fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include <linux/ratelimit.h>
  41 #include "compat.h"
  42 #include "ctree.h"
  43 #include "disk-io.h"
  44 #include "transaction.h"
  45 #include "btrfs_inode.h"
  46 #include "ioctl.h"
  47 #include "print-tree.h"
  48 #include "volumes.h"
  49 #include "ordered-data.h"
  50 #include "xattr.h"
  51 #include "tree-log.h"
  52 #include "compression.h"
  53 #include "locking.h"
  54 #include "free-space-cache.h"
  55 #include "inode-map.h"
  56
  57 struct btrfs_iget_args {
  58         u64 ino;
  59         struct btrfs_root *root;
  60 };
  61
  62 static const struct inode_operations btrfs_dir_inode_operations;
  63 static const struct inode_operations btrfs_symlink_inode_operations;
  64 static const struct inode_operations btrfs_dir_ro_inode_operations;
  65 static const struct inode_operations btrfs_special_inode_operations;
  66 static const struct inode_operations btrfs_file_inode_operations;
  67 static const struct address_space_operations btrfs_aops;
  68 static const struct address_space_operations btrfs_symlink_aops;
  69 static const struct file_operations btrfs_dir_file_operations;
  70 static struct extent_io_ops btrfs_extent_io_ops;
  71
  72 static struct kmem_cache *btrfs_inode_cachep;
  73 struct kmem_cache *btrfs_trans_handle_cachep;
  74 struct kmem_cache *btrfs_transaction_cachep;
  75 struct kmem_cache *btrfs_path_cachep;
  76 struct kmem_cache *btrfs_free_space_cachep;
  77
  78 #define S_SHIFT 12
  79 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  80         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  81         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  82         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  83         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  84         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  85         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  86         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  87 };
  88
  89 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  90 static int btrfs_truncate(struct inode *inode);
  91 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  92 static noinline int cow_file_range(struct inode *inode,
  93                                    struct page *locked_page,
  94                                    u64 start, u64 end, int *page_started,
  95                                    unsigned long *nr_written, int unlock);
  96
  97 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
  98                                      struct inode *inode,  struct inode *dir,
  99                                      const struct qstr *qstr)
 100 {
 101         int err;
 102
 103         err = btrfs_init_acl(trans, inode, dir);
 104         if (!err)
 105                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 106         return err;
 107 }
 108
 109 /*
 110  * this does all the hard work for inserting an inline extent into
 111  * the btree.  The caller should have done a btrfs_drop_extents so that
 112  * no overlapping inline items exist in the btree
 113  */
 114 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 115                                 struct btrfs_root *root, struct inode *inode,
 116                                 u64 start, size_t size, size_t compressed_size,
 117                                 int compress_type,
 118                                 struct page **compressed_pages)
 119 {
 120         struct btrfs_key key;
 121         struct btrfs_path *path;
 122         struct extent_buffer *leaf;
 123         struct page *page = NULL;
 124         char *kaddr;
 125         unsigned long ptr;
 126         struct btrfs_file_extent_item *ei;
 127         int err = 0;
 128         int ret;
 129         size_t cur_size = size;
 130         size_t datasize;
 131         unsigned long offset;
 132
 133         if (compressed_size && compressed_pages)
 134                 cur_size = compressed_size;
 135
 136         path = btrfs_alloc_path();
 137         if (!path)
 138                 return -ENOMEM;
 139
 140         path->leave_spinning = 1;
 141
 142         key.objectid = btrfs_ino(inode);
 143         key.offset = start;
 144         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 145         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 146
 147         inode_add_bytes(inode, size);
 148         ret = btrfs_insert_empty_item(trans, root, path, &key,
 149                                       datasize);
 150         BUG_ON(ret);
 151         if (ret) {
 152                 err = ret;
 153                 goto fail;
 154         }
 155         leaf = path->nodes[0];
 156         ei = btrfs_item_ptr(leaf, path->slots[0],
 157                             struct btrfs_file_extent_item);
 158         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 159         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 160         btrfs_set_file_extent_encryption(leaf, ei, 0);
 161         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 162         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 163         ptr = btrfs_file_extent_inline_start(ei);
 164
 165         if (compress_type != BTRFS_COMPRESS_NONE) {
 166                 struct page *cpage;
 167                 int i = 0;
 168                 while (compressed_size > 0) {
 169                         cpage = compressed_pages[i];
 170                         cur_size = min_t(unsigned long, compressed_size,
 171                                        PAGE_CACHE_SIZE);
 172
 173                         kaddr = kmap_atomic(cpage, KM_USER0);
 174                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 175                         kunmap_atomic(kaddr, KM_USER0);
 176
 177                         i++;
 178                         ptr += cur_size;
 179                         compressed_size -= cur_size;
 180                 }
 181                 btrfs_set_file_extent_compression(leaf, ei,
 182                                                   compress_type);
 183         } else {
 184                 page = find_get_page(inode->i_mapping,
 185                                      start >> PAGE_CACHE_SHIFT);
 186                 btrfs_set_file_extent_compression(leaf, ei, 0);
 187                 kaddr = kmap_atomic(page, KM_USER0);
 188                 offset = start & (PAGE_CACHE_SIZE - 1);
 189                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 190                 kunmap_atomic(kaddr, KM_USER0);
 191                 page_cache_release(page);
 192         }
 193         btrfs_mark_buffer_dirty(leaf);
 194         btrfs_free_path(path);
 195
 196         /*
 197          * we're an inline extent, so nobody can
 198          * extend the file past i_size without locking
 199          * a page we already have locked.
 200          *
 201          * We must do any isize and inode updates
 202          * before we unlock the pages.  Otherwise we
 203          * could end up racing with unlink.
 204          */
 205         BTRFS_I(inode)->disk_i_size = inode->i_size;
 206         btrfs_update_inode(trans, root, inode);
 207
 208         return 0;
 209 fail:
 210         btrfs_free_path(path);
 211         return err;
 212 }
 213
 214
 215 /*
 216  * conditionally insert an inline extent into the file.  This
 217  * does the checks required to make sure the data is small enough
 218  * to fit as an inline extent.
 219  */
 220 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 221                                  struct btrfs_root *root,
 222                                  struct inode *inode, u64 start, u64 end,
 223                                  size_t compressed_size, int compress_type,
 224                                  struct page **compressed_pages)
 225 {
 226         u64 isize = i_size_read(inode);
 227         u64 actual_end = min(end + 1, isize);
 228         u64 inline_len = actual_end - start;
 229         u64 aligned_end = (end + root->sectorsize - 1) &
 230                         ~((u64)root->sectorsize - 1);
 231         u64 hint_byte;
 232         u64 data_len = inline_len;
 233         int ret;
 234
 235         if (compressed_size)
 236                 data_len = compressed_size;
 237
 238         if (start > 0 ||
 239             actual_end >= PAGE_CACHE_SIZE ||
 240             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 241             (!compressed_size &&
 242             (actual_end & (root->sectorsize - 1)) == 0) ||
 243             end + 1 < isize ||
 244             data_len > root->fs_info->max_inline) {
 245                 return 1;
 246         }
 247
 248         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
 249                                  &hint_byte, 1);
 250         BUG_ON(ret);
 251
 252         if (isize > actual_end)
 253                 inline_len = min_t(u64, isize, actual_end);
 254         ret = insert_inline_extent(trans, root, inode, start,
 255                                    inline_len, compressed_size,
 256                                    compress_type, compressed_pages);
 257         BUG_ON(ret);
 258         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 259         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 260         return 0;
 261 }
 262
 263 struct async_extent {
 264         u64 start;
 265         u64 ram_size;
 266         u64 compressed_size;
 267         struct page **pages;
 268         unsigned long nr_pages;
 269         int compress_type;
 270         struct list_head list;
 271 };
 272
 273 struct async_cow {
 274         struct inode *inode;
 275         struct btrfs_root *root;
 276         struct page *locked_page;
 277         u64 start;
 278         u64 end;
 279         struct list_head extents;
 280         struct btrfs_work work;
 281 };
 282
 283 static noinline int add_async_extent(struct async_cow *cow,
 284                                      u64 start, u64 ram_size,
 285                                      u64 compressed_size,
 286                                      struct page **pages,
 287                                      unsigned long nr_pages,
 288                                      int compress_type)
 289 {
 290         struct async_extent *async_extent;
 291
 292         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 293         BUG_ON(!async_extent);
 294         async_extent->start = start;
 295         async_extent->ram_size = ram_size;
 296         async_extent->compressed_size = compressed_size;
 297         async_extent->pages = pages;
 298         async_extent->nr_pages = nr_pages;
 299         async_extent->compress_type = compress_type;
 300         list_add_tail(&async_extent->list, &cow->extents);
 301         return 0;
 302 }
 303
 304 /*
 305  * we create compressed extents in two phases.  The first
 306  * phase compresses a range of pages that have already been
 307  * locked (both pages and state bits are locked).
 308  *
 309  * This is done inside an ordered work queue, and the compression
 310  * is spread across many cpus.  The actual IO submission is step
 311  * two, and the ordered work queue takes care of making sure that
 312  * happens in the same order things were put onto the queue by
 313  * writepages and friends.
 314  *
 315  * If this code finds it can't get good compression, it puts an
 316  * entry onto the work queue to write the uncompressed bytes.  This
 317  * makes sure that both compressed inodes and uncompressed inodes
 318  * are written in the same order that pdflush sent them down.
 319  */
 320 static noinline int compress_file_range(struct inode *inode,
 321                                         struct page *locked_page,
 322                                         u64 start, u64 end,
 323                                         struct async_cow *async_cow,
 324                                         int *num_added)
 325 {
 326         struct btrfs_root *root = BTRFS_I(inode)->root;
 327         struct btrfs_trans_handle *trans;
 328         u64 num_bytes;
 329         u64 blocksize = root->sectorsize;
 330         u64 actual_end;
 331         u64 isize = i_size_read(inode);
 332         int ret = 0;
 333         struct page **pages = NULL;
 334         unsigned long nr_pages;
 335         unsigned long nr_pages_ret = 0;
 336         unsigned long total_compressed = 0;
 337         unsigned long total_in = 0;
 338         unsigned long max_compressed = 128 * 1024;
 339         unsigned long max_uncompressed = 128 * 1024;
 340         int i;
 341         int will_compress;
 342         int compress_type = root->fs_info->compress_type;
 343
 344         /* if this is a small write inside eof, kick off a defragbot */
 345         if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024)
 346                 btrfs_add_inode_defrag(NULL, inode);
 347
 348         actual_end = min_t(u64, isize, end + 1);
 349 again:
 350         will_compress = 0;
 351         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 352         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 353
 354         /*
 355          * we don't want to send crud past the end of i_size through
 356          * compression, that's just a waste of CPU time.  So, if the
 357          * end of the file is before the start of our current
 358          * requested range of bytes, we bail out to the uncompressed
 359          * cleanup code that can deal with all of this.
 360          *
 361          * It isn't really the fastest way to fix things, but this is a
 362          * very uncommon corner.
 363          */
 364         if (actual_end <= start)
 365                 goto cleanup_and_bail_uncompressed;
 366
 367         total_compressed = actual_end - start;
 368
 369         /* we want to make sure that amount of ram required to uncompress
 370          * an extent is reasonable, so we limit the total size in ram
 371          * of a compressed extent to 128k.  This is a crucial number
 372          * because it also controls how easily we can spread reads across
 373          * cpus for decompression.
 374          *
 375          * We also want to make sure the amount of IO required to do
 376          * a random read is reasonably small, so we limit the size of
 377          * a compressed extent to 128k.
 378          */
 379         total_compressed = min(total_compressed, max_uncompressed);
 380         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 381         num_bytes = max(blocksize,  num_bytes);
 382         total_in = 0;
 383         ret = 0;
 384
 385         /*
 386          * we do compression for mount -o compress and when the
 387          * inode has not been flagged as nocompress.  This flag can
 388          * change at any time if we discover bad compression ratios.
 389          */
 390         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 391             (btrfs_test_opt(root, COMPRESS) ||
 392              (BTRFS_I(inode)->force_compress) ||
 393              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 394                 WARN_ON(pages);
 395                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 396                 BUG_ON(!pages);
 397
 398                 if (BTRFS_I(inode)->force_compress)
 399                         compress_type = BTRFS_I(inode)->force_compress;
 400
 401                 ret = btrfs_compress_pages(compress_type,
 402                                            inode->i_mapping, start,
 403                                            total_compressed, pages,
 404                                            nr_pages, &nr_pages_ret,
 405                                            &total_in,
 406                                            &total_compressed,
 407                                            max_compressed);
 408
 409                 if (!ret) {
 410                         unsigned long offset = total_compressed &
 411                                 (PAGE_CACHE_SIZE - 1);
 412                         struct page *page = pages[nr_pages_ret - 1];
 413                         char *kaddr;
 414
 415                         /* zero the tail end of the last page, we might be
 416                          * sending it down to disk
 417                          */
 418                         if (offset) {
 419                                 kaddr = kmap_atomic(page, KM_USER0);
 420                                 memset(kaddr + offset, 0,
 421                                        PAGE_CACHE_SIZE - offset);
 422                                 kunmap_atomic(kaddr, KM_USER0);
 423                         }
 424                         will_compress = 1;
 425                 }
 426         }
 427         if (start == 0) {
 428                 trans = btrfs_join_transaction(root);
 429                 BUG_ON(IS_ERR(trans));
 430                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 431
 432                 /* lets try to make an inline extent */
 433                 if (ret || total_in < (actual_end - start)) {
 434                         /* we didn't compress the entire range, try
 435                          * to make an uncompressed inline extent.
 436                          */
 437                         ret = cow_file_range_inline(trans, root, inode,
 438                                                     start, end, 0, 0, NULL);
 439                 } else {
 440                         /* try making a compressed inline extent */
 441                         ret = cow_file_range_inline(trans, root, inode,
 442                                                     start, end,
 443                                                     total_compressed,
 444                                                     compress_type, pages);
 445                 }
 446                 if (ret == 0) {
 447                         /*
 448                          * inline extent creation worked, we don't need
 449                          * to create any more async work items.  Unlock
 450                          * and free up our temp pages.
 451                          */
 452                         extent_clear_unlock_delalloc(inode,
 453                              &BTRFS_I(inode)->io_tree,
 454                              start, end, NULL,
 455                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 456                              EXTENT_CLEAR_DELALLOC |
 457                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 458
 459                         btrfs_end_transaction(trans, root);
 460                         goto free_pages_out;
 461                 }
 462                 btrfs_end_transaction(trans, root);
 463         }
 464
 465         if (will_compress) {
 466                 /*
 467                  * we aren't doing an inline extent round the compressed size
 468                  * up to a block size boundary so the allocator does sane
 469                  * things
 470                  */
 471                 total_compressed = (total_compressed + blocksize - 1) &
 472                         ~(blocksize - 1);
 473
 474                 /*
 475                  * one last check to make sure the compression is really a
 476                  * win, compare the page count read with the blocks on disk
 477                  */
 478                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 479                         ~(PAGE_CACHE_SIZE - 1);
 480                 if (total_compressed >= total_in) {
 481                         will_compress = 0;
 482                 } else {
 483                         num_bytes = total_in;
 484                 }
 485         }
 486         if (!will_compress && pages) {
 487                 /*
 488                  * the compression code ran but failed to make things smaller,
 489                  * free any pages it allocated and our page pointer array
 490                  */
 491                 for (i = 0; i < nr_pages_ret; i++) {
 492                         WARN_ON(pages[i]->mapping);
 493                         page_cache_release(pages[i]);
 494                 }
 495                 kfree(pages);
 496                 pages = NULL;
 497                 total_compressed = 0;
 498                 nr_pages_ret = 0;
 499
 500                 /* flag the file so we don't compress in the future */
 501                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 502                     !(BTRFS_I(inode)->force_compress)) {
 503                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 504                 }
 505         }
 506         if (will_compress) {
 507                 *num_added += 1;
 508
 509                 /* the async work queues will take care of doing actual
 510                  * allocation on disk for these compressed pages,
 511                  * and will submit them to the elevator.
 512                  */
 513                 add_async_extent(async_cow, start, num_bytes,
 514                                  total_compressed, pages, nr_pages_ret,
 515                                  compress_type);
 516
 517                 if (start + num_bytes < end) {
 518                         start += num_bytes;
 519                         pages = NULL;
 520                         cond_resched();
 521                         goto again;
 522                 }
 523         } else {
 524 cleanup_and_bail_uncompressed:
 525                 /*
 526                  * No compression, but we still need to write the pages in
 527                  * the file we've been given so far.  redirty the locked
 528                  * page if it corresponds to our extent and set things up
 529                  * for the async work queue to run cow_file_range to do
 530                  * the normal delalloc dance
 531                  */
 532                 if (page_offset(locked_page) >= start &&
 533                     page_offset(locked_page) <= end) {
 534                         __set_page_dirty_nobuffers(locked_page);
 535                         /* unlocked later on in the async handlers */
 536                 }
 537                 add_async_extent(async_cow, start, end - start + 1,
 538                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 539                 *num_added += 1;
 540         }
 541
 542 out:
 543         return 0;
 544
 545 free_pages_out:
 546         for (i = 0; i < nr_pages_ret; i++) {
 547                 WARN_ON(pages[i]->mapping);
 548                 page_cache_release(pages[i]);
 549         }
 550         kfree(pages);
 551
 552         goto out;
 553 }
 554
 555 /*
 556  * phase two of compressed writeback.  This is the ordered portion
 557  * of the code, which only gets called in the order the work was
 558  * queued.  We walk all the async extents created by compress_file_range
 559  * and send them down to the disk.
 560  */
 561 static noinline int submit_compressed_extents(struct inode *inode,
 562                                               struct async_cow *async_cow)
 563 {
 564         struct async_extent *async_extent;
 565         u64 alloc_hint = 0;
 566         struct btrfs_trans_handle *trans;
 567         struct btrfs_key ins;
 568         struct extent_map *em;
 569         struct btrfs_root *root = BTRFS_I(inode)->root;
 570         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 571         struct extent_io_tree *io_tree;
 572         int ret = 0;
 573
 574         if (list_empty(&async_cow->extents))
 575                 return 0;
 576
 577
 578         while (!list_empty(&async_cow->extents)) {
 579                 async_extent = list_entry(async_cow->extents.next,
 580                                           struct async_extent, list);
 581                 list_del(&async_extent->list);
 582
 583                 io_tree = &BTRFS_I(inode)->io_tree;
 584
 585 retry:
 586                 /* did the compression code fall back to uncompressed IO? */
 587                 if (!async_extent->pages) {
 588                         int page_started = 0;
 589                         unsigned long nr_written = 0;
 590
 591                         lock_extent(io_tree, async_extent->start,
 592                                          async_extent->start +
 593                                          async_extent->ram_size - 1, GFP_NOFS);
 594
 595                         /* allocate blocks */
 596                         ret = cow_file_range(inode, async_cow->locked_page,
 597                                              async_extent->start,
 598                                              async_extent->start +
 599                                              async_extent->ram_size - 1,
 600                                              &page_started, &nr_written, 0);
 601
 602                         /*
 603                          * if page_started, cow_file_range inserted an
 604                          * inline extent and took care of all the unlocking
 605                          * and IO for us.  Otherwise, we need to submit
 606                          * all those pages down to the drive.
