fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/aio.h>
  36 #include <linux/bit_spinlock.h>
  37 #include <linux/xattr.h>
  38 #include <linux/posix_acl.h>
  39 #include <linux/falloc.h>
  40 #include <linux/slab.h>
  41 #include <linux/ratelimit.h>
  42 #include <linux/mount.h>
  43 #include <linux/btrfs.h>
  44 #include <linux/blkdev.h>
  45 #include "compat.h"
  46 #include "ctree.h"
  47 #include "disk-io.h"
  48 #include "transaction.h"
  49 #include "btrfs_inode.h"
  50 #include "print-tree.h"
  51 #include "ordered-data.h"
  52 #include "xattr.h"
  53 #include "tree-log.h"
  54 #include "volumes.h"
  55 #include "compression.h"
  56 #include "locking.h"
  57 #include "free-space-cache.h"
  58 #include "inode-map.h"
  59 #include "backref.h"
  60
  61 struct btrfs_iget_args {
  62         u64 ino;
  63         struct btrfs_root *root;
  64 };
  65
  66 static const struct inode_operations btrfs_dir_inode_operations;
  67 static const struct inode_operations btrfs_symlink_inode_operations;
  68 static const struct inode_operations btrfs_dir_ro_inode_operations;
  69 static const struct inode_operations btrfs_special_inode_operations;
  70 static const struct inode_operations btrfs_file_inode_operations;
  71 static const struct address_space_operations btrfs_aops;
  72 static const struct address_space_operations btrfs_symlink_aops;
  73 static const struct file_operations btrfs_dir_file_operations;
  74 static struct extent_io_ops btrfs_extent_io_ops;
  75
  76 static struct kmem_cache *btrfs_inode_cachep;
  77 static struct kmem_cache *btrfs_delalloc_work_cachep;
  78 struct kmem_cache *btrfs_trans_handle_cachep;
  79 struct kmem_cache *btrfs_transaction_cachep;
  80 struct kmem_cache *btrfs_path_cachep;
  81 struct kmem_cache *btrfs_free_space_cachep;
  82
  83 #define S_SHIFT 12
  84 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  85         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  86         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  87         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  88         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  89         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  90         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  91         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  92 };
  93
  94 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
  95 static int btrfs_truncate(struct inode *inode);
  96 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  97 static noinline int cow_file_range(struct inode *inode,
  98                                    struct page *locked_page,
  99                                    u64 start, u64 end, int *page_started,
 100                                    unsigned long *nr_written, int unlock);
 101 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
 102                                            u64 len, u64 orig_start,
 103                                            u64 block_start, u64 block_len,
 104                                            u64 orig_block_len, int type);
 105
 106 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 107                                      struct inode *inode,  struct inode *dir,
 108                                      const struct qstr *qstr)
 109 {
 110         int err;
 111
 112         err = btrfs_init_acl(trans, inode, dir);
 113         if (!err)
 114                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 115         return err;
 116 }
 117
 118 /*
 119  * this does all the hard work for inserting an inline extent into
 120  * the btree.  The caller should have done a btrfs_drop_extents so that
 121  * no overlapping inline items exist in the btree
 122  */
 123 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 124                                 struct btrfs_root *root, struct inode *inode,
 125                                 u64 start, size_t size, size_t compressed_size,
 126                                 int compress_type,
 127                                 struct page **compressed_pages)
 128 {
 129         struct btrfs_key key;
 130         struct btrfs_path *path;
 131         struct extent_buffer *leaf;
 132         struct page *page = NULL;
 133         char *kaddr;
 134         unsigned long ptr;
 135         struct btrfs_file_extent_item *ei;
 136         int err = 0;
 137         int ret;
 138         size_t cur_size = size;
 139         size_t datasize;
 140         unsigned long offset;
 141
 142         if (compressed_size && compressed_pages)
 143                 cur_size = compressed_size;
 144
 145         path = btrfs_alloc_path();
 146         if (!path)
 147                 return -ENOMEM;
 148
 149         path->leave_spinning = 1;
 150
 151         key.objectid = btrfs_ino(inode);
 152         key.offset = start;
 153         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 154         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 155
 156         inode_add_bytes(inode, size);
 157         ret = btrfs_insert_empty_item(trans, root, path, &key,
 158                                       datasize);
 159         if (ret) {
 160                 err = ret;
 161                 goto fail;
 162         }
 163         leaf = path->nodes[0];
 164         ei = btrfs_item_ptr(leaf, path->slots[0],
 165                             struct btrfs_file_extent_item);
 166         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 167         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 168         btrfs_set_file_extent_encryption(leaf, ei, 0);
 169         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 170         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 171         ptr = btrfs_file_extent_inline_start(ei);
 172
 173         if (compress_type != BTRFS_COMPRESS_NONE) {
 174                 struct page *cpage;
 175                 int i = 0;
 176                 while (compressed_size > 0) {
 177                         cpage = compressed_pages[i];
 178                         cur_size = min_t(unsigned long, compressed_size,
 179                                        PAGE_CACHE_SIZE);
 180
 181                         kaddr = kmap_atomic(cpage);
 182                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 183                         kunmap_atomic(kaddr);
 184
 185                         i++;
 186                         ptr += cur_size;
 187                         compressed_size -= cur_size;
 188                 }
 189                 btrfs_set_file_extent_compression(leaf, ei,
 190                                                   compress_type);
 191         } else {
 192                 page = find_get_page(inode->i_mapping,
 193                                      start >> PAGE_CACHE_SHIFT);
 194                 btrfs_set_file_extent_compression(leaf, ei, 0);
 195                 kaddr = kmap_atomic(page);
 196                 offset = start & (PAGE_CACHE_SIZE - 1);
 197                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 198                 kunmap_atomic(kaddr);
 199                 page_cache_release(page);
 200         }
 201         btrfs_mark_buffer_dirty(leaf);
 202         btrfs_free_path(path);
 203
 204         /*
 205          * we're an inline extent, so nobody can
 206          * extend the file past i_size without locking
 207          * a page we already have locked.
 208          *
 209          * We must do any isize and inode updates
 210          * before we unlock the pages.  Otherwise we
 211          * could end up racing with unlink.
 212          */
 213         BTRFS_I(inode)->disk_i_size = inode->i_size;
 214         ret = btrfs_update_inode(trans, root, inode);
 215
 216         return ret;
 217 fail:
 218         btrfs_free_path(path);
 219         return err;
 220 }
 221
 222
 223 /*
 224  * conditionally insert an inline extent into the file.  This
 225  * does the checks required to make sure the data is small enough
 226  * to fit as an inline extent.
 227  */
 228 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 229                                  struct btrfs_root *root,
 230                                  struct inode *inode, u64 start, u64 end,
 231                                  size_t compressed_size, int compress_type,
 232                                  struct page **compressed_pages)
 233 {
 234         u64 isize = i_size_read(inode);
 235         u64 actual_end = min(end + 1, isize);
 236         u64 inline_len = actual_end - start;
 237         u64 aligned_end = ALIGN(end, root->sectorsize);
 238         u64 data_len = inline_len;
 239         int ret;
 240
 241         if (compressed_size)
 242                 data_len = compressed_size;
 243
 244         if (start > 0 ||
 245             actual_end >= PAGE_CACHE_SIZE ||
 246             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 247             (!compressed_size &&
 248             (actual_end & (root->sectorsize - 1)) == 0) ||
 249             end + 1 < isize ||
 250             data_len > root->fs_info->max_inline) {
 251                 return 1;
 252         }
 253
 254         ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
 255         if (ret)
 256                 return ret;
 257
 258         if (isize > actual_end)
 259                 inline_len = min_t(u64, isize, actual_end);
 260         ret = insert_inline_extent(trans, root, inode, start,
 261                                    inline_len, compressed_size,
 262                                    compress_type, compressed_pages);
 263         if (ret && ret != -ENOSPC) {
 264                 btrfs_abort_transaction(trans, root, ret);
 265                 return ret;
 266         } else if (ret == -ENOSPC) {
 267                 return 1;
 268         }
 269
 270         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
 271         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 272         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 273         return 0;
 274 }
 275
 276 struct async_extent {
 277         u64 start;
 278         u64 ram_size;
 279         u64 compressed_size;
 280         struct page **pages;
 281         unsigned long nr_pages;
 282         int compress_type;
 283         struct list_head list;
 284 };
 285
 286 struct async_cow {
 287         struct inode *inode;
 288         struct btrfs_root *root;
 289         struct page *locked_page;
 290         u64 start;
 291         u64 end;
 292         struct list_head extents;
 293         struct btrfs_work work;
 294 };
 295
 296 static noinline int add_async_extent(struct async_cow *cow,
 297                                      u64 start, u64 ram_size,
 298                                      u64 compressed_size,
 299                                      struct page **pages,
 300                                      unsigned long nr_pages,
 301                                      int compress_type)
 302 {
 303         struct async_extent *async_extent;
 304
 305         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 306         BUG_ON(!async_extent); /* -ENOMEM */
 307         async_extent->start = start;
 308         async_extent->ram_size = ram_size;
 309         async_extent->compressed_size = compressed_size;
 310         async_extent->pages = pages;
 311         async_extent->nr_pages = nr_pages;
 312         async_extent->compress_type = compress_type;
 313         list_add_tail(&async_extent->list, &cow->extents);
 314         return 0;
 315 }
 316
 317 /*
 318  * we create compressed extents in two phases.  The first
 319  * phase compresses a range of pages that have already been
 320  * locked (both pages and state bits are locked).
 321  *
 322  * This is done inside an ordered work queue, and the compression
 323  * is spread across many cpus.  The actual IO submission is step
 324  * two, and the ordered work queue takes care of making sure that
 325  * happens in the same order things were put onto the queue by
 326  * writepages and friends.
 327  *
 328  * If this code finds it can't get good compression, it puts an
 329  * entry onto the work queue to write the uncompressed bytes.  This
 330  * makes sure that both compressed inodes and uncompressed inodes
 331  * are written in the same order that the flusher thread sent them
 332  * down.
 333  */
 334 static noinline int compress_file_range(struct inode *inode,
 335                                         struct page *locked_page,
 336                                         u64 start, u64 end,
 337                                         struct async_cow *async_cow,
 338                                         int *num_added)
 339 {
 340         struct btrfs_root *root = BTRFS_I(inode)->root;
 341         struct btrfs_trans_handle *trans;
 342         u64 num_bytes;
 343         u64 blocksize = root->sectorsize;
 344         u64 actual_end;
 345         u64 isize = i_size_read(inode);
 346         int ret = 0;
 347         struct page **pages = NULL;
 348         unsigned long nr_pages;
 349         unsigned long nr_pages_ret = 0;
 350         unsigned long total_compressed = 0;
 351         unsigned long total_in = 0;
 352         unsigned long max_compressed = 128 * 1024;
 353         unsigned long max_uncompressed = 128 * 1024;
 354         int i;
 355         int will_compress;
 356         int compress_type = root->fs_info->compress_type;
 357         int redirty = 0;
 358
 359         /* if this is a small write inside eof, kick off a defrag */
 360         if ((end - start + 1) < 16 * 1024 &&
 361             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 362                 btrfs_add_inode_defrag(NULL, inode);
 363
 364         actual_end = min_t(u64, isize, end + 1);
 365 again:
 366         will_compress = 0;
 367         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 368         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 369
 370         /*
 371          * we don't want to send crud past the end of i_size through
 372          * compression, that's just a waste of CPU time.  So, if the
 373          * end of the file is before the start of our current
 374          * requested range of bytes, we bail out to the uncompressed
 375          * cleanup code that can deal with all of this.
 376          *
 377          * It isn't really the fastest way to fix things, but this is a
 378          * very uncommon corner.
 379          */
 380         if (actual_end <= start)
 381                 goto cleanup_and_bail_uncompressed;
 382
 383         total_compressed = actual_end - start;
 384
 385         /* we want to make sure that amount of ram required to uncompress
 386          * an extent is reasonable, so we limit the total size in ram
 387          * of a compressed extent to 128k.  This is a crucial number
 388          * because it also controls how easily we can spread reads across
 389          * cpus for decompression.
 390          *
 391          * We also want to make sure the amount of IO required to do
 392          * a random read is reasonably small, so we limit the size of
 393          * a compressed extent to 128k.
 394          */
 395         total_compressed = min(total_compressed, max_uncompressed);
 396         num_bytes = ALIGN(end - start + 1, blocksize);
 397         num_bytes = max(blocksize,  num_bytes);
 398         total_in = 0;
 399         ret = 0;
 400
 401         /*
 402          * we do compression for mount -o compress and when the
 403          * inode has not been flagged as nocompress.  This flag can
 404          * change at any time if we discover bad compression ratios.
 405          */
 406         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 407             (btrfs_test_opt(root, COMPRESS) ||
 408              (BTRFS_I(inode)->force_compress) ||
 409              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 410                 WARN_ON(pages);
 411                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 412                 if (!pages) {
 413                         /* just bail out to the uncompressed code */
 414                         goto cont;
 415                 }
 416
 417                 if (BTRFS_I(inode)->force_compress)
 418                         compress_type = BTRFS_I(inode)->force_compress;
 419
 420                 /*
 421                  * we need to call clear_page_dirty_for_io on each
 422                  * page in the range.  Otherwise applications with the file
 423                  * mmap'd can wander in and change the page contents while
 424                  * we are compressing them.
 425                  *
 426                  * If the compression fails for any reason, we set the pages
 427                  * dirty again later on.
 428                  */
 429                 extent_range_clear_dirty_for_io(inode, start, end);
 430                 redirty = 1;
 431                 ret = btrfs_compress_pages(compress_type,
 432                                            inode->i_mapping, start,
 433                                            total_compressed, pages,
 434                                            nr_pages, &nr_pages_ret,
 435                                            &total_in,
 436                                            &total_compressed,
 437                                            max_compressed);
 438
 439                 if (!ret) {
 440                         unsigned long offset = total_compressed &
 441                                 (PAGE_CACHE_SIZE - 1);
 442                         struct page *page = pages[nr_pages_ret - 1];
 443                         char *kaddr;
 444
 445                         /* zero the tail end of the last page, we might be
 446                          * sending it down to disk
 447                          */
 448                         if (offset) {
 449                                 kaddr = kmap_atomic(page);
 450                                 memset(kaddr + offset, 0,
 451                                        PAGE_CACHE_SIZE - offset);
 452                                 kunmap_atomic(kaddr);
 453                         }
 454                         will_compress = 1;
 455                 }
 456         }
 457 cont:
 458         if (start == 0) {
 459                 trans = btrfs_join_transaction(root);
 460                 if (IS_ERR(trans)) {
 461                         ret = PTR_ERR(trans);
 462                         trans = NULL;
 463                         goto cleanup_and_out;
 464                 }
 465                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 466
 467                 /* lets try to make an inline extent */
 468                 if (ret || total_in < (actual_end - start)) {
 469                         /* we didn't compress the entire range, try
 470                          * to make an uncompressed inline extent.
 471                          */
 472                         ret = cow_file_range_inline(trans, root, inode,
 473                                                     start, end, 0, 0, NULL);
 474                 } else {
 475                         /* try making a compressed inline extent */
 476                         ret = cow_file_range_inline(trans, root, inode,
 477                                                     start, end,
 478                                                     total_compressed,
 479                                                     compress_type, pages);
 480                 }
 481                 if (ret <= 0) {
 482                         /*
 483                          * inline extent creation worked or returned error,
 484                          * we don't need to create any more async work items.
 485                          * Unlock and free up our temp pages.
 486                          */
 487                         extent_clear_unlock_delalloc(inode,
 488                              &BTRFS_I(inode)->io_tree,
 489                              start, end, NULL,
 490                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 491                              EXTENT_CLEAR_DELALLOC |
 492                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 493
 494                         btrfs_end_transaction(trans, root);
 495                         goto free_pages_out;
 496                 }
 497                 btrfs_end_transaction(trans, root);
 498         }
 499
 500         if (will_compress) {
 501                 /*
 502                  * we aren't doing an inline extent round the compressed size
 503                  * up to a block size boundary so the allocator does sane
 504                  * things
 505                  */
 506                 total_compressed = ALIGN(total_compressed, blocksize);
 507
 508                 /*
 509                  * one last check to make sure the compression is really a
 510                  * win, compare the page count read with the blocks on disk
 511                  */
 512                 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
 513                 if (total_compressed >= total_in) {
 514                         will_compress = 0;
 515                 } else {
 516                         num_bytes = total_in;
 517                 }
 518         }
 519         if (!will_compress && pages) {
 520                 /*
 521                  * the compression code ran but failed to make things smaller,
 522                  * free any pages it allocated and our page pointer array
 523                  */
 524                 for (i = 0; i < nr_pages_ret; i++) {
 525                         WARN_ON(pages[i]->mapping);
 526                         page_cache_release(pages[i]);
 527                 }
 528                 kfree(pages);
 529                 pages = NULL;
 530                 total_compressed = 0;
 531                 nr_pages_ret = 0;
 532
 533                 /* flag the file so we don't compress in the future */
 534                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 535                     !(BTRFS_I(inode)->force_compress)) {
 536                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 537                 }
 538         }
 539         if (will_compress) {
 540                 *num_added += 1;
 541
 542                 /* the async work queues will take care of doing actual
 543                  * allocation on disk for these compressed pages,
 544                  * and will submit them to the elevator.
 545                  */
 546                 add_async_extent(async_cow, start, num_bytes,
 547                                  total_compressed, pages, nr_pages_ret,
 548                                  compress_type);
 549
 550                 if (start + num_bytes < end) {
 551                         start += num_bytes;
 552                         pages = NULL;
 553                         cond_resched();
 554                         goto again;
 555                 }
 556         } else {
 557 cleanup_and_bail_uncompressed:
 558                 /*
 559                  * No compression, but we still need to write the pages in
 560                  * the file we've been given so far.  redirty the locked
 561                  * page if it corresponds to our extent and set things up
 562                  * for the async work queue to run cow_file_range to do
 563                  * the normal delalloc dance
 564                  */
 565                 if (page_offset(locked_page) >= start &&
 566                     page_offset(locked_page) <= end) {
 567                         __set_page_dirty_nobuffers(locked_page);
 568                         /* unlocked later on in the async handlers */
 569                 }
 570                 if (redirty)
 571                         extent_range_redirty_for_io(inode, start, end);
 572                 add_async_extent(async_cow, start, end - start + 1,
 573                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 574                 *num_added += 1;
 575         }
 576
 577 out:
 578         return ret;
 579
 580 free_pages_out:
 581         for (i = 0; i < nr_pages_ret; i++) {
 582                 WARN_ON(pages[i]->mapping);
 583                 page_cache_release(pages[i]);
 584         }
 585         kfree(pages);
 586
 587         goto out;
 588
 589 cleanup_and_out:
 590         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 591                                      start, end, NULL,
 592                                      EXTENT_CLEAR_UNLOCK_PAGE |
 593                                      EXTENT_CLEAR_DIRTY |
 594                                      EXTENT_CLEAR_DELALLOC |
 595                                      EXTENT_SET_WRITEBACK |
 596                                      EXTENT_END_WRITEBACK);
 597         if (!trans || IS_ERR(trans))
 598                 btrfs_error(root->fs_info, ret, "Failed to join transaction");
 599         else
 600                 btrfs_abort_transaction(trans, root, ret);
 601         goto free_pages_out;
 602 }
 603
 604 /*
 605  * phase two of compressed writeback.  This is the ordered portion
 606  * of the code, which only gets called in the order the work was
 607  * queued.  We walk all the async extents created by compress_file_range
 608  * and send them down to the disk.
 609  */
 610 static noinline int submit_compressed_extents(struct inode *inode,
 611                                               struct async_cow *async_cow)
 612 {
 613         struct async_extent *async_extent;
 614         u64 alloc_hint = 0;
 615         struct btrfs_trans_handle *trans;
 616         struct btrfs_key ins;
 617         struct extent_map *em;
 618         struct btrfs_root *root = BTRFS_I(inode)->root;
 619         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 620         struct extent_io_tree *io_tree;
 621         int ret = 0;
 622
 623         if (list_empty(&async_cow->extents))
 624                 return 0;
 625
 626 again:
 627         while (!list_empty(&async_cow->extents)) {
 628                 async_extent = list_entry(async_cow->extents.next,
 629                                           struct async_extent, list);
 630                 list_del(&async_extent->list);
 631
 632                 io_tree = &BTRFS_I(inode)->io_tree;
 633
 634 retry:
 635                 /* did the compression code fall back to uncompressed IO? */
 636                 if (!async_extent->pages) {
 637                         int page_started = 0;
 638                         unsigned long nr_written = 0;
 639
 640                         lock_extent(io_tree, async_extent->start,
 641                                          async_extent->start +
 642                                          async_extent->ram_size - 1);
 643
 644                         /* allocate blocks */
 645                         ret = cow_file_range(inode, async_cow->locked_page,
 646                                              async_extent->start,
 647                                              async_extent->start +
 648                                              async_extent->ram_size - 1,
 649                                              &page_started, &nr_written, 0);
 650
 651                         /* JDM XXX */
 652
 653                         /*
 654                          * if page_started, cow_file_range inserted an
 655                          * inline extent and took care of all the unlocking
 656                          * and IO for us.  Otherwise, we need to submit
 657                          * all those pages down to the drive.
 658                          */
 659                         if (!page_started && !ret)
 660                                 extent_write_locked_range(io_tree,
 661                                                   inode, async_extent->start,
 662                                                   async_extent->start +
 663                                                   async_extent->ram_size - 1,
 664                                                   btrfs_get_extent,
 665                                                   WB_SYNC_ALL);
 666                         else if (ret)
 667                                 unlock_page(async_cow->locked_page);
 668                         kfree(async_extent);
 669                         cond_resched();
 670                         continue;
 671                 }
 672
 673                 lock_extent(io_tree, async_extent->start,
 674                             async_extent->start + async_extent->ram_size - 1);
 675
 676                 trans = btrfs_join_transaction(root);
 677                 if (IS_ERR(trans)) {
 678                         ret = PTR_ERR(trans);
 679                 } else {
 680                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 681                         ret = btrfs_reserve_extent(trans, root,
 682                                            async_extent->compressed_size,
 683                                            async_extent->compressed_size,
 684                                            0, alloc_hint, &ins, 1);
 685                         if (ret && ret != -ENOSPC)
 686                                 btrfs_abort_transaction(trans, root, ret);
 687                         btrfs_end_transaction(trans, root);
 688                 }
 689
 690                 if (ret) {
 691                         int i;
 692
 693                         for (i = 0; i < async_extent->nr_pages; i++) {
 694                                 WARN_ON(async_extent->pages[i]->mapping);
 695                                 page_cache_release(async_extent->pages[i]);
 696                         }
 697                         kfree(async_extent->pages);
 698                         async_extent->nr_pages = 0;
 699                         async_extent->pages = NULL;
 700
 701                         if (ret == -ENOSPC)
 702                                 goto retry;
 703                         goto out_free;
 704                 }
 705
 706                 /*
 707                  * here we're doing allocation and writeback of the
 708                  * compressed pages
 709                  */
 710                 btrfs_drop_extent_cache(inode, async_extent->start,
 711                                         async_extent->start +
 712                                         async_extent->ram_size - 1, 0);
 713
 714                 em = alloc_extent_map();
 715                 if (!em)
 716                         goto out_free_reserve;
 717                 em->start = async_extent->start;
 718                 em->len = async_extent->ram_size;
 719                 em->orig_start = em->start;
 720                 em->mod_start = em->start;
 721                 em->mod_len = em->len;
 722
 723                 em->block_start = ins.objectid;
 724                 em->block_len = ins.offset;
 725                 em->orig_block_len = ins.offset;
 726                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 727                 em->compress_type = async_extent->compress_type;
 728                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 729                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 730                 em->generation = -1;
 731
 732                 while (1) {
 733                         write_lock(&em_tree->lock);
 734                         ret = add_extent_mapping(em_tree, em);
 735                         if (!ret)
 736                                 list_move(&em->list,
 737                                           &em_tree->modified_extents);
 738                         write_unlock(&em_tree->lock);
 739                         if (ret != -EEXIST) {
 740                                 free_extent_map(em);
 741                                 break;
 742                         }
 743                         btrfs_drop_extent_cache(inode, async_extent->start,
 744                                                 async_extent->start +
 745                                                 async_extent->ram_size - 1, 0);
 746                 }
 747
 748                 if (ret)
 749                         goto out_free_reserve;
 750
 751                 ret = btrfs_add_ordered_extent_compress(inode,
 752                                                 async_extent->start,
 753                                                 ins.objectid,
 754                                                 async_extent->ram_size,
 755                                                 ins.offset,
 756                                                 BTRFS_ORDERED_COMPRESSED,
 757                                                 async_extent->compress_type);
 758                 if (ret)
 759                         goto out_free_reserve;
 760
 761                 /*
 762                  * clear dirty, set writeback and unlock the pages.
 763                  */
 764                 extent_clear_unlock_delalloc(inode,
 765                                 &BTRFS_I(inode)->io_tree,
 766                                 async_extent->start,
 767                                 async_extent->start +
 768                                 async_extent->ram_size - 1,
 769                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 770                                 EXTENT_CLEAR_UNLOCK |
 771                                 EXTENT_CLEAR_DELALLOC |
 772                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 773
 774                 ret = btrfs_submit_compressed_write(inode,
 775                                     async_extent->start,
 776                                     async_extent->ram_size,
 777                                     ins.objectid,
 778                                     ins.offset, async_extent->pages,
 779                                     async_extent->nr_pages);
 780                 alloc_hint = ins.objectid + ins.offset;
 781                 kfree(async_extent);
 782                 if (ret)
 783                         goto out;
 784                 cond_resched();
 785         }
 786         ret = 0;
 787 out:
 788         return ret;
 789 out_free_reserve:
 790         btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
 791 out_free:
 792         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 793                                      async_extent->start,
 794                                      async_extent->start +
 795                                      async_extent->ram_size - 1,
 796                                      NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 797                                      EXTENT_CLEAR_UNLOCK |
 798                                      EXTENT_CLEAR_DELALLOC |
 799                                      EXTENT_CLEAR_DIRTY |
 800                                      EXTENT_SET_WRITEBACK |
 801                                      EXTENT_END_WRITEBACK);
 802         kfree(async_extent);
 803         goto again;
 804 }
 805
 806 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 807                                       u64 num_bytes)
 808 {
 809         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 810         struct extent_map *em;
 811         u64 alloc_hint = 0;
 812
 813         read_lock(&em_tree->lock);
 814         em = search_extent_mapping(em_tree, start, num_bytes);
 815         if (em) {
 816                 /*
 817                  * if block start isn't an actual block number then find the
 818                  * first block in this inode and use that as a hint.  If that
 819                  * block is also bogus then just don't worry about it.
 820                  */
 821                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 822                         free_extent_map(em);
 823                         em = search_extent_mapping(em_tree, 0, 0);
 824                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 825                                 alloc_hint = em->block_start;
 826                         if (em)
 827                                 free_extent_map(em);
 828                 } else {
 829                         alloc_hint = em->block_start;
 830                         free_extent_map(em);
 831                 }
 832         }
 833         read_unlock(&em_tree->lock);
 834
 835         return alloc_hint;
 836 }
 837
 838 /*
 839  * when extent_io.c finds a delayed allocation range in the file,
 840  * the call backs end up in this code.  The basic idea is to
 841  * allocate extents on disk for the range, and create ordered data structs
 842  * in ram to track those extents.
 843  *
 844  * locked_page is the page that writepage had locked already.  We use
 845  * it to make sure we don't do extra locks or unlocks.
 846  *
 847  * *page_started is set to one if we unlock locked_page and do everything
 848  * required to start IO on it.  It may be clean and already done with
 849  * IO when we return.
 850  */
 851 static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
 852                                      struct inode *inode,
 853                                      struct btrfs_root *root,
 854                                      struct page *locked_page,
 855                                      u64 start, u64 end, int *page_started,
 856                                      unsigned long *nr_written,
 857                                      int unlock)
 858 {
 859         u64 alloc_hint = 0;
 860         u64 num_bytes;
 861         unsigned long ram_size;
 862         u64 disk_num_bytes;
 863         u64 cur_alloc_size;
 864         u64 blocksize = root->sectorsize;
 865         struct btrfs_key ins;
 866         struct extent_map *em;
 867         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 868         int ret = 0;
 869
 870         BUG_ON(btrfs_is_free_space_inode(inode));
 871
 872         num_bytes = ALIGN(end - start + 1, blocksize);
 873         num_bytes = max(blocksize,  num_bytes);
 874         disk_num_bytes = num_bytes;
 875
 876         /* if this is a small write inside eof, kick off defrag */
 877         if (num_bytes < 64 * 1024 &&
 878             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 879                 btrfs_add_inode_defrag(trans, inode);
 880
 881         if (start == 0) {
 882                 /* lets try to make an inline extent */
 883                 ret = cow_file_range_inline(trans, root, inode,
 884                                             start, end, 0, 0, NULL);
 885                 if (ret == 0) {
 886                         extent_clear_unlock_delalloc(inode,
 887                                      &BTRFS_I(inode)->io_tree,
 888                                      start, end, NULL,
 889                                      EXTENT_CLEAR_UNLOCK_PAGE |
 890                                      EXTENT_CLEAR_UNLOCK |
 891                                      EXTENT_CLEAR_DELALLOC |
 892                                      EXTENT_CLEAR_DIRTY |
 893                                      EXTENT_SET_WRITEBACK |
 894                                      EXTENT_END_WRITEBACK);
 895
 896                         *nr_written = *nr_written +
 897                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 898                         *page_started = 1;
 899                         goto out;
 900                 } else if (ret < 0) {
 901                         btrfs_abort_transaction(trans, root, ret);
 902                         goto out_unlock;
 903                 }
 904         }
 905
 906         BUG_ON(disk_num_bytes >
 907                btrfs_super_total_bytes(root->fs_info->super_copy));
 908
 909         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 910         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 911
 912         while (disk_num_bytes > 0) {
 913                 unsigned long op;
 914
 915                 cur_alloc_size = disk_num_bytes;
 916                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 917                                            root->sectorsize, 0, alloc_hint,
 918                                            &ins, 1);
 919                 if (ret < 0) {
 920                         btrfs_abort_transaction(trans, root, ret);
 921                         goto out_unlock;
 922                 }
 923
 924                 em = alloc_extent_map();
 925                 BUG_ON(!em); /* -ENOMEM */
 926                 em->start = start;
 927                 em->orig_start = em->start;
 928                 ram_size = ins.offset;
 929                 em->len = ins.offset;
 930                 em->mod_start = em->start;
 931                 em->mod_len = em->len;
 932
 933                 em->block_start = ins.objectid;
 934                 em->block_len = ins.offset;
 935                 em->orig_block_len = ins.offset;
 936                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 937                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 938                 em->generation = -1;
 939
 940                 while (1) {
 941                         write_lock(&em_tree->lock);
 942                         ret = add_extent_mapping(em_tree, em);
 943                         if (!ret)
 944                                 list_move(&em->list,
 945                                           &em_tree->modified_extents);
 946                         write_unlock(&em_tree->lock);
 947                         if (ret != -EEXIST) {
 948                                 free_extent_map(em);
 949                                 break;
 950                         }
 951                         btrfs_drop_extent_cache(inode, start,
 952                                                 start + ram_size - 1, 0);
 953                 }
 954
 955                 cur_alloc_size = ins.offset;
 956                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 957                                                ram_size, cur_alloc_size, 0);
 958                 BUG_ON(ret); /* -ENOMEM */
 959
 960                 if (root->root_key.objectid ==
 961                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 962                         ret = btrfs_reloc_clone_csums(inode, start,
 963                                                       cur_alloc_size);
 964                         if (ret) {
 965                                 btrfs_abort_transaction(trans, root, ret);
 966                                 goto out_unlock;
 967                         }
 968                 }
 969
 970                 if (disk_num_bytes < cur_alloc_size)
 971                         break;
 972
 973                 /* we're not doing compressed IO, don't unlock the first
 974                  * page (which the caller expects to stay locked), don't
 975                  * clear any dirty bits and don't set any writeback bits
 976                  *
 977                  * Do set the Private2 bit so we know this page was properly
 978                  * setup for writepage
 979                  */
 980                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 981                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 982                         EXTENT_SET_PRIVATE2;
 983
 984                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 985                                              start, start + ram_size - 1,
 986                                              locked_page, op);
 987                 disk_num_bytes -= cur_alloc_size;
 988                 num_bytes -= cur_alloc_size;
 989                 alloc_hint = ins.objectid + ins.offset;
 990                 start += cur_alloc_size;
 991         }
 992 out:
 993         return ret;
 994
 995 out_unlock:
 996         extent_clear_unlock_delalloc(inode,
 997                      &BTRFS_I(inode)->io_tree,
 998                      start, end, locked_page,
 999                      EXTENT_CLEAR_UNLOCK_PAGE |
1000                      EXTENT_CLEAR_UNLOCK |
1001                      EXTENT_CLEAR_DELALLOC |
1002                      EXTENT_CLEAR_DIRTY |
1003                      EXTENT_SET_WRITEBACK |
1004                      EXTENT_END_WRITEBACK);
1005
1006         goto out;
1007 }
1008
1009 static noinline int cow_file_range(struct inode *inode,
1010                                    struct page *locked_page,
1011                                    u64 start, u64 end, int *page_started,
1012                                    unsigned long *nr_written,
1013                                    int unlock)
1014 {
1015         struct btrfs_trans_handle *trans;
1016         struct btrfs_root *root = BTRFS_I(inode)->root;
1017         int ret;
1018
1019         trans = btrfs_join_transaction(root);
1020         if (IS_ERR(trans)) {
1021                 extent_clear_unlock_delalloc(inode,
1022                              &BTRFS_I(inode)->io_tree,
1023                              start, end, locked_page,
1024                              EXTENT_CLEAR_UNLOCK_PAGE |
1025                              EXTENT_CLEAR_UNLOCK |
1026                              EXTENT_CLEAR_DELALLOC |
1027                              EXTENT_CLEAR_DIRTY |
1028                              EXTENT_SET_WRITEBACK |
1029                              EXTENT_END_WRITEBACK);
1030                 return PTR_ERR(trans);
1031         }
1032         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1033
1034         ret = __cow_file_range(trans, inode, root, locked_page, start, end,
1035                                page_started, nr_written, unlock);
1036
1037         btrfs_end_transaction(trans, root);
1038
1039         return ret;
1040 }
1041
1042 /*
1043  * work queue call back to started compression on a file and pages
1044  */
1045 static noinline void async_cow_start(struct btrfs_work *work)
1046 {
1047         struct async_cow *async_cow;
1048         int num_added = 0;
1049         async_cow = container_of(work, struct async_cow, work);
1050
1051         compress_file_range(async_cow->inode, async_cow->locked_page,
1052                             async_cow->start, async_cow->end, async_cow,
1053                             &num_added);
1054         if (num_added == 0) {
1055                 btrfs_add_delayed_iput(async_cow->inode);
1056                 async_cow->inode = NULL;
1057         }
1058 }
1059
1060 /*
1061  * work queue call back to submit previously compressed pages
1062  */
1063 static noinline void async_cow_submit(struct btrfs_work *work)
1064 {
1065         struct async_cow *async_cow;
1066         struct btrfs_root *root;
1067         unsigned long nr_pages;
1068
1069         async_cow = container_of(work, struct async_cow, work);
1070
1071         root = async_cow->root;
1072         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1073                 PAGE_CACHE_SHIFT;
1074
1075         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1076             5 * 1024 * 1024 &&
1077             waitqueue_active(&root->fs_info->async_submit_wait))
1078                 wake_up(&root->fs_info->async_submit_wait);
1079
1080         if (async_cow->inode)
1081                 submit_compressed_extents(async_cow->inode, async_cow);
1082 }
1083
1084 static noinline void async_cow_free(struct btrfs_work *work)
1085 {
1086         struct async_cow *async_cow;
1087         async_cow = container_of(work, struct async_cow, work);
1088         if (async_cow->inode)
1089                 btrfs_add_delayed_iput(async_cow->inode);
1090         kfree(async_cow);
1091 }
1092
1093 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1094                                 u64 start, u64 end, int *page_started,
1095                                 unsigned long *nr_written)
1096 {
1097         struct async_cow *async_cow;
1098         struct btrfs_root *root = BTRFS_I(inode)->root;
1099         unsigned long nr_pages;
1100         u64 cur_end;
1101         int limit = 10 * 1024 * 1024;
1102
1103         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1104                          1, 0, NULL, GFP_NOFS);
1105         while (start < end) {
1106                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1107                 BUG_ON(!async_cow); /* -ENOMEM */
1108                 async_cow->inode = igrab(inode);
1109                 async_cow->root = root;
1110                 async_cow->locked_page = locked_page;
1111                 async_cow->start = start;
1112
1113                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1114                         cur_end = end;
1115                 else
1116                         cur_end = min(end, start + 512 * 1024 - 1);
1117
1118                 async_cow->end = cur_end;
1119                 INIT_LIST_HEAD(&async_cow->extents);
1120
1121                 async_cow->work.func = async_cow_start;
1122                 async_cow->work.ordered_func = async_cow_submit;
1123                 async_cow->work.ordered_free = async_cow_free;
1124                 async_cow->work.flags = 0;
1125
1126                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1127                         PAGE_CACHE_SHIFT;
1128                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1129
1130                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1131                                    &async_cow->work);
1132
1133                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1134                         wait_event(root->fs_info->async_submit_wait,
1135                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1136                             limit));
1137                 }
1138
1139                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1140                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1141                         wait_event(root->fs_info->async_submit_wait,
1142                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1143                            0));
1144                 }
1145
1146                 *nr_written += nr_pages;
1147                 start = cur_end + 1;
1148         }
1149         *page_started = 1;
1150         return 0;
1151 }
1152
1153 static noinline int csum_exist_in_range(struct btrfs_root *root,
1154                                         u64 bytenr, u64 num_bytes)
1155 {
1156         int ret;
1157         struct btrfs_ordered_sum *sums;
1158         LIST_HEAD(list);
1159
1160         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1161                                        bytenr + num_bytes - 1, &list, 0);
1162         if (ret == 0 && list_empty(&list))
1163                 return 0;
1164
1165         while (!list_empty(&list)) {
1166                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1167                 list_del(&sums->list);
1168                 kfree(sums);
1169         }
1170         return 1;
1171 }
1172
1173 /*
1174  * when nowcow writeback call back.  This checks for snapshots or COW copies
1175  * of the extents that exist in the file, and COWs the file as required.
