fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include <linux/ratelimit.h>
  41 #include <linux/mount.h>
  42 #include <linux/btrfs.h>
  43 #include <linux/blkdev.h>
  44 #include "compat.h"
  45 #include "ctree.h"
  46 #include "disk-io.h"
  47 #include "transaction.h"
  48 #include "btrfs_inode.h"
  49 #include "print-tree.h"
  50 #include "ordered-data.h"
  51 #include "xattr.h"
  52 #include "tree-log.h"
  53 #include "volumes.h"
  54 #include "compression.h"
  55 #include "locking.h"
  56 #include "free-space-cache.h"
  57 #include "inode-map.h"
  58 #include "backref.h"
  59
  60 struct btrfs_iget_args {
  61         u64 ino;
  62         struct btrfs_root *root;
  63 };
  64
  65 static const struct inode_operations btrfs_dir_inode_operations;
  66 static const struct inode_operations btrfs_symlink_inode_operations;
  67 static const struct inode_operations btrfs_dir_ro_inode_operations;
  68 static const struct inode_operations btrfs_special_inode_operations;
  69 static const struct inode_operations btrfs_file_inode_operations;
  70 static const struct address_space_operations btrfs_aops;
  71 static const struct address_space_operations btrfs_symlink_aops;
  72 static const struct file_operations btrfs_dir_file_operations;
  73 static struct extent_io_ops btrfs_extent_io_ops;
  74
  75 static struct kmem_cache *btrfs_inode_cachep;
  76 static struct kmem_cache *btrfs_delalloc_work_cachep;
  77 struct kmem_cache *btrfs_trans_handle_cachep;
  78 struct kmem_cache *btrfs_transaction_cachep;
  79 struct kmem_cache *btrfs_path_cachep;
  80 struct kmem_cache *btrfs_free_space_cachep;
  81
  82 #define S_SHIFT 12
  83 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  84         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  85         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  86         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  87         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  88         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  89         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  90         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  91 };
  92
  93 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
  94 static int btrfs_truncate(struct inode *inode);
  95 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  96 static noinline int cow_file_range(struct inode *inode,
  97                                    struct page *locked_page,
  98                                    u64 start, u64 end, int *page_started,
  99                                    unsigned long *nr_written, int unlock);
 100 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
 101                                            u64 len, u64 orig_start,
 102                                            u64 block_start, u64 block_len,
 103                                            u64 orig_block_len, u64 ram_bytes,
 104                                            int type);
 105
 106 static int btrfs_dirty_inode(struct inode *inode);
 107
 108 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 109                                      struct inode *inode,  struct inode *dir,
 110                                      const struct qstr *qstr)
 111 {
 112         int err;
 113
 114         err = btrfs_init_acl(trans, inode, dir);
 115         if (!err)
 116                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 117         return err;
 118 }
 119
 120 /*
 121  * this does all the hard work for inserting an inline extent into
 122  * the btree.  The caller should have done a btrfs_drop_extents so that
 123  * no overlapping inline items exist in the btree
 124  */
 125 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 126                                 struct btrfs_root *root, struct inode *inode,
 127                                 u64 start, size_t size, size_t compressed_size,
 128                                 int compress_type,
 129                                 struct page **compressed_pages)
 130 {
 131         struct btrfs_key key;
 132         struct btrfs_path *path;
 133         struct extent_buffer *leaf;
 134         struct page *page = NULL;
 135         char *kaddr;
 136         unsigned long ptr;
 137         struct btrfs_file_extent_item *ei;
 138         int err = 0;
 139         int ret;
 140         size_t cur_size = size;
 141         size_t datasize;
 142         unsigned long offset;
 143
 144         if (compressed_size && compressed_pages)
 145                 cur_size = compressed_size;
 146
 147         path = btrfs_alloc_path();
 148         if (!path)
 149                 return -ENOMEM;
 150
 151         path->leave_spinning = 1;
 152
 153         key.objectid = btrfs_ino(inode);
 154         key.offset = start;
 155         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 156         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 157
 158         inode_add_bytes(inode, size);
 159         ret = btrfs_insert_empty_item(trans, root, path, &key,
 160                                       datasize);
 161         if (ret) {
 162                 err = ret;
 163                 goto fail;
 164         }
 165         leaf = path->nodes[0];
 166         ei = btrfs_item_ptr(leaf, path->slots[0],
 167                             struct btrfs_file_extent_item);
 168         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 169         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 170         btrfs_set_file_extent_encryption(leaf, ei, 0);
 171         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 172         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 173         ptr = btrfs_file_extent_inline_start(ei);
 174
 175         if (compress_type != BTRFS_COMPRESS_NONE) {
 176                 struct page *cpage;
 177                 int i = 0;
 178                 while (compressed_size > 0) {
 179                         cpage = compressed_pages[i];
 180                         cur_size = min_t(unsigned long, compressed_size,
 181                                        PAGE_CACHE_SIZE);
 182
 183                         kaddr = kmap_atomic(cpage);
 184                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 185                         kunmap_atomic(kaddr);
 186
 187                         i++;
 188                         ptr += cur_size;
 189                         compressed_size -= cur_size;
 190                 }
 191                 btrfs_set_file_extent_compression(leaf, ei,
 192                                                   compress_type);
 193         } else {
 194                 page = find_get_page(inode->i_mapping,
 195                                      start >> PAGE_CACHE_SHIFT);
 196                 btrfs_set_file_extent_compression(leaf, ei, 0);
 197                 kaddr = kmap_atomic(page);
 198                 offset = start & (PAGE_CACHE_SIZE - 1);
 199                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 200                 kunmap_atomic(kaddr);
 201                 page_cache_release(page);
 202         }
 203         btrfs_mark_buffer_dirty(leaf);
 204         btrfs_free_path(path);
 205
 206         /*
 207          * we're an inline extent, so nobody can
 208          * extend the file past i_size without locking
 209          * a page we already have locked.
 210          *
 211          * We must do any isize and inode updates
 212          * before we unlock the pages.  Otherwise we
 213          * could end up racing with unlink.
 214          */
 215         BTRFS_I(inode)->disk_i_size = inode->i_size;
 216         ret = btrfs_update_inode(trans, root, inode);
 217
 218         return ret;
 219 fail:
 220         btrfs_free_path(path);
 221         return err;
 222 }
 223
 224
 225 /*
 226  * conditionally insert an inline extent into the file.  This
 227  * does the checks required to make sure the data is small enough
 228  * to fit as an inline extent.
 229  */
 230 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 231                                  struct btrfs_root *root,
 232                                  struct inode *inode, u64 start, u64 end,
 233                                  size_t compressed_size, int compress_type,
 234                                  struct page **compressed_pages)
 235 {
 236         u64 isize = i_size_read(inode);
 237         u64 actual_end = min(end + 1, isize);
 238         u64 inline_len = actual_end - start;
 239         u64 aligned_end = ALIGN(end, root->sectorsize);
 240         u64 data_len = inline_len;
 241         int ret;
 242
 243         if (compressed_size)
 244                 data_len = compressed_size;
 245
 246         if (start > 0 ||
 247             actual_end >= PAGE_CACHE_SIZE ||
 248             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 249             (!compressed_size &&
 250             (actual_end & (root->sectorsize - 1)) == 0) ||
 251             end + 1 < isize ||
 252             data_len > root->fs_info->max_inline) {
 253                 return 1;
 254         }
 255
 256         ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
 257         if (ret)
 258                 return ret;
 259
 260         if (isize > actual_end)
 261                 inline_len = min_t(u64, isize, actual_end);
 262         ret = insert_inline_extent(trans, root, inode, start,
 263                                    inline_len, compressed_size,
 264                                    compress_type, compressed_pages);
 265         if (ret && ret != -ENOSPC) {
 266                 btrfs_abort_transaction(trans, root, ret);
 267                 return ret;
 268         } else if (ret == -ENOSPC) {
 269                 return 1;
 270         }
 271
 272         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
 273         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 274         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 275         return 0;
 276 }
 277
 278 struct async_extent {
 279         u64 start;
 280         u64 ram_size;
 281         u64 compressed_size;
 282         struct page **pages;
 283         unsigned long nr_pages;
 284         int compress_type;
 285         struct list_head list;
 286 };
 287
 288 struct async_cow {
 289         struct inode *inode;
 290         struct btrfs_root *root;
 291         struct page *locked_page;
 292         u64 start;
 293         u64 end;
 294         struct list_head extents;
 295         struct btrfs_work work;
 296 };
 297
 298 static noinline int add_async_extent(struct async_cow *cow,
 299                                      u64 start, u64 ram_size,
 300                                      u64 compressed_size,
 301                                      struct page **pages,
 302                                      unsigned long nr_pages,
 303                                      int compress_type)
 304 {
 305         struct async_extent *async_extent;
 306
 307         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 308         BUG_ON(!async_extent); /* -ENOMEM */
 309         async_extent->start = start;
 310         async_extent->ram_size = ram_size;
 311         async_extent->compressed_size = compressed_size;
 312         async_extent->pages = pages;
 313         async_extent->nr_pages = nr_pages;
 314         async_extent->compress_type = compress_type;
 315         list_add_tail(&async_extent->list, &cow->extents);
 316         return 0;
 317 }
 318
 319 /*
 320  * we create compressed extents in two phases.  The first
 321  * phase compresses a range of pages that have already been
 322  * locked (both pages and state bits are locked).
 323  *
 324  * This is done inside an ordered work queue, and the compression
 325  * is spread across many cpus.  The actual IO submission is step
 326  * two, and the ordered work queue takes care of making sure that
 327  * happens in the same order things were put onto the queue by
 328  * writepages and friends.
 329  *
 330  * If this code finds it can't get good compression, it puts an
 331  * entry onto the work queue to write the uncompressed bytes.  This
 332  * makes sure that both compressed inodes and uncompressed inodes
 333  * are written in the same order that the flusher thread sent them
 334  * down.
 335  */
 336 static noinline int compress_file_range(struct inode *inode,
 337                                         struct page *locked_page,
 338                                         u64 start, u64 end,
 339                                         struct async_cow *async_cow,
 340                                         int *num_added)
 341 {
 342         struct btrfs_root *root = BTRFS_I(inode)->root;
 343         struct btrfs_trans_handle *trans;
 344         u64 num_bytes;
 345         u64 blocksize = root->sectorsize;
 346         u64 actual_end;
 347         u64 isize = i_size_read(inode);
 348         int ret = 0;
 349         struct page **pages = NULL;
 350         unsigned long nr_pages;
 351         unsigned long nr_pages_ret = 0;
 352         unsigned long total_compressed = 0;
 353         unsigned long total_in = 0;
 354         unsigned long max_compressed = 128 * 1024;
 355         unsigned long max_uncompressed = 128 * 1024;
 356         int i;
 357         int will_compress;
 358         int compress_type = root->fs_info->compress_type;
 359         int redirty = 0;
 360
 361         /* if this is a small write inside eof, kick off a defrag */
 362         if ((end - start + 1) < 16 * 1024 &&
 363             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 364                 btrfs_add_inode_defrag(NULL, inode);
 365
 366         actual_end = min_t(u64, isize, end + 1);
 367 again:
 368         will_compress = 0;
 369         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 370         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 371
 372         /*
 373          * we don't want to send crud past the end of i_size through
 374          * compression, that's just a waste of CPU time.  So, if the
 375          * end of the file is before the start of our current
 376          * requested range of bytes, we bail out to the uncompressed
 377          * cleanup code that can deal with all of this.
 378          *
 379          * It isn't really the fastest way to fix things, but this is a
 380          * very uncommon corner.
 381          */
 382         if (actual_end <= start)
 383                 goto cleanup_and_bail_uncompressed;
 384
 385         total_compressed = actual_end - start;
 386
 387         /* we want to make sure that amount of ram required to uncompress
 388          * an extent is reasonable, so we limit the total size in ram
 389          * of a compressed extent to 128k.  This is a crucial number
 390          * because it also controls how easily we can spread reads across
 391          * cpus for decompression.
 392          *
 393          * We also want to make sure the amount of IO required to do
 394          * a random read is reasonably small, so we limit the size of
 395          * a compressed extent to 128k.
 396          */
 397         total_compressed = min(total_compressed, max_uncompressed);
 398         num_bytes = ALIGN(end - start + 1, blocksize);
 399         num_bytes = max(blocksize,  num_bytes);
 400         total_in = 0;
 401         ret = 0;
 402
 403         /*
 404          * we do compression for mount -o compress and when the
 405          * inode has not been flagged as nocompress.  This flag can
 406          * change at any time if we discover bad compression ratios.
 407          */
 408         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 409             (btrfs_test_opt(root, COMPRESS) ||
 410              (BTRFS_I(inode)->force_compress) ||
 411              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 412                 WARN_ON(pages);
 413                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 414                 if (!pages) {
 415                         /* just bail out to the uncompressed code */
 416                         goto cont;
 417                 }
 418
 419                 if (BTRFS_I(inode)->force_compress)
 420                         compress_type = BTRFS_I(inode)->force_compress;
 421
 422                 /*
 423                  * we need to call clear_page_dirty_for_io on each
 424                  * page in the range.  Otherwise applications with the file
 425                  * mmap'd can wander in and change the page contents while
 426                  * we are compressing them.
 427                  *
 428                  * If the compression fails for any reason, we set the pages
 429                  * dirty again later on.
 430                  */
 431                 extent_range_clear_dirty_for_io(inode, start, end);
 432                 redirty = 1;
 433                 ret = btrfs_compress_pages(compress_type,
 434                                            inode->i_mapping, start,
 435                                            total_compressed, pages,
 436                                            nr_pages, &nr_pages_ret,
 437                                            &total_in,
 438                                            &total_compressed,
 439                                            max_compressed);
 440
 441                 if (!ret) {
 442                         unsigned long offset = total_compressed &
 443                                 (PAGE_CACHE_SIZE - 1);
 444                         struct page *page = pages[nr_pages_ret - 1];
 445                         char *kaddr;
 446
 447                         /* zero the tail end of the last page, we might be
 448                          * sending it down to disk
 449                          */
 450                         if (offset) {
 451                                 kaddr = kmap_atomic(page);
 452                                 memset(kaddr + offset, 0,
 453                                        PAGE_CACHE_SIZE - offset);
 454                                 kunmap_atomic(kaddr);
 455                         }
 456                         will_compress = 1;
 457                 }
 458         }
 459 cont:
 460         if (start == 0) {
 461                 trans = btrfs_join_transaction(root);
 462                 if (IS_ERR(trans)) {
 463                         ret = PTR_ERR(trans);
 464                         trans = NULL;
 465                         goto cleanup_and_out;
 466                 }
 467                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 468
 469                 /* lets try to make an inline extent */
 470                 if (ret || total_in < (actual_end - start)) {
 471                         /* we didn't compress the entire range, try
 472                          * to make an uncompressed inline extent.
 473                          */
 474                         ret = cow_file_range_inline(trans, root, inode,
 475                                                     start, end, 0, 0, NULL);
 476                 } else {
 477                         /* try making a compressed inline extent */
 478                         ret = cow_file_range_inline(trans, root, inode,
 479                                                     start, end,
 480                                                     total_compressed,
 481                                                     compress_type, pages);
 482                 }
 483                 if (ret <= 0) {
 484                         /*
 485                          * inline extent creation worked or returned error,
 486                          * we don't need to create any more async work items.
 487                          * Unlock and free up our temp pages.
 488                          */
 489                         extent_clear_unlock_delalloc(inode,
 490                              &BTRFS_I(inode)->io_tree,
 491                              start, end, NULL,
 492                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 493                              EXTENT_CLEAR_DELALLOC |
 494                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 495
 496                         btrfs_end_transaction(trans, root);
 497                         goto free_pages_out;
 498                 }
 499                 btrfs_end_transaction(trans, root);
 500         }
 501
 502         if (will_compress) {
 503                 /*
 504                  * we aren't doing an inline extent round the compressed size
 505                  * up to a block size boundary so the allocator does sane
 506                  * things
 507                  */
 508                 total_compressed = ALIGN(total_compressed, blocksize);
 509
 510                 /*
 511                  * one last check to make sure the compression is really a
 512                  * win, compare the page count read with the blocks on disk
 513                  */
 514                 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
 515                 if (total_compressed >= total_in) {
 516                         will_compress = 0;
 517                 } else {
 518                         num_bytes = total_in;
 519                 }
 520         }
 521         if (!will_compress && pages) {
 522                 /*
 523                  * the compression code ran but failed to make things smaller,
 524                  * free any pages it allocated and our page pointer array
 525                  */
 526                 for (i = 0; i < nr_pages_ret; i++) {
 527                         WARN_ON(pages[i]->mapping);
 528                         page_cache_release(pages[i]);
 529                 }
 530                 kfree(pages);
 531                 pages = NULL;
 532                 total_compressed = 0;
 533                 nr_pages_ret = 0;
 534
 535                 /* flag the file so we don't compress in the future */
 536                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 537                     !(BTRFS_I(inode)->force_compress)) {
 538                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 539                 }
 540         }
 541         if (will_compress) {
 542                 *num_added += 1;
 543
 544                 /* the async work queues will take care of doing actual
 545                  * allocation on disk for these compressed pages,
 546                  * and will submit them to the elevator.
 547                  */
 548                 add_async_extent(async_cow, start, num_bytes,
 549                                  total_compressed, pages, nr_pages_ret,
 550                                  compress_type);
 551
 552                 if (start + num_bytes < end) {
 553                         start += num_bytes;
 554                         pages = NULL;
 555                         cond_resched();
 556                         goto again;
 557                 }
 558         } else {
 559 cleanup_and_bail_uncompressed:
 560                 /*
 561                  * No compression, but we still need to write the pages in
 562                  * the file we've been given so far.  redirty the locked
 563                  * page if it corresponds to our extent and set things up
 564                  * for the async work queue to run cow_file_range to do
 565                  * the normal delalloc dance
 566                  */
 567                 if (page_offset(locked_page) >= start &&
 568                     page_offset(locked_page) <= end) {
 569                         __set_page_dirty_nobuffers(locked_page);
 570                         /* unlocked later on in the async handlers */
 571                 }
 572                 if (redirty)
 573                         extent_range_redirty_for_io(inode, start, end);
 574                 add_async_extent(async_cow, start, end - start + 1,
 575                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 576                 *num_added += 1;
 577         }
 578
 579 out:
 580         return ret;
 581
 582 free_pages_out:
 583         for (i = 0; i < nr_pages_ret; i++) {
 584                 WARN_ON(pages[i]->mapping);
 585                 page_cache_release(pages[i]);
 586         }
 587         kfree(pages);
 588
 589         goto out;
 590
 591 cleanup_and_out:
 592         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 593                                      start, end, NULL,
 594                                      EXTENT_CLEAR_UNLOCK_PAGE |
 595                                      EXTENT_CLEAR_DIRTY |
 596                                      EXTENT_CLEAR_DELALLOC |
 597                                      EXTENT_SET_WRITEBACK |
 598                                      EXTENT_END_WRITEBACK);
 599         if (!trans || IS_ERR(trans))
 600                 btrfs_error(root->fs_info, ret, "Failed to join transaction");
 601         else
 602                 btrfs_abort_transaction(trans, root, ret);
 603         goto free_pages_out;
 604 }
 605
 606 /*
 607  * phase two of compressed writeback.  This is the ordered portion
 608  * of the code, which only gets called in the order the work was
 609  * queued.  We walk all the async extents created by compress_file_range
 610  * and send them down to the disk.
 611  */
 612 static noinline int submit_compressed_extents(struct inode *inode,
 613                                               struct async_cow *async_cow)
 614 {
 615         struct async_extent *async_extent;
 616         u64 alloc_hint = 0;
 617         struct btrfs_trans_handle *trans;
 618         struct btrfs_key ins;
 619         struct extent_map *em;
 620         struct btrfs_root *root = BTRFS_I(inode)->root;
 621         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 622         struct extent_io_tree *io_tree;
 623         int ret = 0;
 624
 625         if (list_empty(&async_cow->extents))
 626                 return 0;
 627
 628 again:
 629         while (!list_empty(&async_cow->extents)) {
 630                 async_extent = list_entry(async_cow->extents.next,
 631                                           struct async_extent, list);
 632                 list_del(&async_extent->list);
 633
 634                 io_tree = &BTRFS_I(inode)->io_tree;
 635
 636 retry:
 637                 /* did the compression code fall back to uncompressed IO? */
 638                 if (!async_extent->pages) {
 639                         int page_started = 0;
 640                         unsigned long nr_written = 0;
 641
 642                         lock_extent(io_tree, async_extent->start,
 643                                          async_extent->start +
 644                                          async_extent->ram_size - 1);
 645
 646                         /* allocate blocks */
 647                         ret = cow_file_range(inode, async_cow->locked_page,
 648                                              async_extent->start,
 649                                              async_extent->start +
 650                                              async_extent->ram_size - 1,
 651                                              &page_started, &nr_written, 0);
 652
 653                         /* JDM XXX */
 654
 655                         /*
 656                          * if page_started, cow_file_range inserted an
 657                          * inline extent and took care of all the unlocking
 658                          * and IO for us.  Otherwise, we need to submit
 659                          * all those pages down to the drive.
 660                          */
 661                         if (!page_started && !ret)
 662                                 extent_write_locked_range(io_tree,
 663                                                   inode, async_extent->start,
 664                                                   async_extent->start +
 665                                                   async_extent->ram_size - 1,
 666                                                   btrfs_get_extent,
 667                                                   WB_SYNC_ALL);
 668                         else if (ret)
 669                                 unlock_page(async_cow->locked_page);
 670                         kfree(async_extent);
 671                         cond_resched();
 672                         continue;
 673                 }
 674
 675                 lock_extent(io_tree, async_extent->start,
 676                             async_extent->start + async_extent->ram_size - 1);
 677
 678                 trans = btrfs_join_transaction(root);
 679                 if (IS_ERR(trans)) {
 680                         ret = PTR_ERR(trans);
 681                 } else {
 682                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 683                         ret = btrfs_reserve_extent(trans, root,
 684                                            async_extent->compressed_size,
 685                                            async_extent->compressed_size,
 686                                            0, alloc_hint, &ins, 1);
 687                         if (ret && ret != -ENOSPC)
 688                                 btrfs_abort_transaction(trans, root, ret);
 689                         btrfs_end_transaction(trans, root);
 690                 }
 691
 692                 if (ret) {
 693                         int i;
 694
 695                         for (i = 0; i < async_extent->nr_pages; i++) {
 696                                 WARN_ON(async_extent->pages[i]->mapping);
 697                                 page_cache_release(async_extent->pages[i]);
 698                         }
 699                         kfree(async_extent->pages);
 700                         async_extent->nr_pages = 0;
 701                         async_extent->pages = NULL;
 702
 703                         if (ret == -ENOSPC)
 704                                 goto retry;
 705                         goto out_free;
 706                 }
 707
 708                 /*
 709                  * here we're doing allocation and writeback of the
 710                  * compressed pages
 711                  */
 712                 btrfs_drop_extent_cache(inode, async_extent->start,
 713                                         async_extent->start +
 714                                         async_extent->ram_size - 1, 0);
 715
 716                 em = alloc_extent_map();
 717                 if (!em) {
 718                         ret = -ENOMEM;
 719                         goto out_free_reserve;
 720                 }
 721                 em->start = async_extent->start;
 722                 em->len = async_extent->ram_size;
 723                 em->orig_start = em->start;
 724                 em->mod_start = em->start;
 725                 em->mod_len = em->len;
 726
 727                 em->block_start = ins.objectid;
 728                 em->block_len = ins.offset;
 729                 em->orig_block_len = ins.offset;
 730                 em->ram_bytes = async_extent->ram_size;
 731                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 732                 em->compress_type = async_extent->compress_type;
 733                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 734                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 735                 em->generation = -1;
 736
 737                 while (1) {
 738                         write_lock(&em_tree->lock);
 739                         ret = add_extent_mapping(em_tree, em, 1);
 740                         write_unlock(&em_tree->lock);
 741                         if (ret != -EEXIST) {
 742                                 free_extent_map(em);
 743                                 break;
 744                         }
 745                         btrfs_drop_extent_cache(inode, async_extent->start,
 746                                                 async_extent->start +
 747                                                 async_extent->ram_size - 1, 0);
 748                 }
 749
 750                 if (ret)
 751                         goto out_free_reserve;
 752
 753                 ret = btrfs_add_ordered_extent_compress(inode,
 754                                                 async_extent->start,
 755                                                 ins.objectid,
 756                                                 async_extent->ram_size,
 757                                                 ins.offset,
 758                                                 BTRFS_ORDERED_COMPRESSED,
 759                                                 async_extent->compress_type);
 760                 if (ret)
 761                         goto out_free_reserve;
 762
 763                 /*
 764                  * clear dirty, set writeback and unlock the pages.
 765                  */
 766                 extent_clear_unlock_delalloc(inode,
 767                                 &BTRFS_I(inode)->io_tree,
 768                                 async_extent->start,
 769                                 async_extent->start +
 770                                 async_extent->ram_size - 1,
 771                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 772                                 EXTENT_CLEAR_UNLOCK |
 773                                 EXTENT_CLEAR_DELALLOC |
 774                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 775
 776                 ret = btrfs_submit_compressed_write(inode,
 777                                     async_extent->start,
 778                                     async_extent->ram_size,
 779                                     ins.objectid,
 780                                     ins.offset, async_extent->pages,
 781                                     async_extent->nr_pages);
 782                 alloc_hint = ins.objectid + ins.offset;
 783                 kfree(async_extent);
 784                 if (ret)
 785                         goto out;
 786                 cond_resched();
 787         }
 788         ret = 0;
 789 out:
 790         return ret;
 791 out_free_reserve:
 792         btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
 793 out_free:
 794         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 795                                      async_extent->start,
 796                                      async_extent->start +
 797                                      async_extent->ram_size - 1,
 798                                      NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 799                                      EXTENT_CLEAR_UNLOCK |
 800                                      EXTENT_CLEAR_DELALLOC |
 801                                      EXTENT_CLEAR_DIRTY |
 802                                      EXTENT_SET_WRITEBACK |
 803                                      EXTENT_END_WRITEBACK);
 804         kfree(async_extent);
 805         goto again;
 806 }
 807
 808 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 809                                       u64 num_bytes)
 810 {
 811         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 812         struct extent_map *em;
 813         u64 alloc_hint = 0;
 814
 815         read_lock(&em_tree->lock);
 816         em = search_extent_mapping(em_tree, start, num_bytes);
 817         if (em) {
 818                 /*
 819                  * if block start isn't an actual block number then find the
 820                  * first block in this inode and use that as a hint.  If that
 821                  * block is also bogus then just don't worry about it.
 822                  */
 823                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 824                         free_extent_map(em);
 825                         em = search_extent_mapping(em_tree, 0, 0);
 826                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 827                                 alloc_hint = em->block_start;
 828                         if (em)
 829                                 free_extent_map(em);
 830                 } else {
 831                         alloc_hint = em->block_start;
 832                         free_extent_map(em);
 833                 }
 834         }
 835         read_unlock(&em_tree->lock);
 836
 837         return alloc_hint;
 838 }
 839
 840 /*
 841  * when extent_io.c finds a delayed allocation range in the file,
 842  * the call backs end up in this code.  The basic idea is to
 843  * allocate extents on disk for the range, and create ordered data structs
 844  * in ram to track those extents.
 845  *
 846  * locked_page is the page that writepage had locked already.  We use
 847  * it to make sure we don't do extra locks or unlocks.
 848  *
 849  * *page_started is set to one if we unlock locked_page and do everything
 850  * required to start IO on it.  It may be clean and already done with
 851  * IO when we return.
 852  */
 853 static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
 854                                      struct inode *inode,
 855                                      struct btrfs_root *root,
 856                                      struct page *locked_page,
 857                                      u64 start, u64 end, int *page_started,
 858                                      unsigned long *nr_written,
 859                                      int unlock)
 860 {
 861         u64 alloc_hint = 0;
 862         u64 num_bytes;
 863         unsigned long ram_size;
 864         u64 disk_num_bytes;
 865         u64 cur_alloc_size;
 866         u64 blocksize = root->sectorsize;
 867         struct btrfs_key ins;
 868         struct extent_map *em;
 869         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 870         int ret = 0;
 871
 872         BUG_ON(btrfs_is_free_space_inode(inode));
 873
 874         num_bytes = ALIGN(end - start + 1, blocksize);
 875         num_bytes = max(blocksize,  num_bytes);
 876         disk_num_bytes = num_bytes;
 877
 878         /* if this is a small write inside eof, kick off defrag */
 879         if (num_bytes < 64 * 1024 &&
 880             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 881                 btrfs_add_inode_defrag(trans, inode);
 882
 883         if (start == 0) {
 884                 /* lets try to make an inline extent */
 885                 ret = cow_file_range_inline(trans, root, inode,
 886                                             start, end, 0, 0, NULL);
 887                 if (ret == 0) {
 888                         extent_clear_unlock_delalloc(inode,
 889                                      &BTRFS_I(inode)->io_tree,
 890                                      start, end, NULL,
 891                                      EXTENT_CLEAR_UNLOCK_PAGE |
 892                                      EXTENT_CLEAR_UNLOCK |
 893                                      EXTENT_CLEAR_DELALLOC |
 894                                      EXTENT_CLEAR_DIRTY |
 895                                      EXTENT_SET_WRITEBACK |
 896                                      EXTENT_END_WRITEBACK);
 897
 898                         *nr_written = *nr_written +
 899                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 900                         *page_started = 1;
 901                         goto out;
 902                 } else if (ret < 0) {
 903                         btrfs_abort_transaction(trans, root, ret);
 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(trans, root, cur_alloc_size,
 919                                            root->sectorsize, 0, alloc_hint,
 920                                            &ins, 1);
 921                 if (ret < 0) {
 922                         btrfs_abort_transaction(trans, root, ret);
 923                         goto out_unlock;
 924                 }
 925
 926                 em = alloc_extent_map();
 927                 if (!em) {
 928                         ret = -ENOMEM;
 929                         goto out_reserve;
 930                 }
 931                 em->start = start;
 932                 em->orig_start = em->start;
 933                 ram_size = ins.offset;
 934                 em->len = ins.offset;
 935                 em->mod_start = em->start;
 936                 em->mod_len = em->len;
 937
 938                 em->block_start = ins.objectid;
 939                 em->block_len = ins.offset;
 940                 em->orig_block_len = ins.offset;
 941                 em->ram_bytes = ram_size;
 942                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 943                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 944                 em->generation = -1;
 945
 946                 while (1) {
 947                         write_lock(&em_tree->lock);
 948                         ret = add_extent_mapping(em_tree, em, 1);
 949                         write_unlock(&em_tree->lock);
 950                         if (ret != -EEXIST) {
 951                                 free_extent_map(em);
 952                                 break;
 953                         }
 954                         btrfs_drop_extent_cache(inode, start,
 955                                                 start + ram_size - 1, 0);
 956                 }
 957                 if (ret)
 958                         goto out_reserve;
 959
 960                 cur_alloc_size = ins.offset;
 961                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 962                                                ram_size, cur_alloc_size, 0);
 963                 if (ret)
 964                         goto out_reserve;
 965
 966                 if (root->root_key.objectid ==
 967                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 968                         ret = btrfs_reloc_clone_csums(inode, start,
 969                                                       cur_alloc_size);
 970                         if (ret) {
 971                                 btrfs_abort_transaction(trans, root, ret);
 972                                 goto out_reserve;
 973                         }
 974                 }
 975
 976                 if (disk_num_bytes < cur_alloc_size)
 977                         break;
 978
 979                 /* we're not doing compressed IO, don't unlock the first
 980                  * page (which the caller expects to stay locked), don't
 981                  * clear any dirty bits and don't set any writeback bits
 982                  *
 983                  * Do set the Private2 bit so we know this page was properly
 984                  * setup for writepage
 985                  */
 986                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 987                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 988                         EXTENT_SET_PRIVATE2;
 989
 990                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 991                                              start, start + ram_size - 1,
 992                                              locked_page, op);
 993                 disk_num_bytes -= cur_alloc_size;
 994                 num_bytes -= cur_alloc_size;
 995                 alloc_hint = ins.objectid + ins.offset;
 996                 start += cur_alloc_size;
 997         }
 998 out:
 999         return ret;
1000
1001 out_reserve:
1002         btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
1003 out_unlock:
1004         extent_clear_unlock_delalloc(inode,
1005                      &BTRFS_I(inode)->io_tree,
1006                      start, end, locked_page,
1007                      EXTENT_CLEAR_UNLOCK_PAGE |
1008                      EXTENT_CLEAR_UNLOCK |
1009                      EXTENT_CLEAR_DELALLOC |
1010                      EXTENT_CLEAR_DIRTY |
1011                      EXTENT_SET_WRITEBACK |
1012                      EXTENT_END_WRITEBACK);
1013
1014         goto out;
1015 }
1016
1017 static noinline int cow_file_range(struct inode *inode,
1018                                    struct page *locked_page,
1019                                    u64 start, u64 end, int *page_started,
1020                                    unsigned long *nr_written,
1021                                    int unlock)
1022 {
1023         struct btrfs_trans_handle *trans;
1024         struct btrfs_root *root = BTRFS_I(inode)->root;
1025         int ret;
1026
1027         trans = btrfs_join_transaction(root);
1028         if (IS_ERR(trans)) {
1029                 extent_clear_unlock_delalloc(inode,
1030                              &BTRFS_I(inode)->io_tree,
1031                              start, end, locked_page,
1032                              EXTENT_CLEAR_UNLOCK_PAGE |
1033                              EXTENT_CLEAR_UNLOCK |
1034                              EXTENT_CLEAR_DELALLOC |
1035                              EXTENT_CLEAR_DIRTY |
1036                              EXTENT_SET_WRITEBACK |
1037                              EXTENT_END_WRITEBACK);
1038                 return PTR_ERR(trans);
1039         }
1040         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1041
1042         ret = __cow_file_range(trans, inode, root, locked_page, start, end,
1043                                page_started, nr_written, unlock);
1044
1045         btrfs_end_transaction(trans, root);
1046
1047         return ret;
1048 }
1049
1050 /*
1051  * work queue call back to started compression on a file and pages
1052  */
1053 static noinline void async_cow_start(struct btrfs_work *work)
1054 {
1055         struct async_cow *async_cow;
1056         int num_added = 0;
1057         async_cow = container_of(work, struct async_cow, work);
1058
1059         compress_file_range(async_cow->inode, async_cow->locked_page,
1060                             async_cow->start, async_cow->end, async_cow,
1061                             &num_added);
1062         if (num_added == 0) {
1063                 btrfs_add_delayed_iput(async_cow->inode);
1064                 async_cow->inode = NULL;
1065         }
1066 }
1067
1068 /*
1069  * work queue call back to submit previously compressed pages
1070  */
1071 static noinline void async_cow_submit(struct btrfs_work *work)
1072 {
1073         struct async_cow *async_cow;
1074         struct btrfs_root *root;
1075         unsigned long nr_pages;
1076
1077         async_cow = container_of(work, struct async_cow, work);
1078
1079         root = async_cow->root;
1080         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1081                 PAGE_CACHE_SHIFT;
1082
1083         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1084             5 * 1024 * 1024 &&
1085             waitqueue_active(&root->fs_info->async_submit_wait))
1086                 wake_up(&root->fs_info->async_submit_wait);
1087
1088         if (async_cow->inode)
1089                 submit_compressed_extents(async_cow->inode, async_cow);
1090 }
1091
1092 static noinline void async_cow_free(struct btrfs_work *work)
1093 {
1094         struct async_cow *async_cow;
1095         async_cow = container_of(work, struct async_cow, work);
1096         if (async_cow->inode)
1097                 btrfs_add_delayed_iput(async_cow->inode);
1098         kfree(async_cow);
1099 }
1100
1101 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1102                                 u64 start, u64 end, int *page_started,
1103                                 unsigned long *nr_written)
1104 {
1105         struct async_cow *async_cow;
1106         struct btrfs_root *root = BTRFS_I(inode)->root;
1107         unsigned long nr_pages;
1108         u64 cur_end;
1109         int limit = 10 * 1024 * 1024;
1110
1111         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1112                          1, 0, NULL, GFP_NOFS);
1113         while (start < end) {
1114                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1115                 BUG_ON(!async_cow); /* -ENOMEM */
1116                 async_cow->inode = igrab(inode);
1117                 async_cow->root = root;
1118                 async_cow->locked_page = locked_page;
1119                 async_cow->start = start;
1120
1121                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1122                         cur_end = end;
1123                 else
1124                         cur_end = min(end, start + 512 * 1024 - 1);
1125
1126                 async_cow->end = cur_end;
1127                 INIT_LIST_HEAD(&async_cow->extents);
1128
1129                 async_cow->work.func = async_cow_start;
1130                 async_cow->work.ordered_func = async_cow_submit;
1131                 async_cow->work.ordered_free = async_cow_free;
1132                 async_cow->work.flags = 0;
1133
1134                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1135                         PAGE_CACHE_SHIFT;
1136                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1137
1138                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1139                                    &async_cow->work);
1140
1141                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1142                         wait_event(root->fs_info->async_submit_wait,
1143                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1144                             limit));
1145                 }
1146
1147                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1148                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1149                         wait_event(root->fs_info->async_submit_wait,
1150                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1151                            0));
1152                 }
1153
1154                 *nr_written += nr_pages;
1155                 start = cur_end + 1;
1156         }
1157         *page_started = 1;
1158         return 0;
1159 }
1160
1161 static noinline int csum_exist_in_range(struct btrfs_root *root,
1162                                         u64 bytenr, u64 num_bytes)
1163 {
1164         int ret;
1165         struct btrfs_ordered_sum *sums;
1166         LIST_HEAD(list);
1167
1168         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1169                                        bytenr + num_bytes - 1, &list, 0);
1170         if (ret == 0 && list_empty(&list))
1171                 return 0;
1172
1173         while (!list_empty(&list)) {
1174                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1175                 list_del(&sums->list);
1176                 kfree(sums);
1177         }
1178         return 1;
1179 }
1180
1181 /*
1182  * when nowcow writeback call back.  This checks for snapshots or COW copies
1183  * of the extents that exist in the file, and COWs the file as required.