 607                          */
 608                         if (!page_started && !ret)
 609                                 extent_write_locked_range(io_tree,
 610                                                   inode, async_extent->start,
 611                                                   async_extent->start +
 612                                                   async_extent->ram_size - 1,
 613                                                   btrfs_get_extent,
 614                                                   WB_SYNC_ALL);
 615                         kfree(async_extent);
 616                         cond_resched();
 617                         continue;
 618                 }
 619
 620                 lock_extent(io_tree, async_extent->start,
 621                             async_extent->start + async_extent->ram_size - 1,
 622                             GFP_NOFS);
 623
 624                 trans = btrfs_join_transaction(root);
 625                 BUG_ON(IS_ERR(trans));
 626                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 627                 ret = btrfs_reserve_extent(trans, root,
 628                                            async_extent->compressed_size,
 629                                            async_extent->compressed_size,
 630                                            0, alloc_hint,
 631                                            (u64)-1, &ins, 1);
 632                 btrfs_end_transaction(trans, root);
 633
 634                 if (ret) {
 635                         int i;
 636                         for (i = 0; i < async_extent->nr_pages; i++) {
 637                                 WARN_ON(async_extent->pages[i]->mapping);
 638                                 page_cache_release(async_extent->pages[i]);
 639                         }
 640                         kfree(async_extent->pages);
 641                         async_extent->nr_pages = 0;
 642                         async_extent->pages = NULL;
 643                         unlock_extent(io_tree, async_extent->start,
 644                                       async_extent->start +
 645                                       async_extent->ram_size - 1, GFP_NOFS);
 646                         goto retry;
 647                 }
 648
 649                 /*
 650                  * here we're doing allocation and writeback of the
 651                  * compressed pages
 652                  */
 653                 btrfs_drop_extent_cache(inode, async_extent->start,
 654                                         async_extent->start +
 655                                         async_extent->ram_size - 1, 0);
 656
 657                 em = alloc_extent_map();
 658                 BUG_ON(!em);
 659                 em->start = async_extent->start;
 660                 em->len = async_extent->ram_size;
 661                 em->orig_start = em->start;
 662
 663                 em->block_start = ins.objectid;
 664                 em->block_len = ins.offset;
 665                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 666                 em->compress_type = async_extent->compress_type;
 667                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 668                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 669
 670                 while (1) {
 671                         write_lock(&em_tree->lock);
 672                         ret = add_extent_mapping(em_tree, em);
 673                         write_unlock(&em_tree->lock);
 674                         if (ret != -EEXIST) {
 675                                 free_extent_map(em);
 676                                 break;
 677                         }
 678                         btrfs_drop_extent_cache(inode, async_extent->start,
 679                                                 async_extent->start +
 680                                                 async_extent->ram_size - 1, 0);
 681                 }
 682
 683                 ret = btrfs_add_ordered_extent_compress(inode,
 684                                                 async_extent->start,
 685                                                 ins.objectid,
 686                                                 async_extent->ram_size,
 687                                                 ins.offset,
 688                                                 BTRFS_ORDERED_COMPRESSED,
 689                                                 async_extent->compress_type);
 690                 BUG_ON(ret);
 691
 692                 /*
 693                  * clear dirty, set writeback and unlock the pages.
 694                  */
 695                 extent_clear_unlock_delalloc(inode,
 696                                 &BTRFS_I(inode)->io_tree,
 697                                 async_extent->start,
 698                                 async_extent->start +
 699                                 async_extent->ram_size - 1,
 700                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 701                                 EXTENT_CLEAR_UNLOCK |
 702                                 EXTENT_CLEAR_DELALLOC |
 703                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 704
 705                 ret = btrfs_submit_compressed_write(inode,
 706                                     async_extent->start,
 707                                     async_extent->ram_size,
 708                                     ins.objectid,
 709                                     ins.offset, async_extent->pages,
 710                                     async_extent->nr_pages);
 711
 712                 BUG_ON(ret);
 713                 alloc_hint = ins.objectid + ins.offset;
 714                 kfree(async_extent);
 715                 cond_resched();
 716         }
 717
 718         return 0;
 719 }
 720
 721 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 722                                       u64 num_bytes)
 723 {
 724         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 725         struct extent_map *em;
 726         u64 alloc_hint = 0;
 727
 728         read_lock(&em_tree->lock);
 729         em = search_extent_mapping(em_tree, start, num_bytes);
 730         if (em) {
 731                 /*
 732                  * if block start isn't an actual block number then find the
 733                  * first block in this inode and use that as a hint.  If that
 734                  * block is also bogus then just don't worry about it.
 735                  */
 736                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 737                         free_extent_map(em);
 738                         em = search_extent_mapping(em_tree, 0, 0);
 739                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 740                                 alloc_hint = em->block_start;
 741                         if (em)
 742                                 free_extent_map(em);
 743                 } else {
 744                         alloc_hint = em->block_start;
 745                         free_extent_map(em);
 746                 }
 747         }
 748         read_unlock(&em_tree->lock);
 749
 750         return alloc_hint;
 751 }
 752
 753 static inline bool is_free_space_inode(struct btrfs_root *root,
 754                                        struct inode *inode)
 755 {
 756         if (root == root->fs_info->tree_root ||
 757             BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID)
 758                 return true;
 759         return false;
 760 }
 761
 762 /*
 763  * when extent_io.c finds a delayed allocation range in the file,
 764  * the call backs end up in this code.  The basic idea is to
 765  * allocate extents on disk for the range, and create ordered data structs
 766  * in ram to track those extents.
 767  *
 768  * locked_page is the page that writepage had locked already.  We use
 769  * it to make sure we don't do extra locks or unlocks.
 770  *
 771  * *page_started is set to one if we unlock locked_page and do everything
 772  * required to start IO on it.  It may be clean and already done with
 773  * IO when we return.
 774  */
 775 static noinline int cow_file_range(struct inode *inode,
 776                                    struct page *locked_page,
 777                                    u64 start, u64 end, int *page_started,
 778                                    unsigned long *nr_written,
 779                                    int unlock)
 780 {
 781         struct btrfs_root *root = BTRFS_I(inode)->root;
 782         struct btrfs_trans_handle *trans;
 783         u64 alloc_hint = 0;
 784         u64 num_bytes;
 785         unsigned long ram_size;
 786         u64 disk_num_bytes;
 787         u64 cur_alloc_size;
 788         u64 blocksize = root->sectorsize;
 789         struct btrfs_key ins;
 790         struct extent_map *em;
 791         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 792         int ret = 0;
 793
 794         BUG_ON(is_free_space_inode(root, inode));
 795         trans = btrfs_join_transaction(root);
 796         BUG_ON(IS_ERR(trans));
 797         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 798
 799         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 800         num_bytes = max(blocksize,  num_bytes);
 801         disk_num_bytes = num_bytes;
 802         ret = 0;
 803
 804         /* if this is a small write inside eof, kick off defrag */
 805         if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
 806                 btrfs_add_inode_defrag(trans, inode);
 807
 808         if (start == 0) {
 809                 /* lets try to make an inline extent */
 810                 ret = cow_file_range_inline(trans, root, inode,
 811                                             start, end, 0, 0, NULL);
 812                 if (ret == 0) {
 813                         extent_clear_unlock_delalloc(inode,
 814                                      &BTRFS_I(inode)->io_tree,
 815                                      start, end, NULL,
 816                                      EXTENT_CLEAR_UNLOCK_PAGE |
 817                                      EXTENT_CLEAR_UNLOCK |
 818                                      EXTENT_CLEAR_DELALLOC |
 819                                      EXTENT_CLEAR_DIRTY |
 820                                      EXTENT_SET_WRITEBACK |
 821                                      EXTENT_END_WRITEBACK);
 822
 823                         *nr_written = *nr_written +
 824                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 825                         *page_started = 1;
 826                         ret = 0;
 827                         goto out;
 828                 }
 829         }
 830
 831         BUG_ON(disk_num_bytes >
 832                btrfs_super_total_bytes(&root->fs_info->super_copy));
 833
 834         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 835         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 836
 837         while (disk_num_bytes > 0) {
 838                 unsigned long op;
 839
 840                 cur_alloc_size = disk_num_bytes;
 841                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 842                                            root->sectorsize, 0, alloc_hint,
 843                                            (u64)-1, &ins, 1);
 844                 BUG_ON(ret);
 845
 846                 em = alloc_extent_map();
 847                 BUG_ON(!em);
 848                 em->start = start;
 849                 em->orig_start = em->start;
 850                 ram_size = ins.offset;
 851                 em->len = ins.offset;
 852
 853                 em->block_start = ins.objectid;
 854                 em->block_len = ins.offset;
 855                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 856                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 857
 858                 while (1) {
 859                         write_lock(&em_tree->lock);
 860                         ret = add_extent_mapping(em_tree, em);
 861                         write_unlock(&em_tree->lock);
 862                         if (ret != -EEXIST) {
 863                                 free_extent_map(em);
 864                                 break;
 865                         }
 866                         btrfs_drop_extent_cache(inode, start,
 867                                                 start + ram_size - 1, 0);
 868                 }
 869
 870                 cur_alloc_size = ins.offset;
 871                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 872                                                ram_size, cur_alloc_size, 0);
 873                 BUG_ON(ret);
 874
 875                 if (root->root_key.objectid ==
 876                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 877                         ret = btrfs_reloc_clone_csums(inode, start,
 878                                                       cur_alloc_size);
 879                         BUG_ON(ret);
 880                 }
 881
 882                 if (disk_num_bytes < cur_alloc_size)
 883                         break;
 884
 885                 /* we're not doing compressed IO, don't unlock the first
 886                  * page (which the caller expects to stay locked), don't
 887                  * clear any dirty bits and don't set any writeback bits
 888                  *
 889                  * Do set the Private2 bit so we know this page was properly
 890                  * setup for writepage
 891                  */
 892                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 893                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 894                         EXTENT_SET_PRIVATE2;
 895
 896                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 897                                              start, start + ram_size - 1,
 898                                              locked_page, op);
 899                 disk_num_bytes -= cur_alloc_size;
 900                 num_bytes -= cur_alloc_size;
 901                 alloc_hint = ins.objectid + ins.offset;
 902                 start += cur_alloc_size;
 903         }
 904 out:
 905         ret = 0;
 906         btrfs_end_transaction(trans, root);
 907
 908         return ret;
 909 }
 910
 911 /*
 912  * work queue call back to started compression on a file and pages
 913  */
 914 static noinline void async_cow_start(struct btrfs_work *work)
 915 {
 916         struct async_cow *async_cow;
 917         int num_added = 0;
 918         async_cow = container_of(work, struct async_cow, work);
 919
 920         compress_file_range(async_cow->inode, async_cow->locked_page,
 921                             async_cow->start, async_cow->end, async_cow,
 922                             &num_added);
 923         if (num_added == 0)
 924                 async_cow->inode = NULL;
 925 }
 926
 927 /*
 928  * work queue call back to submit previously compressed pages
 929  */
 930 static noinline void async_cow_submit(struct btrfs_work *work)
 931 {
 932         struct async_cow *async_cow;
 933         struct btrfs_root *root;
 934         unsigned long nr_pages;
 935
 936         async_cow = container_of(work, struct async_cow, work);
 937
 938         root = async_cow->root;
 939         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 940                 PAGE_CACHE_SHIFT;
 941
 942         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 943
 944         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 945             5 * 1042 * 1024 &&
 946             waitqueue_active(&root->fs_info->async_submit_wait))
 947                 wake_up(&root->fs_info->async_submit_wait);
 948
 949         if (async_cow->inode)
 950                 submit_compressed_extents(async_cow->inode, async_cow);
 951 }
 952
 953 static noinline void async_cow_free(struct btrfs_work *work)
 954 {
 955         struct async_cow *async_cow;
 956         async_cow = container_of(work, struct async_cow, work);
 957         kfree(async_cow);
 958 }
 959
 960 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 961                                 u64 start, u64 end, int *page_started,
 962                                 unsigned long *nr_written)
 963 {
 964         struct async_cow *async_cow;
 965         struct btrfs_root *root = BTRFS_I(inode)->root;
 966         unsigned long nr_pages;
 967         u64 cur_end;
 968         int limit = 10 * 1024 * 1042;
 969
 970         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
 971                          1, 0, NULL, GFP_NOFS);
 972         while (start < end) {
 973                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 974                 BUG_ON(!async_cow);
 975                 async_cow->inode = inode;
 976                 async_cow->root = root;
 977                 async_cow->locked_page = locked_page;
 978                 async_cow->start = start;
 979
 980                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 981                         cur_end = end;
 982                 else
 983                         cur_end = min(end, start + 512 * 1024 - 1);
 984
 985                 async_cow->end = cur_end;
 986                 INIT_LIST_HEAD(&async_cow->extents);
 987
 988                 async_cow->work.func = async_cow_start;
 989                 async_cow->work.ordered_func = async_cow_submit;
 990                 async_cow->work.ordered_free = async_cow_free;
 991                 async_cow->work.flags = 0;
 992
 993                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 994                         PAGE_CACHE_SHIFT;
 995                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 996
 997                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 998                                    &async_cow->work);
 999
1000                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1001                         wait_event(root->fs_info->async_submit_wait,
1002                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1003                             limit));
1004                 }
1005
1006                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1007                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1008                         wait_event(root->fs_info->async_submit_wait,
1009                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1010                            0));
1011                 }
1012
1013                 *nr_written += nr_pages;
1014                 start = cur_end + 1;
1015         }
1016         *page_started = 1;
1017         return 0;
1018 }
1019
1020 static noinline int csum_exist_in_range(struct btrfs_root *root,
1021                                         u64 bytenr, u64 num_bytes)
1022 {
1023         int ret;
1024         struct btrfs_ordered_sum *sums;
1025         LIST_HEAD(list);
1026
1027         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1028                                        bytenr + num_bytes - 1, &list, 0);
1029         if (ret == 0 && list_empty(&list))
1030                 return 0;
1031
1032         while (!list_empty(&list)) {
1033                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1034                 list_del(&sums->list);
1035                 kfree(sums);
1036         }
1037         return 1;
1038 }
1039
1040 /*
1041  * when nowcow writeback call back.  This checks for snapshots or COW copies
1042  * of the extents that exist in the file, and COWs the file as required.
1043  *
1044  * If no cow copies or snapshots exist, we write directly to the existing
1045  * blocks on disk
1046  */
1047 static noinline int run_delalloc_nocow(struct inode *inode,
1048                                        struct page *locked_page,
1049                               u64 start, u64 end, int *page_started, int force,
1050                               unsigned long *nr_written)
1051 {
1052         struct btrfs_root *root = BTRFS_I(inode)->root;
1053         struct btrfs_trans_handle *trans;
1054         struct extent_buffer *leaf;
1055         struct btrfs_path *path;
1056         struct btrfs_file_extent_item *fi;
1057         struct btrfs_key found_key;
1058         u64 cow_start;
1059         u64 cur_offset;
1060         u64 extent_end;
1061         u64 extent_offset;
1062         u64 disk_bytenr;
1063         u64 num_bytes;
1064         int extent_type;
1065         int ret;
1066         int type;
1067         int nocow;
1068         int check_prev = 1;
1069         bool nolock;
1070         u64 ino = btrfs_ino(inode);
1071
1072         path = btrfs_alloc_path();
1073         BUG_ON(!path);
1074
1075         nolock = is_free_space_inode(root, inode);
1076
1077         if (nolock)
1078                 trans = btrfs_join_transaction_nolock(root);
1079         else
1080                 trans = btrfs_join_transaction(root);
1081
1082         BUG_ON(IS_ERR(trans));
1083         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1084
1085         cow_start = (u64)-1;
1086         cur_offset = start;
1087         while (1) {
1088                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1089                                                cur_offset, 0);
1090                 BUG_ON(ret < 0);
1091                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1092                         leaf = path->nodes[0];
1093                         btrfs_item_key_to_cpu(leaf, &found_key,
1094                                               path->slots[0] - 1);
1095                         if (found_key.objectid == ino &&
1096                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1097                                 path->slots[0]--;
1098                 }
1099                 check_prev = 0;
1100 next_slot:
1101                 leaf = path->nodes[0];
1102                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1103                         ret = btrfs_next_leaf(root, path);
1104                         if (ret < 0)
1105                                 BUG_ON(1);
1106                         if (ret > 0)
1107                                 break;
1108                         leaf = path->nodes[0];
1109                 }
1110
1111                 nocow = 0;
1112                 disk_bytenr = 0;
1113                 num_bytes = 0;
1114                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1115
1116                 if (found_key.objectid > ino ||
1117                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1118                     found_key.offset > end)
1119                         break;
1120
1121                 if (found_key.offset > cur_offset) {
1122                         extent_end = found_key.offset;
1123                         extent_type = 0;
1124                         goto out_check;
1125                 }
1126
1127                 fi = btrfs_item_ptr(leaf, path->slots[0],
1128                                     struct btrfs_file_extent_item);
1129                 extent_type = btrfs_file_extent_type(leaf, fi);
1130
1131                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1132                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1133                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1134                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1135                         extent_end = found_key.offset +
1136                                 btrfs_file_extent_num_bytes(leaf, fi);
1137                         if (extent_end <= start) {
1138                                 path->slots[0]++;
1139                                 goto next_slot;
1140                         }
1141                         if (disk_bytenr == 0)
1142                                 goto out_check;
1143                         if (btrfs_file_extent_compression(leaf, fi) ||
1144                             btrfs_file_extent_encryption(leaf, fi) ||
1145                             btrfs_file_extent_other_encoding(leaf, fi))
1146                                 goto out_check;
1147                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1148                                 goto out_check;
1149                         if (btrfs_extent_readonly(root, disk_bytenr))
1150                                 goto out_check;
1151                         if (btrfs_cross_ref_exist(trans, root, ino,
1152                                                   found_key.offset -
1153                                                   extent_offset, disk_bytenr))
1154                                 goto out_check;
1155                         disk_bytenr += extent_offset;
1156                         disk_bytenr += cur_offset - found_key.offset;
1157                         num_bytes = min(end + 1, extent_end) - cur_offset;
1158                         /*
1159                          * force cow if csum exists in the range.
1160                          * this ensure that csum for a given extent are
1161                          * either valid or do not exist.