1176  *
1177  * If no cow copies or snapshots exist, we write directly to the existing
1178  * blocks on disk
1179  */
1180 static noinline int run_delalloc_nocow(struct inode *inode,
1181                                        struct page *locked_page,
1182                               u64 start, u64 end, int *page_started, int force,
1183                               unsigned long *nr_written)
1184 {
1185         struct btrfs_root *root = BTRFS_I(inode)->root;
1186         struct btrfs_trans_handle *trans;
1187         struct extent_buffer *leaf;
1188         struct btrfs_path *path;
1189         struct btrfs_file_extent_item *fi;
1190         struct btrfs_key found_key;
1191         u64 cow_start;
1192         u64 cur_offset;
1193         u64 extent_end;
1194         u64 extent_offset;
1195         u64 disk_bytenr;
1196         u64 num_bytes;
1197         u64 disk_num_bytes;
1198         int extent_type;
1199         int ret, err;
1200         int type;
1201         int nocow;
1202         int check_prev = 1;
1203         bool nolock;
1204         u64 ino = btrfs_ino(inode);
1205
1206         path = btrfs_alloc_path();
1207         if (!path) {
1208                 extent_clear_unlock_delalloc(inode,
1209                              &BTRFS_I(inode)->io_tree,
1210                              start, end, locked_page,
1211                              EXTENT_CLEAR_UNLOCK_PAGE |
1212                              EXTENT_CLEAR_UNLOCK |
1213                              EXTENT_CLEAR_DELALLOC |
1214                              EXTENT_CLEAR_DIRTY |
1215                              EXTENT_SET_WRITEBACK |
1216                              EXTENT_END_WRITEBACK);
1217                 return -ENOMEM;
1218         }
1219
1220         nolock = btrfs_is_free_space_inode(inode);
1221
1222         if (nolock)
1223                 trans = btrfs_join_transaction_nolock(root);
1224         else
1225                 trans = btrfs_join_transaction(root);
1226
1227         if (IS_ERR(trans)) {
1228                 extent_clear_unlock_delalloc(inode,
1229                              &BTRFS_I(inode)->io_tree,
1230                              start, end, locked_page,
1231                              EXTENT_CLEAR_UNLOCK_PAGE |
1232                              EXTENT_CLEAR_UNLOCK |
1233                              EXTENT_CLEAR_DELALLOC |
1234                              EXTENT_CLEAR_DIRTY |
1235                              EXTENT_SET_WRITEBACK |
1236                              EXTENT_END_WRITEBACK);
1237                 btrfs_free_path(path);
1238                 return PTR_ERR(trans);
1239         }
1240
1241         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1242
1243         cow_start = (u64)-1;
1244         cur_offset = start;
1245         while (1) {
1246                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1247                                                cur_offset, 0);
1248                 if (ret < 0) {
1249                         btrfs_abort_transaction(trans, root, ret);
1250                         goto error;
1251                 }
1252                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1253                         leaf = path->nodes[0];
1254                         btrfs_item_key_to_cpu(leaf, &found_key,
1255                                               path->slots[0] - 1);
1256                         if (found_key.objectid == ino &&
1257                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1258                                 path->slots[0]--;
1259                 }
1260                 check_prev = 0;
1261 next_slot:
1262                 leaf = path->nodes[0];
1263                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1264                         ret = btrfs_next_leaf(root, path);
1265                         if (ret < 0) {
1266                                 btrfs_abort_transaction(trans, root, ret);
1267                                 goto error;
1268                         }
1269                         if (ret > 0)
1270                                 break;
1271                         leaf = path->nodes[0];
1272                 }
1273
1274                 nocow = 0;
1275                 disk_bytenr = 0;
1276                 num_bytes = 0;
1277                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1278
1279                 if (found_key.objectid > ino ||
1280                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1281                     found_key.offset > end)
1282                         break;
1283
1284                 if (found_key.offset > cur_offset) {
1285                         extent_end = found_key.offset;
1286                         extent_type = 0;
1287                         goto out_check;
1288                 }
1289
1290                 fi = btrfs_item_ptr(leaf, path->slots[0],
1291                                     struct btrfs_file_extent_item);
1292                 extent_type = btrfs_file_extent_type(leaf, fi);
1293
1294                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1295                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1296                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1297                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1298                         extent_end = found_key.offset +
1299                                 btrfs_file_extent_num_bytes(leaf, fi);
1300                         disk_num_bytes =
1301                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1302                         if (extent_end <= start) {
1303                                 path->slots[0]++;
1304                                 goto next_slot;
1305                         }
1306                         if (disk_bytenr == 0)
1307                                 goto out_check;
1308                         if (btrfs_file_extent_compression(leaf, fi) ||
1309                             btrfs_file_extent_encryption(leaf, fi) ||
1310                             btrfs_file_extent_other_encoding(leaf, fi))
1311                                 goto out_check;
1312                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1313                                 goto out_check;
1314                         if (btrfs_extent_readonly(root, disk_bytenr))
1315                                 goto out_check;
1316                         if (btrfs_cross_ref_exist(trans, root, ino,
1317                                                   found_key.offset -
1318                                                   extent_offset, disk_bytenr))
1319                                 goto out_check;
1320                         disk_bytenr += extent_offset;
1321                         disk_bytenr += cur_offset - found_key.offset;
1322                         num_bytes = min(end + 1, extent_end) - cur_offset;
1323                         /*
1324                          * force cow if csum exists in the range.
1325                          * this ensure that csum for a given extent are
1326                          * either valid or do not exist.
1327                          */
1328                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1329                                 goto out_check;
1330                         nocow = 1;
1331                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1332                         extent_end = found_key.offset +
1333                                 btrfs_file_extent_inline_len(leaf, fi);
1334                         extent_end = ALIGN(extent_end, root->sectorsize);
1335                 } else {
1336                         BUG_ON(1);
1337                 }
1338 out_check:
1339                 if (extent_end <= start) {
1340                         path->slots[0]++;
1341                         goto next_slot;
1342                 }
1343                 if (!nocow) {
1344                         if (cow_start == (u64)-1)
1345                                 cow_start = cur_offset;
1346                         cur_offset = extent_end;
1347                         if (cur_offset > end)
1348                                 break;
1349                         path->slots[0]++;
1350                         goto next_slot;
1351                 }
1352
1353                 btrfs_release_path(path);
1354                 if (cow_start != (u64)-1) {
1355                         ret = __cow_file_range(trans, inode, root, locked_page,
1356                                                cow_start, found_key.offset - 1,
1357                                                page_started, nr_written, 1);
1358                         if (ret) {
1359                                 btrfs_abort_transaction(trans, root, ret);
1360                                 goto error;
1361                         }
1362                         cow_start = (u64)-1;
1363                 }
1364
1365                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1366                         struct extent_map *em;
1367                         struct extent_map_tree *em_tree;
1368                         em_tree = &BTRFS_I(inode)->extent_tree;
1369                         em = alloc_extent_map();
1370                         BUG_ON(!em); /* -ENOMEM */
1371                         em->start = cur_offset;
1372                         em->orig_start = found_key.offset - extent_offset;
1373                         em->len = num_bytes;
1374                         em->block_len = num_bytes;
1375                         em->block_start = disk_bytenr;
1376                         em->orig_block_len = disk_num_bytes;
1377                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1378                         em->mod_start = em->start;
1379                         em->mod_len = em->len;
1380                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1381                         set_bit(EXTENT_FLAG_FILLING, &em->flags);
1382                         em->generation = -1;
1383                         while (1) {
1384                                 write_lock(&em_tree->lock);
1385                                 ret = add_extent_mapping(em_tree, em);
1386                                 if (!ret)
1387                                         list_move(&em->list,
1388                                                   &em_tree->modified_extents);
1389                                 write_unlock(&em_tree->lock);
1390                                 if (ret != -EEXIST) {
1391                                         free_extent_map(em);
1392                                         break;
1393                                 }
1394                                 btrfs_drop_extent_cache(inode, em->start,
1395                                                 em->start + em->len - 1, 0);
1396                         }
1397                         type = BTRFS_ORDERED_PREALLOC;
1398                 } else {
1399                         type = BTRFS_ORDERED_NOCOW;
1400                 }
1401
1402                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1403                                                num_bytes, num_bytes, type);
1404                 BUG_ON(ret); /* -ENOMEM */
1405
1406                 if (root->root_key.objectid ==
1407                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1408                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1409                                                       num_bytes);
1410                         if (ret) {
1411                                 btrfs_abort_transaction(trans, root, ret);
1412                                 goto error;
1413                         }
1414                 }
1415
1416                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1417                                 cur_offset, cur_offset + num_bytes - 1,
1418                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1419                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1420                                 EXTENT_SET_PRIVATE2);
1421                 cur_offset = extent_end;
1422                 if (cur_offset > end)
1423                         break;
1424         }
1425         btrfs_release_path(path);
1426
1427         if (cur_offset <= end && cow_start == (u64)-1) {
1428                 cow_start = cur_offset;
1429                 cur_offset = end;
1430         }
1431
1432         if (cow_start != (u64)-1) {
1433                 ret = __cow_file_range(trans, inode, root, locked_page,
1434                                        cow_start, end,
1435                                        page_started, nr_written, 1);
1436                 if (ret) {
1437                         btrfs_abort_transaction(trans, root, ret);
1438                         goto error;
1439                 }
1440         }
1441
1442 error:
1443         err = btrfs_end_transaction(trans, root);
1444         if (!ret)
1445                 ret = err;
1446
1447         if (ret && cur_offset < end)
1448                 extent_clear_unlock_delalloc(inode,
1449                              &BTRFS_I(inode)->io_tree,
1450                              cur_offset, end, locked_page,
1451                              EXTENT_CLEAR_UNLOCK_PAGE |
1452                              EXTENT_CLEAR_UNLOCK |
1453                              EXTENT_CLEAR_DELALLOC |
1454                              EXTENT_CLEAR_DIRTY |
1455                              EXTENT_SET_WRITEBACK |
1456                              EXTENT_END_WRITEBACK);
1457
1458         btrfs_free_path(path);
1459         return ret;
1460 }
1461
1462 /*
1463  * extent_io.c call back to do delayed allocation processing
1464  */
1465 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1466                               u64 start, u64 end, int *page_started,
1467                               unsigned long *nr_written)
1468 {
1469         int ret;
1470         struct btrfs_root *root = BTRFS_I(inode)->root;
1471
1472         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1473                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1474                                          page_started, 1, nr_written);
1475         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1476                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1477                                          page_started, 0, nr_written);
1478         } else if (!btrfs_test_opt(root, COMPRESS) &&
1479                    !(BTRFS_I(inode)->force_compress) &&
1480                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1481                 ret = cow_file_range(inode, locked_page, start, end,
1482                                       page_started, nr_written, 1);
1483         } else {
1484                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1485                         &BTRFS_I(inode)->runtime_flags);
1486                 ret = cow_file_range_async(inode, locked_page, start, end,
1487                                            page_started, nr_written);
1488         }
1489         return ret;
1490 }
1491
1492 static void btrfs_split_extent_hook(struct inode *inode,
1493                                     struct extent_state *orig, u64 split)
1494 {
1495         /* not delalloc, ignore it */
1496         if (!(orig->state & EXTENT_DELALLOC))
1497                 return;
1498
1499         spin_lock(&BTRFS_I(inode)->lock);
1500         BTRFS_I(inode)->outstanding_extents++;
1501         spin_unlock(&BTRFS_I(inode)->lock);
1502 }
1503
1504 /*
1505  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1506  * extents so we can keep track of new extents that are just merged onto old
1507  * extents, such as when we are doing sequential writes, so we can properly
1508  * account for the metadata space we'll need.
1509  */
1510 static void btrfs_merge_extent_hook(struct inode *inode,
1511                                     struct extent_state *new,
1512                                     struct extent_state *other)
1513 {
1514         /* not delalloc, ignore it */
1515         if (!(other->state & EXTENT_DELALLOC))
1516                 return;
1517
1518         spin_lock(&BTRFS_I(inode)->lock);
1519         BTRFS_I(inode)->outstanding_extents--;
1520         spin_unlock(&BTRFS_I(inode)->lock);
1521 }
1522
1523 /*
1524  * extent_io.c set_bit_hook, used to track delayed allocation
1525  * bytes in this file, and to maintain the list of inodes that
1526  * have pending delalloc work to be done.
1527  */
1528 static void btrfs_set_bit_hook(struct inode *inode,
1529                                struct extent_state *state, int *bits)
1530 {
1531
1532         /*
1533          * set_bit and clear bit hooks normally require _irqsave/restore
1534          * but in this case, we are only testing for the DELALLOC
1535          * bit, which is only set or cleared with irqs on
1536          */
1537         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1538                 struct btrfs_root *root = BTRFS_I(inode)->root;
1539                 u64 len = state->end + 1 - state->start;
1540                 bool do_list = !btrfs_is_free_space_inode(inode);
1541
1542                 if (*bits & EXTENT_FIRST_DELALLOC) {
1543                         *bits &= ~EXTENT_FIRST_DELALLOC;
1544                 } else {
1545                         spin_lock(&BTRFS_I(inode)->lock);
1546                         BTRFS_I(inode)->outstanding_extents++;
1547                         spin_unlock(&BTRFS_I(inode)->lock);
1548                 }
1549
1550                 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1551                                      root->fs_info->delalloc_batch);
1552                 spin_lock(&BTRFS_I(inode)->lock);
1553                 BTRFS_I(inode)->delalloc_bytes += len;
1554                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1555                                          &BTRFS_I(inode)->runtime_flags)) {
1556                         spin_lock(&root->fs_info->delalloc_lock);
1557                         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1558                                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1559                                               &root->fs_info->delalloc_inodes);
1560                                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1561                                         &BTRFS_I(inode)->runtime_flags);
1562                         }
1563                         spin_unlock(&root->fs_info->delalloc_lock);
1564                 }
1565                 spin_unlock(&BTRFS_I(inode)->lock);
1566         }
1567 }
1568
1569 /*
1570  * extent_io.c clear_bit_hook, see set_bit_hook for why
1571  */
1572 static void btrfs_clear_bit_hook(struct inode *inode,
1573                                  struct extent_state *state, int *bits)
1574 {
1575         /*
1576          * set_bit and clear bit hooks normally require _irqsave/restore
1577          * but in this case, we are only testing for the DELALLOC
1578          * bit, which is only set or cleared with irqs on
1579          */
1580         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1581                 struct btrfs_root *root = BTRFS_I(inode)->root;
1582                 u64 len = state->end + 1 - state->start;
1583                 bool do_list = !btrfs_is_free_space_inode(inode);
1584
1585                 if (*bits & EXTENT_FIRST_DELALLOC) {
1586                         *bits &= ~EXTENT_FIRST_DELALLOC;
1587                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1588                         spin_lock(&BTRFS_I(inode)->lock);
1589                         BTRFS_I(inode)->outstanding_extents--;
1590                         spin_unlock(&BTRFS_I(inode)->lock);
1591                 }
1592
1593                 if (*bits & EXTENT_DO_ACCOUNTING)
1594                         btrfs_delalloc_release_metadata(inode, len);
1595
1596                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1597                     && do_list)
1598                         btrfs_free_reserved_data_space(inode, len);
1599
1600                 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1601                                      root->fs_info->delalloc_batch);
1602                 spin_lock(&BTRFS_I(inode)->lock);
1603                 BTRFS_I(inode)->delalloc_bytes -= len;
1604                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1605                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1606                              &BTRFS_I(inode)->runtime_flags)) {
1607                         spin_lock(&root->fs_info->delalloc_lock);
1608                         if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1609                                 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1610                                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1611                                           &BTRFS_I(inode)->runtime_flags);
1612                         }
1613                         spin_unlock(&root->fs_info->delalloc_lock);
1614                 }
1615                 spin_unlock(&BTRFS_I(inode)->lock);
1616         }
1617 }
1618
1619 /*
1620  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1621  * we don't create bios that span stripes or chunks
1622  */
1623 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1624                          size_t size, struct bio *bio,
1625                          unsigned long bio_flags)
1626 {
1627         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1628         u64 logical = (u64)bio->bi_sector << 9;
1629         u64 length = 0;
1630         u64 map_length;
1631         int ret;
1632
1633         if (bio_flags & EXTENT_BIO_COMPRESSED)
1634                 return 0;
1635
1636         length = bio->bi_size;
1637         map_length = length;
1638         ret = btrfs_map_block(root->fs_info, rw, logical,
1639                               &map_length, NULL, 0);
1640         /* Will always return 0 with map_multi == NULL */
1641         BUG_ON(ret < 0);
1642         if (map_length < length + size)
1643                 return 1;
1644         return 0;
1645 }
1646
1647 /*
1648  * in order to insert checksums into the metadata in large chunks,
1649  * we wait until bio submission time.   All the pages in the bio are
1650  * checksummed and sums are attached onto the ordered extent record.
1651  *
1652  * At IO completion time the cums attached on the ordered extent record
1653  * are inserted into the btree
1654  */
1655 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1656                                     struct bio *bio, int mirror_num,
1657                                     unsigned long bio_flags,
1658                                     u64 bio_offset)
1659 {
1660         struct btrfs_root *root = BTRFS_I(inode)->root;
1661         int ret = 0;
1662
1663         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1664         BUG_ON(ret); /* -ENOMEM */
1665         return 0;
1666 }
1667
1668 /*
1669  * in order to insert checksums into the metadata in large chunks,
1670  * we wait until bio submission time.   All the pages in the bio are
1671  * checksummed and sums are attached onto the ordered extent record.
1672  *
1673  * At IO completion time the cums attached on the ordered extent record
1674  * are inserted into the btree
1675  */
1676 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1677                           int mirror_num, unsigned long bio_flags,
1678                           u64 bio_offset)
1679 {
1680         struct btrfs_root *root = BTRFS_I(inode)->root;
1681         int ret;
1682
1683         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1684         if (ret)
1685                 bio_endio(bio, ret);
1686         return ret;
1687 }
1688
1689 /*
1690  * extent_io.c submission hook. This does the right thing for csum calculation
1691  * on write, or reading the csums from the tree before a read
1692  */
1693 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1694                           int mirror_num, unsigned long bio_flags,
1695                           u64 bio_offset)
1696 {
1697         struct btrfs_root *root = BTRFS_I(inode)->root;
1698         int ret = 0;
1699         int skip_sum;
1700         int metadata = 0;
1701         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1702
1703         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1704
1705         if (btrfs_is_free_space_inode(inode))
1706                 metadata = 2;
1707
1708         if (!(rw & REQ_WRITE)) {
1709                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1710                 if (ret)
1711                         goto out;
1712
1713                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1714                         ret = btrfs_submit_compressed_read(inode, bio,
1715                                                            mirror_num,
1716                                                            bio_flags);
1717                         goto out;
1718                 } else if (!skip_sum) {
1719                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1720                         if (ret)
1721                                 goto out;
1722                 }
1723                 goto mapit;
1724         } else if (async && !skip_sum) {
1725                 /* csum items have already been cloned */
1726                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1727                         goto mapit;
1728                 /* we're doing a write, do the async checksumming */
1729                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1730                                    inode, rw, bio, mirror_num,
1731                                    bio_flags, bio_offset,
1732                                    __btrfs_submit_bio_start,
1733                                    __btrfs_submit_bio_done);
1734                 goto out;
1735         } else if (!skip_sum) {
1736                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1737                 if (ret)
1738                         goto out;
1739         }
1740
1741 mapit:
1742         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1743
1744 out:
1745         if (ret < 0)
1746                 bio_endio(bio, ret);
1747         return ret;
1748 }
1749
1750 /*
1751  * given a list of ordered sums record them in the inode.  This happens
1752  * at IO completion time based on sums calculated at bio submission time.
1753  */
1754 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1755                              struct inode *inode, u64 file_offset,
1756                              struct list_head *list)
1757 {
1758         struct btrfs_ordered_sum *sum;
1759
1760         list_for_each_entry(sum, list, list) {
1761                 trans->adding_csums = 1;
1762                 btrfs_csum_file_blocks(trans,
1763                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1764                 trans->adding_csums = 0;
1765         }
1766         return 0;
1767 }
1768
1769 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1770                               struct extent_state **cached_state)
1771 {
1772         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1773         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1774                                    cached_state, GFP_NOFS);
1775 }
1776
1777 /* see btrfs_writepage_start_hook for details on why this is required */
1778 struct btrfs_writepage_fixup {
1779         struct page *page;
1780         struct btrfs_work work;
1781 };
1782
1783 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1784 {
1785         struct btrfs_writepage_fixup *fixup;
1786         struct btrfs_ordered_extent *ordered;
1787         struct extent_state *cached_state = NULL;
1788         struct page *page;
1789         struct inode *inode;
1790         u64 page_start;
1791         u64 page_end;
1792         int ret;
1793
1794         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1795         page = fixup->page;
1796 again:
1797         lock_page(page);
1798         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1799                 ClearPageChecked(page);
1800                 goto out_page;
1801         }
1802
1803         inode = page->mapping->host;
1804         page_start = page_offset(page);
1805         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1806
1807         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1808                          &cached_state);
1809
1810         /* already ordered? We're done */
1811         if (PagePrivate2(page))
1812                 goto out;
1813
1814         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1815         if (ordered) {
1816                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1817                                      page_end, &cached_state, GFP_NOFS);
1818                 unlock_page(page);
1819                 btrfs_start_ordered_extent(inode, ordered, 1);
1820                 btrfs_put_ordered_extent(ordered);
1821                 goto again;
1822         }
1823
1824         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1825         if (ret) {
1826                 mapping_set_error(page->mapping, ret);
1827                 end_extent_writepage(page, ret, page_start, page_end);
1828                 ClearPageChecked(page);
1829                 goto out;
1830          }
1831
1832         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1833         ClearPageChecked(page);
1834         set_page_dirty(page);
1835 out:
1836         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1837                              &cached_state, GFP_NOFS);
1838 out_page:
1839         unlock_page(page);
1840         page_cache_release(page);
1841         kfree(fixup);
1842 }
1843
1844 /*
1845  * There are a few paths in the higher layers of the kernel that directly
1846  * set the page dirty bit without asking the filesystem if it is a
1847  * good idea.  This causes problems because we want to make sure COW
1848  * properly happens and the data=ordered rules are followed.
1849  *
1850  * In our case any range that doesn't have the ORDERED bit set
1851  * hasn't been properly setup for IO.  We kick off an async process
1852  * to fix it up.  The async helper will wait for ordered extents, set
1853  * the delalloc bit and make it safe to write the page.
1854  */
1855 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1856 {
1857         struct inode *inode = page->mapping->host;
1858         struct btrfs_writepage_fixup *fixup;
1859         struct btrfs_root *root = BTRFS_I(inode)->root;
1860
1861         /* this page is properly in the ordered list */
1862         if (TestClearPagePrivate2(page))
1863                 return 0;
1864
1865         if (PageChecked(page))
1866                 return -EAGAIN;
1867
1868         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1869         if (!fixup)
1870                 return -EAGAIN;
1871
1872         SetPageChecked(page);
1873         page_cache_get(page);
1874         fixup->work.func = btrfs_writepage_fixup_worker;
1875         fixup->page = page;
1876         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1877         return -EBUSY;
1878 }
1879
1880 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1881                                        struct inode *inode, u64 file_pos,
1882                                        u64 disk_bytenr, u64 disk_num_bytes,
1883                                        u64 num_bytes, u64 ram_bytes,
1884                                        u8 compression, u8 encryption,
1885                                        u16 other_encoding, int extent_type)
1886 {
1887         struct btrfs_root *root = BTRFS_I(inode)->root;
1888         struct btrfs_file_extent_item *fi;
1889         struct btrfs_path *path;
1890         struct extent_buffer *leaf;
1891         struct btrfs_key ins;
1892         int ret;
1893
1894         path = btrfs_alloc_path();
1895         if (!path)
1896                 return -ENOMEM;
1897
1898         path->leave_spinning = 1;
1899
1900         /*
1901          * we may be replacing one extent in the tree with another.
1902          * The new extent is pinned in the extent map, and we don't want
1903          * to drop it from the cache until it is completely in the btree.
1904          *
1905          * So, tell btrfs_drop_extents to leave this extent in the cache.
1906          * the caller is expected to unpin it and allow it to be merged
1907          * with the others.
1908          */
1909         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1910                                  file_pos + num_bytes, 0);
1911         if (ret)
1912                 goto out;
1913
1914         ins.objectid = btrfs_ino(inode);
1915         ins.offset = file_pos;
1916         ins.type = BTRFS_EXTENT_DATA_KEY;
1917         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1918         if (ret)
1919                 goto out;
1920         leaf = path->nodes[0];
1921         fi = btrfs_item_ptr(leaf, path->slots[0],
1922                             struct btrfs_file_extent_item);
1923         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1924         btrfs_set_file_extent_type(leaf, fi, extent_type);
1925         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1926         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1927         btrfs_set_file_extent_offset(leaf, fi, 0);
1928         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1929         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1930         btrfs_set_file_extent_compression(leaf, fi, compression);
1931         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1932         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1933
1934         btrfs_mark_buffer_dirty(leaf);
1935         btrfs_release_path(path);
1936
1937         inode_add_bytes(inode, num_bytes);
1938
1939         ins.objectid = disk_bytenr;
1940         ins.offset = disk_num_bytes;
1941         ins.type = BTRFS_EXTENT_ITEM_KEY;
1942         ret = btrfs_alloc_reserved_file_extent(trans, root,
1943                                         root->root_key.objectid,
1944                                         btrfs_ino(inode), file_pos, &ins);
1945 out:
1946         btrfs_free_path(path);
1947
1948         return ret;
1949 }
1950
1951 /* snapshot-aware defrag */
1952 struct sa_defrag_extent_backref {
1953         struct rb_node node;
1954         struct old_sa_defrag_extent *old;
1955         u64 root_id;
1956         u64 inum;
1957         u64 file_pos;
1958         u64 extent_offset;
1959         u64 num_bytes;
1960         u64 generation;
1961 };
1962
1963 struct old_sa_defrag_extent {
1964         struct list_head list;
1965         struct new_sa_defrag_extent *new;
1966
1967         u64 extent_offset;
1968         u64 bytenr;
1969         u64 offset;
1970         u64 len;
1971         int count;
1972 };
1973
1974 struct new_sa_defrag_extent {
1975         struct rb_root root;
1976         struct list_head head;
1977         struct btrfs_path *path;
1978         struct inode *inode;
1979         u64 file_pos;
1980         u64 len;
1981         u64 bytenr;
1982         u64 disk_len;
1983         u8 compress_type;
1984 };
1985
1986 static int backref_comp(struct sa_defrag_extent_backref *b1,
1987                         struct sa_defrag_extent_backref *b2)
1988 {
1989         if (b1->root_id < b2->root_id)
1990                 return -1;
1991         else if (b1->root_id > b2->root_id)
1992                 return 1;
1993
1994         if (b1->inum < b2->inum)
1995                 return -1;
1996         else if (b1->inum > b2->inum)
1997                 return 1;
1998
1999         if (b1->file_pos < b2->file_pos)
2000                 return -1;
2001         else if (b1->file_pos > b2->file_pos)
2002                 return 1;
2003
2004         /*
2005          * [------------------------------] ===> (a range of space)
2006          *     |<--->|   |<---->| =============> (fs/file tree A)
2007          * |<---------------------------->| ===> (fs/file tree B)
2008          *
2009          * A range of space can refer to two file extents in one tree while
2010          * refer to only one file extent in another tree.
2011          *
2012          * So we may process a disk offset more than one time(two extents in A)
2013          * and locate at the same extent(one extent in B), then insert two same
2014          * backrefs(both refer to the extent in B).
2015          */
2016         return 0;
2017 }
2018
2019 static void backref_insert(struct rb_root *root,
2020                            struct sa_defrag_extent_backref *backref)
2021 {
2022         struct rb_node **p = &root->rb_node;
2023         struct rb_node *parent = NULL;
2024         struct sa_defrag_extent_backref *entry;
2025         int ret;
2026
2027         while (*p) {
2028                 parent = *p;
2029                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2030
2031                 ret = backref_comp(backref, entry);
2032                 if (ret < 0)
2033                         p = &(*p)->rb_left;
2034                 else
2035                         p = &(*p)->rb_right;
2036         }
2037
2038         rb_link_node(&backref->node, parent, p);
2039         rb_insert_color(&backref->node, root);
2040 }
2041
2042 /*
2043  * Note the backref might has changed, and in this case we just return 0.
2044  */
2045 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2046                                        void *ctx)
2047 {
2048         struct btrfs_file_extent_item *extent;
2049         struct btrfs_fs_info *fs_info;
2050         struct old_sa_defrag_extent *old = ctx;
2051         struct new_sa_defrag_extent *new = old->new;
2052         struct btrfs_path *path = new->path;
2053         struct btrfs_key key;
2054         struct btrfs_root *root;
2055         struct sa_defrag_extent_backref *backref;
2056         struct extent_buffer *leaf;
2057         struct inode *inode = new->inode;
2058         int slot;
2059         int ret;
2060         u64 extent_offset;
2061         u64 num_bytes;
2062
2063         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2064             inum == btrfs_ino(inode))
2065                 return 0;
2066
2067         key.objectid = root_id;
2068         key.type = BTRFS_ROOT_ITEM_KEY;
2069         key.offset = (u64)-1;
2070
2071         fs_info = BTRFS_I(inode)->root->fs_info;
2072         root = btrfs_read_fs_root_no_name(fs_info, &key);
2073         if (IS_ERR(root)) {
2074                 if (PTR_ERR(root) == -ENOENT)
2075                         return 0;
2076                 WARN_ON(1);
2077                 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2078                          inum, offset, root_id);
2079                 return PTR_ERR(root);
2080         }
2081
2082         key.objectid = inum;
2083         key.type = BTRFS_EXTENT_DATA_KEY;
2084         if (offset > (u64)-1 << 32)
2085                 key.offset = 0;
2086         else
2087                 key.offset = offset;
2088
2089         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2090         if (ret < 0) {
2091                 WARN_ON(1);
2092                 return ret;
2093         }
2094
2095         while (1) {
2096                 cond_resched();
2097
2098                 leaf = path->nodes[0];
2099                 slot = path->slots[0];
2100
2101                 if (slot >= btrfs_header_nritems(leaf)) {
2102                         ret = btrfs_next_leaf(root, path);
2103                         if (ret < 0) {
2104                                 goto out;
2105                         } else if (ret > 0) {
2106                                 ret = 0;
2107                                 goto out;
2108                         }
2109                         continue;
2110                 }
2111
2112                 path->slots[0]++;
2113
2114                 btrfs_item_key_to_cpu(leaf, &key, slot);
2115
2116                 if (key.objectid > inum)
2117                         goto out;
2118
2119                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2120                         continue;
2121
2122                 extent = btrfs_item_ptr(leaf, slot,
2123                                         struct btrfs_file_extent_item);
2124
2125                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2126                         continue;
2127
2128                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2129                 if (key.offset - extent_offset != offset)
2130                         continue;
2131
2132                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2133                 if (extent_offset >= old->extent_offset + old->offset +
2134                     old->len || extent_offset + num_bytes <=
2135                     old->extent_offset + old->offset)
2136                         continue;
2137
2138                 break;
2139         }
2140
2141         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2142         if (!backref) {
2143                 ret = -ENOENT;
2144                 goto out;
2145         }
2146
2147         backref->root_id = root_id;
2148         backref->inum = inum;
2149         backref->file_pos = offset + extent_offset;
2150         backref->num_bytes = num_bytes;
2151         backref->extent_offset = extent_offset;
2152         backref->generation = btrfs_file_extent_generation(leaf, extent);
2153         backref->old = old;
2154         backref_insert(&new->root, backref);
2155         old->count++;
2156 out:
2157         btrfs_release_path(path);
2158         WARN_ON(ret);
2159         return ret;
2160 }
2161
2162 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2163                                    struct new_sa_defrag_extent *new)
2164 {
2165         struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2166         struct old_sa_defrag_extent *old, *tmp;
2167         int ret;
2168
2169         new->path = path;
2170
2171         list_for_each_entry_safe(old, tmp, &new->head, list) {
2172                 ret = iterate_inodes_from_logical(old->bytenr, fs_info,
2173                                                   path, record_one_backref,
2174                                                   old);
2175                 BUG_ON(ret < 0 && ret != -ENOENT);
2176
2177                 /* no backref to be processed for this extent */
2178                 if (!old->count) {
2179                         list_del(&old->list);
2180                         kfree(old);
2181                 }
2182         }
2183
2184         if (list_empty(&new->head))
2185                 return false;
2186
2187         return true;
2188 }
2189
2190 static int relink_is_mergable(struct extent_buffer *leaf,
2191                               struct btrfs_file_extent_item *fi,
2192                               u64 disk_bytenr)
2193 {
2194         if (btrfs_file_extent_disk_bytenr(leaf, fi) != disk_bytenr)
2195                 return 0;
2196
2197         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2198                 return 0;
2199
2200         if (btrfs_file_extent_compression(leaf, fi) ||
2201             btrfs_file_extent_encryption(leaf, fi) ||
2202             btrfs_file_extent_other_encoding(leaf, fi))
2203                 return 0;
2204
2205         return 1;
2206 }
2207
2208 /*
2209  * Note the backref might has changed, and in this case we just return 0.
