
7947          *
7948          * The inode doesn't have any dirty data here, and so if we commit
7949          * this is a noop.  If someone immediately starts writing to the inode
7950          * it is very likely we'll catch some of their writes in this
7951          * transaction, and the commit will find this file on the ordered
7952          * data list with good things to send down.
7953          *
7954          * This is a best effort solution, there is still a window where
7955          * using truncate to replace the contents of the file will
7956          * end up with a zero length file after a crash.
7957          */
7958         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7959                                            &BTRFS_I(inode)->runtime_flags))
7960                 btrfs_add_ordered_operation(trans, root, inode);
7961
7962         /*
7963          * So if we truncate and then write and fsync we normally would just
7964          * write the extents that changed, which is a problem if we need to
7965          * first truncate that entire inode.  So set this flag so we write out
7966          * all of the extents in the inode to the sync log so we're completely
7967          * safe.
7968          */
7969         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7970         trans->block_rsv = rsv;
7971
7972         while (1) {
7973                 ret = btrfs_truncate_inode_items(trans, root, inode,
7974                                                  inode->i_size,
7975                                                  BTRFS_EXTENT_DATA_KEY);
7976                 if (ret != -ENOSPC) {
7977                         err = ret;
7978                         break;
7979                 }
7980
7981                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7982                 ret = btrfs_update_inode(trans, root, inode);
7983                 if (ret) {
7984                         err = ret;
7985                         break;
7986                 }
7987
7988                 btrfs_end_transaction(trans, root);
7989                 btrfs_btree_balance_dirty(root);
7990
7991                 trans = btrfs_start_transaction(root, 2);
7992                 if (IS_ERR(trans)) {
7993                         ret = err = PTR_ERR(trans);
7994                         trans = NULL;
7995                         break;
7996                 }
7997
7998                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7999                                               rsv, min_size);
8000                 BUG_ON(ret);    /* shouldn't happen */
8001                 trans->block_rsv = rsv;
8002         }
8003
8004         if (ret == 0 && inode->i_nlink > 0) {
8005                 trans->block_rsv = root->orphan_block_rsv;
8006                 ret = btrfs_orphan_del(trans, inode);
8007                 if (ret)
8008                         err = ret;
8009         }
8010
8011         if (trans) {
8012                 trans->block_rsv = &root->fs_info->trans_block_rsv;
8013                 ret = btrfs_update_inode(trans, root, inode);
8014                 if (ret && !err)
8015                         err = ret;
8016
8017                 ret = btrfs_end_transaction(trans, root);
8018                 btrfs_btree_balance_dirty(root);
8019         }
8020
8021 out:
8022         btrfs_free_block_rsv(root, rsv);
8023
8024         if (ret && !err)
8025                 err = ret;
8026
8027         return err;
8028 }
8029
8030 /*
8031  * create a new subvolume directory/inode (helper for the ioctl).
8032  */
8033 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8034                              struct btrfs_root *new_root,
8035                              struct btrfs_root *parent_root,
8036                              u64 new_dirid)
8037 {
8038         struct inode *inode;
8039         int err;
8040         u64 index = 0;
8041
8042         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8043                                 new_dirid, new_dirid,
8044                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
8045                                 &index);
8046         if (IS_ERR(inode))
8047                 return PTR_ERR(inode);
8048         inode->i_op = &btrfs_dir_inode_operations;
8049         inode->i_fop = &btrfs_dir_file_operations;
8050
8051         set_nlink(inode, 1);
8052         btrfs_i_size_write(inode, 0);
8053
8054         err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8055         if (err)
8056                 btrfs_err(new_root->fs_info,
8057                           "error inheriting subvolume %llu properties: %d",
8058                           new_root->root_key.objectid, err);
8059
8060         err = btrfs_update_inode(trans, new_root, inode);
8061
8062         iput(inode);
8063         return err;
8064 }
8065
8066 struct inode *btrfs_alloc_inode(struct super_block *sb)
8067 {
8068         struct btrfs_inode *ei;
8069         struct inode *inode;
8070
8071         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
8072         if (!ei)
8073                 return NULL;
8074
8075         ei->root = NULL;
8076         ei->generation = 0;
8077         ei->last_trans = 0;
8078         ei->last_sub_trans = 0;
8079         ei->logged_trans = 0;
8080         ei->delalloc_bytes = 0;
8081         ei->disk_i_size = 0;
8082         ei->flags = 0;
8083         ei->csum_bytes = 0;
8084         ei->index_cnt = (u64)-1;
8085         ei->dir_index = 0;
8086         ei->last_unlink_trans = 0;
8087         ei->last_log_commit = 0;
8088
8089         spin_lock_init(&ei->lock);
8090         ei->outstanding_extents = 0;
8091         ei->reserved_extents = 0;
8092
8093         ei->runtime_flags = 0;
8094         ei->force_compress = BTRFS_COMPRESS_NONE;
8095
8096         ei->delayed_node = NULL;
8097
8098         inode = &ei->vfs_inode;
8099         extent_map_tree_init(&ei->extent_tree);
8100         extent_io_tree_init(&ei->io_tree, &inode->i_data);
8101         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
8102         ei->io_tree.track_uptodate = 1;
8103         ei->io_failure_tree.track_uptodate = 1;
8104         atomic_set(&ei->sync_writers, 0);
8105         mutex_init(&ei->log_mutex);
8106         mutex_init(&ei->delalloc_mutex);
8107         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8108         INIT_LIST_HEAD(&ei->delalloc_inodes);
8109         INIT_LIST_HEAD(&ei->ordered_operations);
8110         RB_CLEAR_NODE(&ei->rb_node);
8111
8112         return inode;
8113 }
8114
8115 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8116 void btrfs_test_destroy_inode(struct inode *inode)
8117 {
8118         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8119         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8120 }
8121 #endif
8122
8123 static void btrfs_i_callback(struct rcu_head *head)
8124 {
8125         struct inode *inode = container_of(head, struct inode, i_rcu);
8126         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8127 }
8128
8129 void btrfs_destroy_inode(struct inode *inode)
8130 {
8131         struct btrfs_ordered_extent *ordered;
8132         struct btrfs_root *root = BTRFS_I(inode)->root;
8133
8134         WARN_ON(!hlist_empty(&inode->i_dentry));
8135         WARN_ON(inode->i_data.nrpages);
8136         WARN_ON(BTRFS_I(inode)->outstanding_extents);
8137         WARN_ON(BTRFS_I(inode)->reserved_extents);
8138         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8139         WARN_ON(BTRFS_I(inode)->csum_bytes);
8140
8141         /*
8142          * This can happen where we create an inode, but somebody else also
8143          * created the same inode and we need to destroy the one we already
8144          * created.