1162                          */
1163                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1164                                 goto out_check;
1165                         nocow = 1;
1166                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1167                         extent_end = found_key.offset +
1168                                 btrfs_file_extent_inline_len(leaf, fi);
1169                         extent_end = ALIGN(extent_end, root->sectorsize);
1170                 } else {
1171                         BUG_ON(1);
1172                 }
1173 out_check:
1174                 if (extent_end <= start) {
1175                         path->slots[0]++;
1176                         goto next_slot;
1177                 }
1178                 if (!nocow) {
1179                         if (cow_start == (u64)-1)
1180                                 cow_start = cur_offset;
1181                         cur_offset = extent_end;
1182                         if (cur_offset > end)
1183                                 break;
1184                         path->slots[0]++;
1185                         goto next_slot;
1186                 }
1187
1188                 btrfs_release_path(path);
1189                 if (cow_start != (u64)-1) {
1190                         ret = cow_file_range(inode, locked_page, cow_start,
1191                                         found_key.offset - 1, page_started,
1192                                         nr_written, 1);
1193                         BUG_ON(ret);
1194                         cow_start = (u64)-1;
1195                 }
1196
1197                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1198                         struct extent_map *em;
1199                         struct extent_map_tree *em_tree;
1200                         em_tree = &BTRFS_I(inode)->extent_tree;
1201                         em = alloc_extent_map();
1202                         BUG_ON(!em);
1203                         em->start = cur_offset;
1204                         em->orig_start = em->start;
1205                         em->len = num_bytes;
1206                         em->block_len = num_bytes;
1207                         em->block_start = disk_bytenr;
1208                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1209                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1210                         while (1) {
1211                                 write_lock(&em_tree->lock);
1212                                 ret = add_extent_mapping(em_tree, em);
1213                                 write_unlock(&em_tree->lock);
1214                                 if (ret != -EEXIST) {
1215                                         free_extent_map(em);
1216                                         break;
1217                                 }
1218                                 btrfs_drop_extent_cache(inode, em->start,
1219                                                 em->start + em->len - 1, 0);
1220                         }
1221                         type = BTRFS_ORDERED_PREALLOC;
1222                 } else {
1223                         type = BTRFS_ORDERED_NOCOW;
1224                 }
1225
1226                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1227                                                num_bytes, num_bytes, type);
1228                 BUG_ON(ret);
1229
1230                 if (root->root_key.objectid ==
1231                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1232                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1233                                                       num_bytes);
1234                         BUG_ON(ret);
1235                 }
1236
1237                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1238                                 cur_offset, cur_offset + num_bytes - 1,
1239                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1240                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1241                                 EXTENT_SET_PRIVATE2);
1242                 cur_offset = extent_end;
1243                 if (cur_offset > end)
1244                         break;
1245         }
1246         btrfs_release_path(path);
1247
1248         if (cur_offset <= end && cow_start == (u64)-1)
1249                 cow_start = cur_offset;
1250         if (cow_start != (u64)-1) {
1251                 ret = cow_file_range(inode, locked_page, cow_start, end,
1252                                      page_started, nr_written, 1);
1253                 BUG_ON(ret);
1254         }
1255
1256         if (nolock) {
1257                 ret = btrfs_end_transaction_nolock(trans, root);
1258                 BUG_ON(ret);
1259         } else {
1260                 ret = btrfs_end_transaction(trans, root);
1261                 BUG_ON(ret);
1262         }
1263         btrfs_free_path(path);
1264         return 0;
1265 }
1266
1267 /*
1268  * extent_io.c call back to do delayed allocation processing
1269  */
1270 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1271                               u64 start, u64 end, int *page_started,
1272                               unsigned long *nr_written)
1273 {
1274         int ret;
1275         struct btrfs_root *root = BTRFS_I(inode)->root;
1276
1277         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1278                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1279                                          page_started, 1, nr_written);
1280         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1281                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1282                                          page_started, 0, nr_written);
1283         else if (!btrfs_test_opt(root, COMPRESS) &&
1284                  !(BTRFS_I(inode)->force_compress) &&
1285                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1286                 ret = cow_file_range(inode, locked_page, start, end,
1287                                       page_started, nr_written, 1);
1288         else
1289                 ret = cow_file_range_async(inode, locked_page, start, end,
1290                                            page_started, nr_written);
1291         return ret;
1292 }
1293
1294 static int btrfs_split_extent_hook(struct inode *inode,
1295                                    struct extent_state *orig, u64 split)
1296 {
1297         /* not delalloc, ignore it */
1298         if (!(orig->state & EXTENT_DELALLOC))
1299                 return 0;
1300
1301         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1302         return 0;
1303 }
1304
1305 /*
1306  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1307  * extents so we can keep track of new extents that are just merged onto old
1308  * extents, such as when we are doing sequential writes, so we can properly
1309  * account for the metadata space we'll need.
1310  */
1311 static int btrfs_merge_extent_hook(struct inode *inode,
1312                                    struct extent_state *new,
1313                                    struct extent_state *other)
1314 {
1315         /* not delalloc, ignore it */
1316         if (!(other->state & EXTENT_DELALLOC))
1317                 return 0;
1318
1319         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1320         return 0;
1321 }
1322
1323 /*
1324  * extent_io.c set_bit_hook, used to track delayed allocation
1325  * bytes in this file, and to maintain the list of inodes that
1326  * have pending delalloc work to be done.
1327  */
1328 static int btrfs_set_bit_hook(struct inode *inode,
1329                               struct extent_state *state, int *bits)
1330 {
1331
1332         /*
1333          * set_bit and clear bit hooks normally require _irqsave/restore
1334          * but in this case, we are only testing for the DELALLOC
1335          * bit, which is only set or cleared with irqs on
1336          */
1337         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1338                 struct btrfs_root *root = BTRFS_I(inode)->root;
1339                 u64 len = state->end + 1 - state->start;
1340                 bool do_list = !is_free_space_inode(root, inode);
1341
1342                 if (*bits & EXTENT_FIRST_DELALLOC)
1343                         *bits &= ~EXTENT_FIRST_DELALLOC;
1344                 else
1345                         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1346
1347                 spin_lock(&root->fs_info->delalloc_lock);
1348                 BTRFS_I(inode)->delalloc_bytes += len;
1349                 root->fs_info->delalloc_bytes += len;
1350                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1351                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1352                                       &root->fs_info->delalloc_inodes);
1353                 }
1354                 spin_unlock(&root->fs_info->delalloc_lock);
1355         }
1356         return 0;
1357 }
1358
1359 /*
1360  * extent_io.c clear_bit_hook, see set_bit_hook for why
1361  */
1362 static int btrfs_clear_bit_hook(struct inode *inode,
1363                                 struct extent_state *state, int *bits)
1364 {
1365         /*
1366          * set_bit and clear bit hooks normally require _irqsave/restore
1367          * but in this case, we are only testing for the DELALLOC
1368          * bit, which is only set or cleared with irqs on
1369          */
1370         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1371                 struct btrfs_root *root = BTRFS_I(inode)->root;
1372                 u64 len = state->end + 1 - state->start;
1373                 bool do_list = !is_free_space_inode(root, inode);
1374
1375                 if (*bits & EXTENT_FIRST_DELALLOC)
1376                         *bits &= ~EXTENT_FIRST_DELALLOC;
1377                 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1378                         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1379
1380                 if (*bits & EXTENT_DO_ACCOUNTING)
1381                         btrfs_delalloc_release_metadata(inode, len);
1382
1383                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1384                     && do_list)
1385                         btrfs_free_reserved_data_space(inode, len);
1386
1387                 spin_lock(&root->fs_info->delalloc_lock);
1388                 root->fs_info->delalloc_bytes -= len;
1389                 BTRFS_I(inode)->delalloc_bytes -= len;
1390
1391                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1392                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1393                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1394                 }
1395                 spin_unlock(&root->fs_info->delalloc_lock);
1396         }
1397         return 0;
1398 }
1399
1400 /*
1401  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1402  * we don't create bios that span stripes or chunks
1403  */
1404 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1405                          size_t size, struct bio *bio,
1406                          unsigned long bio_flags)
1407 {
1408         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1409         struct btrfs_mapping_tree *map_tree;
1410         u64 logical = (u64)bio->bi_sector << 9;
1411         u64 length = 0;
1412         u64 map_length;
1413         int ret;
1414
1415         if (bio_flags & EXTENT_BIO_COMPRESSED)
1416                 return 0;
1417
1418         length = bio->bi_size;
1419         map_tree = &root->fs_info->mapping_tree;
1420         map_length = length;
1421         ret = btrfs_map_block(map_tree, READ, logical,
1422                               &map_length, NULL, 0);
1423
1424         if (map_length < length + size)
1425                 return 1;
1426         return ret;
1427 }
1428
1429 /*
1430  * in order to insert checksums into the metadata in large chunks,
1431  * we wait until bio submission time.   All the pages in the bio are
1432  * checksummed and sums are attached onto the ordered extent record.
1433  *
1434  * At IO completion time the cums attached on the ordered extent record
1435  * are inserted into the btree
1436  */
1437 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1438                                     struct bio *bio, int mirror_num,
1439                                     unsigned long bio_flags,
1440                                     u64 bio_offset)
1441 {
1442         struct btrfs_root *root = BTRFS_I(inode)->root;
1443         int ret = 0;
1444
1445         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1446         BUG_ON(ret);
1447         return 0;
1448 }
1449
1450 /*
1451  * in order to insert checksums into the metadata in large chunks,
1452  * we wait until bio submission time.   All the pages in the bio are
1453  * checksummed and sums are attached onto the ordered extent record.
1454  *
1455  * At IO completion time the cums attached on the ordered extent record
1456  * are inserted into the btree
1457  */
1458 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1459                           int mirror_num, unsigned long bio_flags,
1460                           u64 bio_offset)
1461 {
1462         struct btrfs_root *root = BTRFS_I(inode)->root;
1463         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1464 }
1465
1466 /*
1467  * extent_io.c submission hook. This does the right thing for csum calculation
1468  * on write, or reading the csums from the tree before a read
1469  */
1470 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1471                           int mirror_num, unsigned long bio_flags,
1472                           u64 bio_offset)
1473 {
1474         struct btrfs_root *root = BTRFS_I(inode)->root;
1475         int ret = 0;
1476         int skip_sum;
1477
1478         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1479
1480         if (is_free_space_inode(root, inode))
1481                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1482         else
1483                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1484         BUG_ON(ret);
1485
1486         if (!(rw & REQ_WRITE)) {
1487                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1488                         return btrfs_submit_compressed_read(inode, bio,
1489                                                     mirror_num, bio_flags);
1490                 } else if (!skip_sum) {
1491                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1492                         if (ret)
1493                                 return ret;
1494                 }
1495                 goto mapit;
1496         } else if (!skip_sum) {
1497                 /* csum items have already been cloned */
1498                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1499                         goto mapit;
1500                 /* we're doing a write, do the async checksumming */
1501                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1502                                    inode, rw, bio, mirror_num,
1503                                    bio_flags, bio_offset,
1504                                    __btrfs_submit_bio_start,
1505                                    __btrfs_submit_bio_done);
1506         }
1507
1508 mapit:
1509         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1510 }
1511
1512 /*
1513  * given a list of ordered sums record them in the inode.  This happens
1514  * at IO completion time based on sums calculated at bio submission time.
1515  */
1516 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1517                              struct inode *inode, u64 file_offset,
1518                              struct list_head *list)
1519 {
1520         struct btrfs_ordered_sum *sum;
1521
1522         list_for_each_entry(sum, list, list) {
1523                 btrfs_csum_file_blocks(trans,
1524                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1525         }
1526         return 0;
1527 }
1528
1529 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1530                               struct extent_state **cached_state)
1531 {
1532         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1533                 WARN_ON(1);
1534         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1535                                    cached_state, GFP_NOFS);
1536 }
1537
1538 /* see btrfs_writepage_start_hook for details on why this is required */
1539 struct btrfs_writepage_fixup {
1540         struct page *page;
1541         struct btrfs_work work;
1542 };
1543
1544 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1545 {
1546         struct btrfs_writepage_fixup *fixup;
1547         struct btrfs_ordered_extent *ordered;
1548         struct extent_state *cached_state = NULL;
1549         struct page *page;
1550         struct inode *inode;
1551         u64 page_start;
1552         u64 page_end;
1553
1554         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1555         page = fixup->page;
1556 again:
1557         lock_page(page);
1558         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1559                 ClearPageChecked(page);
1560                 goto out_page;
1561         }
1562
1563         inode = page->mapping->host;
1564         page_start = page_offset(page);
1565         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1566
1567         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1568                          &cached_state, GFP_NOFS);
1569
1570         /* already ordered? We're done */
1571         if (PagePrivate2(page))
1572                 goto out;
1573
1574         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1575         if (ordered) {
1576                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1577                                      page_end, &cached_state, GFP_NOFS);
1578                 unlock_page(page);
1579                 btrfs_start_ordered_extent(inode, ordered, 1);
1580                 goto again;
1581         }
1582
1583         BUG();
1584         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1585         ClearPageChecked(page);
1586 out:
1587         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1588                              &cached_state, GFP_NOFS);
1589 out_page:
1590         unlock_page(page);
1591         page_cache_release(page);
1592         kfree(fixup);
1593 }
1594
1595 /*
1596  * There are a few paths in the higher layers of the kernel that directly
1597  * set the page dirty bit without asking the filesystem if it is a
1598  * good idea.  This causes problems because we want to make sure COW
1599  * properly happens and the data=ordered rules are followed.
1600  *
1601  * In our case any range that doesn't have the ORDERED bit set
1602  * hasn't been properly setup for IO.  We kick off an async process
1603  * to fix it up.  The async helper will wait for ordered extents, set
1604  * the delalloc bit and make it safe to write the page.
1605  */
1606 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1607 {
1608         struct inode *inode = page->mapping->host;
1609         struct btrfs_writepage_fixup *fixup;
1610         struct btrfs_root *root = BTRFS_I(inode)->root;
1611
1612         /* this page is properly in the ordered list */
1613         if (TestClearPagePrivate2(page))
1614                 return 0;
1615
1616         if (PageChecked(page))
1617                 return -EAGAIN;
1618
1619         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1620         if (!fixup)
1621                 return -EAGAIN;
1622
1623         SetPageChecked(page);
1624         page_cache_get(page);
1625         fixup->work.func = btrfs_writepage_fixup_worker;
1626         fixup->page = page;
1627         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1628         return -EAGAIN;
1629 }
1630
1631 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1632                                        struct inode *inode, u64 file_pos,
1633                                        u64 disk_bytenr, u64 disk_num_bytes,
1634                                        u64 num_bytes, u64 ram_bytes,
1635                                        u8 compression, u8 encryption,
1636                                        u16 other_encoding, int extent_type)
1637 {
1638         struct btrfs_root *root = BTRFS_I(inode)->root;
1639         struct btrfs_file_extent_item *fi;
1640         struct btrfs_path *path;
1641         struct extent_buffer *leaf;
1642         struct btrfs_key ins;
1643         u64 hint;
1644         int ret;
1645
1646         path = btrfs_alloc_path();
1647         BUG_ON(!path);
1648
1649         path->leave_spinning = 1;
1650
1651         /*
1652          * we may be replacing one extent in the tree with another.
1653          * The new extent is pinned in the extent map, and we don't want
1654          * to drop it from the cache until it is completely in the btree.
1655          *
1656          * So, tell btrfs_drop_extents to leave this extent in the cache.
1657          * the caller is expected to unpin it and allow it to be merged
1658          * with the others.
1659          */
1660         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1661                                  &hint, 0);
1662         BUG_ON(ret);
1663
1664         ins.objectid = btrfs_ino(inode);
1665         ins.offset = file_pos;
1666         ins.type = BTRFS_EXTENT_DATA_KEY;
1667         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1668         BUG_ON(ret);
1669         leaf = path->nodes[0];
1670         fi = btrfs_item_ptr(leaf, path->slots[0],
1671                             struct btrfs_file_extent_item);
1672         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1673         btrfs_set_file_extent_type(leaf, fi, extent_type);
1674         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1675         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1676         btrfs_set_file_extent_offset(leaf, fi, 0);
1677         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1678         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1679         btrfs_set_file_extent_compression(leaf, fi, compression);
1680         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1681         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1682
1683         btrfs_unlock_up_safe(path, 1);
1684         btrfs_set_lock_blocking(leaf);
1685
1686         btrfs_mark_buffer_dirty(leaf);
1687
1688         inode_add_bytes(inode, num_bytes);
1689
1690         ins.objectid = disk_bytenr;
1691         ins.offset = disk_num_bytes;
1692         ins.type = BTRFS_EXTENT_ITEM_KEY;
1693         ret = btrfs_alloc_reserved_file_extent(trans, root,
1694                                         root->root_key.objectid,
1695                                         btrfs_ino(inode), file_pos, &ins);
1696         BUG_ON(ret);
1697         btrfs_free_path(path);
1698
1699         return 0;
1700 }
1701
1702 /*
1703  * helper function for btrfs_finish_ordered_io, this
1704  * just reads in some of the csum leaves to prime them into ram
1705  * before we start the transaction.  It limits the amount of btree
1706  * reads required while inside the transaction.
1707  */
1708 /* as ordered data IO finishes, this gets called so we can finish
1709  * an ordered extent if the range of bytes in the file it covers are
1710  * fully written.