2210  */
2211 static noinline int relink_extent_backref(struct btrfs_path *path,
2212                                  struct sa_defrag_extent_backref *prev,
2213                                  struct sa_defrag_extent_backref *backref)
2214 {
2215         struct btrfs_file_extent_item *extent;
2216         struct btrfs_file_extent_item *item;
2217         struct btrfs_ordered_extent *ordered;
2218         struct btrfs_trans_handle *trans;
2219         struct btrfs_fs_info *fs_info;
2220         struct btrfs_root *root;
2221         struct btrfs_key key;
2222         struct extent_buffer *leaf;
2223         struct old_sa_defrag_extent *old = backref->old;
2224         struct new_sa_defrag_extent *new = old->new;
2225         struct inode *src_inode = new->inode;
2226         struct inode *inode;
2227         struct extent_state *cached = NULL;
2228         int ret = 0;
2229         u64 start;
2230         u64 len;
2231         u64 lock_start;
2232         u64 lock_end;
2233         bool merge = false;
2234         int index;
2235
2236         if (prev && prev->root_id == backref->root_id &&
2237             prev->inum == backref->inum &&
2238             prev->file_pos + prev->num_bytes == backref->file_pos)
2239                 merge = true;
2240
2241         /* step 1: get root */
2242         key.objectid = backref->root_id;
2243         key.type = BTRFS_ROOT_ITEM_KEY;
2244         key.offset = (u64)-1;
2245
2246         fs_info = BTRFS_I(src_inode)->root->fs_info;
2247         index = srcu_read_lock(&fs_info->subvol_srcu);
2248
2249         root = btrfs_read_fs_root_no_name(fs_info, &key);
2250         if (IS_ERR(root)) {
2251                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2252                 if (PTR_ERR(root) == -ENOENT)
2253                         return 0;
2254                 return PTR_ERR(root);
2255         }
2256         if (btrfs_root_refs(&root->root_item) == 0) {
2257                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2258                 /* parse ENOENT to 0 */
2259                 return 0;
2260         }
2261
2262         /* step 2: get inode */
2263         key.objectid = backref->inum;
2264         key.type = BTRFS_INODE_ITEM_KEY;
2265         key.offset = 0;
2266
2267         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2268         if (IS_ERR(inode)) {
2269                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2270                 return 0;
2271         }
2272
2273         srcu_read_unlock(&fs_info->subvol_srcu, index);
2274
2275         /* step 3: relink backref */
2276         lock_start = backref->file_pos;
2277         lock_end = backref->file_pos + backref->num_bytes - 1;
2278         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2279                          0, &cached);
2280
2281         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2282         if (ordered) {
2283                 btrfs_put_ordered_extent(ordered);
2284                 goto out_unlock;
2285         }
2286
2287         trans = btrfs_join_transaction(root);
2288         if (IS_ERR(trans)) {
2289                 ret = PTR_ERR(trans);
2290                 goto out_unlock;
2291         }
2292
2293         key.objectid = backref->inum;
2294         key.type = BTRFS_EXTENT_DATA_KEY;
2295         key.offset = backref->file_pos;
2296
2297         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2298         if (ret < 0) {
2299                 goto out_free_path;
2300         } else if (ret > 0) {
2301                 ret = 0;
2302                 goto out_free_path;
2303         }
2304
2305         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2306                                 struct btrfs_file_extent_item);
2307
2308         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2309             backref->generation)
2310                 goto out_free_path;
2311
2312         btrfs_release_path(path);
2313
2314         start = backref->file_pos;
2315         if (backref->extent_offset < old->extent_offset + old->offset)
2316                 start += old->extent_offset + old->offset -
2317                          backref->extent_offset;
2318
2319         len = min(backref->extent_offset + backref->num_bytes,
2320                   old->extent_offset + old->offset + old->len);
2321         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2322
2323         ret = btrfs_drop_extents(trans, root, inode, start,
2324                                  start + len, 1);
2325         if (ret)
2326                 goto out_free_path;
2327 again:
2328         key.objectid = btrfs_ino(inode);
2329         key.type = BTRFS_EXTENT_DATA_KEY;
2330         key.offset = start;
2331
2332         path->leave_spinning = 1;
2333         if (merge) {
2334                 struct btrfs_file_extent_item *fi;
2335                 u64 extent_len;
2336                 struct btrfs_key found_key;
2337
2338                 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
2339                 if (ret < 0)
2340                         goto out_free_path;
2341
2342                 path->slots[0]--;
2343                 leaf = path->nodes[0];
2344                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2345
2346                 fi = btrfs_item_ptr(leaf, path->slots[0],
2347                                     struct btrfs_file_extent_item);
2348                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2349
2350                 if (relink_is_mergable(leaf, fi, new->bytenr) &&
2351                     extent_len + found_key.offset == start) {
2352                         btrfs_set_file_extent_num_bytes(leaf, fi,
2353                                                         extent_len + len);
2354                         btrfs_mark_buffer_dirty(leaf);
2355                         inode_add_bytes(inode, len);
2356
2357                         ret = 1;
2358                         goto out_free_path;
2359                 } else {
2360                         merge = false;
2361                         btrfs_release_path(path);
2362                         goto again;
2363                 }
2364         }
2365
2366         ret = btrfs_insert_empty_item(trans, root, path, &key,
2367                                         sizeof(*extent));
2368         if (ret) {
2369                 btrfs_abort_transaction(trans, root, ret);
2370                 goto out_free_path;
2371         }
2372
2373         leaf = path->nodes[0];
2374         item = btrfs_item_ptr(leaf, path->slots[0],
2375                                 struct btrfs_file_extent_item);
2376         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2377         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2378         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2379         btrfs_set_file_extent_num_bytes(leaf, item, len);
2380         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2381         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2382         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2383         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2384         btrfs_set_file_extent_encryption(leaf, item, 0);
2385         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2386
2387         btrfs_mark_buffer_dirty(leaf);
2388         inode_add_bytes(inode, len);
2389         btrfs_release_path(path);
2390
2391         ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2392                         new->disk_len, 0,
2393                         backref->root_id, backref->inum,
2394                         new->file_pos, 0);      /* start - extent_offset */
2395         if (ret) {
2396                 btrfs_abort_transaction(trans, root, ret);
2397                 goto out_free_path;
2398         }
2399
2400         ret = 1;
2401 out_free_path:
2402         btrfs_release_path(path);
2403         path->leave_spinning = 0;
2404         btrfs_end_transaction(trans, root);
2405 out_unlock:
2406         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2407                              &cached, GFP_NOFS);
2408         iput(inode);
2409         return ret;
2410 }
2411
2412 static void relink_file_extents(struct new_sa_defrag_extent *new)
2413 {
2414         struct btrfs_path *path;
2415         struct old_sa_defrag_extent *old, *tmp;
2416         struct sa_defrag_extent_backref *backref;
2417         struct sa_defrag_extent_backref *prev = NULL;
2418         struct inode *inode;
2419         struct btrfs_root *root;
2420         struct rb_node *node;
2421         int ret;
2422
2423         inode = new->inode;
2424         root = BTRFS_I(inode)->root;
2425
2426         path = btrfs_alloc_path();
2427         if (!path)
2428                 return;
2429
2430         if (!record_extent_backrefs(path, new)) {
2431                 btrfs_free_path(path);
2432                 goto out;
2433         }
2434         btrfs_release_path(path);
2435
2436         while (1) {
2437                 node = rb_first(&new->root);
2438                 if (!node)
2439                         break;
2440                 rb_erase(node, &new->root);
2441
2442                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2443
2444                 ret = relink_extent_backref(path, prev, backref);
2445                 WARN_ON(ret < 0);
2446
2447                 kfree(prev);
2448
2449                 if (ret == 1)
2450                         prev = backref;
2451                 else
2452                         prev = NULL;
2453                 cond_resched();
2454         }
2455         kfree(prev);
2456
2457         btrfs_free_path(path);
2458
2459         list_for_each_entry_safe(old, tmp, &new->head, list) {
2460                 list_del(&old->list);
2461                 kfree(old);
2462         }
2463 out:
2464         atomic_dec(&root->fs_info->defrag_running);
2465         wake_up(&root->fs_info->transaction_wait);
2466
2467         kfree(new);
2468 }
2469
2470 static struct new_sa_defrag_extent *
2471 record_old_file_extents(struct inode *inode,
2472                         struct btrfs_ordered_extent *ordered)
2473 {
2474         struct btrfs_root *root = BTRFS_I(inode)->root;
2475         struct btrfs_path *path;
2476         struct btrfs_key key;
2477         struct old_sa_defrag_extent *old, *tmp;
2478         struct new_sa_defrag_extent *new;
2479         int ret;
2480
2481         new = kmalloc(sizeof(*new), GFP_NOFS);
2482         if (!new)
2483                 return NULL;
2484
2485         new->inode = inode;
2486         new->file_pos = ordered->file_offset;
2487         new->len = ordered->len;
2488         new->bytenr = ordered->start;
2489         new->disk_len = ordered->disk_len;
2490         new->compress_type = ordered->compress_type;
2491         new->root = RB_ROOT;
2492         INIT_LIST_HEAD(&new->head);
2493
2494         path = btrfs_alloc_path();
2495         if (!path)
2496                 goto out_kfree;
2497
2498         key.objectid = btrfs_ino(inode);
2499         key.type = BTRFS_EXTENT_DATA_KEY;
2500         key.offset = new->file_pos;
2501
2502         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2503         if (ret < 0)
2504                 goto out_free_path;
2505         if (ret > 0 && path->slots[0] > 0)
2506                 path->slots[0]--;
2507
2508         /* find out all the old extents for the file range */
2509         while (1) {
2510                 struct btrfs_file_extent_item *extent;
2511                 struct extent_buffer *l;
2512                 int slot;
2513                 u64 num_bytes;
2514                 u64 offset;
2515                 u64 end;
2516                 u64 disk_bytenr;
2517                 u64 extent_offset;
2518
2519                 l = path->nodes[0];
2520                 slot = path->slots[0];
2521
2522                 if (slot >= btrfs_header_nritems(l)) {
2523                         ret = btrfs_next_leaf(root, path);
2524                         if (ret < 0)
2525                                 goto out_free_list;
2526                         else if (ret > 0)
2527                                 break;
2528                         continue;
2529                 }
2530
2531                 btrfs_item_key_to_cpu(l, &key, slot);
2532
2533                 if (key.objectid != btrfs_ino(inode))
2534                         break;
2535                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2536                         break;
2537                 if (key.offset >= new->file_pos + new->len)
2538                         break;
2539
2540                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2541
2542                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2543                 if (key.offset + num_bytes < new->file_pos)
2544                         goto next;
2545
2546                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2547                 if (!disk_bytenr)
2548                         goto next;
2549
2550                 extent_offset = btrfs_file_extent_offset(l, extent);
2551
2552                 old = kmalloc(sizeof(*old), GFP_NOFS);
2553                 if (!old)
2554                         goto out_free_list;
2555
2556                 offset = max(new->file_pos, key.offset);
2557                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2558
2559                 old->bytenr = disk_bytenr;
2560                 old->extent_offset = extent_offset;
2561                 old->offset = offset - key.offset;
2562                 old->len = end - offset;
2563                 old->new = new;
2564                 old->count = 0;
2565                 list_add_tail(&old->list, &new->head);
2566 next:
2567                 path->slots[0]++;
2568                 cond_resched();
2569         }
2570
2571         btrfs_free_path(path);
2572         atomic_inc(&root->fs_info->defrag_running);
2573
2574         return new;
2575
2576 out_free_list:
2577         list_for_each_entry_safe(old, tmp, &new->head, list) {
2578                 list_del(&old->list);
2579                 kfree(old);
2580         }
2581 out_free_path:
2582         btrfs_free_path(path);
2583 out_kfree:
2584         kfree(new);
2585         return NULL;
2586 }
2587
2588 /*
2589  * helper function for btrfs_finish_ordered_io, this
2590  * just reads in some of the csum leaves to prime them into ram
2591  * before we start the transaction.  It limits the amount of btree
2592  * reads required while inside the transaction.
2593  */
2594 /* as ordered data IO finishes, this gets called so we can finish
2595  * an ordered extent if the range of bytes in the file it covers are
2596  * fully written.
2597  */
2598 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2599 {
2600         struct inode *inode = ordered_extent->inode;
2601         struct btrfs_root *root = BTRFS_I(inode)->root;
2602         struct btrfs_trans_handle *trans = NULL;
2603         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2604         struct extent_state *cached_state = NULL;
2605         struct new_sa_defrag_extent *new = NULL;
2606         int compress_type = 0;
2607         int ret;
2608         bool nolock;
2609
2610         nolock = btrfs_is_free_space_inode(inode);
2611
2612         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2613                 ret = -EIO;
2614                 goto out;
2615         }
2616
2617         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2618                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2619                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2620                 if (nolock)
2621                         trans = btrfs_join_transaction_nolock(root);
2622                 else
2623                         trans = btrfs_join_transaction(root);
2624                 if (IS_ERR(trans)) {
2625                         ret = PTR_ERR(trans);
2626                         trans = NULL;
2627                         goto out;
2628                 }
2629                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2630                 ret = btrfs_update_inode_fallback(trans, root, inode);
2631                 if (ret) /* -ENOMEM or corruption */
2632                         btrfs_abort_transaction(trans, root, ret);
2633                 goto out;
2634         }
2635
2636         lock_extent_bits(io_tree, ordered_extent->file_offset,
2637                          ordered_extent->file_offset + ordered_extent->len - 1,
2638                          0, &cached_state);
2639
2640         ret = test_range_bit(io_tree, ordered_extent->file_offset,
2641                         ordered_extent->file_offset + ordered_extent->len - 1,
2642                         EXTENT_DEFRAG, 1, cached_state);
2643         if (ret) {
2644                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2645                 if (last_snapshot >= BTRFS_I(inode)->generation)
2646                         /* the inode is shared */
2647                         new = record_old_file_extents(inode, ordered_extent);
2648
2649                 clear_extent_bit(io_tree, ordered_extent->file_offset,
2650                         ordered_extent->file_offset + ordered_extent->len - 1,
2651                         EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2652         }
2653
2654         if (nolock)
2655                 trans = btrfs_join_transaction_nolock(root);
2656         else
2657                 trans = btrfs_join_transaction(root);
2658         if (IS_ERR(trans)) {
2659                 ret = PTR_ERR(trans);
2660                 trans = NULL;
2661                 goto out_unlock;
2662         }
2663         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2664
2665         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2666                 compress_type = ordered_extent->compress_type;
2667         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2668                 BUG_ON(compress_type);
2669                 ret = btrfs_mark_extent_written(trans, inode,
2670                                                 ordered_extent->file_offset,
2671                                                 ordered_extent->file_offset +
2672                                                 ordered_extent->len);
2673         } else {
2674                 BUG_ON(root == root->fs_info->tree_root);
2675                 ret = insert_reserved_file_extent(trans, inode,
2676                                                 ordered_extent->file_offset,
2677                                                 ordered_extent->start,
2678                                                 ordered_extent->disk_len,
2679                                                 ordered_extent->len,
2680                                                 ordered_extent->len,
2681                                                 compress_type, 0, 0,
2682                                                 BTRFS_FILE_EXTENT_REG);
2683         }
2684         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2685                            ordered_extent->file_offset, ordered_extent->len,
2686                            trans->transid);
2687         if (ret < 0) {
2688                 btrfs_abort_transaction(trans, root, ret);
2689                 goto out_unlock;
2690         }
2691
2692         add_pending_csums(trans, inode, ordered_extent->file_offset,
2693                           &ordered_extent->list);
2694
2695         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2696         ret = btrfs_update_inode_fallback(trans, root, inode);
2697         if (ret) { /* -ENOMEM or corruption */
2698                 btrfs_abort_transaction(trans, root, ret);
2699                 goto out_unlock;
2700         }
2701         ret = 0;
2702 out_unlock:
2703         unlock_extent_cached(io_tree, ordered_extent->file_offset,
2704                              ordered_extent->file_offset +
2705                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
2706 out:
2707         if (root != root->fs_info->tree_root)
2708                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2709         if (trans)
2710                 btrfs_end_transaction(trans, root);
2711
2712         if (ret) {
2713                 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
2714                                       ordered_extent->file_offset +
2715                                       ordered_extent->len - 1, NULL, GFP_NOFS);
2716
2717                 /*
2718                  * If the ordered extent had an IOERR or something else went
2719                  * wrong we need to return the space for this ordered extent
2720                  * back to the allocator.
2721                  */
2722                 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2723                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2724                         btrfs_free_reserved_extent(root, ordered_extent->start,
2725                                                    ordered_extent->disk_len);
2726         }
2727
2728
2729         /*
2730          * This needs to be done to make sure anybody waiting knows we are done
2731          * updating everything for this ordered extent.
2732          */
2733         btrfs_remove_ordered_extent(inode, ordered_extent);
2734
2735         /* for snapshot-aware defrag */
2736         if (new)
2737                 relink_file_extents(new);
2738
2739         /* once for us */
2740         btrfs_put_ordered_extent(ordered_extent);
2741         /* once for the tree */
2742         btrfs_put_ordered_extent(ordered_extent);
2743
2744         return ret;
2745 }
2746
2747 static void finish_ordered_fn(struct btrfs_work *work)
2748 {
2749         struct btrfs_ordered_extent *ordered_extent;
2750         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2751         btrfs_finish_ordered_io(ordered_extent);
2752 }
2753
2754 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2755                                 struct extent_state *state, int uptodate)
2756 {
2757         struct inode *inode = page->mapping->host;
2758         struct btrfs_root *root = BTRFS_I(inode)->root;
2759         struct btrfs_ordered_extent *ordered_extent = NULL;
2760         struct btrfs_workers *workers;
2761
2762         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2763
2764         ClearPagePrivate2(page);
2765         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2766                                             end - start + 1, uptodate))
2767                 return 0;
2768
2769         ordered_extent->work.func = finish_ordered_fn;
2770         ordered_extent->work.flags = 0;
2771
2772         if (btrfs_is_free_space_inode(inode))
2773                 workers = &root->fs_info->endio_freespace_worker;
2774         else
2775                 workers = &root->fs_info->endio_write_workers;
2776         btrfs_queue_worker(workers, &ordered_extent->work);
2777
2778         return 0;
2779 }
2780
2781 /*
2782  * when reads are done, we need to check csums to verify the data is correct
2783  * if there's a match, we allow the bio to finish.  If not, the code in
2784  * extent_io.c will try to find good copies for us.
2785  */
2786 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2787                                struct extent_state *state, int mirror)
2788 {
2789         size_t offset = start - page_offset(page);
2790         struct inode *inode = page->mapping->host;
2791         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2792         char *kaddr;
2793         u64 private = ~(u32)0;
2794         int ret;
2795         struct btrfs_root *root = BTRFS_I(inode)->root;
2796         u32 csum = ~(u32)0;
2797
2798         if (PageChecked(page)) {
2799                 ClearPageChecked(page);
2800                 goto good;
2801         }
2802
2803         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2804                 goto good;
2805
2806         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2807             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2808                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2809                                   GFP_NOFS);
2810                 return 0;
2811         }
2812
2813         if (state && state->start == start) {
2814                 private = state->private;
2815                 ret = 0;
2816         } else {
2817                 ret = get_state_private(io_tree, start, &private);
2818         }
2819         kaddr = kmap_atomic(page);
2820         if (ret)
2821                 goto zeroit;
2822
2823         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2824         btrfs_csum_final(csum, (char *)&csum);
2825         if (csum != private)
2826                 goto zeroit;
2827
2828         kunmap_atomic(kaddr);
2829 good:
2830         return 0;
2831
2832 zeroit:
2833         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2834                        "private %llu\n",
2835                        (unsigned long long)btrfs_ino(page->mapping->host),
2836                        (unsigned long long)start, csum,
2837                        (unsigned long long)private);
2838         memset(kaddr + offset, 1, end - start + 1);
2839         flush_dcache_page(page);
2840         kunmap_atomic(kaddr);
2841         if (private == 0)
2842                 return 0;
2843         return -EIO;
2844 }
2845
2846 struct delayed_iput {
2847         struct list_head list;
2848         struct inode *inode;
2849 };
2850
2851 /* JDM: If this is fs-wide, why can't we add a pointer to
2852  * btrfs_inode instead and avoid the allocation? */
2853 void btrfs_add_delayed_iput(struct inode *inode)
2854 {
2855         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2856         struct delayed_iput *delayed;
2857
2858         if (atomic_add_unless(&inode->i_count, -1, 1))
2859                 return;
2860
2861         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2862         delayed->inode = inode;
2863
2864         spin_lock(&fs_info->delayed_iput_lock);
2865         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2866         spin_unlock(&fs_info->delayed_iput_lock);
2867 }
2868
2869 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2870 {
2871         LIST_HEAD(list);
2872         struct btrfs_fs_info *fs_info = root->fs_info;
2873         struct delayed_iput *delayed;
2874         int empty;
2875
2876         spin_lock(&fs_info->delayed_iput_lock);
2877         empty = list_empty(&fs_info->delayed_iputs);
2878         spin_unlock(&fs_info->delayed_iput_lock);
2879         if (empty)
2880                 return;
2881
2882         spin_lock(&fs_info->delayed_iput_lock);
2883         list_splice_init(&fs_info->delayed_iputs, &list);
2884         spin_unlock(&fs_info->delayed_iput_lock);
2885
2886         while (!list_empty(&list)) {
2887                 delayed = list_entry(list.next, struct delayed_iput, list);
2888                 list_del(&delayed->list);
2889                 iput(delayed->inode);
2890                 kfree(delayed);
2891         }
2892 }
2893
2894 /*
2895  * This is called in transaction commit time. If there are no orphan
2896  * files in the subvolume, it removes orphan item and frees block_rsv
2897  * structure.
2898  */
2899 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2900                               struct btrfs_root *root)
2901 {
2902         struct btrfs_block_rsv *block_rsv;
2903         int ret;
2904
2905         if (atomic_read(&root->orphan_inodes) ||
2906             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2907                 return;
2908
2909         spin_lock(&root->orphan_lock);
2910         if (atomic_read(&root->orphan_inodes)) {
2911                 spin_unlock(&root->orphan_lock);
2912                 return;
2913         }
2914
2915         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2916                 spin_unlock(&root->orphan_lock);
2917                 return;
2918         }
2919
2920         block_rsv = root->orphan_block_rsv;
2921         root->orphan_block_rsv = NULL;
2922         spin_unlock(&root->orphan_lock);
2923
2924         if (root->orphan_item_inserted &&
2925             btrfs_root_refs(&root->root_item) > 0) {
2926                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2927                                             root->root_key.objectid);
2928                 BUG_ON(ret);
2929                 root->orphan_item_inserted = 0;
2930         }
2931
2932         if (block_rsv) {
2933                 WARN_ON(block_rsv->size > 0);
2934                 btrfs_free_block_rsv(root, block_rsv);
2935         }
2936 }
2937
2938 /*
2939  * This creates an orphan entry for the given inode in case something goes
2940  * wrong in the middle of an unlink/truncate.
2941  *
2942  * NOTE: caller of this function should reserve 5 units of metadata for
2943  *       this function.
2944  */
2945 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2946 {
2947         struct btrfs_root *root = BTRFS_I(inode)->root;
2948         struct btrfs_block_rsv *block_rsv = NULL;
2949         int reserve = 0;
2950         int insert = 0;
2951         int ret;
2952
2953         if (!root->orphan_block_rsv) {
2954                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2955                 if (!block_rsv)
2956                         return -ENOMEM;
2957         }
2958
2959         spin_lock(&root->orphan_lock);
2960         if (!root->orphan_block_rsv) {
2961                 root->orphan_block_rsv = block_rsv;
2962         } else if (block_rsv) {
2963                 btrfs_free_block_rsv(root, block_rsv);
2964                 block_rsv = NULL;
2965         }
2966
2967         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2968                               &BTRFS_I(inode)->runtime_flags)) {
2969 #if 0
2970                 /*
2971                  * For proper ENOSPC handling, we should do orphan
2972                  * cleanup when mounting. But this introduces backward
2973                  * compatibility issue.
2974                  */
2975                 if (!xchg(&root->orphan_item_inserted, 1))
2976                         insert = 2;
2977                 else
2978                         insert = 1;
2979 #endif
2980                 insert = 1;
2981                 atomic_inc(&root->orphan_inodes);
2982         }
2983
2984         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2985                               &BTRFS_I(inode)->runtime_flags))
2986                 reserve = 1;
2987         spin_unlock(&root->orphan_lock);
2988
2989         /* grab metadata reservation from transaction handle */
2990         if (reserve) {
2991                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2992                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2993         }
2994
2995         /* insert an orphan item to track this unlinked/truncated file */
2996         if (insert >= 1) {
2997                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2998                 if (ret && ret != -EEXIST) {
2999                         clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3000                                   &BTRFS_I(inode)->runtime_flags);
3001                         btrfs_abort_transaction(trans, root, ret);
3002                         return ret;
3003                 }
3004                 ret = 0;
3005         }
3006
3007         /* insert an orphan item to track subvolume contains orphan files */
3008         if (insert >= 2) {
3009                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3010                                                root->root_key.objectid);
3011                 if (ret && ret != -EEXIST) {
3012                         btrfs_abort_transaction(trans, root, ret);
3013                         return ret;
3014                 }
3015         }
3016         return 0;
3017 }
3018
3019 /*
3020  * We have done the truncate/delete so we can go ahead and remove the orphan
3021  * item for this particular inode.
3022  */
3023 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
3024 {
3025         struct btrfs_root *root = BTRFS_I(inode)->root;
3026         int delete_item = 0;
3027         int release_rsv = 0;
3028         int ret = 0;
3029
3030         spin_lock(&root->orphan_lock);
3031         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3032                                &BTRFS_I(inode)->runtime_flags))
3033                 delete_item = 1;
3034
3035         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3036                                &BTRFS_I(inode)->runtime_flags))
3037                 release_rsv = 1;
3038         spin_unlock(&root->orphan_lock);
3039
3040         if (trans && delete_item) {
3041                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
3042                 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3043         }
3044
3045         if (release_rsv) {
3046                 btrfs_orphan_release_metadata(inode);
3047                 atomic_dec(&root->orphan_inodes);
3048         }
3049
3050         return 0;
3051 }
3052
3053 /*
3054  * this cleans up any orphans that may be left on the list from the last use
3055  * of this root.
3056  */
3057 int btrfs_orphan_cleanup(struct btrfs_root *root)
3058 {
3059         struct btrfs_path *path;
3060         struct extent_buffer *leaf;
3061         struct btrfs_key key, found_key;
3062         struct btrfs_trans_handle *trans;
3063         struct inode *inode;
3064         u64 last_objectid = 0;
3065         int ret = 0, nr_unlink = 0, nr_truncate = 0;
3066
3067         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3068                 return 0;
3069
3070         path = btrfs_alloc_path();
3071         if (!path) {
3072                 ret = -ENOMEM;
3073                 goto out;
3074         }
3075         path->reada = -1;
3076
3077         key.objectid = BTRFS_ORPHAN_OBJECTID;
3078         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3079         key.offset = (u64)-1;
3080
3081         while (1) {
3082                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3083                 if (ret < 0)
3084                         goto out;
3085
3086                 /*
3087                  * if ret == 0 means we found what we were searching for, which
3088                  * is weird, but possible, so only screw with path if we didn't
3089                  * find the key and see if we have stuff that matches
3090                  */
3091                 if (ret > 0) {
3092                         ret = 0;
3093                         if (path->slots[0] == 0)
3094                                 break;
3095                         path->slots[0]--;
3096                 }
3097
3098                 /* pull out the item */
3099                 leaf = path->nodes[0];
3100                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3101
3102                 /* make sure the item matches what we want */
3103                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3104                         break;
3105                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3106                         break;
3107
3108                 /* release the path since we're done with it */
3109                 btrfs_release_path(path);
3110
3111                 /*
3112                  * this is where we are basically btrfs_lookup, without the
3113                  * crossing root thing.  we store the inode number in the
3114                  * offset of the orphan item.
3115                  */
3116
3117                 if (found_key.offset == last_objectid) {
3118                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
3119                                "stopping orphan cleanup\n");
3120                         ret = -EINVAL;
3121                         goto out;
3122                 }
3123
3124                 last_objectid = found_key.offset;
3125
3126                 found_key.objectid = found_key.offset;
3127                 found_key.type = BTRFS_INODE_ITEM_KEY;
3128                 found_key.offset = 0;
3129                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3130                 ret = PTR_RET(inode);
3131                 if (ret && ret != -ESTALE)
3132                         goto out;
3133
3134                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3135                         struct btrfs_root *dead_root;
3136                         struct btrfs_fs_info *fs_info = root->fs_info;
3137                         int is_dead_root = 0;
3138
3139                         /*
3140                          * this is an orphan in the tree root. Currently these
3141                          * could come from 2 sources:
3142                          *  a) a snapshot deletion in progress
3143                          *  b) a free space cache inode
3144                          * We need to distinguish those two, as the snapshot
3145                          * orphan must not get deleted.
3146                          * find_dead_roots already ran before us, so if this
3147                          * is a snapshot deletion, we should find the root
3148                          * in the dead_roots list
3149                          */
3150                         spin_lock(&fs_info->trans_lock);
3151                         list_for_each_entry(dead_root, &fs_info->dead_roots,
3152                                             root_list) {
3153                                 if (dead_root->root_key.objectid ==
3154                                     found_key.objectid) {
3155                                         is_dead_root = 1;
3156                                         break;
3157                                 }
3158                         }
3159                         spin_unlock(&fs_info->trans_lock);
3160                         if (is_dead_root) {
3161                                 /* prevent this orphan from being found again */
3162                                 key.offset = found_key.objectid - 1;
3163                                 continue;
3164                         }
3165                 }
3166                 /*
3167                  * Inode is already gone but the orphan item is still there,
3168                  * kill the orphan item.
3169                  */
3170                 if (ret == -ESTALE) {
3171                         trans = btrfs_start_transaction(root, 1);
3172                         if (IS_ERR(trans)) {
3173                                 ret = PTR_ERR(trans);
3174                                 goto out;
3175                         }
3176                         printk(KERN_ERR "auto deleting %Lu\n",
3177                                found_key.objectid);
3178                         ret = btrfs_del_orphan_item(trans, root,
3179                                                     found_key.objectid);
3180                         BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3181                         btrfs_end_transaction(trans, root);
3182                         continue;
3183                 }
3184
3185                 /*
3186                  * add this inode to the orphan list so btrfs_orphan_del does
3187                  * the proper thing when we hit it
3188                  */
3189                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3190                         &BTRFS_I(inode)->runtime_flags);
3191                 atomic_inc(&root->orphan_inodes);
3192
3193                 /* if we have links, this was a truncate, lets do that */
3194                 if (inode->i_nlink) {
3195                         if (!S_ISREG(inode->i_mode)) {
3196                                 WARN_ON(1);
3197                                 iput(inode);
3198                                 continue;
3199                         }
3200                         nr_truncate++;
3201
3202                         /* 1 for the orphan item deletion. */
3203                         trans = btrfs_start_transaction(root, 1);
3204                         if (IS_ERR(trans)) {
3205                                 ret = PTR_ERR(trans);
3206                                 goto out;
3207                         }
3208                         ret = btrfs_orphan_add(trans, inode);
3209                         btrfs_end_transaction(trans, root);
3210                         if (ret)
3211                                 goto out;
3212
3213                         ret = btrfs_truncate(inode);
3214                         if (ret)
3215                                 btrfs_orphan_del(NULL, inode);
3216                 } else {
3217                         nr_unlink++;
3218                 }
3219
3220                 /* this will do delete_inode and everything for us */
3221                 iput(inode);
3222                 if (ret)
3223                         goto out;
3224         }
3225         /* release the path since we're done with it */
3226         btrfs_release_path(path);
3227
3228         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3229
3230         if (root->orphan_block_rsv)
3231                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3232                                         (u64)-1);
3233
3234         if (root->orphan_block_rsv || root->orphan_item_inserted) {
3235                 trans = btrfs_join_transaction(root);
3236                 if (!IS_ERR(trans))
3237                         btrfs_end_transaction(trans, root);
3238         }
3239
3240         if (nr_unlink)
3241                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
3242         if (nr_truncate)
3243                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
3244
3245 out:
3246         if (ret)
3247                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
3248         btrfs_free_path(path);
3249         return ret;
3250 }
3251
3252 /*
3253  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3254  * don't find any xattrs, we know there can't be any acls.