8145          */
8146         if (!root)
8147                 goto free;
8148
8149         /*
8150          * Make sure we're properly removed from the ordered operation
8151          * lists.
8152          */
8153         smp_mb();
8154         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
8155                 spin_lock(&root->fs_info->ordered_root_lock);
8156                 list_del_init(&BTRFS_I(inode)->ordered_operations);
8157                 spin_unlock(&root->fs_info->ordered_root_lock);
8158         }
8159
8160         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8161                      &BTRFS_I(inode)->runtime_flags)) {
8162                 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
8163                         btrfs_ino(inode));
8164                 atomic_dec(&root->orphan_inodes);
8165         }
8166
8167         while (1) {
8168                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8169                 if (!ordered)
8170                         break;
8171                 else {
8172                         btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
8173                                 ordered->file_offset, ordered->len);
8174                         btrfs_remove_ordered_extent(inode, ordered);
8175                         btrfs_put_ordered_extent(ordered);
8176                         btrfs_put_ordered_extent(ordered);
8177                 }
8178         }
8179         inode_tree_del(inode);
8180         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8181 free:
8182         call_rcu(&inode->i_rcu, btrfs_i_callback);
8183 }
8184
8185 int btrfs_drop_inode(struct inode *inode)
8186 {
8187         struct btrfs_root *root = BTRFS_I(inode)->root;
8188
8189         if (root == NULL)
8190                 return 1;
8191
8192         /* the snap/subvol tree is on deleting */
8193         if (btrfs_root_refs(&root->root_item) == 0)
8194                 return 1;
8195         else
8196                 return generic_drop_inode(inode);
8197 }
8198
8199 static void init_once(void *foo)
8200 {
8201         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8202
8203         inode_init_once(&ei->vfs_inode);
8204 }
8205
8206 void btrfs_destroy_cachep(void)
8207 {
8208         /*
8209          * Make sure all delayed rcu free inodes are flushed before we
8210          * destroy cache.
8211          */
8212         rcu_barrier();
8213         if (btrfs_inode_cachep)
8214                 kmem_cache_destroy(btrfs_inode_cachep);
8215         if (btrfs_trans_handle_cachep)
8216                 kmem_cache_destroy(btrfs_trans_handle_cachep);
8217         if (btrfs_transaction_cachep)
8218                 kmem_cache_destroy(btrfs_transaction_cachep);
8219         if (btrfs_path_cachep)
8220                 kmem_cache_destroy(btrfs_path_cachep);
8221         if (btrfs_free_space_cachep)
8222                 kmem_cache_destroy(btrfs_free_space_cachep);
8223         if (btrfs_delalloc_work_cachep)
8224                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8225 }
8226
8227 int btrfs_init_cachep(void)
8228 {
8229         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8230                         sizeof(struct btrfs_inode), 0,
8231                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8232         if (!btrfs_inode_cachep)
8233                 goto fail;
8234
8235         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8236                         sizeof(struct btrfs_trans_handle), 0,
8237                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8238         if (!btrfs_trans_handle_cachep)
8239                 goto fail;
8240
8241         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8242                         sizeof(struct btrfs_transaction), 0,
8243                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8244         if (!btrfs_transaction_cachep)
8245                 goto fail;
8246
8247         btrfs_path_cachep = kmem_cache_create("btrfs_path",
8248                         sizeof(struct btrfs_path), 0,
8249                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8250         if (!btrfs_path_cachep)
8251                 goto fail;
8252
8253         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8254                         sizeof(struct btrfs_free_space), 0,
8255                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8256         if (!btrfs_free_space_cachep)
8257                 goto fail;
8258
8259         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8260                         sizeof(struct btrfs_delalloc_work), 0,
8261                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8262                         NULL);
8263         if (!btrfs_delalloc_work_cachep)
8264                 goto fail;
8265
8266         return 0;
8267 fail:
8268         btrfs_destroy_cachep();
8269         return -ENOMEM;
8270 }
8271
8272 static int btrfs_getattr(struct vfsmount *mnt,
8273                          struct dentry *dentry, struct kstat *stat)
8274 {
8275         u64 delalloc_bytes;
8276         struct inode *inode = dentry->d_inode;
8277         u32 blocksize = inode->i_sb->s_blocksize;
8278
8279         generic_fillattr(inode, stat);
8280         stat->dev = BTRFS_I(inode)->root->anon_dev;