1711  */
1712 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1713 {
1714         struct btrfs_root *root = BTRFS_I(inode)->root;
1715         struct btrfs_trans_handle *trans = NULL;
1716         struct btrfs_ordered_extent *ordered_extent = NULL;
1717         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1718         struct extent_state *cached_state = NULL;
1719         int compress_type = 0;
1720         int ret;
1721         bool nolock;
1722
1723         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1724                                              end - start + 1);
1725         if (!ret)
1726                 return 0;
1727         BUG_ON(!ordered_extent);
1728
1729         nolock = is_free_space_inode(root, inode);
1730
1731         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1732                 BUG_ON(!list_empty(&ordered_extent->list));
1733                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1734                 if (!ret) {
1735                         if (nolock)
1736                                 trans = btrfs_join_transaction_nolock(root);
1737                         else
1738                                 trans = btrfs_join_transaction(root);
1739                         BUG_ON(IS_ERR(trans));
1740                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1741                         ret = btrfs_update_inode(trans, root, inode);
1742                         BUG_ON(ret);
1743                 }
1744                 goto out;
1745         }
1746
1747         lock_extent_bits(io_tree, ordered_extent->file_offset,
1748                          ordered_extent->file_offset + ordered_extent->len - 1,
1749                          0, &cached_state, GFP_NOFS);
1750
1751         if (nolock)
1752                 trans = btrfs_join_transaction_nolock(root);
1753         else
1754                 trans = btrfs_join_transaction(root);
1755         BUG_ON(IS_ERR(trans));
1756         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1757
1758         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1759                 compress_type = ordered_extent->compress_type;
1760         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1761                 BUG_ON(compress_type);
1762                 ret = btrfs_mark_extent_written(trans, inode,
1763                                                 ordered_extent->file_offset,
1764                                                 ordered_extent->file_offset +
1765                                                 ordered_extent->len);
1766                 BUG_ON(ret);
1767         } else {
1768                 BUG_ON(root == root->fs_info->tree_root);
1769                 ret = insert_reserved_file_extent(trans, inode,
1770                                                 ordered_extent->file_offset,
1771                                                 ordered_extent->start,
1772                                                 ordered_extent->disk_len,
1773                                                 ordered_extent->len,
1774                                                 ordered_extent->len,
1775                                                 compress_type, 0, 0,
1776                                                 BTRFS_FILE_EXTENT_REG);
1777                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1778                                    ordered_extent->file_offset,
1779                                    ordered_extent->len);
1780                 BUG_ON(ret);
1781         }
1782         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1783                              ordered_extent->file_offset +
1784                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1785
1786         add_pending_csums(trans, inode, ordered_extent->file_offset,
1787                           &ordered_extent->list);
1788
1789         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1790         if (!ret) {
1791                 ret = btrfs_update_inode(trans, root, inode);
1792                 BUG_ON(ret);
1793         }
1794         ret = 0;
1795 out:
1796         if (nolock) {
1797                 if (trans)
1798                         btrfs_end_transaction_nolock(trans, root);
1799         } else {
1800                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1801                 if (trans)
1802                         btrfs_end_transaction(trans, root);
1803         }
1804
1805         /* once for us */
1806         btrfs_put_ordered_extent(ordered_extent);
1807         /* once for the tree */
1808         btrfs_put_ordered_extent(ordered_extent);
1809
1810         return 0;
1811 }
1812
1813 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1814                                 struct extent_state *state, int uptodate)
1815 {
1816         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1817
1818         ClearPagePrivate2(page);
1819         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1820 }
1821
1822 /*
1823  * When IO fails, either with EIO or csum verification fails, we
1824  * try other mirrors that might have a good copy of the data.  This
1825  * io_failure_record is used to record state as we go through all the
1826  * mirrors.  If another mirror has good data, the page is set up to date
1827  * and things continue.  If a good mirror can't be found, the original
1828  * bio end_io callback is called to indicate things have failed.
1829  */
1830 struct io_failure_record {
1831         struct page *page;
1832         u64 start;
1833         u64 len;
1834         u64 logical;
1835         unsigned long bio_flags;
1836         int last_mirror;
1837 };
1838
1839 static int btrfs_io_failed_hook(struct bio *failed_bio,
1840                          struct page *page, u64 start, u64 end,
1841                          struct extent_state *state)
1842 {
1843         struct io_failure_record *failrec = NULL;
1844         u64 private;
1845         struct extent_map *em;
1846         struct inode *inode = page->mapping->host;
1847         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1848         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1849         struct bio *bio;
1850         int num_copies;
1851         int ret;
1852         int rw;
1853         u64 logical;
1854
1855         ret = get_state_private(failure_tree, start, &private);
1856         if (ret) {
1857                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1858                 if (!failrec)
1859                         return -ENOMEM;
1860                 failrec->start = start;
1861                 failrec->len = end - start + 1;
1862                 failrec->last_mirror = 0;
1863                 failrec->bio_flags = 0;
1864
1865                 read_lock(&em_tree->lock);
1866                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1867                 if (em->start > start || em->start + em->len < start) {
1868                         free_extent_map(em);
1869                         em = NULL;
1870                 }
1871                 read_unlock(&em_tree->lock);
1872
1873                 if (IS_ERR_OR_NULL(em)) {
1874                         kfree(failrec);
1875                         return -EIO;
1876                 }
1877                 logical = start - em->start;
1878                 logical = em->block_start + logical;
1879                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1880                         logical = em->block_start;
1881                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1882                         extent_set_compress_type(&failrec->bio_flags,
1883                                                  em->compress_type);
1884                 }
1885                 failrec->logical = logical;
1886                 free_extent_map(em);
1887                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1888                                 EXTENT_DIRTY, GFP_NOFS);
1889                 set_state_private(failure_tree, start,
1890                                  (u64)(unsigned long)failrec);
1891         } else {
1892                 failrec = (struct io_failure_record *)(unsigned long)private;
1893         }
1894         num_copies = btrfs_num_copies(
1895                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1896                               failrec->logical, failrec->len);
1897         failrec->last_mirror++;
1898         if (!state) {
1899                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1900                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1901                                                     failrec->start,
1902                                                     EXTENT_LOCKED);
1903                 if (state && state->start != failrec->start)
1904                         state = NULL;
1905                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1906         }
1907         if (!state || failrec->last_mirror > num_copies) {
1908                 set_state_private(failure_tree, failrec->start, 0);
1909                 clear_extent_bits(failure_tree, failrec->start,
1910                                   failrec->start + failrec->len - 1,
1911                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1912                 kfree(failrec);
1913                 return -EIO;
1914         }
1915         bio = bio_alloc(GFP_NOFS, 1);
1916         bio->bi_private = state;
1917         bio->bi_end_io = failed_bio->bi_end_io;
1918         bio->bi_sector = failrec->logical >> 9;
1919         bio->bi_bdev = failed_bio->bi_bdev;
1920         bio->bi_size = 0;
1921
1922         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1923         if (failed_bio->bi_rw & REQ_WRITE)
1924                 rw = WRITE;
1925         else
1926                 rw = READ;
1927
1928         ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1929                                                       failrec->last_mirror,
1930                                                       failrec->bio_flags, 0);
1931         return ret;
1932 }
1933
1934 /*
1935  * each time an IO finishes, we do a fast check in the IO failure tree
1936  * to see if we need to process or clean up an io_failure_record
1937  */
1938 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1939 {
1940         u64 private;
1941         u64 private_failure;
1942         struct io_failure_record *failure;
1943         int ret;
1944
1945         private = 0;
1946         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1947                              (u64)-1, 1, EXTENT_DIRTY, 0)) {
1948                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1949                                         start, &private_failure);
1950                 if (ret == 0) {
1951                         failure = (struct io_failure_record *)(unsigned long)
1952                                    private_failure;
1953                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1954                                           failure->start, 0);
1955                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1956                                           failure->start,
1957                                           failure->start + failure->len - 1,
1958                                           EXTENT_DIRTY | EXTENT_LOCKED,
1959                                           GFP_NOFS);
1960                         kfree(failure);
1961                 }
1962         }
1963         return 0;
1964 }
1965
1966 /*
1967  * when reads are done, we need to check csums to verify the data is correct
1968  * if there's a match, we allow the bio to finish.  If not, we go through
1969  * the io_failure_record routines to find good copies
1970  */
1971 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1972                                struct extent_state *state)
1973 {
1974         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1975         struct inode *inode = page->mapping->host;
1976         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1977         char *kaddr;
1978         u64 private = ~(u32)0;
1979         int ret;
1980         struct btrfs_root *root = BTRFS_I(inode)->root;
1981         u32 csum = ~(u32)0;
1982
1983         if (PageChecked(page)) {
1984                 ClearPageChecked(page);
1985                 goto good;
1986         }
1987
1988         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1989                 goto good;
1990
1991         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1992             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1993                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1994                                   GFP_NOFS);
1995                 return 0;
1996         }
1997
1998         if (state && state->start == start) {
1999                 private = state->private;
2000                 ret = 0;
2001         } else {
2002                 ret = get_state_private(io_tree, start, &private);
2003         }
2004         kaddr = kmap_atomic(page, KM_USER0);
2005         if (ret)
2006                 goto zeroit;
2007
2008         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2009         btrfs_csum_final(csum, (char *)&csum);
2010         if (csum != private)
2011                 goto zeroit;
2012
2013         kunmap_atomic(kaddr, KM_USER0);
2014 good:
2015         /* if the io failure tree for this inode is non-empty,
2016          * check to see if we've recovered from a failed IO
2017          */
2018         btrfs_clean_io_failures(inode, start);
2019         return 0;
2020
2021 zeroit:
2022         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2023                        "private %llu\n",
2024                        (unsigned long long)btrfs_ino(page->mapping->host),
2025                        (unsigned long long)start, csum,
2026                        (unsigned long long)private);
2027         memset(kaddr + offset, 1, end - start + 1);
2028         flush_dcache_page(page);
2029         kunmap_atomic(kaddr, KM_USER0);
2030         if (private == 0)
2031                 return 0;
2032         return -EIO;
2033 }
2034
2035 struct delayed_iput {
2036         struct list_head list;
2037         struct inode *inode;
2038 };
2039
2040 void btrfs_add_delayed_iput(struct inode *inode)
2041 {
2042         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2043         struct delayed_iput *delayed;
2044
2045         if (atomic_add_unless(&inode->i_count, -1, 1))
2046                 return;
2047
2048         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2049         delayed->inode = inode;
2050
2051         spin_lock(&fs_info->delayed_iput_lock);
2052         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2053         spin_unlock(&fs_info->delayed_iput_lock);
2054 }
2055
2056 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2057 {
2058         LIST_HEAD(list);
2059         struct btrfs_fs_info *fs_info = root->fs_info;
2060         struct delayed_iput *delayed;
2061         int empty;
2062
2063         spin_lock(&fs_info->delayed_iput_lock);
2064         empty = list_empty(&fs_info->delayed_iputs);
2065         spin_unlock(&fs_info->delayed_iput_lock);
2066         if (empty)
2067                 return;
2068
2069         down_read(&root->fs_info->cleanup_work_sem);
2070         spin_lock(&fs_info->delayed_iput_lock);
2071         list_splice_init(&fs_info->delayed_iputs, &list);
2072         spin_unlock(&fs_info->delayed_iput_lock);
2073
2074         while (!list_empty(&list)) {
2075                 delayed = list_entry(list.next, struct delayed_iput, list);
2076                 list_del(&delayed->list);
2077                 iput(delayed->inode);
2078                 kfree(delayed);
2079         }
2080         up_read(&root->fs_info->cleanup_work_sem);
2081 }
2082
2083 /*
2084  * calculate extra metadata reservation when snapshotting a subvolume
2085  * contains orphan files.
2086  */
2087 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2088                                 struct btrfs_pending_snapshot *pending,
2089                                 u64 *bytes_to_reserve)
2090 {
2091         struct btrfs_root *root;
2092         struct btrfs_block_rsv *block_rsv;
2093         u64 num_bytes;
2094         int index;
2095
2096         root = pending->root;
2097         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2098                 return;
2099
2100         block_rsv = root->orphan_block_rsv;
2101
2102         /* orphan block reservation for the snapshot */
2103         num_bytes = block_rsv->size;
2104
2105         /*
2106          * after the snapshot is created, COWing tree blocks may use more
2107          * space than it frees. So we should make sure there is enough
2108          * reserved space.
2109          */
2110         index = trans->transid & 0x1;
2111         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2112                 num_bytes += block_rsv->size -
2113                              (block_rsv->reserved + block_rsv->freed[index]);
2114         }
2115
2116         *bytes_to_reserve += num_bytes;
2117 }
2118
2119 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2120                                 struct btrfs_pending_snapshot *pending)
2121 {
2122         struct btrfs_root *root = pending->root;
2123         struct btrfs_root *snap = pending->snap;
2124         struct btrfs_block_rsv *block_rsv;
2125         u64 num_bytes;
2126         int index;
2127         int ret;
2128
2129         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2130                 return;
2131
2132         /* refill source subvolume's orphan block reservation */
2133         block_rsv = root->orphan_block_rsv;
2134         index = trans->transid & 0x1;
2135         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2136                 num_bytes = block_rsv->size -
2137                             (block_rsv->reserved + block_rsv->freed[index]);
2138                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2139                                               root->orphan_block_rsv,
2140                                               num_bytes);
2141                 BUG_ON(ret);
2142         }
2143
2144         /* setup orphan block reservation for the snapshot */
2145         block_rsv = btrfs_alloc_block_rsv(snap);
2146         BUG_ON(!block_rsv);
2147
2148         btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2149         snap->orphan_block_rsv = block_rsv;
2150
2151         num_bytes = root->orphan_block_rsv->size;
2152         ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2153                                       block_rsv, num_bytes);
2154         BUG_ON(ret);
2155
2156 #if 0
2157         /* insert orphan item for the snapshot */
2158         WARN_ON(!root->orphan_item_inserted);
2159         ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2160                                        snap->root_key.objectid);
2161         BUG_ON(ret);
2162         snap->orphan_item_inserted = 1;
2163 #endif
2164 }
2165
2166 enum btrfs_orphan_cleanup_state {
2167         ORPHAN_CLEANUP_STARTED  = 1,
2168         ORPHAN_CLEANUP_DONE     = 2,
2169 };
2170
2171 /*
2172  * This is called in transaction commmit time. If there are no orphan
2173  * files in the subvolume, it removes orphan item and frees block_rsv
2174  * structure.
2175  */
2176 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2177                               struct btrfs_root *root)
2178 {
2179         int ret;
2180
2181         if (!list_empty(&root->orphan_list) ||
2182             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2183                 return;
2184
2185         if (root->orphan_item_inserted &&
2186             btrfs_root_refs(&root->root_item) > 0) {
2187                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2188                                             root->root_key.objectid);
2189                 BUG_ON(ret);
2190                 root->orphan_item_inserted = 0;
2191         }
2192
2193         if (root->orphan_block_rsv) {
2194                 WARN_ON(root->orphan_block_rsv->size > 0);
2195                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2196                 root->orphan_block_rsv = NULL;
2197         }
2198 }
2199
2200 /*
2201  * This creates an orphan entry for the given inode in case something goes
2202  * wrong in the middle of an unlink/truncate.
2203  *
2204  * NOTE: caller of this function should reserve 5 units of metadata for
2205  *       this function.
2206  */
2207 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2208 {
2209         struct btrfs_root *root = BTRFS_I(inode)->root;
2210         struct btrfs_block_rsv *block_rsv = NULL;
2211         int reserve = 0;
2212         int insert = 0;
2213         int ret;
2214
2215         if (!root->orphan_block_rsv) {
2216                 block_rsv = btrfs_alloc_block_rsv(root);
2217                 BUG_ON(!block_rsv);
2218         }
2219
2220         spin_lock(&root->orphan_lock);
2221         if (!root->orphan_block_rsv) {
2222                 root->orphan_block_rsv = block_rsv;
2223         } else if (block_rsv) {
2224                 btrfs_free_block_rsv(root, block_rsv);
2225                 block_rsv = NULL;
2226         }
2227
2228         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2229                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2230 #if 0
2231                 /*
2232                  * For proper ENOSPC handling, we should do orphan
2233                  * cleanup when mounting. But this introduces backward
2234                  * compatibility issue.
2235                  */
2236                 if (!xchg(&root->orphan_item_inserted, 1))
2237                         insert = 2;
2238                 else
2239                         insert = 1;
2240 #endif
2241                 insert = 1;
2242         }
2243
2244         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2245                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2246                 reserve = 1;
2247         }
2248         spin_unlock(&root->orphan_lock);
2249
2250         if (block_rsv)
2251                 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2252
2253         /* grab metadata reservation from transaction handle */
2254         if (reserve) {
2255                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2256                 BUG_ON(ret);
2257         }
2258
2259         /* insert an orphan item to track this unlinked/truncated file */
2260         if (insert >= 1) {
2261                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2262                 BUG_ON(ret);
2263         }
2264
2265         /* insert an orphan item to track subvolume contains orphan files */
2266         if (insert >= 2) {
2267                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2268                                                root->root_key.objectid);
2269                 BUG_ON(ret);
2270         }
2271         return 0;
2272 }
2273
2274 /*
2275  * We have done the truncate/delete so we can go ahead and remove the orphan
2276  * item for this particular inode.
2277  */
2278 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2279 {
2280         struct btrfs_root *root = BTRFS_I(inode)->root;
2281         int delete_item = 0;
2282         int release_rsv = 0;
2283         int ret = 0;
2284
2285         spin_lock(&root->orphan_lock);
2286         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2287                 list_del_init(&BTRFS_I(inode)->i_orphan);
2288                 delete_item = 1;
2289         }
2290
2291         if (BTRFS_I(inode)->orphan_meta_reserved) {
2292                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2293                 release_rsv = 1;
2294         }
2295         spin_unlock(&root->orphan_lock);
2296
2297         if (trans && delete_item) {
2298                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2299                 BUG_ON(ret);
2300         }
2301
2302         if (release_rsv)
2303                 btrfs_orphan_release_metadata(inode);
2304
2305         return 0;
2306 }
2307
2308 /*
2309  * this cleans up any orphans that may be left on the list from the last use
2310  * of this root.
2311  */
2312 int btrfs_orphan_cleanup(struct btrfs_root *root)
2313 {
2314         struct btrfs_path *path;
2315         struct extent_buffer *leaf;
2316         struct btrfs_key key, found_key;
2317         struct btrfs_trans_handle *trans;
2318         struct inode *inode;
2319         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2320
2321         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2322                 return 0;
2323
2324         path = btrfs_alloc_path();
2325         if (!path) {
2326                 ret = -ENOMEM;
2327                 goto out;
2328         }
2329         path->reada = -1;
2330
2331         key.objectid = BTRFS_ORPHAN_OBJECTID;
2332         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2333         key.offset = (u64)-1;
2334
2335         while (1) {
2336                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2337                 if (ret < 0)
2338                         goto out;
2339
2340                 /*
2341                  * if ret == 0 means we found what we were searching for, which
2342                  * is weird, but possible, so only screw with path if we didn't
2343                  * find the key and see if we have stuff that matches
2344                  */
2345                 if (ret > 0) {
2346                         ret = 0;
2347                         if (path->slots[0] == 0)
2348                                 break;
2349                         path->slots[0]--;
2350                 }
2351
2352                 /* pull out the item */
2353                 leaf = path->nodes[0];
2354                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2355
2356                 /* make sure the item matches what we want */
2357                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2358                         break;
2359                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2360                         break;
2361
2362                 /* release the path since we're done with it */
2363                 btrfs_release_path(path);
2364
2365                 /*
2366                  * this is where we are basically btrfs_lookup, without the
2367                  * crossing root thing.  we store the inode number in the
2368                  * offset of the orphan item.