3255  *
3256  * slot is the slot the inode is in, objectid is the objectid of the inode
3257  */
3258 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3259                                           int slot, u64 objectid)
3260 {
3261         u32 nritems = btrfs_header_nritems(leaf);
3262         struct btrfs_key found_key;
3263         int scanned = 0;
3264
3265         slot++;
3266         while (slot < nritems) {
3267                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3268
3269                 /* we found a different objectid, there must not be acls */
3270                 if (found_key.objectid != objectid)
3271                         return 0;
3272
3273                 /* we found an xattr, assume we've got an acl */
3274                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
3275                         return 1;
3276
3277                 /*
3278                  * we found a key greater than an xattr key, there can't
3279                  * be any acls later on
3280                  */
3281                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3282                         return 0;
3283
3284                 slot++;
3285                 scanned++;
3286
3287                 /*
3288                  * it goes inode, inode backrefs, xattrs, extents,
3289                  * so if there are a ton of hard links to an inode there can
3290                  * be a lot of backrefs.  Don't waste time searching too hard,
3291                  * this is just an optimization
3292                  */
3293                 if (scanned >= 8)
3294                         break;
3295         }
3296         /* we hit the end of the leaf before we found an xattr or
3297          * something larger than an xattr.  We have to assume the inode
3298          * has acls
3299          */
3300         return 1;
3301 }
3302
3303 /*
3304  * read an inode from the btree into the in-memory inode
3305  */
3306 static void btrfs_read_locked_inode(struct inode *inode)
3307 {
3308         struct btrfs_path *path;
3309         struct extent_buffer *leaf;
3310         struct btrfs_inode_item *inode_item;
3311         struct btrfs_timespec *tspec;
3312         struct btrfs_root *root = BTRFS_I(inode)->root;
3313         struct btrfs_key location;
3314         int maybe_acls;
3315         u32 rdev;
3316         int ret;
3317         bool filled = false;
3318
3319         ret = btrfs_fill_inode(inode, &rdev);
3320         if (!ret)
3321                 filled = true;
3322
3323         path = btrfs_alloc_path();
3324         if (!path)
3325                 goto make_bad;
3326
3327         path->leave_spinning = 1;
3328         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3329
3330         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3331         if (ret)
3332                 goto make_bad;
3333
3334         leaf = path->nodes[0];
3335
3336         if (filled)
3337                 goto cache_acl;
3338
3339         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3340                                     struct btrfs_inode_item);
3341         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3342         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3343         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3344         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3345         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3346
3347         tspec = btrfs_inode_atime(inode_item);
3348         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3349         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3350
3351         tspec = btrfs_inode_mtime(inode_item);
3352         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3353         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3354
3355         tspec = btrfs_inode_ctime(inode_item);
3356         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3357         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3358
3359         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3360         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3361         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3362
3363         /*
3364          * If we were modified in the current generation and evicted from memory
3365          * and then re-read we need to do a full sync since we don't have any
3366          * idea about which extents were modified before we were evicted from
3367          * cache.
3368          */
3369         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3370                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3371                         &BTRFS_I(inode)->runtime_flags);
3372
3373         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3374         inode->i_generation = BTRFS_I(inode)->generation;
3375         inode->i_rdev = 0;
3376         rdev = btrfs_inode_rdev(leaf, inode_item);
3377
3378         BTRFS_I(inode)->index_cnt = (u64)-1;
3379         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3380 cache_acl:
3381         /*
3382          * try to precache a NULL acl entry for files that don't have
3383          * any xattrs or acls
3384          */
3385         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3386                                            btrfs_ino(inode));
3387         if (!maybe_acls)
3388                 cache_no_acl(inode);
3389
3390         btrfs_free_path(path);
3391
3392         switch (inode->i_mode & S_IFMT) {
3393         case S_IFREG:
3394                 inode->i_mapping->a_ops = &btrfs_aops;
3395                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3396                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3397                 inode->i_fop = &btrfs_file_operations;
3398                 inode->i_op = &btrfs_file_inode_operations;
3399                 break;
3400         case S_IFDIR:
3401                 inode->i_fop = &btrfs_dir_file_operations;
3402                 if (root == root->fs_info->tree_root)
3403                         inode->i_op = &btrfs_dir_ro_inode_operations;
3404                 else
3405                         inode->i_op = &btrfs_dir_inode_operations;
3406                 break;
3407         case S_IFLNK:
3408                 inode->i_op = &btrfs_symlink_inode_operations;
3409                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3410                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3411                 break;
3412         default:
3413                 inode->i_op = &btrfs_special_inode_operations;
3414                 init_special_inode(inode, inode->i_mode, rdev);
3415                 break;
3416         }
3417
3418         btrfs_update_iflags(inode);
3419         return;
3420
3421 make_bad:
3422         btrfs_free_path(path);
3423         make_bad_inode(inode);
3424 }
3425
3426 /*
3427  * given a leaf and an inode, copy the inode fields into the leaf
3428  */
3429 static void fill_inode_item(struct btrfs_trans_handle *trans,
3430                             struct extent_buffer *leaf,
3431                             struct btrfs_inode_item *item,
3432                             struct inode *inode)
3433 {
3434         struct btrfs_map_token token;
3435
3436         btrfs_init_map_token(&token);
3437
3438         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3439         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3440         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3441                                    &token);
3442         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3443         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3444
3445         btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3446                                      inode->i_atime.tv_sec, &token);
3447         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3448                                       inode->i_atime.tv_nsec, &token);
3449
3450         btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3451                                      inode->i_mtime.tv_sec, &token);
3452         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3453                                       inode->i_mtime.tv_nsec, &token);
3454
3455         btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3456                                      inode->i_ctime.tv_sec, &token);
3457         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3458                                       inode->i_ctime.tv_nsec, &token);
3459
3460         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3461                                      &token);
3462         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3463                                          &token);
3464         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3465         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3466         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3467         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3468         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3469 }
3470
3471 /*
3472  * copy everything in the in-memory inode into the btree.
3473  */
3474 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3475                                 struct btrfs_root *root, struct inode *inode)
3476 {
3477         struct btrfs_inode_item *inode_item;
3478         struct btrfs_path *path;
3479         struct extent_buffer *leaf;
3480         int ret;
3481
3482         path = btrfs_alloc_path();
3483         if (!path)
3484                 return -ENOMEM;
3485
3486         path->leave_spinning = 1;
3487         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3488                                  1);
3489         if (ret) {
3490                 if (ret > 0)
3491                         ret = -ENOENT;
3492                 goto failed;
3493         }
3494
3495         btrfs_unlock_up_safe(path, 1);
3496         leaf = path->nodes[0];
3497         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3498                                     struct btrfs_inode_item);
3499
3500         fill_inode_item(trans, leaf, inode_item, inode);
3501         btrfs_mark_buffer_dirty(leaf);
3502         btrfs_set_inode_last_trans(trans, inode);
3503         ret = 0;
3504 failed:
3505         btrfs_free_path(path);
3506         return ret;
3507 }
3508
3509 /*
3510  * copy everything in the in-memory inode into the btree.
3511  */
3512 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3513                                 struct btrfs_root *root, struct inode *inode)
3514 {
3515         int ret;
3516
3517         /*
3518          * If the inode is a free space inode, we can deadlock during commit
3519          * if we put it into the delayed code.
3520          *
3521          * The data relocation inode should also be directly updated
3522          * without delay
3523          */
3524         if (!btrfs_is_free_space_inode(inode)
3525             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3526                 btrfs_update_root_times(trans, root);
3527
3528                 ret = btrfs_delayed_update_inode(trans, root, inode);
3529                 if (!ret)
3530                         btrfs_set_inode_last_trans(trans, inode);
3531                 return ret;
3532         }
3533
3534         return btrfs_update_inode_item(trans, root, inode);
3535 }
3536
3537 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3538                                          struct btrfs_root *root,
3539                                          struct inode *inode)
3540 {
3541         int ret;
3542
3543         ret = btrfs_update_inode(trans, root, inode);
3544         if (ret == -ENOSPC)
3545                 return btrfs_update_inode_item(trans, root, inode);
3546         return ret;
3547 }
3548
3549 /*
3550  * unlink helper that gets used here in inode.c and in the tree logging
3551  * recovery code.  It remove a link in a directory with a given name, and
3552  * also drops the back refs in the inode to the directory
3553  */
3554 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3555                                 struct btrfs_root *root,
3556                                 struct inode *dir, struct inode *inode,
3557                                 const char *name, int name_len)
3558 {
3559         struct btrfs_path *path;
3560         int ret = 0;
3561         struct extent_buffer *leaf;
3562         struct btrfs_dir_item *di;
3563         struct btrfs_key key;
3564         u64 index;
3565         u64 ino = btrfs_ino(inode);
3566         u64 dir_ino = btrfs_ino(dir);
3567
3568         path = btrfs_alloc_path();
3569         if (!path) {
3570                 ret = -ENOMEM;
3571                 goto out;
3572         }
3573
3574         path->leave_spinning = 1;
3575         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3576                                     name, name_len, -1);
3577         if (IS_ERR(di)) {
3578                 ret = PTR_ERR(di);
3579                 goto err;
3580         }
3581         if (!di) {
3582                 ret = -ENOENT;
3583                 goto err;
3584         }
3585         leaf = path->nodes[0];
3586         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3587         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3588         if (ret)
3589                 goto err;
3590         btrfs_release_path(path);
3591
3592         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3593                                   dir_ino, &index);
3594         if (ret) {
3595                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
3596                        "inode %llu parent %llu\n", name_len, name,
3597                        (unsigned long long)ino, (unsigned long long)dir_ino);
3598                 btrfs_abort_transaction(trans, root, ret);
3599                 goto err;
3600         }
3601
3602         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3603         if (ret) {
3604                 btrfs_abort_transaction(trans, root, ret);
3605                 goto err;
3606         }
3607
3608         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3609                                          inode, dir_ino);
3610         if (ret != 0 && ret != -ENOENT) {
3611                 btrfs_abort_transaction(trans, root, ret);
3612                 goto err;
3613         }
3614
3615         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3616                                            dir, index);
3617         if (ret == -ENOENT)
3618                 ret = 0;
3619 err:
3620         btrfs_free_path(path);
3621         if (ret)
3622                 goto out;
3623
3624         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3625         inode_inc_iversion(inode);
3626         inode_inc_iversion(dir);
3627         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3628         ret = btrfs_update_inode(trans, root, dir);
3629 out:
3630         return ret;
3631 }
3632
3633 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3634                        struct btrfs_root *root,
3635                        struct inode *dir, struct inode *inode,
3636                        const char *name, int name_len)
3637 {
3638         int ret;
3639         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3640         if (!ret) {
3641                 btrfs_drop_nlink(inode);
3642                 ret = btrfs_update_inode(trans, root, inode);
3643         }
3644         return ret;
3645 }
3646
3647
3648 /* helper to check if there is any shared block in the path */
3649 static int check_path_shared(struct btrfs_root *root,
3650                              struct btrfs_path *path)
3651 {
3652         struct extent_buffer *eb;
3653         int level;
3654         u64 refs = 1;
3655
3656         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
3657                 int ret;
3658
3659                 if (!path->nodes[level])
3660                         break;
3661                 eb = path->nodes[level];
3662                 if (!btrfs_block_can_be_shared(root, eb))
3663                         continue;
3664                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
3665                                                &refs, NULL);
3666                 if (refs > 1)
3667                         return 1;
3668         }
3669         return 0;
3670 }
3671
3672 /*
3673  * helper to start transaction for unlink and rmdir.
3674  *
3675  * unlink and rmdir are special in btrfs, they do not always free space.
3676  * so in enospc case, we should make sure they will free space before
3677  * allowing them to use the global metadata reservation.
3678  */
3679 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
3680                                                        struct dentry *dentry)
3681 {
3682         struct btrfs_trans_handle *trans;
3683         struct btrfs_root *root = BTRFS_I(dir)->root;
3684         struct btrfs_path *path;
3685         struct btrfs_dir_item *di;
3686         struct inode *inode = dentry->d_inode;
3687         u64 index;
3688         int check_link = 1;
3689         int err = -ENOSPC;
3690         int ret;
3691         u64 ino = btrfs_ino(inode);
3692         u64 dir_ino = btrfs_ino(dir);
3693
3694         /*
3695          * 1 for the possible orphan item
3696          * 1 for the dir item
3697          * 1 for the dir index
3698          * 1 for the inode ref
3699          * 1 for the inode
3700          */
3701         trans = btrfs_start_transaction(root, 5);
3702         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3703                 return trans;
3704
3705         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
3706                 return ERR_PTR(-ENOSPC);
3707
3708         /* check if there is someone else holds reference */
3709         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
3710                 return ERR_PTR(-ENOSPC);
3711
3712         if (atomic_read(&inode->i_count) > 2)
3713                 return ERR_PTR(-ENOSPC);
3714
3715         if (xchg(&root->fs_info->enospc_unlink, 1))
3716                 return ERR_PTR(-ENOSPC);
3717
3718         path = btrfs_alloc_path();
3719         if (!path) {
3720                 root->fs_info->enospc_unlink = 0;
3721                 return ERR_PTR(-ENOMEM);
3722         }
3723
3724         /* 1 for the orphan item */
3725         trans = btrfs_start_transaction(root, 1);
3726         if (IS_ERR(trans)) {
3727                 btrfs_free_path(path);
3728                 root->fs_info->enospc_unlink = 0;
3729                 return trans;
3730         }
3731
3732         path->skip_locking = 1;
3733         path->search_commit_root = 1;
3734
3735         ret = btrfs_lookup_inode(trans, root, path,
3736                                 &BTRFS_I(dir)->location, 0);
3737         if (ret < 0) {
3738                 err = ret;
3739                 goto out;
3740         }
3741         if (ret == 0) {
3742                 if (check_path_shared(root, path))
3743                         goto out;
3744         } else {
3745                 check_link = 0;
3746         }
3747         btrfs_release_path(path);
3748
3749         ret = btrfs_lookup_inode(trans, root, path,
3750                                 &BTRFS_I(inode)->location, 0);
3751         if (ret < 0) {
3752                 err = ret;
3753                 goto out;
3754         }
3755         if (ret == 0) {
3756                 if (check_path_shared(root, path))
3757                         goto out;
3758         } else {
3759                 check_link = 0;
3760         }
3761         btrfs_release_path(path);
3762
3763         if (ret == 0 && S_ISREG(inode->i_mode)) {
3764                 ret = btrfs_lookup_file_extent(trans, root, path,
3765                                                ino, (u64)-1, 0);
3766                 if (ret < 0) {
3767                         err = ret;
3768                         goto out;
3769                 }
3770                 BUG_ON(ret == 0); /* Corruption */
3771                 if (check_path_shared(root, path))
3772                         goto out;
3773                 btrfs_release_path(path);
3774         }
3775
3776         if (!check_link) {
3777                 err = 0;
3778                 goto out;
3779         }
3780
3781         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3782                                 dentry->d_name.name, dentry->d_name.len, 0);
3783         if (IS_ERR(di)) {
3784                 err = PTR_ERR(di);
3785                 goto out;
3786         }
3787         if (di) {
3788                 if (check_path_shared(root, path))
3789                         goto out;
3790         } else {
3791                 err = 0;
3792                 goto out;
3793         }
3794         btrfs_release_path(path);
3795
3796         ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3797                                         dentry->d_name.len, ino, dir_ino, 0,
3798                                         &index);
3799         if (ret) {
3800                 err = ret;
3801                 goto out;
3802         }
3803
3804         if (check_path_shared(root, path))
3805                 goto out;
3806
3807         btrfs_release_path(path);
3808
3809         /*
3810          * This is a commit root search, if we can lookup inode item and other
3811          * relative items in the commit root, it means the transaction of
3812          * dir/file creation has been committed, and the dir index item that we
3813          * delay to insert has also been inserted into the commit root. So
3814          * we needn't worry about the delayed insertion of the dir index item
3815          * here.
3816          */
3817         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
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         BUG_ON(ret == -ENOENT);
3824         if (check_path_shared(root, path))
3825                 goto out;
3826
3827         err = 0;
3828 out:
3829         btrfs_free_path(path);
3830         /* Migrate the orphan reservation over */
3831         if (!err)
3832                 err = btrfs_block_rsv_migrate(trans->block_rsv,
3833                                 &root->fs_info->global_block_rsv,
3834                                 trans->bytes_reserved);
3835
3836         if (err) {
3837                 btrfs_end_transaction(trans, root);
3838                 root->fs_info->enospc_unlink = 0;
3839                 return ERR_PTR(err);
3840         }
3841
3842         trans->block_rsv = &root->fs_info->global_block_rsv;
3843         return trans;
3844 }
3845
3846 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3847                                struct btrfs_root *root)
3848 {
3849         if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3850                 btrfs_block_rsv_release(root, trans->block_rsv,
3851                                         trans->bytes_reserved);
3852                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3853                 BUG_ON(!root->fs_info->enospc_unlink);
3854                 root->fs_info->enospc_unlink = 0;
3855         }
3856         btrfs_end_transaction(trans, root);
3857 }
3858
3859 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3860 {
3861         struct btrfs_root *root = BTRFS_I(dir)->root;
3862         struct btrfs_trans_handle *trans;
3863         struct inode *inode = dentry->d_inode;
3864         int ret;
3865
3866         trans = __unlink_start_trans(dir, dentry);
3867         if (IS_ERR(trans))
3868                 return PTR_ERR(trans);
3869
3870         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3871
3872         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3873                                  dentry->d_name.name, dentry->d_name.len);
3874         if (ret)
3875                 goto out;
3876
3877         if (inode->i_nlink == 0) {
3878                 ret = btrfs_orphan_add(trans, inode);
3879                 if (ret)
3880                         goto out;
3881         }
3882
3883 out:
3884         __unlink_end_trans(trans, root);
3885         btrfs_btree_balance_dirty(root);
3886         return ret;
3887 }
3888
3889 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3890                         struct btrfs_root *root,
3891                         struct inode *dir, u64 objectid,
3892                         const char *name, int name_len)
3893 {
3894         struct btrfs_path *path;
3895         struct extent_buffer *leaf;
3896         struct btrfs_dir_item *di;
3897         struct btrfs_key key;
3898         u64 index;
3899         int ret;
3900         u64 dir_ino = btrfs_ino(dir);
3901
3902         path = btrfs_alloc_path();
3903         if (!path)
3904                 return -ENOMEM;
3905
3906         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3907                                    name, name_len, -1);
3908         if (IS_ERR_OR_NULL(di)) {
3909                 if (!di)
3910                         ret = -ENOENT;
3911                 else
3912                         ret = PTR_ERR(di);
3913                 goto out;
3914         }
3915
3916         leaf = path->nodes[0];
3917         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3918         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3919         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3920         if (ret) {
3921                 btrfs_abort_transaction(trans, root, ret);
3922                 goto out;
3923         }
3924         btrfs_release_path(path);
3925
3926         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3927                                  objectid, root->root_key.objectid,
3928                                  dir_ino, &index, name, name_len);
3929         if (ret < 0) {
3930                 if (ret != -ENOENT) {
3931                         btrfs_abort_transaction(trans, root, ret);
3932                         goto out;
3933                 }
3934                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3935                                                  name, name_len);
3936                 if (IS_ERR_OR_NULL(di)) {
3937                         if (!di)
3938                                 ret = -ENOENT;
3939                         else
3940                                 ret = PTR_ERR(di);
3941                         btrfs_abort_transaction(trans, root, ret);
3942                         goto out;
3943                 }
3944
3945                 leaf = path->nodes[0];
3946                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3947                 btrfs_release_path(path);
3948                 index = key.offset;
3949         }
3950         btrfs_release_path(path);
3951
3952         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3953         if (ret) {
3954                 btrfs_abort_transaction(trans, root, ret);
3955                 goto out;
3956         }
3957
3958         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3959         inode_inc_iversion(dir);
3960         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3961         ret = btrfs_update_inode_fallback(trans, root, dir);
3962         if (ret)
3963                 btrfs_abort_transaction(trans, root, ret);
3964 out:
3965         btrfs_free_path(path);
3966         return ret;
3967 }
3968
3969 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3970 {
3971         struct inode *inode = dentry->d_inode;
3972         int err = 0;
3973         struct btrfs_root *root = BTRFS_I(dir)->root;
3974         struct btrfs_trans_handle *trans;
3975
3976         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3977                 return -ENOTEMPTY;
3978         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3979                 return -EPERM;
3980
3981         trans = __unlink_start_trans(dir, dentry);
3982         if (IS_ERR(trans))
3983                 return PTR_ERR(trans);
3984
3985         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3986                 err = btrfs_unlink_subvol(trans, root, dir,
3987                                           BTRFS_I(inode)->location.objectid,
3988                                           dentry->d_name.name,
3989                                           dentry->d_name.len);
3990                 goto out;
3991         }
3992
3993         err = btrfs_orphan_add(trans, inode);
3994         if (err)
3995                 goto out;
3996
3997         /* now the directory is empty */
3998         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3999                                  dentry->d_name.name, dentry->d_name.len);
4000         if (!err)
4001                 btrfs_i_size_write(inode, 0);
4002 out:
4003         __unlink_end_trans(trans, root);
4004         btrfs_btree_balance_dirty(root);
4005
4006         return err;
4007 }
4008
4009 /*
4010  * this can truncate away extent items, csum items and directory items.
4011  * It starts at a high offset and removes keys until it can't find
4012  * any higher than new_size
4013  *
4014  * csum items that cross the new i_size are truncated to the new size
4015  * as well.
4016  *
4017  * min_type is the minimum key type to truncate down to.  If set to 0, this
4018  * will kill all the items on this inode, including the INODE_ITEM_KEY.
4019  */
4020 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4021                                struct btrfs_root *root,
4022                                struct inode *inode,
4023                                u64 new_size, u32 min_type)
4024 {
4025         struct btrfs_path *path;
4026         struct extent_buffer *leaf;
4027         struct btrfs_file_extent_item *fi;
4028         struct btrfs_key key;
4029         struct btrfs_key found_key;
4030         u64 extent_start = 0;
4031         u64 extent_num_bytes = 0;
4032         u64 extent_offset = 0;
4033         u64 item_end = 0;
4034         u32 found_type = (u8)-1;
4035         int found_extent;
4036         int del_item;
4037         int pending_del_nr = 0;
4038         int pending_del_slot = 0;
4039         int extent_type = -1;
4040         int ret;
4041         int err = 0;
4042         u64 ino = btrfs_ino(inode);
4043
4044         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4045
4046         path = btrfs_alloc_path();
4047         if (!path)
4048                 return -ENOMEM;
4049         path->reada = -1;
4050
4051         /*
4052          * We want to drop from the next block forward in case this new size is
4053          * not block aligned since we will be keeping the last block of the
4054          * extent just the way it is.
4055          */
4056         if (root->ref_cows || root == root->fs_info->tree_root)
4057                 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4058                                         root->sectorsize), (u64)-1, 0);
4059
4060         /*
4061          * This function is also used to drop the items in the log tree before
4062          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4063          * it is used to drop the loged items. So we shouldn't kill the delayed
4064          * items.
4065          */
4066         if (min_type == 0 && root == BTRFS_I(inode)->root)
4067                 btrfs_kill_delayed_inode_items(inode);
4068
4069         key.objectid = ino;
4070         key.offset = (u64)-1;
4071         key.type = (u8)-1;
4072
4073 search_again:
4074         path->leave_spinning = 1;
4075         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4076         if (ret < 0) {
4077                 err = ret;
4078                 goto out;
4079         }
4080
4081         if (ret > 0) {
4082                 /* there are no items in the tree for us to truncate, we're
4083                  * done
4084                  */
4085                 if (path->slots[0] == 0)
4086                         goto out;
4087                 path->slots[0]--;
4088         }
4089
4090         while (1) {
4091                 fi = NULL;
4092                 leaf = path->nodes[0];
4093                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4094                 found_type = btrfs_key_type(&found_key);
4095
4096                 if (found_key.objectid != ino)
4097                         break;
4098
4099                 if (found_type < min_type)
4100                         break;
4101
4102                 item_end = found_key.offset;
4103                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4104                         fi = btrfs_item_ptr(leaf, path->slots[0],
4105                                             struct btrfs_file_extent_item);
4106                         extent_type = btrfs_file_extent_type(leaf, fi);
4107                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4108                                 item_end +=
4109                                     btrfs_file_extent_num_bytes(leaf, fi);
4110                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4111                                 item_end += btrfs_file_extent_inline_len(leaf,
4112                                                                          fi);
4113                         }
4114                         item_end--;
4115                 }
4116                 if (found_type > min_type) {
4117                         del_item = 1;
4118                 } else {
4119                         if (item_end < new_size)
4120                                 break;
4121                         if (found_key.offset >= new_size)
4122                                 del_item = 1;
4123                         else
4124                                 del_item = 0;
4125                 }
4126                 found_extent = 0;
4127                 /* FIXME, shrink the extent if the ref count is only 1 */
4128                 if (found_type != BTRFS_EXTENT_DATA_KEY)
4129                         goto delete;
4130
4131                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4132                         u64 num_dec;
4133                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4134                         if (!del_item) {
4135                                 u64 orig_num_bytes =
4136                                         btrfs_file_extent_num_bytes(leaf, fi);
4137                                 extent_num_bytes = ALIGN(new_size -
4138                                                 found_key.offset,
4139                                                 root->sectorsize);
4140                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4141                                                          extent_num_bytes);
4142                                 num_dec = (orig_num_bytes -
4143                                            extent_num_bytes);
4144                                 if (root->ref_cows && extent_start != 0)
4145                                         inode_sub_bytes(inode, num_dec);
4146                                 btrfs_mark_buffer_dirty(leaf);
4147                         } else {
4148                                 extent_num_bytes =
4149                                         btrfs_file_extent_disk_num_bytes(leaf,
4150                                                                          fi);
4151                                 extent_offset = found_key.offset -
4152                                         btrfs_file_extent_offset(leaf, fi);
4153
4154                                 /* FIXME blocksize != 4096 */
4155                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4156                                 if (extent_start != 0) {
4157                                         found_extent = 1;
4158                                         if (root->ref_cows)
4159                                                 inode_sub_bytes(inode, num_dec);
4160                                 }
4161                         }
4162                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4163                         /*
4164                          * we can't truncate inline items that have had
4165                          * special encodings
4166                          */
4167                         if (!del_item &&
4168                             btrfs_file_extent_compression(leaf, fi) == 0 &&
4169                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
4170                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4171                                 u32 size = new_size - found_key.offset;
4172
4173                                 if (root->ref_cows) {
4174                                         inode_sub_bytes(inode, item_end + 1 -
4175                                                         new_size);
4176                                 }
4177                                 size =
4178                                     btrfs_file_extent_calc_inline_size(size);
4179                                 btrfs_truncate_item(trans, root, path,
4180                                                     size, 1);
4181                         } else if (root->ref_cows) {
4182                                 inode_sub_bytes(inode, item_end + 1 -
4183                                                 found_key.offset);
4184                         }
4185                 }
4186 delete:
4187                 if (del_item) {
4188                         if (!pending_del_nr) {
4189                                 /* no pending yet, add ourselves */
4190                                 pending_del_slot = path->slots[0];
4191                                 pending_del_nr = 1;
4192                         } else if (pending_del_nr &&
4193                                    path->slots[0] + 1 == pending_del_slot) {
4194                                 /* hop on the pending chunk */
4195                                 pending_del_nr++;
4196                                 pending_del_slot = path->slots[0];
4197                         } else {
4198                                 BUG();
4199                         }
4200                 } else {
4201                         break;
4202                 }
4203                 if (found_extent && (root->ref_cows ||
4204                                      root == root->fs_info->tree_root)) {
4205                         btrfs_set_path_blocking(path);
4206                         ret = btrfs_free_extent(trans, root, extent_start,
4207                                                 extent_num_bytes, 0,
4208                                                 btrfs_header_owner(leaf),
4209                                                 ino, extent_offset, 0);
4210                         BUG_ON(ret);
4211                 }
4212
4213                 if (found_type == BTRFS_INODE_ITEM_KEY)
4214                         break;
4215
4216                 if (path->slots[0] == 0 ||
4217                     path->slots[0] != pending_del_slot) {
4218                         if (pending_del_nr) {
4219                                 ret = btrfs_del_items(trans, root, path,
4220                                                 pending_del_slot,
4221                                                 pending_del_nr);
4222                                 if (ret) {
4223                                         btrfs_abort_transaction(trans,
4224                                                                 root, ret);
4225                                         goto error;
4226                                 }
4227                                 pending_del_nr = 0;
4228                         }
4229                         btrfs_release_path(path);
4230                         goto search_again;
4231                 } else {
4232                         path->slots[0]--;
4233                 }
4234         }
4235 out:
4236         if (pending_del_nr) {
4237                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4238                                       pending_del_nr);
4239                 if (ret)
4240                         btrfs_abort_transaction(trans, root, ret);
4241         }
4242 error:
4243         btrfs_free_path(path);
4244         return err;
4245 }
4246
4247 /*
4248  * btrfs_truncate_page - read, zero a chunk and write a page
4249  * @inode - inode that we're zeroing
4250  * @from - the offset to start zeroing
4251  * @len - the length to zero, 0 to zero the entire range respective to the
4252  *      offset
4253  * @front - zero up to the offset instead of from the offset on
4254  *
4255  * This will find the page for the "from" offset and cow the page and zero the
4256  * part we want to zero.  This is used with truncate and hole punching.
4257  */
4258 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4259                         int front)
4260 {
4261         struct address_space *mapping = inode->i_mapping;
4262         struct btrfs_root *root = BTRFS_I(inode)->root;
4263         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4264         struct btrfs_ordered_extent *ordered;
4265         struct extent_state *cached_state = NULL;
4266         char *kaddr;
4267         u32 blocksize = root->sectorsize;
4268         pgoff_t index = from >> PAGE_CACHE_SHIFT;
4269         unsigned offset = from & (PAGE_CACHE_SIZE-1);
4270         struct page *page;
4271         gfp_t mask = btrfs_alloc_write_mask(mapping);
4272         int ret = 0;
4273         u64 page_start;
4274         u64 page_end;
4275
4276         if ((offset & (blocksize - 1)) == 0 &&
4277             (!len || ((len & (blocksize - 1)) == 0)))
4278                 goto out;
4279         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4280         if (ret)
4281                 goto out;
4282
4283 again:
4284         page = find_or_create_page(mapping, index, mask);
4285         if (!page) {
4286                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4287                 ret = -ENOMEM;
4288                 goto out;
4289         }
4290
4291         page_start = page_offset(page);
4292         page_end = page_start + PAGE_CACHE_SIZE - 1;
4293
4294         if (!PageUptodate(page)) {
4295                 ret = btrfs_readpage(NULL, page);
4296                 lock_page(page);
4297                 if (page->mapping != mapping) {
4298                         unlock_page(page);
4299                         page_cache_release(page);
4300                         goto again;
4301                 }
4302                 if (!PageUptodate(page)) {
4303                         ret = -EIO;
4304                         goto out_unlock;
4305                 }
4306         }
4307         wait_on_page_writeback(page);
4308
4309         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4310         set_page_extent_mapped(page);
4311
4312         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4313         if (ordered) {
4314                 unlock_extent_cached(io_tree, page_start, page_end,
4315                                      &cached_state, GFP_NOFS);
4316                 unlock_page(page);
4317                 page_cache_release(page);
4318                 btrfs_start_ordered_extent(inode, ordered, 1);
4319                 btrfs_put_ordered_extent(ordered);
4320                 goto again;
4321         }
4322
4323         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4324                           EXTENT_DIRTY | EXTENT_DELALLOC |
4325                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4326                           0, 0, &cached_state, GFP_NOFS);
4327
4328         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4329                                         &cached_state);
4330         if (ret) {
4331                 unlock_extent_cached(io_tree, page_start, page_end,
4332                                      &cached_state, GFP_NOFS);
4333                 goto out_unlock;
4334         }
4335
4336         if (offset != PAGE_CACHE_SIZE) {
4337                 if (!len)
4338                         len = PAGE_CACHE_SIZE - offset;
4339                 kaddr = kmap(page);
4340                 if (front)
4341                         memset(kaddr, 0, offset);
4342                 else
4343                         memset(kaddr + offset, 0, len);
4344                 flush_dcache_page(page);
4345                 kunmap(page);
4346         }
4347         ClearPageChecked(page);
4348         set_page_dirty(page);
4349         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4350                              GFP_NOFS);
4351
4352 out_unlock:
4353         if (ret)
4354                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4355         unlock_page(page);
4356         page_cache_release(page);
4357 out:
4358         return ret;
4359 }
4360
4361 /*
4362  * This function puts in dummy file extents for the area we're creating a hole
4363  * for.  So if we are truncating this file to a larger size we need to insert
4364  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4365  * the range between oldsize and size
4366  */
4367 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4368 {
4369         struct btrfs_trans_handle *trans;
4370         struct btrfs_root *root = BTRFS_I(inode)->root;
4371         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4372         struct extent_map *em = NULL;
4373         struct extent_state *cached_state = NULL;
4374         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4375         u64 hole_start = ALIGN(oldsize, root->sectorsize);
4376         u64 block_end = ALIGN(size, root->sectorsize);
4377         u64 last_byte;
4378         u64 cur_offset;
4379         u64 hole_size;
4380         int err = 0;
4381
4382         if (size <= hole_start)
4383                 return 0;
4384
4385         while (1) {
4386                 struct btrfs_ordered_extent *ordered;
4387                 btrfs_wait_ordered_range(inode, hole_start,
4388                                          block_end - hole_start);
4389                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4390                                  &cached_state);
4391                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
4392                 if (!ordered)
4393                         break;
4394                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4395                                      &cached_state, GFP_NOFS);
4396                 btrfs_put_ordered_extent(ordered);
4397         }
4398
4399         cur_offset = hole_start;
4400         while (1) {
4401                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4402                                 block_end - cur_offset, 0);
4403                 if (IS_ERR(em)) {
4404                         err = PTR_ERR(em);
4405                         em = NULL;
4406                         break;
4407                 }
4408                 last_byte = min(extent_map_end(em), block_end);
4409                 last_byte = ALIGN(last_byte , root->sectorsize);
4410                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4411                         struct extent_map *hole_em;
4412                         hole_size = last_byte - cur_offset;
4413
4414                         trans = btrfs_start_transaction(root, 3);
4415                         if (IS_ERR(trans)) {
4416                                 err = PTR_ERR(trans);
4417                                 break;
4418                         }
4419
4420                         err = btrfs_drop_extents(trans, root, inode,
4421                                                  cur_offset,
4422                                                  cur_offset + hole_size, 1);
4423                         if (err) {
4424                                 btrfs_abort_transaction(trans, root, err);
4425                                 btrfs_end_transaction(trans, root);
4426                                 break;
4427                         }
4428
4429                         err = btrfs_insert_file_extent(trans, root,
4430                                         btrfs_ino(inode), cur_offset, 0,
4431                                         0, hole_size, 0, hole_size,
4432                                         0, 0, 0);
4433                         if (err) {
4434                                 btrfs_abort_transaction(trans, root, err);
4435                                 btrfs_end_transaction(trans, root);
4436                                 break;
4437                         }
4438
4439                         btrfs_drop_extent_cache(inode, cur_offset,
4440                                                 cur_offset + hole_size - 1, 0);
4441                         hole_em = alloc_extent_map();
4442                         if (!hole_em) {
4443                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4444                                         &BTRFS_I(inode)->runtime_flags);
4445                                 goto next;
4446                         }
4447                         hole_em->start = cur_offset;
4448                         hole_em->len = hole_size;
4449                         hole_em->orig_start = cur_offset;
4450
4451                         hole_em->block_start = EXTENT_MAP_HOLE;
4452                         hole_em->block_len = 0;
4453                         hole_em->orig_block_len = 0;
4454                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4455                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
4456                         hole_em->generation = trans->transid;
4457
4458                         while (1) {
4459                                 write_lock(&em_tree->lock);
4460                                 err = add_extent_mapping(em_tree, hole_em);
4461                                 if (!err)
4462                                         list_move(&hole_em->list,
4463                                                   &em_tree->modified_extents);
4464                                 write_unlock(&em_tree->lock);
4465                                 if (err != -EEXIST)
4466                                         break;
4467                                 btrfs_drop_extent_cache(inode, cur_offset,
4468                                                         cur_offset +
4469                                                         hole_size - 1, 0);
4470                         }
4471                         free_extent_map(hole_em);
4472 next:
4473                         btrfs_update_inode(trans, root, inode);
4474                         btrfs_end_transaction(trans, root);
4475                 }
4476                 free_extent_map(em);
4477                 em = NULL;
4478                 cur_offset = last_byte;
4479                 if (cur_offset >= block_end)
4480                         break;
4481         }
4482
4483         free_extent_map(em);
4484         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4485                              GFP_NOFS);
4486         return err;
4487 }
4488
4489 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4490 {
4491         struct btrfs_root *root = BTRFS_I(inode)->root;
4492         struct btrfs_trans_handle *trans;
4493         loff_t oldsize = i_size_read(inode);
4494         loff_t newsize = attr->ia_size;
4495         int mask = attr->ia_valid;
4496         int ret;
4497
4498         if (newsize == oldsize)
4499                 return 0;
4500
4501         /*
4502          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4503          * special case where we need to update the times despite not having
4504          * these flags set.  For all other operations the VFS set these flags
4505          * explicitly if it wants a timestamp update.