1184  *
1185  * If no cow copies or snapshots exist, we write directly to the existing
1186  * blocks on disk
1187  */
1188 static noinline int run_delalloc_nocow(struct inode *inode,
1189                                        struct page *locked_page,
1190                               u64 start, u64 end, int *page_started, int force,
1191                               unsigned long *nr_written)
1192 {
1193         struct btrfs_root *root = BTRFS_I(inode)->root;
1194         struct btrfs_trans_handle *trans;
1195         struct extent_buffer *leaf;
1196         struct btrfs_path *path;
1197         struct btrfs_file_extent_item *fi;
1198         struct btrfs_key found_key;
1199         u64 cow_start;
1200         u64 cur_offset;
1201         u64 extent_end;
1202         u64 extent_offset;
1203         u64 disk_bytenr;
1204         u64 num_bytes;
1205         u64 disk_num_bytes;
1206         u64 ram_bytes;
1207         int extent_type;
1208         int ret, err;
1209         int type;
1210         int nocow;
1211         int check_prev = 1;
1212         bool nolock;
1213         u64 ino = btrfs_ino(inode);
1214
1215         path = btrfs_alloc_path();
1216         if (!path) {
1217                 extent_clear_unlock_delalloc(inode,
1218                              &BTRFS_I(inode)->io_tree,
1219                              start, end, locked_page,
1220                              EXTENT_CLEAR_UNLOCK_PAGE |
1221                              EXTENT_CLEAR_UNLOCK |
1222                              EXTENT_CLEAR_DELALLOC |
1223                              EXTENT_CLEAR_DIRTY |
1224                              EXTENT_SET_WRITEBACK |
1225                              EXTENT_END_WRITEBACK);
1226                 return -ENOMEM;
1227         }
1228
1229         nolock = btrfs_is_free_space_inode(inode);
1230
1231         if (nolock)
1232                 trans = btrfs_join_transaction_nolock(root);
1233         else
1234                 trans = btrfs_join_transaction(root);
1235
1236         if (IS_ERR(trans)) {
1237                 extent_clear_unlock_delalloc(inode,
1238                              &BTRFS_I(inode)->io_tree,
1239                              start, end, locked_page,
1240                              EXTENT_CLEAR_UNLOCK_PAGE |
1241                              EXTENT_CLEAR_UNLOCK |
1242                              EXTENT_CLEAR_DELALLOC |
1243                              EXTENT_CLEAR_DIRTY |
1244                              EXTENT_SET_WRITEBACK |
1245                              EXTENT_END_WRITEBACK);
1246                 btrfs_free_path(path);
1247                 return PTR_ERR(trans);
1248         }
1249
1250         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1251
1252         cow_start = (u64)-1;
1253         cur_offset = start;
1254         while (1) {
1255                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1256                                                cur_offset, 0);
1257                 if (ret < 0) {
1258                         btrfs_abort_transaction(trans, root, ret);
1259                         goto error;
1260                 }
1261                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1262                         leaf = path->nodes[0];
1263                         btrfs_item_key_to_cpu(leaf, &found_key,
1264                                               path->slots[0] - 1);
1265                         if (found_key.objectid == ino &&
1266                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1267                                 path->slots[0]--;
1268                 }
1269                 check_prev = 0;
1270 next_slot:
1271                 leaf = path->nodes[0];
1272                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1273                         ret = btrfs_next_leaf(root, path);
1274                         if (ret < 0) {
1275                                 btrfs_abort_transaction(trans, root, ret);
1276                                 goto error;
1277                         }
1278                         if (ret > 0)
1279                                 break;
1280                         leaf = path->nodes[0];
1281                 }
1282
1283                 nocow = 0;
1284                 disk_bytenr = 0;
1285                 num_bytes = 0;
1286                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1287
1288                 if (found_key.objectid > ino ||
1289                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1290                     found_key.offset > end)
1291                         break;
1292
1293                 if (found_key.offset > cur_offset) {
1294                         extent_end = found_key.offset;
1295                         extent_type = 0;
1296                         goto out_check;
1297                 }
1298
1299                 fi = btrfs_item_ptr(leaf, path->slots[0],
1300                                     struct btrfs_file_extent_item);
1301                 extent_type = btrfs_file_extent_type(leaf, fi);
1302
1303                 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1304                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1305                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1306                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1307                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1308                         extent_end = found_key.offset +
1309                                 btrfs_file_extent_num_bytes(leaf, fi);
1310                         disk_num_bytes =
1311                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1312                         if (extent_end <= start) {
1313                                 path->slots[0]++;
1314                                 goto next_slot;
1315                         }
1316                         if (disk_bytenr == 0)
1317                                 goto out_check;
1318                         if (btrfs_file_extent_compression(leaf, fi) ||
1319                             btrfs_file_extent_encryption(leaf, fi) ||
1320                             btrfs_file_extent_other_encoding(leaf, fi))
1321                                 goto out_check;
1322                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1323                                 goto out_check;
1324                         if (btrfs_extent_readonly(root, disk_bytenr))
1325                                 goto out_check;
1326                         if (btrfs_cross_ref_exist(trans, root, ino,
1327                                                   found_key.offset -
1328                                                   extent_offset, disk_bytenr))
1329                                 goto out_check;
1330                         disk_bytenr += extent_offset;
1331                         disk_bytenr += cur_offset - found_key.offset;
1332                         num_bytes = min(end + 1, extent_end) - cur_offset;
1333                         /*
1334                          * force cow if csum exists in the range.
1335                          * this ensure that csum for a given extent are
1336                          * either valid or do not exist.
1337                          */
1338                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1339                                 goto out_check;
1340                         nocow = 1;
1341                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1342                         extent_end = found_key.offset +
1343                                 btrfs_file_extent_inline_len(leaf, fi);
1344                         extent_end = ALIGN(extent_end, root->sectorsize);
1345                 } else {
1346                         BUG_ON(1);
1347                 }
1348 out_check:
1349                 if (extent_end <= start) {
1350                         path->slots[0]++;
1351                         goto next_slot;
1352                 }
1353                 if (!nocow) {
1354                         if (cow_start == (u64)-1)
1355                                 cow_start = cur_offset;
1356                         cur_offset = extent_end;
1357                         if (cur_offset > end)
1358                                 break;
1359                         path->slots[0]++;
1360                         goto next_slot;
1361                 }
1362
1363                 btrfs_release_path(path);
1364                 if (cow_start != (u64)-1) {
1365                         ret = __cow_file_range(trans, inode, root, locked_page,
1366                                                cow_start, found_key.offset - 1,
1367                                                page_started, nr_written, 1);
1368                         if (ret) {
1369                                 btrfs_abort_transaction(trans, root, ret);
1370                                 goto error;
1371                         }
1372                         cow_start = (u64)-1;
1373                 }
1374
1375                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1376                         struct extent_map *em;
1377                         struct extent_map_tree *em_tree;
1378                         em_tree = &BTRFS_I(inode)->extent_tree;
1379                         em = alloc_extent_map();
1380                         BUG_ON(!em); /* -ENOMEM */
1381                         em->start = cur_offset;
1382                         em->orig_start = found_key.offset - extent_offset;
1383                         em->len = num_bytes;
1384                         em->block_len = num_bytes;
1385                         em->block_start = disk_bytenr;
1386                         em->orig_block_len = disk_num_bytes;
1387                         em->ram_bytes = ram_bytes;
1388                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1389                         em->mod_start = em->start;
1390                         em->mod_len = em->len;
1391                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1392                         set_bit(EXTENT_FLAG_FILLING, &em->flags);
1393                         em->generation = -1;
1394                         while (1) {
1395                                 write_lock(&em_tree->lock);
1396                                 ret = add_extent_mapping(em_tree, em, 1);
1397                                 write_unlock(&em_tree->lock);
1398                                 if (ret != -EEXIST) {
1399                                         free_extent_map(em);
1400                                         break;
1401                                 }
1402                                 btrfs_drop_extent_cache(inode, em->start,
1403                                                 em->start + em->len - 1, 0);
1404                         }
1405                         type = BTRFS_ORDERED_PREALLOC;
1406                 } else {
1407                         type = BTRFS_ORDERED_NOCOW;
1408                 }
1409
1410                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1411                                                num_bytes, num_bytes, type);
1412                 BUG_ON(ret); /* -ENOMEM */
1413
1414                 if (root->root_key.objectid ==
1415                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1416                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1417                                                       num_bytes);
1418                         if (ret) {
1419                                 btrfs_abort_transaction(trans, root, ret);
1420                                 goto error;
1421                         }
1422                 }
1423
1424                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1425                                 cur_offset, cur_offset + num_bytes - 1,
1426                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1427                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1428                                 EXTENT_SET_PRIVATE2);
1429                 cur_offset = extent_end;
1430                 if (cur_offset > end)
1431                         break;
1432         }
1433         btrfs_release_path(path);
1434
1435         if (cur_offset <= end && cow_start == (u64)-1) {
1436                 cow_start = cur_offset;
1437                 cur_offset = end;
1438         }
1439
1440         if (cow_start != (u64)-1) {
1441                 ret = __cow_file_range(trans, inode, root, locked_page,
1442                                        cow_start, end,
1443                                        page_started, nr_written, 1);
1444                 if (ret) {
1445                         btrfs_abort_transaction(trans, root, ret);
1446                         goto error;
1447                 }
1448         }
1449
1450 error:
1451         err = btrfs_end_transaction(trans, root);
1452         if (!ret)
1453                 ret = err;
1454
1455         if (ret && cur_offset < end)
1456                 extent_clear_unlock_delalloc(inode,
1457                              &BTRFS_I(inode)->io_tree,
1458                              cur_offset, end, locked_page,
1459                              EXTENT_CLEAR_UNLOCK_PAGE |
1460                              EXTENT_CLEAR_UNLOCK |
1461                              EXTENT_CLEAR_DELALLOC |
1462                              EXTENT_CLEAR_DIRTY |
1463                              EXTENT_SET_WRITEBACK |
1464                              EXTENT_END_WRITEBACK);
1465
1466         btrfs_free_path(path);
1467         return ret;
1468 }
1469
1470 /*
1471  * extent_io.c call back to do delayed allocation processing
1472  */
1473 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1474                               u64 start, u64 end, int *page_started,
1475                               unsigned long *nr_written)
1476 {
1477         int ret;
1478         struct btrfs_root *root = BTRFS_I(inode)->root;
1479
1480         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1481                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1482                                          page_started, 1, nr_written);
1483         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1484                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1485                                          page_started, 0, nr_written);
1486         } else if (!btrfs_test_opt(root, COMPRESS) &&
1487                    !(BTRFS_I(inode)->force_compress) &&
1488                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1489                 ret = cow_file_range(inode, locked_page, start, end,
1490                                       page_started, nr_written, 1);
1491         } else {
1492                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1493                         &BTRFS_I(inode)->runtime_flags);
1494                 ret = cow_file_range_async(inode, locked_page, start, end,
1495                                            page_started, nr_written);
1496         }
1497         return ret;
1498 }
1499
1500 static void btrfs_split_extent_hook(struct inode *inode,
1501                                     struct extent_state *orig, u64 split)
1502 {
1503         /* not delalloc, ignore it */
1504         if (!(orig->state & EXTENT_DELALLOC))
1505                 return;
1506
1507         spin_lock(&BTRFS_I(inode)->lock);
1508         BTRFS_I(inode)->outstanding_extents++;
1509         spin_unlock(&BTRFS_I(inode)->lock);
1510 }
1511
1512 /*
1513  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1514  * extents so we can keep track of new extents that are just merged onto old
1515  * extents, such as when we are doing sequential writes, so we can properly
1516  * account for the metadata space we'll need.
1517  */
1518 static void btrfs_merge_extent_hook(struct inode *inode,
1519                                     struct extent_state *new,
1520                                     struct extent_state *other)
1521 {
1522         /* not delalloc, ignore it */
1523         if (!(other->state & EXTENT_DELALLOC))
1524                 return;
1525
1526         spin_lock(&BTRFS_I(inode)->lock);
1527         BTRFS_I(inode)->outstanding_extents--;
1528         spin_unlock(&BTRFS_I(inode)->lock);
1529 }
1530
1531 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1532                                       struct inode *inode)
1533 {
1534         spin_lock(&root->delalloc_lock);
1535         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1536                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1537                               &root->delalloc_inodes);
1538                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1539                         &BTRFS_I(inode)->runtime_flags);
1540                 root->nr_delalloc_inodes++;
1541                 if (root->nr_delalloc_inodes == 1) {
1542                         spin_lock(&root->fs_info->delalloc_root_lock);
1543                         BUG_ON(!list_empty(&root->delalloc_root));
1544                         list_add_tail(&root->delalloc_root,
1545                                       &root->fs_info->delalloc_roots);
1546                         spin_unlock(&root->fs_info->delalloc_root_lock);
1547                 }
1548         }
1549         spin_unlock(&root->delalloc_lock);
1550 }
1551
1552 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1553                                      struct inode *inode)
1554 {
1555         spin_lock(&root->delalloc_lock);
1556         if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1557                 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1558                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1559                           &BTRFS_I(inode)->runtime_flags);
1560                 root->nr_delalloc_inodes--;
1561                 if (!root->nr_delalloc_inodes) {
1562                         spin_lock(&root->fs_info->delalloc_root_lock);
1563                         BUG_ON(list_empty(&root->delalloc_root));
1564                         list_del_init(&root->delalloc_root);
1565                         spin_unlock(&root->fs_info->delalloc_root_lock);
1566                 }
1567         }
1568         spin_unlock(&root->delalloc_lock);
1569 }
1570
1571 /*
1572  * extent_io.c set_bit_hook, used to track delayed allocation
1573  * bytes in this file, and to maintain the list of inodes that
1574  * have pending delalloc work to be done.
1575  */
1576 static void btrfs_set_bit_hook(struct inode *inode,
1577                                struct extent_state *state, unsigned long *bits)
1578 {
1579
1580         /*
1581          * set_bit and clear bit hooks normally require _irqsave/restore
1582          * but in this case, we are only testing for the DELALLOC
1583          * bit, which is only set or cleared with irqs on
1584          */
1585         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1586                 struct btrfs_root *root = BTRFS_I(inode)->root;
1587                 u64 len = state->end + 1 - state->start;
1588                 bool do_list = !btrfs_is_free_space_inode(inode);
1589
1590                 if (*bits & EXTENT_FIRST_DELALLOC) {
1591                         *bits &= ~EXTENT_FIRST_DELALLOC;
1592                 } else {
1593                         spin_lock(&BTRFS_I(inode)->lock);
1594                         BTRFS_I(inode)->outstanding_extents++;
1595                         spin_unlock(&BTRFS_I(inode)->lock);
1596                 }
1597
1598                 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1599                                      root->fs_info->delalloc_batch);
1600                 spin_lock(&BTRFS_I(inode)->lock);
1601                 BTRFS_I(inode)->delalloc_bytes += len;
1602                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1603                                          &BTRFS_I(inode)->runtime_flags))
1604                         btrfs_add_delalloc_inodes(root, inode);
1605                 spin_unlock(&BTRFS_I(inode)->lock);
1606         }
1607 }
1608
1609 /*
1610  * extent_io.c clear_bit_hook, see set_bit_hook for why
1611  */
1612 static void btrfs_clear_bit_hook(struct inode *inode,
1613                                  struct extent_state *state,
1614                                  unsigned long *bits)
1615 {
1616         /*
1617          * set_bit and clear bit hooks normally require _irqsave/restore
1618          * but in this case, we are only testing for the DELALLOC
1619          * bit, which is only set or cleared with irqs on
1620          */
1621         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1622                 struct btrfs_root *root = BTRFS_I(inode)->root;
1623                 u64 len = state->end + 1 - state->start;
1624                 bool do_list = !btrfs_is_free_space_inode(inode);
1625
1626                 if (*bits & EXTENT_FIRST_DELALLOC) {
1627                         *bits &= ~EXTENT_FIRST_DELALLOC;
1628                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1629                         spin_lock(&BTRFS_I(inode)->lock);
1630                         BTRFS_I(inode)->outstanding_extents--;
1631                         spin_unlock(&BTRFS_I(inode)->lock);
1632                 }
1633
1634                 if (*bits & EXTENT_DO_ACCOUNTING)
1635                         btrfs_delalloc_release_metadata(inode, len);
1636
1637                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1638                     && do_list)
1639                         btrfs_free_reserved_data_space(inode, len);
1640
1641                 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1642                                      root->fs_info->delalloc_batch);
1643                 spin_lock(&BTRFS_I(inode)->lock);
1644                 BTRFS_I(inode)->delalloc_bytes -= len;
1645                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1646                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1647                              &BTRFS_I(inode)->runtime_flags))
1648                         btrfs_del_delalloc_inode(root, inode);
1649                 spin_unlock(&BTRFS_I(inode)->lock);
1650         }
1651 }
1652
1653 /*
1654  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1655  * we don't create bios that span stripes or chunks
1656  */
1657 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1658                          size_t size, struct bio *bio,
1659                          unsigned long bio_flags)
1660 {
1661         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1662         u64 logical = (u64)bio->bi_sector << 9;
1663         u64 length = 0;
1664         u64 map_length;
1665         int ret;
1666
1667         if (bio_flags & EXTENT_BIO_COMPRESSED)
1668                 return 0;
1669
1670         length = bio->bi_size;
1671         map_length = length;
1672         ret = btrfs_map_block(root->fs_info, rw, logical,
1673                               &map_length, NULL, 0);
1674         /* Will always return 0 with map_multi == NULL */
1675         BUG_ON(ret < 0);
1676         if (map_length < length + size)
1677                 return 1;
1678         return 0;
1679 }
1680
1681 /*
1682  * in order to insert checksums into the metadata in large chunks,
1683  * we wait until bio submission time.   All the pages in the bio are
1684  * checksummed and sums are attached onto the ordered extent record.
1685  *
1686  * At IO completion time the cums attached on the ordered extent record
1687  * are inserted into the btree
1688  */
1689 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1690                                     struct bio *bio, int mirror_num,
1691                                     unsigned long bio_flags,
1692                                     u64 bio_offset)
1693 {
1694         struct btrfs_root *root = BTRFS_I(inode)->root;
1695         int ret = 0;
1696
1697         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1698         BUG_ON(ret); /* -ENOMEM */
1699         return 0;
1700 }
1701
1702 /*
1703  * in order to insert checksums into the metadata in large chunks,
1704  * we wait until bio submission time.   All the pages in the bio are
1705  * checksummed and sums are attached onto the ordered extent record.
1706  *
1707  * At IO completion time the cums attached on the ordered extent record
1708  * are inserted into the btree
1709  */
1710 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1711                           int mirror_num, unsigned long bio_flags,
1712                           u64 bio_offset)
1713 {
1714         struct btrfs_root *root = BTRFS_I(inode)->root;
1715         int ret;
1716
1717         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1718         if (ret)
1719                 bio_endio(bio, ret);
1720         return ret;
1721 }
1722
1723 /*
1724  * extent_io.c submission hook. This does the right thing for csum calculation
1725  * on write, or reading the csums from the tree before a read
1726  */
1727 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1728                           int mirror_num, unsigned long bio_flags,
1729                           u64 bio_offset)
1730 {
1731         struct btrfs_root *root = BTRFS_I(inode)->root;
1732         int ret = 0;
1733         int skip_sum;
1734         int metadata = 0;
1735         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1736
1737         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1738
1739         if (btrfs_is_free_space_inode(inode))
1740                 metadata = 2;
1741
1742         if (!(rw & REQ_WRITE)) {
1743                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1744                 if (ret)
1745                         goto out;
1746
1747                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1748                         ret = btrfs_submit_compressed_read(inode, bio,
1749                                                            mirror_num,
1750                                                            bio_flags);
1751                         goto out;
1752                 } else if (!skip_sum) {
1753                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1754                         if (ret)
1755                                 goto out;
1756                 }
1757                 goto mapit;
1758         } else if (async && !skip_sum) {
1759                 /* csum items have already been cloned */
1760                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1761                         goto mapit;
1762                 /* we're doing a write, do the async checksumming */
1763                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1764                                    inode, rw, bio, mirror_num,
1765                                    bio_flags, bio_offset,
1766                                    __btrfs_submit_bio_start,
1767                                    __btrfs_submit_bio_done);
1768                 goto out;
1769         } else if (!skip_sum) {
1770                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1771                 if (ret)
1772                         goto out;
1773         }
1774
1775 mapit:
1776         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1777
1778 out:
1779         if (ret < 0)
1780                 bio_endio(bio, ret);
1781         return ret;
1782 }
1783
1784 /*
1785  * given a list of ordered sums record them in the inode.  This happens
1786  * at IO completion time based on sums calculated at bio submission time.
1787  */
1788 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1789                              struct inode *inode, u64 file_offset,
1790                              struct list_head *list)
1791 {
1792         struct btrfs_ordered_sum *sum;
1793
1794         list_for_each_entry(sum, list, list) {
1795                 trans->adding_csums = 1;
1796                 btrfs_csum_file_blocks(trans,
1797                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1798                 trans->adding_csums = 0;
1799         }
1800         return 0;
1801 }
1802
1803 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1804                               struct extent_state **cached_state)
1805 {
1806         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1807         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1808                                    cached_state, GFP_NOFS);
1809 }
1810
1811 /* see btrfs_writepage_start_hook for details on why this is required */
1812 struct btrfs_writepage_fixup {
1813         struct page *page;
1814         struct btrfs_work work;
1815 };
1816
1817 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1818 {
1819         struct btrfs_writepage_fixup *fixup;
1820         struct btrfs_ordered_extent *ordered;
1821         struct extent_state *cached_state = NULL;
1822         struct page *page;
1823         struct inode *inode;
1824         u64 page_start;
1825         u64 page_end;
1826         int ret;
1827
1828         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1829         page = fixup->page;
1830 again:
1831         lock_page(page);
1832         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1833                 ClearPageChecked(page);
1834                 goto out_page;
1835         }
1836
1837         inode = page->mapping->host;
1838         page_start = page_offset(page);
1839         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1840
1841         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1842                          &cached_state);
1843
1844         /* already ordered? We're done */
1845         if (PagePrivate2(page))
1846                 goto out;
1847
1848         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1849         if (ordered) {
1850                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1851                                      page_end, &cached_state, GFP_NOFS);
1852                 unlock_page(page);
1853                 btrfs_start_ordered_extent(inode, ordered, 1);
1854                 btrfs_put_ordered_extent(ordered);
1855                 goto again;
1856         }
1857
1858         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1859         if (ret) {
1860                 mapping_set_error(page->mapping, ret);
1861                 end_extent_writepage(page, ret, page_start, page_end);
1862                 ClearPageChecked(page);
1863                 goto out;
1864          }
1865
1866         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1867         ClearPageChecked(page);
1868         set_page_dirty(page);
1869 out:
1870         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1871                              &cached_state, GFP_NOFS);
1872 out_page:
1873         unlock_page(page);
1874         page_cache_release(page);
1875         kfree(fixup);
1876 }
1877
1878 /*
1879  * There are a few paths in the higher layers of the kernel that directly
1880  * set the page dirty bit without asking the filesystem if it is a
1881  * good idea.  This causes problems because we want to make sure COW
1882  * properly happens and the data=ordered rules are followed.
1883  *
1884  * In our case any range that doesn't have the ORDERED bit set
1885  * hasn't been properly setup for IO.  We kick off an async process
1886  * to fix it up.  The async helper will wait for ordered extents, set
1887  * the delalloc bit and make it safe to write the page.
1888  */
1889 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1890 {
1891         struct inode *inode = page->mapping->host;
1892         struct btrfs_writepage_fixup *fixup;
1893         struct btrfs_root *root = BTRFS_I(inode)->root;
1894
1895         /* this page is properly in the ordered list */
1896         if (TestClearPagePrivate2(page))
1897                 return 0;
1898
1899         if (PageChecked(page))
1900                 return -EAGAIN;
1901
1902         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1903         if (!fixup)
1904                 return -EAGAIN;
1905
1906         SetPageChecked(page);
1907         page_cache_get(page);
1908         fixup->work.func = btrfs_writepage_fixup_worker;
1909         fixup->page = page;
1910         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1911         return -EBUSY;
1912 }
1913
1914 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1915                                        struct inode *inode, u64 file_pos,
1916                                        u64 disk_bytenr, u64 disk_num_bytes,
1917                                        u64 num_bytes, u64 ram_bytes,
1918                                        u8 compression, u8 encryption,
1919                                        u16 other_encoding, int extent_type)
1920 {
1921         struct btrfs_root *root = BTRFS_I(inode)->root;
1922         struct btrfs_file_extent_item *fi;
1923         struct btrfs_path *path;
1924         struct extent_buffer *leaf;
1925         struct btrfs_key ins;
1926         int ret;
1927
1928         path = btrfs_alloc_path();
1929         if (!path)
1930                 return -ENOMEM;
1931
1932         path->leave_spinning = 1;
1933
1934         /*
1935          * we may be replacing one extent in the tree with another.
1936          * The new extent is pinned in the extent map, and we don't want
1937          * to drop it from the cache until it is completely in the btree.
1938          *
1939          * So, tell btrfs_drop_extents to leave this extent in the cache.
1940          * the caller is expected to unpin it and allow it to be merged
1941          * with the others.
1942          */
1943         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1944                                  file_pos + num_bytes, 0);
1945         if (ret)
1946                 goto out;
1947
1948         ins.objectid = btrfs_ino(inode);
1949         ins.offset = file_pos;
1950         ins.type = BTRFS_EXTENT_DATA_KEY;
1951         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1952         if (ret)
1953                 goto out;
1954         leaf = path->nodes[0];
1955         fi = btrfs_item_ptr(leaf, path->slots[0],
1956                             struct btrfs_file_extent_item);
1957         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1958         btrfs_set_file_extent_type(leaf, fi, extent_type);
1959         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1960         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1961         btrfs_set_file_extent_offset(leaf, fi, 0);
1962         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1963         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1964         btrfs_set_file_extent_compression(leaf, fi, compression);
1965         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1966         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1967
1968         btrfs_mark_buffer_dirty(leaf);
1969         btrfs_release_path(path);
1970
1971         inode_add_bytes(inode, num_bytes);
1972
1973         ins.objectid = disk_bytenr;
1974         ins.offset = disk_num_bytes;
1975         ins.type = BTRFS_EXTENT_ITEM_KEY;
1976         ret = btrfs_alloc_reserved_file_extent(trans, root,
1977                                         root->root_key.objectid,
1978                                         btrfs_ino(inode), file_pos, &ins);
1979 out:
1980         btrfs_free_path(path);
1981
1982         return ret;
1983 }
1984
1985 /* snapshot-aware defrag */
1986 struct sa_defrag_extent_backref {
1987         struct rb_node node;
1988         struct old_sa_defrag_extent *old;
1989         u64 root_id;
1990         u64 inum;
1991         u64 file_pos;
1992         u64 extent_offset;
1993         u64 num_bytes;
1994         u64 generation;
1995 };
1996
1997 struct old_sa_defrag_extent {
1998         struct list_head list;
1999         struct new_sa_defrag_extent *new;
2000
2001         u64 extent_offset;
2002         u64 bytenr;
2003         u64 offset;
2004         u64 len;
2005         int count;
2006 };
2007
2008 struct new_sa_defrag_extent {
2009         struct rb_root root;
2010         struct list_head head;
2011         struct btrfs_path *path;
2012         struct inode *inode;
2013         u64 file_pos;
2014         u64 len;
2015         u64 bytenr;
2016         u64 disk_len;
2017         u8 compress_type;
2018 };
2019
2020 static int backref_comp(struct sa_defrag_extent_backref *b1,
2021                         struct sa_defrag_extent_backref *b2)
2022 {
2023         if (b1->root_id < b2->root_id)
2024                 return -1;
2025         else if (b1->root_id > b2->root_id)
2026                 return 1;
2027
2028         if (b1->inum < b2->inum)
2029                 return -1;
2030         else if (b1->inum > b2->inum)
2031                 return 1;
2032
2033         if (b1->file_pos < b2->file_pos)
2034                 return -1;
2035         else if (b1->file_pos > b2->file_pos)
2036                 return 1;
2037
2038         /*
2039          * [------------------------------] ===> (a range of space)
2040          *     |<--->|   |<---->| =============> (fs/file tree A)
2041          * |<---------------------------->| ===> (fs/file tree B)
2042          *
2043          * A range of space can refer to two file extents in one tree while
2044          * refer to only one file extent in another tree.
2045          *
2046          * So we may process a disk offset more than one time(two extents in A)
2047          * and locate at the same extent(one extent in B), then insert two same
2048          * backrefs(both refer to the extent in B).
2049          */
2050         return 0;
2051 }
2052
2053 static void backref_insert(struct rb_root *root,
2054                            struct sa_defrag_extent_backref *backref)
2055 {
2056         struct rb_node **p = &root->rb_node;
2057         struct rb_node *parent = NULL;
2058         struct sa_defrag_extent_backref *entry;
2059         int ret;
2060
2061         while (*p) {
2062                 parent = *p;
2063                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2064
2065                 ret = backref_comp(backref, entry);
2066                 if (ret < 0)
2067                         p = &(*p)->rb_left;
2068                 else
2069                         p = &(*p)->rb_right;
2070         }
2071
2072         rb_link_node(&backref->node, parent, p);
2073         rb_insert_color(&backref->node, root);
2074 }
2075
2076 /*
2077  * Note the backref might has changed, and in this case we just return 0.
2078  */
2079 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2080                                        void *ctx)
2081 {
2082         struct btrfs_file_extent_item *extent;
2083         struct btrfs_fs_info *fs_info;
2084         struct old_sa_defrag_extent *old = ctx;
2085         struct new_sa_defrag_extent *new = old->new;
2086         struct btrfs_path *path = new->path;
2087         struct btrfs_key key;
2088         struct btrfs_root *root;
2089         struct sa_defrag_extent_backref *backref;
2090         struct extent_buffer *leaf;
2091         struct inode *inode = new->inode;
2092         int slot;
2093         int ret;
2094         u64 extent_offset;
2095         u64 num_bytes;
2096
2097         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2098             inum == btrfs_ino(inode))
2099                 return 0;
2100
2101         key.objectid = root_id;
2102         key.type = BTRFS_ROOT_ITEM_KEY;
2103         key.offset = (u64)-1;
2104
2105         fs_info = BTRFS_I(inode)->root->fs_info;
2106         root = btrfs_read_fs_root_no_name(fs_info, &key);
2107         if (IS_ERR(root)) {
2108                 if (PTR_ERR(root) == -ENOENT)
2109                         return 0;
2110                 WARN_ON(1);
2111                 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2112                          inum, offset, root_id);
2113                 return PTR_ERR(root);
2114         }
2115
2116         key.objectid = inum;
2117         key.type = BTRFS_EXTENT_DATA_KEY;
2118         if (offset > (u64)-1 << 32)
2119                 key.offset = 0;
2120         else
2121                 key.offset = offset;
2122
2123         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2124         if (ret < 0) {
2125                 WARN_ON(1);
2126                 return ret;
2127         }
2128
2129         while (1) {
2130                 cond_resched();
2131
2132                 leaf = path->nodes[0];
2133                 slot = path->slots[0];
2134
2135                 if (slot >= btrfs_header_nritems(leaf)) {
2136                         ret = btrfs_next_leaf(root, path);
2137                         if (ret < 0) {
2138                                 goto out;
2139                         } else if (ret > 0) {
2140                                 ret = 0;
2141                                 goto out;
2142                         }
2143                         continue;
2144                 }
2145
2146                 path->slots[0]++;
2147
2148                 btrfs_item_key_to_cpu(leaf, &key, slot);
2149
2150                 if (key.objectid > inum)
2151                         goto out;
2152
2153                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2154                         continue;
2155
2156                 extent = btrfs_item_ptr(leaf, slot,
2157                                         struct btrfs_file_extent_item);
2158
2159                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2160                         continue;
2161
2162                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2163                 if (key.offset - extent_offset != offset)
2164                         continue;
2165
2166                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2167                 if (extent_offset >= old->extent_offset + old->offset +
2168                     old->len || extent_offset + num_bytes <=
2169                     old->extent_offset + old->offset)
2170                         continue;
2171
2172                 break;
2173         }
2174
2175         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2176         if (!backref) {
2177                 ret = -ENOENT;
2178                 goto out;
2179         }
2180
2181         backref->root_id = root_id;
2182         backref->inum = inum;
2183         backref->file_pos = offset + extent_offset;
2184         backref->num_bytes = num_bytes;
2185         backref->extent_offset = extent_offset;
2186         backref->generation = btrfs_file_extent_generation(leaf, extent);
2187         backref->old = old;
2188         backref_insert(&new->root, backref);
2189         old->count++;
2190 out:
2191         btrfs_release_path(path);
2192         WARN_ON(ret);
2193         return ret;
2194 }
2195
2196 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2197                                    struct new_sa_defrag_extent *new)
2198 {
2199         struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2200         struct old_sa_defrag_extent *old, *tmp;
2201         int ret;
2202
2203         new->path = path;
2204
2205         list_for_each_entry_safe(old, tmp, &new->head, list) {
2206                 ret = iterate_inodes_from_logical(old->bytenr, fs_info,
2207                                                   path, record_one_backref,
2208                                                   old);
2209                 BUG_ON(ret < 0 && ret != -ENOENT);
2210
2211                 /* no backref to be processed for this extent */
2212                 if (!old->count) {
2213                         list_del(&old->list);
2214                         kfree(old);
2215                 }
2216         }
2217
2218         if (list_empty(&new->head))
2219                 return false;
2220
2221         return true;
2222 }
2223
2224 static int relink_is_mergable(struct extent_buffer *leaf,
2225                               struct btrfs_file_extent_item *fi,
2226                               u64 disk_bytenr)
2227 {
2228         if (btrfs_file_extent_disk_bytenr(leaf, fi) != disk_bytenr)
2229                 return 0;
2230
2231         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2232                 return 0;
2233
2234         if (btrfs_file_extent_compression(leaf, fi) ||
2235             btrfs_file_extent_encryption(leaf, fi) ||
2236             btrfs_file_extent_other_encoding(leaf, fi))
2237                 return 0;
2238
2239         return 1;
2240 }
2241
2242 /*
2243  * Note the backref might has changed, and in this case we just return 0.