2369                  */
2370                 found_key.objectid = found_key.offset;
2371                 found_key.type = BTRFS_INODE_ITEM_KEY;
2372                 found_key.offset = 0;
2373                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2374                 if (IS_ERR(inode)) {
2375                         ret = PTR_ERR(inode);
2376                         goto out;
2377                 }
2378
2379                 /*
2380                  * add this inode to the orphan list so btrfs_orphan_del does
2381                  * the proper thing when we hit it
2382                  */
2383                 spin_lock(&root->orphan_lock);
2384                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2385                 spin_unlock(&root->orphan_lock);
2386
2387                 /*
2388                  * if this is a bad inode, means we actually succeeded in
2389                  * removing the inode, but not the orphan record, which means
2390                  * we need to manually delete the orphan since iput will just
2391                  * do a destroy_inode
2392                  */
2393                 if (is_bad_inode(inode)) {
2394                         trans = btrfs_start_transaction(root, 0);
2395                         if (IS_ERR(trans)) {
2396                                 ret = PTR_ERR(trans);
2397                                 goto out;
2398                         }
2399                         btrfs_orphan_del(trans, inode);
2400                         btrfs_end_transaction(trans, root);
2401                         iput(inode);
2402                         continue;
2403                 }
2404
2405                 /* if we have links, this was a truncate, lets do that */
2406                 if (inode->i_nlink) {
2407                         if (!S_ISREG(inode->i_mode)) {
2408                                 WARN_ON(1);
2409                                 iput(inode);
2410                                 continue;
2411                         }
2412                         nr_truncate++;
2413                         ret = btrfs_truncate(inode);
2414                 } else {
2415                         nr_unlink++;
2416                 }
2417
2418                 /* this will do delete_inode and everything for us */
2419                 iput(inode);
2420                 if (ret)
2421                         goto out;
2422         }
2423         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2424
2425         if (root->orphan_block_rsv)
2426                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2427                                         (u64)-1);
2428
2429         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2430                 trans = btrfs_join_transaction(root);
2431                 if (!IS_ERR(trans))
2432                         btrfs_end_transaction(trans, root);
2433         }
2434
2435         if (nr_unlink)
2436                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2437         if (nr_truncate)
2438                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2439
2440 out:
2441         if (ret)
2442                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2443         btrfs_free_path(path);
2444         return ret;
2445 }
2446
2447 /*
2448  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2449  * don't find any xattrs, we know there can't be any acls.
2450  *
2451  * slot is the slot the inode is in, objectid is the objectid of the inode
2452  */
2453 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2454                                           int slot, u64 objectid)
2455 {
2456         u32 nritems = btrfs_header_nritems(leaf);
2457         struct btrfs_key found_key;
2458         int scanned = 0;
2459
2460         slot++;
2461         while (slot < nritems) {
2462                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2463
2464                 /* we found a different objectid, there must not be acls */
2465                 if (found_key.objectid != objectid)
2466                         return 0;
2467
2468                 /* we found an xattr, assume we've got an acl */
2469                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2470                         return 1;
2471
2472                 /*
2473                  * we found a key greater than an xattr key, there can't
2474                  * be any acls later on
2475                  */
2476                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2477                         return 0;
2478
2479                 slot++;
2480                 scanned++;
2481
2482                 /*
2483                  * it goes inode, inode backrefs, xattrs, extents,
2484                  * so if there are a ton of hard links to an inode there can
2485                  * be a lot of backrefs.  Don't waste time searching too hard,
2486                  * this is just an optimization
2487                  */
2488                 if (scanned >= 8)
2489                         break;
2490         }
2491         /* we hit the end of the leaf before we found an xattr or
2492          * something larger than an xattr.  We have to assume the inode
2493          * has acls
2494          */
2495         return 1;
2496 }
2497
2498 /*
2499  * read an inode from the btree into the in-memory inode
2500  */
2501 static void btrfs_read_locked_inode(struct inode *inode)
2502 {
2503         struct btrfs_path *path;
2504         struct extent_buffer *leaf;
2505         struct btrfs_inode_item *inode_item;
2506         struct btrfs_timespec *tspec;
2507         struct btrfs_root *root = BTRFS_I(inode)->root;
2508         struct btrfs_key location;
2509         int maybe_acls;
2510         u32 rdev;
2511         int ret;
2512
2513         path = btrfs_alloc_path();
2514         BUG_ON(!path);
2515         path->leave_spinning = 1;
2516         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2517
2518         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2519         if (ret)
2520                 goto make_bad;
2521
2522         leaf = path->nodes[0];
2523         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2524                                     struct btrfs_inode_item);
2525         if (!leaf->map_token)
2526                 map_private_extent_buffer(leaf, (unsigned long)inode_item,
2527                                           sizeof(struct btrfs_inode_item),
2528                                           &leaf->map_token, &leaf->kaddr,
2529                                           &leaf->map_start, &leaf->map_len,
2530                                           KM_USER1);
2531
2532         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2533         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2534         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2535         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2536         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2537
2538         tspec = btrfs_inode_atime(inode_item);
2539         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2540         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2541
2542         tspec = btrfs_inode_mtime(inode_item);
2543         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2544         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2545
2546         tspec = btrfs_inode_ctime(inode_item);
2547         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2548         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2549
2550         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2551         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2552         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2553         inode->i_generation = BTRFS_I(inode)->generation;
2554         inode->i_rdev = 0;
2555         rdev = btrfs_inode_rdev(leaf, inode_item);
2556
2557         BTRFS_I(inode)->index_cnt = (u64)-1;
2558         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2559
2560         /*
2561          * try to precache a NULL acl entry for files that don't have
2562          * any xattrs or acls
2563          */
2564         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2565                                            btrfs_ino(inode));
2566         if (!maybe_acls)
2567                 cache_no_acl(inode);
2568
2569         if (leaf->map_token) {
2570                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2571                 leaf->map_token = NULL;
2572         }
2573
2574         btrfs_free_path(path);
2575         inode_item = NULL;
2576
2577         switch (inode->i_mode & S_IFMT) {
2578         case S_IFREG:
2579                 inode->i_mapping->a_ops = &btrfs_aops;
2580                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2581                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2582                 inode->i_fop = &btrfs_file_operations;
2583                 inode->i_op = &btrfs_file_inode_operations;
2584                 break;
2585         case S_IFDIR:
2586                 inode->i_fop = &btrfs_dir_file_operations;
2587                 if (root == root->fs_info->tree_root)
2588                         inode->i_op = &btrfs_dir_ro_inode_operations;
2589                 else
2590                         inode->i_op = &btrfs_dir_inode_operations;
2591                 break;
2592         case S_IFLNK:
2593                 inode->i_op = &btrfs_symlink_inode_operations;
2594                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2595                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2596                 break;
2597         default:
2598                 inode->i_op = &btrfs_special_inode_operations;
2599                 init_special_inode(inode, inode->i_mode, rdev);
2600                 break;
2601         }
2602
2603         btrfs_update_iflags(inode);
2604         return;
2605
2606 make_bad:
2607         btrfs_free_path(path);
2608         make_bad_inode(inode);
2609 }
2610
2611 /*
2612  * given a leaf and an inode, copy the inode fields into the leaf
2613  */
2614 static void fill_inode_item(struct btrfs_trans_handle *trans,
2615                             struct extent_buffer *leaf,
2616                             struct btrfs_inode_item *item,
2617                             struct inode *inode)
2618 {
2619         if (!leaf->map_token)
2620                 map_private_extent_buffer(leaf, (unsigned long)item,
2621                                           sizeof(struct btrfs_inode_item),
2622                                           &leaf->map_token, &leaf->kaddr,
2623                                           &leaf->map_start, &leaf->map_len,
2624                                           KM_USER1);
2625
2626         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2627         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2628         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2629         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2630         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2631
2632         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2633                                inode->i_atime.tv_sec);
2634         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2635                                 inode->i_atime.tv_nsec);
2636
2637         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2638                                inode->i_mtime.tv_sec);
2639         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2640                                 inode->i_mtime.tv_nsec);
2641
2642         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2643                                inode->i_ctime.tv_sec);
2644         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2645                                 inode->i_ctime.tv_nsec);
2646
2647         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2648         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2649         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2650         btrfs_set_inode_transid(leaf, item, trans->transid);
2651         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2652         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2653         btrfs_set_inode_block_group(leaf, item, 0);
2654
2655         if (leaf->map_token) {
2656                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2657                 leaf->map_token = NULL;
2658         }
2659 }
2660
2661 /*
2662  * copy everything in the in-memory inode into the btree.
2663  */
2664 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2665                                 struct btrfs_root *root, struct inode *inode)
2666 {
2667         struct btrfs_inode_item *inode_item;
2668         struct btrfs_path *path;
2669         struct extent_buffer *leaf;
2670         int ret;
2671
2672         /*
2673          * If root is tree root, it means this inode is used to
2674          * store free space information. And these inodes are updated
2675          * when committing the transaction, so they needn't delaye to
2676          * be updated, or deadlock will occured.
2677          */
2678         if (!is_free_space_inode(root, inode)) {
2679                 ret = btrfs_delayed_update_inode(trans, root, inode);
2680                 if (!ret)
2681                         btrfs_set_inode_last_trans(trans, inode);
2682                 return ret;
2683         }
2684
2685         path = btrfs_alloc_path();
2686         if (!path)
2687                 return -ENOMEM;
2688
2689         path->leave_spinning = 1;
2690         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2691                                  1);
2692         if (ret) {
2693                 if (ret > 0)
2694                         ret = -ENOENT;
2695                 goto failed;
2696         }
2697
2698         btrfs_unlock_up_safe(path, 1);
2699         leaf = path->nodes[0];
2700         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2701                                     struct btrfs_inode_item);
2702
2703         fill_inode_item(trans, leaf, inode_item, inode);
2704         btrfs_mark_buffer_dirty(leaf);
2705         btrfs_set_inode_last_trans(trans, inode);
2706         ret = 0;
2707 failed:
2708         btrfs_free_path(path);
2709         return ret;
2710 }
2711
2712 /*
2713  * unlink helper that gets used here in inode.c and in the tree logging
2714  * recovery code.  It remove a link in a directory with a given name, and
2715  * also drops the back refs in the inode to the directory
2716  */
2717 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2718                                 struct btrfs_root *root,
2719                                 struct inode *dir, struct inode *inode,
2720                                 const char *name, int name_len)
2721 {
2722         struct btrfs_path *path;
2723         int ret = 0;
2724         struct extent_buffer *leaf;
2725         struct btrfs_dir_item *di;
2726         struct btrfs_key key;
2727         u64 index;
2728         u64 ino = btrfs_ino(inode);
2729         u64 dir_ino = btrfs_ino(dir);
2730
2731         path = btrfs_alloc_path();
2732         if (!path) {
2733                 ret = -ENOMEM;
2734                 goto out;
2735         }
2736
2737         path->leave_spinning = 1;
2738         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2739                                     name, name_len, -1);
2740         if (IS_ERR(di)) {
2741                 ret = PTR_ERR(di);
2742                 goto err;
2743         }
2744         if (!di) {
2745                 ret = -ENOENT;
2746                 goto err;
2747         }
2748         leaf = path->nodes[0];
2749         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2750         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2751         if (ret)
2752                 goto err;
2753         btrfs_release_path(path);
2754
2755         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2756                                   dir_ino, &index);
2757         if (ret) {
2758                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2759                        "inode %llu parent %llu\n", name_len, name,
2760                        (unsigned long long)ino, (unsigned long long)dir_ino);
2761                 goto err;
2762         }
2763
2764         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2765         if (ret)
2766                 goto err;
2767
2768         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2769                                          inode, dir_ino);
2770         BUG_ON(ret != 0 && ret != -ENOENT);
2771
2772         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2773                                            dir, index);
2774         if (ret == -ENOENT)
2775                 ret = 0;
2776 err:
2777         btrfs_free_path(path);
2778         if (ret)
2779                 goto out;
2780
2781         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2782         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2783         btrfs_update_inode(trans, root, dir);
2784 out:
2785         return ret;
2786 }
2787
2788 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2789                        struct btrfs_root *root,
2790                        struct inode *dir, struct inode *inode,
2791                        const char *name, int name_len)
2792 {
2793         int ret;
2794         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2795         if (!ret) {
2796                 btrfs_drop_nlink(inode);
2797                 ret = btrfs_update_inode(trans, root, inode);
2798         }
2799         return ret;
2800 }
2801
2802
2803 /* helper to check if there is any shared block in the path */
2804 static int check_path_shared(struct btrfs_root *root,
2805                              struct btrfs_path *path)
2806 {
2807         struct extent_buffer *eb;
2808         int level;
2809         u64 refs = 1;
2810
2811         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2812                 int ret;
2813
2814                 if (!path->nodes[level])
2815                         break;
2816                 eb = path->nodes[level];
2817                 if (!btrfs_block_can_be_shared(root, eb))
2818                         continue;
2819                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2820                                                &refs, NULL);
2821                 if (refs > 1)
2822                         return 1;
2823         }
2824         return 0;
2825 }
2826
2827 /*
2828  * helper to start transaction for unlink and rmdir.
2829  *
2830  * unlink and rmdir are special in btrfs, they do not always free space.
2831  * so in enospc case, we should make sure they will free space before
2832  * allowing them to use the global metadata reservation.
2833  */
2834 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2835                                                        struct dentry *dentry)
2836 {
2837         struct btrfs_trans_handle *trans;
2838         struct btrfs_root *root = BTRFS_I(dir)->root;
2839         struct btrfs_path *path;
2840         struct btrfs_inode_ref *ref;
2841         struct btrfs_dir_item *di;
2842         struct inode *inode = dentry->d_inode;
2843         u64 index;
2844         int check_link = 1;
2845         int err = -ENOSPC;
2846         int ret;
2847         u64 ino = btrfs_ino(inode);
2848         u64 dir_ino = btrfs_ino(dir);
2849
2850         trans = btrfs_start_transaction(root, 10);
2851         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2852                 return trans;
2853
2854         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2855                 return ERR_PTR(-ENOSPC);
2856
2857         /* check if there is someone else holds reference */
2858         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2859                 return ERR_PTR(-ENOSPC);
2860
2861         if (atomic_read(&inode->i_count) > 2)
2862                 return ERR_PTR(-ENOSPC);
2863
2864         if (xchg(&root->fs_info->enospc_unlink, 1))
2865                 return ERR_PTR(-ENOSPC);
2866
2867         path = btrfs_alloc_path();
2868         if (!path) {
2869                 root->fs_info->enospc_unlink = 0;
2870                 return ERR_PTR(-ENOMEM);
2871         }
2872
2873         trans = btrfs_start_transaction(root, 0);
2874         if (IS_ERR(trans)) {
2875                 btrfs_free_path(path);
2876                 root->fs_info->enospc_unlink = 0;
2877                 return trans;
2878         }
2879
2880         path->skip_locking = 1;
2881         path->search_commit_root = 1;
2882
2883         ret = btrfs_lookup_inode(trans, root, path,
2884                                 &BTRFS_I(dir)->location, 0);
2885         if (ret < 0) {
2886                 err = ret;
2887                 goto out;
2888         }
2889         if (ret == 0) {
2890                 if (check_path_shared(root, path))
2891                         goto out;
2892         } else {
2893                 check_link = 0;
2894         }
2895         btrfs_release_path(path);
2896
2897         ret = btrfs_lookup_inode(trans, root, path,
2898                                 &BTRFS_I(inode)->location, 0);
2899         if (ret < 0) {
2900                 err = ret;
2901                 goto out;
2902         }
2903         if (ret == 0) {
2904                 if (check_path_shared(root, path))
2905                         goto out;
2906         } else {
2907                 check_link = 0;
2908         }
2909         btrfs_release_path(path);
2910
2911         if (ret == 0 && S_ISREG(inode->i_mode)) {
2912                 ret = btrfs_lookup_file_extent(trans, root, path,
2913                                                ino, (u64)-1, 0);
2914                 if (ret < 0) {
2915                         err = ret;
2916                         goto out;
2917                 }
2918                 BUG_ON(ret == 0);
2919                 if (check_path_shared(root, path))
2920                         goto out;
2921                 btrfs_release_path(path);
2922         }
2923
2924         if (!check_link) {
2925                 err = 0;
2926                 goto out;
2927         }
2928
2929         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2930                                 dentry->d_name.name, dentry->d_name.len, 0);
2931         if (IS_ERR(di)) {
2932                 err = PTR_ERR(di);
2933                 goto out;
2934         }
2935         if (di) {
2936                 if (check_path_shared(root, path))
2937                         goto out;
2938         } else {
2939                 err = 0;
2940                 goto out;
2941         }
2942         btrfs_release_path(path);
2943
2944         ref = btrfs_lookup_inode_ref(trans, root, path,
2945                                 dentry->d_name.name, dentry->d_name.len,
2946                                 ino, dir_ino, 0);
2947         if (IS_ERR(ref)) {
2948                 err = PTR_ERR(ref);
2949                 goto out;
2950         }
2951         BUG_ON(!ref);
2952         if (check_path_shared(root, path))
2953                 goto out;
2954         index = btrfs_inode_ref_index(path->nodes[0], ref);
2955         btrfs_release_path(path);
2956
2957         /*
2958          * This is a commit root search, if we can lookup inode item and other
2959          * relative items in the commit root, it means the transaction of
2960          * dir/file creation has been committed, and the dir index item that we
2961          * delay to insert has also been inserted into the commit root. So
2962          * we needn't worry about the delayed insertion of the dir index item
2963          * here.