4506          */
4507         if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4508                 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4509
4510         if (newsize > oldsize) {
4511                 truncate_pagecache(inode, oldsize, newsize);
4512                 ret = btrfs_cont_expand(inode, oldsize, newsize);
4513                 if (ret)
4514                         return ret;
4515
4516                 trans = btrfs_start_transaction(root, 1);
4517                 if (IS_ERR(trans))
4518                         return PTR_ERR(trans);
4519
4520                 i_size_write(inode, newsize);
4521                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4522                 ret = btrfs_update_inode(trans, root, inode);
4523                 btrfs_end_transaction(trans, root);
4524         } else {
4525
4526                 /*
4527                  * We're truncating a file that used to have good data down to
4528                  * zero. Make sure it gets into the ordered flush list so that
4529                  * any new writes get down to disk quickly.
4530                  */
4531                 if (newsize == 0)
4532                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4533                                 &BTRFS_I(inode)->runtime_flags);
4534
4535                 /*
4536                  * 1 for the orphan item we're going to add
4537                  * 1 for the orphan item deletion.
4538                  */
4539                 trans = btrfs_start_transaction(root, 2);
4540                 if (IS_ERR(trans))
4541                         return PTR_ERR(trans);
4542
4543                 /*
4544                  * We need to do this in case we fail at _any_ point during the
4545                  * actual truncate.  Once we do the truncate_setsize we could
4546                  * invalidate pages which forces any outstanding ordered io to
4547                  * be instantly completed which will give us extents that need
4548                  * to be truncated.  If we fail to get an orphan inode down we
4549                  * could have left over extents that were never meant to live,
4550                  * so we need to garuntee from this point on that everything
4551                  * will be consistent.
4552                  */
4553                 ret = btrfs_orphan_add(trans, inode);
4554                 btrfs_end_transaction(trans, root);
4555                 if (ret)
4556                         return ret;
4557
4558                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4559                 truncate_setsize(inode, newsize);
4560
4561                 /* Disable nonlocked read DIO to avoid the end less truncate */
4562                 btrfs_inode_block_unlocked_dio(inode);
4563                 inode_dio_wait(inode);
4564                 btrfs_inode_resume_unlocked_dio(inode);
4565
4566                 ret = btrfs_truncate(inode);
4567                 if (ret && inode->i_nlink)
4568                         btrfs_orphan_del(NULL, inode);
4569         }
4570
4571         return ret;
4572 }
4573
4574 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4575 {
4576         struct inode *inode = dentry->d_inode;
4577         struct btrfs_root *root = BTRFS_I(inode)->root;
4578         int err;
4579
4580         if (btrfs_root_readonly(root))
4581                 return -EROFS;
4582
4583         err = inode_change_ok(inode, attr);
4584         if (err)
4585                 return err;
4586
4587         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4588                 err = btrfs_setsize(inode, attr);
4589                 if (err)
4590                         return err;
4591         }
4592
4593         if (attr->ia_valid) {
4594                 setattr_copy(inode, attr);
4595                 inode_inc_iversion(inode);
4596                 err = btrfs_dirty_inode(inode);
4597
4598                 if (!err && attr->ia_valid & ATTR_MODE)
4599                         err = btrfs_acl_chmod(inode);
4600         }
4601
4602         return err;
4603 }
4604
4605 void btrfs_evict_inode(struct inode *inode)
4606 {
4607         struct btrfs_trans_handle *trans;
4608         struct btrfs_root *root = BTRFS_I(inode)->root;
4609         struct btrfs_block_rsv *rsv, *global_rsv;
4610         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4611         int ret;
4612
4613         trace_btrfs_inode_evict(inode);
4614
4615         truncate_inode_pages(&inode->i_data, 0);
4616         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
4617                                btrfs_is_free_space_inode(inode)))
4618                 goto no_delete;
4619
4620         if (is_bad_inode(inode)) {
4621                 btrfs_orphan_del(NULL, inode);
4622                 goto no_delete;
4623         }
4624         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4625         btrfs_wait_ordered_range(inode, 0, (u64)-1);
4626
4627         if (root->fs_info->log_root_recovering) {
4628                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4629                                  &BTRFS_I(inode)->runtime_flags));
4630                 goto no_delete;
4631         }
4632
4633         if (inode->i_nlink > 0) {
4634                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
4635                 goto no_delete;
4636         }
4637
4638         ret = btrfs_commit_inode_delayed_inode(inode);
4639         if (ret) {
4640                 btrfs_orphan_del(NULL, inode);
4641                 goto no_delete;
4642         }
4643
4644         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4645         if (!rsv) {
4646                 btrfs_orphan_del(NULL, inode);
4647                 goto no_delete;
4648         }
4649         rsv->size = min_size;
4650         rsv->failfast = 1;
4651         global_rsv = &root->fs_info->global_block_rsv;
4652
4653         btrfs_i_size_write(inode, 0);
4654
4655         /*
4656          * This is a bit simpler than btrfs_truncate since we've already
4657          * reserved our space for our orphan item in the unlink, so we just
4658          * need to reserve some slack space in case we add bytes and update
4659          * inode item when doing the truncate.
4660          */
4661         while (1) {
4662                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4663                                              BTRFS_RESERVE_FLUSH_LIMIT);
4664
4665                 /*
4666                  * Try and steal from the global reserve since we will
4667                  * likely not use this space anyway, we want to try as
4668                  * hard as possible to get this to work.
4669                  */
4670                 if (ret)
4671                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4672
4673                 if (ret) {
4674                         printk(KERN_WARNING "Could not get space for a "
4675                                "delete, will truncate on mount %d\n", ret);
4676                         btrfs_orphan_del(NULL, inode);
4677                         btrfs_free_block_rsv(root, rsv);
4678                         goto no_delete;
4679                 }
4680
4681                 trans = btrfs_join_transaction(root);
4682                 if (IS_ERR(trans)) {
4683                         btrfs_orphan_del(NULL, inode);
4684                         btrfs_free_block_rsv(root, rsv);
4685                         goto no_delete;
4686                 }
4687
4688                 trans->block_rsv = rsv;
4689
4690                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4691                 if (ret != -ENOSPC)
4692                         break;
4693
4694                 trans->block_rsv = &root->fs_info->trans_block_rsv;
4695                 btrfs_end_transaction(trans, root);
4696                 trans = NULL;
4697                 btrfs_btree_balance_dirty(root);
4698         }
4699
4700         btrfs_free_block_rsv(root, rsv);
4701
4702         if (ret == 0) {
4703                 trans->block_rsv = root->orphan_block_rsv;
4704                 ret = btrfs_orphan_del(trans, inode);
4705                 BUG_ON(ret);
4706         }
4707
4708         trans->block_rsv = &root->fs_info->trans_block_rsv;
4709         if (!(root == root->fs_info->tree_root ||
4710               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4711                 btrfs_return_ino(root, btrfs_ino(inode));
4712
4713         btrfs_end_transaction(trans, root);
4714         btrfs_btree_balance_dirty(root);
4715 no_delete:
4716         clear_inode(inode);
4717         return;
4718 }
4719
4720 /*
4721  * this returns the key found in the dir entry in the location pointer.
4722  * If no dir entries were found, location->objectid is 0.
4723  */
4724 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4725                                struct btrfs_key *location)
4726 {
4727         const char *name = dentry->d_name.name;
4728         int namelen = dentry->d_name.len;
4729         struct btrfs_dir_item *di;
4730         struct btrfs_path *path;
4731         struct btrfs_root *root = BTRFS_I(dir)->root;
4732         int ret = 0;
4733
4734         path = btrfs_alloc_path();
4735         if (!path)
4736                 return -ENOMEM;
4737
4738         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4739                                     namelen, 0);
4740         if (IS_ERR(di))
4741                 ret = PTR_ERR(di);
4742
4743         if (IS_ERR_OR_NULL(di))
4744                 goto out_err;
4745
4746         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4747 out:
4748         btrfs_free_path(path);
4749         return ret;
4750 out_err:
4751         location->objectid = 0;
4752         goto out;
4753 }
4754
4755 /*
4756  * when we hit a tree root in a directory, the btrfs part of the inode
4757  * needs to be changed to reflect the root directory of the tree root.  This
4758  * is kind of like crossing a mount point.
4759  */
4760 static int fixup_tree_root_location(struct btrfs_root *root,
4761                                     struct inode *dir,
4762                                     struct dentry *dentry,
4763                                     struct btrfs_key *location,
4764                                     struct btrfs_root **sub_root)
4765 {
4766         struct btrfs_path *path;
4767         struct btrfs_root *new_root;
4768         struct btrfs_root_ref *ref;
4769         struct extent_buffer *leaf;
4770         int ret;
4771         int err = 0;
4772
4773         path = btrfs_alloc_path();
4774         if (!path) {
4775                 err = -ENOMEM;
4776                 goto out;
4777         }
4778
4779         err = -ENOENT;
4780         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4781                                   BTRFS_I(dir)->root->root_key.objectid,
4782                                   location->objectid);
4783         if (ret) {
4784                 if (ret < 0)
4785                         err = ret;
4786                 goto out;
4787         }
4788
4789         leaf = path->nodes[0];
4790         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4791         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4792             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4793                 goto out;
4794
4795         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4796                                    (unsigned long)(ref + 1),
4797                                    dentry->d_name.len);
4798         if (ret)
4799                 goto out;
4800
4801         btrfs_release_path(path);
4802
4803         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4804         if (IS_ERR(new_root)) {
4805                 err = PTR_ERR(new_root);
4806                 goto out;
4807         }
4808
4809         if (btrfs_root_refs(&new_root->root_item) == 0) {
4810                 err = -ENOENT;
4811                 goto out;
4812         }
4813
4814         *sub_root = new_root;
4815         location->objectid = btrfs_root_dirid(&new_root->root_item);
4816         location->type = BTRFS_INODE_ITEM_KEY;
4817         location->offset = 0;
4818         err = 0;
4819 out:
4820         btrfs_free_path(path);
4821         return err;
4822 }
4823
4824 static void inode_tree_add(struct inode *inode)
4825 {
4826         struct btrfs_root *root = BTRFS_I(inode)->root;
4827         struct btrfs_inode *entry;
4828         struct rb_node **p;
4829         struct rb_node *parent;
4830         u64 ino = btrfs_ino(inode);
4831 again:
4832         p = &root->inode_tree.rb_node;
4833         parent = NULL;
4834
4835         if (inode_unhashed(inode))
4836                 return;
4837
4838         spin_lock(&root->inode_lock);
4839         while (*p) {
4840                 parent = *p;
4841                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4842
4843                 if (ino < btrfs_ino(&entry->vfs_inode))
4844                         p = &parent->rb_left;
4845                 else if (ino > btrfs_ino(&entry->vfs_inode))
4846                         p = &parent->rb_right;
4847                 else {
4848                         WARN_ON(!(entry->vfs_inode.i_state &
4849                                   (I_WILL_FREE | I_FREEING)));
4850                         rb_erase(parent, &root->inode_tree);
4851                         RB_CLEAR_NODE(parent);
4852                         spin_unlock(&root->inode_lock);
4853                         goto again;
4854                 }
4855         }
4856         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4857         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4858         spin_unlock(&root->inode_lock);
4859 }
4860
4861 static void inode_tree_del(struct inode *inode)
4862 {
4863         struct btrfs_root *root = BTRFS_I(inode)->root;
4864         int empty = 0;
4865
4866         spin_lock(&root->inode_lock);
4867         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4868                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4869                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4870                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4871         }
4872         spin_unlock(&root->inode_lock);
4873
4874         /*
4875          * Free space cache has inodes in the tree root, but the tree root has a
4876          * root_refs of 0, so this could end up dropping the tree root as a
4877          * snapshot, so we need the extra !root->fs_info->tree_root check to
4878          * make sure we don't drop it.
4879          */
4880         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4881             root != root->fs_info->tree_root) {
4882                 synchronize_srcu(&root->fs_info->subvol_srcu);
4883                 spin_lock(&root->inode_lock);
4884                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4885                 spin_unlock(&root->inode_lock);
4886                 if (empty)
4887                         btrfs_add_dead_root(root);
4888         }
4889 }
4890
4891 void btrfs_invalidate_inodes(struct btrfs_root *root)
4892 {
4893         struct rb_node *node;
4894         struct rb_node *prev;
4895         struct btrfs_inode *entry;
4896         struct inode *inode;
4897         u64 objectid = 0;
4898
4899         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4900
4901         spin_lock(&root->inode_lock);
4902 again:
4903         node = root->inode_tree.rb_node;
4904         prev = NULL;
4905         while (node) {
4906                 prev = node;
4907                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4908
4909                 if (objectid < btrfs_ino(&entry->vfs_inode))
4910                         node = node->rb_left;
4911                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4912                         node = node->rb_right;
4913                 else
4914                         break;
4915         }
4916         if (!node) {
4917                 while (prev) {
4918                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4919                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4920                                 node = prev;
4921                                 break;
4922                         }
4923                         prev = rb_next(prev);
4924                 }
4925         }
4926         while (node) {
4927                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4928                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4929                 inode = igrab(&entry->vfs_inode);
4930                 if (inode) {
4931                         spin_unlock(&root->inode_lock);
4932                         if (atomic_read(&inode->i_count) > 1)
4933                                 d_prune_aliases(inode);
4934                         /*
4935                          * btrfs_drop_inode will have it removed from
4936                          * the inode cache when its usage count
4937                          * hits zero.
4938                          */
4939                         iput(inode);
4940                         cond_resched();
4941                         spin_lock(&root->inode_lock);
4942                         goto again;
4943                 }
4944
4945                 if (cond_resched_lock(&root->inode_lock))
4946                         goto again;
4947
4948                 node = rb_next(node);
4949         }
4950         spin_unlock(&root->inode_lock);
4951 }
4952
4953 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4954 {
4955         struct btrfs_iget_args *args = p;
4956         inode->i_ino = args->ino;
4957         BTRFS_I(inode)->root = args->root;
4958         return 0;
4959 }
4960
4961 static int btrfs_find_actor(struct inode *inode, void *opaque)
4962 {
4963         struct btrfs_iget_args *args = opaque;
4964         return args->ino == btrfs_ino(inode) &&
4965                 args->root == BTRFS_I(inode)->root;
4966 }
4967
4968 static struct inode *btrfs_iget_locked(struct super_block *s,
4969                                        u64 objectid,
4970                                        struct btrfs_root *root)
4971 {
4972         struct inode *inode;
4973         struct btrfs_iget_args args;
4974         args.ino = objectid;
4975         args.root = root;
4976
4977         inode = iget5_locked(s, objectid, btrfs_find_actor,
4978                              btrfs_init_locked_inode,
4979                              (void *)&args);
4980         return inode;
4981 }
4982
4983 /* Get an inode object given its location and corresponding root.
4984  * Returns in *is_new if the inode was read from disk
4985  */
4986 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4987                          struct btrfs_root *root, int *new)
4988 {
4989         struct inode *inode;
4990
4991         inode = btrfs_iget_locked(s, location->objectid, root);
4992         if (!inode)
4993                 return ERR_PTR(-ENOMEM);
4994
4995         if (inode->i_state & I_NEW) {
4996                 BTRFS_I(inode)->root = root;
4997                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4998                 btrfs_read_locked_inode(inode);
4999                 if (!is_bad_inode(inode)) {
5000                         inode_tree_add(inode);
5001                         unlock_new_inode(inode);
5002                         if (new)
5003                                 *new = 1;
5004                 } else {
5005                         unlock_new_inode(inode);
5006                         iput(inode);
5007                         inode = ERR_PTR(-ESTALE);
5008                 }
5009         }
5010
5011         return inode;
5012 }
5013
5014 static struct inode *new_simple_dir(struct super_block *s,
5015                                     struct btrfs_key *key,
5016                                     struct btrfs_root *root)
5017 {
5018         struct inode *inode = new_inode(s);
5019
5020         if (!inode)
5021                 return ERR_PTR(-ENOMEM);
5022
5023         BTRFS_I(inode)->root = root;
5024         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5025         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5026
5027         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5028         inode->i_op = &btrfs_dir_ro_inode_operations;
5029         inode->i_fop = &simple_dir_operations;
5030         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5031         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5032
5033         return inode;
5034 }
5035
5036 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5037 {
5038         struct inode *inode;
5039         struct btrfs_root *root = BTRFS_I(dir)->root;
5040         struct btrfs_root *sub_root = root;
5041         struct btrfs_key location;
5042         int index;
5043         int ret = 0;
5044
5045         if (dentry->d_name.len > BTRFS_NAME_LEN)
5046                 return ERR_PTR(-ENAMETOOLONG);
5047
5048         ret = btrfs_inode_by_name(dir, dentry, &location);
5049         if (ret < 0)
5050                 return ERR_PTR(ret);
5051
5052         if (location.objectid == 0)
5053                 return NULL;
5054
5055         if (location.type == BTRFS_INODE_ITEM_KEY) {
5056                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5057                 return inode;
5058         }
5059
5060         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5061
5062         index = srcu_read_lock(&root->fs_info->subvol_srcu);
5063         ret = fixup_tree_root_location(root, dir, dentry,
5064                                        &location, &sub_root);
5065         if (ret < 0) {
5066                 if (ret != -ENOENT)
5067                         inode = ERR_PTR(ret);
5068                 else
5069                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
5070         } else {
5071                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5072         }
5073         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5074
5075         if (!IS_ERR(inode) && root != sub_root) {
5076                 down_read(&root->fs_info->cleanup_work_sem);
5077                 if (!(inode->i_sb->s_flags & MS_RDONLY))
5078                         ret = btrfs_orphan_cleanup(sub_root);
5079                 up_read(&root->fs_info->cleanup_work_sem);
5080                 if (ret)
5081                         inode = ERR_PTR(ret);
5082         }
5083
5084         return inode;
5085 }
5086
5087 static int btrfs_dentry_delete(const struct dentry *dentry)
5088 {
5089         struct btrfs_root *root;
5090         struct inode *inode = dentry->d_inode;
5091
5092         if (!inode && !IS_ROOT(dentry))
5093                 inode = dentry->d_parent->d_inode;
5094
5095         if (inode) {
5096                 root = BTRFS_I(inode)->root;
5097                 if (btrfs_root_refs(&root->root_item) == 0)
5098                         return 1;
5099
5100                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5101                         return 1;
5102         }
5103         return 0;
5104 }
5105
5106 static void btrfs_dentry_release(struct dentry *dentry)
5107 {
5108         if (dentry->d_fsdata)
5109                 kfree(dentry->d_fsdata);
5110 }
5111
5112 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5113                                    unsigned int flags)
5114 {
5115         struct dentry *ret;
5116
5117         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
5118         return ret;
5119 }
5120
5121 unsigned char btrfs_filetype_table[] = {
5122         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5123 };
5124
5125 static int btrfs_real_readdir(struct file *filp, void *dirent,
5126                               filldir_t filldir)
5127 {
5128         struct inode *inode = file_inode(filp);
5129         struct btrfs_root *root = BTRFS_I(inode)->root;
5130         struct btrfs_item *item;
5131         struct btrfs_dir_item *di;
5132         struct btrfs_key key;
5133         struct btrfs_key found_key;
5134         struct btrfs_path *path;
5135         struct list_head ins_list;
5136         struct list_head del_list;
5137         int ret;
5138         struct extent_buffer *leaf;
5139         int slot;
5140         unsigned char d_type;
5141         int over = 0;
5142         u32 di_cur;
5143         u32 di_total;
5144         u32 di_len;
5145         int key_type = BTRFS_DIR_INDEX_KEY;
5146         char tmp_name[32];
5147         char *name_ptr;
5148         int name_len;
5149         int is_curr = 0;        /* filp->f_pos points to the current index? */
5150
5151         /* FIXME, use a real flag for deciding about the key type */
5152         if (root->fs_info->tree_root == root)
5153                 key_type = BTRFS_DIR_ITEM_KEY;
5154
5155         /* special case for "." */
5156         if (filp->f_pos == 0) {
5157                 over = filldir(dirent, ".", 1,
5158                                filp->f_pos, btrfs_ino(inode), DT_DIR);
5159                 if (over)
5160                         return 0;
5161                 filp->f_pos = 1;
5162         }
5163         /* special case for .., just use the back ref */
5164         if (filp->f_pos == 1) {
5165                 u64 pino = parent_ino(filp->f_path.dentry);
5166                 over = filldir(dirent, "..", 2,
5167                                filp->f_pos, pino, DT_DIR);
5168                 if (over)
5169                         return 0;
5170                 filp->f_pos = 2;
5171         }
5172         path = btrfs_alloc_path();
5173         if (!path)
5174                 return -ENOMEM;
5175
5176         path->reada = 1;
5177
5178         if (key_type == BTRFS_DIR_INDEX_KEY) {
5179                 INIT_LIST_HEAD(&ins_list);
5180                 INIT_LIST_HEAD(&del_list);
5181                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5182         }
5183
5184         btrfs_set_key_type(&key, key_type);
5185         key.offset = filp->f_pos;
5186         key.objectid = btrfs_ino(inode);
5187
5188         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5189         if (ret < 0)
5190                 goto err;
5191
5192         while (1) {
5193                 leaf = path->nodes[0];
5194                 slot = path->slots[0];
5195                 if (slot >= btrfs_header_nritems(leaf)) {
5196                         ret = btrfs_next_leaf(root, path);
5197                         if (ret < 0)
5198                                 goto err;
5199                         else if (ret > 0)
5200                                 break;
5201                         continue;
5202                 }
5203
5204                 item = btrfs_item_nr(leaf, slot);
5205                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5206
5207                 if (found_key.objectid != key.objectid)
5208                         break;
5209                 if (btrfs_key_type(&found_key) != key_type)
5210                         break;
5211                 if (found_key.offset < filp->f_pos)
5212                         goto next;
5213                 if (key_type == BTRFS_DIR_INDEX_KEY &&
5214                     btrfs_should_delete_dir_index(&del_list,
5215                                                   found_key.offset))
5216                         goto next;
5217
5218                 filp->f_pos = found_key.offset;
5219                 is_curr = 1;
5220
5221                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5222                 di_cur = 0;
5223                 di_total = btrfs_item_size(leaf, item);
5224
5225                 while (di_cur < di_total) {
5226                         struct btrfs_key location;
5227
5228                         if (verify_dir_item(root, leaf, di))
5229                                 break;
5230
5231                         name_len = btrfs_dir_name_len(leaf, di);
5232                         if (name_len <= sizeof(tmp_name)) {
5233                                 name_ptr = tmp_name;
5234                         } else {
5235                                 name_ptr = kmalloc(name_len, GFP_NOFS);
5236                                 if (!name_ptr) {
5237                                         ret = -ENOMEM;
5238                                         goto err;
5239                                 }
5240                         }
5241                         read_extent_buffer(leaf, name_ptr,
5242                                            (unsigned long)(di + 1), name_len);
5243
5244                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5245                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
5246
5247
5248                         /* is this a reference to our own snapshot? If so
5249                          * skip it.
5250                          *
5251                          * In contrast to old kernels, we insert the snapshot's
5252                          * dir item and dir index after it has been created, so
5253                          * we won't find a reference to our own snapshot. We
5254                          * still keep the following code for backward
5255                          * compatibility.
5256                          */
5257                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
5258                             location.objectid == root->root_key.objectid) {
5259                                 over = 0;
5260                                 goto skip;
5261                         }
5262                         over = filldir(dirent, name_ptr, name_len,
5263                                        found_key.offset, location.objectid,
5264                                        d_type);
5265
5266 skip:
5267                         if (name_ptr != tmp_name)
5268                                 kfree(name_ptr);
5269
5270                         if (over)
5271                                 goto nopos;
5272                         di_len = btrfs_dir_name_len(leaf, di) +
5273                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
5274                         di_cur += di_len;
5275                         di = (struct btrfs_dir_item *)((char *)di + di_len);
5276                 }
5277 next:
5278                 path->slots[0]++;
5279         }
5280
5281         if (key_type == BTRFS_DIR_INDEX_KEY) {
5282                 if (is_curr)
5283                         filp->f_pos++;
5284                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
5285                                                       &ins_list);
5286                 if (ret)
5287                         goto nopos;
5288         }
5289
5290         /* Reached end of directory/root. Bump pos past the last item. */
5291         if (key_type == BTRFS_DIR_INDEX_KEY)
5292                 /*
5293                  * 32-bit glibc will use getdents64, but then strtol -
5294                  * so the last number we can serve is this.
5295                  */
5296                 filp->f_pos = 0x7fffffff;
5297         else
5298                 filp->f_pos++;
5299 nopos:
5300         ret = 0;
5301 err:
5302         if (key_type == BTRFS_DIR_INDEX_KEY)
5303                 btrfs_put_delayed_items(&ins_list, &del_list);
5304         btrfs_free_path(path);
5305         return ret;
5306 }
5307
5308 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5309 {
5310         struct btrfs_root *root = BTRFS_I(inode)->root;
5311         struct btrfs_trans_handle *trans;
5312         int ret = 0;
5313         bool nolock = false;
5314
5315         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5316                 return 0;
5317
5318         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5319                 nolock = true;
5320
5321         if (wbc->sync_mode == WB_SYNC_ALL) {
5322                 if (nolock)
5323                         trans = btrfs_join_transaction_nolock(root);
5324                 else
5325                         trans = btrfs_join_transaction(root);
5326                 if (IS_ERR(trans))
5327                         return PTR_ERR(trans);
5328                 ret = btrfs_commit_transaction(trans, root);
5329         }
5330         return ret;
5331 }
5332
5333 /*
5334  * This is somewhat expensive, updating the tree every time the
5335  * inode changes.  But, it is most likely to find the inode in cache.
5336  * FIXME, needs more benchmarking...there are no reasons other than performance
5337  * to keep or drop this code.
5338  */
5339 int btrfs_dirty_inode(struct inode *inode)
5340 {
5341         struct btrfs_root *root = BTRFS_I(inode)->root;
5342         struct btrfs_trans_handle *trans;
5343         int ret;
5344
5345         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5346                 return 0;
5347
5348         trans = btrfs_join_transaction(root);
5349         if (IS_ERR(trans))
5350                 return PTR_ERR(trans);
5351
5352         ret = btrfs_update_inode(trans, root, inode);
5353         if (ret && ret == -ENOSPC) {
5354                 /* whoops, lets try again with the full transaction */
5355                 btrfs_end_transaction(trans, root);
5356                 trans = btrfs_start_transaction(root, 1);
5357                 if (IS_ERR(trans))
5358                         return PTR_ERR(trans);
5359
5360                 ret = btrfs_update_inode(trans, root, inode);
5361         }
5362         btrfs_end_transaction(trans, root);
5363         if (BTRFS_I(inode)->delayed_node)
5364                 btrfs_balance_delayed_items(root);
5365
5366         return ret;
5367 }
5368
5369 /*
5370  * This is a copy of file_update_time.  We need this so we can return error on
5371  * ENOSPC for updating the inode in the case of file write and mmap writes.
5372  */
5373 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5374                              int flags)
5375 {
5376         struct btrfs_root *root = BTRFS_I(inode)->root;
5377
5378         if (btrfs_root_readonly(root))
5379                 return -EROFS;
5380
5381         if (flags & S_VERSION)
5382                 inode_inc_iversion(inode);
5383         if (flags & S_CTIME)
5384                 inode->i_ctime = *now;
5385         if (flags & S_MTIME)
5386                 inode->i_mtime = *now;
5387         if (flags & S_ATIME)
5388                 inode->i_atime = *now;
5389         return btrfs_dirty_inode(inode);
5390 }
5391
5392 /*
5393  * find the highest existing sequence number in a directory
5394  * and then set the in-memory index_cnt variable to reflect
5395  * free sequence numbers
5396  */
5397 static int btrfs_set_inode_index_count(struct inode *inode)
5398 {
5399         struct btrfs_root *root = BTRFS_I(inode)->root;
5400         struct btrfs_key key, found_key;
5401         struct btrfs_path *path;
5402         struct extent_buffer *leaf;
5403         int ret;
5404
5405         key.objectid = btrfs_ino(inode);
5406         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5407         key.offset = (u64)-1;
5408
5409         path = btrfs_alloc_path();
5410         if (!path)
5411                 return -ENOMEM;
5412
5413         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5414         if (ret < 0)
5415                 goto out;
5416         /* FIXME: we should be able to handle this */
5417         if (ret == 0)
5418                 goto out;
5419         ret = 0;
5420
5421         /*
5422          * MAGIC NUMBER EXPLANATION:
5423          * since we search a directory based on f_pos we have to start at 2
5424          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5425          * else has to start at 2
5426          */
5427         if (path->slots[0] == 0) {
5428                 BTRFS_I(inode)->index_cnt = 2;
5429                 goto out;
5430         }
5431
5432         path->slots[0]--;
5433
5434         leaf = path->nodes[0];
5435         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5436
5437         if (found_key.objectid != btrfs_ino(inode) ||
5438             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5439                 BTRFS_I(inode)->index_cnt = 2;
5440                 goto out;
5441         }
5442
5443         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5444 out:
5445         btrfs_free_path(path);
5446         return ret;
5447 }
5448
5449 /*
5450  * helper to find a free sequence number in a given directory.  This current
5451  * code is very simple, later versions will do smarter things in the btree
5452  */
5453 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5454 {
5455         int ret = 0;
5456
5457         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5458                 ret = btrfs_inode_delayed_dir_index_count(dir);
5459                 if (ret) {
5460                         ret = btrfs_set_inode_index_count(dir);
5461                         if (ret)
5462                                 return ret;
5463                 }
5464         }
5465
5466         *index = BTRFS_I(dir)->index_cnt;
5467         BTRFS_I(dir)->index_cnt++;
5468
5469         return ret;
5470 }
5471
5472 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5473                                      struct btrfs_root *root,
5474                                      struct inode *dir,
5475                                      const char *name, int name_len,
5476                                      u64 ref_objectid, u64 objectid,
5477                                      umode_t mode, u64 *index)
5478 {
5479         struct inode *inode;
5480         struct btrfs_inode_item *inode_item;
5481         struct btrfs_key *location;
5482         struct btrfs_path *path;
5483         struct btrfs_inode_ref *ref;
5484         struct btrfs_key key[2];
5485         u32 sizes[2];
5486         unsigned long ptr;
5487         int ret;
5488         int owner;
5489
5490         path = btrfs_alloc_path();
5491         if (!path)
5492                 return ERR_PTR(-ENOMEM);
5493
5494         inode = new_inode(root->fs_info->sb);
5495         if (!inode) {
5496                 btrfs_free_path(path);
5497                 return ERR_PTR(-ENOMEM);
5498         }
5499
5500         /*
5501          * we have to initialize this early, so we can reclaim the inode
5502          * number if we fail afterwards in this function.
5503          */
5504         inode->i_ino = objectid;
5505
5506         if (dir) {
5507                 trace_btrfs_inode_request(dir);
5508
5509                 ret = btrfs_set_inode_index(dir, index);
5510                 if (ret) {
5511                         btrfs_free_path(path);
5512                         iput(inode);
5513                         return ERR_PTR(ret);
5514                 }
5515         }
5516         /*
5517          * index_cnt is ignored for everything but a dir,
5518          * btrfs_get_inode_index_count has an explanation for the magic
5519          * number
5520          */
5521         BTRFS_I(inode)->index_cnt = 2;
5522         BTRFS_I(inode)->root = root;
5523         BTRFS_I(inode)->generation = trans->transid;
5524         inode->i_generation = BTRFS_I(inode)->generation;
5525
5526         /*
5527          * We could have gotten an inode number from somebody who was fsynced
5528          * and then removed in this same transaction, so let's just set full
5529          * sync since it will be a full sync anyway and this will blow away the
5530          * old info in the log.
5531          */
5532         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5533
5534         if (S_ISDIR(mode))
5535                 owner = 0;
5536         else
5537                 owner = 1;
5538
5539         key[0].objectid = objectid;
5540         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5541         key[0].offset = 0;
5542
5543         /*
5544          * Start new inodes with an inode_ref. This is slightly more
5545          * efficient for small numbers of hard links since they will
5546          * be packed into one item. Extended refs will kick in if we
5547          * add more hard links than can fit in the ref item.