2244  */
2245 static noinline int relink_extent_backref(struct btrfs_path *path,
2246                                  struct sa_defrag_extent_backref *prev,
2247                                  struct sa_defrag_extent_backref *backref)
2248 {
2249         struct btrfs_file_extent_item *extent;
2250         struct btrfs_file_extent_item *item;
2251         struct btrfs_ordered_extent *ordered;
2252         struct btrfs_trans_handle *trans;
2253         struct btrfs_fs_info *fs_info;
2254         struct btrfs_root *root;
2255         struct btrfs_key key;
2256         struct extent_buffer *leaf;
2257         struct old_sa_defrag_extent *old = backref->old;
2258         struct new_sa_defrag_extent *new = old->new;
2259         struct inode *src_inode = new->inode;
2260         struct inode *inode;
2261         struct extent_state *cached = NULL;
2262         int ret = 0;
2263         u64 start;
2264         u64 len;
2265         u64 lock_start;
2266         u64 lock_end;
2267         bool merge = false;
2268         int index;
2269
2270         if (prev && prev->root_id == backref->root_id &&
2271             prev->inum == backref->inum &&
2272             prev->file_pos + prev->num_bytes == backref->file_pos)
2273                 merge = true;
2274
2275         /* step 1: get root */
2276         key.objectid = backref->root_id;
2277         key.type = BTRFS_ROOT_ITEM_KEY;
2278         key.offset = (u64)-1;
2279
2280         fs_info = BTRFS_I(src_inode)->root->fs_info;
2281         index = srcu_read_lock(&fs_info->subvol_srcu);
2282
2283         root = btrfs_read_fs_root_no_name(fs_info, &key);
2284         if (IS_ERR(root)) {
2285                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2286                 if (PTR_ERR(root) == -ENOENT)
2287                         return 0;
2288                 return PTR_ERR(root);
2289         }
2290
2291         /* step 2: get inode */
2292         key.objectid = backref->inum;
2293         key.type = BTRFS_INODE_ITEM_KEY;
2294         key.offset = 0;
2295
2296         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2297         if (IS_ERR(inode)) {
2298                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2299                 return 0;
2300         }
2301
2302         srcu_read_unlock(&fs_info->subvol_srcu, index);
2303
2304         /* step 3: relink backref */
2305         lock_start = backref->file_pos;
2306         lock_end = backref->file_pos + backref->num_bytes - 1;
2307         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2308                          0, &cached);
2309
2310         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2311         if (ordered) {
2312                 btrfs_put_ordered_extent(ordered);
2313                 goto out_unlock;
2314         }
2315
2316         trans = btrfs_join_transaction(root);
2317         if (IS_ERR(trans)) {
2318                 ret = PTR_ERR(trans);
2319                 goto out_unlock;
2320         }
2321
2322         key.objectid = backref->inum;
2323         key.type = BTRFS_EXTENT_DATA_KEY;
2324         key.offset = backref->file_pos;
2325
2326         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2327         if (ret < 0) {
2328                 goto out_free_path;
2329         } else if (ret > 0) {
2330                 ret = 0;
2331                 goto out_free_path;
2332         }
2333
2334         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2335                                 struct btrfs_file_extent_item);
2336
2337         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2338             backref->generation)
2339                 goto out_free_path;
2340
2341         btrfs_release_path(path);
2342
2343         start = backref->file_pos;
2344         if (backref->extent_offset < old->extent_offset + old->offset)
2345                 start += old->extent_offset + old->offset -
2346                          backref->extent_offset;
2347
2348         len = min(backref->extent_offset + backref->num_bytes,
2349                   old->extent_offset + old->offset + old->len);
2350         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2351
2352         ret = btrfs_drop_extents(trans, root, inode, start,
2353                                  start + len, 1);
2354         if (ret)
2355                 goto out_free_path;
2356 again:
2357         key.objectid = btrfs_ino(inode);
2358         key.type = BTRFS_EXTENT_DATA_KEY;
2359         key.offset = start;
2360
2361         path->leave_spinning = 1;
2362         if (merge) {
2363                 struct btrfs_file_extent_item *fi;
2364                 u64 extent_len;
2365                 struct btrfs_key found_key;
2366
2367                 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
2368                 if (ret < 0)
2369                         goto out_free_path;
2370
2371                 path->slots[0]--;
2372                 leaf = path->nodes[0];
2373                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2374
2375                 fi = btrfs_item_ptr(leaf, path->slots[0],
2376                                     struct btrfs_file_extent_item);
2377                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2378
2379                 if (relink_is_mergable(leaf, fi, new->bytenr) &&
2380                     extent_len + found_key.offset == start) {
2381                         btrfs_set_file_extent_num_bytes(leaf, fi,
2382                                                         extent_len + len);
2383                         btrfs_mark_buffer_dirty(leaf);
2384                         inode_add_bytes(inode, len);
2385
2386                         ret = 1;
2387                         goto out_free_path;
2388                 } else {
2389                         merge = false;
2390                         btrfs_release_path(path);
2391                         goto again;
2392                 }
2393         }
2394
2395         ret = btrfs_insert_empty_item(trans, root, path, &key,
2396                                         sizeof(*extent));
2397         if (ret) {
2398                 btrfs_abort_transaction(trans, root, ret);
2399                 goto out_free_path;
2400         }
2401
2402         leaf = path->nodes[0];
2403         item = btrfs_item_ptr(leaf, path->slots[0],
2404                                 struct btrfs_file_extent_item);
2405         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2406         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2407         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2408         btrfs_set_file_extent_num_bytes(leaf, item, len);
2409         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2410         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2411         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2412         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2413         btrfs_set_file_extent_encryption(leaf, item, 0);
2414         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2415
2416         btrfs_mark_buffer_dirty(leaf);
2417         inode_add_bytes(inode, len);
2418         btrfs_release_path(path);
2419
2420         ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2421                         new->disk_len, 0,
2422                         backref->root_id, backref->inum,
2423                         new->file_pos, 0);      /* start - extent_offset */
2424         if (ret) {
2425                 btrfs_abort_transaction(trans, root, ret);
2426                 goto out_free_path;
2427         }
2428
2429         ret = 1;
2430 out_free_path:
2431         btrfs_release_path(path);
2432         path->leave_spinning = 0;
2433         btrfs_end_transaction(trans, root);
2434 out_unlock:
2435         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2436                              &cached, GFP_NOFS);
2437         iput(inode);
2438         return ret;
2439 }
2440
2441 static void relink_file_extents(struct new_sa_defrag_extent *new)
2442 {
2443         struct btrfs_path *path;
2444         struct old_sa_defrag_extent *old, *tmp;
2445         struct sa_defrag_extent_backref *backref;
2446         struct sa_defrag_extent_backref *prev = NULL;
2447         struct inode *inode;
2448         struct btrfs_root *root;
2449         struct rb_node *node;
2450         int ret;
2451
2452         inode = new->inode;
2453         root = BTRFS_I(inode)->root;
2454
2455         path = btrfs_alloc_path();
2456         if (!path)
2457                 return;
2458
2459         if (!record_extent_backrefs(path, new)) {
2460                 btrfs_free_path(path);
2461                 goto out;
2462         }
2463         btrfs_release_path(path);
2464
2465         while (1) {
2466                 node = rb_first(&new->root);
2467                 if (!node)
2468                         break;
2469                 rb_erase(node, &new->root);
2470
2471                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2472
2473                 ret = relink_extent_backref(path, prev, backref);
2474                 WARN_ON(ret < 0);
2475
2476                 kfree(prev);
2477
2478                 if (ret == 1)
2479                         prev = backref;
2480                 else
2481                         prev = NULL;
2482                 cond_resched();
2483         }
2484         kfree(prev);
2485
2486         btrfs_free_path(path);
2487
2488         list_for_each_entry_safe(old, tmp, &new->head, list) {
2489                 list_del(&old->list);
2490                 kfree(old);
2491         }
2492 out:
2493         atomic_dec(&root->fs_info->defrag_running);
2494         wake_up(&root->fs_info->transaction_wait);
2495
2496         kfree(new);
2497 }
2498
2499 static struct new_sa_defrag_extent *
2500 record_old_file_extents(struct inode *inode,
2501                         struct btrfs_ordered_extent *ordered)
2502 {
2503         struct btrfs_root *root = BTRFS_I(inode)->root;
2504         struct btrfs_path *path;
2505         struct btrfs_key key;
2506         struct old_sa_defrag_extent *old, *tmp;
2507         struct new_sa_defrag_extent *new;
2508         int ret;
2509
2510         new = kmalloc(sizeof(*new), GFP_NOFS);
2511         if (!new)
2512                 return NULL;
2513
2514         new->inode = inode;
2515         new->file_pos = ordered->file_offset;
2516         new->len = ordered->len;
2517         new->bytenr = ordered->start;
2518         new->disk_len = ordered->disk_len;
2519         new->compress_type = ordered->compress_type;
2520         new->root = RB_ROOT;
2521         INIT_LIST_HEAD(&new->head);
2522
2523         path = btrfs_alloc_path();
2524         if (!path)
2525                 goto out_kfree;
2526
2527         key.objectid = btrfs_ino(inode);
2528         key.type = BTRFS_EXTENT_DATA_KEY;
2529         key.offset = new->file_pos;
2530
2531         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2532         if (ret < 0)
2533                 goto out_free_path;
2534         if (ret > 0 && path->slots[0] > 0)
2535                 path->slots[0]--;
2536
2537         /* find out all the old extents for the file range */
2538         while (1) {
2539                 struct btrfs_file_extent_item *extent;
2540                 struct extent_buffer *l;
2541                 int slot;
2542                 u64 num_bytes;
2543                 u64 offset;
2544                 u64 end;
2545                 u64 disk_bytenr;
2546                 u64 extent_offset;
2547
2548                 l = path->nodes[0];
2549                 slot = path->slots[0];
2550
2551                 if (slot >= btrfs_header_nritems(l)) {
2552                         ret = btrfs_next_leaf(root, path);
2553                         if (ret < 0)
2554                                 goto out_free_list;
2555                         else if (ret > 0)
2556                                 break;
2557                         continue;
2558                 }
2559
2560                 btrfs_item_key_to_cpu(l, &key, slot);
2561
2562                 if (key.objectid != btrfs_ino(inode))
2563                         break;
2564                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2565                         break;
2566                 if (key.offset >= new->file_pos + new->len)
2567                         break;
2568
2569                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2570
2571                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2572                 if (key.offset + num_bytes < new->file_pos)
2573                         goto next;
2574
2575                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2576                 if (!disk_bytenr)
2577                         goto next;
2578
2579                 extent_offset = btrfs_file_extent_offset(l, extent);
2580
2581                 old = kmalloc(sizeof(*old), GFP_NOFS);
2582                 if (!old)
2583                         goto out_free_list;
2584
2585                 offset = max(new->file_pos, key.offset);
2586                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2587
2588                 old->bytenr = disk_bytenr;
2589                 old->extent_offset = extent_offset;
2590                 old->offset = offset - key.offset;
2591                 old->len = end - offset;
2592                 old->new = new;
2593                 old->count = 0;
2594                 list_add_tail(&old->list, &new->head);
2595 next:
2596                 path->slots[0]++;
2597                 cond_resched();
2598         }
2599
2600         btrfs_free_path(path);
2601         atomic_inc(&root->fs_info->defrag_running);
2602
2603         return new;
2604
2605 out_free_list:
2606         list_for_each_entry_safe(old, tmp, &new->head, list) {
2607                 list_del(&old->list);
2608                 kfree(old);
2609         }
2610 out_free_path:
2611         btrfs_free_path(path);
2612 out_kfree:
2613         kfree(new);
2614         return NULL;
2615 }
2616
2617 /*
2618  * helper function for btrfs_finish_ordered_io, this
2619  * just reads in some of the csum leaves to prime them into ram
2620  * before we start the transaction.  It limits the amount of btree
2621  * reads required while inside the transaction.
2622  */
2623 /* as ordered data IO finishes, this gets called so we can finish
2624  * an ordered extent if the range of bytes in the file it covers are
2625  * fully written.
2626  */
2627 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2628 {
2629         struct inode *inode = ordered_extent->inode;
2630         struct btrfs_root *root = BTRFS_I(inode)->root;
2631         struct btrfs_trans_handle *trans = NULL;
2632         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2633         struct extent_state *cached_state = NULL;
2634         struct new_sa_defrag_extent *new = NULL;
2635         int compress_type = 0;
2636         int ret;
2637         bool nolock;
2638
2639         nolock = btrfs_is_free_space_inode(inode);
2640
2641         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2642                 ret = -EIO;
2643                 goto out;
2644         }
2645
2646         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2647                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2648                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2649                 if (nolock)
2650                         trans = btrfs_join_transaction_nolock(root);
2651                 else
2652                         trans = btrfs_join_transaction(root);
2653                 if (IS_ERR(trans)) {
2654                         ret = PTR_ERR(trans);
2655                         trans = NULL;
2656                         goto out;
2657                 }
2658                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2659                 ret = btrfs_update_inode_fallback(trans, root, inode);
2660                 if (ret) /* -ENOMEM or corruption */
2661                         btrfs_abort_transaction(trans, root, ret);
2662                 goto out;
2663         }
2664
2665         lock_extent_bits(io_tree, ordered_extent->file_offset,
2666                          ordered_extent->file_offset + ordered_extent->len - 1,
2667                          0, &cached_state);
2668
2669         ret = test_range_bit(io_tree, ordered_extent->file_offset,
2670                         ordered_extent->file_offset + ordered_extent->len - 1,
2671                         EXTENT_DEFRAG, 1, cached_state);
2672         if (ret) {
2673                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2674                 if (last_snapshot >= BTRFS_I(inode)->generation)
2675                         /* the inode is shared */
2676                         new = record_old_file_extents(inode, ordered_extent);
2677
2678                 clear_extent_bit(io_tree, ordered_extent->file_offset,
2679                         ordered_extent->file_offset + ordered_extent->len - 1,
2680                         EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2681         }
2682
2683         if (nolock)
2684                 trans = btrfs_join_transaction_nolock(root);
2685         else
2686                 trans = btrfs_join_transaction(root);
2687         if (IS_ERR(trans)) {
2688                 ret = PTR_ERR(trans);
2689                 trans = NULL;
2690                 goto out_unlock;
2691         }
2692         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2693
2694         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2695                 compress_type = ordered_extent->compress_type;
2696         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2697                 BUG_ON(compress_type);
2698                 ret = btrfs_mark_extent_written(trans, inode,
2699                                                 ordered_extent->file_offset,
2700                                                 ordered_extent->file_offset +
2701                                                 ordered_extent->len);
2702         } else {
2703                 BUG_ON(root == root->fs_info->tree_root);
2704                 ret = insert_reserved_file_extent(trans, inode,
2705                                                 ordered_extent->file_offset,
2706                                                 ordered_extent->start,
2707                                                 ordered_extent->disk_len,
2708                                                 ordered_extent->len,
2709                                                 ordered_extent->len,
2710                                                 compress_type, 0, 0,
2711                                                 BTRFS_FILE_EXTENT_REG);
2712         }
2713         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2714                            ordered_extent->file_offset, ordered_extent->len,
2715                            trans->transid);
2716         if (ret < 0) {
2717                 btrfs_abort_transaction(trans, root, ret);
2718                 goto out_unlock;
2719         }
2720
2721         add_pending_csums(trans, inode, ordered_extent->file_offset,
2722                           &ordered_extent->list);
2723
2724         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2725         ret = btrfs_update_inode_fallback(trans, root, inode);
2726         if (ret) { /* -ENOMEM or corruption */
2727                 btrfs_abort_transaction(trans, root, ret);
2728                 goto out_unlock;
2729         }
2730         ret = 0;
2731 out_unlock:
2732         unlock_extent_cached(io_tree, ordered_extent->file_offset,
2733                              ordered_extent->file_offset +
2734                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
2735 out:
2736         if (root != root->fs_info->tree_root)
2737                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2738         if (trans)
2739                 btrfs_end_transaction(trans, root);
2740
2741         if (ret) {
2742                 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
2743                                       ordered_extent->file_offset +
2744                                       ordered_extent->len - 1, NULL, GFP_NOFS);
2745
2746                 /*
2747                  * If the ordered extent had an IOERR or something else went
2748                  * wrong we need to return the space for this ordered extent
2749                  * back to the allocator.
2750                  */
2751                 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2752                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2753                         btrfs_free_reserved_extent(root, ordered_extent->start,
2754                                                    ordered_extent->disk_len);
2755         }
2756
2757
2758         /*
2759          * This needs to be done to make sure anybody waiting knows we are done
2760          * updating everything for this ordered extent.
2761          */
2762         btrfs_remove_ordered_extent(inode, ordered_extent);
2763
2764         /* for snapshot-aware defrag */
2765         if (new)
2766                 relink_file_extents(new);
2767
2768         /* once for us */
2769         btrfs_put_ordered_extent(ordered_extent);
2770         /* once for the tree */
2771         btrfs_put_ordered_extent(ordered_extent);
2772
2773         return ret;
2774 }
2775
2776 static void finish_ordered_fn(struct btrfs_work *work)
2777 {
2778         struct btrfs_ordered_extent *ordered_extent;
2779         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2780         btrfs_finish_ordered_io(ordered_extent);
2781 }
2782
2783 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2784                                 struct extent_state *state, int uptodate)
2785 {
2786         struct inode *inode = page->mapping->host;
2787         struct btrfs_root *root = BTRFS_I(inode)->root;
2788         struct btrfs_ordered_extent *ordered_extent = NULL;
2789         struct btrfs_workers *workers;
2790
2791         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2792
2793         ClearPagePrivate2(page);
2794         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2795                                             end - start + 1, uptodate))
2796                 return 0;
2797
2798         ordered_extent->work.func = finish_ordered_fn;
2799         ordered_extent->work.flags = 0;
2800
2801         if (btrfs_is_free_space_inode(inode))
2802                 workers = &root->fs_info->endio_freespace_worker;
2803         else
2804                 workers = &root->fs_info->endio_write_workers;
2805         btrfs_queue_worker(workers, &ordered_extent->work);
2806
2807         return 0;
2808 }
2809
2810 /*
2811  * when reads are done, we need to check csums to verify the data is correct
2812  * if there's a match, we allow the bio to finish.  If not, the code in
2813  * extent_io.c will try to find good copies for us.
2814  */
2815 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2816                                struct extent_state *state, int mirror)
2817 {
2818         size_t offset = start - page_offset(page);
2819         struct inode *inode = page->mapping->host;
2820         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2821         char *kaddr;
2822         u64 private = ~(u32)0;
2823         int ret;
2824         struct btrfs_root *root = BTRFS_I(inode)->root;
2825         u32 csum = ~(u32)0;
2826         static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2827                                       DEFAULT_RATELIMIT_BURST);
2828
2829         if (PageChecked(page)) {
2830                 ClearPageChecked(page);
2831                 goto good;
2832         }
2833
2834         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2835                 goto good;
2836
2837         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2838             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2839                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2840                                   GFP_NOFS);
2841                 return 0;
2842         }
2843
2844         if (state && state->start == start) {
2845                 private = state->private;
2846                 ret = 0;
2847         } else {
2848                 ret = get_state_private(io_tree, start, &private);
2849         }
2850         kaddr = kmap_atomic(page);
2851         if (ret)
2852                 goto zeroit;
2853
2854         csum = btrfs_csum_data(kaddr + offset, csum,  end - start + 1);
2855         btrfs_csum_final(csum, (char *)&csum);
2856         if (csum != private)
2857                 goto zeroit;
2858
2859         kunmap_atomic(kaddr);
2860 good:
2861         return 0;
2862
2863 zeroit:
2864         if (__ratelimit(&_rs))
2865                 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u private %llu",
2866                         (unsigned long long)btrfs_ino(page->mapping->host),
2867                         (unsigned long long)start, csum,
2868                         (unsigned long long)private);
2869         memset(kaddr + offset, 1, end - start + 1);
2870         flush_dcache_page(page);
2871         kunmap_atomic(kaddr);
2872         if (private == 0)
2873                 return 0;
2874         return -EIO;
2875 }
2876
2877 struct delayed_iput {
2878         struct list_head list;
2879         struct inode *inode;
2880 };
2881
2882 /* JDM: If this is fs-wide, why can't we add a pointer to
2883  * btrfs_inode instead and avoid the allocation? */
2884 void btrfs_add_delayed_iput(struct inode *inode)
2885 {
2886         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2887         struct delayed_iput *delayed;
2888
2889         if (atomic_add_unless(&inode->i_count, -1, 1))
2890                 return;
2891
2892         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2893         delayed->inode = inode;
2894
2895         spin_lock(&fs_info->delayed_iput_lock);
2896         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2897         spin_unlock(&fs_info->delayed_iput_lock);
2898 }
2899
2900 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2901 {
2902         LIST_HEAD(list);
2903         struct btrfs_fs_info *fs_info = root->fs_info;
2904         struct delayed_iput *delayed;
2905         int empty;
2906
2907         spin_lock(&fs_info->delayed_iput_lock);
2908         empty = list_empty(&fs_info->delayed_iputs);
2909         spin_unlock(&fs_info->delayed_iput_lock);
2910         if (empty)
2911                 return;
2912
2913         spin_lock(&fs_info->delayed_iput_lock);
2914         list_splice_init(&fs_info->delayed_iputs, &list);
2915         spin_unlock(&fs_info->delayed_iput_lock);
2916
2917         while (!list_empty(&list)) {
2918                 delayed = list_entry(list.next, struct delayed_iput, list);
2919                 list_del(&delayed->list);
2920                 iput(delayed->inode);
2921                 kfree(delayed);
2922         }
2923 }
2924
2925 /*
2926  * This is called in transaction commit time. If there are no orphan
2927  * files in the subvolume, it removes orphan item and frees block_rsv
2928  * structure.
2929  */
2930 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2931                               struct btrfs_root *root)
2932 {
2933         struct btrfs_block_rsv *block_rsv;
2934         int ret;
2935
2936         if (atomic_read(&root->orphan_inodes) ||
2937             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2938                 return;
2939
2940         spin_lock(&root->orphan_lock);
2941         if (atomic_read(&root->orphan_inodes)) {
2942                 spin_unlock(&root->orphan_lock);
2943                 return;
2944         }
2945
2946         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2947                 spin_unlock(&root->orphan_lock);
2948                 return;
2949         }
2950
2951         block_rsv = root->orphan_block_rsv;
2952         root->orphan_block_rsv = NULL;
2953         spin_unlock(&root->orphan_lock);
2954
2955         if (root->orphan_item_inserted &&
2956             btrfs_root_refs(&root->root_item) > 0) {
2957                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2958                                             root->root_key.objectid);
2959                 BUG_ON(ret);
2960                 root->orphan_item_inserted = 0;
2961         }
2962
2963         if (block_rsv) {
2964                 WARN_ON(block_rsv->size > 0);
2965                 btrfs_free_block_rsv(root, block_rsv);
2966         }
2967 }
2968
2969 /*
2970  * This creates an orphan entry for the given inode in case something goes
2971  * wrong in the middle of an unlink/truncate.
2972  *
2973  * NOTE: caller of this function should reserve 5 units of metadata for
2974  *       this function.
2975  */
2976 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2977 {
2978         struct btrfs_root *root = BTRFS_I(inode)->root;
2979         struct btrfs_block_rsv *block_rsv = NULL;
2980         int reserve = 0;
2981         int insert = 0;
2982         int ret;
2983
2984         if (!root->orphan_block_rsv) {
2985                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2986                 if (!block_rsv)
2987                         return -ENOMEM;
2988         }
2989
2990         spin_lock(&root->orphan_lock);
2991         if (!root->orphan_block_rsv) {
2992                 root->orphan_block_rsv = block_rsv;
2993         } else if (block_rsv) {
2994                 btrfs_free_block_rsv(root, block_rsv);
2995                 block_rsv = NULL;
2996         }
2997
2998         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2999                               &BTRFS_I(inode)->runtime_flags)) {
3000 #if 0
3001                 /*
3002                  * For proper ENOSPC handling, we should do orphan
3003                  * cleanup when mounting. But this introduces backward
3004                  * compatibility issue.
3005                  */
3006                 if (!xchg(&root->orphan_item_inserted, 1))
3007                         insert = 2;
3008                 else
3009                         insert = 1;
3010 #endif
3011                 insert = 1;
3012                 atomic_inc(&root->orphan_inodes);
3013         }
3014
3015         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3016                               &BTRFS_I(inode)->runtime_flags))
3017                 reserve = 1;
3018         spin_unlock(&root->orphan_lock);
3019
3020         /* grab metadata reservation from transaction handle */
3021         if (reserve) {
3022                 ret = btrfs_orphan_reserve_metadata(trans, inode);
3023                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3024         }
3025
3026         /* insert an orphan item to track this unlinked/truncated file */
3027         if (insert >= 1) {
3028                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3029                 if (ret && ret != -EEXIST) {
3030                         clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3031                                   &BTRFS_I(inode)->runtime_flags);
3032                         btrfs_abort_transaction(trans, root, ret);
3033                         return ret;
3034                 }
3035                 ret = 0;
3036         }
3037
3038         /* insert an orphan item to track subvolume contains orphan files */
3039         if (insert >= 2) {
3040                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3041                                                root->root_key.objectid);
3042                 if (ret && ret != -EEXIST) {
3043                         btrfs_abort_transaction(trans, root, ret);
3044                         return ret;
3045                 }
3046         }
3047         return 0;
3048 }
3049
3050 /*
3051  * We have done the truncate/delete so we can go ahead and remove the orphan
3052  * item for this particular inode.
3053  */
3054 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3055                             struct inode *inode)
3056 {
3057         struct btrfs_root *root = BTRFS_I(inode)->root;
3058         int delete_item = 0;
3059         int release_rsv = 0;
3060         int ret = 0;
3061
3062         spin_lock(&root->orphan_lock);
3063         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3064                                &BTRFS_I(inode)->runtime_flags))
3065                 delete_item = 1;
3066
3067         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3068                                &BTRFS_I(inode)->runtime_flags))
3069                 release_rsv = 1;
3070         spin_unlock(&root->orphan_lock);
3071
3072         if (trans && delete_item) {
3073                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
3074                 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3075         }
3076
3077         if (release_rsv) {
3078                 btrfs_orphan_release_metadata(inode);
3079                 atomic_dec(&root->orphan_inodes);
3080         }
3081
3082         return 0;
3083 }
3084
3085 /*
3086  * this cleans up any orphans that may be left on the list from the last use
3087  * of this root.
3088  */
3089 int btrfs_orphan_cleanup(struct btrfs_root *root)
3090 {
3091         struct btrfs_path *path;
3092         struct extent_buffer *leaf;
3093         struct btrfs_key key, found_key;
3094         struct btrfs_trans_handle *trans;
3095         struct inode *inode;
3096         u64 last_objectid = 0;
3097         int ret = 0, nr_unlink = 0, nr_truncate = 0;
3098
3099         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3100                 return 0;
3101
3102         path = btrfs_alloc_path();
3103         if (!path) {
3104                 ret = -ENOMEM;
3105                 goto out;
3106         }
3107         path->reada = -1;
3108
3109         key.objectid = BTRFS_ORPHAN_OBJECTID;
3110         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3111         key.offset = (u64)-1;
3112
3113         while (1) {
3114                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3115                 if (ret < 0)
3116                         goto out;
3117
3118                 /*
3119                  * if ret == 0 means we found what we were searching for, which
3120                  * is weird, but possible, so only screw with path if we didn't
3121                  * find the key and see if we have stuff that matches
3122                  */
3123                 if (ret > 0) {
3124                         ret = 0;
3125                         if (path->slots[0] == 0)
3126                                 break;
3127                         path->slots[0]--;
3128                 }
3129
3130                 /* pull out the item */
3131                 leaf = path->nodes[0];
3132                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3133
3134                 /* make sure the item matches what we want */
3135                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3136                         break;
3137                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3138                         break;
3139
3140                 /* release the path since we're done with it */
3141                 btrfs_release_path(path);
3142
3143                 /*
3144                  * this is where we are basically btrfs_lookup, without the
3145                  * crossing root thing.  we store the inode number in the
3146                  * offset of the orphan item.
3147                  */
3148
3149                 if (found_key.offset == last_objectid) {
3150                         btrfs_err(root->fs_info,
3151                                 "Error removing orphan entry, stopping orphan cleanup");
3152                         ret = -EINVAL;
3153                         goto out;
3154                 }
3155
3156                 last_objectid = found_key.offset;
3157
3158                 found_key.objectid = found_key.offset;
3159                 found_key.type = BTRFS_INODE_ITEM_KEY;
3160                 found_key.offset = 0;
3161                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3162                 ret = PTR_RET(inode);
3163                 if (ret && ret != -ESTALE)
3164                         goto out;
3165
3166                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3167                         struct btrfs_root *dead_root;
3168                         struct btrfs_fs_info *fs_info = root->fs_info;
3169                         int is_dead_root = 0;
3170
3171                         /*
3172                          * this is an orphan in the tree root. Currently these
3173                          * could come from 2 sources:
3174                          *  a) a snapshot deletion in progress
3175                          *  b) a free space cache inode
3176                          * We need to distinguish those two, as the snapshot
3177                          * orphan must not get deleted.
3178                          * find_dead_roots already ran before us, so if this
3179                          * is a snapshot deletion, we should find the root
3180                          * in the dead_roots list
3181                          */
3182                         spin_lock(&fs_info->trans_lock);
3183                         list_for_each_entry(dead_root, &fs_info->dead_roots,
3184                                             root_list) {
3185                                 if (dead_root->root_key.objectid ==
3186                                     found_key.objectid) {
3187                                         is_dead_root = 1;
3188                                         break;
3189                                 }
3190                         }
3191                         spin_unlock(&fs_info->trans_lock);
3192                         if (is_dead_root) {
3193                                 /* prevent this orphan from being found again */
3194                                 key.offset = found_key.objectid - 1;
3195                                 continue;
3196                         }
3197                 }
3198                 /*
3199                  * Inode is already gone but the orphan item is still there,
3200                  * kill the orphan item.
3201                  */
3202                 if (ret == -ESTALE) {
3203                         trans = btrfs_start_transaction(root, 1);
3204                         if (IS_ERR(trans)) {
3205                                 ret = PTR_ERR(trans);
3206                                 goto out;
3207                         }
3208                         btrfs_debug(root->fs_info, "auto deleting %Lu",
3209                                 found_key.objectid);
3210                         ret = btrfs_del_orphan_item(trans, root,
3211                                                     found_key.objectid);
3212                         BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3213                         btrfs_end_transaction(trans, root);
3214                         continue;
3215                 }
3216
3217                 /*
3218                  * add this inode to the orphan list so btrfs_orphan_del does
3219                  * the proper thing when we hit it
3220                  */
3221                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3222                         &BTRFS_I(inode)->runtime_flags);
3223                 atomic_inc(&root->orphan_inodes);
3224
3225                 /* if we have links, this was a truncate, lets do that */
3226                 if (inode->i_nlink) {
3227                         if (!S_ISREG(inode->i_mode)) {
3228                                 WARN_ON(1);
3229                                 iput(inode);
3230                                 continue;
3231                         }
3232                         nr_truncate++;
3233
3234                         /* 1 for the orphan item deletion. */
3235                         trans = btrfs_start_transaction(root, 1);
3236                         if (IS_ERR(trans)) {
3237                                 ret = PTR_ERR(trans);
3238                                 goto out;
3239                         }
3240                         ret = btrfs_orphan_add(trans, inode);
3241                         btrfs_end_transaction(trans, root);
3242                         if (ret)
3243                                 goto out;
3244
3245                         ret = btrfs_truncate(inode);
3246                         if (ret)
3247                                 btrfs_orphan_del(NULL, inode);
3248                 } else {
3249                         nr_unlink++;
3250                 }
3251
3252                 /* this will do delete_inode and everything for us */
3253                 iput(inode);
3254                 if (ret)
3255                         goto out;
3256         }
3257         /* release the path since we're done with it */
3258         btrfs_release_path(path);
3259
3260         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3261
3262         if (root->orphan_block_rsv)
3263                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3264                                         (u64)-1);
3265
3266         if (root->orphan_block_rsv || root->orphan_item_inserted) {
3267                 trans = btrfs_join_transaction(root);
3268                 if (!IS_ERR(trans))
3269                         btrfs_end_transaction(trans, root);
3270         }
3271
3272         if (nr_unlink)
3273                 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3274         if (nr_truncate)
3275                 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3276
3277 out:
3278         if (ret)
3279                 btrfs_crit(root->fs_info,
3280                         "could not do orphan cleanup %d", ret);
3281         btrfs_free_path(path);
3282         return ret;
3283 }
3284
3285 /*
3286  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3287  * don't find any xattrs, we know there can't be any acls.