8281         stat->blksize = PAGE_CACHE_SIZE;
8282
8283         spin_lock(&BTRFS_I(inode)->lock);
8284         delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8285         spin_unlock(&BTRFS_I(inode)->lock);
8286         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8287                         ALIGN(delalloc_bytes, blocksize)) >> 9;
8288         return 0;
8289 }
8290
8291 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8292                            struct inode *new_dir, struct dentry *new_dentry)
8293 {
8294         struct btrfs_trans_handle *trans;
8295         struct btrfs_root *root = BTRFS_I(old_dir)->root;
8296         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8297         struct inode *new_inode = new_dentry->d_inode;
8298         struct inode *old_inode = old_dentry->d_inode;
8299         struct timespec ctime = CURRENT_TIME;
8300         u64 index = 0;
8301         u64 root_objectid;
8302         int ret;
8303         u64 old_ino = btrfs_ino(old_inode);
8304
8305         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8306                 return -EPERM;
8307
8308         /* we only allow rename subvolume link between subvolumes */
8309         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8310                 return -EXDEV;
8311
8312         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8313             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8314                 return -ENOTEMPTY;
8315
8316         if (S_ISDIR(old_inode->i_mode) && new_inode &&
8317             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8318                 return -ENOTEMPTY;
8319
8320
8321         /* check for collisions, even if the  name isn't there */
8322         ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
8323                              new_dentry->d_name.name,
8324                              new_dentry->d_name.len);
8325
8326         if (ret) {
8327                 if (ret == -EEXIST) {
8328                         /* we shouldn't get
8329                          * eexist without a new_inode */
8330                         if (WARN_ON(!new_inode)) {
8331                                 return ret;
8332                         }
8333                 } else {
8334                         /* maybe -EOVERFLOW */
8335                         return ret;
8336                 }
8337         }
8338         ret = 0;
8339
8340         /*
8341          * we're using rename to replace one file with another.
8342          * and the replacement file is large.  Start IO on it now so
8343          * we don't add too much work to the end of the transaction
8344          */
8345         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8346             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8347                 filemap_flush(old_inode->i_mapping);
8348
8349         /* close the racy window with snapshot create/destroy ioctl */
8350         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8351                 down_read(&root->fs_info->subvol_sem);
8352         /*
8353          * We want to reserve the absolute worst case amount of items.  So if
8354          * both inodes are subvols and we need to unlink them then that would
8355          * require 4 item modifications, but if they are both normal inodes it
8356          * would require 5 item modifications, so we'll assume their normal
8357          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8358          * should cover the worst case number of items we'll modify.
8359          */
8360         trans = btrfs_start_transaction(root, 11);
8361         if (IS_ERR(trans)) {
8362                 ret = PTR_ERR(trans);
8363                 goto out_notrans;
8364         }
8365
8366         if (dest != root)
8367                 btrfs_record_root_in_trans(trans, dest);
8368
8369         ret = btrfs_set_inode_index(new_dir, &index);
8370         if (ret)
8371                 goto out_fail;
8372
8373         BTRFS_I(old_inode)->dir_index = 0ULL;
8374         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8375                 /* force full log commit if subvolume involved. */
8376                 btrfs_set_log_full_commit(root->fs_info, trans);
8377         } else {
8378                 ret = btrfs_insert_inode_ref(trans, dest,
8379                                              new_dentry->d_name.name,
8380                                              new_dentry->d_name.len,
8381                                              old_ino,
8382                                              btrfs_ino(new_dir), index);
8383                 if (ret)
8384                         goto out_fail;
8385                 /*
8386                  * this is an ugly little race, but the rename is required
8387                  * to make sure that if we crash, the inode is either at the
8388                  * old name or the new one.  pinning the log transaction lets
8389                  * us make sure we don't allow a log commit to come in after
8390                  * we unlink the name but before we add the new name back in.
8391                  */
8392                 btrfs_pin_log_trans(root);
8393         }
8394         /*
8395          * make sure the inode gets flushed if it is replacing
8396          * something.