5548          */
5549         key[1].objectid = objectid;
5550         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5551         key[1].offset = ref_objectid;
5552
5553         sizes[0] = sizeof(struct btrfs_inode_item);
5554         sizes[1] = name_len + sizeof(*ref);
5555
5556         path->leave_spinning = 1;
5557         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5558         if (ret != 0)
5559                 goto fail;
5560
5561         inode_init_owner(inode, dir, mode);
5562         inode_set_bytes(inode, 0);
5563         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5564         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5565                                   struct btrfs_inode_item);
5566         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5567                              sizeof(*inode_item));
5568         fill_inode_item(trans, path->nodes[0], inode_item, inode);
5569
5570         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5571                              struct btrfs_inode_ref);
5572         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5573         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5574         ptr = (unsigned long)(ref + 1);
5575         write_extent_buffer(path->nodes[0], name, ptr, name_len);
5576
5577         btrfs_mark_buffer_dirty(path->nodes[0]);
5578         btrfs_free_path(path);
5579
5580         location = &BTRFS_I(inode)->location;
5581         location->objectid = objectid;
5582         location->offset = 0;
5583         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5584
5585         btrfs_inherit_iflags(inode, dir);
5586
5587         if (S_ISREG(mode)) {
5588                 if (btrfs_test_opt(root, NODATASUM))
5589                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5590                 if (btrfs_test_opt(root, NODATACOW))
5591                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5592                                 BTRFS_INODE_NODATASUM;
5593         }
5594
5595         insert_inode_hash(inode);
5596         inode_tree_add(inode);
5597
5598         trace_btrfs_inode_new(inode);
5599         btrfs_set_inode_last_trans(trans, inode);
5600
5601         btrfs_update_root_times(trans, root);
5602
5603         return inode;
5604 fail:
5605         if (dir)
5606                 BTRFS_I(dir)->index_cnt--;
5607         btrfs_free_path(path);
5608         iput(inode);
5609         return ERR_PTR(ret);
5610 }
5611
5612 static inline u8 btrfs_inode_type(struct inode *inode)
5613 {
5614         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5615 }
5616
5617 /*
5618  * utility function to add 'inode' into 'parent_inode' with
5619  * a give name and a given sequence number.
5620  * if 'add_backref' is true, also insert a backref from the
5621  * inode to the parent directory.
5622  */
5623 int btrfs_add_link(struct btrfs_trans_handle *trans,
5624                    struct inode *parent_inode, struct inode *inode,
5625                    const char *name, int name_len, int add_backref, u64 index)
5626 {
5627         int ret = 0;
5628         struct btrfs_key key;
5629         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5630         u64 ino = btrfs_ino(inode);
5631         u64 parent_ino = btrfs_ino(parent_inode);
5632
5633         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5634                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5635         } else {
5636                 key.objectid = ino;
5637                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5638                 key.offset = 0;
5639         }
5640
5641         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5642                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5643                                          key.objectid, root->root_key.objectid,
5644                                          parent_ino, index, name, name_len);
5645         } else if (add_backref) {
5646                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5647                                              parent_ino, index);
5648         }
5649
5650         /* Nothing to clean up yet */
5651         if (ret)
5652                 return ret;
5653
5654         ret = btrfs_insert_dir_item(trans, root, name, name_len,
5655                                     parent_inode, &key,
5656                                     btrfs_inode_type(inode), index);
5657         if (ret == -EEXIST || ret == -EOVERFLOW)
5658                 goto fail_dir_item;
5659         else if (ret) {
5660                 btrfs_abort_transaction(trans, root, ret);
5661                 return ret;
5662         }
5663
5664         btrfs_i_size_write(parent_inode, parent_inode->i_size +
5665                            name_len * 2);
5666         inode_inc_iversion(parent_inode);
5667         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5668         ret = btrfs_update_inode(trans, root, parent_inode);
5669         if (ret)
5670                 btrfs_abort_transaction(trans, root, ret);
5671         return ret;
5672
5673 fail_dir_item:
5674         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5675                 u64 local_index;
5676                 int err;
5677                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5678                                  key.objectid, root->root_key.objectid,
5679                                  parent_ino, &local_index, name, name_len);
5680
5681         } else if (add_backref) {
5682                 u64 local_index;
5683                 int err;
5684
5685                 err = btrfs_del_inode_ref(trans, root, name, name_len,
5686                                           ino, parent_ino, &local_index);
5687         }
5688         return ret;
5689 }
5690
5691 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5692                             struct inode *dir, struct dentry *dentry,
5693                             struct inode *inode, int backref, u64 index)
5694 {
5695         int err = btrfs_add_link(trans, dir, inode,
5696                                  dentry->d_name.name, dentry->d_name.len,
5697                                  backref, index);
5698         if (err > 0)
5699                 err = -EEXIST;
5700         return err;
5701 }
5702
5703 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5704                         umode_t mode, dev_t rdev)
5705 {
5706         struct btrfs_trans_handle *trans;
5707         struct btrfs_root *root = BTRFS_I(dir)->root;
5708         struct inode *inode = NULL;
5709         int err;
5710         int drop_inode = 0;
5711         u64 objectid;
5712         u64 index = 0;
5713
5714         if (!new_valid_dev(rdev))
5715                 return -EINVAL;
5716
5717         /*
5718          * 2 for inode item and ref
5719          * 2 for dir items
5720          * 1 for xattr if selinux is on
5721          */
5722         trans = btrfs_start_transaction(root, 5);
5723         if (IS_ERR(trans))
5724                 return PTR_ERR(trans);
5725
5726         err = btrfs_find_free_ino(root, &objectid);
5727         if (err)
5728                 goto out_unlock;
5729
5730         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5731                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5732                                 mode, &index);
5733         if (IS_ERR(inode)) {
5734                 err = PTR_ERR(inode);
5735                 goto out_unlock;
5736         }
5737
5738         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5739         if (err) {
5740                 drop_inode = 1;
5741                 goto out_unlock;
5742         }
5743
5744         /*
5745         * If the active LSM wants to access the inode during
5746         * d_instantiate it needs these. Smack checks to see
5747         * if the filesystem supports xattrs by looking at the
5748         * ops vector.
5749         */
5750
5751         inode->i_op = &btrfs_special_inode_operations;
5752         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5753         if (err)
5754                 drop_inode = 1;
5755         else {
5756                 init_special_inode(inode, inode->i_mode, rdev);
5757                 btrfs_update_inode(trans, root, inode);
5758                 d_instantiate(dentry, inode);
5759         }
5760 out_unlock:
5761         btrfs_end_transaction(trans, root);
5762         btrfs_btree_balance_dirty(root);
5763         if (drop_inode) {
5764                 inode_dec_link_count(inode);
5765                 iput(inode);
5766         }
5767         return err;
5768 }
5769
5770 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5771                         umode_t mode, bool excl)
5772 {
5773         struct btrfs_trans_handle *trans;
5774         struct btrfs_root *root = BTRFS_I(dir)->root;
5775         struct inode *inode = NULL;
5776         int drop_inode_on_err = 0;
5777         int err;
5778         u64 objectid;
5779         u64 index = 0;
5780
5781         /*
5782          * 2 for inode item and ref
5783          * 2 for dir items
5784          * 1 for xattr if selinux is on
5785          */
5786         trans = btrfs_start_transaction(root, 5);
5787         if (IS_ERR(trans))
5788                 return PTR_ERR(trans);
5789
5790         err = btrfs_find_free_ino(root, &objectid);
5791         if (err)
5792                 goto out_unlock;
5793
5794         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5795                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5796                                 mode, &index);
5797         if (IS_ERR(inode)) {
5798                 err = PTR_ERR(inode);
5799                 goto out_unlock;
5800         }
5801         drop_inode_on_err = 1;
5802
5803         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5804         if (err)
5805                 goto out_unlock;
5806
5807         err = btrfs_update_inode(trans, root, inode);
5808         if (err)
5809                 goto out_unlock;
5810
5811         /*
5812         * If the active LSM wants to access the inode during
5813         * d_instantiate it needs these. Smack checks to see
5814         * if the filesystem supports xattrs by looking at the
5815         * ops vector.
5816         */
5817         inode->i_fop = &btrfs_file_operations;
5818         inode->i_op = &btrfs_file_inode_operations;
5819
5820         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5821         if (err)
5822                 goto out_unlock;
5823
5824         inode->i_mapping->a_ops = &btrfs_aops;
5825         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5826         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5827         d_instantiate(dentry, inode);
5828
5829 out_unlock:
5830         btrfs_end_transaction(trans, root);
5831         if (err && drop_inode_on_err) {
5832                 inode_dec_link_count(inode);
5833                 iput(inode);
5834         }
5835         btrfs_btree_balance_dirty(root);
5836         return err;
5837 }
5838
5839 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5840                       struct dentry *dentry)
5841 {
5842         struct btrfs_trans_handle *trans;
5843         struct btrfs_root *root = BTRFS_I(dir)->root;
5844         struct inode *inode = old_dentry->d_inode;
5845         u64 index;
5846         int err;
5847         int drop_inode = 0;
5848
5849         /* do not allow sys_link's with other subvols of the same device */
5850         if (root->objectid != BTRFS_I(inode)->root->objectid)
5851                 return -EXDEV;
5852
5853         if (inode->i_nlink >= BTRFS_LINK_MAX)
5854                 return -EMLINK;
5855
5856         err = btrfs_set_inode_index(dir, &index);
5857         if (err)
5858                 goto fail;
5859
5860         /*
5861          * 2 items for inode and inode ref
5862          * 2 items for dir items
5863          * 1 item for parent inode
5864          */
5865         trans = btrfs_start_transaction(root, 5);
5866         if (IS_ERR(trans)) {
5867                 err = PTR_ERR(trans);
5868                 goto fail;
5869         }
5870
5871         btrfs_inc_nlink(inode);
5872         inode_inc_iversion(inode);
5873         inode->i_ctime = CURRENT_TIME;
5874         ihold(inode);
5875         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5876
5877         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5878
5879         if (err) {
5880                 drop_inode = 1;
5881         } else {
5882                 struct dentry *parent = dentry->d_parent;
5883                 err = btrfs_update_inode(trans, root, inode);
5884                 if (err)
5885                         goto fail;
5886                 d_instantiate(dentry, inode);
5887                 btrfs_log_new_name(trans, inode, NULL, parent);
5888         }
5889
5890         btrfs_end_transaction(trans, root);
5891 fail:
5892         if (drop_inode) {
5893                 inode_dec_link_count(inode);
5894                 iput(inode);
5895         }
5896         btrfs_btree_balance_dirty(root);
5897         return err;
5898 }
5899
5900 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5901 {
5902         struct inode *inode = NULL;
5903         struct btrfs_trans_handle *trans;
5904         struct btrfs_root *root = BTRFS_I(dir)->root;
5905         int err = 0;
5906         int drop_on_err = 0;
5907         u64 objectid = 0;
5908         u64 index = 0;
5909
5910         /*
5911          * 2 items for inode and ref
5912          * 2 items for dir items
5913          * 1 for xattr if selinux is on
5914          */
5915         trans = btrfs_start_transaction(root, 5);
5916         if (IS_ERR(trans))
5917                 return PTR_ERR(trans);
5918
5919         err = btrfs_find_free_ino(root, &objectid);
5920         if (err)
5921                 goto out_fail;
5922
5923         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5924                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5925                                 S_IFDIR | mode, &index);
5926         if (IS_ERR(inode)) {
5927                 err = PTR_ERR(inode);
5928                 goto out_fail;
5929         }
5930
5931         drop_on_err = 1;
5932
5933         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5934         if (err)
5935                 goto out_fail;
5936
5937         inode->i_op = &btrfs_dir_inode_operations;
5938         inode->i_fop = &btrfs_dir_file_operations;
5939
5940         btrfs_i_size_write(inode, 0);
5941         err = btrfs_update_inode(trans, root, inode);
5942         if (err)
5943                 goto out_fail;
5944
5945         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5946                              dentry->d_name.len, 0, index);
5947         if (err)
5948                 goto out_fail;
5949
5950         d_instantiate(dentry, inode);
5951         drop_on_err = 0;
5952
5953 out_fail:
5954         btrfs_end_transaction(trans, root);
5955         if (drop_on_err)
5956                 iput(inode);
5957         btrfs_btree_balance_dirty(root);
5958         return err;
5959 }
5960
5961 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5962  * and an extent that you want to insert, deal with overlap and insert
5963  * the new extent into the tree.
5964  */
5965 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5966                                 struct extent_map *existing,
5967                                 struct extent_map *em,
5968                                 u64 map_start, u64 map_len)
5969 {
5970         u64 start_diff;
5971
5972         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5973         start_diff = map_start - em->start;
5974         em->start = map_start;
5975         em->len = map_len;
5976         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5977             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5978                 em->block_start += start_diff;
5979                 em->block_len -= start_diff;
5980         }
5981         return add_extent_mapping(em_tree, em);
5982 }
5983
5984 static noinline int uncompress_inline(struct btrfs_path *path,
5985                                       struct inode *inode, struct page *page,
5986                                       size_t pg_offset, u64 extent_offset,
5987                                       struct btrfs_file_extent_item *item)
5988 {
5989         int ret;
5990         struct extent_buffer *leaf = path->nodes[0];
5991         char *tmp;
5992         size_t max_size;
5993         unsigned long inline_size;
5994         unsigned long ptr;
5995         int compress_type;
5996
5997         WARN_ON(pg_offset != 0);
5998         compress_type = btrfs_file_extent_compression(leaf, item);
5999         max_size = btrfs_file_extent_ram_bytes(leaf, item);
6000         inline_size = btrfs_file_extent_inline_item_len(leaf,
6001                                         btrfs_item_nr(leaf, path->slots[0]));
6002         tmp = kmalloc(inline_size, GFP_NOFS);
6003         if (!tmp)
6004                 return -ENOMEM;
6005         ptr = btrfs_file_extent_inline_start(item);
6006
6007         read_extent_buffer(leaf, tmp, ptr, inline_size);
6008
6009         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6010         ret = btrfs_decompress(compress_type, tmp, page,
6011                                extent_offset, inline_size, max_size);
6012         if (ret) {
6013                 char *kaddr = kmap_atomic(page);
6014                 unsigned long copy_size = min_t(u64,
6015                                   PAGE_CACHE_SIZE - pg_offset,
6016                                   max_size - extent_offset);
6017                 memset(kaddr + pg_offset, 0, copy_size);
6018                 kunmap_atomic(kaddr);
6019         }
6020         kfree(tmp);
6021         return 0;
6022 }
6023
6024 /*
6025  * a bit scary, this does extent mapping from logical file offset to the disk.
6026  * the ugly parts come from merging extents from the disk with the in-ram
6027  * representation.  This gets more complex because of the data=ordered code,
6028  * where the in-ram extents might be locked pending data=ordered completion.
6029  *
6030  * This also copies inline extents directly into the page.
6031  */
6032
6033 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6034                                     size_t pg_offset, u64 start, u64 len,
6035                                     int create)
6036 {
6037         int ret;
6038         int err = 0;
6039         u64 bytenr;
6040         u64 extent_start = 0;
6041         u64 extent_end = 0;
6042         u64 objectid = btrfs_ino(inode);
6043         u32 found_type;
6044         struct btrfs_path *path = NULL;
6045         struct btrfs_root *root = BTRFS_I(inode)->root;
6046         struct btrfs_file_extent_item *item;
6047         struct extent_buffer *leaf;
6048         struct btrfs_key found_key;
6049         struct extent_map *em = NULL;
6050         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6051         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6052         struct btrfs_trans_handle *trans = NULL;
6053         int compress_type;
6054
6055 again:
6056         read_lock(&em_tree->lock);
6057         em = lookup_extent_mapping(em_tree, start, len);
6058         if (em)
6059                 em->bdev = root->fs_info->fs_devices->latest_bdev;
6060         read_unlock(&em_tree->lock);
6061
6062         if (em) {
6063                 if (em->start > start || em->start + em->len <= start)
6064                         free_extent_map(em);
6065                 else if (em->block_start == EXTENT_MAP_INLINE && page)
6066                         free_extent_map(em);
6067                 else
6068                         goto out;
6069         }
6070         em = alloc_extent_map();
6071         if (!em) {
6072                 err = -ENOMEM;
6073                 goto out;
6074         }
6075         em->bdev = root->fs_info->fs_devices->latest_bdev;
6076         em->start = EXTENT_MAP_HOLE;
6077         em->orig_start = EXTENT_MAP_HOLE;
6078         em->len = (u64)-1;
6079         em->block_len = (u64)-1;
6080
6081         if (!path) {
6082                 path = btrfs_alloc_path();
6083                 if (!path) {
6084                         err = -ENOMEM;
6085                         goto out;
6086                 }
6087                 /*
6088                  * Chances are we'll be called again, so go ahead and do
6089                  * readahead
6090                  */
6091                 path->reada = 1;
6092         }
6093
6094         ret = btrfs_lookup_file_extent(trans, root, path,
6095                                        objectid, start, trans != NULL);
6096         if (ret < 0) {
6097                 err = ret;
6098                 goto out;
6099         }
6100
6101         if (ret != 0) {
6102                 if (path->slots[0] == 0)
6103                         goto not_found;
6104                 path->slots[0]--;
6105         }
6106
6107         leaf = path->nodes[0];
6108         item = btrfs_item_ptr(leaf, path->slots[0],
6109                               struct btrfs_file_extent_item);
6110         /* are we inside the extent that was found? */
6111         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6112         found_type = btrfs_key_type(&found_key);
6113         if (found_key.objectid != objectid ||
6114             found_type != BTRFS_EXTENT_DATA_KEY) {
6115                 goto not_found;
6116         }
6117
6118         found_type = btrfs_file_extent_type(leaf, item);
6119         extent_start = found_key.offset;
6120         compress_type = btrfs_file_extent_compression(leaf, item);
6121         if (found_type == BTRFS_FILE_EXTENT_REG ||
6122             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6123                 extent_end = extent_start +
6124                        btrfs_file_extent_num_bytes(leaf, item);
6125         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6126                 size_t size;
6127                 size = btrfs_file_extent_inline_len(leaf, item);
6128                 extent_end = ALIGN(extent_start + size, root->sectorsize);
6129         }
6130
6131         if (start >= extent_end) {
6132                 path->slots[0]++;
6133                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6134                         ret = btrfs_next_leaf(root, path);
6135                         if (ret < 0) {
6136                                 err = ret;
6137                                 goto out;
6138                         }
6139                         if (ret > 0)
6140                                 goto not_found;
6141                         leaf = path->nodes[0];
6142                 }
6143                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6144                 if (found_key.objectid != objectid ||
6145                     found_key.type != BTRFS_EXTENT_DATA_KEY)
6146                         goto not_found;
6147                 if (start + len <= found_key.offset)
6148                         goto not_found;
6149                 em->start = start;
6150                 em->orig_start = start;
6151                 em->len = found_key.offset - start;
6152                 goto not_found_em;
6153         }
6154
6155         if (found_type == BTRFS_FILE_EXTENT_REG ||
6156             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6157                 em->start = extent_start;
6158                 em->len = extent_end - extent_start;
6159                 em->orig_start = extent_start -
6160                                  btrfs_file_extent_offset(leaf, item);
6161                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6162                                                                       item);
6163                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6164                 if (bytenr == 0) {
6165                         em->block_start = EXTENT_MAP_HOLE;
6166                         goto insert;
6167                 }
6168                 if (compress_type != BTRFS_COMPRESS_NONE) {
6169                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6170                         em->compress_type = compress_type;
6171                         em->block_start = bytenr;
6172                         em->block_len = em->orig_block_len;
6173                 } else {
6174                         bytenr += btrfs_file_extent_offset(leaf, item);
6175                         em->block_start = bytenr;
6176                         em->block_len = em->len;
6177                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6178                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6179                 }
6180                 goto insert;
6181         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6182                 unsigned long ptr;
6183                 char *map;
6184                 size_t size;
6185                 size_t extent_offset;
6186                 size_t copy_size;
6187
6188                 em->block_start = EXTENT_MAP_INLINE;
6189                 if (!page || create) {
6190                         em->start = extent_start;
6191                         em->len = extent_end - extent_start;
6192                         goto out;
6193                 }
6194
6195                 size = btrfs_file_extent_inline_len(leaf, item);
6196                 extent_offset = page_offset(page) + pg_offset - extent_start;
6197                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6198                                 size - extent_offset);
6199                 em->start = extent_start + extent_offset;
6200                 em->len = ALIGN(copy_size, root->sectorsize);
6201                 em->orig_block_len = em->len;
6202                 em->orig_start = em->start;
6203                 if (compress_type) {
6204                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6205                         em->compress_type = compress_type;
6206                 }
6207                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6208                 if (create == 0 && !PageUptodate(page)) {
6209                         if (btrfs_file_extent_compression(leaf, item) !=
6210                             BTRFS_COMPRESS_NONE) {
6211                                 ret = uncompress_inline(path, inode, page,
6212                                                         pg_offset,
6213                                                         extent_offset, item);
6214                                 BUG_ON(ret); /* -ENOMEM */
6215                         } else {
6216                                 map = kmap(page);
6217                                 read_extent_buffer(leaf, map + pg_offset, ptr,
6218                                                    copy_size);
6219                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6220                                         memset(map + pg_offset + copy_size, 0,
6221                                                PAGE_CACHE_SIZE - pg_offset -
6222                                                copy_size);
6223                                 }
6224                                 kunmap(page);
6225                         }
6226                         flush_dcache_page(page);
6227                 } else if (create && PageUptodate(page)) {
6228                         BUG();
6229                         if (!trans) {
6230                                 kunmap(page);
6231                                 free_extent_map(em);
6232                                 em = NULL;
6233
6234                                 btrfs_release_path(path);
6235                                 trans = btrfs_join_transaction(root);
6236
6237                                 if (IS_ERR(trans))
6238                                         return ERR_CAST(trans);
6239                                 goto again;
6240                         }
6241                         map = kmap(page);
6242                         write_extent_buffer(leaf, map + pg_offset, ptr,
6243                                             copy_size);
6244                         kunmap(page);
6245                         btrfs_mark_buffer_dirty(leaf);
6246                 }
6247                 set_extent_uptodate(io_tree, em->start,
6248                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
6249                 goto insert;
6250         } else {
6251                 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6252         }
6253 not_found:
6254         em->start = start;
6255         em->orig_start = start;
6256         em->len = len;
6257 not_found_em:
6258         em->block_start = EXTENT_MAP_HOLE;
6259         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6260 insert:
6261         btrfs_release_path(path);
6262         if (em->start > start || extent_map_end(em) <= start) {
6263                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
6264                        "[%llu %llu]\n", (unsigned long long)em->start,
6265                        (unsigned long long)em->len,
6266                        (unsigned long long)start,
6267                        (unsigned long long)len);
6268                 err = -EIO;
6269                 goto out;
6270         }
6271
6272         err = 0;
6273         write_lock(&em_tree->lock);
6274         ret = add_extent_mapping(em_tree, em);
6275         /* it is possible that someone inserted the extent into the tree
6276          * while we had the lock dropped.  It is also possible that
6277          * an overlapping map exists in the tree
6278          */
6279         if (ret == -EEXIST) {
6280                 struct extent_map *existing;
6281
6282                 ret = 0;
6283
6284                 existing = lookup_extent_mapping(em_tree, start, len);
6285                 if (existing && (existing->start > start ||
6286                     existing->start + existing->len <= start)) {
6287                         free_extent_map(existing);
6288                         existing = NULL;
6289                 }
6290                 if (!existing) {
6291                         existing = lookup_extent_mapping(em_tree, em->start,
6292                                                          em->len);
6293                         if (existing) {
6294                                 err = merge_extent_mapping(em_tree, existing,
6295                                                            em, start,
6296                                                            root->sectorsize);
6297                                 free_extent_map(existing);
6298                                 if (err) {
6299                                         free_extent_map(em);
6300                                         em = NULL;
6301                                 }
6302                         } else {
6303                                 err = -EIO;
6304                                 free_extent_map(em);
6305                                 em = NULL;
6306                         }
6307                 } else {
6308                         free_extent_map(em);
6309                         em = existing;
6310                         err = 0;
6311                 }
6312         }
6313         write_unlock(&em_tree->lock);
6314 out:
6315
6316         if (em)
6317                 trace_btrfs_get_extent(root, em);
6318
6319         if (path)
6320                 btrfs_free_path(path);
6321         if (trans) {
6322                 ret = btrfs_end_transaction(trans, root);
6323                 if (!err)
6324                         err = ret;
6325         }
6326         if (err) {
6327                 free_extent_map(em);
6328                 return ERR_PTR(err);
6329         }
6330         BUG_ON(!em); /* Error is always set */
6331         return em;
6332 }
6333
6334 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6335                                            size_t pg_offset, u64 start, u64 len,
6336                                            int create)
6337 {
6338         struct extent_map *em;
6339         struct extent_map *hole_em = NULL;
6340         u64 range_start = start;
6341         u64 end;
6342         u64 found;
6343         u64 found_end;
6344         int err = 0;
6345
6346         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6347         if (IS_ERR(em))
6348                 return em;
6349         if (em) {
6350                 /*
6351                  * if our em maps to
6352                  * -  a hole or
6353                  * -  a pre-alloc extent,
6354                  * there might actually be delalloc bytes behind it.
6355                  */
6356                 if (em->block_start != EXTENT_MAP_HOLE &&
6357                     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6358                         return em;
6359                 else
6360                         hole_em = em;
6361         }
6362
6363         /* check to see if we've wrapped (len == -1 or similar) */
6364         end = start + len;
6365         if (end < start)
6366                 end = (u64)-1;
6367         else
6368                 end -= 1;
6369
6370         em = NULL;
6371
6372         /* ok, we didn't find anything, lets look for delalloc */
6373         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6374                                  end, len, EXTENT_DELALLOC, 1);
6375         found_end = range_start + found;
6376         if (found_end < range_start)
6377                 found_end = (u64)-1;
6378
6379         /*
6380          * we didn't find anything useful, return
6381          * the original results from get_extent()
6382          */
6383         if (range_start > end || found_end <= start) {
6384                 em = hole_em;
6385                 hole_em = NULL;
6386                 goto out;
6387         }
6388
6389         /* adjust the range_start to make sure it doesn't
6390          * go backwards from the start they passed in
6391          */
6392         range_start = max(start,range_start);
6393         found = found_end - range_start;
6394
6395         if (found > 0) {
6396                 u64 hole_start = start;
6397                 u64 hole_len = len;
6398
6399                 em = alloc_extent_map();
6400                 if (!em) {
6401                         err = -ENOMEM;
6402                         goto out;
6403                 }
6404                 /*
6405                  * when btrfs_get_extent can't find anything it
6406                  * returns one huge hole
6407                  *
6408                  * make sure what it found really fits our range, and
6409                  * adjust to make sure it is based on the start from
6410                  * the caller
6411                  */
6412                 if (hole_em) {
6413                         u64 calc_end = extent_map_end(hole_em);
6414
6415                         if (calc_end <= start || (hole_em->start > end)) {
6416                                 free_extent_map(hole_em);
6417                                 hole_em = NULL;
6418                         } else {
6419                                 hole_start = max(hole_em->start, start);
6420                                 hole_len = calc_end - hole_start;
6421                         }
6422                 }
6423                 em->bdev = NULL;
6424                 if (hole_em && range_start > hole_start) {
6425                         /* our hole starts before our delalloc, so we
6426                          * have to return just the parts of the hole
6427                          * that go until  the delalloc starts
6428                          */
6429                         em->len = min(hole_len,
6430                                       range_start - hole_start);
6431                         em->start = hole_start;
6432                         em->orig_start = hole_start;
6433                         /*
6434                          * don't adjust block start at all,
6435                          * it is fixed at EXTENT_MAP_HOLE
6436                          */
6437                         em->block_start = hole_em->block_start;
6438                         em->block_len = hole_len;
6439                         if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6440                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6441                 } else {
6442                         em->start = range_start;
6443                         em->len = found;
6444                         em->orig_start = range_start;
6445                         em->block_start = EXTENT_MAP_DELALLOC;
6446                         em->block_len = found;
6447                 }
6448         } else if (hole_em) {
6449                 return hole_em;
6450         }
6451 out:
6452
6453         free_extent_map(hole_em);
6454         if (err) {
6455                 free_extent_map(em);
6456                 return ERR_PTR(err);
6457         }
6458         return em;
6459 }
6460
6461 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6462                                                   u64 start, u64 len)
6463 {
6464         struct btrfs_root *root = BTRFS_I(inode)->root;
6465         struct btrfs_trans_handle *trans;
6466         struct extent_map *em;
6467         struct btrfs_key ins;
6468         u64 alloc_hint;
6469         int ret;
6470
6471         trans = btrfs_join_transaction(root);
6472         if (IS_ERR(trans))
6473                 return ERR_CAST(trans);
6474
6475         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
6476
6477         alloc_hint = get_extent_allocation_hint(inode, start, len);
6478         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
6479                                    alloc_hint, &ins, 1);
6480         if (ret) {
6481                 em = ERR_PTR(ret);
6482                 goto out;
6483         }
6484
6485         em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6486                               ins.offset, ins.offset, 0);
6487         if (IS_ERR(em))
6488                 goto out;
6489
6490         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6491                                            ins.offset, ins.offset, 0);
6492         if (ret) {
6493                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6494                 em = ERR_PTR(ret);
6495         }
6496 out:
6497         btrfs_end_transaction(trans, root);
6498         return em;
6499 }
6500
6501 /*
6502  * returns 1 when the nocow is safe, < 1 on error, 0 if the
6503  * block must be cow'd
6504  */
6505 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
6506                                       struct inode *inode, u64 offset, u64 len)
6507 {
6508         struct btrfs_path *path;
6509         int ret;
6510         struct extent_buffer *leaf;
6511         struct btrfs_root *root = BTRFS_I(inode)->root;
6512         struct btrfs_file_extent_item *fi;
6513         struct btrfs_key key;
6514         u64 disk_bytenr;
6515         u64 backref_offset;
6516         u64 extent_end;
6517         u64 num_bytes;
6518         int slot;
6519         int found_type;
6520
6521         path = btrfs_alloc_path();
6522         if (!path)
6523                 return -ENOMEM;
6524
6525         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
6526                                        offset, 0);
6527         if (ret < 0)
6528                 goto out;
6529
6530         slot = path->slots[0];
6531         if (ret == 1) {
6532                 if (slot == 0) {
6533                         /* can't find the item, must cow */
6534                         ret = 0;
6535                         goto out;
6536                 }
6537                 slot--;
6538         }
6539         ret = 0;
6540         leaf = path->nodes[0];
6541         btrfs_item_key_to_cpu(leaf, &key, slot);
6542         if (key.objectid != btrfs_ino(inode) ||
6543             key.type != BTRFS_EXTENT_DATA_KEY) {
6544                 /* not our file or wrong item type, must cow */
6545                 goto out;
6546         }
6547
6548         if (key.offset > offset) {
6549                 /* Wrong offset, must cow */
6550                 goto out;
6551         }
6552
6553         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6554         found_type = btrfs_file_extent_type(leaf, fi);
6555         if (found_type != BTRFS_FILE_EXTENT_REG &&
6556             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6557                 /* not a regular extent, must cow */
6558                 goto out;
6559         }
6560         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6561         backref_offset = btrfs_file_extent_offset(leaf, fi);
6562
6563         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6564         if (extent_end < offset + len) {
6565                 /* extent doesn't include our full range, must cow */
6566                 goto out;
6567         }
6568
6569         if (btrfs_extent_readonly(root, disk_bytenr))
6570                 goto out;
6571
6572         /*
6573          * look for other files referencing this extent, if we
6574          * find any we must cow
6575          */
6576         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6577                                   key.offset - backref_offset, disk_bytenr))
6578                 goto out;
6579
6580         /*
6581          * adjust disk_bytenr and num_bytes to cover just the bytes
6582          * in this extent we are about to write.  If there
6583          * are any csums in that range we have to cow in order
6584          * to keep the csums correct
6585          */
6586         disk_bytenr += backref_offset;
6587         disk_bytenr += offset - key.offset;
6588         num_bytes = min(offset + len, extent_end) - offset;
6589         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6590                                 goto out;
6591         /*
6592          * all of the above have passed, it is safe to overwrite this extent
6593          * without cow
6594          */
6595         ret = 1;
6596 out:
6597         btrfs_free_path(path);
6598         return ret;
6599 }
6600
6601 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6602                               struct extent_state **cached_state, int writing)
6603 {
6604         struct btrfs_ordered_extent *ordered;
6605         int ret = 0;
6606
6607         while (1) {
6608                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6609                                  0, cached_state);
6610                 /*
6611                  * We're concerned with the entire range that we're going to be
6612                  * doing DIO to, so we need to make sure theres no ordered
6613                  * extents in this range.
6614                  */
6615                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6616                                                      lockend - lockstart + 1);
6617
6618                 /*
6619                  * We need to make sure there are no buffered pages in this
6620                  * range either, we could have raced between the invalidate in
6621                  * generic_file_direct_write and locking the extent.  The
6622                  * invalidate needs to happen so that reads after a write do not
6623                  * get stale data.
6624                  */
6625                 if (!ordered && (!writing ||
6626                     !test_range_bit(&BTRFS_I(inode)->io_tree,
6627                                     lockstart, lockend, EXTENT_UPTODATE, 0,
6628                                     *cached_state)))
6629                         break;
6630
6631                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6632                                      cached_state, GFP_NOFS);
6633
6634                 if (ordered) {
6635                         btrfs_start_ordered_extent(inode, ordered, 1);
6636                         btrfs_put_ordered_extent(ordered);
6637                 } else {
6638                         /* Screw you mmap */
6639                         ret = filemap_write_and_wait_range(inode->i_mapping,
6640                                                            lockstart,
6641                                                            lockend);
6642                         if (ret)
6643                                 break;
6644
6645                         /*
6646                          * If we found a page that couldn't be invalidated just
6647                          * fall back to buffered.