3288  *
3289  * slot is the slot the inode is in, objectid is the objectid of the inode
3290  */
3291 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3292                                           int slot, u64 objectid)
3293 {
3294         u32 nritems = btrfs_header_nritems(leaf);
3295         struct btrfs_key found_key;
3296         int scanned = 0;
3297
3298         slot++;
3299         while (slot < nritems) {
3300                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3301
3302                 /* we found a different objectid, there must not be acls */
3303                 if (found_key.objectid != objectid)
3304                         return 0;
3305
3306                 /* we found an xattr, assume we've got an acl */
3307                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
3308                         return 1;
3309
3310                 /*
3311                  * we found a key greater than an xattr key, there can't
3312                  * be any acls later on
3313                  */
3314                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3315                         return 0;
3316
3317                 slot++;
3318                 scanned++;
3319
3320                 /*
3321                  * it goes inode, inode backrefs, xattrs, extents,
3322                  * so if there are a ton of hard links to an inode there can
3323                  * be a lot of backrefs.  Don't waste time searching too hard,
3324                  * this is just an optimization
3325                  */
3326                 if (scanned >= 8)
3327                         break;
3328         }
3329         /* we hit the end of the leaf before we found an xattr or
3330          * something larger than an xattr.  We have to assume the inode
3331          * has acls
3332          */
3333         return 1;
3334 }
3335
3336 /*
3337  * read an inode from the btree into the in-memory inode
3338  */
3339 static void btrfs_read_locked_inode(struct inode *inode)
3340 {
3341         struct btrfs_path *path;
3342         struct extent_buffer *leaf;
3343         struct btrfs_inode_item *inode_item;
3344         struct btrfs_timespec *tspec;
3345         struct btrfs_root *root = BTRFS_I(inode)->root;
3346         struct btrfs_key location;
3347         int maybe_acls;
3348         u32 rdev;
3349         int ret;
3350         bool filled = false;
3351
3352         ret = btrfs_fill_inode(inode, &rdev);
3353         if (!ret)
3354                 filled = true;
3355
3356         path = btrfs_alloc_path();
3357         if (!path)
3358                 goto make_bad;
3359
3360         path->leave_spinning = 1;
3361         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3362
3363         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3364         if (ret)
3365                 goto make_bad;
3366
3367         leaf = path->nodes[0];
3368
3369         if (filled)
3370                 goto cache_acl;
3371
3372         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3373                                     struct btrfs_inode_item);
3374         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3375         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3376         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3377         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3378         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3379
3380         tspec = btrfs_inode_atime(inode_item);
3381         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3382         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3383
3384         tspec = btrfs_inode_mtime(inode_item);
3385         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3386         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3387
3388         tspec = btrfs_inode_ctime(inode_item);
3389         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3390         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3391
3392         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3393         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3394         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3395
3396         /*
3397          * If we were modified in the current generation and evicted from memory
3398          * and then re-read we need to do a full sync since we don't have any
3399          * idea about which extents were modified before we were evicted from
3400          * cache.
3401          */
3402         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3403                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3404                         &BTRFS_I(inode)->runtime_flags);
3405
3406         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3407         inode->i_generation = BTRFS_I(inode)->generation;
3408         inode->i_rdev = 0;
3409         rdev = btrfs_inode_rdev(leaf, inode_item);
3410
3411         BTRFS_I(inode)->index_cnt = (u64)-1;
3412         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3413 cache_acl:
3414         /*
3415          * try to precache a NULL acl entry for files that don't have
3416          * any xattrs or acls
3417          */
3418         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3419                                            btrfs_ino(inode));
3420         if (!maybe_acls)
3421                 cache_no_acl(inode);
3422
3423         btrfs_free_path(path);
3424
3425         switch (inode->i_mode & S_IFMT) {
3426         case S_IFREG:
3427                 inode->i_mapping->a_ops = &btrfs_aops;
3428                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3429                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3430                 inode->i_fop = &btrfs_file_operations;
3431                 inode->i_op = &btrfs_file_inode_operations;
3432                 break;
3433         case S_IFDIR:
3434                 inode->i_fop = &btrfs_dir_file_operations;
3435                 if (root == root->fs_info->tree_root)
3436                         inode->i_op = &btrfs_dir_ro_inode_operations;
3437                 else
3438                         inode->i_op = &btrfs_dir_inode_operations;
3439                 break;
3440         case S_IFLNK:
3441                 inode->i_op = &btrfs_symlink_inode_operations;
3442                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3443                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3444                 break;
3445         default:
3446                 inode->i_op = &btrfs_special_inode_operations;
3447                 init_special_inode(inode, inode->i_mode, rdev);
3448                 break;
3449         }
3450
3451         btrfs_update_iflags(inode);
3452         return;
3453
3454 make_bad:
3455         btrfs_free_path(path);
3456         make_bad_inode(inode);
3457 }
3458
3459 /*
3460  * given a leaf and an inode, copy the inode fields into the leaf
3461  */
3462 static void fill_inode_item(struct btrfs_trans_handle *trans,
3463                             struct extent_buffer *leaf,
3464                             struct btrfs_inode_item *item,
3465                             struct inode *inode)
3466 {
3467         struct btrfs_map_token token;
3468
3469         btrfs_init_map_token(&token);
3470
3471         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3472         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3473         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3474                                    &token);
3475         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3476         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3477
3478         btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3479                                      inode->i_atime.tv_sec, &token);
3480         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3481                                       inode->i_atime.tv_nsec, &token);
3482
3483         btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3484                                      inode->i_mtime.tv_sec, &token);
3485         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3486                                       inode->i_mtime.tv_nsec, &token);
3487
3488         btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3489                                      inode->i_ctime.tv_sec, &token);
3490         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3491                                       inode->i_ctime.tv_nsec, &token);
3492
3493         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3494                                      &token);
3495         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3496                                          &token);
3497         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3498         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3499         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3500         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3501         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3502 }
3503
3504 /*
3505  * copy everything in the in-memory inode into the btree.
3506  */
3507 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3508                                 struct btrfs_root *root, struct inode *inode)
3509 {
3510         struct btrfs_inode_item *inode_item;
3511         struct btrfs_path *path;
3512         struct extent_buffer *leaf;
3513         int ret;
3514
3515         path = btrfs_alloc_path();
3516         if (!path)
3517                 return -ENOMEM;
3518
3519         path->leave_spinning = 1;
3520         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3521                                  1);
3522         if (ret) {
3523                 if (ret > 0)
3524                         ret = -ENOENT;
3525                 goto failed;
3526         }
3527
3528         btrfs_unlock_up_safe(path, 1);
3529         leaf = path->nodes[0];
3530         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3531                                     struct btrfs_inode_item);
3532
3533         fill_inode_item(trans, leaf, inode_item, inode);
3534         btrfs_mark_buffer_dirty(leaf);
3535         btrfs_set_inode_last_trans(trans, inode);
3536         ret = 0;
3537 failed:
3538         btrfs_free_path(path);
3539         return ret;
3540 }
3541
3542 /*
3543  * copy everything in the in-memory inode into the btree.
3544  */
3545 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3546                                 struct btrfs_root *root, struct inode *inode)
3547 {
3548         int ret;
3549
3550         /*
3551          * If the inode is a free space inode, we can deadlock during commit
3552          * if we put it into the delayed code.
3553          *
3554          * The data relocation inode should also be directly updated
3555          * without delay
3556          */
3557         if (!btrfs_is_free_space_inode(inode)
3558             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3559                 btrfs_update_root_times(trans, root);
3560
3561                 ret = btrfs_delayed_update_inode(trans, root, inode);
3562                 if (!ret)
3563                         btrfs_set_inode_last_trans(trans, inode);
3564                 return ret;
3565         }
3566
3567         return btrfs_update_inode_item(trans, root, inode);
3568 }
3569
3570 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3571                                          struct btrfs_root *root,
3572                                          struct inode *inode)
3573 {
3574         int ret;
3575
3576         ret = btrfs_update_inode(trans, root, inode);
3577         if (ret == -ENOSPC)
3578                 return btrfs_update_inode_item(trans, root, inode);
3579         return ret;
3580 }
3581
3582 /*
3583  * unlink helper that gets used here in inode.c and in the tree logging
3584  * recovery code.  It remove a link in a directory with a given name, and
3585  * also drops the back refs in the inode to the directory
3586  */
3587 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3588                                 struct btrfs_root *root,
3589                                 struct inode *dir, struct inode *inode,
3590                                 const char *name, int name_len)
3591 {
3592         struct btrfs_path *path;
3593         int ret = 0;
3594         struct extent_buffer *leaf;
3595         struct btrfs_dir_item *di;
3596         struct btrfs_key key;
3597         u64 index;
3598         u64 ino = btrfs_ino(inode);
3599         u64 dir_ino = btrfs_ino(dir);
3600
3601         path = btrfs_alloc_path();
3602         if (!path) {
3603                 ret = -ENOMEM;
3604                 goto out;
3605         }
3606
3607         path->leave_spinning = 1;
3608         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3609                                     name, name_len, -1);
3610         if (IS_ERR(di)) {
3611                 ret = PTR_ERR(di);
3612                 goto err;
3613         }
3614         if (!di) {
3615                 ret = -ENOENT;
3616                 goto err;
3617         }
3618         leaf = path->nodes[0];
3619         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3620         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3621         if (ret)
3622                 goto err;
3623         btrfs_release_path(path);
3624
3625         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3626                                   dir_ino, &index);
3627         if (ret) {
3628                 btrfs_info(root->fs_info,
3629                         "failed to delete reference to %.*s, inode %llu parent %llu",
3630                         name_len, name,
3631                         (unsigned long long)ino, (unsigned long long)dir_ino);
3632                 btrfs_abort_transaction(trans, root, ret);
3633                 goto err;
3634         }
3635
3636         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3637         if (ret) {
3638                 btrfs_abort_transaction(trans, root, ret);
3639                 goto err;
3640         }
3641
3642         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3643                                          inode, dir_ino);
3644         if (ret != 0 && ret != -ENOENT) {
3645                 btrfs_abort_transaction(trans, root, ret);
3646                 goto err;
3647         }
3648
3649         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3650                                            dir, index);
3651         if (ret == -ENOENT)
3652                 ret = 0;
3653         else if (ret)
3654                 btrfs_abort_transaction(trans, root, ret);
3655 err:
3656         btrfs_free_path(path);
3657         if (ret)
3658                 goto out;
3659
3660         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3661         inode_inc_iversion(inode);
3662         inode_inc_iversion(dir);
3663         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3664         ret = btrfs_update_inode(trans, root, dir);
3665 out:
3666         return ret;
3667 }
3668
3669 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3670                        struct btrfs_root *root,
3671                        struct inode *dir, struct inode *inode,
3672                        const char *name, int name_len)
3673 {
3674         int ret;
3675         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3676         if (!ret) {
3677                 btrfs_drop_nlink(inode);
3678                 ret = btrfs_update_inode(trans, root, inode);
3679         }
3680         return ret;
3681 }
3682
3683
3684 /* helper to check if there is any shared block in the path */
3685 static int check_path_shared(struct btrfs_root *root,
3686                              struct btrfs_path *path)
3687 {
3688         struct extent_buffer *eb;
3689         int level;
3690         u64 refs = 1;
3691
3692         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
3693                 int ret;
3694
3695                 if (!path->nodes[level])
3696                         break;
3697                 eb = path->nodes[level];
3698                 if (!btrfs_block_can_be_shared(root, eb))
3699                         continue;
3700                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, level, 1,
3701                                                &refs, NULL);
3702                 if (refs > 1)
3703                         return 1;
3704         }
3705         return 0;
3706 }
3707
3708 /*
3709  * helper to start transaction for unlink and rmdir.
3710  *
3711  * unlink and rmdir are special in btrfs, they do not always free space.
3712  * so in enospc case, we should make sure they will free space before
3713  * allowing them to use the global metadata reservation.
3714  */
3715 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
3716                                                        struct dentry *dentry)
3717 {
3718         struct btrfs_trans_handle *trans;
3719         struct btrfs_root *root = BTRFS_I(dir)->root;
3720         struct btrfs_path *path;
3721         struct btrfs_dir_item *di;
3722         struct inode *inode = dentry->d_inode;
3723         u64 index;
3724         int check_link = 1;
3725         int err = -ENOSPC;
3726         int ret;
3727         u64 ino = btrfs_ino(inode);
3728         u64 dir_ino = btrfs_ino(dir);
3729
3730         /*
3731          * 1 for the possible orphan item
3732          * 1 for the dir item
3733          * 1 for the dir index
3734          * 1 for the inode ref
3735          * 1 for the inode
3736          */
3737         trans = btrfs_start_transaction(root, 5);
3738         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3739                 return trans;
3740
3741         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
3742                 return ERR_PTR(-ENOSPC);
3743
3744         /* check if there is someone else holds reference */
3745         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
3746                 return ERR_PTR(-ENOSPC);
3747
3748         if (atomic_read(&inode->i_count) > 2)
3749                 return ERR_PTR(-ENOSPC);
3750
3751         if (xchg(&root->fs_info->enospc_unlink, 1))
3752                 return ERR_PTR(-ENOSPC);
3753
3754         path = btrfs_alloc_path();
3755         if (!path) {
3756                 root->fs_info->enospc_unlink = 0;
3757                 return ERR_PTR(-ENOMEM);
3758         }
3759
3760         /* 1 for the orphan item */
3761         trans = btrfs_start_transaction(root, 1);
3762         if (IS_ERR(trans)) {
3763                 btrfs_free_path(path);
3764                 root->fs_info->enospc_unlink = 0;
3765                 return trans;
3766         }
3767
3768         path->skip_locking = 1;
3769         path->search_commit_root = 1;
3770
3771         ret = btrfs_lookup_inode(trans, root, path,
3772                                 &BTRFS_I(dir)->location, 0);
3773         if (ret < 0) {
3774                 err = ret;
3775                 goto out;
3776         }
3777         if (ret == 0) {
3778                 if (check_path_shared(root, path))
3779                         goto out;
3780         } else {
3781                 check_link = 0;
3782         }
3783         btrfs_release_path(path);
3784
3785         ret = btrfs_lookup_inode(trans, root, path,
3786                                 &BTRFS_I(inode)->location, 0);
3787         if (ret < 0) {
3788                 err = ret;
3789                 goto out;
3790         }
3791         if (ret == 0) {
3792                 if (check_path_shared(root, path))
3793                         goto out;
3794         } else {
3795                 check_link = 0;
3796         }
3797         btrfs_release_path(path);
3798
3799         if (ret == 0 && S_ISREG(inode->i_mode)) {
3800                 ret = btrfs_lookup_file_extent(trans, root, path,
3801                                                ino, (u64)-1, 0);
3802                 if (ret < 0) {
3803                         err = ret;
3804                         goto out;
3805                 }
3806                 BUG_ON(ret == 0); /* Corruption */
3807                 if (check_path_shared(root, path))
3808                         goto out;
3809                 btrfs_release_path(path);
3810         }
3811
3812         if (!check_link) {
3813                 err = 0;
3814                 goto out;
3815         }
3816
3817         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3818                                 dentry->d_name.name, dentry->d_name.len, 0);
3819         if (IS_ERR(di)) {
3820                 err = PTR_ERR(di);
3821                 goto out;
3822         }
3823         if (di) {
3824                 if (check_path_shared(root, path))
3825                         goto out;
3826         } else {
3827                 err = 0;
3828                 goto out;
3829         }
3830         btrfs_release_path(path);
3831
3832         ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3833                                         dentry->d_name.len, ino, dir_ino, 0,
3834                                         &index);
3835         if (ret) {
3836                 err = ret;
3837                 goto out;
3838         }
3839
3840         if (check_path_shared(root, path))
3841                 goto out;
3842
3843         btrfs_release_path(path);
3844
3845         /*
3846          * This is a commit root search, if we can lookup inode item and other
3847          * relative items in the commit root, it means the transaction of
3848          * dir/file creation has been committed, and the dir index item that we
3849          * delay to insert has also been inserted into the commit root. So
3850          * we needn't worry about the delayed insertion of the dir index item
3851          * here.
3852          */
3853         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3854                                 dentry->d_name.name, dentry->d_name.len, 0);
3855         if (IS_ERR(di)) {
3856                 err = PTR_ERR(di);
3857                 goto out;
3858         }
3859         BUG_ON(ret == -ENOENT);
3860         if (check_path_shared(root, path))
3861                 goto out;
3862
3863         err = 0;
3864 out:
3865         btrfs_free_path(path);
3866         /* Migrate the orphan reservation over */
3867         if (!err)
3868                 err = btrfs_block_rsv_migrate(trans->block_rsv,
3869                                 &root->fs_info->global_block_rsv,
3870                                 trans->bytes_reserved);
3871
3872         if (err) {
3873                 btrfs_end_transaction(trans, root);
3874                 root->fs_info->enospc_unlink = 0;
3875                 return ERR_PTR(err);
3876         }
3877
3878         trans->block_rsv = &root->fs_info->global_block_rsv;
3879         return trans;
3880 }
3881
3882 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3883                                struct btrfs_root *root)
3884 {
3885         if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3886                 btrfs_block_rsv_release(root, trans->block_rsv,
3887                                         trans->bytes_reserved);
3888                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3889                 BUG_ON(!root->fs_info->enospc_unlink);
3890                 root->fs_info->enospc_unlink = 0;
3891         }
3892         btrfs_end_transaction(trans, root);
3893 }
3894
3895 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3896 {
3897         struct btrfs_root *root = BTRFS_I(dir)->root;
3898         struct btrfs_trans_handle *trans;
3899         struct inode *inode = dentry->d_inode;
3900         int ret;
3901
3902         trans = __unlink_start_trans(dir, dentry);
3903         if (IS_ERR(trans))
3904                 return PTR_ERR(trans);
3905
3906         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3907
3908         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3909                                  dentry->d_name.name, dentry->d_name.len);
3910         if (ret)
3911                 goto out;
3912
3913         if (inode->i_nlink == 0) {
3914                 ret = btrfs_orphan_add(trans, inode);
3915                 if (ret)
3916                         goto out;
3917         }
3918
3919 out:
3920         __unlink_end_trans(trans, root);
3921         btrfs_btree_balance_dirty(root);
3922         return ret;
3923 }
3924
3925 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3926                         struct btrfs_root *root,
3927                         struct inode *dir, u64 objectid,
3928                         const char *name, int name_len)
3929 {
3930         struct btrfs_path *path;
3931         struct extent_buffer *leaf;
3932         struct btrfs_dir_item *di;
3933         struct btrfs_key key;
3934         u64 index;
3935         int ret;
3936         u64 dir_ino = btrfs_ino(dir);
3937
3938         path = btrfs_alloc_path();
3939         if (!path)
3940                 return -ENOMEM;
3941
3942         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3943                                    name, name_len, -1);
3944         if (IS_ERR_OR_NULL(di)) {
3945                 if (!di)
3946                         ret = -ENOENT;
3947                 else
3948                         ret = PTR_ERR(di);
3949                 goto out;
3950         }
3951
3952         leaf = path->nodes[0];
3953         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3954         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3955         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3956         if (ret) {
3957                 btrfs_abort_transaction(trans, root, ret);
3958                 goto out;
3959         }
3960         btrfs_release_path(path);
3961
3962         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3963                                  objectid, root->root_key.objectid,
3964                                  dir_ino, &index, name, name_len);
3965         if (ret < 0) {
3966                 if (ret != -ENOENT) {
3967                         btrfs_abort_transaction(trans, root, ret);
3968                         goto out;
3969                 }
3970                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3971                                                  name, name_len);
3972                 if (IS_ERR_OR_NULL(di)) {
3973                         if (!di)
3974                                 ret = -ENOENT;
3975                         else
3976                                 ret = PTR_ERR(di);
3977                         btrfs_abort_transaction(trans, root, ret);
3978                         goto out;
3979                 }
3980
3981                 leaf = path->nodes[0];
3982                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3983                 btrfs_release_path(path);
3984                 index = key.offset;
3985         }
3986         btrfs_release_path(path);
3987
3988         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3989         if (ret) {
3990                 btrfs_abort_transaction(trans, root, ret);
3991                 goto out;
3992         }
3993
3994         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3995         inode_inc_iversion(dir);
3996         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3997         ret = btrfs_update_inode_fallback(trans, root, dir);
3998         if (ret)
3999                 btrfs_abort_transaction(trans, root, ret);
4000 out:
4001         btrfs_free_path(path);
4002         return ret;
4003 }
4004
4005 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4006 {
4007         struct inode *inode = dentry->d_inode;
4008         int err = 0;
4009         struct btrfs_root *root = BTRFS_I(dir)->root;
4010         struct btrfs_trans_handle *trans;
4011
4012         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4013                 return -ENOTEMPTY;
4014         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4015                 return -EPERM;
4016
4017         trans = __unlink_start_trans(dir, dentry);
4018         if (IS_ERR(trans))
4019                 return PTR_ERR(trans);
4020
4021         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4022                 err = btrfs_unlink_subvol(trans, root, dir,
4023                                           BTRFS_I(inode)->location.objectid,
4024                                           dentry->d_name.name,
4025                                           dentry->d_name.len);
4026                 goto out;
4027         }
4028
4029         err = btrfs_orphan_add(trans, inode);
4030         if (err)
4031                 goto out;
4032
4033         /* now the directory is empty */
4034         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
4035                                  dentry->d_name.name, dentry->d_name.len);
4036         if (!err)
4037                 btrfs_i_size_write(inode, 0);
4038 out:
4039         __unlink_end_trans(trans, root);
4040         btrfs_btree_balance_dirty(root);
4041
4042         return err;
4043 }
4044
4045 /*
4046  * this can truncate away extent items, csum items and directory items.
4047  * It starts at a high offset and removes keys until it can't find
4048  * any higher than new_size
4049  *
4050  * csum items that cross the new i_size are truncated to the new size
4051  * as well.
4052  *
4053  * min_type is the minimum key type to truncate down to.  If set to 0, this
4054  * will kill all the items on this inode, including the INODE_ITEM_KEY.
4055  */
4056 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4057                                struct btrfs_root *root,
4058                                struct inode *inode,
4059                                u64 new_size, u32 min_type)
4060 {
4061         struct btrfs_path *path;
4062         struct extent_buffer *leaf;
4063         struct btrfs_file_extent_item *fi;
4064         struct btrfs_key key;
4065         struct btrfs_key found_key;
4066         u64 extent_start = 0;
4067         u64 extent_num_bytes = 0;
4068         u64 extent_offset = 0;
4069         u64 item_end = 0;
4070         u32 found_type = (u8)-1;
4071         int found_extent;
4072         int del_item;
4073         int pending_del_nr = 0;
4074         int pending_del_slot = 0;
4075         int extent_type = -1;
4076         int ret;
4077         int err = 0;
4078         u64 ino = btrfs_ino(inode);
4079
4080         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4081
4082         path = btrfs_alloc_path();
4083         if (!path)
4084                 return -ENOMEM;
4085         path->reada = -1;
4086
4087         /*
4088          * We want to drop from the next block forward in case this new size is
4089          * not block aligned since we will be keeping the last block of the
4090          * extent just the way it is.
4091          */
4092         if (root->ref_cows || root == root->fs_info->tree_root)
4093                 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4094                                         root->sectorsize), (u64)-1, 0);
4095
4096         /*
4097          * This function is also used to drop the items in the log tree before
4098          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4099          * it is used to drop the loged items. So we shouldn't kill the delayed
4100          * items.
4101          */
4102         if (min_type == 0 && root == BTRFS_I(inode)->root)
4103                 btrfs_kill_delayed_inode_items(inode);
4104
4105         key.objectid = ino;
4106         key.offset = (u64)-1;
4107         key.type = (u8)-1;
4108
4109 search_again:
4110         path->leave_spinning = 1;
4111         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4112         if (ret < 0) {
4113                 err = ret;
4114                 goto out;
4115         }
4116
4117         if (ret > 0) {
4118                 /* there are no items in the tree for us to truncate, we're
4119                  * done
4120                  */
4121                 if (path->slots[0] == 0)
4122                         goto out;
4123                 path->slots[0]--;
4124         }
4125
4126         while (1) {
4127                 fi = NULL;
4128                 leaf = path->nodes[0];
4129                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4130                 found_type = btrfs_key_type(&found_key);
4131
4132                 if (found_key.objectid != ino)
4133                         break;
4134
4135                 if (found_type < min_type)
4136                         break;
4137
4138                 item_end = found_key.offset;
4139                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4140                         fi = btrfs_item_ptr(leaf, path->slots[0],
4141                                             struct btrfs_file_extent_item);
4142                         extent_type = btrfs_file_extent_type(leaf, fi);
4143                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4144                                 item_end +=
4145                                     btrfs_file_extent_num_bytes(leaf, fi);
4146                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4147                                 item_end += btrfs_file_extent_inline_len(leaf,
4148                                                                          fi);
4149                         }
4150                         item_end--;
4151                 }
4152                 if (found_type > min_type) {
4153                         del_item = 1;
4154                 } else {
4155                         if (item_end < new_size)
4156                                 break;
4157                         if (found_key.offset >= new_size)
4158                                 del_item = 1;
4159                         else
4160                                 del_item = 0;
4161                 }
4162                 found_extent = 0;
4163                 /* FIXME, shrink the extent if the ref count is only 1 */
4164                 if (found_type != BTRFS_EXTENT_DATA_KEY)
4165                         goto delete;
4166
4167                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4168                         u64 num_dec;
4169                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4170                         if (!del_item) {
4171                                 u64 orig_num_bytes =
4172                                         btrfs_file_extent_num_bytes(leaf, fi);
4173                                 extent_num_bytes = ALIGN(new_size -
4174                                                 found_key.offset,
4175                                                 root->sectorsize);
4176                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4177                                                          extent_num_bytes);
4178                                 num_dec = (orig_num_bytes -
4179                                            extent_num_bytes);
4180                                 if (root->ref_cows && extent_start != 0)
4181                                         inode_sub_bytes(inode, num_dec);
4182                                 btrfs_mark_buffer_dirty(leaf);
4183                         } else {
4184                                 extent_num_bytes =
4185                                         btrfs_file_extent_disk_num_bytes(leaf,
4186                                                                          fi);
4187                                 extent_offset = found_key.offset -
4188                                         btrfs_file_extent_offset(leaf, fi);
4189
4190                                 /* FIXME blocksize != 4096 */
4191                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4192                                 if (extent_start != 0) {
4193                                         found_extent = 1;
4194                                         if (root->ref_cows)
4195                                                 inode_sub_bytes(inode, num_dec);
4196                                 }
4197                         }
4198                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4199                         /*
4200                          * we can't truncate inline items that have had
4201                          * special encodings
4202                          */
4203                         if (!del_item &&
4204                             btrfs_file_extent_compression(leaf, fi) == 0 &&
4205                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
4206                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4207                                 u32 size = new_size - found_key.offset;
4208
4209                                 if (root->ref_cows) {
4210                                         inode_sub_bytes(inode, item_end + 1 -
4211                                                         new_size);
4212                                 }
4213                                 size =
4214                                     btrfs_file_extent_calc_inline_size(size);
4215                                 btrfs_truncate_item(root, path, size, 1);
4216                         } else if (root->ref_cows) {
4217                                 inode_sub_bytes(inode, item_end + 1 -
4218                                                 found_key.offset);
4219                         }
4220                 }
4221 delete:
4222                 if (del_item) {
4223                         if (!pending_del_nr) {
4224                                 /* no pending yet, add ourselves */
4225                                 pending_del_slot = path->slots[0];
4226                                 pending_del_nr = 1;
4227                         } else if (pending_del_nr &&
4228                                    path->slots[0] + 1 == pending_del_slot) {
4229                                 /* hop on the pending chunk */
4230                                 pending_del_nr++;
4231                                 pending_del_slot = path->slots[0];
4232                         } else {
4233                                 BUG();
4234                         }
4235                 } else {
4236                         break;
4237                 }
4238                 if (found_extent && (root->ref_cows ||
4239                                      root == root->fs_info->tree_root)) {
4240                         btrfs_set_path_blocking(path);
4241                         ret = btrfs_free_extent(trans, root, extent_start,
4242                                                 extent_num_bytes, 0,
4243                                                 btrfs_header_owner(leaf),
4244                                                 ino, extent_offset, 0);
4245                         BUG_ON(ret);
4246                 }
4247
4248                 if (found_type == BTRFS_INODE_ITEM_KEY)
4249                         break;
4250
4251                 if (path->slots[0] == 0 ||
4252                     path->slots[0] != pending_del_slot) {
4253                         if (pending_del_nr) {
4254                                 ret = btrfs_del_items(trans, root, path,
4255                                                 pending_del_slot,
4256                                                 pending_del_nr);
4257                                 if (ret) {
4258                                         btrfs_abort_transaction(trans,
4259                                                                 root, ret);
4260                                         goto error;
4261                                 }
4262                                 pending_del_nr = 0;
4263                         }
4264                         btrfs_release_path(path);
4265                         goto search_again;
4266                 } else {
4267                         path->slots[0]--;
4268                 }
4269         }
4270 out:
4271         if (pending_del_nr) {
4272                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4273                                       pending_del_nr);
4274                 if (ret)
4275                         btrfs_abort_transaction(trans, root, ret);
4276         }
4277 error:
4278         btrfs_free_path(path);
4279         return err;
4280 }
4281
4282 /*
4283  * btrfs_truncate_page - read, zero a chunk and write a page
4284  * @inode - inode that we're zeroing
4285  * @from - the offset to start zeroing
4286  * @len - the length to zero, 0 to zero the entire range respective to the
4287  *      offset
4288  * @front - zero up to the offset instead of from the offset on
4289  *
4290  * This will find the page for the "from" offset and cow the page and zero the
4291  * part we want to zero.  This is used with truncate and hole punching.
4292  */
4293 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4294                         int front)
4295 {
4296         struct address_space *mapping = inode->i_mapping;
4297         struct btrfs_root *root = BTRFS_I(inode)->root;
4298         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4299         struct btrfs_ordered_extent *ordered;
4300         struct extent_state *cached_state = NULL;
4301         char *kaddr;
4302         u32 blocksize = root->sectorsize;
4303         pgoff_t index = from >> PAGE_CACHE_SHIFT;
4304         unsigned offset = from & (PAGE_CACHE_SIZE-1);
4305         struct page *page;
4306         gfp_t mask = btrfs_alloc_write_mask(mapping);
4307         int ret = 0;
4308         u64 page_start;
4309         u64 page_end;
4310
4311         if ((offset & (blocksize - 1)) == 0 &&
4312             (!len || ((len & (blocksize - 1)) == 0)))
4313                 goto out;
4314         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4315         if (ret)
4316                 goto out;
4317
4318 again:
4319         page = find_or_create_page(mapping, index, mask);
4320         if (!page) {
4321                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4322                 ret = -ENOMEM;
4323                 goto out;
4324         }
4325
4326         page_start = page_offset(page);
4327         page_end = page_start + PAGE_CACHE_SIZE - 1;
4328
4329         if (!PageUptodate(page)) {
4330                 ret = btrfs_readpage(NULL, page);
4331                 lock_page(page);
4332                 if (page->mapping != mapping) {
4333                         unlock_page(page);
4334                         page_cache_release(page);
4335                         goto again;
4336                 }
4337                 if (!PageUptodate(page)) {
4338                         ret = -EIO;
4339                         goto out_unlock;
4340                 }
4341         }
4342         wait_on_page_writeback(page);
4343
4344         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4345         set_page_extent_mapped(page);
4346
4347         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4348         if (ordered) {
4349                 unlock_extent_cached(io_tree, page_start, page_end,
4350                                      &cached_state, GFP_NOFS);
4351                 unlock_page(page);
4352                 page_cache_release(page);
4353                 btrfs_start_ordered_extent(inode, ordered, 1);
4354                 btrfs_put_ordered_extent(ordered);
4355                 goto again;
4356         }
4357
4358         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4359                           EXTENT_DIRTY | EXTENT_DELALLOC |
4360                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4361                           0, 0, &cached_state, GFP_NOFS);
4362
4363         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4364                                         &cached_state);
4365         if (ret) {
4366                 unlock_extent_cached(io_tree, page_start, page_end,
4367                                      &cached_state, GFP_NOFS);
4368                 goto out_unlock;
4369         }
4370
4371         if (offset != PAGE_CACHE_SIZE) {
4372                 if (!len)
4373                         len = PAGE_CACHE_SIZE - offset;
4374                 kaddr = kmap(page);
4375                 if (front)
4376                         memset(kaddr, 0, offset);
4377                 else
4378                         memset(kaddr + offset, 0, len);
4379                 flush_dcache_page(page);
4380                 kunmap(page);
4381         }
4382         ClearPageChecked(page);
4383         set_page_dirty(page);
4384         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4385                              GFP_NOFS);
4386
4387 out_unlock:
4388         if (ret)
4389                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4390         unlock_page(page);
4391         page_cache_release(page);
4392 out:
4393         return ret;
4394 }
4395
4396 /*
4397  * This function puts in dummy file extents for the area we're creating a hole
4398  * for.  So if we are truncating this file to a larger size we need to insert
4399  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4400  * the range between oldsize and size
4401  */
4402 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4403 {
4404         struct btrfs_trans_handle *trans;
4405         struct btrfs_root *root = BTRFS_I(inode)->root;
4406         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4407         struct extent_map *em = NULL;
4408         struct extent_state *cached_state = NULL;
4409         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4410         u64 hole_start = ALIGN(oldsize, root->sectorsize);
4411         u64 block_end = ALIGN(size, root->sectorsize);
4412         u64 last_byte;
4413         u64 cur_offset;
4414         u64 hole_size;
4415         int err = 0;
4416
4417         if (size <= hole_start)
4418                 return 0;
4419
4420         while (1) {
4421                 struct btrfs_ordered_extent *ordered;
4422                 btrfs_wait_ordered_range(inode, hole_start,
4423                                          block_end - hole_start);
4424                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4425                                  &cached_state);
4426                 ordered = btrfs_lookup_ordered_extent(inode, 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_put_ordered_extent(ordered);
4432         }
4433
4434         cur_offset = hole_start;
4435         while (1) {
4436                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4437                                 block_end - cur_offset, 0);
4438                 if (IS_ERR(em)) {
4439                         err = PTR_ERR(em);
4440                         em = NULL;
4441                         break;
4442                 }
4443                 last_byte = min(extent_map_end(em), block_end);
4444                 last_byte = ALIGN(last_byte , root->sectorsize);
4445                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4446                         struct extent_map *hole_em;
4447                         hole_size = last_byte - cur_offset;
4448
4449                         trans = btrfs_start_transaction(root, 3);
4450                         if (IS_ERR(trans)) {
4451                                 err = PTR_ERR(trans);
4452                                 break;
4453                         }
4454
4455                         err = btrfs_drop_extents(trans, root, inode,
4456                                                  cur_offset,
4457                                                  cur_offset + hole_size, 1);
4458                         if (err) {
4459                                 btrfs_abort_transaction(trans, root, err);
4460                                 btrfs_end_transaction(trans, root);
4461                                 break;
4462                         }
4463
4464                         err = btrfs_insert_file_extent(trans, root,
4465                                         btrfs_ino(inode), cur_offset, 0,
4466                                         0, hole_size, 0, hole_size,
4467                                         0, 0, 0);
4468                         if (err) {
4469                                 btrfs_abort_transaction(trans, root, err);
4470                                 btrfs_end_transaction(trans, root);
4471                                 break;
4472                         }
4473
4474                         btrfs_drop_extent_cache(inode, cur_offset,
4475                                                 cur_offset + hole_size - 1, 0);
4476                         hole_em = alloc_extent_map();
4477                         if (!hole_em) {
4478                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4479                                         &BTRFS_I(inode)->runtime_flags);
4480                                 goto next;
4481                         }
4482                         hole_em->start = cur_offset;
4483                         hole_em->len = hole_size;
4484                         hole_em->orig_start = cur_offset;
4485
4486                         hole_em->block_start = EXTENT_MAP_HOLE;
4487                         hole_em->block_len = 0;
4488                         hole_em->orig_block_len = 0;
4489                         hole_em->ram_bytes = hole_size;
4490                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4491                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
4492                         hole_em->generation = trans->transid;
4493
4494                         while (1) {
4495                                 write_lock(&em_tree->lock);
4496                                 err = add_extent_mapping(em_tree, hole_em, 1);
4497                                 write_unlock(&em_tree->lock);
4498                                 if (err != -EEXIST)
4499                                         break;
4500                                 btrfs_drop_extent_cache(inode, cur_offset,
4501                                                         cur_offset +
4502                                                         hole_size - 1, 0);
4503                         }
4504                         free_extent_map(hole_em);
4505 next:
4506                         btrfs_update_inode(trans, root, inode);
4507                         btrfs_end_transaction(trans, root);
4508                 }
4509                 free_extent_map(em);
4510                 em = NULL;
4511                 cur_offset = last_byte;
4512                 if (cur_offset >= block_end)
4513                         break;
4514         }
4515
4516         free_extent_map(em);
4517         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4518                              GFP_NOFS);
4519         return err;
4520 }
4521
4522 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4523 {
4524         struct btrfs_root *root = BTRFS_I(inode)->root;
4525         struct btrfs_trans_handle *trans;
4526         loff_t oldsize = i_size_read(inode);
4527         loff_t newsize = attr->ia_size;
4528         int mask = attr->ia_valid;
4529         int ret;
4530
4531         if (newsize == oldsize)
4532                 return 0;
4533
4534         /*
4535          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4536          * special case where we need to update the times despite not having
4537          * these flags set.  For all other operations the VFS set these flags
4538          * explicitly if it wants a timestamp update.