8397          */
8398         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8399                 btrfs_add_ordered_operation(trans, root, old_inode);
8400
8401         inode_inc_iversion(old_dir);
8402         inode_inc_iversion(new_dir);
8403         inode_inc_iversion(old_inode);
8404         old_dir->i_ctime = old_dir->i_mtime = ctime;
8405         new_dir->i_ctime = new_dir->i_mtime = ctime;
8406         old_inode->i_ctime = ctime;
8407
8408         if (old_dentry->d_parent != new_dentry->d_parent)
8409                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8410
8411         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8412                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8413                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8414                                         old_dentry->d_name.name,
8415                                         old_dentry->d_name.len);
8416         } else {
8417                 ret = __btrfs_unlink_inode(trans, root, old_dir,
8418                                         old_dentry->d_inode,
8419                                         old_dentry->d_name.name,
8420                                         old_dentry->d_name.len);
8421                 if (!ret)
8422                         ret = btrfs_update_inode(trans, root, old_inode);
8423         }
8424         if (ret) {
8425                 btrfs_abort_transaction(trans, root, ret);
8426                 goto out_fail;
8427         }
8428
8429         if (new_inode) {
8430                 inode_inc_iversion(new_inode);
8431                 new_inode->i_ctime = CURRENT_TIME;
8432                 if (unlikely(btrfs_ino(new_inode) ==
8433                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8434                         root_objectid = BTRFS_I(new_inode)->location.objectid;
8435                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
8436                                                 root_objectid,
8437                                                 new_dentry->d_name.name,
8438                                                 new_dentry->d_name.len);
8439                         BUG_ON(new_inode->i_nlink == 0);
8440                 } else {
8441                         ret = btrfs_unlink_inode(trans, dest, new_dir,
8442                                                  new_dentry->d_inode,
8443                                                  new_dentry->d_name.name,
8444                                                  new_dentry->d_name.len);
8445                 }
8446                 if (!ret && new_inode->i_nlink == 0)
8447                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8448                 if (ret) {
8449                         btrfs_abort_transaction(trans, root, ret);
8450                         goto out_fail;
8451                 }
8452         }
8453
8454         ret = btrfs_add_link(trans, new_dir, old_inode,
8455                              new_dentry->d_name.name,
8456                              new_dentry->d_name.len, 0, index);
8457         if (ret) {
8458                 btrfs_abort_transaction(trans, root, ret);
8459                 goto out_fail;
8460         }
8461
8462         if (old_inode->i_nlink == 1)
8463                 BTRFS_I(old_inode)->dir_index = index;
8464
8465         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8466                 struct dentry *parent = new_dentry->d_parent;
8467                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8468                 btrfs_end_log_trans(root);
8469         }
8470 out_fail:
8471         btrfs_end_transaction(trans, root);
8472 out_notrans:
8473         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8474                 up_read(&root->fs_info->subvol_sem);
8475
8476         return ret;
8477 }
8478
8479 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8480 {
8481         struct btrfs_delalloc_work *delalloc_work;
8482         struct inode *inode;
8483
8484         delalloc_work = container_of(work, struct btrfs_delalloc_work,
8485                                      work);
8486         inode = delalloc_work->inode;
8487         if (delalloc_work->wait) {
8488                 btrfs_wait_ordered_range(inode, 0, (u64)-1);
8489         } else {
8490                 filemap_flush(inode->i_mapping);
8491                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8492                              &BTRFS_I(inode)->runtime_flags))
8493                         filemap_flush(inode->i_mapping);
8494         }
8495
8496         if (delalloc_work->delay_iput)
8497                 btrfs_add_delayed_iput(inode);
8498         else
8499                 iput(inode);
8500         complete(&delalloc_work->completion);
8501 }
8502
8503 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8504                                                     int wait, int delay_iput)
8505 {
8506         struct btrfs_delalloc_work *work;
8507
8508         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8509         if (!work)
8510                 return NULL;
8511
8512         init_completion(&work->completion);
8513         INIT_LIST_HEAD(&work->list);
8514         work->inode = inode;
8515         work->wait = wait;
8516         work->delay_iput = delay_iput;
8517         btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
8518
8519         return work;
8520 }
8521
8522 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8523 {
8524         wait_for_completion(&work->completion);
8525         kmem_cache_free(btrfs_delalloc_work_cachep, work);
8526 }
8527
8528 /*
8529  * some fairly slow code that needs optimization. This walks the list
8530  * of all the inodes with pending delalloc and forces them to disk.
8531  */
8532 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
8533                                    int nr)
8534 {
8535         struct btrfs_inode *binode;
8536         struct inode *inode;
8537         struct btrfs_delalloc_work *work, *next;
8538         struct list_head works;
8539         struct list_head splice;
8540         int ret = 0;
8541
8542         INIT_LIST_HEAD(&works);
8543         INIT_LIST_HEAD(&splice);
8544
8545         mutex_lock(&root->delalloc_mutex);
8546         spin_lock(&root->delalloc_lock);
8547         list_splice_init(&root->delalloc_inodes, &splice);
8548         while (!list_empty(&splice)) {
8549                 binode = list_entry(splice.next, struct btrfs_inode,
8550                                     delalloc_inodes);
8551
8552                 list_move_tail(&binode->delalloc_inodes,
8553                                &root->delalloc_inodes);
8554                 inode = igrab(&binode->vfs_inode);
8555                 if (!inode) {
8556                         cond_resched_lock(&root->delalloc_lock);
8557                         continue;
8558                 }