6648                          */
6649                         ret = invalidate_inode_pages2_range(inode->i_mapping,
6650                                         lockstart >> PAGE_CACHE_SHIFT,
6651                                         lockend >> PAGE_CACHE_SHIFT);
6652                         if (ret)
6653                                 break;
6654                 }
6655
6656                 cond_resched();
6657         }
6658
6659         return ret;
6660 }
6661
6662 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6663                                            u64 len, u64 orig_start,
6664                                            u64 block_start, u64 block_len,
6665                                            u64 orig_block_len, int type)
6666 {
6667         struct extent_map_tree *em_tree;
6668         struct extent_map *em;
6669         struct btrfs_root *root = BTRFS_I(inode)->root;
6670         int ret;
6671
6672         em_tree = &BTRFS_I(inode)->extent_tree;
6673         em = alloc_extent_map();
6674         if (!em)
6675                 return ERR_PTR(-ENOMEM);
6676
6677         em->start = start;
6678         em->orig_start = orig_start;
6679         em->mod_start = start;
6680         em->mod_len = len;
6681         em->len = len;
6682         em->block_len = block_len;
6683         em->block_start = block_start;
6684         em->bdev = root->fs_info->fs_devices->latest_bdev;
6685         em->orig_block_len = orig_block_len;
6686         em->generation = -1;
6687         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6688         if (type == BTRFS_ORDERED_PREALLOC)
6689                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6690
6691         do {
6692                 btrfs_drop_extent_cache(inode, em->start,
6693                                 em->start + em->len - 1, 0);
6694                 write_lock(&em_tree->lock);
6695                 ret = add_extent_mapping(em_tree, em);
6696                 if (!ret)
6697                         list_move(&em->list,
6698                                   &em_tree->modified_extents);
6699                 write_unlock(&em_tree->lock);
6700         } while (ret == -EEXIST);
6701
6702         if (ret) {
6703                 free_extent_map(em);
6704                 return ERR_PTR(ret);
6705         }
6706
6707         return em;
6708 }
6709
6710
6711 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6712                                    struct buffer_head *bh_result, int create)
6713 {
6714         struct extent_map *em;
6715         struct btrfs_root *root = BTRFS_I(inode)->root;
6716         struct extent_state *cached_state = NULL;
6717         u64 start = iblock << inode->i_blkbits;
6718         u64 lockstart, lockend;
6719         u64 len = bh_result->b_size;
6720         struct btrfs_trans_handle *trans;
6721         int unlock_bits = EXTENT_LOCKED;
6722         int ret = 0;
6723
6724         if (create)
6725                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6726         else
6727                 len = min_t(u64, len, root->sectorsize);
6728
6729         lockstart = start;
6730         lockend = start + len - 1;
6731
6732         /*
6733          * If this errors out it's because we couldn't invalidate pagecache for
6734          * this range and we need to fallback to buffered.
6735          */
6736         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6737                 return -ENOTBLK;
6738
6739         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6740         if (IS_ERR(em)) {
6741                 ret = PTR_ERR(em);
6742                 goto unlock_err;
6743         }
6744
6745         /*
6746          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6747          * io.  INLINE is special, and we could probably kludge it in here, but
6748          * it's still buffered so for safety lets just fall back to the generic
6749          * buffered path.
6750          *
6751          * For COMPRESSED we _have_ to read the entire extent in so we can
6752          * decompress it, so there will be buffering required no matter what we
6753          * do, so go ahead and fallback to buffered.
6754          *
6755          * We return -ENOTBLK because thats what makes DIO go ahead and go back
6756          * to buffered IO.  Don't blame me, this is the price we pay for using
6757          * the generic code.
6758          */
6759         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6760             em->block_start == EXTENT_MAP_INLINE) {
6761                 free_extent_map(em);
6762                 ret = -ENOTBLK;
6763                 goto unlock_err;
6764         }
6765
6766         /* Just a good old fashioned hole, return */
6767         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6768                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6769                 free_extent_map(em);
6770                 goto unlock_err;
6771         }
6772
6773         /*
6774          * We don't allocate a new extent in the following cases
6775          *
6776          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6777          * existing extent.
6778          * 2) The extent is marked as PREALLOC.  We're good to go here and can
6779          * just use the extent.
6780          *
6781          */
6782         if (!create) {
6783                 len = min(len, em->len - (start - em->start));
6784                 lockstart = start + len;
6785                 goto unlock;
6786         }
6787
6788         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6789             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6790              em->block_start != EXTENT_MAP_HOLE)) {
6791                 int type;
6792                 int ret;
6793                 u64 block_start;
6794
6795                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6796                         type = BTRFS_ORDERED_PREALLOC;
6797                 else
6798                         type = BTRFS_ORDERED_NOCOW;
6799                 len = min(len, em->len - (start - em->start));
6800                 block_start = em->block_start + (start - em->start);
6801
6802                 /*
6803                  * we're not going to log anything, but we do need
6804                  * to make sure the current transaction stays open
6805                  * while we look for nocow cross refs
6806                  */
6807                 trans = btrfs_join_transaction(root);
6808                 if (IS_ERR(trans))
6809                         goto must_cow;
6810
6811                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
6812                         u64 orig_start = em->orig_start;
6813                         u64 orig_block_len = em->orig_block_len;
6814
6815                         if (type == BTRFS_ORDERED_PREALLOC) {
6816                                 free_extent_map(em);
6817                                 em = create_pinned_em(inode, start, len,
6818                                                        orig_start,
6819                                                        block_start, len,
6820                                                        orig_block_len, type);
6821                                 if (IS_ERR(em)) {
6822                                         btrfs_end_transaction(trans, root);
6823                                         goto unlock_err;
6824                                 }
6825                         }
6826
6827                         ret = btrfs_add_ordered_extent_dio(inode, start,
6828                                            block_start, len, len, type);
6829                         btrfs_end_transaction(trans, root);
6830                         if (ret) {
6831                                 free_extent_map(em);
6832                                 goto unlock_err;
6833                         }
6834                         goto unlock;
6835                 }
6836                 btrfs_end_transaction(trans, root);
6837         }
6838 must_cow:
6839         /*
6840          * this will cow the extent, reset the len in case we changed
6841          * it above
6842          */
6843         len = bh_result->b_size;
6844         free_extent_map(em);
6845         em = btrfs_new_extent_direct(inode, start, len);
6846         if (IS_ERR(em)) {
6847                 ret = PTR_ERR(em);
6848                 goto unlock_err;
6849         }
6850         len = min(len, em->len - (start - em->start));
6851 unlock:
6852         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6853                 inode->i_blkbits;
6854         bh_result->b_size = len;
6855         bh_result->b_bdev = em->bdev;
6856         set_buffer_mapped(bh_result);
6857         if (create) {
6858                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6859                         set_buffer_new(bh_result);
6860
6861                 /*
6862                  * Need to update the i_size under the extent lock so buffered
6863                  * readers will get the updated i_size when we unlock.
6864                  */
6865                 if (start + len > i_size_read(inode))
6866                         i_size_write(inode, start + len);
6867
6868                 spin_lock(&BTRFS_I(inode)->lock);
6869                 BTRFS_I(inode)->outstanding_extents++;
6870                 spin_unlock(&BTRFS_I(inode)->lock);
6871
6872                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6873                                      lockstart + len - 1, EXTENT_DELALLOC, NULL,
6874                                      &cached_state, GFP_NOFS);
6875                 BUG_ON(ret);
6876         }
6877
6878         /*
6879          * In the case of write we need to clear and unlock the entire range,
6880          * in the case of read we need to unlock only the end area that we
6881          * aren't using if there is any left over space.
6882          */
6883         if (lockstart < lockend) {
6884                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6885                                  lockend, unlock_bits, 1, 0,
6886                                  &cached_state, GFP_NOFS);
6887         } else {
6888                 free_extent_state(cached_state);
6889         }
6890
6891         free_extent_map(em);
6892
6893         return 0;
6894
6895 unlock_err:
6896         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6897                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6898         return ret;
6899 }
6900
6901 struct btrfs_dio_private {
6902         struct inode *inode;
6903         u64 logical_offset;
6904         u64 disk_bytenr;
6905         u64 bytes;
6906         void *private;
6907
6908         /* number of bios pending for this dio */
6909         atomic_t pending_bios;
6910
6911         /* IO errors */
6912         int errors;
6913
6914         struct bio *orig_bio;
6915 };
6916
6917 static void btrfs_endio_direct_read(struct bio *bio, int err)
6918 {
6919         struct btrfs_dio_private *dip = bio->bi_private;
6920         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6921         struct bio_vec *bvec = bio->bi_io_vec;
6922         struct inode *inode = dip->inode;
6923         struct btrfs_root *root = BTRFS_I(inode)->root;
6924         u64 start;
6925
6926         start = dip->logical_offset;
6927         do {
6928                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6929                         struct page *page = bvec->bv_page;
6930                         char *kaddr;
6931                         u32 csum = ~(u32)0;
6932                         u64 private = ~(u32)0;
6933                         unsigned long flags;
6934
6935                         if (get_state_private(&BTRFS_I(inode)->io_tree,
6936                                               start, &private))
6937                                 goto failed;
6938                         local_irq_save(flags);
6939                         kaddr = kmap_atomic(page);
6940                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
6941                                                csum, bvec->bv_len);
6942                         btrfs_csum_final(csum, (char *)&csum);
6943                         kunmap_atomic(kaddr);
6944                         local_irq_restore(flags);
6945
6946                         flush_dcache_page(bvec->bv_page);
6947                         if (csum != private) {
6948 failed:
6949                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
6950                                       " %llu csum %u private %u\n",
6951                                       (unsigned long long)btrfs_ino(inode),
6952                                       (unsigned long long)start,
6953                                       csum, (unsigned)private);
6954                                 err = -EIO;
6955                         }
6956                 }
6957
6958                 start += bvec->bv_len;
6959                 bvec++;
6960         } while (bvec <= bvec_end);
6961
6962         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6963                       dip->logical_offset + dip->bytes - 1);
6964         bio->bi_private = dip->private;
6965
6966         kfree(dip);
6967
6968         /* If we had a csum failure make sure to clear the uptodate flag */
6969         if (err)
6970                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6971         dio_end_io(bio, err);
6972 }
6973
6974 static void btrfs_endio_direct_write(struct bio *bio, int err)
6975 {
6976         struct btrfs_dio_private *dip = bio->bi_private;
6977         struct inode *inode = dip->inode;
6978         struct btrfs_root *root = BTRFS_I(inode)->root;
6979         struct btrfs_ordered_extent *ordered = NULL;
6980         u64 ordered_offset = dip->logical_offset;
6981         u64 ordered_bytes = dip->bytes;
6982         int ret;
6983
6984         if (err)
6985                 goto out_done;
6986 again:
6987         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6988                                                    &ordered_offset,
6989                                                    ordered_bytes, !err);
6990         if (!ret)
6991                 goto out_test;
6992
6993         ordered->work.func = finish_ordered_fn;
6994         ordered->work.flags = 0;
6995         btrfs_queue_worker(&root->fs_info->endio_write_workers,
6996                            &ordered->work);
6997 out_test:
6998         /*
6999          * our bio might span multiple ordered extents.  If we haven't
7000          * completed the accounting for the whole dio, go back and try again
7001          */
7002         if (ordered_offset < dip->logical_offset + dip->bytes) {
7003                 ordered_bytes = dip->logical_offset + dip->bytes -
7004                         ordered_offset;
7005                 ordered = NULL;
7006                 goto again;
7007         }
7008 out_done:
7009         bio->bi_private = dip->private;
7010
7011         kfree(dip);
7012
7013         /* If we had an error make sure to clear the uptodate flag */
7014         if (err)
7015                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
7016         dio_end_io(bio, err);
7017 }
7018
7019 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7020                                     struct bio *bio, int mirror_num,
7021                                     unsigned long bio_flags, u64 offset)
7022 {
7023         int ret;
7024         struct btrfs_root *root = BTRFS_I(inode)->root;
7025         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7026         BUG_ON(ret); /* -ENOMEM */
7027         return 0;
7028 }
7029
7030 static void btrfs_end_dio_bio(struct bio *bio, int err)
7031 {
7032         struct btrfs_dio_private *dip = bio->bi_private;
7033
7034         if (err) {
7035                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
7036                       "sector %#Lx len %u err no %d\n",
7037                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
7038                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
7039                 dip->errors = 1;
7040
7041                 /*
7042                  * before atomic variable goto zero, we must make sure
7043                  * dip->errors is perceived to be set.
7044                  */
7045                 smp_mb__before_atomic_dec();
7046         }
7047
7048         /* if there are more bios still pending for this dio, just exit */
7049         if (!atomic_dec_and_test(&dip->pending_bios))
7050                 goto out;
7051
7052         if (dip->errors)
7053                 bio_io_error(dip->orig_bio);
7054         else {
7055                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
7056                 bio_endio(dip->orig_bio, 0);
7057         }
7058 out:
7059         bio_put(bio);
7060 }
7061
7062 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7063                                        u64 first_sector, gfp_t gfp_flags)
7064 {
7065         int nr_vecs = bio_get_nr_vecs(bdev);
7066         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7067 }
7068
7069 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7070                                          int rw, u64 file_offset, int skip_sum,
7071                                          int async_submit)
7072 {
7073         int write = rw & REQ_WRITE;
7074         struct btrfs_root *root = BTRFS_I(inode)->root;
7075         int ret;
7076
7077         if (async_submit)
7078                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7079
7080         bio_get(bio);
7081
7082         if (!write) {
7083                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7084                 if (ret)
7085                         goto err;
7086         }
7087
7088         if (skip_sum)
7089                 goto map;
7090
7091         if (write && async_submit) {
7092                 ret = btrfs_wq_submit_bio(root->fs_info,
7093                                    inode, rw, bio, 0, 0,
7094                                    file_offset,
7095                                    __btrfs_submit_bio_start_direct_io,
7096                                    __btrfs_submit_bio_done);
7097                 goto err;
7098         } else if (write) {
7099                 /*
7100                  * If we aren't doing async submit, calculate the csum of the
7101                  * bio now.
7102                  */
7103                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7104                 if (ret)
7105                         goto err;
7106         } else if (!skip_sum) {
7107                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
7108                 if (ret)
7109                         goto err;
7110         }
7111
7112 map:
7113         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7114 err:
7115         bio_put(bio);
7116         return ret;
7117 }
7118
7119 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7120                                     int skip_sum)
7121 {
7122         struct inode *inode = dip->inode;
7123         struct btrfs_root *root = BTRFS_I(inode)->root;
7124         struct bio *bio;
7125         struct bio *orig_bio = dip->orig_bio;
7126         struct bio_vec *bvec = orig_bio->bi_io_vec;
7127         u64 start_sector = orig_bio->bi_sector;
7128         u64 file_offset = dip->logical_offset;
7129         u64 submit_len = 0;
7130         u64 map_length;
7131         int nr_pages = 0;
7132         int ret = 0;
7133         int async_submit = 0;
7134
7135         map_length = orig_bio->bi_size;
7136         ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7137                               &map_length, NULL, 0);
7138         if (ret) {
7139                 bio_put(orig_bio);
7140                 return -EIO;
7141         }
7142         if (map_length >= orig_bio->bi_size) {
7143                 bio = orig_bio;
7144                 goto submit;
7145         }
7146
7147         /* async crcs make it difficult to collect full stripe writes. */
7148         if (btrfs_get_alloc_profile(root, 1) &
7149             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7150                 async_submit = 0;
7151         else
7152                 async_submit = 1;
7153
7154         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7155         if (!bio)
7156                 return -ENOMEM;
7157         bio->bi_private = dip;
7158         bio->bi_end_io = btrfs_end_dio_bio;
7159         atomic_inc(&dip->pending_bios);
7160
7161         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7162                 if (unlikely(map_length < submit_len + bvec->bv_len ||
7163                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7164                                  bvec->bv_offset) < bvec->bv_len)) {
7165                         /*
7166                          * inc the count before we submit the bio so
7167                          * we know the end IO handler won't happen before
7168                          * we inc the count. Otherwise, the dip might get freed
7169                          * before we're done setting it up
7170                          */
7171                         atomic_inc(&dip->pending_bios);
7172                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
7173                                                      file_offset, skip_sum,
7174                                                      async_submit);
7175                         if (ret) {
7176                                 bio_put(bio);
7177                                 atomic_dec(&dip->pending_bios);
7178                                 goto out_err;
7179                         }
7180
7181                         start_sector += submit_len >> 9;
7182                         file_offset += submit_len;
7183
7184                         submit_len = 0;
7185                         nr_pages = 0;
7186
7187                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7188                                                   start_sector, GFP_NOFS);
7189                         if (!bio)
7190                                 goto out_err;
7191                         bio->bi_private = dip;
7192                         bio->bi_end_io = btrfs_end_dio_bio;
7193
7194                         map_length = orig_bio->bi_size;
7195                         ret = btrfs_map_block(root->fs_info, rw,
7196                                               start_sector << 9,
7197                                               &map_length, NULL, 0);
7198                         if (ret) {
7199                                 bio_put(bio);
7200                                 goto out_err;
7201                         }
7202                 } else {
7203                         submit_len += bvec->bv_len;
7204                         nr_pages ++;
7205                         bvec++;
7206                 }
7207         }
7208
7209 submit:
7210         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7211                                      async_submit);
7212         if (!ret)
7213                 return 0;
7214
7215         bio_put(bio);
7216 out_err:
7217         dip->errors = 1;
7218         /*
7219          * before atomic variable goto zero, we must
7220          * make sure dip->errors is perceived to be set.
7221          */
7222         smp_mb__before_atomic_dec();
7223         if (atomic_dec_and_test(&dip->pending_bios))
7224                 bio_io_error(dip->orig_bio);
7225
7226         /* bio_end_io() will handle error, so we needn't return it */
7227         return 0;
7228 }
7229
7230 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
7231                                 loff_t file_offset)
7232 {
7233         struct btrfs_root *root = BTRFS_I(inode)->root;
7234         struct btrfs_dio_private *dip;
7235         struct bio_vec *bvec = bio->bi_io_vec;
7236         int skip_sum;
7237         int write = rw & REQ_WRITE;
7238         int ret = 0;
7239
7240         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7241
7242         dip = kmalloc(sizeof(*dip), GFP_NOFS);
7243         if (!dip) {
7244                 ret = -ENOMEM;
7245                 goto free_ordered;
7246         }
7247
7248         dip->private = bio->bi_private;
7249         dip->inode = inode;
7250         dip->logical_offset = file_offset;
7251
7252         dip->bytes = 0;
7253         do {
7254                 dip->bytes += bvec->bv_len;
7255                 bvec++;
7256         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
7257
7258         dip->disk_bytenr = (u64)bio->bi_sector << 9;
7259         bio->bi_private = dip;
7260         dip->errors = 0;
7261         dip->orig_bio = bio;
7262         atomic_set(&dip->pending_bios, 0);
7263
7264         if (write)
7265                 bio->bi_end_io = btrfs_endio_direct_write;
7266         else
7267                 bio->bi_end_io = btrfs_endio_direct_read;
7268
7269         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7270         if (!ret)
7271                 return;
7272 free_ordered:
7273         /*
7274          * If this is a write, we need to clean up the reserved space and kill
7275          * the ordered extent.
7276          */
7277         if (write) {
7278                 struct btrfs_ordered_extent *ordered;
7279                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7280                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7281                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7282                         btrfs_free_reserved_extent(root, ordered->start,
7283                                                    ordered->disk_len);
7284                 btrfs_put_ordered_extent(ordered);
7285                 btrfs_put_ordered_extent(ordered);
7286         }
7287         bio_endio(bio, ret);
7288 }
7289
7290 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7291                         const struct iovec *iov, loff_t offset,
7292                         unsigned long nr_segs)
7293 {
7294         int seg;
7295         int i;
7296         size_t size;
7297         unsigned long addr;
7298         unsigned blocksize_mask = root->sectorsize - 1;
7299         ssize_t retval = -EINVAL;
7300         loff_t end = offset;
7301
7302         if (offset & blocksize_mask)
7303                 goto out;
7304
7305         /* Check the memory alignment.  Blocks cannot straddle pages */
7306         for (seg = 0; seg < nr_segs; seg++) {
7307                 addr = (unsigned long)iov[seg].iov_base;
7308                 size = iov[seg].iov_len;
7309                 end += size;
7310                 if ((addr & blocksize_mask) || (size & blocksize_mask))
7311                         goto out;
7312
7313                 /* If this is a write we don't need to check anymore */
7314                 if (rw & WRITE)
7315                         continue;
7316
7317                 /*
7318                  * Check to make sure we don't have duplicate iov_base's in this
7319                  * iovec, if so return EINVAL, otherwise we'll get csum errors
7320                  * when reading back.
7321                  */
7322                 for (i = seg + 1; i < nr_segs; i++) {
7323                         if (iov[seg].iov_base == iov[i].iov_base)
7324                                 goto out;
7325                 }
7326         }
7327         retval = 0;
7328 out:
7329         return retval;
7330 }
7331
7332 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7333                         const struct iovec *iov, loff_t offset,
7334                         unsigned long nr_segs)
7335 {
7336         struct file *file = iocb->ki_filp;
7337         struct inode *inode = file->f_mapping->host;
7338         size_t count = 0;
7339         int flags = 0;
7340         bool wakeup = true;
7341         bool relock = false;
7342         ssize_t ret;
7343
7344         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7345                             offset, nr_segs))
7346                 return 0;
7347
7348         atomic_inc(&inode->i_dio_count);
7349         smp_mb__after_atomic_inc();
7350
7351         if (rw & WRITE) {
7352                 count = iov_length(iov, nr_segs);
7353                 /*
7354                  * If the write DIO is beyond the EOF, we need update
7355                  * the isize, but it is protected by i_mutex. So we can
7356                  * not unlock the i_mutex at this case.
7357                  */
7358                 if (offset + count <= inode->i_size) {
7359                         mutex_unlock(&inode->i_mutex);
7360                         relock = true;
7361                 }
7362                 ret = btrfs_delalloc_reserve_space(inode, count);
7363                 if (ret)
7364                         goto out;
7365         } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7366                                      &BTRFS_I(inode)->runtime_flags))) {
7367                 inode_dio_done(inode);
7368                 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7369                 wakeup = false;
7370         }
7371
7372         ret = __blockdev_direct_IO(rw, iocb, inode,
7373                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7374                         iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7375                         btrfs_submit_direct, flags);
7376         if (rw & WRITE) {
7377                 if (ret < 0 && ret != -EIOCBQUEUED)
7378                         btrfs_delalloc_release_space(inode, count);
7379                 else if (ret >= 0 && (size_t)ret < count)
7380                         btrfs_delalloc_release_space(inode,
7381                                                      count - (size_t)ret);
7382                 else
7383                         btrfs_delalloc_release_metadata(inode, 0);
7384         }
7385 out:
7386         if (wakeup)
7387                 inode_dio_done(inode);
7388         if (relock)
7389                 mutex_lock(&inode->i_mutex);
7390
7391         return ret;
7392 }
7393
7394 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
7395
7396 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7397                 __u64 start, __u64 len)
7398 {
7399         int     ret;
7400
7401         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7402         if (ret)
7403                 return ret;
7404
7405         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7406 }
7407
7408 int btrfs_readpage(struct file *file, struct page *page)
7409 {
7410         struct extent_io_tree *tree;
7411         tree = &BTRFS_I(page->mapping->host)->io_tree;
7412         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7413 }
7414
7415 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7416 {
7417         struct extent_io_tree *tree;
7418
7419
7420         if (current->flags & PF_MEMALLOC) {
7421                 redirty_page_for_writepage(wbc, page);
7422                 unlock_page(page);
7423                 return 0;
7424         }
7425         tree = &BTRFS_I(page->mapping->host)->io_tree;
7426         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7427 }
7428
7429 int btrfs_writepages(struct address_space *mapping,
7430                      struct writeback_control *wbc)
7431 {
7432         struct extent_io_tree *tree;
7433
7434         tree = &BTRFS_I(mapping->host)->io_tree;
7435         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7436 }
7437
7438 static int
7439 btrfs_readpages(struct file *file, struct address_space *mapping,
7440                 struct list_head *pages, unsigned nr_pages)
7441 {
7442         struct extent_io_tree *tree;
7443         tree = &BTRFS_I(mapping->host)->io_tree;
7444         return extent_readpages(tree, mapping, pages, nr_pages,
7445                                 btrfs_get_extent);
7446 }
7447 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7448 {
7449         struct extent_io_tree *tree;
7450         struct extent_map_tree *map;
7451         int ret;
7452
7453         tree = &BTRFS_I(page->mapping->host)->io_tree;
7454         map = &BTRFS_I(page->mapping->host)->extent_tree;
7455         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7456         if (ret == 1) {
7457                 ClearPagePrivate(page);
7458                 set_page_private(page, 0);
7459                 page_cache_release(page);
7460         }
7461         return ret;
7462 }
7463
7464 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7465 {
7466         if (PageWriteback(page) || PageDirty(page))
7467                 return 0;
7468         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7469 }
7470
7471 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
7472 {
7473         struct inode *inode = page->mapping->host;
7474         struct extent_io_tree *tree;
7475         struct btrfs_ordered_extent *ordered;
7476         struct extent_state *cached_state = NULL;
7477         u64 page_start = page_offset(page);
7478         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7479
7480         /*
7481          * we have the page locked, so new writeback can't start,
7482          * and the dirty bit won't be cleared while we are here.
7483          *
7484          * Wait for IO on this page so that we can safely clear
7485          * the PagePrivate2 bit and do ordered accounting
7486          */
7487         wait_on_page_writeback(page);
7488
7489         tree = &BTRFS_I(inode)->io_tree;
7490         if (offset) {
7491                 btrfs_releasepage(page, GFP_NOFS);
7492                 return;
7493         }
7494         lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7495         ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
7496         if (ordered) {
7497                 /*
7498                  * IO on this page will never be started, so we need
7499                  * to account for any ordered extents now
7500                  */
7501                 clear_extent_bit(tree, page_start, page_end,
7502                                  EXTENT_DIRTY | EXTENT_DELALLOC |
7503                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7504                                  EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
7505                 /*
7506                  * whoever cleared the private bit is responsible
7507                  * for the finish_ordered_io
7508                  */
7509                 if (TestClearPagePrivate2(page) &&
7510                     btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
7511                                                    PAGE_CACHE_SIZE, 1)) {
7512                         btrfs_finish_ordered_io(ordered);
7513                 }
7514                 btrfs_put_ordered_extent(ordered);
7515                 cached_state = NULL;
7516                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7517         }
7518         clear_extent_bit(tree, page_start, page_end,
7519                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
7520                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
7521                  &cached_state, GFP_NOFS);
7522         __btrfs_releasepage(page, GFP_NOFS);
7523
7524         ClearPageChecked(page);
7525         if (PagePrivate(page)) {
7526                 ClearPagePrivate(page);
7527                 set_page_private(page, 0);
7528                 page_cache_release(page);
7529         }
7530 }
7531
7532 /*
7533  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7534  * called from a page fault handler when a page is first dirtied. Hence we must
7535  * be careful to check for EOF conditions here. We set the page up correctly
7536  * for a written page which means we get ENOSPC checking when writing into
7537  * holes and correct delalloc and unwritten extent mapping on filesystems that
7538  * support these features.
7539  *
7540  * We are not allowed to take the i_mutex here so we have to play games to
7541  * protect against truncate races as the page could now be beyond EOF.  Because
7542  * vmtruncate() writes the inode size before removing pages, once we have the
7543  * page lock we can determine safely if the page is beyond EOF. If it is not
7544  * beyond EOF, then the page is guaranteed safe against truncation until we
7545  * unlock the page.
7546  */
7547 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7548 {
7549         struct page *page = vmf->page;
7550         struct inode *inode = file_inode(vma->vm_file);
7551         struct btrfs_root *root = BTRFS_I(inode)->root;
7552         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7553         struct btrfs_ordered_extent *ordered;
7554         struct extent_state *cached_state = NULL;
7555         char *kaddr;
7556         unsigned long zero_start;
7557         loff_t size;
7558         int ret;
7559         int reserved = 0;
7560         u64 page_start;
7561         u64 page_end;
7562
7563         sb_start_pagefault(inode->i_sb);
7564         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7565         if (!ret) {
7566                 ret = file_update_time(vma->vm_file);
7567                 reserved = 1;
7568         }
7569         if (ret) {
7570                 if (ret == -ENOMEM)
7571                         ret = VM_FAULT_OOM;
7572                 else /* -ENOSPC, -EIO, etc */
7573                         ret = VM_FAULT_SIGBUS;
7574                 if (reserved)
7575                         goto out;
7576                 goto out_noreserve;
7577         }
7578
7579         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7580 again:
7581         lock_page(page);
7582         size = i_size_read(inode);
7583         page_start = page_offset(page);
7584         page_end = page_start + PAGE_CACHE_SIZE - 1;
7585
7586         if ((page->mapping != inode->i_mapping) ||
7587             (page_start >= size)) {
7588                 /* page got truncated out from underneath us */
7589                 goto out_unlock;
7590         }
7591         wait_on_page_writeback(page);
7592
7593         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7594         set_page_extent_mapped(page);
7595
7596         /*
7597          * we can't set the delalloc bits if there are pending ordered
7598          * extents.  Drop our locks and wait for them to finish
7599          */
7600         ordered = btrfs_lookup_ordered_extent(inode, page_start);
7601         if (ordered) {
7602                 unlock_extent_cached(io_tree, page_start, page_end,
7603                                      &cached_state, GFP_NOFS);
7604                 unlock_page(page);
7605                 btrfs_start_ordered_extent(inode, ordered, 1);
7606                 btrfs_put_ordered_extent(ordered);
7607                 goto again;
7608         }
7609
7610         /*
7611          * XXX - page_mkwrite gets called every time the page is dirtied, even
7612          * if it was already dirty, so for space accounting reasons we need to
7613          * clear any delalloc bits for the range we are fixing to save.  There
7614          * is probably a better way to do this, but for now keep consistent with
7615          * prepare_pages in the normal write path.
7616          */
7617         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7618                           EXTENT_DIRTY | EXTENT_DELALLOC |
7619                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7620                           0, 0, &cached_state, GFP_NOFS);
7621
7622         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7623                                         &cached_state);
7624         if (ret) {
7625                 unlock_extent_cached(io_tree, page_start, page_end,
7626                                      &cached_state, GFP_NOFS);
7627                 ret = VM_FAULT_SIGBUS;
7628                 goto out_unlock;
7629         }
7630         ret = 0;
7631
7632         /* page is wholly or partially inside EOF */
7633         if (page_start + PAGE_CACHE_SIZE > size)
7634                 zero_start = size & ~PAGE_CACHE_MASK;
7635         else
7636                 zero_start = PAGE_CACHE_SIZE;
7637
7638         if (zero_start != PAGE_CACHE_SIZE) {
7639                 kaddr = kmap(page);
7640                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7641                 flush_dcache_page(page);
7642                 kunmap(page);
7643         }
7644         ClearPageChecked(page);
7645         set_page_dirty(page);
7646         SetPageUptodate(page);
7647
7648         BTRFS_I(inode)->last_trans = root->fs_info->generation;
7649         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7650         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7651
7652         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7653
7654 out_unlock:
7655         if (!ret) {
7656                 sb_end_pagefault(inode->i_sb);
7657                 return VM_FAULT_LOCKED;
7658         }
7659         unlock_page(page);
7660 out:
7661         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7662 out_noreserve:
7663         sb_end_pagefault(inode->i_sb);
7664         return ret;
7665 }
7666
7667 static int btrfs_truncate(struct inode *inode)
7668 {
7669         struct btrfs_root *root = BTRFS_I(inode)->root;
7670         struct btrfs_block_rsv *rsv;
7671         int ret;
7672         int err = 0;
7673         struct btrfs_trans_handle *trans;
7674         u64 mask = root->sectorsize - 1;
7675         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7676
7677         ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
7678         if (ret)
7679                 return ret;
7680
7681         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
7682         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
7683
7684         /*
7685          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
7686          * 3 things going on here
7687          *
7688          * 1) We need to reserve space for our orphan item and the space to
7689          * delete our orphan item.  Lord knows we don't want to have a dangling
7690          * orphan item because we didn't reserve space to remove it.
7691          *
7692          * 2) We need to reserve space to update our inode.
7693          *
7694          * 3) We need to have something to cache all the space that is going to
7695          * be free'd up by the truncate operation, but also have some slack
7696          * space reserved in case it uses space during the truncate (thank you
7697          * very much snapshotting).
7698          *
7699          * And we need these to all be seperate.  The fact is we can use alot of
7700          * space doing the truncate, and we have no earthly idea how much space
7701          * we will use, so we need the truncate reservation to be seperate so it
7702          * doesn't end up using space reserved for updating the inode or
7703          * removing the orphan item.  We also need to be able to stop the
7704          * transaction and start a new one, which means we need to be able to
7705          * update the inode several times, and we have no idea of knowing how
7706          * many times that will be, so we can't just reserve 1 item for the
7707          * entirety of the opration, so that has to be done seperately as well.
7708          * Then there is the orphan item, which does indeed need to be held on
7709          * to for the whole operation, and we need nobody to touch this reserved
7710          * space except the orphan code.
7711          *
7712          * So that leaves us with
7713          *
7714          * 1) root->orphan_block_rsv - for the orphan deletion.
7715          * 2) rsv - for the truncate reservation, which we will steal from the
7716          * transaction reservation.
7717          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7718          * updating the inode.
7719          */
7720         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7721         if (!rsv)
7722                 return -ENOMEM;
7723         rsv->size = min_size;
7724         rsv->failfast = 1;
7725
7726         /*
7727          * 1 for the truncate slack space
7728          * 1 for updating the inode.
7729          */
7730         trans = btrfs_start_transaction(root, 2);
7731         if (IS_ERR(trans)) {
7732                 err = PTR_ERR(trans);
7733                 goto out;
7734         }
7735
7736         /* Migrate the slack space for the truncate to our reserve */
7737         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7738                                       min_size);
7739         BUG_ON(ret);
7740
7741         /*
7742          * setattr is responsible for setting the ordered_data_close flag,
7743          * but that is only tested during the last file release.  That
7744          * could happen well after the next commit, leaving a great big
7745          * window where new writes may get lost if someone chooses to write
7746          * to this file after truncating to zero
7747          *
7748          * The inode doesn't have any dirty data here, and so if we commit
7749          * this is a noop.  If someone immediately starts writing to the inode
7750          * it is very likely we'll catch some of their writes in this
7751          * transaction, and the commit will find this file on the ordered
7752          * data list with good things to send down.
7753          *
7754          * This is a best effort solution, there is still a window where
7755          * using truncate to replace the contents of the file will
7756          * end up with a zero length file after a crash.
7757          */
7758         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7759                                            &BTRFS_I(inode)->runtime_flags))
7760                 btrfs_add_ordered_operation(trans, root, inode);
7761
7762         /*
7763          * So if we truncate and then write and fsync we normally would just
7764          * write the extents that changed, which is a problem if we need to
7765          * first truncate that entire inode.  So set this flag so we write out
7766          * all of the extents in the inode to the sync log so we're completely
7767          * safe.