2964          */
2965         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2966                                 dentry->d_name.name, dentry->d_name.len, 0);
2967         if (IS_ERR(di)) {
2968                 err = PTR_ERR(di);
2969                 goto out;
2970         }
2971         BUG_ON(ret == -ENOENT);
2972         if (check_path_shared(root, path))
2973                 goto out;
2974
2975         err = 0;
2976 out:
2977         btrfs_free_path(path);
2978         if (err) {
2979                 btrfs_end_transaction(trans, root);
2980                 root->fs_info->enospc_unlink = 0;
2981                 return ERR_PTR(err);
2982         }
2983
2984         trans->block_rsv = &root->fs_info->global_block_rsv;
2985         return trans;
2986 }
2987
2988 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2989                                struct btrfs_root *root)
2990 {
2991         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2992                 BUG_ON(!root->fs_info->enospc_unlink);
2993                 root->fs_info->enospc_unlink = 0;
2994         }
2995         btrfs_end_transaction_throttle(trans, root);
2996 }
2997
2998 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2999 {
3000         struct btrfs_root *root = BTRFS_I(dir)->root;
3001         struct btrfs_trans_handle *trans;
3002         struct inode *inode = dentry->d_inode;
3003         int ret;
3004         unsigned long nr = 0;
3005
3006         trans = __unlink_start_trans(dir, dentry);
3007         if (IS_ERR(trans))
3008                 return PTR_ERR(trans);
3009
3010         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3011
3012         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3013                                  dentry->d_name.name, dentry->d_name.len);
3014         BUG_ON(ret);
3015
3016         if (inode->i_nlink == 0) {
3017                 ret = btrfs_orphan_add(trans, inode);
3018                 BUG_ON(ret);
3019         }
3020
3021         nr = trans->blocks_used;
3022         __unlink_end_trans(trans, root);
3023         btrfs_btree_balance_dirty(root, nr);
3024         return ret;
3025 }
3026
3027 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3028                         struct btrfs_root *root,
3029                         struct inode *dir, u64 objectid,
3030                         const char *name, int name_len)
3031 {
3032         struct btrfs_path *path;
3033         struct extent_buffer *leaf;
3034         struct btrfs_dir_item *di;
3035         struct btrfs_key key;
3036         u64 index;
3037         int ret;
3038         u64 dir_ino = btrfs_ino(dir);
3039
3040         path = btrfs_alloc_path();
3041         if (!path)
3042                 return -ENOMEM;
3043
3044         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3045                                    name, name_len, -1);
3046         BUG_ON(IS_ERR_OR_NULL(di));
3047
3048         leaf = path->nodes[0];
3049         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3050         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3051         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3052         BUG_ON(ret);
3053         btrfs_release_path(path);
3054
3055         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3056                                  objectid, root->root_key.objectid,
3057                                  dir_ino, &index, name, name_len);
3058         if (ret < 0) {
3059                 BUG_ON(ret != -ENOENT);
3060                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3061                                                  name, name_len);
3062                 BUG_ON(IS_ERR_OR_NULL(di));
3063
3064                 leaf = path->nodes[0];
3065                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3066                 btrfs_release_path(path);
3067                 index = key.offset;
3068         }
3069         btrfs_release_path(path);
3070
3071         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3072         BUG_ON(ret);
3073
3074         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3075         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3076         ret = btrfs_update_inode(trans, root, dir);
3077         BUG_ON(ret);
3078
3079         btrfs_free_path(path);
3080         return 0;
3081 }
3082
3083 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3084 {
3085         struct inode *inode = dentry->d_inode;
3086         int err = 0;
3087         struct btrfs_root *root = BTRFS_I(dir)->root;
3088         struct btrfs_trans_handle *trans;
3089         unsigned long nr = 0;
3090
3091         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3092             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3093                 return -ENOTEMPTY;
3094
3095         trans = __unlink_start_trans(dir, dentry);
3096         if (IS_ERR(trans))
3097                 return PTR_ERR(trans);
3098
3099         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3100                 err = btrfs_unlink_subvol(trans, root, dir,
3101                                           BTRFS_I(inode)->location.objectid,
3102                                           dentry->d_name.name,
3103                                           dentry->d_name.len);
3104                 goto out;
3105         }
3106
3107         err = btrfs_orphan_add(trans, inode);
3108         if (err)
3109                 goto out;
3110
3111         /* now the directory is empty */
3112         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3113                                  dentry->d_name.name, dentry->d_name.len);
3114         if (!err)
3115                 btrfs_i_size_write(inode, 0);
3116 out:
3117         nr = trans->blocks_used;
3118         __unlink_end_trans(trans, root);
3119         btrfs_btree_balance_dirty(root, nr);
3120
3121         return err;
3122 }
3123
3124 /*
3125  * this can truncate away extent items, csum items and directory items.
3126  * It starts at a high offset and removes keys until it can't find
3127  * any higher than new_size
3128  *
3129  * csum items that cross the new i_size are truncated to the new size
3130  * as well.
3131  *
3132  * min_type is the minimum key type to truncate down to.  If set to 0, this
3133  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3134  */
3135 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3136                                struct btrfs_root *root,
3137                                struct inode *inode,
3138                                u64 new_size, u32 min_type)
3139 {
3140         struct btrfs_path *path;
3141         struct extent_buffer *leaf;
3142         struct btrfs_file_extent_item *fi;
3143         struct btrfs_key key;
3144         struct btrfs_key found_key;
3145         u64 extent_start = 0;
3146         u64 extent_num_bytes = 0;
3147         u64 extent_offset = 0;
3148         u64 item_end = 0;
3149         u64 mask = root->sectorsize - 1;
3150         u32 found_type = (u8)-1;
3151         int found_extent;
3152         int del_item;
3153         int pending_del_nr = 0;
3154         int pending_del_slot = 0;
3155         int extent_type = -1;
3156         int encoding;
3157         int ret;
3158         int err = 0;
3159         u64 ino = btrfs_ino(inode);
3160
3161         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3162
3163         if (root->ref_cows || root == root->fs_info->tree_root)
3164                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3165
3166         /*
3167          * This function is also used to drop the items in the log tree before
3168          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3169          * it is used to drop the loged items. So we shouldn't kill the delayed
3170          * items.
3171          */
3172         if (min_type == 0 && root == BTRFS_I(inode)->root)
3173                 btrfs_kill_delayed_inode_items(inode);
3174
3175         path = btrfs_alloc_path();
3176         BUG_ON(!path);
3177         path->reada = -1;
3178
3179         key.objectid = ino;
3180         key.offset = (u64)-1;
3181         key.type = (u8)-1;
3182
3183 search_again:
3184         path->leave_spinning = 1;
3185         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3186         if (ret < 0) {
3187                 err = ret;
3188                 goto out;
3189         }
3190
3191         if (ret > 0) {
3192                 /* there are no items in the tree for us to truncate, we're
3193                  * done
3194                  */
3195                 if (path->slots[0] == 0)
3196                         goto out;
3197                 path->slots[0]--;
3198         }
3199
3200         while (1) {
3201                 fi = NULL;
3202                 leaf = path->nodes[0];
3203                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3204                 found_type = btrfs_key_type(&found_key);
3205                 encoding = 0;
3206
3207                 if (found_key.objectid != ino)
3208                         break;
3209
3210                 if (found_type < min_type)
3211                         break;
3212
3213                 item_end = found_key.offset;
3214                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3215                         fi = btrfs_item_ptr(leaf, path->slots[0],
3216                                             struct btrfs_file_extent_item);
3217                         extent_type = btrfs_file_extent_type(leaf, fi);
3218                         encoding = btrfs_file_extent_compression(leaf, fi);
3219                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3220                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3221
3222                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3223                                 item_end +=
3224                                     btrfs_file_extent_num_bytes(leaf, fi);
3225                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3226                                 item_end += btrfs_file_extent_inline_len(leaf,
3227                                                                          fi);
3228                         }
3229                         item_end--;
3230                 }
3231                 if (found_type > min_type) {
3232                         del_item = 1;
3233                 } else {
3234                         if (item_end < new_size)
3235                                 break;
3236                         if (found_key.offset >= new_size)
3237                                 del_item = 1;
3238                         else
3239                                 del_item = 0;
3240                 }
3241                 found_extent = 0;
3242                 /* FIXME, shrink the extent if the ref count is only 1 */
3243                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3244                         goto delete;
3245
3246                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3247                         u64 num_dec;
3248                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3249                         if (!del_item && !encoding) {
3250                                 u64 orig_num_bytes =
3251                                         btrfs_file_extent_num_bytes(leaf, fi);
3252                                 extent_num_bytes = new_size -
3253                                         found_key.offset + root->sectorsize - 1;
3254                                 extent_num_bytes = extent_num_bytes &
3255                                         ~((u64)root->sectorsize - 1);
3256                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3257                                                          extent_num_bytes);
3258                                 num_dec = (orig_num_bytes -
3259                                            extent_num_bytes);
3260                                 if (root->ref_cows && extent_start != 0)
3261                                         inode_sub_bytes(inode, num_dec);
3262                                 btrfs_mark_buffer_dirty(leaf);
3263                         } else {
3264                                 extent_num_bytes =
3265                                         btrfs_file_extent_disk_num_bytes(leaf,
3266                                                                          fi);
3267                                 extent_offset = found_key.offset -
3268                                         btrfs_file_extent_offset(leaf, fi);
3269
3270                                 /* FIXME blocksize != 4096 */
3271                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3272                                 if (extent_start != 0) {
3273                                         found_extent = 1;
3274                                         if (root->ref_cows)
3275                                                 inode_sub_bytes(inode, num_dec);
3276                                 }
3277                         }
3278                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3279                         /*
3280                          * we can't truncate inline items that have had
3281                          * special encodings
3282                          */
3283                         if (!del_item &&
3284                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3285                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3286                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3287                                 u32 size = new_size - found_key.offset;
3288
3289                                 if (root->ref_cows) {
3290                                         inode_sub_bytes(inode, item_end + 1 -
3291                                                         new_size);
3292                                 }
3293                                 size =
3294                                     btrfs_file_extent_calc_inline_size(size);
3295                                 ret = btrfs_truncate_item(trans, root, path,
3296                                                           size, 1);
3297                         } else if (root->ref_cows) {
3298                                 inode_sub_bytes(inode, item_end + 1 -
3299                                                 found_key.offset);
3300                         }
3301                 }
3302 delete:
3303                 if (del_item) {
3304                         if (!pending_del_nr) {
3305                                 /* no pending yet, add ourselves */
3306                                 pending_del_slot = path->slots[0];
3307                                 pending_del_nr = 1;
3308                         } else if (pending_del_nr &&
3309                                    path->slots[0] + 1 == pending_del_slot) {
3310                                 /* hop on the pending chunk */
3311                                 pending_del_nr++;
3312                                 pending_del_slot = path->slots[0];
3313                         } else {
3314                                 BUG();
3315                         }
3316                 } else {
3317                         break;
3318                 }
3319                 if (found_extent && (root->ref_cows ||
3320                                      root == root->fs_info->tree_root)) {
3321                         btrfs_set_path_blocking(path);
3322                         ret = btrfs_free_extent(trans, root, extent_start,
3323                                                 extent_num_bytes, 0,
3324                                                 btrfs_header_owner(leaf),
3325                                                 ino, extent_offset);
3326                         BUG_ON(ret);
3327                 }
3328
3329                 if (found_type == BTRFS_INODE_ITEM_KEY)
3330                         break;
3331
3332                 if (path->slots[0] == 0 ||
3333                     path->slots[0] != pending_del_slot) {
3334                         if (root->ref_cows &&
3335                             BTRFS_I(inode)->location.objectid !=
3336                                                 BTRFS_FREE_INO_OBJECTID) {
3337                                 err = -EAGAIN;
3338                                 goto out;
3339                         }
3340                         if (pending_del_nr) {
3341                                 ret = btrfs_del_items(trans, root, path,
3342                                                 pending_del_slot,
3343                                                 pending_del_nr);
3344                                 BUG_ON(ret);
3345                                 pending_del_nr = 0;
3346                         }
3347                         btrfs_release_path(path);
3348                         goto search_again;
3349                 } else {
3350                         path->slots[0]--;
3351                 }
3352         }
3353 out:
3354         if (pending_del_nr) {
3355                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3356                                       pending_del_nr);
3357                 BUG_ON(ret);
3358         }
3359         btrfs_free_path(path);
3360         return err;
3361 }
3362
3363 /*
3364  * taken from block_truncate_page, but does cow as it zeros out
3365  * any bytes left in the last page in the file.
3366  */
3367 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3368 {
3369         struct inode *inode = mapping->host;
3370         struct btrfs_root *root = BTRFS_I(inode)->root;
3371         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3372         struct btrfs_ordered_extent *ordered;
3373         struct extent_state *cached_state = NULL;
3374         char *kaddr;
3375         u32 blocksize = root->sectorsize;
3376         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3377         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3378         struct page *page;
3379         int ret = 0;
3380         u64 page_start;
3381         u64 page_end;
3382
3383         if ((offset & (blocksize - 1)) == 0)
3384                 goto out;
3385         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3386         if (ret)
3387                 goto out;
3388
3389         ret = -ENOMEM;
3390 again:
3391         page = grab_cache_page(mapping, index);
3392         if (!page) {
3393                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3394                 goto out;
3395         }
3396
3397         page_start = page_offset(page);
3398         page_end = page_start + PAGE_CACHE_SIZE - 1;
3399
3400         if (!PageUptodate(page)) {
3401                 ret = btrfs_readpage(NULL, page);
3402                 lock_page(page);
3403                 if (page->mapping != mapping) {
3404                         unlock_page(page);
3405                         page_cache_release(page);
3406                         goto again;
3407                 }
3408                 if (!PageUptodate(page)) {
3409                         ret = -EIO;
3410                         goto out_unlock;
3411                 }
3412         }
3413         wait_on_page_writeback(page);
3414
3415         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3416                          GFP_NOFS);
3417         set_page_extent_mapped(page);
3418
3419         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3420         if (ordered) {
3421                 unlock_extent_cached(io_tree, page_start, page_end,
3422                                      &cached_state, GFP_NOFS);
3423                 unlock_page(page);
3424                 page_cache_release(page);
3425                 btrfs_start_ordered_extent(inode, ordered, 1);
3426                 btrfs_put_ordered_extent(ordered);
3427                 goto again;
3428         }
3429
3430         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3431                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3432                           0, 0, &cached_state, GFP_NOFS);
3433
3434         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3435                                         &cached_state);
3436         if (ret) {
3437                 unlock_extent_cached(io_tree, page_start, page_end,
3438                                      &cached_state, GFP_NOFS);
3439                 goto out_unlock;
3440         }
3441
3442         ret = 0;
3443         if (offset != PAGE_CACHE_SIZE) {
3444                 kaddr = kmap(page);
3445                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3446                 flush_dcache_page(page);
3447                 kunmap(page);
3448         }
3449         ClearPageChecked(page);
3450         set_page_dirty(page);
3451         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3452                              GFP_NOFS);
3453
3454 out_unlock:
3455         if (ret)
3456                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3457         unlock_page(page);
3458         page_cache_release(page);
3459 out:
3460         return ret;
3461 }
3462
3463 /*
3464  * This function puts in dummy file extents for the area we're creating a hole
3465  * for.  So if we are truncating this file to a larger size we need to insert
3466  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3467  * the range between oldsize and size
3468  */
3469 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3470 {
3471         struct btrfs_trans_handle *trans;
3472         struct btrfs_root *root = BTRFS_I(inode)->root;
3473         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3474         struct extent_map *em = NULL;
3475         struct extent_state *cached_state = NULL;
3476         u64 mask = root->sectorsize - 1;
3477         u64 hole_start = (oldsize + mask) & ~mask;
3478         u64 block_end = (size + mask) & ~mask;
3479         u64 last_byte;
3480         u64 cur_offset;
3481         u64 hole_size;
3482         int err = 0;
3483
3484         if (size <= hole_start)
3485                 return 0;
3486
3487         while (1) {
3488                 struct btrfs_ordered_extent *ordered;
3489                 btrfs_wait_ordered_range(inode, hole_start,
3490                                          block_end - hole_start);
3491                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3492                                  &cached_state, GFP_NOFS);
3493                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3494                 if (!ordered)
3495                         break;
3496                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3497                                      &cached_state, GFP_NOFS);
3498                 btrfs_put_ordered_extent(ordered);
3499         }
3500
3501         cur_offset = hole_start;
3502         while (1) {
3503                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3504                                 block_end - cur_offset, 0);
3505                 BUG_ON(IS_ERR_OR_NULL(em));
3506                 last_byte = min(extent_map_end(em), block_end);
3507                 last_byte = (last_byte + mask) & ~mask;
3508                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3509                         u64 hint_byte = 0;
3510                         hole_size = last_byte - cur_offset;
3511
3512                         trans = btrfs_start_transaction(root, 2);
3513                         if (IS_ERR(trans)) {
3514                                 err = PTR_ERR(trans);
3515                                 break;
3516                         }
3517
3518                         err = btrfs_drop_extents(trans, inode, cur_offset,
3519                                                  cur_offset + hole_size,
3520                                                  &hint_byte, 1);
3521                         if (err)
3522                                 break;
3523
3524                         err = btrfs_insert_file_extent(trans, root,
3525                                         btrfs_ino(inode), cur_offset, 0,
3526                                         0, hole_size, 0, hole_size,
3527                                         0, 0, 0);
3528                         if (err)
3529                                 break;
3530
3531                         btrfs_drop_extent_cache(inode, hole_start,
3532                                         last_byte - 1, 0);
3533
3534                         btrfs_end_transaction(trans, root);
3535                 }
3536                 free_extent_map(em);
3537                 em = NULL;
3538                 cur_offset = last_byte;
3539                 if (cur_offset >= block_end)
3540                         break;
3541         }
3542
3543         free_extent_map(em);
3544         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3545                              GFP_NOFS);
3546         return err;
3547 }
3548
3549 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3550 {
3551         loff_t oldsize = i_size_read(inode);
3552         int ret;
3553
3554         if (newsize == oldsize)
3555                 return 0;
3556
3557         if (newsize > oldsize) {
3558                 i_size_write(inode, newsize);
3559                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3560                 truncate_pagecache(inode, oldsize, newsize);
3561                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3562                 if (ret) {
3563                         btrfs_setsize(inode, oldsize);
3564                         return ret;
3565                 }
3566
3567                 mark_inode_dirty(inode);
3568         } else {
3569
3570                 /*
3571                  * We're truncating a file that used to have good data down to
3572                  * zero. Make sure it gets into the ordered flush list so that
3573                  * any new writes get down to disk quickly.
3574                  */
3575                 if (newsize == 0)
3576                         BTRFS_I(inode)->ordered_data_close = 1;
3577
3578                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3579                 truncate_setsize(inode, newsize);
3580                 ret = btrfs_truncate(inode);
3581         }
3582
3583         return ret;
3584 }
3585
3586 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3587 {
3588         struct inode *inode = dentry->d_inode;
3589         struct btrfs_root *root = BTRFS_I(inode)->root;
3590         int err;
3591
3592         if (btrfs_root_readonly(root))
3593                 return -EROFS;
3594
3595         err = inode_change_ok(inode, attr);
3596         if (err)
3597                 return err;
3598
3599         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3600                 err = btrfs_setsize(inode, attr->ia_size);
3601                 if (err)
3602                         return err;
3603         }
3604
3605         if (attr->ia_valid) {
3606                 setattr_copy(inode, attr);
3607                 mark_inode_dirty(inode);
3608
3609                 if (attr->ia_valid & ATTR_MODE)
3610                         err = btrfs_acl_chmod(inode);
3611         }
3612
3613         return err;
3614 }
3615
3616 void btrfs_evict_inode(struct inode *inode)
3617 {
3618         struct btrfs_trans_handle *trans;
3619         struct btrfs_root *root = BTRFS_I(inode)->root;
3620         unsigned long nr;
3621         int ret;
3622
3623         trace_btrfs_inode_evict(inode);
3624
3625         truncate_inode_pages(&inode->i_data, 0);
3626         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3627                                is_free_space_inode(root, inode)))
3628                 goto no_delete;
3629
3630         if (is_bad_inode(inode)) {
3631                 btrfs_orphan_del(NULL, inode);
3632                 goto no_delete;
3633         }
3634         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3635         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3636
3637         if (root->fs_info->log_root_recovering) {
3638                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3639                 goto no_delete;
3640         }
3641
3642         if (inode->i_nlink > 0) {
3643                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3644                 goto no_delete;
3645         }
3646
3647         btrfs_i_size_write(inode, 0);
3648
3649         while (1) {
3650                 trans = btrfs_join_transaction(root);
3651                 BUG_ON(IS_ERR(trans));
3652                 trans->block_rsv = root->orphan_block_rsv;
3653
3654                 ret = btrfs_block_rsv_check(trans, root,
3655                                             root->orphan_block_rsv, 0, 5);
3656                 if (ret) {
3657                         BUG_ON(ret != -EAGAIN);
3658                         ret = btrfs_commit_transaction(trans, root);
3659                         BUG_ON(ret);
3660                         continue;
3661                 }
3662
3663                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3664                 if (ret != -EAGAIN)
3665                         break;
3666
3667                 nr = trans->blocks_used;
3668                 btrfs_end_transaction(trans, root);
3669                 trans = NULL;
3670                 btrfs_btree_balance_dirty(root, nr);
3671
3672         }
3673
3674         if (ret == 0) {
3675                 ret = btrfs_orphan_del(trans, inode);
3676                 BUG_ON(ret);
3677         }
3678
3679         if (!(root == root->fs_info->tree_root ||
3680               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3681                 btrfs_return_ino(root, btrfs_ino(inode));
3682
3683         nr = trans->blocks_used;
3684         btrfs_end_transaction(trans, root);
3685         btrfs_btree_balance_dirty(root, nr);
3686 no_delete:
3687         end_writeback(inode);
3688         return;
3689 }
3690
3691 /*
3692  * this returns the key found in the dir entry in the location pointer.