4539          */
4540         if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4541                 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4542
4543         if (newsize > oldsize) {
4544                 truncate_pagecache(inode, oldsize, newsize);
4545                 ret = btrfs_cont_expand(inode, oldsize, newsize);
4546                 if (ret)
4547                         return ret;
4548
4549                 trans = btrfs_start_transaction(root, 1);
4550                 if (IS_ERR(trans))
4551                         return PTR_ERR(trans);
4552
4553                 i_size_write(inode, newsize);
4554                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4555                 ret = btrfs_update_inode(trans, root, inode);
4556                 btrfs_end_transaction(trans, root);
4557         } else {
4558
4559                 /*
4560                  * We're truncating a file that used to have good data down to
4561                  * zero. Make sure it gets into the ordered flush list so that
4562                  * any new writes get down to disk quickly.
4563                  */
4564                 if (newsize == 0)
4565                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4566                                 &BTRFS_I(inode)->runtime_flags);
4567
4568                 /*
4569                  * 1 for the orphan item we're going to add
4570                  * 1 for the orphan item deletion.
4571                  */
4572                 trans = btrfs_start_transaction(root, 2);
4573                 if (IS_ERR(trans))
4574                         return PTR_ERR(trans);
4575
4576                 /*
4577                  * We need to do this in case we fail at _any_ point during the
4578                  * actual truncate.  Once we do the truncate_setsize we could
4579                  * invalidate pages which forces any outstanding ordered io to
4580                  * be instantly completed which will give us extents that need
4581                  * to be truncated.  If we fail to get an orphan inode down we
4582                  * could have left over extents that were never meant to live,
4583                  * so we need to garuntee from this point on that everything
4584                  * will be consistent.
4585                  */
4586                 ret = btrfs_orphan_add(trans, inode);
4587                 btrfs_end_transaction(trans, root);
4588                 if (ret)
4589                         return ret;
4590
4591                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4592                 truncate_setsize(inode, newsize);
4593
4594                 /* Disable nonlocked read DIO to avoid the end less truncate */
4595                 btrfs_inode_block_unlocked_dio(inode);
4596                 inode_dio_wait(inode);
4597                 btrfs_inode_resume_unlocked_dio(inode);
4598
4599                 ret = btrfs_truncate(inode);
4600                 if (ret && inode->i_nlink)
4601                         btrfs_orphan_del(NULL, inode);
4602         }
4603
4604         return ret;
4605 }
4606
4607 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4608 {
4609         struct inode *inode = dentry->d_inode;
4610         struct btrfs_root *root = BTRFS_I(inode)->root;
4611         int err;
4612
4613         if (btrfs_root_readonly(root))
4614                 return -EROFS;
4615
4616         err = inode_change_ok(inode, attr);
4617         if (err)
4618                 return err;
4619
4620         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4621                 err = btrfs_setsize(inode, attr);
4622                 if (err)
4623                         return err;
4624         }
4625
4626         if (attr->ia_valid) {
4627                 setattr_copy(inode, attr);
4628                 inode_inc_iversion(inode);
4629                 err = btrfs_dirty_inode(inode);
4630
4631                 if (!err && attr->ia_valid & ATTR_MODE)
4632                         err = btrfs_acl_chmod(inode);
4633         }
4634
4635         return err;
4636 }
4637
4638 void btrfs_evict_inode(struct inode *inode)
4639 {
4640         struct btrfs_trans_handle *trans;
4641         struct btrfs_root *root = BTRFS_I(inode)->root;
4642         struct btrfs_block_rsv *rsv, *global_rsv;
4643         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4644         int ret;
4645
4646         trace_btrfs_inode_evict(inode);
4647
4648         truncate_inode_pages(&inode->i_data, 0);
4649         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
4650                                btrfs_is_free_space_inode(inode)))
4651                 goto no_delete;
4652
4653         if (is_bad_inode(inode)) {
4654                 btrfs_orphan_del(NULL, inode);
4655                 goto no_delete;
4656         }
4657         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4658         btrfs_wait_ordered_range(inode, 0, (u64)-1);
4659
4660         if (root->fs_info->log_root_recovering) {
4661                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4662                                  &BTRFS_I(inode)->runtime_flags));
4663                 goto no_delete;
4664         }
4665
4666         if (inode->i_nlink > 0) {
4667                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
4668                 goto no_delete;
4669         }
4670
4671         ret = btrfs_commit_inode_delayed_inode(inode);
4672         if (ret) {
4673                 btrfs_orphan_del(NULL, inode);
4674                 goto no_delete;
4675         }
4676
4677         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4678         if (!rsv) {
4679                 btrfs_orphan_del(NULL, inode);
4680                 goto no_delete;
4681         }
4682         rsv->size = min_size;
4683         rsv->failfast = 1;
4684         global_rsv = &root->fs_info->global_block_rsv;
4685
4686         btrfs_i_size_write(inode, 0);
4687
4688         /*
4689          * This is a bit simpler than btrfs_truncate since we've already
4690          * reserved our space for our orphan item in the unlink, so we just
4691          * need to reserve some slack space in case we add bytes and update
4692          * inode item when doing the truncate.
4693          */
4694         while (1) {
4695                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4696                                              BTRFS_RESERVE_FLUSH_LIMIT);
4697
4698                 /*
4699                  * Try and steal from the global reserve since we will
4700                  * likely not use this space anyway, we want to try as
4701                  * hard as possible to get this to work.
4702                  */
4703                 if (ret)
4704                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4705
4706                 if (ret) {
4707                         btrfs_warn(root->fs_info,
4708                                 "Could not get space for a delete, will truncate on mount %d",
4709                                 ret);
4710                         btrfs_orphan_del(NULL, inode);
4711                         btrfs_free_block_rsv(root, rsv);
4712                         goto no_delete;
4713                 }
4714
4715                 trans = btrfs_join_transaction(root);
4716                 if (IS_ERR(trans)) {
4717                         btrfs_orphan_del(NULL, inode);
4718                         btrfs_free_block_rsv(root, rsv);
4719                         goto no_delete;
4720                 }
4721
4722                 trans->block_rsv = rsv;
4723
4724                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4725                 if (ret != -ENOSPC)
4726                         break;
4727
4728                 trans->block_rsv = &root->fs_info->trans_block_rsv;
4729                 btrfs_end_transaction(trans, root);
4730                 trans = NULL;
4731                 btrfs_btree_balance_dirty(root);
4732         }
4733
4734         btrfs_free_block_rsv(root, rsv);
4735
4736         if (ret == 0) {
4737                 trans->block_rsv = root->orphan_block_rsv;
4738                 ret = btrfs_orphan_del(trans, inode);
4739                 BUG_ON(ret);
4740         }
4741
4742         trans->block_rsv = &root->fs_info->trans_block_rsv;
4743         if (!(root == root->fs_info->tree_root ||
4744               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4745                 btrfs_return_ino(root, btrfs_ino(inode));
4746
4747         btrfs_end_transaction(trans, root);
4748         btrfs_btree_balance_dirty(root);
4749 no_delete:
4750         btrfs_remove_delayed_node(inode);
4751         clear_inode(inode);
4752         return;
4753 }
4754
4755 /*
4756  * this returns the key found in the dir entry in the location pointer.
4757  * If no dir entries were found, location->objectid is 0.
4758  */
4759 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4760                                struct btrfs_key *location)
4761 {
4762         const char *name = dentry->d_name.name;
4763         int namelen = dentry->d_name.len;
4764         struct btrfs_dir_item *di;
4765         struct btrfs_path *path;
4766         struct btrfs_root *root = BTRFS_I(dir)->root;
4767         int ret = 0;
4768
4769         path = btrfs_alloc_path();
4770         if (!path)
4771                 return -ENOMEM;
4772
4773         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4774                                     namelen, 0);
4775         if (IS_ERR(di))
4776                 ret = PTR_ERR(di);
4777
4778         if (IS_ERR_OR_NULL(di))
4779                 goto out_err;
4780
4781         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4782 out:
4783         btrfs_free_path(path);
4784         return ret;
4785 out_err:
4786         location->objectid = 0;
4787         goto out;
4788 }
4789
4790 /*
4791  * when we hit a tree root in a directory, the btrfs part of the inode
4792  * needs to be changed to reflect the root directory of the tree root.  This
4793  * is kind of like crossing a mount point.
4794  */
4795 static int fixup_tree_root_location(struct btrfs_root *root,
4796                                     struct inode *dir,
4797                                     struct dentry *dentry,
4798                                     struct btrfs_key *location,
4799                                     struct btrfs_root **sub_root)
4800 {
4801         struct btrfs_path *path;
4802         struct btrfs_root *new_root;
4803         struct btrfs_root_ref *ref;
4804         struct extent_buffer *leaf;
4805         int ret;
4806         int err = 0;
4807
4808         path = btrfs_alloc_path();
4809         if (!path) {
4810                 err = -ENOMEM;
4811                 goto out;
4812         }
4813
4814         err = -ENOENT;
4815         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4816                                   BTRFS_I(dir)->root->root_key.objectid,
4817                                   location->objectid);
4818         if (ret) {
4819                 if (ret < 0)
4820                         err = ret;
4821                 goto out;
4822         }
4823
4824         leaf = path->nodes[0];
4825         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4826         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4827             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4828                 goto out;
4829
4830         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4831                                    (unsigned long)(ref + 1),
4832                                    dentry->d_name.len);
4833         if (ret)
4834                 goto out;
4835
4836         btrfs_release_path(path);
4837
4838         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4839         if (IS_ERR(new_root)) {
4840                 err = PTR_ERR(new_root);
4841                 goto out;
4842         }
4843
4844         *sub_root = new_root;
4845         location->objectid = btrfs_root_dirid(&new_root->root_item);
4846         location->type = BTRFS_INODE_ITEM_KEY;
4847         location->offset = 0;
4848         err = 0;
4849 out:
4850         btrfs_free_path(path);
4851         return err;
4852 }
4853
4854 static void inode_tree_add(struct inode *inode)
4855 {
4856         struct btrfs_root *root = BTRFS_I(inode)->root;
4857         struct btrfs_inode *entry;
4858         struct rb_node **p;
4859         struct rb_node *parent;
4860         u64 ino = btrfs_ino(inode);
4861
4862         if (inode_unhashed(inode))
4863                 return;
4864 again:
4865         parent = NULL;
4866         spin_lock(&root->inode_lock);
4867         p = &root->inode_tree.rb_node;
4868         while (*p) {
4869                 parent = *p;
4870                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4871
4872                 if (ino < btrfs_ino(&entry->vfs_inode))
4873                         p = &parent->rb_left;
4874                 else if (ino > btrfs_ino(&entry->vfs_inode))
4875                         p = &parent->rb_right;
4876                 else {
4877                         WARN_ON(!(entry->vfs_inode.i_state &
4878                                   (I_WILL_FREE | I_FREEING)));
4879                         rb_erase(parent, &root->inode_tree);
4880                         RB_CLEAR_NODE(parent);
4881                         spin_unlock(&root->inode_lock);
4882                         goto again;
4883                 }
4884         }
4885         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4886         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4887         spin_unlock(&root->inode_lock);
4888 }
4889
4890 static void inode_tree_del(struct inode *inode)
4891 {
4892         struct btrfs_root *root = BTRFS_I(inode)->root;
4893         int empty = 0;
4894
4895         spin_lock(&root->inode_lock);
4896         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4897                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4898                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4899                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4900         }
4901         spin_unlock(&root->inode_lock);
4902
4903         /*
4904          * Free space cache has inodes in the tree root, but the tree root has a
4905          * root_refs of 0, so this could end up dropping the tree root as a
4906          * snapshot, so we need the extra !root->fs_info->tree_root check to
4907          * make sure we don't drop it.
4908          */
4909         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4910             root != root->fs_info->tree_root) {
4911                 synchronize_srcu(&root->fs_info->subvol_srcu);
4912                 spin_lock(&root->inode_lock);
4913                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4914                 spin_unlock(&root->inode_lock);
4915                 if (empty)
4916                         btrfs_add_dead_root(root);
4917         }
4918 }
4919
4920 void btrfs_invalidate_inodes(struct btrfs_root *root)
4921 {
4922         struct rb_node *node;
4923         struct rb_node *prev;
4924         struct btrfs_inode *entry;
4925         struct inode *inode;
4926         u64 objectid = 0;
4927
4928         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4929
4930         spin_lock(&root->inode_lock);
4931 again:
4932         node = root->inode_tree.rb_node;
4933         prev = NULL;
4934         while (node) {
4935                 prev = node;
4936                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4937
4938                 if (objectid < btrfs_ino(&entry->vfs_inode))
4939                         node = node->rb_left;
4940                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4941                         node = node->rb_right;
4942                 else
4943                         break;
4944         }
4945         if (!node) {
4946                 while (prev) {
4947                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4948                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4949                                 node = prev;
4950                                 break;
4951                         }
4952                         prev = rb_next(prev);
4953                 }
4954         }
4955         while (node) {
4956                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4957                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4958                 inode = igrab(&entry->vfs_inode);
4959                 if (inode) {
4960                         spin_unlock(&root->inode_lock);
4961                         if (atomic_read(&inode->i_count) > 1)
4962                                 d_prune_aliases(inode);
4963                         /*
4964                          * btrfs_drop_inode will have it removed from
4965                          * the inode cache when its usage count
4966                          * hits zero.
4967                          */
4968                         iput(inode);
4969                         cond_resched();
4970                         spin_lock(&root->inode_lock);
4971                         goto again;
4972                 }
4973
4974                 if (cond_resched_lock(&root->inode_lock))
4975                         goto again;
4976
4977                 node = rb_next(node);
4978         }
4979         spin_unlock(&root->inode_lock);
4980 }
4981
4982 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4983 {
4984         struct btrfs_iget_args *args = p;
4985         inode->i_ino = args->ino;
4986         BTRFS_I(inode)->root = args->root;
4987         return 0;
4988 }
4989
4990 static int btrfs_find_actor(struct inode *inode, void *opaque)
4991 {
4992         struct btrfs_iget_args *args = opaque;
4993         return args->ino == btrfs_ino(inode) &&
4994                 args->root == BTRFS_I(inode)->root;
4995 }
4996
4997 static struct inode *btrfs_iget_locked(struct super_block *s,
4998                                        u64 objectid,
4999                                        struct btrfs_root *root)
5000 {
5001         struct inode *inode;
5002         struct btrfs_iget_args args;
5003         args.ino = objectid;
5004         args.root = root;
5005
5006         inode = iget5_locked(s, objectid, btrfs_find_actor,
5007                              btrfs_init_locked_inode,
5008                              (void *)&args);
5009         return inode;
5010 }
5011
5012 /* Get an inode object given its location and corresponding root.
5013  * Returns in *is_new if the inode was read from disk
5014  */
5015 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5016                          struct btrfs_root *root, int *new)
5017 {
5018         struct inode *inode;
5019
5020         inode = btrfs_iget_locked(s, location->objectid, root);
5021         if (!inode)
5022                 return ERR_PTR(-ENOMEM);
5023
5024         if (inode->i_state & I_NEW) {
5025                 BTRFS_I(inode)->root = root;
5026                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
5027                 btrfs_read_locked_inode(inode);
5028                 if (!is_bad_inode(inode)) {
5029                         inode_tree_add(inode);
5030                         unlock_new_inode(inode);
5031                         if (new)
5032                                 *new = 1;
5033                 } else {
5034                         unlock_new_inode(inode);
5035                         iput(inode);
5036                         inode = ERR_PTR(-ESTALE);
5037                 }
5038         }
5039
5040         return inode;
5041 }
5042
5043 static struct inode *new_simple_dir(struct super_block *s,
5044                                     struct btrfs_key *key,
5045                                     struct btrfs_root *root)
5046 {
5047         struct inode *inode = new_inode(s);
5048
5049         if (!inode)
5050                 return ERR_PTR(-ENOMEM);
5051
5052         BTRFS_I(inode)->root = root;
5053         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5054         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5055
5056         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5057         inode->i_op = &btrfs_dir_ro_inode_operations;
5058         inode->i_fop = &simple_dir_operations;
5059         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5060         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5061
5062         return inode;
5063 }
5064
5065 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5066 {
5067         struct inode *inode;
5068         struct btrfs_root *root = BTRFS_I(dir)->root;
5069         struct btrfs_root *sub_root = root;
5070         struct btrfs_key location;
5071         int index;
5072         int ret = 0;
5073
5074         if (dentry->d_name.len > BTRFS_NAME_LEN)
5075                 return ERR_PTR(-ENAMETOOLONG);
5076
5077         ret = btrfs_inode_by_name(dir, dentry, &location);
5078         if (ret < 0)
5079                 return ERR_PTR(ret);
5080
5081         if (location.objectid == 0)
5082                 return NULL;
5083
5084         if (location.type == BTRFS_INODE_ITEM_KEY) {
5085                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5086                 return inode;
5087         }
5088
5089         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5090
5091         index = srcu_read_lock(&root->fs_info->subvol_srcu);
5092         ret = fixup_tree_root_location(root, dir, dentry,
5093                                        &location, &sub_root);
5094         if (ret < 0) {
5095                 if (ret != -ENOENT)
5096                         inode = ERR_PTR(ret);
5097                 else
5098                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
5099         } else {
5100                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5101         }
5102         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5103
5104         if (!IS_ERR(inode) && root != sub_root) {
5105                 down_read(&root->fs_info->cleanup_work_sem);
5106                 if (!(inode->i_sb->s_flags & MS_RDONLY))
5107                         ret = btrfs_orphan_cleanup(sub_root);
5108                 up_read(&root->fs_info->cleanup_work_sem);
5109                 if (ret)
5110                         inode = ERR_PTR(ret);
5111         }
5112
5113         return inode;
5114 }
5115
5116 static int btrfs_dentry_delete(const struct dentry *dentry)
5117 {
5118         struct btrfs_root *root;
5119         struct inode *inode = dentry->d_inode;
5120
5121         if (!inode && !IS_ROOT(dentry))
5122                 inode = dentry->d_parent->d_inode;
5123
5124         if (inode) {
5125                 root = BTRFS_I(inode)->root;
5126                 if (btrfs_root_refs(&root->root_item) == 0)
5127                         return 1;
5128
5129                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5130                         return 1;
5131         }
5132         return 0;
5133 }
5134
5135 static void btrfs_dentry_release(struct dentry *dentry)
5136 {
5137         if (dentry->d_fsdata)
5138                 kfree(dentry->d_fsdata);
5139 }
5140
5141 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5142                                    unsigned int flags)
5143 {
5144         struct dentry *ret;
5145
5146         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
5147         return ret;
5148 }
5149
5150 unsigned char btrfs_filetype_table[] = {
5151         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5152 };
5153
5154 static int btrfs_real_readdir(struct file *filp, void *dirent,
5155                               filldir_t filldir)
5156 {
5157         struct inode *inode = file_inode(filp);
5158         struct btrfs_root *root = BTRFS_I(inode)->root;
5159         struct btrfs_item *item;
5160         struct btrfs_dir_item *di;
5161         struct btrfs_key key;
5162         struct btrfs_key found_key;
5163         struct btrfs_path *path;
5164         struct list_head ins_list;
5165         struct list_head del_list;
5166         int ret;
5167         struct extent_buffer *leaf;
5168         int slot;
5169         unsigned char d_type;
5170         int over = 0;
5171         u32 di_cur;
5172         u32 di_total;
5173         u32 di_len;
5174         int key_type = BTRFS_DIR_INDEX_KEY;
5175         char tmp_name[32];
5176         char *name_ptr;
5177         int name_len;
5178         int is_curr = 0;        /* filp->f_pos points to the current index? */
5179
5180         /* FIXME, use a real flag for deciding about the key type */
5181         if (root->fs_info->tree_root == root)
5182                 key_type = BTRFS_DIR_ITEM_KEY;
5183
5184         /* special case for "." */
5185         if (filp->f_pos == 0) {
5186                 over = filldir(dirent, ".", 1,
5187                                filp->f_pos, btrfs_ino(inode), DT_DIR);
5188                 if (over)
5189                         return 0;
5190                 filp->f_pos = 1;
5191         }
5192         /* special case for .., just use the back ref */
5193         if (filp->f_pos == 1) {
5194                 u64 pino = parent_ino(filp->f_path.dentry);
5195                 over = filldir(dirent, "..", 2,
5196                                filp->f_pos, pino, DT_DIR);
5197                 if (over)
5198                         return 0;
5199                 filp->f_pos = 2;
5200         }
5201         path = btrfs_alloc_path();
5202         if (!path)
5203                 return -ENOMEM;
5204
5205         path->reada = 1;
5206
5207         if (key_type == BTRFS_DIR_INDEX_KEY) {
5208                 INIT_LIST_HEAD(&ins_list);
5209                 INIT_LIST_HEAD(&del_list);
5210                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5211         }
5212
5213         btrfs_set_key_type(&key, key_type);
5214         key.offset = filp->f_pos;
5215         key.objectid = btrfs_ino(inode);
5216
5217         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5218         if (ret < 0)
5219                 goto err;
5220
5221         while (1) {
5222                 leaf = path->nodes[0];
5223                 slot = path->slots[0];
5224                 if (slot >= btrfs_header_nritems(leaf)) {
5225                         ret = btrfs_next_leaf(root, path);
5226                         if (ret < 0)
5227                                 goto err;
5228                         else if (ret > 0)
5229                                 break;
5230                         continue;
5231                 }
5232
5233                 item = btrfs_item_nr(leaf, slot);
5234                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5235
5236                 if (found_key.objectid != key.objectid)
5237                         break;
5238                 if (btrfs_key_type(&found_key) != key_type)
5239                         break;
5240                 if (found_key.offset < filp->f_pos)
5241                         goto next;
5242                 if (key_type == BTRFS_DIR_INDEX_KEY &&
5243                     btrfs_should_delete_dir_index(&del_list,
5244                                                   found_key.offset))
5245                         goto next;
5246
5247                 filp->f_pos = found_key.offset;
5248                 is_curr = 1;
5249
5250                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5251                 di_cur = 0;
5252                 di_total = btrfs_item_size(leaf, item);
5253
5254                 while (di_cur < di_total) {
5255                         struct btrfs_key location;
5256
5257                         if (verify_dir_item(root, leaf, di))
5258                                 break;
5259
5260                         name_len = btrfs_dir_name_len(leaf, di);
5261                         if (name_len <= sizeof(tmp_name)) {
5262                                 name_ptr = tmp_name;
5263                         } else {
5264                                 name_ptr = kmalloc(name_len, GFP_NOFS);
5265                                 if (!name_ptr) {
5266                                         ret = -ENOMEM;
5267                                         goto err;
5268                                 }
5269                         }
5270                         read_extent_buffer(leaf, name_ptr,
5271                                            (unsigned long)(di + 1), name_len);
5272
5273                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5274                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
5275
5276
5277                         /* is this a reference to our own snapshot? If so
5278                          * skip it.
5279                          *
5280                          * In contrast to old kernels, we insert the snapshot's
5281                          * dir item and dir index after it has been created, so
5282                          * we won't find a reference to our own snapshot. We
5283                          * still keep the following code for backward
5284                          * compatibility.
5285                          */
5286                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
5287                             location.objectid == root->root_key.objectid) {
5288                                 over = 0;
5289                                 goto skip;
5290                         }
5291                         over = filldir(dirent, name_ptr, name_len,
5292                                        found_key.offset, location.objectid,
5293                                        d_type);
5294
5295 skip:
5296                         if (name_ptr != tmp_name)
5297                                 kfree(name_ptr);
5298
5299                         if (over)
5300                                 goto nopos;
5301                         di_len = btrfs_dir_name_len(leaf, di) +
5302                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
5303                         di_cur += di_len;
5304                         di = (struct btrfs_dir_item *)((char *)di + di_len);
5305                 }
5306 next:
5307                 path->slots[0]++;
5308         }
5309
5310         if (key_type == BTRFS_DIR_INDEX_KEY) {
5311                 if (is_curr)
5312                         filp->f_pos++;
5313                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
5314                                                       &ins_list);
5315                 if (ret)
5316                         goto nopos;
5317         }
5318
5319         /* Reached end of directory/root. Bump pos past the last item. */
5320         if (key_type == BTRFS_DIR_INDEX_KEY)
5321                 /*
5322                  * 32-bit glibc will use getdents64, but then strtol -
5323                  * so the last number we can serve is this.
5324                  */
5325                 filp->f_pos = 0x7fffffff;
5326         else
5327                 filp->f_pos++;
5328 nopos:
5329         ret = 0;
5330 err:
5331         if (key_type == BTRFS_DIR_INDEX_KEY)
5332                 btrfs_put_delayed_items(&ins_list, &del_list);
5333         btrfs_free_path(path);
5334         return ret;
5335 }
5336
5337 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5338 {
5339         struct btrfs_root *root = BTRFS_I(inode)->root;
5340         struct btrfs_trans_handle *trans;
5341         int ret = 0;
5342         bool nolock = false;
5343
5344         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5345                 return 0;
5346
5347         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5348                 nolock = true;
5349
5350         if (wbc->sync_mode == WB_SYNC_ALL) {
5351                 if (nolock)
5352                         trans = btrfs_join_transaction_nolock(root);
5353                 else
5354                         trans = btrfs_join_transaction(root);
5355                 if (IS_ERR(trans))
5356                         return PTR_ERR(trans);
5357                 ret = btrfs_commit_transaction(trans, root);
5358         }
5359         return ret;
5360 }
5361
5362 /*
5363  * This is somewhat expensive, updating the tree every time the
5364  * inode changes.  But, it is most likely to find the inode in cache.
5365  * FIXME, needs more benchmarking...there are no reasons other than performance
5366  * to keep or drop this code.
5367  */
5368 static int btrfs_dirty_inode(struct inode *inode)
5369 {
5370         struct btrfs_root *root = BTRFS_I(inode)->root;
5371         struct btrfs_trans_handle *trans;
5372         int ret;
5373
5374         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5375                 return 0;
5376
5377         trans = btrfs_join_transaction(root);
5378         if (IS_ERR(trans))
5379                 return PTR_ERR(trans);
5380
5381         ret = btrfs_update_inode(trans, root, inode);
5382         if (ret && ret == -ENOSPC) {
5383                 /* whoops, lets try again with the full transaction */
5384                 btrfs_end_transaction(trans, root);
5385                 trans = btrfs_start_transaction(root, 1);
5386                 if (IS_ERR(trans))
5387                         return PTR_ERR(trans);
5388
5389                 ret = btrfs_update_inode(trans, root, inode);
5390         }
5391         btrfs_end_transaction(trans, root);
5392         if (BTRFS_I(inode)->delayed_node)
5393                 btrfs_balance_delayed_items(root);
5394
5395         return ret;
5396 }
5397
5398 /*
5399  * This is a copy of file_update_time.  We need this so we can return error on
5400  * ENOSPC for updating the inode in the case of file write and mmap writes.
5401  */
5402 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5403                              int flags)
5404 {
5405         struct btrfs_root *root = BTRFS_I(inode)->root;
5406
5407         if (btrfs_root_readonly(root))
5408                 return -EROFS;
5409
5410         if (flags & S_VERSION)
5411                 inode_inc_iversion(inode);
5412         if (flags & S_CTIME)
5413                 inode->i_ctime = *now;
5414         if (flags & S_MTIME)
5415                 inode->i_mtime = *now;
5416         if (flags & S_ATIME)
5417                 inode->i_atime = *now;
5418         return btrfs_dirty_inode(inode);
5419 }
5420
5421 /*
5422  * find the highest existing sequence number in a directory
5423  * and then set the in-memory index_cnt variable to reflect
5424  * free sequence numbers
5425  */
5426 static int btrfs_set_inode_index_count(struct inode *inode)
5427 {
5428         struct btrfs_root *root = BTRFS_I(inode)->root;
5429         struct btrfs_key key, found_key;
5430         struct btrfs_path *path;
5431         struct extent_buffer *leaf;
5432         int ret;
5433
5434         key.objectid = btrfs_ino(inode);
5435         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5436         key.offset = (u64)-1;
5437
5438         path = btrfs_alloc_path();
5439         if (!path)
5440                 return -ENOMEM;
5441
5442         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5443         if (ret < 0)
5444                 goto out;
5445         /* FIXME: we should be able to handle this */
5446         if (ret == 0)
5447                 goto out;
5448         ret = 0;
5449
5450         /*
5451          * MAGIC NUMBER EXPLANATION:
5452          * since we search a directory based on f_pos we have to start at 2
5453          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5454          * else has to start at 2
5455          */
5456         if (path->slots[0] == 0) {
5457                 BTRFS_I(inode)->index_cnt = 2;
5458                 goto out;
5459         }
5460
5461         path->slots[0]--;
5462
5463         leaf = path->nodes[0];
5464         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5465
5466         if (found_key.objectid != btrfs_ino(inode) ||
5467             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5468                 BTRFS_I(inode)->index_cnt = 2;
5469                 goto out;
5470         }
5471
5472         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5473 out:
5474         btrfs_free_path(path);
5475         return ret;
5476 }
5477
5478 /*
5479  * helper to find a free sequence number in a given directory.  This current
5480  * code is very simple, later versions will do smarter things in the btree
5481  */
5482 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5483 {
5484         int ret = 0;
5485
5486         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5487                 ret = btrfs_inode_delayed_dir_index_count(dir);
5488                 if (ret) {
5489                         ret = btrfs_set_inode_index_count(dir);
5490                         if (ret)
5491                                 return ret;
5492                 }
5493         }
5494
5495         *index = BTRFS_I(dir)->index_cnt;
5496         BTRFS_I(dir)->index_cnt++;
5497
5498         return ret;
5499 }
5500
5501 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5502                                      struct btrfs_root *root,
5503                                      struct inode *dir,
5504                                      const char *name, int name_len,
5505                                      u64 ref_objectid, u64 objectid,
5506                                      umode_t mode, u64 *index)
5507 {
5508         struct inode *inode;
5509         struct btrfs_inode_item *inode_item;
5510         struct btrfs_key *location;
5511         struct btrfs_path *path;
5512         struct btrfs_inode_ref *ref;
5513         struct btrfs_key key[2];
5514         u32 sizes[2];
5515         unsigned long ptr;
5516         int ret;
5517         int owner;
5518
5519         path = btrfs_alloc_path();
5520         if (!path)
5521                 return ERR_PTR(-ENOMEM);
5522
5523         inode = new_inode(root->fs_info->sb);
5524         if (!inode) {
5525                 btrfs_free_path(path);
5526                 return ERR_PTR(-ENOMEM);
5527         }
5528
5529         /*
5530          * we have to initialize this early, so we can reclaim the inode
5531          * number if we fail afterwards in this function.
5532          */
5533         inode->i_ino = objectid;
5534
5535         if (dir) {
5536                 trace_btrfs_inode_request(dir);
5537
5538                 ret = btrfs_set_inode_index(dir, index);
5539                 if (ret) {
5540                         btrfs_free_path(path);
5541                         iput(inode);
5542                         return ERR_PTR(ret);
5543                 }
5544         }
5545         /*
5546          * index_cnt is ignored for everything but a dir,
5547          * btrfs_get_inode_index_count has an explanation for the magic
5548          * number
5549          */
5550         BTRFS_I(inode)->index_cnt = 2;
5551         BTRFS_I(inode)->root = root;
5552         BTRFS_I(inode)->generation = trans->transid;
5553         inode->i_generation = BTRFS_I(inode)->generation;
5554
5555         /*
5556          * We could have gotten an inode number from somebody who was fsynced
5557          * and then removed in this same transaction, so let's just set full
5558          * sync since it will be a full sync anyway and this will blow away the
5559          * old info in the log.
5560          */
5561         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5562
5563         if (S_ISDIR(mode))
5564                 owner = 0;
5565         else
5566                 owner = 1;
5567
5568         key[0].objectid = objectid;
5569         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5570         key[0].offset = 0;
5571
5572         /*
5573          * Start new inodes with an inode_ref. This is slightly more
5574          * efficient for small numbers of hard links since they will
5575          * be packed into one item. Extended refs will kick in if we
5576          * add more hard links than can fit in the ref item.
5577          */
5578         key[1].objectid = objectid;
5579         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5580         key[1].offset = ref_objectid;
5581
5582         sizes[0] = sizeof(struct btrfs_inode_item);
5583         sizes[1] = name_len + sizeof(*ref);
5584
5585         path->leave_spinning = 1;
5586         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5587         if (ret != 0)
5588                 goto fail;
5589
5590         inode_init_owner(inode, dir, mode);
5591         inode_set_bytes(inode, 0);
5592         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5593         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5594                                   struct btrfs_inode_item);
5595         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5596                              sizeof(*inode_item));
5597         fill_inode_item(trans, path->nodes[0], inode_item, inode);
5598
5599         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5600                              struct btrfs_inode_ref);
5601         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5602         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5603         ptr = (unsigned long)(ref + 1);
5604         write_extent_buffer(path->nodes[0], name, ptr, name_len);
5605
5606         btrfs_mark_buffer_dirty(path->nodes[0]);
5607         btrfs_free_path(path);
5608
5609         location = &BTRFS_I(inode)->location;
5610         location->objectid = objectid;
5611         location->offset = 0;
5612         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5613
5614         btrfs_inherit_iflags(inode, dir);
5615
5616         if (S_ISREG(mode)) {
5617                 if (btrfs_test_opt(root, NODATASUM))
5618                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5619                 if (btrfs_test_opt(root, NODATACOW))
5620                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5621                                 BTRFS_INODE_NODATASUM;
5622         }
5623
5624         insert_inode_hash(inode);
5625         inode_tree_add(inode);
5626
5627         trace_btrfs_inode_new(inode);
5628         btrfs_set_inode_last_trans(trans, inode);
5629
5630         btrfs_update_root_times(trans, root);
5631
5632         return inode;
5633 fail:
5634         if (dir)
5635                 BTRFS_I(dir)->index_cnt--;
5636         btrfs_free_path(path);
5637         iput(inode);
5638         return ERR_PTR(ret);
5639 }
5640
5641 static inline u8 btrfs_inode_type(struct inode *inode)
5642 {
5643         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5644 }
5645
5646 /*
5647  * utility function to add 'inode' into 'parent_inode' with
5648  * a give name and a given sequence number.