7768          */
7769         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7770         trans->block_rsv = rsv;
7771
7772         while (1) {
7773                 ret = btrfs_truncate_inode_items(trans, root, inode,
7774                                                  inode->i_size,
7775                                                  BTRFS_EXTENT_DATA_KEY);
7776                 if (ret != -ENOSPC) {
7777                         err = ret;
7778                         break;
7779                 }
7780
7781                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7782                 ret = btrfs_update_inode(trans, root, inode);
7783                 if (ret) {
7784                         err = ret;
7785                         break;
7786                 }
7787
7788                 btrfs_end_transaction(trans, root);
7789                 btrfs_btree_balance_dirty(root);
7790
7791                 trans = btrfs_start_transaction(root, 2);
7792                 if (IS_ERR(trans)) {
7793                         ret = err = PTR_ERR(trans);
7794                         trans = NULL;
7795                         break;
7796                 }
7797
7798                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7799                                               rsv, min_size);
7800                 BUG_ON(ret);    /* shouldn't happen */
7801                 trans->block_rsv = rsv;
7802         }
7803
7804         if (ret == 0 && inode->i_nlink > 0) {
7805                 trans->block_rsv = root->orphan_block_rsv;
7806                 ret = btrfs_orphan_del(trans, inode);
7807                 if (ret)
7808                         err = ret;
7809         }
7810
7811         if (trans) {
7812                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7813                 ret = btrfs_update_inode(trans, root, inode);
7814                 if (ret && !err)
7815                         err = ret;
7816
7817                 ret = btrfs_end_transaction(trans, root);
7818                 btrfs_btree_balance_dirty(root);
7819         }
7820
7821 out:
7822         btrfs_free_block_rsv(root, rsv);
7823
7824         if (ret && !err)
7825                 err = ret;
7826
7827         return err;
7828 }
7829
7830 /*
7831  * create a new subvolume directory/inode (helper for the ioctl).
7832  */
7833 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7834                              struct btrfs_root *new_root, u64 new_dirid)
7835 {
7836         struct inode *inode;
7837         int err;
7838         u64 index = 0;
7839
7840         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7841                                 new_dirid, new_dirid,
7842                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
7843                                 &index);
7844         if (IS_ERR(inode))
7845                 return PTR_ERR(inode);
7846         inode->i_op = &btrfs_dir_inode_operations;
7847         inode->i_fop = &btrfs_dir_file_operations;
7848
7849         set_nlink(inode, 1);
7850         btrfs_i_size_write(inode, 0);
7851
7852         err = btrfs_update_inode(trans, new_root, inode);
7853
7854         iput(inode);
7855         return err;
7856 }
7857
7858 struct inode *btrfs_alloc_inode(struct super_block *sb)
7859 {
7860         struct btrfs_inode *ei;
7861         struct inode *inode;
7862
7863         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7864         if (!ei)
7865                 return NULL;
7866
7867         ei->root = NULL;
7868         ei->generation = 0;
7869         ei->last_trans = 0;
7870         ei->last_sub_trans = 0;
7871         ei->logged_trans = 0;
7872         ei->delalloc_bytes = 0;
7873         ei->disk_i_size = 0;
7874         ei->flags = 0;
7875         ei->csum_bytes = 0;
7876         ei->index_cnt = (u64)-1;
7877         ei->last_unlink_trans = 0;
7878         ei->last_log_commit = 0;
7879
7880         spin_lock_init(&ei->lock);
7881         ei->outstanding_extents = 0;
7882         ei->reserved_extents = 0;
7883
7884         ei->runtime_flags = 0;
7885         ei->force_compress = BTRFS_COMPRESS_NONE;
7886
7887         ei->delayed_node = NULL;
7888
7889         inode = &ei->vfs_inode;
7890         extent_map_tree_init(&ei->extent_tree);
7891         extent_io_tree_init(&ei->io_tree, &inode->i_data);
7892         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7893         ei->io_tree.track_uptodate = 1;
7894         ei->io_failure_tree.track_uptodate = 1;
7895         atomic_set(&ei->sync_writers, 0);
7896         mutex_init(&ei->log_mutex);
7897         mutex_init(&ei->delalloc_mutex);
7898         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7899         INIT_LIST_HEAD(&ei->delalloc_inodes);
7900         INIT_LIST_HEAD(&ei->ordered_operations);
7901         RB_CLEAR_NODE(&ei->rb_node);
7902
7903         return inode;
7904 }
7905
7906 static void btrfs_i_callback(struct rcu_head *head)
7907 {
7908         struct inode *inode = container_of(head, struct inode, i_rcu);
7909         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7910 }
7911
7912 void btrfs_destroy_inode(struct inode *inode)
7913 {
7914         struct btrfs_ordered_extent *ordered;
7915         struct btrfs_root *root = BTRFS_I(inode)->root;
7916
7917         WARN_ON(!hlist_empty(&inode->i_dentry));
7918         WARN_ON(inode->i_data.nrpages);
7919         WARN_ON(BTRFS_I(inode)->outstanding_extents);
7920         WARN_ON(BTRFS_I(inode)->reserved_extents);
7921         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7922         WARN_ON(BTRFS_I(inode)->csum_bytes);
7923
7924         /*
7925          * This can happen where we create an inode, but somebody else also
7926          * created the same inode and we need to destroy the one we already
7927          * created.
7928          */
7929         if (!root)
7930                 goto free;
7931
7932         /*
7933          * Make sure we're properly removed from the ordered operation
7934          * lists.
7935          */
7936         smp_mb();
7937         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7938                 spin_lock(&root->fs_info->ordered_extent_lock);
7939                 list_del_init(&BTRFS_I(inode)->ordered_operations);
7940                 spin_unlock(&root->fs_info->ordered_extent_lock);
7941         }
7942
7943         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7944                      &BTRFS_I(inode)->runtime_flags)) {
7945                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7946                        (unsigned long long)btrfs_ino(inode));
7947                 atomic_dec(&root->orphan_inodes);
7948         }
7949
7950         while (1) {
7951                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7952                 if (!ordered)
7953                         break;
7954                 else {
7955                         printk(KERN_ERR "btrfs found ordered "
7956                                "extent %llu %llu on inode cleanup\n",
7957                                (unsigned long long)ordered->file_offset,
7958                                (unsigned long long)ordered->len);
7959                         btrfs_remove_ordered_extent(inode, ordered);
7960                         btrfs_put_ordered_extent(ordered);
7961                         btrfs_put_ordered_extent(ordered);
7962                 }
7963         }
7964         inode_tree_del(inode);
7965         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7966 free:
7967         btrfs_remove_delayed_node(inode);
7968         call_rcu(&inode->i_rcu, btrfs_i_callback);
7969 }
7970
7971 int btrfs_drop_inode(struct inode *inode)
7972 {
7973         struct btrfs_root *root = BTRFS_I(inode)->root;
7974
7975         /* the snap/subvol tree is on deleting */
7976         if (btrfs_root_refs(&root->root_item) == 0 &&
7977             root != root->fs_info->tree_root)
7978                 return 1;
7979         else
7980                 return generic_drop_inode(inode);
7981 }
7982
7983 static void init_once(void *foo)
7984 {
7985         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7986
7987         inode_init_once(&ei->vfs_inode);
7988 }
7989
7990 void btrfs_destroy_cachep(void)
7991 {
7992         /*
7993          * Make sure all delayed rcu free inodes are flushed before we
7994          * destroy cache.
7995          */
7996         rcu_barrier();
7997         if (btrfs_inode_cachep)
7998                 kmem_cache_destroy(btrfs_inode_cachep);
7999         if (btrfs_trans_handle_cachep)
8000                 kmem_cache_destroy(btrfs_trans_handle_cachep);
8001         if (btrfs_transaction_cachep)
8002                 kmem_cache_destroy(btrfs_transaction_cachep);
8003         if (btrfs_path_cachep)
8004                 kmem_cache_destroy(btrfs_path_cachep);
8005         if (btrfs_free_space_cachep)
8006                 kmem_cache_destroy(btrfs_free_space_cachep);
8007         if (btrfs_delalloc_work_cachep)
8008                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8009 }
8010
8011 int btrfs_init_cachep(void)
8012 {
8013         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8014                         sizeof(struct btrfs_inode), 0,
8015                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8016         if (!btrfs_inode_cachep)
8017                 goto fail;
8018
8019         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8020                         sizeof(struct btrfs_trans_handle), 0,
8021                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8022         if (!btrfs_trans_handle_cachep)
8023                 goto fail;
8024
8025         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8026                         sizeof(struct btrfs_transaction), 0,
8027                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8028         if (!btrfs_transaction_cachep)
8029                 goto fail;
8030
8031         btrfs_path_cachep = kmem_cache_create("btrfs_path",
8032                         sizeof(struct btrfs_path), 0,
8033                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8034         if (!btrfs_path_cachep)
8035                 goto fail;
8036
8037         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8038                         sizeof(struct btrfs_free_space), 0,
8039                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8040         if (!btrfs_free_space_cachep)
8041                 goto fail;
8042
8043         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8044                         sizeof(struct btrfs_delalloc_work), 0,
8045                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8046                         NULL);
8047         if (!btrfs_delalloc_work_cachep)
8048                 goto fail;
8049
8050         return 0;
8051 fail:
8052         btrfs_destroy_cachep();
8053         return -ENOMEM;
8054 }
8055
8056 static int btrfs_getattr(struct vfsmount *mnt,
8057                          struct dentry *dentry, struct kstat *stat)
8058 {
8059         u64 delalloc_bytes;
8060         struct inode *inode = dentry->d_inode;
8061         u32 blocksize = inode->i_sb->s_blocksize;
8062
8063         generic_fillattr(inode, stat);
8064         stat->dev = BTRFS_I(inode)->root->anon_dev;
8065         stat->blksize = PAGE_CACHE_SIZE;
8066
8067         spin_lock(&BTRFS_I(inode)->lock);
8068         delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8069         spin_unlock(&BTRFS_I(inode)->lock);
8070         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8071                         ALIGN(delalloc_bytes, blocksize)) >> 9;
8072         return 0;
8073 }
8074
8075 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8076                            struct inode *new_dir, struct dentry *new_dentry)
8077 {
8078         struct btrfs_trans_handle *trans;
8079         struct btrfs_root *root = BTRFS_I(old_dir)->root;
8080         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8081         struct inode *new_inode = new_dentry->d_inode;
8082         struct inode *old_inode = old_dentry->d_inode;
8083         struct timespec ctime = CURRENT_TIME;
8084         u64 index = 0;
8085         u64 root_objectid;
8086         int ret;
8087         u64 old_ino = btrfs_ino(old_inode);
8088
8089         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8090                 return -EPERM;
8091
8092         /* we only allow rename subvolume link between subvolumes */
8093         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8094                 return -EXDEV;
8095
8096         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8097             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8098                 return -ENOTEMPTY;
8099
8100         if (S_ISDIR(old_inode->i_mode) && new_inode &&
8101             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8102                 return -ENOTEMPTY;
8103
8104
8105         /* check for collisions, even if the  name isn't there */
8106         ret = btrfs_check_dir_item_collision(root, new_dir->i_ino,
8107                              new_dentry->d_name.name,
8108                              new_dentry->d_name.len);
8109
8110         if (ret) {
8111                 if (ret == -EEXIST) {
8112                         /* we shouldn't get
8113                          * eexist without a new_inode */
8114                         if (!new_inode) {
8115                                 WARN_ON(1);
8116                                 return ret;
8117                         }
8118                 } else {
8119                         /* maybe -EOVERFLOW */
8120                         return ret;
8121                 }
8122         }
8123         ret = 0;
8124
8125         /*
8126          * we're using rename to replace one file with another.
8127          * and the replacement file is large.  Start IO on it now so
8128          * we don't add too much work to the end of the transaction
8129          */
8130         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8131             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8132                 filemap_flush(old_inode->i_mapping);
8133
8134         /* close the racy window with snapshot create/destroy ioctl */
8135         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8136                 down_read(&root->fs_info->subvol_sem);
8137         /*
8138          * We want to reserve the absolute worst case amount of items.  So if
8139          * both inodes are subvols and we need to unlink them then that would
8140          * require 4 item modifications, but if they are both normal inodes it
8141          * would require 5 item modifications, so we'll assume their normal
8142          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8143          * should cover the worst case number of items we'll modify.
8144          */
8145         trans = btrfs_start_transaction(root, 11);
8146         if (IS_ERR(trans)) {
8147                 ret = PTR_ERR(trans);
8148                 goto out_notrans;
8149         }
8150
8151         if (dest != root)
8152                 btrfs_record_root_in_trans(trans, dest);
8153
8154         ret = btrfs_set_inode_index(new_dir, &index);
8155         if (ret)
8156                 goto out_fail;
8157
8158         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8159                 /* force full log commit if subvolume involved. */
8160                 root->fs_info->last_trans_log_full_commit = trans->transid;
8161         } else {
8162                 ret = btrfs_insert_inode_ref(trans, dest,
8163                                              new_dentry->d_name.name,
8164                                              new_dentry->d_name.len,
8165                                              old_ino,
8166                                              btrfs_ino(new_dir), index);
8167                 if (ret)
8168                         goto out_fail;
8169                 /*
8170                  * this is an ugly little race, but the rename is required
8171                  * to make sure that if we crash, the inode is either at the
8172                  * old name or the new one.  pinning the log transaction lets
8173                  * us make sure we don't allow a log commit to come in after
8174                  * we unlink the name but before we add the new name back in.
8175                  */
8176                 btrfs_pin_log_trans(root);
8177         }
8178         /*
8179          * make sure the inode gets flushed if it is replacing
8180          * something.
8181          */
8182         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8183                 btrfs_add_ordered_operation(trans, root, old_inode);
8184
8185         inode_inc_iversion(old_dir);
8186         inode_inc_iversion(new_dir);
8187         inode_inc_iversion(old_inode);
8188         old_dir->i_ctime = old_dir->i_mtime = ctime;
8189         new_dir->i_ctime = new_dir->i_mtime = ctime;
8190         old_inode->i_ctime = ctime;
8191
8192         if (old_dentry->d_parent != new_dentry->d_parent)
8193                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8194
8195         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8196                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8197                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8198                                         old_dentry->d_name.name,
8199                                         old_dentry->d_name.len);
8200         } else {
8201                 ret = __btrfs_unlink_inode(trans, root, old_dir,
8202                                         old_dentry->d_inode,
8203                                         old_dentry->d_name.name,
8204                                         old_dentry->d_name.len);
8205                 if (!ret)
8206                         ret = btrfs_update_inode(trans, root, old_inode);
8207         }
8208         if (ret) {
8209                 btrfs_abort_transaction(trans, root, ret);
8210                 goto out_fail;
8211         }
8212
8213         if (new_inode) {
8214                 inode_inc_iversion(new_inode);
8215                 new_inode->i_ctime = CURRENT_TIME;
8216                 if (unlikely(btrfs_ino(new_inode) ==
8217                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8218                         root_objectid = BTRFS_I(new_inode)->location.objectid;
8219                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
8220                                                 root_objectid,
8221                                                 new_dentry->d_name.name,
8222                                                 new_dentry->d_name.len);
8223                         BUG_ON(new_inode->i_nlink == 0);
8224                 } else {
8225                         ret = btrfs_unlink_inode(trans, dest, new_dir,
8226                                                  new_dentry->d_inode,
8227                                                  new_dentry->d_name.name,
8228                                                  new_dentry->d_name.len);
8229                 }
8230                 if (!ret && new_inode->i_nlink == 0) {
8231                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8232                         BUG_ON(ret);
8233                 }
8234                 if (ret) {
8235                         btrfs_abort_transaction(trans, root, ret);
8236                         goto out_fail;
8237                 }
8238         }
8239
8240         ret = btrfs_add_link(trans, new_dir, old_inode,
8241                              new_dentry->d_name.name,
8242                              new_dentry->d_name.len, 0, index);
8243         if (ret) {
8244                 btrfs_abort_transaction(trans, root, ret);
8245                 goto out_fail;
8246         }
8247
8248         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8249                 struct dentry *parent = new_dentry->d_parent;
8250                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8251                 btrfs_end_log_trans(root);
8252         }
8253 out_fail:
8254         btrfs_end_transaction(trans, root);
8255 out_notrans:
8256         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8257                 up_read(&root->fs_info->subvol_sem);
8258
8259         return ret;
8260 }
8261
8262 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8263 {
8264         struct btrfs_delalloc_work *delalloc_work;
8265
8266         delalloc_work = container_of(work, struct btrfs_delalloc_work,
8267                                      work);
8268         if (delalloc_work->wait)
8269                 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
8270         else
8271                 filemap_flush(delalloc_work->inode->i_mapping);
8272
8273         if (delalloc_work->delay_iput)
8274                 btrfs_add_delayed_iput(delalloc_work->inode);
8275         else
8276                 iput(delalloc_work->inode);
8277         complete(&delalloc_work->completion);
8278 }
8279
8280 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8281                                                     int wait, int delay_iput)
8282 {
8283         struct btrfs_delalloc_work *work;
8284
8285         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8286         if (!work)
8287                 return NULL;
8288
8289         init_completion(&work->completion);
8290         INIT_LIST_HEAD(&work->list);
8291         work->inode = inode;
8292         work->wait = wait;
8293         work->delay_iput = delay_iput;
8294         work->work.func = btrfs_run_delalloc_work;
8295
8296         return work;
8297 }
8298
8299 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8300 {
8301         wait_for_completion(&work->completion);
8302         kmem_cache_free(btrfs_delalloc_work_cachep, work);
8303 }
8304
8305 /*
8306  * some fairly slow code that needs optimization. This walks the list
8307  * of all the inodes with pending delalloc and forces them to disk.
8308  */
8309 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8310 {
8311         struct btrfs_inode *binode;
8312         struct inode *inode;
8313         struct btrfs_delalloc_work *work, *next;
8314         struct list_head works;
8315         struct list_head splice;
8316         int ret = 0;
8317
8318         if (root->fs_info->sb->s_flags & MS_RDONLY)
8319                 return -EROFS;
8320
8321         INIT_LIST_HEAD(&works);
8322         INIT_LIST_HEAD(&splice);
8323
8324         spin_lock(&root->fs_info->delalloc_lock);
8325         list_splice_init(&root->fs_info->delalloc_inodes, &splice);
8326         while (!list_empty(&splice)) {
8327                 binode = list_entry(splice.next, struct btrfs_inode,
8328                                     delalloc_inodes);
8329
8330                 list_del_init(&binode->delalloc_inodes);
8331
8332                 inode = igrab(&binode->vfs_inode);
8333                 if (!inode) {
8334                         clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
8335                                   &binode->runtime_flags);
8336                         continue;
8337                 }
8338
8339                 list_add_tail(&binode->delalloc_inodes,
8340                               &root->fs_info->delalloc_inodes);
8341                 spin_unlock(&root->fs_info->delalloc_lock);
8342
8343                 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8344                 if (unlikely(!work)) {
8345                         ret = -ENOMEM;
8346                         goto out;
8347                 }
8348                 list_add_tail(&work->list, &works);
8349                 btrfs_queue_worker(&root->fs_info->flush_workers,
8350                                    &work->work);
8351
8352                 cond_resched();
8353                 spin_lock(&root->fs_info->delalloc_lock);
8354         }
8355         spin_unlock(&root->fs_info->delalloc_lock);
8356
8357         list_for_each_entry_safe(work, next, &works, list) {
8358                 list_del_init(&work->list);
8359                 btrfs_wait_and_free_delalloc_work(work);
8360         }
8361
8362         /* the filemap_flush will queue IO into the worker threads, but
8363          * we have to make sure the IO is actually started and that
8364          * ordered extents get created before we return
8365          */
8366         atomic_inc(&root->fs_info->async_submit_draining);
8367         while (atomic_read(&root->fs_info->nr_async_submits) ||
8368               atomic_read(&root->fs_info->async_delalloc_pages)) {
8369                 wait_event(root->fs_info->async_submit_wait,
8370                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8371                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8372         }
8373         atomic_dec(&root->fs_info->async_submit_draining);
8374         return 0;
8375 out:
8376         list_for_each_entry_safe(work, next, &works, list) {
8377                 list_del_init(&work->list);
8378                 btrfs_wait_and_free_delalloc_work(work);
8379         }
8380
8381         if (!list_empty_careful(&splice)) {
8382                 spin_lock(&root->fs_info->delalloc_lock);
8383                 list_splice_tail(&splice, &root->fs_info->delalloc_inodes);
8384                 spin_unlock(&root->fs_info->delalloc_lock);
8385         }
8386         return ret;
8387 }
8388
8389 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8390                          const char *symname)
8391 {
8392         struct btrfs_trans_handle *trans;
8393         struct btrfs_root *root = BTRFS_I(dir)->root;
8394         struct btrfs_path *path;
8395         struct btrfs_key key;
8396         struct inode *inode = NULL;
8397         int err;
8398         int drop_inode = 0;
8399         u64 objectid;
8400         u64 index = 0 ;
8401         int name_len;
8402         int datasize;
8403         unsigned long ptr;
8404         struct btrfs_file_extent_item *ei;
8405         struct extent_buffer *leaf;
8406
8407         name_len = strlen(symname) + 1;
8408         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8409                 return -ENAMETOOLONG;
8410
8411         /*
8412          * 2 items for inode item and ref
8413          * 2 items for dir items
8414          * 1 item for xattr if selinux is on
8415          */
8416         trans = btrfs_start_transaction(root, 5);
8417         if (IS_ERR(trans))
8418                 return PTR_ERR(trans);
8419
8420         err = btrfs_find_free_ino(root, &objectid);
8421         if (err)
8422                 goto out_unlock;
8423
8424         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8425                                 dentry->d_name.len, btrfs_ino(dir), objectid,
8426                                 S_IFLNK|S_IRWXUGO, &index);
8427         if (IS_ERR(inode)) {
8428                 err = PTR_ERR(inode);
8429                 goto out_unlock;
8430         }
8431
8432         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8433         if (err) {
8434                 drop_inode = 1;
8435                 goto out_unlock;
8436         }
8437
8438         /*
8439         * If the active LSM wants to access the inode during
8440         * d_instantiate it needs these. Smack checks to see
8441         * if the filesystem supports xattrs by looking at the
8442         * ops vector.
8443         */
8444         inode->i_fop = &btrfs_file_operations;
8445         inode->i_op = &btrfs_file_inode_operations;
8446
8447         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8448         if (err)
8449                 drop_inode = 1;
8450         else {
8451                 inode->i_mapping->a_ops = &btrfs_aops;
8452                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8453                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8454         }
8455         if (drop_inode)
8456                 goto out_unlock;
8457
8458         path = btrfs_alloc_path();
8459         if (!path) {
8460                 err = -ENOMEM;
8461                 drop_inode = 1;
8462                 goto out_unlock;
8463         }
8464         key.objectid = btrfs_ino(inode);
8465         key.offset = 0;
8466         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8467         datasize = btrfs_file_extent_calc_inline_size(name_len);
8468         err = btrfs_insert_empty_item(trans, root, path, &key,
8469                                       datasize);
8470         if (err) {
8471                 drop_inode = 1;
8472                 btrfs_free_path(path);
8473                 goto out_unlock;
8474         }
8475         leaf = path->nodes[0];
8476         ei = btrfs_item_ptr(leaf, path->slots[0],
8477                             struct btrfs_file_extent_item);
8478         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8479         btrfs_set_file_extent_type(leaf, ei,
8480                                    BTRFS_FILE_EXTENT_INLINE);
8481         btrfs_set_file_extent_encryption(leaf, ei, 0);
8482         btrfs_set_file_extent_compression(leaf, ei, 0);
8483         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8484         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8485
8486         ptr = btrfs_file_extent_inline_start(ei);
8487         write_extent_buffer(leaf, symname, ptr, name_len);
8488         btrfs_mark_buffer_dirty(leaf);
8489         btrfs_free_path(path);
8490
8491         inode->i_op = &btrfs_symlink_inode_operations;
8492         inode->i_mapping->a_ops = &btrfs_symlink_aops;
8493         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8494         inode_set_bytes(inode, name_len);
8495         btrfs_i_size_write(inode, name_len - 1);
8496         err = btrfs_update_inode(trans, root, inode);
8497         if (err)
8498                 drop_inode = 1;
8499
8500 out_unlock:
8501         if (!err)
8502                 d_instantiate(dentry, inode);
8503         btrfs_end_transaction(trans, root);
8504         if (drop_inode) {
8505                 inode_dec_link_count(inode);
8506                 iput(inode);
8507         }
8508         btrfs_btree_balance_dirty(root);
8509         return err;
8510 }
8511
8512 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8513                                        u64 start, u64 num_bytes, u64 min_size,
8514                                        loff_t actual_len, u64 *alloc_hint,
8515                                        struct btrfs_trans_handle *trans)
8516 {
8517         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8518         struct extent_map *em;
8519         struct btrfs_root *root = BTRFS_I(inode)->root;
8520         struct btrfs_key ins;
8521         u64 cur_offset = start;
8522         u64 i_size;
8523         u64 cur_bytes;
8524         int ret = 0;
8525         bool own_trans = true;
8526
8527         if (trans)
8528                 own_trans = false;
8529         while (num_bytes > 0) {
8530                 if (own_trans) {
8531                         trans = btrfs_start_transaction(root, 3);
8532                         if (IS_ERR(trans)) {
8533                                 ret = PTR_ERR(trans);
8534                                 break;
8535                         }
8536                 }
8537
8538                 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8539                 cur_bytes = max(cur_bytes, min_size);
8540                 ret = btrfs_reserve_extent(trans, root, cur_bytes,
8541                                            min_size, 0, *alloc_hint, &ins, 1);
8542                 if (ret) {
8543                         if (own_trans)
8544                                 btrfs_end_transaction(trans, root);
8545                         break;
8546                 }
8547
8548                 ret = insert_reserved_file_extent(trans, inode,
8549                                                   cur_offset, ins.objectid,
8550                                                   ins.offset, ins.offset,
8551                                                   ins.offset, 0, 0, 0,
8552                                                   BTRFS_FILE_EXTENT_PREALLOC);
8553                 if (ret) {
8554                         btrfs_abort_transaction(trans, root, ret);
8555                         if (own_trans)
8556                                 btrfs_end_transaction(trans, root);
8557                         break;
8558                 }
8559                 btrfs_drop_extent_cache(inode, cur_offset,
8560                                         cur_offset + ins.offset -1, 0);
8561
8562                 em = alloc_extent_map();
8563                 if (!em) {
8564                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8565                                 &BTRFS_I(inode)->runtime_flags);
8566                         goto next;
8567                 }
8568
8569                 em->start = cur_offset;
8570                 em->orig_start = cur_offset;
8571                 em->len = ins.offset;
8572                 em->block_start = ins.objectid;
8573                 em->block_len = ins.offset;
8574                 em->orig_block_len = ins.offset;
8575                 em->bdev = root->fs_info->fs_devices->latest_bdev;
8576                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8577                 em->generation = trans->transid;
8578
8579                 while (1) {
8580                         write_lock(&em_tree->lock);
8581                         ret = add_extent_mapping(em_tree, em);
8582                         if (!ret)
8583                                 list_move(&em->list,
8584                                           &em_tree->modified_extents);
8585                         write_unlock(&em_tree->lock);
8586                         if (ret != -EEXIST)
8587                                 break;
8588                         btrfs_drop_extent_cache(inode, cur_offset,
8589                                                 cur_offset + ins.offset - 1,
8590                                                 0);
8591                 }
8592                 free_extent_map(em);
8593 next:
8594                 num_bytes -= ins.offset;
8595                 cur_offset += ins.offset;
8596                 *alloc_hint = ins.objectid + ins.offset;
8597
8598                 inode_inc_iversion(inode);
8599                 inode->i_ctime = CURRENT_TIME;
8600                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8601                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8602                     (actual_len > inode->i_size) &&
8603                     (cur_offset > inode->i_size)) {
8604                         if (cur_offset > actual_len)
8605                                 i_size = actual_len;
8606                         else
8607                                 i_size = cur_offset;
8608                         i_size_write(inode, i_size);
8609                         btrfs_ordered_update_i_size(inode, i_size, NULL);
8610                 }
8611
8612                 ret = btrfs_update_inode(trans, root, inode);
8613
8614                 if (ret) {
8615                         btrfs_abort_transaction(trans, root, ret);
8616                         if (own_trans)
8617                                 btrfs_end_transaction(trans, root);
8618                         break;
8619                 }
8620
8621                 if (own_trans)
8622                         btrfs_end_transaction(trans, root);
8623         }
8624         return ret;
8625 }
8626
8627 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8628                               u64 start, u64 num_bytes, u64 min_size,
8629                               loff_t actual_len, u64 *alloc_hint)
8630 {
8631         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8632                                            min_size, actual_len, alloc_hint,
8633                                            NULL);
8634 }
8635
8636 int btrfs_prealloc_file_range_trans(struct inode *inode,
8637                                     struct btrfs_trans_handle *trans, int mode,
8638                                     u64 start, u64 num_bytes, u64 min_size,
8639                                     loff_t actual_len, u64 *alloc_hint)
8640 {
8641         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8642                                            min_size, actual_len, alloc_hint, trans);
8643 }
8644
8645 static int btrfs_set_page_dirty(struct page *page)
8646 {
8647         return __set_page_dirty_nobuffers(page);
8648 }
8649
8650 static int btrfs_permission(struct inode *inode, int mask)
8651 {
8652         struct btrfs_root *root = BTRFS_I(inode)->root;
8653         umode_t mode = inode->i_mode;
8654
8655         if (mask & MAY_WRITE &&
8656             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8657                 if (btrfs_root_readonly(root))
8658                         return -EROFS;
8659                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8660                         return -EACCES;
8661         }
8662         return generic_permission(inode, mask);
8663 }
8664
8665 static const struct inode_operations btrfs_dir_inode_operations = {
8666         .getattr        = btrfs_getattr,
8667         .lookup         = btrfs_lookup,
8668         .create         = btrfs_create,
8669         .unlink         = btrfs_unlink,
8670         .link           = btrfs_link,
8671         .mkdir          = btrfs_mkdir,
8672         .rmdir          = btrfs_rmdir,
8673         .rename         = btrfs_rename,
8674         .symlink        = btrfs_symlink,
8675         .setattr        = btrfs_setattr,
8676         .mknod          = btrfs_mknod,
8677         .setxattr       = btrfs_setxattr,
8678         .getxattr       = btrfs_getxattr,
8679         .listxattr      = btrfs_listxattr,
8680         .removexattr    = btrfs_removexattr,
8681         .permission     = btrfs_permission,
8682         .get_acl        = btrfs_get_acl,
8683 };
8684 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8685         .lookup         = btrfs_lookup,
8686         .permission     = btrfs_permission,
8687         .get_acl        = btrfs_get_acl,
8688 };
8689
8690 static const struct file_operations btrfs_dir_file_operations = {
8691         .llseek         = generic_file_llseek,
8692         .read           = generic_read_dir,
8693         .readdir        = btrfs_real_readdir,
8694         .unlocked_ioctl = btrfs_ioctl,
8695 #ifdef CONFIG_COMPAT
8696         .compat_ioctl   = btrfs_ioctl,
8697 #endif
8698         .release        = btrfs_release_file,
8699         .fsync          = btrfs_sync_file,
8700 };
8701
8702 static struct extent_io_ops btrfs_extent_io_ops = {
8703         .fill_delalloc = run_delalloc_range,
8704         .submit_bio_hook = btrfs_submit_bio_hook,
8705         .merge_bio_hook = btrfs_merge_bio_hook,
8706         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
8707         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
8708         .writepage_start_hook = btrfs_writepage_start_hook,
8709         .set_bit_hook = btrfs_set_bit_hook,
8710         .clear_bit_hook = btrfs_clear_bit_hook,
8711         .merge_extent_hook = btrfs_merge_extent_hook,
8712         .split_extent_hook = btrfs_split_extent_hook,
8713 };
8714
8715 /*
8716  * btrfs doesn't support the bmap operation because swapfiles
8717  * use bmap to make a mapping of extents in the file.  They assume
8718  * these extents won't change over the life of the file and they
8719  * use the bmap result to do IO directly to the drive.
8720  *
8721  * the btrfs bmap call would return logical addresses that aren't
8722  * suitable for IO and they also will change frequently as COW
8723  * operations happen.  So, swapfile + btrfs == corruption.
8724  *
8725  * For now we're avoiding this by dropping bmap.
8726  */
8727 static const struct address_space_operations btrfs_aops = {
8728         .readpage       = btrfs_readpage,
8729         .writepage      = btrfs_writepage,
8730         .writepages     = btrfs_writepages,
8731         .readpages      = btrfs_readpages,
8732         .direct_IO      = btrfs_direct_IO,
8733         .invalidatepage = btrfs_invalidatepage,
8734         .releasepage    = btrfs_releasepage,
8735         .set_page_dirty = btrfs_set_page_dirty,
8736         .error_remove_page = generic_error_remove_page,
8737 };
8738
8739 static const struct address_space_operations btrfs_symlink_aops = {
8740         .readpage       = btrfs_readpage,
8741         .writepage      = btrfs_writepage,
8742         .invalidatepage = btrfs_invalidatepage,
8743         .releasepage    = btrfs_releasepage,
8744 };
8745
8746 static const struct inode_operations btrfs_file_inode_operations = {
8747         .getattr        = btrfs_getattr,
8748         .setattr        = btrfs_setattr,
8749         .setxattr       = btrfs_setxattr,
8750         .getxattr       = btrfs_getxattr,
8751         .listxattr      = btrfs_listxattr,
8752         .removexattr    = btrfs_removexattr,
8753         .permission     = btrfs_permission,
8754         .fiemap         = btrfs_fiemap,
8755         .get_acl        = btrfs_get_acl,
8756         .update_time    = btrfs_update_time,
8757 };
8758 static const struct inode_operations btrfs_special_inode_operations = {
8759         .getattr        = btrfs_getattr,
8760         .setattr        = btrfs_setattr,
8761         .permission     = btrfs_permission,
8762         .setxattr       = btrfs_setxattr,
8763         .getxattr       = btrfs_getxattr,
8764         .listxattr      = btrfs_listxattr,
8765         .removexattr    = btrfs_removexattr,
8766         .get_acl        = btrfs_get_acl,
8767         .update_time    = btrfs_update_time,
8768 };
8769 static const struct inode_operations btrfs_symlink_inode_operations = {
8770         .readlink       = generic_readlink,
8771         .follow_link    = page_follow_link_light,
8772         .put_link       = page_put_link,
8773         .getattr        = btrfs_getattr,
8774         .setattr        = btrfs_setattr,
8775         .permission     = btrfs_permission,
8776         .setxattr       = btrfs_setxattr,
8777         .getxattr       = btrfs_getxattr,
8778         .listxattr      = btrfs_listxattr,
8779         .removexattr    = btrfs_removexattr,
8780         .get_acl        = btrfs_get_acl,
8781         .update_time    = btrfs_update_time,
8782 };
8783
8784 const struct dentry_operations btrfs_dentry_operations = {
8785         .d_delete       = btrfs_dentry_delete,
8786         .d_release      = btrfs_dentry_release,
8787 };