8559                 spin_unlock(&root->delalloc_lock);
8560
8561                 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8562                 if (unlikely(!work)) {
8563                         if (delay_iput)
8564                                 btrfs_add_delayed_iput(inode);
8565                         else
8566                                 iput(inode);
8567                         ret = -ENOMEM;
8568                         goto out;
8569                 }
8570                 list_add_tail(&work->list, &works);
8571                 btrfs_queue_work(root->fs_info->flush_workers,
8572                                  &work->work);
8573                 ret++;
8574                 if (nr != -1 && ret >= nr)
8575                         goto out;
8576                 cond_resched();
8577                 spin_lock(&root->delalloc_lock);
8578         }
8579         spin_unlock(&root->delalloc_lock);
8580
8581 out:
8582         list_for_each_entry_safe(work, next, &works, list) {
8583                 list_del_init(&work->list);
8584                 btrfs_wait_and_free_delalloc_work(work);
8585         }
8586
8587         if (!list_empty_careful(&splice)) {
8588                 spin_lock(&root->delalloc_lock);
8589                 list_splice_tail(&splice, &root->delalloc_inodes);
8590                 spin_unlock(&root->delalloc_lock);
8591         }
8592         mutex_unlock(&root->delalloc_mutex);
8593         return ret;
8594 }
8595
8596 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8597 {
8598         int ret;
8599
8600         if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
8601                 return -EROFS;
8602
8603         ret = __start_delalloc_inodes(root, delay_iput, -1);
8604         if (ret > 0)
8605                 ret = 0;
8606         /*
8607          * the filemap_flush will queue IO into the worker threads, but
8608          * we have to make sure the IO is actually started and that
8609          * ordered extents get created before we return
8610          */
8611         atomic_inc(&root->fs_info->async_submit_draining);
8612         while (atomic_read(&root->fs_info->nr_async_submits) ||
8613               atomic_read(&root->fs_info->async_delalloc_pages)) {
8614                 wait_event(root->fs_info->async_submit_wait,
8615                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8616                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8617         }
8618         atomic_dec(&root->fs_info->async_submit_draining);
8619         return ret;
8620 }
8621
8622 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
8623                                int nr)
8624 {
8625         struct btrfs_root *root;
8626         struct list_head splice;
8627         int ret;
8628
8629         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
8630                 return -EROFS;
8631
8632         INIT_LIST_HEAD(&splice);
8633
8634         mutex_lock(&fs_info->delalloc_root_mutex);
8635         spin_lock(&fs_info->delalloc_root_lock);
8636         list_splice_init(&fs_info->delalloc_roots, &splice);
8637         while (!list_empty(&splice) && nr) {
8638                 root = list_first_entry(&splice, struct btrfs_root,
8639                                         delalloc_root);
8640                 root = btrfs_grab_fs_root(root);
8641                 BUG_ON(!root);
8642                 list_move_tail(&root->delalloc_root,
8643                                &fs_info->delalloc_roots);
8644                 spin_unlock(&fs_info->delalloc_root_lock);
8645
8646                 ret = __start_delalloc_inodes(root, delay_iput, nr);
8647                 btrfs_put_fs_root(root);
8648                 if (ret < 0)
8649                         goto out;
8650
8651                 if (nr != -1) {
8652                         nr -= ret;
8653                         WARN_ON(nr < 0);
8654                 }
8655                 spin_lock(&fs_info->delalloc_root_lock);
8656         }
8657         spin_unlock(&fs_info->delalloc_root_lock);
8658
8659         ret = 0;
8660         atomic_inc(&fs_info->async_submit_draining);
8661         while (atomic_read(&fs_info->nr_async_submits) ||
8662               atomic_read(&fs_info->async_delalloc_pages)) {
8663                 wait_event(fs_info->async_submit_wait,
8664                    (atomic_read(&fs_info->nr_async_submits) == 0 &&
8665                     atomic_read(&fs_info->async_delalloc_pages) == 0));
8666         }
8667         atomic_dec(&fs_info->async_submit_draining);
8668 out:
8669         if (!list_empty_careful(&splice)) {
8670                 spin_lock(&fs_info->delalloc_root_lock);
8671                 list_splice_tail(&splice, &fs_info->delalloc_roots);
8672                 spin_unlock(&fs_info->delalloc_root_lock);
8673         }
8674         mutex_unlock(&fs_info->delalloc_root_mutex);
8675         return ret;
8676 }
8677
8678 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8679                          const char *symname)
8680 {
8681         struct btrfs_trans_handle *trans;
8682         struct btrfs_root *root = BTRFS_I(dir)->root;
8683         struct btrfs_path *path;
8684         struct btrfs_key key;
8685         struct inode *inode = NULL;
8686         int err;
8687         int drop_inode = 0;
8688         u64 objectid;
8689         u64 index = 0;
8690         int name_len;
8691         int datasize;
8692         unsigned long ptr;
8693         struct btrfs_file_extent_item *ei;
8694         struct extent_buffer *leaf;
8695
8696         name_len = strlen(symname);
8697         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8698                 return -ENAMETOOLONG;
8699
8700         /*
8701          * 2 items for inode item and ref
8702          * 2 items for dir items
8703          * 1 item for xattr if selinux is on
8704          */
8705         trans = btrfs_start_transaction(root, 5);
8706         if (IS_ERR(trans))
8707                 return PTR_ERR(trans);
8708
8709         err = btrfs_find_free_ino(root, &objectid);
8710         if (err)
8711                 goto out_unlock;
8712
8713         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8714                                 dentry->d_name.len, btrfs_ino(dir), objectid,
8715                                 S_IFLNK|S_IRWXUGO, &index);
8716         if (IS_ERR(inode)) {
8717                 err = PTR_ERR(inode);
8718                 goto out_unlock;
8719         }
8720
8721         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8722         if (err) {
8723                 drop_inode = 1;
8724                 goto out_unlock;
8725         }
8726
8727         /*
8728         * If the active LSM wants to access the inode during
8729         * d_instantiate it needs these. Smack checks to see
8730         * if the filesystem supports xattrs by looking at the
8731         * ops vector.