5649  * if 'add_backref' is true, also insert a backref from the
5650  * inode to the parent directory.
5651  */
5652 int btrfs_add_link(struct btrfs_trans_handle *trans,
5653                    struct inode *parent_inode, struct inode *inode,
5654                    const char *name, int name_len, int add_backref, u64 index)
5655 {
5656         int ret = 0;
5657         struct btrfs_key key;
5658         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5659         u64 ino = btrfs_ino(inode);
5660         u64 parent_ino = btrfs_ino(parent_inode);
5661
5662         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5663                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5664         } else {
5665                 key.objectid = ino;
5666                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5667                 key.offset = 0;
5668         }
5669
5670         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5671                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5672                                          key.objectid, root->root_key.objectid,
5673                                          parent_ino, index, name, name_len);
5674         } else if (add_backref) {
5675                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5676                                              parent_ino, index);
5677         }
5678
5679         /* Nothing to clean up yet */
5680         if (ret)
5681                 return ret;
5682
5683         ret = btrfs_insert_dir_item(trans, root, name, name_len,
5684                                     parent_inode, &key,
5685                                     btrfs_inode_type(inode), index);
5686         if (ret == -EEXIST || ret == -EOVERFLOW)
5687                 goto fail_dir_item;
5688         else if (ret) {
5689                 btrfs_abort_transaction(trans, root, ret);
5690                 return ret;
5691         }
5692
5693         btrfs_i_size_write(parent_inode, parent_inode->i_size +
5694                            name_len * 2);
5695         inode_inc_iversion(parent_inode);
5696         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5697         ret = btrfs_update_inode(trans, root, parent_inode);
5698         if (ret)
5699                 btrfs_abort_transaction(trans, root, ret);
5700         return ret;
5701
5702 fail_dir_item:
5703         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5704                 u64 local_index;
5705                 int err;
5706                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5707                                  key.objectid, root->root_key.objectid,
5708                                  parent_ino, &local_index, name, name_len);
5709
5710         } else if (add_backref) {
5711                 u64 local_index;
5712                 int err;
5713
5714                 err = btrfs_del_inode_ref(trans, root, name, name_len,
5715                                           ino, parent_ino, &local_index);
5716         }
5717         return ret;
5718 }
5719
5720 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5721                             struct inode *dir, struct dentry *dentry,
5722                             struct inode *inode, int backref, u64 index)
5723 {
5724         int err = btrfs_add_link(trans, dir, inode,
5725                                  dentry->d_name.name, dentry->d_name.len,
5726                                  backref, index);
5727         if (err > 0)
5728                 err = -EEXIST;
5729         return err;
5730 }
5731
5732 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5733                         umode_t mode, dev_t rdev)
5734 {
5735         struct btrfs_trans_handle *trans;
5736         struct btrfs_root *root = BTRFS_I(dir)->root;
5737         struct inode *inode = NULL;
5738         int err;
5739         int drop_inode = 0;
5740         u64 objectid;
5741         u64 index = 0;
5742
5743         if (!new_valid_dev(rdev))
5744                 return -EINVAL;
5745
5746         /*
5747          * 2 for inode item and ref
5748          * 2 for dir items
5749          * 1 for xattr if selinux is on
5750          */
5751         trans = btrfs_start_transaction(root, 5);
5752         if (IS_ERR(trans))
5753                 return PTR_ERR(trans);
5754
5755         err = btrfs_find_free_ino(root, &objectid);
5756         if (err)
5757                 goto out_unlock;
5758
5759         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5760                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5761                                 mode, &index);
5762         if (IS_ERR(inode)) {
5763                 err = PTR_ERR(inode);
5764                 goto out_unlock;
5765         }
5766
5767         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5768         if (err) {
5769                 drop_inode = 1;
5770                 goto out_unlock;
5771         }
5772
5773         /*
5774         * If the active LSM wants to access the inode during
5775         * d_instantiate it needs these. Smack checks to see
5776         * if the filesystem supports xattrs by looking at the
5777         * ops vector.
5778         */
5779
5780         inode->i_op = &btrfs_special_inode_operations;
5781         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5782         if (err)
5783                 drop_inode = 1;
5784         else {
5785                 init_special_inode(inode, inode->i_mode, rdev);
5786                 btrfs_update_inode(trans, root, inode);
5787                 d_instantiate(dentry, inode);
5788         }
5789 out_unlock:
5790         btrfs_end_transaction(trans, root);
5791         btrfs_btree_balance_dirty(root);
5792         if (drop_inode) {
5793                 inode_dec_link_count(inode);
5794                 iput(inode);
5795         }
5796         return err;
5797 }
5798
5799 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5800                         umode_t mode, bool excl)
5801 {
5802         struct btrfs_trans_handle *trans;
5803         struct btrfs_root *root = BTRFS_I(dir)->root;
5804         struct inode *inode = NULL;
5805         int drop_inode_on_err = 0;
5806         int err;
5807         u64 objectid;
5808         u64 index = 0;
5809
5810         /*
5811          * 2 for inode item and ref
5812          * 2 for dir items
5813          * 1 for xattr if selinux is on
5814          */
5815         trans = btrfs_start_transaction(root, 5);
5816         if (IS_ERR(trans))
5817                 return PTR_ERR(trans);
5818
5819         err = btrfs_find_free_ino(root, &objectid);
5820         if (err)
5821                 goto out_unlock;
5822
5823         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5824                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5825                                 mode, &index);
5826         if (IS_ERR(inode)) {
5827                 err = PTR_ERR(inode);
5828                 goto out_unlock;
5829         }
5830         drop_inode_on_err = 1;
5831
5832         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5833         if (err)
5834                 goto out_unlock;
5835
5836         err = btrfs_update_inode(trans, root, inode);
5837         if (err)
5838                 goto out_unlock;
5839
5840         /*
5841         * If the active LSM wants to access the inode during
5842         * d_instantiate it needs these. Smack checks to see
5843         * if the filesystem supports xattrs by looking at the
5844         * ops vector.
5845         */
5846         inode->i_fop = &btrfs_file_operations;
5847         inode->i_op = &btrfs_file_inode_operations;
5848
5849         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5850         if (err)
5851                 goto out_unlock;
5852
5853         inode->i_mapping->a_ops = &btrfs_aops;
5854         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5855         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5856         d_instantiate(dentry, inode);
5857
5858 out_unlock:
5859         btrfs_end_transaction(trans, root);
5860         if (err && drop_inode_on_err) {
5861                 inode_dec_link_count(inode);
5862                 iput(inode);
5863         }
5864         btrfs_btree_balance_dirty(root);
5865         return err;
5866 }
5867
5868 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5869                       struct dentry *dentry)
5870 {
5871         struct btrfs_trans_handle *trans;
5872         struct btrfs_root *root = BTRFS_I(dir)->root;
5873         struct inode *inode = old_dentry->d_inode;
5874         u64 index;
5875         int err;
5876         int drop_inode = 0;
5877
5878         /* do not allow sys_link's with other subvols of the same device */
5879         if (root->objectid != BTRFS_I(inode)->root->objectid)
5880                 return -EXDEV;
5881
5882         if (inode->i_nlink >= BTRFS_LINK_MAX)
5883                 return -EMLINK;
5884
5885         err = btrfs_set_inode_index(dir, &index);
5886         if (err)
5887                 goto fail;
5888
5889         /*
5890          * 2 items for inode and inode ref
5891          * 2 items for dir items
5892          * 1 item for parent inode
5893          */
5894         trans = btrfs_start_transaction(root, 5);
5895         if (IS_ERR(trans)) {
5896                 err = PTR_ERR(trans);
5897                 goto fail;
5898         }
5899
5900         btrfs_inc_nlink(inode);
5901         inode_inc_iversion(inode);
5902         inode->i_ctime = CURRENT_TIME;
5903         ihold(inode);
5904         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5905
5906         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5907
5908         if (err) {
5909                 drop_inode = 1;
5910         } else {
5911                 struct dentry *parent = dentry->d_parent;
5912                 err = btrfs_update_inode(trans, root, inode);
5913                 if (err)
5914                         goto fail;
5915                 d_instantiate(dentry, inode);
5916                 btrfs_log_new_name(trans, inode, NULL, parent);
5917         }
5918
5919         btrfs_end_transaction(trans, root);
5920 fail:
5921         if (drop_inode) {
5922                 inode_dec_link_count(inode);
5923                 iput(inode);
5924         }
5925         btrfs_btree_balance_dirty(root);
5926         return err;
5927 }
5928
5929 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5930 {
5931         struct inode *inode = NULL;
5932         struct btrfs_trans_handle *trans;
5933         struct btrfs_root *root = BTRFS_I(dir)->root;
5934         int err = 0;
5935         int drop_on_err = 0;
5936         u64 objectid = 0;
5937         u64 index = 0;
5938
5939         /*
5940          * 2 items for inode and ref
5941          * 2 items for dir items
5942          * 1 for xattr if selinux is on
5943          */
5944         trans = btrfs_start_transaction(root, 5);
5945         if (IS_ERR(trans))
5946                 return PTR_ERR(trans);
5947
5948         err = btrfs_find_free_ino(root, &objectid);
5949         if (err)
5950                 goto out_fail;
5951
5952         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5953                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5954                                 S_IFDIR | mode, &index);
5955         if (IS_ERR(inode)) {
5956                 err = PTR_ERR(inode);
5957                 goto out_fail;
5958         }
5959
5960         drop_on_err = 1;
5961
5962         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5963         if (err)
5964                 goto out_fail;
5965
5966         inode->i_op = &btrfs_dir_inode_operations;
5967         inode->i_fop = &btrfs_dir_file_operations;
5968
5969         btrfs_i_size_write(inode, 0);
5970         err = btrfs_update_inode(trans, root, inode);
5971         if (err)
5972                 goto out_fail;
5973
5974         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5975                              dentry->d_name.len, 0, index);
5976         if (err)
5977                 goto out_fail;
5978
5979         d_instantiate(dentry, inode);
5980         drop_on_err = 0;
5981
5982 out_fail:
5983         btrfs_end_transaction(trans, root);
5984         if (drop_on_err)
5985                 iput(inode);
5986         btrfs_btree_balance_dirty(root);
5987         return err;
5988 }
5989
5990 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5991  * and an extent that you want to insert, deal with overlap and insert
5992  * the new extent into the tree.
5993  */
5994 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5995                                 struct extent_map *existing,
5996                                 struct extent_map *em,
5997                                 u64 map_start, u64 map_len)
5998 {
5999         u64 start_diff;
6000
6001         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6002         start_diff = map_start - em->start;
6003         em->start = map_start;
6004         em->len = map_len;
6005         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6006             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6007                 em->block_start += start_diff;
6008                 em->block_len -= start_diff;
6009         }
6010         return add_extent_mapping(em_tree, em, 0);
6011 }
6012
6013 static noinline int uncompress_inline(struct btrfs_path *path,
6014                                       struct inode *inode, struct page *page,
6015                                       size_t pg_offset, u64 extent_offset,
6016                                       struct btrfs_file_extent_item *item)
6017 {
6018         int ret;
6019         struct extent_buffer *leaf = path->nodes[0];
6020         char *tmp;
6021         size_t max_size;
6022         unsigned long inline_size;
6023         unsigned long ptr;
6024         int compress_type;
6025
6026         WARN_ON(pg_offset != 0);
6027         compress_type = btrfs_file_extent_compression(leaf, item);
6028         max_size = btrfs_file_extent_ram_bytes(leaf, item);
6029         inline_size = btrfs_file_extent_inline_item_len(leaf,
6030                                         btrfs_item_nr(leaf, path->slots[0]));
6031         tmp = kmalloc(inline_size, GFP_NOFS);
6032         if (!tmp)
6033                 return -ENOMEM;
6034         ptr = btrfs_file_extent_inline_start(item);
6035
6036         read_extent_buffer(leaf, tmp, ptr, inline_size);
6037
6038         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6039         ret = btrfs_decompress(compress_type, tmp, page,
6040                                extent_offset, inline_size, max_size);
6041         if (ret) {
6042                 char *kaddr = kmap_atomic(page);
6043                 unsigned long copy_size = min_t(u64,
6044                                   PAGE_CACHE_SIZE - pg_offset,
6045                                   max_size - extent_offset);
6046                 memset(kaddr + pg_offset, 0, copy_size);
6047                 kunmap_atomic(kaddr);
6048         }
6049         kfree(tmp);
6050         return 0;
6051 }
6052
6053 /*
6054  * a bit scary, this does extent mapping from logical file offset to the disk.
6055  * the ugly parts come from merging extents from the disk with the in-ram
6056  * representation.  This gets more complex because of the data=ordered code,
6057  * where the in-ram extents might be locked pending data=ordered completion.
6058  *
6059  * This also copies inline extents directly into the page.
6060  */
6061
6062 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6063                                     size_t pg_offset, u64 start, u64 len,
6064                                     int create)
6065 {
6066         int ret;
6067         int err = 0;
6068         u64 bytenr;
6069         u64 extent_start = 0;
6070         u64 extent_end = 0;
6071         u64 objectid = btrfs_ino(inode);
6072         u32 found_type;
6073         struct btrfs_path *path = NULL;
6074         struct btrfs_root *root = BTRFS_I(inode)->root;
6075         struct btrfs_file_extent_item *item;
6076         struct extent_buffer *leaf;
6077         struct btrfs_key found_key;
6078         struct extent_map *em = NULL;
6079         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6080         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6081         struct btrfs_trans_handle *trans = NULL;
6082         int compress_type;
6083
6084 again:
6085         read_lock(&em_tree->lock);
6086         em = lookup_extent_mapping(em_tree, start, len);
6087         if (em)
6088                 em->bdev = root->fs_info->fs_devices->latest_bdev;
6089         read_unlock(&em_tree->lock);
6090
6091         if (em) {
6092                 if (em->start > start || em->start + em->len <= start)
6093                         free_extent_map(em);
6094                 else if (em->block_start == EXTENT_MAP_INLINE && page)
6095                         free_extent_map(em);
6096                 else
6097                         goto out;
6098         }
6099         em = alloc_extent_map();
6100         if (!em) {
6101                 err = -ENOMEM;
6102                 goto out;
6103         }
6104         em->bdev = root->fs_info->fs_devices->latest_bdev;
6105         em->start = EXTENT_MAP_HOLE;
6106         em->orig_start = EXTENT_MAP_HOLE;
6107         em->len = (u64)-1;
6108         em->block_len = (u64)-1;
6109
6110         if (!path) {
6111                 path = btrfs_alloc_path();
6112                 if (!path) {
6113                         err = -ENOMEM;
6114                         goto out;
6115                 }
6116                 /*
6117                  * Chances are we'll be called again, so go ahead and do
6118                  * readahead
6119                  */
6120                 path->reada = 1;
6121         }
6122
6123         ret = btrfs_lookup_file_extent(trans, root, path,
6124                                        objectid, start, trans != NULL);
6125         if (ret < 0) {
6126                 err = ret;
6127                 goto out;
6128         }
6129
6130         if (ret != 0) {
6131                 if (path->slots[0] == 0)
6132                         goto not_found;
6133                 path->slots[0]--;
6134         }
6135
6136         leaf = path->nodes[0];
6137         item = btrfs_item_ptr(leaf, path->slots[0],
6138                               struct btrfs_file_extent_item);
6139         /* are we inside the extent that was found? */
6140         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6141         found_type = btrfs_key_type(&found_key);
6142         if (found_key.objectid != objectid ||
6143             found_type != BTRFS_EXTENT_DATA_KEY) {
6144                 goto not_found;
6145         }
6146
6147         found_type = btrfs_file_extent_type(leaf, item);
6148         extent_start = found_key.offset;
6149         compress_type = btrfs_file_extent_compression(leaf, item);
6150         if (found_type == BTRFS_FILE_EXTENT_REG ||
6151             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6152                 extent_end = extent_start +
6153                        btrfs_file_extent_num_bytes(leaf, item);
6154         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6155                 size_t size;
6156                 size = btrfs_file_extent_inline_len(leaf, item);
6157                 extent_end = ALIGN(extent_start + size, root->sectorsize);
6158         }
6159
6160         if (start >= extent_end) {
6161                 path->slots[0]++;
6162                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6163                         ret = btrfs_next_leaf(root, path);
6164                         if (ret < 0) {
6165                                 err = ret;
6166                                 goto out;
6167                         }
6168                         if (ret > 0)
6169                                 goto not_found;
6170                         leaf = path->nodes[0];
6171                 }
6172                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6173                 if (found_key.objectid != objectid ||
6174                     found_key.type != BTRFS_EXTENT_DATA_KEY)
6175                         goto not_found;
6176                 if (start + len <= found_key.offset)
6177                         goto not_found;
6178                 em->start = start;
6179                 em->orig_start = start;
6180                 em->len = found_key.offset - start;
6181                 goto not_found_em;
6182         }
6183
6184         em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
6185         if (found_type == BTRFS_FILE_EXTENT_REG ||
6186             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6187                 em->start = extent_start;
6188                 em->len = extent_end - extent_start;
6189                 em->orig_start = extent_start -
6190                                  btrfs_file_extent_offset(leaf, item);
6191                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6192                                                                       item);
6193                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6194                 if (bytenr == 0) {
6195                         em->block_start = EXTENT_MAP_HOLE;
6196                         goto insert;
6197                 }
6198                 if (compress_type != BTRFS_COMPRESS_NONE) {
6199                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6200                         em->compress_type = compress_type;
6201                         em->block_start = bytenr;
6202                         em->block_len = em->orig_block_len;
6203                 } else {
6204                         bytenr += btrfs_file_extent_offset(leaf, item);
6205                         em->block_start = bytenr;
6206                         em->block_len = em->len;
6207                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6208                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6209                 }
6210                 goto insert;
6211         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6212                 unsigned long ptr;
6213                 char *map;
6214                 size_t size;
6215                 size_t extent_offset;
6216                 size_t copy_size;
6217
6218                 em->block_start = EXTENT_MAP_INLINE;
6219                 if (!page || create) {
6220                         em->start = extent_start;
6221                         em->len = extent_end - extent_start;
6222                         goto out;
6223                 }
6224
6225                 size = btrfs_file_extent_inline_len(leaf, item);
6226                 extent_offset = page_offset(page) + pg_offset - extent_start;
6227                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6228                                 size - extent_offset);
6229                 em->start = extent_start + extent_offset;
6230                 em->len = ALIGN(copy_size, root->sectorsize);
6231                 em->orig_block_len = em->len;
6232                 em->orig_start = em->start;
6233                 if (compress_type) {
6234                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6235                         em->compress_type = compress_type;
6236                 }
6237                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6238                 if (create == 0 && !PageUptodate(page)) {
6239                         if (btrfs_file_extent_compression(leaf, item) !=
6240                             BTRFS_COMPRESS_NONE) {
6241                                 ret = uncompress_inline(path, inode, page,
6242                                                         pg_offset,
6243                                                         extent_offset, item);
6244                                 BUG_ON(ret); /* -ENOMEM */
6245                         } else {
6246                                 map = kmap(page);
6247                                 read_extent_buffer(leaf, map + pg_offset, ptr,
6248                                                    copy_size);
6249                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6250                                         memset(map + pg_offset + copy_size, 0,
6251                                                PAGE_CACHE_SIZE - pg_offset -
6252                                                copy_size);
6253                                 }
6254                                 kunmap(page);
6255                         }
6256                         flush_dcache_page(page);
6257                 } else if (create && PageUptodate(page)) {
6258                         BUG();
6259                         if (!trans) {
6260                                 kunmap(page);
6261                                 free_extent_map(em);
6262                                 em = NULL;
6263
6264                                 btrfs_release_path(path);
6265                                 trans = btrfs_join_transaction(root);
6266
6267                                 if (IS_ERR(trans))
6268                                         return ERR_CAST(trans);
6269                                 goto again;
6270                         }
6271                         map = kmap(page);
6272                         write_extent_buffer(leaf, map + pg_offset, ptr,
6273                                             copy_size);
6274                         kunmap(page);
6275                         btrfs_mark_buffer_dirty(leaf);
6276                 }
6277                 set_extent_uptodate(io_tree, em->start,
6278                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
6279                 goto insert;
6280         } else {
6281                 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6282         }
6283 not_found:
6284         em->start = start;
6285         em->orig_start = start;
6286         em->len = len;
6287 not_found_em:
6288         em->block_start = EXTENT_MAP_HOLE;
6289         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6290 insert:
6291         btrfs_release_path(path);
6292         if (em->start > start || extent_map_end(em) <= start) {
6293                 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6294                         (unsigned long long)em->start,
6295                         (unsigned long long)em->len,
6296                         (unsigned long long)start,
6297                         (unsigned long long)len);
6298                 err = -EIO;
6299                 goto out;
6300         }
6301
6302         err = 0;
6303         write_lock(&em_tree->lock);
6304         ret = add_extent_mapping(em_tree, em, 0);
6305         /* it is possible that someone inserted the extent into the tree
6306          * while we had the lock dropped.  It is also possible that
6307          * an overlapping map exists in the tree
6308          */
6309         if (ret == -EEXIST) {
6310                 struct extent_map *existing;
6311
6312                 ret = 0;
6313
6314                 existing = lookup_extent_mapping(em_tree, start, len);
6315                 if (existing && (existing->start > start ||
6316                     existing->start + existing->len <= start)) {
6317                         free_extent_map(existing);
6318                         existing = NULL;
6319                 }
6320                 if (!existing) {
6321                         existing = lookup_extent_mapping(em_tree, em->start,
6322                                                          em->len);
6323                         if (existing) {
6324                                 err = merge_extent_mapping(em_tree, existing,
6325                                                            em, start,
6326                                                            root->sectorsize);
6327                                 free_extent_map(existing);
6328                                 if (err) {
6329                                         free_extent_map(em);
6330                                         em = NULL;
6331                                 }
6332                         } else {
6333                                 err = -EIO;
6334                                 free_extent_map(em);
6335                                 em = NULL;
6336                         }
6337                 } else {
6338                         free_extent_map(em);
6339                         em = existing;
6340                         err = 0;
6341                 }
6342         }
6343         write_unlock(&em_tree->lock);
6344 out:
6345
6346         if (em)
6347                 trace_btrfs_get_extent(root, em);
6348
6349         if (path)
6350                 btrfs_free_path(path);
6351         if (trans) {
6352                 ret = btrfs_end_transaction(trans, root);
6353                 if (!err)
6354                         err = ret;
6355         }
6356         if (err) {
6357                 free_extent_map(em);
6358                 return ERR_PTR(err);
6359         }
6360         BUG_ON(!em); /* Error is always set */
6361         return em;
6362 }
6363
6364 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6365                                            size_t pg_offset, u64 start, u64 len,
6366                                            int create)
6367 {
6368         struct extent_map *em;
6369         struct extent_map *hole_em = NULL;
6370         u64 range_start = start;
6371         u64 end;
6372         u64 found;
6373         u64 found_end;
6374         int err = 0;
6375
6376         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6377         if (IS_ERR(em))
6378                 return em;
6379         if (em) {
6380                 /*
6381                  * if our em maps to
6382                  * -  a hole or
6383                  * -  a pre-alloc extent,
6384                  * there might actually be delalloc bytes behind it.
6385                  */
6386                 if (em->block_start != EXTENT_MAP_HOLE &&
6387                     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6388                         return em;
6389                 else
6390                         hole_em = em;
6391         }
6392
6393         /* check to see if we've wrapped (len == -1 or similar) */
6394         end = start + len;
6395         if (end < start)
6396                 end = (u64)-1;
6397         else
6398                 end -= 1;
6399
6400         em = NULL;
6401
6402         /* ok, we didn't find anything, lets look for delalloc */
6403         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6404                                  end, len, EXTENT_DELALLOC, 1);
6405         found_end = range_start + found;
6406         if (found_end < range_start)
6407                 found_end = (u64)-1;
6408
6409         /*
6410          * we didn't find anything useful, return
6411          * the original results from get_extent()
6412          */
6413         if (range_start > end || found_end <= start) {
6414                 em = hole_em;
6415                 hole_em = NULL;
6416                 goto out;
6417         }
6418
6419         /* adjust the range_start to make sure it doesn't
6420          * go backwards from the start they passed in
6421          */
6422         range_start = max(start,range_start);
6423         found = found_end - range_start;
6424
6425         if (found > 0) {
6426                 u64 hole_start = start;
6427                 u64 hole_len = len;
6428
6429                 em = alloc_extent_map();
6430                 if (!em) {
6431                         err = -ENOMEM;
6432                         goto out;
6433                 }
6434                 /*
6435                  * when btrfs_get_extent can't find anything it
6436                  * returns one huge hole
6437                  *
6438                  * make sure what it found really fits our range, and
6439                  * adjust to make sure it is based on the start from
6440                  * the caller
6441                  */
6442                 if (hole_em) {
6443                         u64 calc_end = extent_map_end(hole_em);
6444
6445                         if (calc_end <= start || (hole_em->start > end)) {
6446                                 free_extent_map(hole_em);
6447                                 hole_em = NULL;
6448                         } else {
6449                                 hole_start = max(hole_em->start, start);
6450                                 hole_len = calc_end - hole_start;
6451                         }
6452                 }
6453                 em->bdev = NULL;
6454                 if (hole_em && range_start > hole_start) {
6455                         /* our hole starts before our delalloc, so we
6456                          * have to return just the parts of the hole
6457                          * that go until  the delalloc starts
6458                          */
6459                         em->len = min(hole_len,
6460                                       range_start - hole_start);
6461                         em->start = hole_start;
6462                         em->orig_start = hole_start;
6463                         /*
6464                          * don't adjust block start at all,
6465                          * it is fixed at EXTENT_MAP_HOLE
6466                          */
6467                         em->block_start = hole_em->block_start;
6468                         em->block_len = hole_len;
6469                         if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6470                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6471                 } else {
6472                         em->start = range_start;
6473                         em->len = found;
6474                         em->orig_start = range_start;
6475                         em->block_start = EXTENT_MAP_DELALLOC;
6476                         em->block_len = found;
6477                 }
6478         } else if (hole_em) {
6479                 return hole_em;
6480         }
6481 out:
6482
6483         free_extent_map(hole_em);
6484         if (err) {
6485                 free_extent_map(em);
6486                 return ERR_PTR(err);
6487         }
6488         return em;
6489 }
6490
6491 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6492                                                   u64 start, u64 len)
6493 {
6494         struct btrfs_root *root = BTRFS_I(inode)->root;
6495         struct btrfs_trans_handle *trans;
6496         struct extent_map *em;
6497         struct btrfs_key ins;
6498         u64 alloc_hint;
6499         int ret;
6500
6501         trans = btrfs_join_transaction(root);
6502         if (IS_ERR(trans))
6503                 return ERR_CAST(trans);
6504
6505         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
6506
6507         alloc_hint = get_extent_allocation_hint(inode, start, len);
6508         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
6509                                    alloc_hint, &ins, 1);
6510         if (ret) {
6511                 em = ERR_PTR(ret);
6512                 goto out;
6513         }
6514
6515         em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6516                               ins.offset, ins.offset, ins.offset, 0);
6517         if (IS_ERR(em))
6518                 goto out;
6519
6520         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6521                                            ins.offset, ins.offset, 0);
6522         if (ret) {
6523                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6524                 em = ERR_PTR(ret);
6525         }
6526 out:
6527         btrfs_end_transaction(trans, root);
6528         return em;
6529 }
6530
6531 /*
6532  * returns 1 when the nocow is safe, < 1 on error, 0 if the
6533  * block must be cow'd
6534  */
6535 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
6536                                       struct inode *inode, u64 offset, u64 *len,
6537                                       u64 *orig_start, u64 *orig_block_len,
6538                                       u64 *ram_bytes)
6539 {
6540         struct btrfs_path *path;
6541         int ret;
6542         struct extent_buffer *leaf;
6543         struct btrfs_root *root = BTRFS_I(inode)->root;
6544         struct btrfs_file_extent_item *fi;
6545         struct btrfs_key key;
6546         u64 disk_bytenr;
6547         u64 backref_offset;
6548         u64 extent_end;
6549         u64 num_bytes;
6550         int slot;
6551         int found_type;
6552
6553         path = btrfs_alloc_path();
6554         if (!path)
6555                 return -ENOMEM;
6556
6557         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
6558                                        offset, 0);
6559         if (ret < 0)
6560                 goto out;
6561
6562         slot = path->slots[0];
6563         if (ret == 1) {
6564                 if (slot == 0) {
6565                         /* can't find the item, must cow */
6566                         ret = 0;
6567                         goto out;
6568                 }
6569                 slot--;
6570         }
6571         ret = 0;
6572         leaf = path->nodes[0];
6573         btrfs_item_key_to_cpu(leaf, &key, slot);
6574         if (key.objectid != btrfs_ino(inode) ||
6575             key.type != BTRFS_EXTENT_DATA_KEY) {
6576                 /* not our file or wrong item type, must cow */
6577                 goto out;
6578         }
6579
6580         if (key.offset > offset) {
6581                 /* Wrong offset, must cow */
6582                 goto out;
6583         }
6584
6585         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6586         found_type = btrfs_file_extent_type(leaf, fi);
6587         if (found_type != BTRFS_FILE_EXTENT_REG &&
6588             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6589                 /* not a regular extent, must cow */
6590                 goto out;
6591         }
6592         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6593         backref_offset = btrfs_file_extent_offset(leaf, fi);
6594
6595         *orig_start = key.offset - backref_offset;
6596         *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6597         *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6598
6599         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6600         if (extent_end < offset + *len) {
6601                 /* extent doesn't include our full range, must cow */
6602                 goto out;
6603         }
6604
6605         if (btrfs_extent_readonly(root, disk_bytenr))
6606                 goto out;
6607
6608         /*
6609          * look for other files referencing this extent, if we
6610          * find any we must cow
6611          */
6612         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6613                                   key.offset - backref_offset, disk_bytenr))
6614                 goto out;
6615
6616         /*
6617          * adjust disk_bytenr and num_bytes to cover just the bytes
6618          * in this extent we are about to write.  If there
6619          * are any csums in that range we have to cow in order
6620          * to keep the csums correct
6621          */
6622         disk_bytenr += backref_offset;
6623         disk_bytenr += offset - key.offset;
6624         num_bytes = min(offset + *len, extent_end) - offset;
6625         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6626                                 goto out;
6627         /*
6628          * all of the above have passed, it is safe to overwrite this extent
6629          * without cow
6630          */
6631         *len = num_bytes;
6632         ret = 1;
6633 out:
6634         btrfs_free_path(path);
6635         return ret;
6636 }
6637
6638 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6639                               struct extent_state **cached_state, int writing)
6640 {
6641         struct btrfs_ordered_extent *ordered;
6642         int ret = 0;
6643
6644         while (1) {
6645                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6646                                  0, cached_state);
6647                 /*
6648                  * We're concerned with the entire range that we're going to be
6649                  * doing DIO to, so we need to make sure theres no ordered
6650                  * extents in this range.
6651                  */
6652                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6653                                                      lockend - lockstart + 1);
6654
6655                 /*
6656                  * We need to make sure there are no buffered pages in this
6657                  * range either, we could have raced between the invalidate in
6658                  * generic_file_direct_write and locking the extent.  The
6659                  * invalidate needs to happen so that reads after a write do not
6660                  * get stale data.
6661                  */
6662                 if (!ordered && (!writing ||
6663                     !test_range_bit(&BTRFS_I(inode)->io_tree,
6664                                     lockstart, lockend, EXTENT_UPTODATE, 0,
6665                                     *cached_state)))
6666                         break;
6667
6668                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6669                                      cached_state, GFP_NOFS);
6670
6671                 if (ordered) {
6672                         btrfs_start_ordered_extent(inode, ordered, 1);
6673                         btrfs_put_ordered_extent(ordered);
6674                 } else {
6675                         /* Screw you mmap */
6676                         ret = filemap_write_and_wait_range(inode->i_mapping,
6677                                                            lockstart,
6678                                                            lockend);
6679                         if (ret)
6680                                 break;
6681
6682                         /*
6683                          * If we found a page that couldn't be invalidated just
6684                          * fall back to buffered.
6685                          */
6686                         ret = invalidate_inode_pages2_range(inode->i_mapping,
6687                                         lockstart >> PAGE_CACHE_SHIFT,
6688                                         lockend >> PAGE_CACHE_SHIFT);
6689                         if (ret)
6690                                 break;
6691                 }
6692
6693                 cond_resched();
6694         }
6695
6696         return ret;
6697 }
6698
6699 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6700                                            u64 len, u64 orig_start,
6701                                            u64 block_start, u64 block_len,
6702                                            u64 orig_block_len, u64 ram_bytes,
6703                                            int type)
6704 {
6705         struct extent_map_tree *em_tree;
6706         struct extent_map *em;
6707         struct btrfs_root *root = BTRFS_I(inode)->root;
6708         int ret;
6709
6710         em_tree = &BTRFS_I(inode)->extent_tree;
6711         em = alloc_extent_map();
6712         if (!em)
6713                 return ERR_PTR(-ENOMEM);
6714
6715         em->start = start;
6716         em->orig_start = orig_start;
6717         em->mod_start = start;
6718         em->mod_len = len;
6719         em->len = len;
6720         em->block_len = block_len;
6721         em->block_start = block_start;
6722         em->bdev = root->fs_info->fs_devices->latest_bdev;
6723         em->orig_block_len = orig_block_len;
6724         em->ram_bytes = ram_bytes;
6725         em->generation = -1;
6726         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6727         if (type == BTRFS_ORDERED_PREALLOC)
6728                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6729
6730         do {
6731                 btrfs_drop_extent_cache(inode, em->start,
6732                                 em->start + em->len - 1, 0);
6733                 write_lock(&em_tree->lock);
6734                 ret = add_extent_mapping(em_tree, em, 1);
6735                 write_unlock(&em_tree->lock);
6736         } while (ret == -EEXIST);
6737
6738         if (ret) {
6739                 free_extent_map(em);
6740                 return ERR_PTR(ret);
6741         }
6742
6743         return em;
6744 }
6745
6746
6747 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6748                                    struct buffer_head *bh_result, int create)
6749 {
6750         struct extent_map *em;
6751         struct btrfs_root *root = BTRFS_I(inode)->root;
6752         struct extent_state *cached_state = NULL;
6753         u64 start = iblock << inode->i_blkbits;
6754         u64 lockstart, lockend;
6755         u64 len = bh_result->b_size;
6756         struct btrfs_trans_handle *trans;
6757         int unlock_bits = EXTENT_LOCKED;
6758         int ret = 0;
6759
6760         if (create)
6761                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6762         else
6763                 len = min_t(u64, len, root->sectorsize);
6764
6765         lockstart = start;
6766         lockend = start + len - 1;
6767
6768         /*
6769          * If this errors out it's because we couldn't invalidate pagecache for
6770          * this range and we need to fallback to buffered.