3693  * If no dir entries were found, location->objectid is 0.
3694  */
3695 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3696                                struct btrfs_key *location)
3697 {
3698         const char *name = dentry->d_name.name;
3699         int namelen = dentry->d_name.len;
3700         struct btrfs_dir_item *di;
3701         struct btrfs_path *path;
3702         struct btrfs_root *root = BTRFS_I(dir)->root;
3703         int ret = 0;
3704
3705         path = btrfs_alloc_path();
3706         BUG_ON(!path);
3707
3708         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3709                                     namelen, 0);
3710         if (IS_ERR(di))
3711                 ret = PTR_ERR(di);
3712
3713         if (IS_ERR_OR_NULL(di))
3714                 goto out_err;
3715
3716         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3717 out:
3718         btrfs_free_path(path);
3719         return ret;
3720 out_err:
3721         location->objectid = 0;
3722         goto out;
3723 }
3724
3725 /*
3726  * when we hit a tree root in a directory, the btrfs part of the inode
3727  * needs to be changed to reflect the root directory of the tree root.  This
3728  * is kind of like crossing a mount point.
3729  */
3730 static int fixup_tree_root_location(struct btrfs_root *root,
3731                                     struct inode *dir,
3732                                     struct dentry *dentry,
3733                                     struct btrfs_key *location,
3734                                     struct btrfs_root **sub_root)
3735 {
3736         struct btrfs_path *path;
3737         struct btrfs_root *new_root;
3738         struct btrfs_root_ref *ref;
3739         struct extent_buffer *leaf;
3740         int ret;
3741         int err = 0;
3742
3743         path = btrfs_alloc_path();
3744         if (!path) {
3745                 err = -ENOMEM;
3746                 goto out;
3747         }
3748
3749         err = -ENOENT;
3750         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3751                                   BTRFS_I(dir)->root->root_key.objectid,
3752                                   location->objectid);
3753         if (ret) {
3754                 if (ret < 0)
3755                         err = ret;
3756                 goto out;
3757         }
3758
3759         leaf = path->nodes[0];
3760         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3761         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3762             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3763                 goto out;
3764
3765         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3766                                    (unsigned long)(ref + 1),
3767                                    dentry->d_name.len);
3768         if (ret)
3769                 goto out;
3770
3771         btrfs_release_path(path);
3772
3773         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3774         if (IS_ERR(new_root)) {
3775                 err = PTR_ERR(new_root);
3776                 goto out;
3777         }
3778
3779         if (btrfs_root_refs(&new_root->root_item) == 0) {
3780                 err = -ENOENT;
3781                 goto out;
3782         }
3783
3784         *sub_root = new_root;
3785         location->objectid = btrfs_root_dirid(&new_root->root_item);
3786         location->type = BTRFS_INODE_ITEM_KEY;
3787         location->offset = 0;
3788         err = 0;
3789 out:
3790         btrfs_free_path(path);
3791         return err;
3792 }
3793
3794 static void inode_tree_add(struct inode *inode)
3795 {
3796         struct btrfs_root *root = BTRFS_I(inode)->root;
3797         struct btrfs_inode *entry;
3798         struct rb_node **p;
3799         struct rb_node *parent;
3800         u64 ino = btrfs_ino(inode);
3801 again:
3802         p = &root->inode_tree.rb_node;
3803         parent = NULL;
3804
3805         if (inode_unhashed(inode))
3806                 return;
3807
3808         spin_lock(&root->inode_lock);
3809         while (*p) {
3810                 parent = *p;
3811                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3812
3813                 if (ino < btrfs_ino(&entry->vfs_inode))
3814                         p = &parent->rb_left;
3815                 else if (ino > btrfs_ino(&entry->vfs_inode))
3816                         p = &parent->rb_right;
3817                 else {
3818                         WARN_ON(!(entry->vfs_inode.i_state &
3819                                   (I_WILL_FREE | I_FREEING)));
3820                         rb_erase(parent, &root->inode_tree);
3821                         RB_CLEAR_NODE(parent);
3822                         spin_unlock(&root->inode_lock);
3823                         goto again;
3824                 }
3825         }
3826         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3827         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3828         spin_unlock(&root->inode_lock);
3829 }
3830
3831 static void inode_tree_del(struct inode *inode)
3832 {
3833         struct btrfs_root *root = BTRFS_I(inode)->root;
3834         int empty = 0;
3835
3836         spin_lock(&root->inode_lock);
3837         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3838                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3839                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3840                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3841         }
3842         spin_unlock(&root->inode_lock);
3843
3844         /*
3845          * Free space cache has inodes in the tree root, but the tree root has a
3846          * root_refs of 0, so this could end up dropping the tree root as a
3847          * snapshot, so we need the extra !root->fs_info->tree_root check to
3848          * make sure we don't drop it.
3849          */
3850         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3851             root != root->fs_info->tree_root) {
3852                 synchronize_srcu(&root->fs_info->subvol_srcu);
3853                 spin_lock(&root->inode_lock);
3854                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3855                 spin_unlock(&root->inode_lock);
3856                 if (empty)
3857                         btrfs_add_dead_root(root);
3858         }
3859 }
3860
3861 int btrfs_invalidate_inodes(struct btrfs_root *root)
3862 {
3863         struct rb_node *node;
3864         struct rb_node *prev;
3865         struct btrfs_inode *entry;
3866         struct inode *inode;
3867         u64 objectid = 0;
3868
3869         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3870
3871         spin_lock(&root->inode_lock);
3872 again:
3873         node = root->inode_tree.rb_node;
3874         prev = NULL;
3875         while (node) {
3876                 prev = node;
3877                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3878
3879                 if (objectid < btrfs_ino(&entry->vfs_inode))
3880                         node = node->rb_left;
3881                 else if (objectid > btrfs_ino(&entry->vfs_inode))
3882                         node = node->rb_right;
3883                 else
3884                         break;
3885         }
3886         if (!node) {
3887                 while (prev) {
3888                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3889                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3890                                 node = prev;
3891                                 break;
3892                         }
3893                         prev = rb_next(prev);
3894                 }
3895         }
3896         while (node) {
3897                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3898                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3899                 inode = igrab(&entry->vfs_inode);
3900                 if (inode) {
3901                         spin_unlock(&root->inode_lock);
3902                         if (atomic_read(&inode->i_count) > 1)
3903                                 d_prune_aliases(inode);
3904                         /*
3905                          * btrfs_drop_inode will have it removed from
3906                          * the inode cache when its usage count
3907                          * hits zero.
3908                          */
3909                         iput(inode);
3910                         cond_resched();
3911                         spin_lock(&root->inode_lock);
3912                         goto again;
3913                 }
3914
3915                 if (cond_resched_lock(&root->inode_lock))
3916                         goto again;
3917
3918                 node = rb_next(node);
3919         }
3920         spin_unlock(&root->inode_lock);
3921         return 0;
3922 }
3923
3924 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3925 {
3926         struct btrfs_iget_args *args = p;
3927         inode->i_ino = args->ino;
3928         BTRFS_I(inode)->root = args->root;
3929         btrfs_set_inode_space_info(args->root, inode);
3930         return 0;
3931 }
3932
3933 static int btrfs_find_actor(struct inode *inode, void *opaque)
3934 {
3935         struct btrfs_iget_args *args = opaque;
3936         return args->ino == btrfs_ino(inode) &&
3937                 args->root == BTRFS_I(inode)->root;
3938 }
3939
3940 static struct inode *btrfs_iget_locked(struct super_block *s,
3941                                        u64 objectid,
3942                                        struct btrfs_root *root)
3943 {
3944         struct inode *inode;
3945         struct btrfs_iget_args args;
3946         args.ino = objectid;
3947         args.root = root;
3948
3949         inode = iget5_locked(s, objectid, btrfs_find_actor,
3950                              btrfs_init_locked_inode,
3951                              (void *)&args);
3952         return inode;
3953 }
3954
3955 /* Get an inode object given its location and corresponding root.
3956  * Returns in *is_new if the inode was read from disk
3957  */
3958 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3959                          struct btrfs_root *root, int *new)
3960 {
3961         struct inode *inode;
3962
3963         inode = btrfs_iget_locked(s, location->objectid, root);
3964         if (!inode)
3965                 return ERR_PTR(-ENOMEM);
3966
3967         if (inode->i_state & I_NEW) {
3968                 BTRFS_I(inode)->root = root;
3969                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3970                 btrfs_read_locked_inode(inode);
3971                 inode_tree_add(inode);
3972                 unlock_new_inode(inode);
3973                 if (new)
3974                         *new = 1;
3975         }
3976
3977         return inode;
3978 }
3979
3980 static struct inode *new_simple_dir(struct super_block *s,
3981                                     struct btrfs_key *key,
3982                                     struct btrfs_root *root)
3983 {
3984         struct inode *inode = new_inode(s);
3985
3986         if (!inode)
3987                 return ERR_PTR(-ENOMEM);
3988
3989         BTRFS_I(inode)->root = root;
3990         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3991         BTRFS_I(inode)->dummy_inode = 1;
3992
3993         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3994         inode->i_op = &simple_dir_inode_operations;
3995         inode->i_fop = &simple_dir_operations;
3996         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3997         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3998
3999         return inode;
4000 }
4001
4002 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4003 {
4004         struct inode *inode;
4005         struct btrfs_root *root = BTRFS_I(dir)->root;
4006         struct btrfs_root *sub_root = root;
4007         struct btrfs_key location;
4008         int index;
4009         int ret;
4010
4011         if (dentry->d_name.len > BTRFS_NAME_LEN)
4012                 return ERR_PTR(-ENAMETOOLONG);
4013
4014         ret = btrfs_inode_by_name(dir, dentry, &location);
4015
4016         if (ret < 0)
4017                 return ERR_PTR(ret);
4018
4019         if (location.objectid == 0)
4020                 return NULL;
4021
4022         if (location.type == BTRFS_INODE_ITEM_KEY) {
4023                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4024                 return inode;
4025         }
4026
4027         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4028
4029         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4030         ret = fixup_tree_root_location(root, dir, dentry,
4031                                        &location, &sub_root);
4032         if (ret < 0) {
4033                 if (ret != -ENOENT)
4034                         inode = ERR_PTR(ret);
4035                 else
4036                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4037         } else {
4038                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4039         }
4040         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4041
4042         if (!IS_ERR(inode) && root != sub_root) {
4043                 down_read(&root->fs_info->cleanup_work_sem);
4044                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4045                         ret = btrfs_orphan_cleanup(sub_root);
4046                 up_read(&root->fs_info->cleanup_work_sem);
4047                 if (ret)
4048                         inode = ERR_PTR(ret);
4049         }
4050
4051         return inode;
4052 }
4053
4054 static int btrfs_dentry_delete(const struct dentry *dentry)
4055 {
4056         struct btrfs_root *root;
4057
4058         if (!dentry->d_inode && !IS_ROOT(dentry))
4059                 dentry = dentry->d_parent;
4060
4061         if (dentry->d_inode) {
4062                 root = BTRFS_I(dentry->d_inode)->root;
4063                 if (btrfs_root_refs(&root->root_item) == 0)
4064                         return 1;
4065         }
4066         return 0;
4067 }
4068
4069 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4070                                    struct nameidata *nd)
4071 {
4072         struct inode *inode;
4073
4074         inode = btrfs_lookup_dentry(dir, dentry);
4075         if (IS_ERR(inode))
4076                 return ERR_CAST(inode);
4077
4078         return d_splice_alias(inode, dentry);
4079 }
4080
4081 unsigned char btrfs_filetype_table[] = {
4082         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4083 };
4084
4085 static int btrfs_real_readdir(struct file *filp, void *dirent,
4086                               filldir_t filldir)
4087 {
4088         struct inode *inode = filp->f_dentry->d_inode;
4089         struct btrfs_root *root = BTRFS_I(inode)->root;
4090         struct btrfs_item *item;
4091         struct btrfs_dir_item *di;
4092         struct btrfs_key key;
4093         struct btrfs_key found_key;
4094         struct btrfs_path *path;
4095         struct list_head ins_list;
4096         struct list_head del_list;
4097         int ret;
4098         struct extent_buffer *leaf;
4099         int slot;
4100         unsigned char d_type;
4101         int over = 0;
4102         u32 di_cur;
4103         u32 di_total;
4104         u32 di_len;
4105         int key_type = BTRFS_DIR_INDEX_KEY;
4106         char tmp_name[32];
4107         char *name_ptr;
4108         int name_len;
4109         int is_curr = 0;        /* filp->f_pos points to the current index? */
4110
4111         /* FIXME, use a real flag for deciding about the key type */
4112         if (root->fs_info->tree_root == root)
4113                 key_type = BTRFS_DIR_ITEM_KEY;
4114
4115         /* special case for "." */
4116         if (filp->f_pos == 0) {
4117                 over = filldir(dirent, ".", 1, 1, btrfs_ino(inode), DT_DIR);
4118                 if (over)
4119                         return 0;
4120                 filp->f_pos = 1;
4121         }
4122         /* special case for .., just use the back ref */
4123         if (filp->f_pos == 1) {
4124                 u64 pino = parent_ino(filp->f_path.dentry);
4125                 over = filldir(dirent, "..", 2,
4126                                2, pino, DT_DIR);
4127                 if (over)
4128                         return 0;
4129                 filp->f_pos = 2;
4130         }
4131         path = btrfs_alloc_path();
4132         if (!path)
4133                 return -ENOMEM;
4134
4135         path->reada = 1;
4136
4137         if (key_type == BTRFS_DIR_INDEX_KEY) {
4138                 INIT_LIST_HEAD(&ins_list);
4139                 INIT_LIST_HEAD(&del_list);
4140                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4141         }
4142
4143         btrfs_set_key_type(&key, key_type);
4144         key.offset = filp->f_pos;
4145         key.objectid = btrfs_ino(inode);
4146
4147         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4148         if (ret < 0)
4149                 goto err;
4150
4151         while (1) {
4152                 leaf = path->nodes[0];
4153                 slot = path->slots[0];
4154                 if (slot >= btrfs_header_nritems(leaf)) {
4155                         ret = btrfs_next_leaf(root, path);
4156                         if (ret < 0)
4157                                 goto err;
4158                         else if (ret > 0)
4159                                 break;
4160                         continue;
4161                 }
4162
4163                 item = btrfs_item_nr(leaf, slot);
4164                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4165
4166                 if (found_key.objectid != key.objectid)
4167                         break;
4168                 if (btrfs_key_type(&found_key) != key_type)
4169                         break;
4170                 if (found_key.offset < filp->f_pos)
4171                         goto next;
4172                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4173                     btrfs_should_delete_dir_index(&del_list,
4174                                                   found_key.offset))
4175                         goto next;
4176
4177                 filp->f_pos = found_key.offset;
4178                 is_curr = 1;
4179
4180                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4181                 di_cur = 0;
4182                 di_total = btrfs_item_size(leaf, item);
4183
4184                 while (di_cur < di_total) {
4185                         struct btrfs_key location;
4186
4187                         if (verify_dir_item(root, leaf, di))
4188                                 break;
4189
4190                         name_len = btrfs_dir_name_len(leaf, di);
4191                         if (name_len <= sizeof(tmp_name)) {
4192                                 name_ptr = tmp_name;
4193                         } else {
4194                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4195                                 if (!name_ptr) {
4196                                         ret = -ENOMEM;
4197                                         goto err;
4198                                 }
4199                         }
4200                         read_extent_buffer(leaf, name_ptr,
4201                                            (unsigned long)(di + 1), name_len);
4202
4203                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4204                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4205
4206                         /* is this a reference to our own snapshot? If so
4207                          * skip it
4208                          */
4209                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4210                             location.objectid == root->root_key.objectid) {
4211                                 over = 0;
4212                                 goto skip;
4213                         }
4214                         over = filldir(dirent, name_ptr, name_len,
4215                                        found_key.offset, location.objectid,
4216                                        d_type);
4217
4218 skip:
4219                         if (name_ptr != tmp_name)
4220                                 kfree(name_ptr);
4221
4222                         if (over)
4223                                 goto nopos;
4224                         di_len = btrfs_dir_name_len(leaf, di) +
4225                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4226                         di_cur += di_len;
4227                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4228                 }
4229 next:
4230                 path->slots[0]++;
4231         }
4232
4233         if (key_type == BTRFS_DIR_INDEX_KEY) {
4234                 if (is_curr)
4235                         filp->f_pos++;
4236                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4237                                                       &ins_list);
4238                 if (ret)
4239                         goto nopos;
4240         }
4241
4242         /* Reached end of directory/root. Bump pos past the last item. */
4243         if (key_type == BTRFS_DIR_INDEX_KEY)
4244                 /*
4245                  * 32-bit glibc will use getdents64, but then strtol -
4246                  * so the last number we can serve is this.
4247                  */
4248                 filp->f_pos = 0x7fffffff;
4249         else
4250                 filp->f_pos++;
4251 nopos:
4252         ret = 0;
4253 err:
4254         if (key_type == BTRFS_DIR_INDEX_KEY)
4255                 btrfs_put_delayed_items(&ins_list, &del_list);
4256         btrfs_free_path(path);
4257         return ret;
4258 }
4259
4260 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4261 {
4262         struct btrfs_root *root = BTRFS_I(inode)->root;
4263         struct btrfs_trans_handle *trans;
4264         int ret = 0;
4265         bool nolock = false;
4266
4267         if (BTRFS_I(inode)->dummy_inode)
4268                 return 0;
4269
4270         if (btrfs_fs_closing(root->fs_info) && is_free_space_inode(root, inode))
4271                 nolock = true;
4272
4273         if (wbc->sync_mode == WB_SYNC_ALL) {
4274                 if (nolock)
4275                         trans = btrfs_join_transaction_nolock(root);
4276                 else
4277                         trans = btrfs_join_transaction(root);
4278                 if (IS_ERR(trans))
4279                         return PTR_ERR(trans);
4280                 if (nolock)
4281                         ret = btrfs_end_transaction_nolock(trans, root);
4282                 else
4283                         ret = btrfs_commit_transaction(trans, root);
4284         }
4285         return ret;
4286 }
4287
4288 /*
4289  * This is somewhat expensive, updating the tree every time the
4290  * inode changes.  But, it is most likely to find the inode in cache.