8732         */
8733         inode->i_fop = &btrfs_file_operations;
8734         inode->i_op = &btrfs_file_inode_operations;
8735
8736         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8737         if (err)
8738                 drop_inode = 1;
8739         else {
8740                 inode->i_mapping->a_ops = &btrfs_aops;
8741                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8742                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8743         }
8744         if (drop_inode)
8745                 goto out_unlock;
8746
8747         path = btrfs_alloc_path();
8748         if (!path) {
8749                 err = -ENOMEM;
8750                 drop_inode = 1;
8751                 goto out_unlock;
8752         }
8753         key.objectid = btrfs_ino(inode);
8754         key.offset = 0;
8755         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8756         datasize = btrfs_file_extent_calc_inline_size(name_len);
8757         err = btrfs_insert_empty_item(trans, root, path, &key,
8758                                       datasize);
8759         if (err) {
8760                 drop_inode = 1;
8761                 btrfs_free_path(path);
8762                 goto out_unlock;
8763         }
8764         leaf = path->nodes[0];
8765         ei = btrfs_item_ptr(leaf, path->slots[0],
8766                             struct btrfs_file_extent_item);
8767         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8768         btrfs_set_file_extent_type(leaf, ei,
8769                                    BTRFS_FILE_EXTENT_INLINE);
8770         btrfs_set_file_extent_encryption(leaf, ei, 0);
8771         btrfs_set_file_extent_compression(leaf, ei, 0);
8772         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8773         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8774
8775         ptr = btrfs_file_extent_inline_start(ei);
8776         write_extent_buffer(leaf, symname, ptr, name_len);
8777         btrfs_mark_buffer_dirty(leaf);
8778         btrfs_free_path(path);
8779
8780         inode->i_op = &btrfs_symlink_inode_operations;
8781         inode->i_mapping->a_ops = &btrfs_symlink_aops;
8782         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8783         inode_set_bytes(inode, name_len);
8784         btrfs_i_size_write(inode, name_len);
8785         err = btrfs_update_inode(trans, root, inode);
8786         if (err)
8787                 drop_inode = 1;
8788
8789 out_unlock:
8790         if (!err)
8791                 d_instantiate(dentry, inode);
8792         btrfs_end_transaction(trans, root);
8793         if (drop_inode) {
8794                 inode_dec_link_count(inode);
8795                 iput(inode);
8796         }
8797         btrfs_btree_balance_dirty(root);
8798         return err;
8799 }
8800
8801 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8802                                        u64 start, u64 num_bytes, u64 min_size,
8803                                        loff_t actual_len, u64 *alloc_hint,
8804                                        struct btrfs_trans_handle *trans)
8805 {
8806         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8807         struct extent_map *em;
8808         struct btrfs_root *root = BTRFS_I(inode)->root;
8809         struct btrfs_key ins;
8810         u64 cur_offset = start;
8811         u64 i_size;
8812         u64 cur_bytes;
8813         int ret = 0;
8814         bool own_trans = true;
8815
8816         if (trans)
8817                 own_trans = false;
8818         while (num_bytes > 0) {
8819                 if (own_trans) {
8820                         trans = btrfs_start_transaction(root, 3);
8821                         if (IS_ERR(trans)) {
8822                                 ret = PTR_ERR(trans);
8823                                 break;
8824                         }
8825                 }
8826
8827                 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8828                 cur_bytes = max(cur_bytes, min_size);
8829                 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
8830                                            *alloc_hint, &ins, 1, 0);
8831                 if (ret) {
8832                         if (own_trans)
8833                                 btrfs_end_transaction(trans, root);
8834                         break;
8835                 }
8836
8837                 ret = insert_reserved_file_extent(trans, inode,
8838                                                   cur_offset, ins.objectid,
8839                                                   ins.offset, ins.offset,
8840                                                   ins.offset, 0, 0, 0,
8841                                                   BTRFS_FILE_EXTENT_PREALLOC);
8842                 if (ret) {
8843                         btrfs_free_reserved_extent(root, ins.objectid,
8844                                                    ins.offset, 0);
8845                         btrfs_abort_transaction(trans, root, ret);
8846                         if (own_trans)
8847                                 btrfs_end_transaction(trans, root);
8848                         break;
8849                 }
8850                 btrfs_drop_extent_cache(inode, cur_offset,
8851                                         cur_offset + ins.offset -1, 0);
8852
8853                 em = alloc_extent_map();
8854                 if (!em) {
8855                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8856                                 &BTRFS_I(inode)->runtime_flags);
8857                         goto next;
8858                 }
8859
8860                 em->start = cur_offset;
8861                 em->orig_start = cur_offset;
8862                 em->len = ins.offset;
8863                 em->block_start = ins.objectid;