6771          */
6772         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6773                 return -ENOTBLK;
6774
6775         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6776         if (IS_ERR(em)) {
6777                 ret = PTR_ERR(em);
6778                 goto unlock_err;
6779         }
6780
6781         /*
6782          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6783          * io.  INLINE is special, and we could probably kludge it in here, but
6784          * it's still buffered so for safety lets just fall back to the generic
6785          * buffered path.
6786          *
6787          * For COMPRESSED we _have_ to read the entire extent in so we can
6788          * decompress it, so there will be buffering required no matter what we
6789          * do, so go ahead and fallback to buffered.
6790          *
6791          * We return -ENOTBLK because thats what makes DIO go ahead and go back
6792          * to buffered IO.  Don't blame me, this is the price we pay for using
6793          * the generic code.
6794          */
6795         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6796             em->block_start == EXTENT_MAP_INLINE) {
6797                 free_extent_map(em);
6798                 ret = -ENOTBLK;
6799                 goto unlock_err;
6800         }
6801
6802         /* Just a good old fashioned hole, return */
6803         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6804                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6805                 free_extent_map(em);
6806                 goto unlock_err;
6807         }
6808
6809         /*
6810          * We don't allocate a new extent in the following cases
6811          *
6812          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6813          * existing extent.
6814          * 2) The extent is marked as PREALLOC.  We're good to go here and can
6815          * just use the extent.
6816          *
6817          */
6818         if (!create) {
6819                 len = min(len, em->len - (start - em->start));
6820                 lockstart = start + len;
6821                 goto unlock;
6822         }
6823
6824         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6825             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6826              em->block_start != EXTENT_MAP_HOLE)) {
6827                 int type;
6828                 int ret;
6829                 u64 block_start, orig_start, orig_block_len, ram_bytes;
6830
6831                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6832                         type = BTRFS_ORDERED_PREALLOC;
6833                 else
6834                         type = BTRFS_ORDERED_NOCOW;
6835                 len = min(len, em->len - (start - em->start));
6836                 block_start = em->block_start + (start - em->start);
6837
6838                 /*
6839                  * we're not going to log anything, but we do need
6840                  * to make sure the current transaction stays open
6841                  * while we look for nocow cross refs
6842                  */
6843                 trans = btrfs_join_transaction(root);
6844                 if (IS_ERR(trans))
6845                         goto must_cow;
6846
6847                 if (can_nocow_odirect(trans, inode, start, &len, &orig_start,
6848                                       &orig_block_len, &ram_bytes) == 1) {
6849                         if (type == BTRFS_ORDERED_PREALLOC) {
6850                                 free_extent_map(em);
6851                                 em = create_pinned_em(inode, start, len,
6852                                                        orig_start,
6853                                                        block_start, len,
6854                                                        orig_block_len,
6855                                                        ram_bytes, type);
6856                                 if (IS_ERR(em)) {
6857                                         btrfs_end_transaction(trans, root);
6858                                         goto unlock_err;
6859                                 }
6860                         }
6861
6862                         ret = btrfs_add_ordered_extent_dio(inode, start,
6863                                            block_start, len, len, type);
6864                         btrfs_end_transaction(trans, root);
6865                         if (ret) {
6866                                 free_extent_map(em);
6867                                 goto unlock_err;
6868                         }
6869                         goto unlock;
6870                 }
6871                 btrfs_end_transaction(trans, root);
6872         }
6873 must_cow:
6874         /*
6875          * this will cow the extent, reset the len in case we changed
6876          * it above
6877          */
6878         len = bh_result->b_size;
6879         free_extent_map(em);
6880         em = btrfs_new_extent_direct(inode, start, len);
6881         if (IS_ERR(em)) {
6882                 ret = PTR_ERR(em);
6883                 goto unlock_err;
6884         }
6885         len = min(len, em->len - (start - em->start));
6886 unlock:
6887         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6888                 inode->i_blkbits;
6889         bh_result->b_size = len;
6890         bh_result->b_bdev = em->bdev;
6891         set_buffer_mapped(bh_result);
6892         if (create) {
6893                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6894                         set_buffer_new(bh_result);
6895
6896                 /*
6897                  * Need to update the i_size under the extent lock so buffered
6898                  * readers will get the updated i_size when we unlock.
6899                  */
6900                 if (start + len > i_size_read(inode))
6901                         i_size_write(inode, start + len);
6902
6903                 spin_lock(&BTRFS_I(inode)->lock);
6904                 BTRFS_I(inode)->outstanding_extents++;
6905                 spin_unlock(&BTRFS_I(inode)->lock);
6906
6907                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6908                                      lockstart + len - 1, EXTENT_DELALLOC, NULL,
6909                                      &cached_state, GFP_NOFS);
6910                 BUG_ON(ret);
6911         }
6912
6913         /*
6914          * In the case of write we need to clear and unlock the entire range,
6915          * in the case of read we need to unlock only the end area that we
6916          * aren't using if there is any left over space.
6917          */
6918         if (lockstart < lockend) {
6919                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6920                                  lockend, unlock_bits, 1, 0,
6921                                  &cached_state, GFP_NOFS);
6922         } else {
6923                 free_extent_state(cached_state);
6924         }
6925
6926         free_extent_map(em);
6927
6928         return 0;
6929
6930 unlock_err:
6931         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6932                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6933         return ret;
6934 }
6935
6936 struct btrfs_dio_private {
6937         struct inode *inode;
6938         u64 logical_offset;
6939         u64 disk_bytenr;
6940         u64 bytes;
6941         void *private;
6942
6943         /* number of bios pending for this dio */
6944         atomic_t pending_bios;
6945
6946         /* IO errors */
6947         int errors;
6948
6949         /* orig_bio is our btrfs_io_bio */
6950         struct bio *orig_bio;
6951
6952         /* dio_bio came from fs/direct-io.c */
6953         struct bio *dio_bio;
6954 };
6955
6956 static void btrfs_endio_direct_read(struct bio *bio, int err)
6957 {
6958         struct btrfs_dio_private *dip = bio->bi_private;
6959         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6960         struct bio_vec *bvec = bio->bi_io_vec;
6961         struct inode *inode = dip->inode;
6962         struct btrfs_root *root = BTRFS_I(inode)->root;
6963         struct bio *dio_bio;
6964         u64 start;
6965
6966         start = dip->logical_offset;
6967         do {
6968                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6969                         struct page *page = bvec->bv_page;
6970                         char *kaddr;
6971                         u32 csum = ~(u32)0;
6972                         u64 private = ~(u32)0;
6973                         unsigned long flags;
6974
6975                         if (get_state_private(&BTRFS_I(inode)->io_tree,
6976                                               start, &private))
6977                                 goto failed;
6978                         local_irq_save(flags);
6979                         kaddr = kmap_atomic(page);
6980                         csum = btrfs_csum_data(kaddr + bvec->bv_offset,
6981                                                csum, bvec->bv_len);
6982                         btrfs_csum_final(csum, (char *)&csum);
6983                         kunmap_atomic(kaddr);
6984                         local_irq_restore(flags);
6985
6986                         flush_dcache_page(bvec->bv_page);
6987                         if (csum != private) {
6988 failed:
6989                                 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u private %u",
6990                                         (unsigned long long)btrfs_ino(inode),
6991                                         (unsigned long long)start,
6992                                         csum, (unsigned)private);
6993                                 err = -EIO;
6994                         }
6995                 }
6996
6997                 start += bvec->bv_len;
6998                 bvec++;
6999         } while (bvec <= bvec_end);
7000
7001         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
7002                       dip->logical_offset + dip->bytes - 1);
7003         dio_bio = dip->dio_bio;
7004
7005         kfree(dip);
7006
7007         /* If we had a csum failure make sure to clear the uptodate flag */
7008         if (err)
7009                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7010         dio_end_io(dio_bio, err);
7011         bio_put(bio);
7012 }
7013
7014 static void btrfs_endio_direct_write(struct bio *bio, int err)
7015 {
7016         struct btrfs_dio_private *dip = bio->bi_private;
7017         struct inode *inode = dip->inode;
7018         struct btrfs_root *root = BTRFS_I(inode)->root;
7019         struct btrfs_ordered_extent *ordered = NULL;
7020         u64 ordered_offset = dip->logical_offset;
7021         u64 ordered_bytes = dip->bytes;
7022         struct bio *dio_bio;
7023         int ret;
7024
7025         if (err)
7026                 goto out_done;
7027 again:
7028         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7029                                                    &ordered_offset,
7030                                                    ordered_bytes, !err);
7031         if (!ret)
7032                 goto out_test;
7033
7034         ordered->work.func = finish_ordered_fn;
7035         ordered->work.flags = 0;
7036         btrfs_queue_worker(&root->fs_info->endio_write_workers,
7037                            &ordered->work);
7038 out_test:
7039         /*
7040          * our bio might span multiple ordered extents.  If we haven't
7041          * completed the accounting for the whole dio, go back and try again
7042          */
7043         if (ordered_offset < dip->logical_offset + dip->bytes) {
7044                 ordered_bytes = dip->logical_offset + dip->bytes -
7045                         ordered_offset;
7046                 ordered = NULL;
7047                 goto again;
7048         }
7049 out_done:
7050         dio_bio = dip->dio_bio;
7051
7052         kfree(dip);
7053
7054         /* If we had an error make sure to clear the uptodate flag */
7055         if (err)
7056                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7057         dio_end_io(dio_bio, err);
7058         bio_put(bio);
7059 }
7060
7061 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7062                                     struct bio *bio, int mirror_num,
7063                                     unsigned long bio_flags, u64 offset)
7064 {
7065         int ret;
7066         struct btrfs_root *root = BTRFS_I(inode)->root;
7067         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7068         BUG_ON(ret); /* -ENOMEM */
7069         return 0;
7070 }
7071
7072 static void btrfs_end_dio_bio(struct bio *bio, int err)
7073 {
7074         struct btrfs_dio_private *dip = bio->bi_private;
7075
7076         if (err) {
7077                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
7078                       "sector %#Lx len %u err no %d\n",
7079                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
7080                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
7081                 dip->errors = 1;
7082
7083                 /*
7084                  * before atomic variable goto zero, we must make sure
7085                  * dip->errors is perceived to be set.
7086                  */
7087                 smp_mb__before_atomic_dec();
7088         }
7089
7090         /* if there are more bios still pending for this dio, just exit */
7091         if (!atomic_dec_and_test(&dip->pending_bios))
7092                 goto out;
7093
7094         if (dip->errors) {
7095                 bio_io_error(dip->orig_bio);
7096         } else {
7097                 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7098                 bio_endio(dip->orig_bio, 0);
7099         }
7100 out:
7101         bio_put(bio);
7102 }
7103
7104 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7105                                        u64 first_sector, gfp_t gfp_flags)
7106 {
7107         int nr_vecs = bio_get_nr_vecs(bdev);
7108         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7109 }
7110
7111 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7112                                          int rw, u64 file_offset, int skip_sum,
7113                                          int async_submit)
7114 {
7115         int write = rw & REQ_WRITE;
7116         struct btrfs_root *root = BTRFS_I(inode)->root;
7117         int ret;
7118
7119         if (async_submit)
7120                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7121
7122         bio_get(bio);
7123
7124         if (!write) {
7125                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7126                 if (ret)
7127                         goto err;
7128         }
7129
7130         if (skip_sum)
7131                 goto map;
7132
7133         if (write && async_submit) {
7134                 ret = btrfs_wq_submit_bio(root->fs_info,
7135                                    inode, rw, bio, 0, 0,
7136                                    file_offset,
7137                                    __btrfs_submit_bio_start_direct_io,
7138                                    __btrfs_submit_bio_done);
7139                 goto err;
7140         } else if (write) {
7141                 /*
7142                  * If we aren't doing async submit, calculate the csum of the
7143                  * bio now.
7144                  */
7145                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7146                 if (ret)
7147                         goto err;
7148         } else if (!skip_sum) {
7149                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
7150                 if (ret)
7151                         goto err;
7152         }
7153
7154 map:
7155         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7156 err:
7157         bio_put(bio);
7158         return ret;
7159 }
7160
7161 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7162                                     int skip_sum)
7163 {
7164         struct inode *inode = dip->inode;
7165         struct btrfs_root *root = BTRFS_I(inode)->root;
7166         struct bio *bio;
7167         struct bio *orig_bio = dip->orig_bio;
7168         struct bio_vec *bvec = orig_bio->bi_io_vec;
7169         u64 start_sector = orig_bio->bi_sector;
7170         u64 file_offset = dip->logical_offset;
7171         u64 submit_len = 0;
7172         u64 map_length;
7173         int nr_pages = 0;
7174         int ret = 0;
7175         int async_submit = 0;
7176
7177         map_length = orig_bio->bi_size;
7178         ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7179                               &map_length, NULL, 0);
7180         if (ret) {
7181                 bio_put(orig_bio);
7182                 return -EIO;
7183         }
7184         if (map_length >= orig_bio->bi_size) {
7185                 bio = orig_bio;
7186                 goto submit;
7187         }
7188
7189         /* async crcs make it difficult to collect full stripe writes. */
7190         if (btrfs_get_alloc_profile(root, 1) &
7191             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7192                 async_submit = 0;
7193         else
7194                 async_submit = 1;
7195
7196         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7197         if (!bio)
7198                 return -ENOMEM;
7199         bio->bi_private = dip;
7200         bio->bi_end_io = btrfs_end_dio_bio;
7201         atomic_inc(&dip->pending_bios);
7202
7203         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7204                 if (unlikely(map_length < submit_len + bvec->bv_len ||
7205                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7206                                  bvec->bv_offset) < bvec->bv_len)) {
7207                         /*
7208                          * inc the count before we submit the bio so
7209                          * we know the end IO handler won't happen before
7210                          * we inc the count. Otherwise, the dip might get freed
7211                          * before we're done setting it up
7212                          */
7213                         atomic_inc(&dip->pending_bios);
7214                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
7215                                                      file_offset, skip_sum,
7216                                                      async_submit);
7217                         if (ret) {
7218                                 bio_put(bio);
7219                                 atomic_dec(&dip->pending_bios);
7220                                 goto out_err;
7221                         }
7222
7223                         start_sector += submit_len >> 9;
7224                         file_offset += submit_len;
7225
7226                         submit_len = 0;
7227                         nr_pages = 0;
7228
7229                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7230                                                   start_sector, GFP_NOFS);
7231                         if (!bio)
7232                                 goto out_err;
7233                         bio->bi_private = dip;
7234                         bio->bi_end_io = btrfs_end_dio_bio;
7235
7236                         map_length = orig_bio->bi_size;
7237                         ret = btrfs_map_block(root->fs_info, rw,
7238                                               start_sector << 9,
7239                                               &map_length, NULL, 0);
7240                         if (ret) {
7241                                 bio_put(bio);
7242                                 goto out_err;
7243                         }
7244                 } else {
7245                         submit_len += bvec->bv_len;
7246                         nr_pages ++;
7247                         bvec++;
7248                 }
7249         }
7250
7251 submit:
7252         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7253                                      async_submit);
7254         if (!ret)
7255                 return 0;
7256
7257         bio_put(bio);
7258 out_err:
7259         dip->errors = 1;
7260         /*
7261          * before atomic variable goto zero, we must
7262          * make sure dip->errors is perceived to be set.
7263          */
7264         smp_mb__before_atomic_dec();
7265         if (atomic_dec_and_test(&dip->pending_bios))
7266                 bio_io_error(dip->orig_bio);
7267
7268         /* bio_end_io() will handle error, so we needn't return it */
7269         return 0;
7270 }
7271
7272 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7273                                 struct inode *inode, loff_t file_offset)
7274 {
7275         struct btrfs_root *root = BTRFS_I(inode)->root;
7276         struct btrfs_dio_private *dip;
7277         struct bio_vec *bvec = dio_bio->bi_io_vec;
7278         struct bio *io_bio;
7279         int skip_sum;
7280         int write = rw & REQ_WRITE;
7281         int ret = 0;
7282
7283         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7284
7285         io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7286
7287         if (!io_bio) {
7288                 ret = -ENOMEM;
7289                 goto free_ordered;
7290         }
7291
7292         dip = kmalloc(sizeof(*dip), GFP_NOFS);
7293         if (!dip) {
7294                 ret = -ENOMEM;
7295                 goto free_io_bio;
7296         }
7297
7298         dip->private = dio_bio->bi_private;
7299         io_bio->bi_private = dio_bio->bi_private;
7300         dip->inode = inode;
7301         dip->logical_offset = file_offset;
7302
7303         dip->bytes = 0;
7304         do {
7305                 dip->bytes += bvec->bv_len;
7306                 bvec++;
7307         } while (bvec <= (dio_bio->bi_io_vec + dio_bio->bi_vcnt - 1));
7308
7309         dip->disk_bytenr = (u64)dio_bio->bi_sector << 9;
7310         io_bio->bi_private = dip;
7311         dip->errors = 0;
7312         dip->orig_bio = io_bio;
7313         dip->dio_bio = dio_bio;
7314         atomic_set(&dip->pending_bios, 0);
7315
7316         if (write)
7317                 io_bio->bi_end_io = btrfs_endio_direct_write;
7318         else
7319                 io_bio->bi_end_io = btrfs_endio_direct_read;
7320
7321         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7322         if (!ret)
7323                 return;
7324
7325 free_io_bio:
7326         bio_put(io_bio);
7327
7328 free_ordered:
7329         /*
7330          * If this is a write, we need to clean up the reserved space and kill
7331          * the ordered extent.
7332          */
7333         if (write) {
7334                 struct btrfs_ordered_extent *ordered;
7335                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7336                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7337                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7338                         btrfs_free_reserved_extent(root, ordered->start,
7339                                                    ordered->disk_len);
7340                 btrfs_put_ordered_extent(ordered);
7341                 btrfs_put_ordered_extent(ordered);
7342         }
7343         bio_endio(dio_bio, ret);
7344 }
7345
7346 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7347                         const struct iovec *iov, loff_t offset,
7348                         unsigned long nr_segs)
7349 {
7350         int seg;
7351         int i;
7352         size_t size;
7353         unsigned long addr;
7354         unsigned blocksize_mask = root->sectorsize - 1;
7355         ssize_t retval = -EINVAL;
7356         loff_t end = offset;
7357
7358         if (offset & blocksize_mask)
7359                 goto out;
7360
7361         /* Check the memory alignment.  Blocks cannot straddle pages */
7362         for (seg = 0; seg < nr_segs; seg++) {
7363                 addr = (unsigned long)iov[seg].iov_base;
7364                 size = iov[seg].iov_len;
7365                 end += size;
7366                 if ((addr & blocksize_mask) || (size & blocksize_mask))
7367                         goto out;
7368
7369                 /* If this is a write we don't need to check anymore */
7370                 if (rw & WRITE)
7371                         continue;
7372
7373                 /*
7374                  * Check to make sure we don't have duplicate iov_base's in this
7375                  * iovec, if so return EINVAL, otherwise we'll get csum errors
7376                  * when reading back.
7377                  */
7378                 for (i = seg + 1; i < nr_segs; i++) {
7379                         if (iov[seg].iov_base == iov[i].iov_base)
7380                                 goto out;
7381                 }
7382         }
7383         retval = 0;
7384 out:
7385         return retval;
7386 }
7387
7388 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7389                         const struct iovec *iov, loff_t offset,
7390                         unsigned long nr_segs)
7391 {
7392         struct file *file = iocb->ki_filp;
7393         struct inode *inode = file->f_mapping->host;
7394         size_t count = 0;
7395         int flags = 0;
7396         bool wakeup = true;
7397         bool relock = false;
7398         ssize_t ret;
7399
7400         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7401                             offset, nr_segs))
7402                 return 0;
7403
7404         atomic_inc(&inode->i_dio_count);
7405         smp_mb__after_atomic_inc();
7406
7407         if (rw & WRITE) {
7408                 count = iov_length(iov, nr_segs);
7409                 /*
7410                  * If the write DIO is beyond the EOF, we need update
7411                  * the isize, but it is protected by i_mutex. So we can
7412                  * not unlock the i_mutex at this case.
7413                  */
7414                 if (offset + count <= inode->i_size) {
7415                         mutex_unlock(&inode->i_mutex);
7416                         relock = true;
7417                 }
7418                 ret = btrfs_delalloc_reserve_space(inode, count);
7419                 if (ret)
7420                         goto out;
7421         } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7422                                      &BTRFS_I(inode)->runtime_flags))) {
7423                 inode_dio_done(inode);
7424                 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7425                 wakeup = false;
7426         }
7427
7428         ret = __blockdev_direct_IO(rw, iocb, inode,
7429                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7430                         iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7431                         btrfs_submit_direct, flags);
7432         if (rw & WRITE) {
7433                 if (ret < 0 && ret != -EIOCBQUEUED)
7434                         btrfs_delalloc_release_space(inode, count);
7435                 else if (ret >= 0 && (size_t)ret < count)
7436                         btrfs_delalloc_release_space(inode,
7437                                                      count - (size_t)ret);
7438                 else
7439                         btrfs_delalloc_release_metadata(inode, 0);
7440         }
7441 out:
7442         if (wakeup)
7443                 inode_dio_done(inode);
7444         if (relock)
7445                 mutex_lock(&inode->i_mutex);
7446
7447         return ret;
7448 }
7449
7450 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
7451
7452 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7453                 __u64 start, __u64 len)
7454 {
7455         int     ret;
7456
7457         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7458         if (ret)
7459                 return ret;
7460
7461         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7462 }
7463
7464 int btrfs_readpage(struct file *file, struct page *page)
7465 {
7466         struct extent_io_tree *tree;
7467         tree = &BTRFS_I(page->mapping->host)->io_tree;
7468         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7469 }
7470
7471 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7472 {
7473         struct extent_io_tree *tree;
7474
7475
7476         if (current->flags & PF_MEMALLOC) {
7477                 redirty_page_for_writepage(wbc, page);
7478                 unlock_page(page);
7479                 return 0;
7480         }
7481         tree = &BTRFS_I(page->mapping->host)->io_tree;
7482         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7483 }
7484
7485 static int btrfs_writepages(struct address_space *mapping,
7486                             struct writeback_control *wbc)
7487 {
7488         struct extent_io_tree *tree;
7489
7490         tree = &BTRFS_I(mapping->host)->io_tree;
7491         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7492 }
7493
7494 static int
7495 btrfs_readpages(struct file *file, struct address_space *mapping,
7496                 struct list_head *pages, unsigned nr_pages)
7497 {
7498         struct extent_io_tree *tree;
7499         tree = &BTRFS_I(mapping->host)->io_tree;
7500         return extent_readpages(tree, mapping, pages, nr_pages,
7501                                 btrfs_get_extent);
7502 }
7503 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7504 {
7505         struct extent_io_tree *tree;
7506         struct extent_map_tree *map;
7507         int ret;
7508
7509         tree = &BTRFS_I(page->mapping->host)->io_tree;
7510         map = &BTRFS_I(page->mapping->host)->extent_tree;
7511         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7512         if (ret == 1) {
7513                 ClearPagePrivate(page);
7514                 set_page_private(page, 0);
7515                 page_cache_release(page);
7516         }
7517         return ret;
7518 }
7519
7520 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7521 {
7522         if (PageWriteback(page) || PageDirty(page))
7523                 return 0;
7524         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7525 }
7526
7527 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
7528 {
7529         struct inode *inode = page->mapping->host;
7530         struct extent_io_tree *tree;
7531         struct btrfs_ordered_extent *ordered;
7532         struct extent_state *cached_state = NULL;
7533         u64 page_start = page_offset(page);
7534         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7535
7536         /*
7537          * we have the page locked, so new writeback can't start,
7538          * and the dirty bit won't be cleared while we are here.
7539          *
7540          * Wait for IO on this page so that we can safely clear
7541          * the PagePrivate2 bit and do ordered accounting
7542          */
7543         wait_on_page_writeback(page);
7544
7545         tree = &BTRFS_I(inode)->io_tree;
7546         if (offset) {
7547                 btrfs_releasepage(page, GFP_NOFS);
7548                 return;
7549         }
7550         lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7551         ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
7552         if (ordered) {
7553                 /*
7554                  * IO on this page will never be started, so we need
7555                  * to account for any ordered extents now
7556                  */
7557                 clear_extent_bit(tree, page_start, page_end,
7558                                  EXTENT_DIRTY | EXTENT_DELALLOC |
7559                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7560                                  EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
7561                 /*
7562                  * whoever cleared the private bit is responsible
7563                  * for the finish_ordered_io
7564                  */
7565                 if (TestClearPagePrivate2(page) &&
7566                     btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
7567                                                    PAGE_CACHE_SIZE, 1)) {
7568                         btrfs_finish_ordered_io(ordered);
7569                 }
7570                 btrfs_put_ordered_extent(ordered);
7571                 cached_state = NULL;
7572                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7573         }
7574         clear_extent_bit(tree, page_start, page_end,
7575                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
7576                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
7577                  &cached_state, GFP_NOFS);
7578         __btrfs_releasepage(page, GFP_NOFS);
7579
7580         ClearPageChecked(page);
7581         if (PagePrivate(page)) {
7582                 ClearPagePrivate(page);
7583                 set_page_private(page, 0);
7584                 page_cache_release(page);
7585         }
7586 }
7587
7588 /*
7589  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7590  * called from a page fault handler when a page is first dirtied. Hence we must
7591  * be careful to check for EOF conditions here. We set the page up correctly
7592  * for a written page which means we get ENOSPC checking when writing into
7593  * holes and correct delalloc and unwritten extent mapping on filesystems that
7594  * support these features.
7595  *
7596  * We are not allowed to take the i_mutex here so we have to play games to
7597  * protect against truncate races as the page could now be beyond EOF.  Because
7598  * vmtruncate() writes the inode size before removing pages, once we have the
7599  * page lock we can determine safely if the page is beyond EOF. If it is not
7600  * beyond EOF, then the page is guaranteed safe against truncation until we
7601  * unlock the page.
7602  */
7603 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7604 {
7605         struct page *page = vmf->page;
7606         struct inode *inode = file_inode(vma->vm_file);
7607         struct btrfs_root *root = BTRFS_I(inode)->root;
7608         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7609         struct btrfs_ordered_extent *ordered;
7610         struct extent_state *cached_state = NULL;
7611         char *kaddr;
7612         unsigned long zero_start;
7613         loff_t size;
7614         int ret;
7615         int reserved = 0;
7616         u64 page_start;
7617         u64 page_end;
7618
7619         sb_start_pagefault(inode->i_sb);
7620         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7621         if (!ret) {
7622                 ret = file_update_time(vma->vm_file);
7623                 reserved = 1;
7624         }
7625         if (ret) {
7626                 if (ret == -ENOMEM)
7627                         ret = VM_FAULT_OOM;
7628                 else /* -ENOSPC, -EIO, etc */
7629                         ret = VM_FAULT_SIGBUS;
7630                 if (reserved)
7631                         goto out;
7632                 goto out_noreserve;
7633         }
7634
7635         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7636 again:
7637         lock_page(page);
7638         size = i_size_read(inode);
7639         page_start = page_offset(page);
7640         page_end = page_start + PAGE_CACHE_SIZE - 1;
7641
7642         if ((page->mapping != inode->i_mapping) ||
7643             (page_start >= size)) {
7644                 /* page got truncated out from underneath us */
7645                 goto out_unlock;
7646         }
7647         wait_on_page_writeback(page);
7648
7649         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7650         set_page_extent_mapped(page);
7651
7652         /*
7653          * we can't set the delalloc bits if there are pending ordered
7654          * extents.  Drop our locks and wait for them to finish
7655          */
7656         ordered = btrfs_lookup_ordered_extent(inode, page_start);
7657         if (ordered) {
7658                 unlock_extent_cached(io_tree, page_start, page_end,
7659                                      &cached_state, GFP_NOFS);
7660                 unlock_page(page);
7661                 btrfs_start_ordered_extent(inode, ordered, 1);
7662                 btrfs_put_ordered_extent(ordered);
7663                 goto again;
7664         }
7665
7666         /*
7667          * XXX - page_mkwrite gets called every time the page is dirtied, even
7668          * if it was already dirty, so for space accounting reasons we need to
7669          * clear any delalloc bits for the range we are fixing to save.  There
7670          * is probably a better way to do this, but for now keep consistent with
7671          * prepare_pages in the normal write path.
7672          */
7673         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7674                           EXTENT_DIRTY | EXTENT_DELALLOC |
7675                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7676                           0, 0, &cached_state, GFP_NOFS);
7677
7678         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7679                                         &cached_state);
7680         if (ret) {
7681                 unlock_extent_cached(io_tree, page_start, page_end,
7682                                      &cached_state, GFP_NOFS);
7683                 ret = VM_FAULT_SIGBUS;
7684                 goto out_unlock;
7685         }
7686         ret = 0;
7687
7688         /* page is wholly or partially inside EOF */
7689         if (page_start + PAGE_CACHE_SIZE > size)
7690                 zero_start = size & ~PAGE_CACHE_MASK;
7691         else
7692                 zero_start = PAGE_CACHE_SIZE;
7693
7694         if (zero_start != PAGE_CACHE_SIZE) {
7695                 kaddr = kmap(page);
7696                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7697                 flush_dcache_page(page);
7698                 kunmap(page);
7699         }
7700         ClearPageChecked(page);
7701         set_page_dirty(page);
7702         SetPageUptodate(page);
7703
7704         BTRFS_I(inode)->last_trans = root->fs_info->generation;
7705         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7706         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7707
7708         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7709
7710 out_unlock:
7711         if (!ret) {
7712                 sb_end_pagefault(inode->i_sb);
7713                 return VM_FAULT_LOCKED;
7714         }
7715         unlock_page(page);
7716 out:
7717         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7718 out_noreserve:
7719         sb_end_pagefault(inode->i_sb);
7720         return ret;
7721 }
7722
7723 static int btrfs_truncate(struct inode *inode)
7724 {
7725         struct btrfs_root *root = BTRFS_I(inode)->root;
7726         struct btrfs_block_rsv *rsv;
7727         int ret;
7728         int err = 0;
7729         struct btrfs_trans_handle *trans;
7730         u64 mask = root->sectorsize - 1;
7731         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7732
7733         ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
7734         if (ret)
7735                 return ret;
7736
7737         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
7738         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
7739
7740         /*
7741          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
7742          * 3 things going on here
7743          *
7744          * 1) We need to reserve space for our orphan item and the space to
7745          * delete our orphan item.  Lord knows we don't want to have a dangling
7746          * orphan item because we didn't reserve space to remove it.
7747          *
7748          * 2) We need to reserve space to update our inode.
7749          *
7750          * 3) We need to have something to cache all the space that is going to
7751          * be free'd up by the truncate operation, but also have some slack
7752          * space reserved in case it uses space during the truncate (thank you
7753          * very much snapshotting).
7754          *
7755          * And we need these to all be seperate.  The fact is we can use alot of
7756          * space doing the truncate, and we have no earthly idea how much space
7757          * we will use, so we need the truncate reservation to be seperate so it
7758          * doesn't end up using space reserved for updating the inode or
7759          * removing the orphan item.  We also need to be able to stop the
7760          * transaction and start a new one, which means we need to be able to
7761          * update the inode several times, and we have no idea of knowing how
7762          * many times that will be, so we can't just reserve 1 item for the
7763          * entirety of the opration, so that has to be done seperately as well.
7764          * Then there is the orphan item, which does indeed need to be held on
7765          * to for the whole operation, and we need nobody to touch this reserved
7766          * space except the orphan code.
7767          *
7768          * So that leaves us with
7769          *
7770          * 1) root->orphan_block_rsv - for the orphan deletion.
7771          * 2) rsv - for the truncate reservation, which we will steal from the
7772          * transaction reservation.
7773          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7774          * updating the inode.
7775          */
7776         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7777         if (!rsv)
7778                 return -ENOMEM;
7779         rsv->size = min_size;
7780         rsv->failfast = 1;
7781
7782         /*
7783          * 1 for the truncate slack space
7784          * 1 for updating the inode.
7785          */
7786         trans = btrfs_start_transaction(root, 2);
7787         if (IS_ERR(trans)) {
7788                 err = PTR_ERR(trans);
7789                 goto out;
7790         }
7791
7792         /* Migrate the slack space for the truncate to our reserve */
7793         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7794                                       min_size);
7795         BUG_ON(ret);
7796
7797         /*
7798          * setattr is responsible for setting the ordered_data_close flag,
7799          * but that is only tested during the last file release.  That
7800          * could happen well after the next commit, leaving a great big
7801          * window where new writes may get lost if someone chooses to write
7802          * to this file after truncating to zero
7803          *
7804          * The inode doesn't have any dirty data here, and so if we commit
7805          * this is a noop.  If someone immediately starts writing to the inode
7806          * it is very likely we'll catch some of their writes in this
7807          * transaction, and the commit will find this file on the ordered
7808          * data list with good things to send down.
7809          *
7810          * This is a best effort solution, there is still a window where
7811          * using truncate to replace the contents of the file will
7812          * end up with a zero length file after a crash.
7813          */
7814         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7815                                            &BTRFS_I(inode)->runtime_flags))
7816                 btrfs_add_ordered_operation(trans, root, inode);
7817
7818         /*
7819          * So if we truncate and then write and fsync we normally would just
7820          * write the extents that changed, which is a problem if we need to
7821          * first truncate that entire inode.  So set this flag so we write out
7822          * all of the extents in the inode to the sync log so we're completely
7823          * safe.
7824          */
7825         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7826         trans->block_rsv = rsv;
7827
7828         while (1) {
7829                 ret = btrfs_truncate_inode_items(trans, root, inode,
7830                                                  inode->i_size,
7831                                                  BTRFS_EXTENT_DATA_KEY);
7832                 if (ret != -ENOSPC) {
7833                         err = ret;
7834                         break;
7835                 }
7836
7837                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7838                 ret = btrfs_update_inode(trans, root, inode);
7839                 if (ret) {
7840                         err = ret;
7841                         break;
7842                 }
7843
7844                 btrfs_end_transaction(trans, root);
7845                 btrfs_btree_balance_dirty(root);
7846
7847                 trans = btrfs_start_transaction(root, 2);
7848                 if (IS_ERR(trans)) {
7849                         ret = err = PTR_ERR(trans);
7850                         trans = NULL;
7851                         break;
7852                 }
7853
7854                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7855                                               rsv, min_size);
7856                 BUG_ON(ret);    /* shouldn't happen */
7857                 trans->block_rsv = rsv;
7858         }
7859
7860         if (ret == 0 && inode->i_nlink > 0) {
7861                 trans->block_rsv = root->orphan_block_rsv;
7862                 ret = btrfs_orphan_del(trans, inode);
7863                 if (ret)
7864                         err = ret;
7865         }
7866
7867         if (trans) {
7868                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7869                 ret = btrfs_update_inode(trans, root, inode);
7870                 if (ret && !err)
7871                         err = ret;
7872
7873                 ret = btrfs_end_transaction(trans, root);
7874                 btrfs_btree_balance_dirty(root);
7875         }
7876
7877 out:
7878         btrfs_free_block_rsv(root, rsv);
7879
7880         if (ret && !err)
7881                 err = ret;
7882
7883         return err;
7884 }
7885
7886 /*
7887  * create a new subvolume directory/inode (helper for the ioctl).