8864                 em->block_len = ins.offset;
8865                 em->orig_block_len = ins.offset;
8866                 em->ram_bytes = ins.offset;
8867                 em->bdev = root->fs_info->fs_devices->latest_bdev;
8868                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8869                 em->generation = trans->transid;
8870
8871                 while (1) {
8872                         write_lock(&em_tree->lock);
8873                         ret = add_extent_mapping(em_tree, em, 1);
8874                         write_unlock(&em_tree->lock);
8875                         if (ret != -EEXIST)
8876                                 break;
8877                         btrfs_drop_extent_cache(inode, cur_offset,
8878                                                 cur_offset + ins.offset - 1,
8879                                                 0);
8880                 }
8881                 free_extent_map(em);
8882 next:
8883                 num_bytes -= ins.offset;
8884                 cur_offset += ins.offset;
8885                 *alloc_hint = ins.objectid + ins.offset;
8886
8887                 inode_inc_iversion(inode);
8888                 inode->i_ctime = CURRENT_TIME;
8889                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8890                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8891                     (actual_len > inode->i_size) &&
8892                     (cur_offset > inode->i_size)) {
8893                         if (cur_offset > actual_len)
8894                                 i_size = actual_len;
8895                         else
8896                                 i_size = cur_offset;
8897                         i_size_write(inode, i_size);
8898                         btrfs_ordered_update_i_size(inode, i_size, NULL);
8899                 }
8900
8901                 ret = btrfs_update_inode(trans, root, inode);
8902
8903                 if (ret) {
8904                         btrfs_abort_transaction(trans, root, ret);
8905                         if (own_trans)
8906                                 btrfs_end_transaction(trans, root);
8907                         break;
8908                 }
8909
8910                 if (own_trans)
8911                         btrfs_end_transaction(trans, root);
8912         }
8913         return ret;
8914 }
8915
8916 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8917                               u64 start, u64 num_bytes, u64 min_size,
8918                               loff_t actual_len, u64 *alloc_hint)
8919 {
8920         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8921                                            min_size, actual_len, alloc_hint,
8922                                            NULL);
8923 }
8924
8925 int btrfs_prealloc_file_range_trans(struct inode *inode,
8926                                     struct btrfs_trans_handle *trans, int mode,
8927                                     u64 start, u64 num_bytes, u64 min_size,
8928                                     loff_t actual_len, u64 *alloc_hint)
8929 {
8930         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8931                                            min_size, actual_len, alloc_hint, trans);
8932 }
8933
8934 static int btrfs_set_page_dirty(struct page *page)
8935 {
8936         return __set_page_dirty_nobuffers(page);
8937 }
8938
8939 static int btrfs_permission(struct inode *inode, int mask)
8940 {
8941         struct btrfs_root *root = BTRFS_I(inode)->root;
8942         umode_t mode = inode->i_mode;
8943
8944         if (mask & MAY_WRITE &&
8945             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8946                 if (btrfs_root_readonly(root))
8947                         return -EROFS;
8948                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8949                         return -EACCES;
8950         }
8951         return generic_permission(inode, mask);
8952 }
8953
8954 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
8955 {
8956         struct btrfs_trans_handle *trans;
8957         struct btrfs_root *root = BTRFS_I(dir)->root;
8958         struct inode *inode = NULL;
8959         u64 objectid;
8960         u64 index;
8961         int ret = 0;
8962
8963         /*
8964          * 5 units required for adding orphan entry
8965          */
8966         trans = btrfs_start_transaction(root, 5);
8967         if (IS_ERR(trans))
8968                 return PTR_ERR(trans);
8969
8970         ret = btrfs_find_free_ino(root, &objectid);
8971         if (ret)
8972                 goto out;
8973
8974         inode = btrfs_new_inode(trans, root, dir, NULL, 0,
8975                                 btrfs_ino(dir), objectid, mode, &index);
8976         if (IS_ERR(inode)) {
8977                 ret = PTR_ERR(inode);
8978                 inode = NULL;
8979                 goto out;
8980         }
8981
8982         ret = btrfs_init_inode_security(trans, inode, dir, NULL);
8983         if (ret)
8984                 goto out;
8985
8986         ret = btrfs_update_inode(trans, root, inode);
8987         if (ret)
8988                 goto out;
8989
8990         inode->i_fop = &btrfs_file_operations;
8991         inode->i_op = &btrfs_file_inode_operations;
8992
8993         inode->i_mapping->a_ops = &btrfs_aops;
8994         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8995         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8996
8997         ret = btrfs_orphan_add(trans, inode);
8998         if (ret)
8999                 goto out;
9000
9001         d_tmpfile(dentry, inode);
9002         mark_inode_dirty(inode);
9003
9004 out:
9005         btrfs_end_transaction(trans, root);
9006         if (ret)
9007                 iput(inode);
9008         btrfs_balance_delayed_items(root);
9009         btrfs_btree_balance_dirty(root);
9010
9011         return ret;
9012 }
9013
9014 static const struct inode_operations btrfs_dir_inode_operations = {
9015         .getattr        = btrfs_getattr,
9016         .lookup         = btrfs_lookup,
9017         .create         = btrfs_create,
9018         .unlink         = btrfs_unlink,
9019         .link           = btrfs_link,
9020         .mkdir          = btrfs_mkdir,
9021         .rmdir          = btrfs_rmdir,
9022         .rename         = btrfs_rename,
9023         .symlink        = btrfs_symlink,
9024         .setattr        = btrfs_setattr,
9025         .mknod          = btrfs_mknod,
9026         .setxattr       = btrfs_setxattr,
9027         .getxattr       = btrfs_getxattr,
9028         .listxattr      = btrfs_listxattr,
9029         .removexattr    = btrfs_removexattr,
9030         .permission     = btrfs_permission,
9031         .get_acl        = btrfs_get_acl,
9032         .set_acl        = btrfs_set_acl,
9033         .update_time    = btrfs_update_time,
9034         .tmpfile        = btrfs_tmpfile,
9035 };
9036 static const struct inode_operations btrfs_dir_ro_inode_operations = {
9037         .lookup         = btrfs_lookup,
9038         .permission     = btrfs_permission,
9039         .get_acl        = btrfs_get_acl,
9040         .set_acl        = btrfs_set_acl,
9041         .update_time    = btrfs_update_time,
9042 };
9043
9044 static const struct file_operations btrfs_dir_file_operations = {
9045         .llseek         = generic_file_llseek,
9046         .read           = generic_read_dir,
9047         .iterate        = btrfs_real_readdir,
9048         .unlocked_ioctl = btrfs_ioctl,
9049 #ifdef CONFIG_COMPAT
9050         .compat_ioctl   = btrfs_ioctl,
9051 #endif
9052         .release        = btrfs_release_file,
9053         .fsync          = btrfs_sync_file,
9054 };
9055
9056 static struct extent_io_ops btrfs_extent_io_ops = {
9057         .fill_delalloc = run_delalloc_range,
9058         .submit_bio_hook = btrfs_submit_bio_hook,
9059         .merge_bio_hook = btrfs_merge_bio_hook,
9060         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
9061         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
9062         .writepage_start_hook = btrfs_writepage_start_hook,
9063         .set_bit_hook = btrfs_set_bit_hook,
9064         .clear_bit_hook = btrfs_clear_bit_hook,
9065         .merge_extent_hook = btrfs_merge_extent_hook,
9066         .split_extent_hook = btrfs_split_extent_hook,
9067 };
9068
9069 /*
9070  * btrfs doesn't support the bmap operation because swapfiles
9071  * use bmap to make a mapping of extents in the file.  They assume
9072  * these extents won't change over the life of the file and they
9073  * use the bmap result to do IO directly to the drive.
9074  *
9075  * the btrfs bmap call would return logical addresses that aren't
9076  * suitable for IO and they also will change frequently as COW
9077  * operations happen.  So, swapfile + btrfs == corruption.
9078  *
9079  * For now we're avoiding this by dropping bmap.
9080  */
9081 static const struct address_space_operations btrfs_aops = {
9082         .readpage       = btrfs_readpage,
9083         .writepage      = btrfs_writepage,
9084         .writepages     = btrfs_writepages,
9085         .readpages      = btrfs_readpages,
9086         .direct_IO      = btrfs_direct_IO,
9087         .invalidatepage = btrfs_invalidatepage,
9088         .releasepage    = btrfs_releasepage,
9089         .set_page_dirty = btrfs_set_page_dirty,
9090         .error_remove_page = generic_error_remove_page,
9091 };
9092
9093 static const struct address_space_operations btrfs_symlink_aops = {
9094         .readpage       = btrfs_readpage,
9095         .writepage      = btrfs_writepage,
9096         .invalidatepage = btrfs_invalidatepage,
9097         .releasepage    = btrfs_releasepage,
9098 };
9099
9100 static const struct inode_operations btrfs_file_inode_operations = {
9101         .getattr        = btrfs_getattr,
9102         .setattr        = btrfs_setattr,
9103         .setxattr       = btrfs_setxattr,
9104         .getxattr       = btrfs_getxattr,
9105         .listxattr      = btrfs_listxattr,
9106         .removexattr    = btrfs_removexattr,
9107         .permission     = btrfs_permission,
9108         .fiemap         = btrfs_fiemap,
9109         .get_acl        = btrfs_get_acl,
9110         .set_acl        = btrfs_set_acl,
9111         .update_time    = btrfs_update_time,
9112 };
9113 static const struct inode_operations btrfs_special_inode_operations = {
9114         .getattr        = btrfs_getattr,
9115         .setattr        = btrfs_setattr,
9116         .permission     = btrfs_permission,
9117         .setxattr       = btrfs_setxattr,
9118         .getxattr       = btrfs_getxattr,
9119         .listxattr      = btrfs_listxattr,
9120         .removexattr    = btrfs_removexattr,
9121         .get_acl        = btrfs_get_acl,
9122         .set_acl        = btrfs_set_acl,
9123         .update_time    = btrfs_update_time,
9124 };
9125 static const struct inode_operations btrfs_symlink_inode_operations = {
9126         .readlink       = generic_readlink,
9127         .follow_link    = page_follow_link_light,
9128         .put_link       = page_put_link,
9129         .getattr        = btrfs_getattr,
9130         .setattr        = btrfs_setattr,
9131         .permission     = btrfs_permission,
9132         .setxattr       = btrfs_setxattr,
9133         .getxattr       = btrfs_getxattr,
9134         .listxattr      = btrfs_listxattr,
9135         .removexattr    = btrfs_removexattr,
9136         .update_time    = btrfs_update_time,
9137 };
9138
9139 const struct dentry_operations btrfs_dentry_operations = {
9140         .d_delete       = btrfs_dentry_delete,
9141         .d_release      = btrfs_dentry_release,
9142 };