7888  */
7889 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7890                              struct btrfs_root *new_root, u64 new_dirid)
7891 {
7892         struct inode *inode;
7893         int err;
7894         u64 index = 0;
7895
7896         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7897                                 new_dirid, new_dirid,
7898                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
7899                                 &index);
7900         if (IS_ERR(inode))
7901                 return PTR_ERR(inode);
7902         inode->i_op = &btrfs_dir_inode_operations;
7903         inode->i_fop = &btrfs_dir_file_operations;
7904
7905         set_nlink(inode, 1);
7906         btrfs_i_size_write(inode, 0);
7907
7908         err = btrfs_update_inode(trans, new_root, inode);
7909
7910         iput(inode);
7911         return err;
7912 }
7913
7914 struct inode *btrfs_alloc_inode(struct super_block *sb)
7915 {
7916         struct btrfs_inode *ei;
7917         struct inode *inode;
7918
7919         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7920         if (!ei)
7921                 return NULL;
7922
7923         ei->root = NULL;
7924         ei->generation = 0;
7925         ei->last_trans = 0;
7926         ei->last_sub_trans = 0;
7927         ei->logged_trans = 0;
7928         ei->delalloc_bytes = 0;
7929         ei->disk_i_size = 0;
7930         ei->flags = 0;
7931         ei->csum_bytes = 0;
7932         ei->index_cnt = (u64)-1;
7933         ei->last_unlink_trans = 0;
7934         ei->last_log_commit = 0;
7935
7936         spin_lock_init(&ei->lock);
7937         ei->outstanding_extents = 0;
7938         ei->reserved_extents = 0;
7939
7940         ei->runtime_flags = 0;
7941         ei->force_compress = BTRFS_COMPRESS_NONE;
7942
7943         ei->delayed_node = NULL;
7944
7945         inode = &ei->vfs_inode;
7946         extent_map_tree_init(&ei->extent_tree);
7947         extent_io_tree_init(&ei->io_tree, &inode->i_data);
7948         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7949         ei->io_tree.track_uptodate = 1;
7950         ei->io_failure_tree.track_uptodate = 1;
7951         atomic_set(&ei->sync_writers, 0);
7952         mutex_init(&ei->log_mutex);
7953         mutex_init(&ei->delalloc_mutex);
7954         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7955         INIT_LIST_HEAD(&ei->delalloc_inodes);
7956         INIT_LIST_HEAD(&ei->ordered_operations);
7957         RB_CLEAR_NODE(&ei->rb_node);
7958
7959         return inode;
7960 }
7961
7962 static void btrfs_i_callback(struct rcu_head *head)
7963 {
7964         struct inode *inode = container_of(head, struct inode, i_rcu);
7965         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7966 }
7967
7968 void btrfs_destroy_inode(struct inode *inode)
7969 {
7970         struct btrfs_ordered_extent *ordered;
7971         struct btrfs_root *root = BTRFS_I(inode)->root;
7972
7973         WARN_ON(!hlist_empty(&inode->i_dentry));
7974         WARN_ON(inode->i_data.nrpages);
7975         WARN_ON(BTRFS_I(inode)->outstanding_extents);
7976         WARN_ON(BTRFS_I(inode)->reserved_extents);
7977         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7978         WARN_ON(BTRFS_I(inode)->csum_bytes);
7979
7980         /*
7981          * This can happen where we create an inode, but somebody else also
7982          * created the same inode and we need to destroy the one we already
7983          * created.
7984          */
7985         if (!root)
7986                 goto free;
7987
7988         /*
7989          * Make sure we're properly removed from the ordered operation
7990          * lists.
7991          */
7992         smp_mb();
7993         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7994                 spin_lock(&root->fs_info->ordered_extent_lock);
7995                 list_del_init(&BTRFS_I(inode)->ordered_operations);
7996                 spin_unlock(&root->fs_info->ordered_extent_lock);
7997         }
7998
7999         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8000                      &BTRFS_I(inode)->runtime_flags)) {
8001                 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
8002                         (unsigned long long)btrfs_ino(inode));
8003                 atomic_dec(&root->orphan_inodes);
8004         }
8005
8006         while (1) {
8007                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8008                 if (!ordered)
8009                         break;
8010                 else {
8011                         btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
8012                                 (unsigned long long)ordered->file_offset,
8013                                 (unsigned long long)ordered->len);
8014                         btrfs_remove_ordered_extent(inode, ordered);
8015                         btrfs_put_ordered_extent(ordered);
8016                         btrfs_put_ordered_extent(ordered);
8017                 }
8018         }
8019         inode_tree_del(inode);
8020         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8021 free:
8022         call_rcu(&inode->i_rcu, btrfs_i_callback);
8023 }
8024
8025 int btrfs_drop_inode(struct inode *inode)
8026 {
8027         struct btrfs_root *root = BTRFS_I(inode)->root;
8028
8029         if (root == NULL)
8030                 return 1;
8031
8032         /* the snap/subvol tree is on deleting */
8033         if (btrfs_root_refs(&root->root_item) == 0 &&
8034             root != root->fs_info->tree_root)
8035                 return 1;
8036         else
8037                 return generic_drop_inode(inode);
8038 }
8039
8040 static void init_once(void *foo)
8041 {
8042         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8043
8044         inode_init_once(&ei->vfs_inode);
8045 }
8046
8047 void btrfs_destroy_cachep(void)
8048 {
8049         /*
8050          * Make sure all delayed rcu free inodes are flushed before we
8051          * destroy cache.
8052          */
8053         rcu_barrier();
8054         if (btrfs_inode_cachep)
8055                 kmem_cache_destroy(btrfs_inode_cachep);
8056         if (btrfs_trans_handle_cachep)
8057                 kmem_cache_destroy(btrfs_trans_handle_cachep);
8058         if (btrfs_transaction_cachep)
8059                 kmem_cache_destroy(btrfs_transaction_cachep);
8060         if (btrfs_path_cachep)
8061                 kmem_cache_destroy(btrfs_path_cachep);
8062         if (btrfs_free_space_cachep)
8063                 kmem_cache_destroy(btrfs_free_space_cachep);
8064         if (btrfs_delalloc_work_cachep)
8065                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8066 }
8067
8068 int btrfs_init_cachep(void)
8069 {
8070         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8071                         sizeof(struct btrfs_inode), 0,
8072                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8073         if (!btrfs_inode_cachep)
8074                 goto fail;
8075
8076         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8077                         sizeof(struct btrfs_trans_handle), 0,
8078                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8079         if (!btrfs_trans_handle_cachep)
8080                 goto fail;
8081
8082         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8083                         sizeof(struct btrfs_transaction), 0,
8084                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8085         if (!btrfs_transaction_cachep)
8086                 goto fail;
8087
8088         btrfs_path_cachep = kmem_cache_create("btrfs_path",
8089                         sizeof(struct btrfs_path), 0,
8090                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8091         if (!btrfs_path_cachep)
8092                 goto fail;
8093
8094         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8095                         sizeof(struct btrfs_free_space), 0,
8096                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8097         if (!btrfs_free_space_cachep)
8098                 goto fail;
8099
8100         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8101                         sizeof(struct btrfs_delalloc_work), 0,
8102                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8103                         NULL);
8104         if (!btrfs_delalloc_work_cachep)
8105                 goto fail;
8106
8107         return 0;
8108 fail:
8109         btrfs_destroy_cachep();
8110         return -ENOMEM;
8111 }
8112
8113 static int btrfs_getattr(struct vfsmount *mnt,
8114                          struct dentry *dentry, struct kstat *stat)
8115 {
8116         u64 delalloc_bytes;
8117         struct inode *inode = dentry->d_inode;
8118         u32 blocksize = inode->i_sb->s_blocksize;
8119
8120         generic_fillattr(inode, stat);
8121         stat->dev = BTRFS_I(inode)->root->anon_dev;
8122         stat->blksize = PAGE_CACHE_SIZE;
8123
8124         spin_lock(&BTRFS_I(inode)->lock);
8125         delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8126         spin_unlock(&BTRFS_I(inode)->lock);
8127         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8128                         ALIGN(delalloc_bytes, blocksize)) >> 9;
8129         return 0;
8130 }
8131
8132 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8133                            struct inode *new_dir, struct dentry *new_dentry)
8134 {
8135         struct btrfs_trans_handle *trans;
8136         struct btrfs_root *root = BTRFS_I(old_dir)->root;
8137         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8138         struct inode *new_inode = new_dentry->d_inode;
8139         struct inode *old_inode = old_dentry->d_inode;
8140         struct timespec ctime = CURRENT_TIME;
8141         u64 index = 0;
8142         u64 root_objectid;
8143         int ret;
8144         u64 old_ino = btrfs_ino(old_inode);
8145
8146         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8147                 return -EPERM;
8148
8149         /* we only allow rename subvolume link between subvolumes */
8150         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8151                 return -EXDEV;
8152
8153         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8154             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8155                 return -ENOTEMPTY;
8156
8157         if (S_ISDIR(old_inode->i_mode) && new_inode &&
8158             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8159                 return -ENOTEMPTY;
8160
8161
8162         /* check for collisions, even if the  name isn't there */
8163         ret = btrfs_check_dir_item_collision(root, new_dir->i_ino,
8164                              new_dentry->d_name.name,
8165                              new_dentry->d_name.len);
8166
8167         if (ret) {
8168                 if (ret == -EEXIST) {
8169                         /* we shouldn't get
8170                          * eexist without a new_inode */
8171                         if (!new_inode) {
8172                                 WARN_ON(1);
8173                                 return ret;
8174                         }
8175                 } else {
8176                         /* maybe -EOVERFLOW */
8177                         return ret;
8178                 }
8179         }
8180         ret = 0;
8181
8182         /*
8183          * we're using rename to replace one file with another.
8184          * and the replacement file is large.  Start IO on it now so
8185          * we don't add too much work to the end of the transaction
8186          */
8187         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8188             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8189                 filemap_flush(old_inode->i_mapping);
8190
8191         /* close the racy window with snapshot create/destroy ioctl */
8192         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8193                 down_read(&root->fs_info->subvol_sem);
8194         /*
8195          * We want to reserve the absolute worst case amount of items.  So if
8196          * both inodes are subvols and we need to unlink them then that would
8197          * require 4 item modifications, but if they are both normal inodes it
8198          * would require 5 item modifications, so we'll assume their normal
8199          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8200          * should cover the worst case number of items we'll modify.
8201          */
8202         trans = btrfs_start_transaction(root, 11);
8203         if (IS_ERR(trans)) {
8204                 ret = PTR_ERR(trans);
8205                 goto out_notrans;
8206         }
8207
8208         if (dest != root)
8209                 btrfs_record_root_in_trans(trans, dest);
8210
8211         ret = btrfs_set_inode_index(new_dir, &index);
8212         if (ret)
8213                 goto out_fail;
8214
8215         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8216                 /* force full log commit if subvolume involved. */
8217                 root->fs_info->last_trans_log_full_commit = trans->transid;
8218         } else {
8219                 ret = btrfs_insert_inode_ref(trans, dest,
8220                                              new_dentry->d_name.name,
8221                                              new_dentry->d_name.len,
8222                                              old_ino,
8223                                              btrfs_ino(new_dir), index);
8224                 if (ret)
8225                         goto out_fail;
8226                 /*
8227                  * this is an ugly little race, but the rename is required
8228                  * to make sure that if we crash, the inode is either at the
8229                  * old name or the new one.  pinning the log transaction lets
8230                  * us make sure we don't allow a log commit to come in after
8231                  * we unlink the name but before we add the new name back in.
8232                  */
8233                 btrfs_pin_log_trans(root);
8234         }
8235         /*
8236          * make sure the inode gets flushed if it is replacing
8237          * something.
8238          */
8239         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8240                 btrfs_add_ordered_operation(trans, root, old_inode);
8241
8242         inode_inc_iversion(old_dir);
8243         inode_inc_iversion(new_dir);
8244         inode_inc_iversion(old_inode);
8245         old_dir->i_ctime = old_dir->i_mtime = ctime;
8246         new_dir->i_ctime = new_dir->i_mtime = ctime;
8247         old_inode->i_ctime = ctime;
8248
8249         if (old_dentry->d_parent != new_dentry->d_parent)
8250                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8251
8252         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8253                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8254                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8255                                         old_dentry->d_name.name,
8256                                         old_dentry->d_name.len);
8257         } else {
8258                 ret = __btrfs_unlink_inode(trans, root, old_dir,
8259                                         old_dentry->d_inode,
8260                                         old_dentry->d_name.name,
8261                                         old_dentry->d_name.len);
8262                 if (!ret)
8263                         ret = btrfs_update_inode(trans, root, old_inode);
8264         }
8265         if (ret) {
8266                 btrfs_abort_transaction(trans, root, ret);
8267                 goto out_fail;
8268         }
8269
8270         if (new_inode) {
8271                 inode_inc_iversion(new_inode);
8272                 new_inode->i_ctime = CURRENT_TIME;
8273                 if (unlikely(btrfs_ino(new_inode) ==
8274                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8275                         root_objectid = BTRFS_I(new_inode)->location.objectid;
8276                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
8277                                                 root_objectid,
8278                                                 new_dentry->d_name.name,
8279                                                 new_dentry->d_name.len);
8280                         BUG_ON(new_inode->i_nlink == 0);
8281                 } else {
8282                         ret = btrfs_unlink_inode(trans, dest, new_dir,
8283                                                  new_dentry->d_inode,
8284                                                  new_dentry->d_name.name,
8285                                                  new_dentry->d_name.len);
8286                 }
8287                 if (!ret && new_inode->i_nlink == 0) {
8288                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8289                         BUG_ON(ret);
8290                 }
8291                 if (ret) {
8292                         btrfs_abort_transaction(trans, root, ret);
8293                         goto out_fail;
8294                 }
8295         }
8296
8297         ret = btrfs_add_link(trans, new_dir, old_inode,
8298                              new_dentry->d_name.name,
8299                              new_dentry->d_name.len, 0, index);
8300         if (ret) {
8301                 btrfs_abort_transaction(trans, root, ret);
8302                 goto out_fail;
8303         }
8304
8305         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8306                 struct dentry *parent = new_dentry->d_parent;
8307                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8308                 btrfs_end_log_trans(root);
8309         }
8310 out_fail:
8311         btrfs_end_transaction(trans, root);
8312 out_notrans:
8313         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8314                 up_read(&root->fs_info->subvol_sem);
8315
8316         return ret;
8317 }
8318
8319 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8320 {
8321         struct btrfs_delalloc_work *delalloc_work;
8322
8323         delalloc_work = container_of(work, struct btrfs_delalloc_work,
8324                                      work);
8325         if (delalloc_work->wait)
8326                 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
8327         else
8328                 filemap_flush(delalloc_work->inode->i_mapping);
8329
8330         if (delalloc_work->delay_iput)
8331                 btrfs_add_delayed_iput(delalloc_work->inode);
8332         else
8333                 iput(delalloc_work->inode);
8334         complete(&delalloc_work->completion);
8335 }
8336
8337 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8338                                                     int wait, int delay_iput)
8339 {
8340         struct btrfs_delalloc_work *work;
8341
8342         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8343         if (!work)
8344                 return NULL;
8345
8346         init_completion(&work->completion);
8347         INIT_LIST_HEAD(&work->list);
8348         work->inode = inode;
8349         work->wait = wait;
8350         work->delay_iput = delay_iput;
8351         work->work.func = btrfs_run_delalloc_work;
8352
8353         return work;
8354 }
8355
8356 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8357 {
8358         wait_for_completion(&work->completion);
8359         kmem_cache_free(btrfs_delalloc_work_cachep, work);
8360 }
8361
8362 /*
8363  * some fairly slow code that needs optimization. This walks the list
8364  * of all the inodes with pending delalloc and forces them to disk.
8365  */
8366 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8367 {
8368         struct btrfs_inode *binode;
8369         struct inode *inode;
8370         struct btrfs_delalloc_work *work, *next;
8371         struct list_head works;
8372         struct list_head splice;
8373         int ret = 0;
8374
8375         INIT_LIST_HEAD(&works);
8376         INIT_LIST_HEAD(&splice);
8377
8378         spin_lock(&root->delalloc_lock);
8379         list_splice_init(&root->delalloc_inodes, &splice);
8380         while (!list_empty(&splice)) {
8381                 binode = list_entry(splice.next, struct btrfs_inode,
8382                                     delalloc_inodes);
8383
8384                 list_move_tail(&binode->delalloc_inodes,
8385                                &root->delalloc_inodes);
8386                 inode = igrab(&binode->vfs_inode);
8387                 if (!inode) {
8388                         cond_resched_lock(&root->delalloc_lock);
8389                         continue;
8390                 }
8391                 spin_unlock(&root->delalloc_lock);
8392
8393                 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8394                 if (unlikely(!work)) {
8395                         ret = -ENOMEM;
8396                         goto out;
8397                 }
8398                 list_add_tail(&work->list, &works);
8399                 btrfs_queue_worker(&root->fs_info->flush_workers,
8400                                    &work->work);
8401
8402                 cond_resched();
8403                 spin_lock(&root->delalloc_lock);
8404         }
8405         spin_unlock(&root->delalloc_lock);
8406
8407         list_for_each_entry_safe(work, next, &works, list) {
8408                 list_del_init(&work->list);
8409                 btrfs_wait_and_free_delalloc_work(work);
8410         }
8411         return 0;
8412 out:
8413         list_for_each_entry_safe(work, next, &works, list) {
8414                 list_del_init(&work->list);
8415                 btrfs_wait_and_free_delalloc_work(work);
8416         }
8417
8418         if (!list_empty_careful(&splice)) {
8419                 spin_lock(&root->delalloc_lock);
8420                 list_splice_tail(&splice, &root->delalloc_inodes);
8421                 spin_unlock(&root->delalloc_lock);
8422         }
8423         return ret;
8424 }
8425
8426 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8427 {
8428         int ret;
8429
8430         if (root->fs_info->sb->s_flags & MS_RDONLY)
8431                 return -EROFS;
8432
8433         ret = __start_delalloc_inodes(root, delay_iput);
8434         /*
8435          * the filemap_flush will queue IO into the worker threads, but
8436          * we have to make sure the IO is actually started and that
8437          * ordered extents get created before we return
8438          */
8439         atomic_inc(&root->fs_info->async_submit_draining);
8440         while (atomic_read(&root->fs_info->nr_async_submits) ||
8441               atomic_read(&root->fs_info->async_delalloc_pages)) {
8442                 wait_event(root->fs_info->async_submit_wait,
8443                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8444                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8445         }
8446         atomic_dec(&root->fs_info->async_submit_draining);
8447         return ret;
8448 }
8449
8450 int btrfs_start_all_delalloc_inodes(struct btrfs_fs_info *fs_info,
8451                                     int delay_iput)
8452 {
8453         struct btrfs_root *root;
8454         struct list_head splice;
8455         int ret;
8456
8457         if (fs_info->sb->s_flags & MS_RDONLY)
8458                 return -EROFS;
8459
8460         INIT_LIST_HEAD(&splice);
8461
8462         spin_lock(&fs_info->delalloc_root_lock);
8463         list_splice_init(&fs_info->delalloc_roots, &splice);
8464         while (!list_empty(&splice)) {
8465                 root = list_first_entry(&splice, struct btrfs_root,
8466                                         delalloc_root);
8467                 root = btrfs_grab_fs_root(root);
8468                 BUG_ON(!root);
8469                 list_move_tail(&root->delalloc_root,
8470                                &fs_info->delalloc_roots);
8471                 spin_unlock(&fs_info->delalloc_root_lock);
8472
8473                 ret = __start_delalloc_inodes(root, delay_iput);
8474                 btrfs_put_fs_root(root);
8475                 if (ret)
8476                         goto out;
8477
8478                 spin_lock(&fs_info->delalloc_root_lock);
8479         }
8480         spin_unlock(&fs_info->delalloc_root_lock);
8481
8482         atomic_inc(&fs_info->async_submit_draining);
8483         while (atomic_read(&fs_info->nr_async_submits) ||
8484               atomic_read(&fs_info->async_delalloc_pages)) {
8485                 wait_event(fs_info->async_submit_wait,
8486                    (atomic_read(&fs_info->nr_async_submits) == 0 &&
8487                     atomic_read(&fs_info->async_delalloc_pages) == 0));
8488         }
8489         atomic_dec(&fs_info->async_submit_draining);
8490         return 0;
8491 out:
8492         if (!list_empty_careful(&splice)) {
8493                 spin_lock(&fs_info->delalloc_root_lock);
8494                 list_splice_tail(&splice, &fs_info->delalloc_roots);
8495                 spin_unlock(&fs_info->delalloc_root_lock);
8496         }
8497         return ret;
8498 }
8499
8500 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8501                          const char *symname)
8502 {
8503         struct btrfs_trans_handle *trans;
8504         struct btrfs_root *root = BTRFS_I(dir)->root;
8505         struct btrfs_path *path;
8506         struct btrfs_key key;
8507         struct inode *inode = NULL;
8508         int err;
8509         int drop_inode = 0;
8510         u64 objectid;
8511         u64 index = 0 ;
8512         int name_len;
8513         int datasize;
8514         unsigned long ptr;
8515         struct btrfs_file_extent_item *ei;
8516         struct extent_buffer *leaf;
8517
8518         name_len = strlen(symname) + 1;
8519         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8520                 return -ENAMETOOLONG;
8521
8522         /*
8523          * 2 items for inode item and ref
8524          * 2 items for dir items
8525          * 1 item for xattr if selinux is on
8526          */
8527         trans = btrfs_start_transaction(root, 5);
8528         if (IS_ERR(trans))
8529                 return PTR_ERR(trans);
8530
8531         err = btrfs_find_free_ino(root, &objectid);
8532         if (err)
8533                 goto out_unlock;
8534
8535         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8536                                 dentry->d_name.len, btrfs_ino(dir), objectid,
8537                                 S_IFLNK|S_IRWXUGO, &index);
8538         if (IS_ERR(inode)) {
8539                 err = PTR_ERR(inode);
8540                 goto out_unlock;
8541         }
8542
8543         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8544         if (err) {
8545                 drop_inode = 1;
8546                 goto out_unlock;
8547         }
8548
8549         /*
8550         * If the active LSM wants to access the inode during
8551         * d_instantiate it needs these. Smack checks to see
8552         * if the filesystem supports xattrs by looking at the
8553         * ops vector.
8554         */
8555         inode->i_fop = &btrfs_file_operations;
8556         inode->i_op = &btrfs_file_inode_operations;
8557
8558         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8559         if (err)
8560                 drop_inode = 1;
8561         else {
8562                 inode->i_mapping->a_ops = &btrfs_aops;
8563                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8564                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8565         }
8566         if (drop_inode)
8567                 goto out_unlock;
8568
8569         path = btrfs_alloc_path();
8570         if (!path) {
8571                 err = -ENOMEM;
8572                 drop_inode = 1;
8573                 goto out_unlock;
8574         }
8575         key.objectid = btrfs_ino(inode);
8576         key.offset = 0;
8577         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8578         datasize = btrfs_file_extent_calc_inline_size(name_len);
8579         err = btrfs_insert_empty_item(trans, root, path, &key,
8580                                       datasize);
8581         if (err) {
8582                 drop_inode = 1;
8583                 btrfs_free_path(path);
8584                 goto out_unlock;
8585         }
8586         leaf = path->nodes[0];
8587         ei = btrfs_item_ptr(leaf, path->slots[0],
8588                             struct btrfs_file_extent_item);
8589         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8590         btrfs_set_file_extent_type(leaf, ei,
8591                                    BTRFS_FILE_EXTENT_INLINE);
8592         btrfs_set_file_extent_encryption(leaf, ei, 0);
8593         btrfs_set_file_extent_compression(leaf, ei, 0);
8594         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8595         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8596
8597         ptr = btrfs_file_extent_inline_start(ei);
8598         write_extent_buffer(leaf, symname, ptr, name_len);
8599         btrfs_mark_buffer_dirty(leaf);
8600         btrfs_free_path(path);
8601
8602         inode->i_op = &btrfs_symlink_inode_operations;
8603         inode->i_mapping->a_ops = &btrfs_symlink_aops;
8604         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8605         inode_set_bytes(inode, name_len);
8606         btrfs_i_size_write(inode, name_len - 1);
8607         err = btrfs_update_inode(trans, root, inode);
8608         if (err)
8609                 drop_inode = 1;
8610
8611 out_unlock:
8612         if (!err)
8613                 d_instantiate(dentry, inode);
8614         btrfs_end_transaction(trans, root);
8615         if (drop_inode) {
8616                 inode_dec_link_count(inode);
8617                 iput(inode);
8618         }
8619         btrfs_btree_balance_dirty(root);
8620         return err;
8621 }
8622
8623 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8624                                        u64 start, u64 num_bytes, u64 min_size,
8625                                        loff_t actual_len, u64 *alloc_hint,
8626                                        struct btrfs_trans_handle *trans)
8627 {
8628         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8629         struct extent_map *em;
8630         struct btrfs_root *root = BTRFS_I(inode)->root;
8631         struct btrfs_key ins;
8632         u64 cur_offset = start;
8633         u64 i_size;
8634         u64 cur_bytes;
8635         int ret = 0;
8636         bool own_trans = true;
8637
8638         if (trans)
8639                 own_trans = false;
8640         while (num_bytes > 0) {
8641                 if (own_trans) {
8642                         trans = btrfs_start_transaction(root, 3);
8643                         if (IS_ERR(trans)) {
8644                                 ret = PTR_ERR(trans);
8645                                 break;
8646                         }
8647                 }
8648
8649                 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8650                 cur_bytes = max(cur_bytes, min_size);
8651                 ret = btrfs_reserve_extent(trans, root, cur_bytes,
8652                                            min_size, 0, *alloc_hint, &ins, 1);
8653                 if (ret) {
8654                         if (own_trans)
8655                                 btrfs_end_transaction(trans, root);
8656                         break;
8657                 }
8658
8659                 ret = insert_reserved_file_extent(trans, inode,
8660                                                   cur_offset, ins.objectid,
8661                                                   ins.offset, ins.offset,
8662                                                   ins.offset, 0, 0, 0,
8663                                                   BTRFS_FILE_EXTENT_PREALLOC);
8664                 if (ret) {
8665                         btrfs_abort_transaction(trans, root, ret);
8666                         if (own_trans)
8667                                 btrfs_end_transaction(trans, root);
8668                         break;
8669                 }
8670                 btrfs_drop_extent_cache(inode, cur_offset,
8671                                         cur_offset + ins.offset -1, 0);
8672
8673                 em = alloc_extent_map();
8674                 if (!em) {
8675                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8676                                 &BTRFS_I(inode)->runtime_flags);
8677                         goto next;
8678                 }
8679
8680                 em->start = cur_offset;
8681                 em->orig_start = cur_offset;
8682                 em->len = ins.offset;
8683                 em->block_start = ins.objectid;
8684                 em->block_len = ins.offset;
8685                 em->orig_block_len = ins.offset;
8686                 em->ram_bytes = ins.offset;
8687                 em->bdev = root->fs_info->fs_devices->latest_bdev;
8688                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8689                 em->generation = trans->transid;
8690
8691                 while (1) {
8692                         write_lock(&em_tree->lock);
8693                         ret = add_extent_mapping(em_tree, em, 1);
8694                         write_unlock(&em_tree->lock);
8695                         if (ret != -EEXIST)
8696                                 break;
8697                         btrfs_drop_extent_cache(inode, cur_offset,
8698                                                 cur_offset + ins.offset - 1,
8699                                                 0);
8700                 }
8701                 free_extent_map(em);
8702 next:
8703                 num_bytes -= ins.offset;
8704                 cur_offset += ins.offset;
8705                 *alloc_hint = ins.objectid + ins.offset;
8706
8707                 inode_inc_iversion(inode);
8708                 inode->i_ctime = CURRENT_TIME;
8709                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8710                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8711                     (actual_len > inode->i_size) &&
8712                     (cur_offset > inode->i_size)) {
8713                         if (cur_offset > actual_len)
8714                                 i_size = actual_len;
8715                         else
8716                                 i_size = cur_offset;
8717                         i_size_write(inode, i_size);
8718                         btrfs_ordered_update_i_size(inode, i_size, NULL);
8719                 }
8720
8721                 ret = btrfs_update_inode(trans, root, inode);
8722
8723                 if (ret) {
8724                         btrfs_abort_transaction(trans, root, ret);
8725                         if (own_trans)
8726                                 btrfs_end_transaction(trans, root);
8727                         break;
8728                 }
8729
8730                 if (own_trans)
8731                         btrfs_end_transaction(trans, root);
8732         }
8733         return ret;
8734 }
8735
8736 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8737                               u64 start, u64 num_bytes, u64 min_size,
8738                               loff_t actual_len, u64 *alloc_hint)
8739 {
8740         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8741                                            min_size, actual_len, alloc_hint,
8742                                            NULL);
8743 }
8744
8745 int btrfs_prealloc_file_range_trans(struct inode *inode,
8746                                     struct btrfs_trans_handle *trans, int mode,
8747                                     u64 start, u64 num_bytes, u64 min_size,
8748                                     loff_t actual_len, u64 *alloc_hint)
8749 {
8750         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8751                                            min_size, actual_len, alloc_hint, trans);
8752 }
8753
8754 static int btrfs_set_page_dirty(struct page *page)
8755 {
8756         return __set_page_dirty_nobuffers(page);
8757 }
8758
8759 static int btrfs_permission(struct inode *inode, int mask)
8760 {
8761         struct btrfs_root *root = BTRFS_I(inode)->root;
8762         umode_t mode = inode->i_mode;
8763
8764         if (mask & MAY_WRITE &&
8765             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8766                 if (btrfs_root_readonly(root))
8767                         return -EROFS;
8768                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8769                         return -EACCES;
8770         }
8771         return generic_permission(inode, mask);
8772 }
8773
8774 static const struct inode_operations btrfs_dir_inode_operations = {
8775         .getattr        = btrfs_getattr,
8776         .lookup         = btrfs_lookup,
8777         .create         = btrfs_create,
8778         .unlink         = btrfs_unlink,
8779         .link           = btrfs_link,
8780         .mkdir          = btrfs_mkdir,
8781         .rmdir          = btrfs_rmdir,
8782         .rename         = btrfs_rename,
8783         .symlink        = btrfs_symlink,
8784         .setattr        = btrfs_setattr,
8785         .mknod          = btrfs_mknod,
8786         .setxattr       = btrfs_setxattr,
8787         .getxattr       = btrfs_getxattr,
8788         .listxattr      = btrfs_listxattr,
8789         .removexattr    = btrfs_removexattr,
8790         .permission     = btrfs_permission,
8791         .get_acl        = btrfs_get_acl,
8792 };
8793 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8794         .lookup         = btrfs_lookup,
8795         .permission     = btrfs_permission,
8796         .get_acl        = btrfs_get_acl,
8797 };
8798
8799 static const struct file_operations btrfs_dir_file_operations = {
8800         .llseek         = generic_file_llseek,
8801         .read           = generic_read_dir,
8802         .readdir        = btrfs_real_readdir,
8803         .unlocked_ioctl = btrfs_ioctl,
8804 #ifdef CONFIG_COMPAT
8805         .compat_ioctl   = btrfs_ioctl,
8806 #endif
8807         .release        = btrfs_release_file,
8808         .fsync          = btrfs_sync_file,
8809 };
8810
8811 static struct extent_io_ops btrfs_extent_io_ops = {
8812         .fill_delalloc = run_delalloc_range,
8813         .submit_bio_hook = btrfs_submit_bio_hook,
8814         .merge_bio_hook = btrfs_merge_bio_hook,
8815         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
8816         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
8817         .writepage_start_hook = btrfs_writepage_start_hook,
8818         .set_bit_hook = btrfs_set_bit_hook,
8819         .clear_bit_hook = btrfs_clear_bit_hook,
8820         .merge_extent_hook = btrfs_merge_extent_hook,
8821         .split_extent_hook = btrfs_split_extent_hook,
8822 };
8823
8824 /*
8825  * btrfs doesn't support the bmap operation because swapfiles
8826  * use bmap to make a mapping of extents in the file.  They assume
8827  * these extents won't change over the life of the file and they
8828  * use the bmap result to do IO directly to the drive.
8829  *
8830  * the btrfs bmap call would return logical addresses that aren't
8831  * suitable for IO and they also will change frequently as COW
8832  * operations happen.  So, swapfile + btrfs == corruption.
8833  *
8834  * For now we're avoiding this by dropping bmap.
8835  */
8836 static const struct address_space_operations btrfs_aops = {
8837         .readpage       = btrfs_readpage,
8838         .writepage      = btrfs_writepage,
8839         .writepages     = btrfs_writepages,
8840         .readpages      = btrfs_readpages,
8841         .direct_IO      = btrfs_direct_IO,
8842         .invalidatepage = btrfs_invalidatepage,
8843         .releasepage    = btrfs_releasepage,
8844         .set_page_dirty = btrfs_set_page_dirty,
8845         .error_remove_page = generic_error_remove_page,
8846 };
8847
8848 static const struct address_space_operations btrfs_symlink_aops = {
8849         .readpage       = btrfs_readpage,
8850         .writepage      = btrfs_writepage,
8851         .invalidatepage = btrfs_invalidatepage,
8852         .releasepage    = btrfs_releasepage,
8853 };
8854
8855 static const struct inode_operations btrfs_file_inode_operations = {
8856         .getattr        = btrfs_getattr,
8857         .setattr        = btrfs_setattr,
8858         .setxattr       = btrfs_setxattr,
8859         .getxattr       = btrfs_getxattr,
8860         .listxattr      = btrfs_listxattr,
8861         .removexattr    = btrfs_removexattr,
8862         .permission     = btrfs_permission,
8863         .fiemap         = btrfs_fiemap,
8864         .get_acl        = btrfs_get_acl,
8865         .update_time    = btrfs_update_time,
8866 };
8867 static const struct inode_operations btrfs_special_inode_operations = {
8868         .getattr        = btrfs_getattr,
8869         .setattr        = btrfs_setattr,
8870         .permission     = btrfs_permission,
8871         .setxattr       = btrfs_setxattr,
8872         .getxattr       = btrfs_getxattr,
8873         .listxattr      = btrfs_listxattr,
8874         .removexattr    = btrfs_removexattr,
8875         .get_acl        = btrfs_get_acl,
8876         .update_time    = btrfs_update_time,
8877 };
8878 static const struct inode_operations btrfs_symlink_inode_operations = {
8879         .readlink       = generic_readlink,
8880         .follow_link    = page_follow_link_light,
8881         .put_link       = page_put_link,
8882         .getattr        = btrfs_getattr,
8883         .setattr        = btrfs_setattr,
8884         .permission     = btrfs_permission,
8885         .setxattr       = btrfs_setxattr,
8886         .getxattr       = btrfs_getxattr,
8887         .listxattr      = btrfs_listxattr,
8888         .removexattr    = btrfs_removexattr,
8889         .get_acl        = btrfs_get_acl,
8890         .update_time    = btrfs_update_time,
8891 };
8892
8893 const struct dentry_operations btrfs_dentry_operations = {
8894         .d_delete       = btrfs_dentry_delete,
8895         .d_release      = btrfs_dentry_release,
8896 };