4291  * FIXME, needs more benchmarking...there are no reasons other than performance
4292  * to keep or drop this code.
4293  */
4294 void btrfs_dirty_inode(struct inode *inode, int flags)
4295 {
4296         struct btrfs_root *root = BTRFS_I(inode)->root;
4297         struct btrfs_trans_handle *trans;
4298         int ret;
4299
4300         if (BTRFS_I(inode)->dummy_inode)
4301                 return;
4302
4303         trans = btrfs_join_transaction(root);
4304         BUG_ON(IS_ERR(trans));
4305
4306         ret = btrfs_update_inode(trans, root, inode);
4307         if (ret && ret == -ENOSPC) {
4308                 /* whoops, lets try again with the full transaction */
4309                 btrfs_end_transaction(trans, root);
4310                 trans = btrfs_start_transaction(root, 1);
4311                 if (IS_ERR(trans)) {
4312                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4313                                        "dirty  inode %llu error %ld\n",
4314                                        (unsigned long long)btrfs_ino(inode),
4315                                        PTR_ERR(trans));
4316                         return;
4317                 }
4318
4319                 ret = btrfs_update_inode(trans, root, inode);
4320                 if (ret) {
4321                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4322                                        "dirty  inode %llu error %d\n",
4323                                        (unsigned long long)btrfs_ino(inode),
4324                                        ret);
4325                 }
4326         }
4327         btrfs_end_transaction(trans, root);
4328         if (BTRFS_I(inode)->delayed_node)
4329                 btrfs_balance_delayed_items(root);
4330 }
4331
4332 /*
4333  * find the highest existing sequence number in a directory
4334  * and then set the in-memory index_cnt variable to reflect
4335  * free sequence numbers
4336  */
4337 static int btrfs_set_inode_index_count(struct inode *inode)
4338 {
4339         struct btrfs_root *root = BTRFS_I(inode)->root;
4340         struct btrfs_key key, found_key;
4341         struct btrfs_path *path;
4342         struct extent_buffer *leaf;
4343         int ret;
4344
4345         key.objectid = btrfs_ino(inode);
4346         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4347         key.offset = (u64)-1;
4348
4349         path = btrfs_alloc_path();
4350         if (!path)
4351                 return -ENOMEM;
4352
4353         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4354         if (ret < 0)
4355                 goto out;
4356         /* FIXME: we should be able to handle this */
4357         if (ret == 0)
4358                 goto out;
4359         ret = 0;
4360
4361         /*
4362          * MAGIC NUMBER EXPLANATION:
4363          * since we search a directory based on f_pos we have to start at 2
4364          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4365          * else has to start at 2
4366          */
4367         if (path->slots[0] == 0) {
4368                 BTRFS_I(inode)->index_cnt = 2;
4369                 goto out;
4370         }
4371
4372         path->slots[0]--;
4373
4374         leaf = path->nodes[0];
4375         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4376
4377         if (found_key.objectid != btrfs_ino(inode) ||
4378             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4379                 BTRFS_I(inode)->index_cnt = 2;
4380                 goto out;
4381         }
4382
4383         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4384 out:
4385         btrfs_free_path(path);
4386         return ret;
4387 }
4388
4389 /*
4390  * helper to find a free sequence number in a given directory.  This current
4391  * code is very simple, later versions will do smarter things in the btree
4392  */
4393 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4394 {
4395         int ret = 0;
4396
4397         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4398                 ret = btrfs_inode_delayed_dir_index_count(dir);
4399                 if (ret) {
4400                         ret = btrfs_set_inode_index_count(dir);
4401                         if (ret)
4402                                 return ret;
4403                 }
4404         }
4405
4406         *index = BTRFS_I(dir)->index_cnt;
4407         BTRFS_I(dir)->index_cnt++;
4408
4409         return ret;
4410 }
4411
4412 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4413                                      struct btrfs_root *root,
4414                                      struct inode *dir,
4415                                      const char *name, int name_len,
4416                                      u64 ref_objectid, u64 objectid, int mode,
4417                                      u64 *index)
4418 {
4419         struct inode *inode;
4420         struct btrfs_inode_item *inode_item;
4421         struct btrfs_key *location;
4422         struct btrfs_path *path;
4423         struct btrfs_inode_ref *ref;
4424         struct btrfs_key key[2];
4425         u32 sizes[2];
4426         unsigned long ptr;
4427         int ret;
4428         int owner;
4429
4430         path = btrfs_alloc_path();
4431         BUG_ON(!path);
4432
4433         inode = new_inode(root->fs_info->sb);
4434         if (!inode) {
4435                 btrfs_free_path(path);
4436                 return ERR_PTR(-ENOMEM);
4437         }
4438
4439         /*
4440          * we have to initialize this early, so we can reclaim the inode
4441          * number if we fail afterwards in this function.
4442          */
4443         inode->i_ino = objectid;
4444
4445         if (dir) {
4446                 trace_btrfs_inode_request(dir);
4447
4448                 ret = btrfs_set_inode_index(dir, index);
4449                 if (ret) {
4450                         btrfs_free_path(path);
4451                         iput(inode);
4452                         return ERR_PTR(ret);
4453                 }
4454         }
4455         /*
4456          * index_cnt is ignored for everything but a dir,
4457          * btrfs_get_inode_index_count has an explanation for the magic
4458          * number
4459          */
4460         BTRFS_I(inode)->index_cnt = 2;
4461         BTRFS_I(inode)->root = root;
4462         BTRFS_I(inode)->generation = trans->transid;
4463         inode->i_generation = BTRFS_I(inode)->generation;
4464         btrfs_set_inode_space_info(root, inode);
4465
4466         if (mode & S_IFDIR)
4467                 owner = 0;
4468         else
4469                 owner = 1;
4470
4471         key[0].objectid = objectid;
4472         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4473         key[0].offset = 0;
4474
4475         key[1].objectid = objectid;
4476         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4477         key[1].offset = ref_objectid;
4478
4479         sizes[0] = sizeof(struct btrfs_inode_item);
4480         sizes[1] = name_len + sizeof(*ref);
4481
4482         path->leave_spinning = 1;
4483         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4484         if (ret != 0)
4485                 goto fail;
4486
4487         inode_init_owner(inode, dir, mode);
4488         inode_set_bytes(inode, 0);
4489         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4490         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4491                                   struct btrfs_inode_item);
4492         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4493
4494         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4495                              struct btrfs_inode_ref);
4496         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4497         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4498         ptr = (unsigned long)(ref + 1);
4499         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4500
4501         btrfs_mark_buffer_dirty(path->nodes[0]);
4502         btrfs_free_path(path);
4503
4504         location = &BTRFS_I(inode)->location;
4505         location->objectid = objectid;
4506         location->offset = 0;
4507         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4508
4509         btrfs_inherit_iflags(inode, dir);
4510
4511         if ((mode & S_IFREG)) {
4512                 if (btrfs_test_opt(root, NODATASUM))
4513                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4514                 if (btrfs_test_opt(root, NODATACOW) ||
4515                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4516                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4517         }
4518
4519         insert_inode_hash(inode);
4520         inode_tree_add(inode);
4521
4522         trace_btrfs_inode_new(inode);
4523
4524         return inode;
4525 fail:
4526         if (dir)
4527                 BTRFS_I(dir)->index_cnt--;
4528         btrfs_free_path(path);
4529         iput(inode);
4530         return ERR_PTR(ret);
4531 }
4532
4533 static inline u8 btrfs_inode_type(struct inode *inode)
4534 {
4535         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4536 }
4537
4538 /*
4539  * utility function to add 'inode' into 'parent_inode' with
4540  * a give name and a given sequence number.
4541  * if 'add_backref' is true, also insert a backref from the
4542  * inode to the parent directory.
4543  */
4544 int btrfs_add_link(struct btrfs_trans_handle *trans,
4545                    struct inode *parent_inode, struct inode *inode,
4546                    const char *name, int name_len, int add_backref, u64 index)
4547 {
4548         int ret = 0;
4549         struct btrfs_key key;
4550         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4551         u64 ino = btrfs_ino(inode);
4552         u64 parent_ino = btrfs_ino(parent_inode);
4553
4554         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4555                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4556         } else {
4557                 key.objectid = ino;
4558                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4559                 key.offset = 0;
4560         }
4561
4562         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4563                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4564                                          key.objectid, root->root_key.objectid,
4565                                          parent_ino, index, name, name_len);
4566         } else if (add_backref) {
4567                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4568                                              parent_ino, index);
4569         }
4570
4571         if (ret == 0) {
4572                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4573                                             parent_inode, &key,
4574                                             btrfs_inode_type(inode), index);
4575                 BUG_ON(ret);
4576
4577                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4578                                    name_len * 2);
4579                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4580                 ret = btrfs_update_inode(trans, root, parent_inode);
4581         }
4582         return ret;
4583 }
4584
4585 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4586                             struct inode *dir, struct dentry *dentry,
4587                             struct inode *inode, int backref, u64 index)
4588 {
4589         int err = btrfs_add_link(trans, dir, inode,
4590                                  dentry->d_name.name, dentry->d_name.len,
4591                                  backref, index);
4592         if (!err) {
4593                 d_instantiate(dentry, inode);
4594                 return 0;
4595         }
4596         if (err > 0)
4597                 err = -EEXIST;
4598         return err;
4599 }
4600
4601 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4602                         int mode, dev_t rdev)
4603 {
4604         struct btrfs_trans_handle *trans;
4605         struct btrfs_root *root = BTRFS_I(dir)->root;
4606         struct inode *inode = NULL;
4607         int err;
4608         int drop_inode = 0;
4609         u64 objectid;
4610         unsigned long nr = 0;
4611         u64 index = 0;
4612
4613         if (!new_valid_dev(rdev))
4614                 return -EINVAL;
4615
4616         /*
4617          * 2 for inode item and ref
4618          * 2 for dir items
4619          * 1 for xattr if selinux is on
4620          */
4621         trans = btrfs_start_transaction(root, 5);
4622         if (IS_ERR(trans))
4623                 return PTR_ERR(trans);
4624
4625         err = btrfs_find_free_ino(root, &objectid);
4626         if (err)
4627                 goto out_unlock;
4628
4629         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4630                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4631                                 mode, &index);
4632         if (IS_ERR(inode)) {
4633                 err = PTR_ERR(inode);
4634                 goto out_unlock;
4635         }
4636
4637         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4638         if (err) {
4639                 drop_inode = 1;
4640                 goto out_unlock;
4641         }
4642
4643         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4644         if (err)
4645                 drop_inode = 1;
4646         else {
4647                 inode->i_op = &btrfs_special_inode_operations;
4648                 init_special_inode(inode, inode->i_mode, rdev);
4649                 btrfs_update_inode(trans, root, inode);
4650         }
4651 out_unlock:
4652         nr = trans->blocks_used;
4653         btrfs_end_transaction_throttle(trans, root);
4654         btrfs_btree_balance_dirty(root, nr);
4655         if (drop_inode) {
4656                 inode_dec_link_count(inode);
4657                 iput(inode);
4658         }
4659         return err;
4660 }
4661
4662 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4663                         int mode, struct nameidata *nd)
4664 {
4665         struct btrfs_trans_handle *trans;
4666         struct btrfs_root *root = BTRFS_I(dir)->root;
4667         struct inode *inode = NULL;
4668         int drop_inode = 0;
4669         int err;
4670         unsigned long nr = 0;
4671         u64 objectid;
4672         u64 index = 0;
4673
4674         /*
4675          * 2 for inode item and ref
4676          * 2 for dir items
4677          * 1 for xattr if selinux is on
4678          */
4679         trans = btrfs_start_transaction(root, 5);
4680         if (IS_ERR(trans))
4681                 return PTR_ERR(trans);
4682
4683         err = btrfs_find_free_ino(root, &objectid);
4684         if (err)
4685                 goto out_unlock;
4686
4687         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4688                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4689                                 mode, &index);
4690         if (IS_ERR(inode)) {
4691                 err = PTR_ERR(inode);
4692                 goto out_unlock;
4693         }
4694
4695         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4696         if (err) {
4697                 drop_inode = 1;
4698                 goto out_unlock;
4699         }
4700
4701         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4702         if (err)
4703                 drop_inode = 1;
4704         else {
4705                 inode->i_mapping->a_ops = &btrfs_aops;
4706                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4707                 inode->i_fop = &btrfs_file_operations;
4708                 inode->i_op = &btrfs_file_inode_operations;
4709                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4710         }
4711 out_unlock:
4712         nr = trans->blocks_used;
4713         btrfs_end_transaction_throttle(trans, root);
4714         if (drop_inode) {
4715                 inode_dec_link_count(inode);
4716                 iput(inode);
4717         }
4718         btrfs_btree_balance_dirty(root, nr);
4719         return err;
4720 }
4721
4722 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4723                       struct dentry *dentry)
4724 {
4725         struct btrfs_trans_handle *trans;
4726         struct btrfs_root *root = BTRFS_I(dir)->root;
4727         struct inode *inode = old_dentry->d_inode;
4728         u64 index;
4729         unsigned long nr = 0;
4730         int err;
4731         int drop_inode = 0;
4732
4733         /* do not allow sys_link's with other subvols of the same device */
4734         if (root->objectid != BTRFS_I(inode)->root->objectid)
4735                 return -EXDEV;
4736
4737         if (inode->i_nlink == ~0U)
4738                 return -EMLINK;
4739
4740         err = btrfs_set_inode_index(dir, &index);
4741         if (err)
4742                 goto fail;
4743
4744         /*
4745          * 2 items for inode and inode ref
4746          * 2 items for dir items
4747          * 1 item for parent inode
4748          */
4749         trans = btrfs_start_transaction(root, 5);
4750         if (IS_ERR(trans)) {
4751                 err = PTR_ERR(trans);
4752                 goto fail;
4753         }
4754
4755         btrfs_inc_nlink(inode);
4756         inode->i_ctime = CURRENT_TIME;
4757         ihold(inode);
4758
4759         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4760
4761         if (err) {
4762                 drop_inode = 1;
4763         } else {
4764                 struct dentry *parent = dget_parent(dentry);
4765                 err = btrfs_update_inode(trans, root, inode);
4766                 BUG_ON(err);
4767                 btrfs_log_new_name(trans, inode, NULL, parent);
4768                 dput(parent);
4769         }
4770
4771         nr = trans->blocks_used;
4772         btrfs_end_transaction_throttle(trans, root);
4773 fail:
4774         if (drop_inode) {
4775                 inode_dec_link_count(inode);
4776                 iput(inode);
4777         }
4778         btrfs_btree_balance_dirty(root, nr);
4779         return err;
4780 }
4781
4782 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4783 {
4784         struct inode *inode = NULL;
4785         struct btrfs_trans_handle *trans;
4786         struct btrfs_root *root = BTRFS_I(dir)->root;
4787         int err = 0;
4788         int drop_on_err = 0;
4789         u64 objectid = 0;
4790         u64 index = 0;
4791         unsigned long nr = 1;
4792
4793         /*
4794          * 2 items for inode and ref
4795          * 2 items for dir items
4796          * 1 for xattr if selinux is on
4797          */
4798         trans = btrfs_start_transaction(root, 5);
4799         if (IS_ERR(trans))
4800                 return PTR_ERR(trans);
4801
4802         err = btrfs_find_free_ino(root, &objectid);
4803         if (err)
4804                 goto out_fail;
4805
4806         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4807                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4808                                 S_IFDIR | mode, &index);
4809         if (IS_ERR(inode)) {
4810                 err = PTR_ERR(inode);
4811                 goto out_fail;
4812         }
4813
4814         drop_on_err = 1;
4815
4816         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4817         if (err)
4818                 goto out_fail;
4819
4820         inode->i_op = &btrfs_dir_inode_operations;
4821         inode->i_fop = &btrfs_dir_file_operations;
4822
4823         btrfs_i_size_write(inode, 0);
4824         err = btrfs_update_inode(trans, root, inode);
4825         if (err)
4826                 goto out_fail;
4827
4828         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4829                              dentry->d_name.len, 0, index);
4830         if (err)
4831                 goto out_fail;
4832
4833         d_instantiate(dentry, inode);
4834         drop_on_err = 0;
4835
4836 out_fail:
4837         nr = trans->blocks_used;
4838         btrfs_end_transaction_throttle(trans, root);
4839         if (drop_on_err)
4840                 iput(inode);
4841         btrfs_btree_balance_dirty(root, nr);
4842         return err;
4843 }
4844
4845 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4846  * and an extent that you want to insert, deal with overlap and insert
4847  * the new extent into the tree.
4848  */
4849 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4850                                 struct extent_map *existing,
4851                                 struct extent_map *em,
4852                                 u64 map_start, u64 map_len)
4853 {
4854         u64 start_diff;
4855
4856         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4857         start_diff = map_start - em->start;
4858         em->start = map_start;
4859         em->len = map_len;
4860         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4861             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4862                 em->block_start += start_diff;
4863                 em->block_len -= start_diff;
4864         }
4865         return add_extent_mapping(em_tree, em);
4866 }
4867
4868 static noinline int uncompress_inline(struct btrfs_path *path,
4869                                       struct inode *inode, struct page *page,
4870                                       size_t pg_offset, u64 extent_offset,
4871                                       struct btrfs_file_extent_item *item)
4872 {
4873         int ret;
4874         struct extent_buffer *leaf = path->nodes[0];
4875         char *tmp;
4876         size_t max_size;
4877         unsigned long inline_size;
4878         unsigned long ptr;
4879         int compress_type;
4880
4881         WARN_ON(pg_offset != 0);
4882         compress_type = btrfs_file_extent_compression(leaf, item);
4883         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4884         inline_size = btrfs_file_extent_inline_item_len(leaf,
4885                                         btrfs_item_nr(leaf, path->slots[0]));
4886         tmp = kmalloc(inline_size, GFP_NOFS);
4887         if (!tmp)
4888                 return -ENOMEM;
4889         ptr = btrfs_file_extent_inline_start(item);
4890
4891         read_extent_buffer(leaf, tmp, ptr, inline_size);
4892
4893         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4894         ret = btrfs_decompress(compress_type, tmp, page,
4895                                extent_offset, inline_size, max_size);
4896         if (ret) {
4897                 char *kaddr = kmap_atomic(page, KM_USER0);
4898                 unsigned long copy_size = min_t(u64,
4899                                   PAGE_CACHE_SIZE - pg_offset,
4900                                   max_size - extent_offset);
4901                 memset(kaddr + pg_offset, 0, copy_size);
4902                 kunmap_atomic(kaddr, KM_USER0);
4903         }
4904         kfree(tmp);
4905         return 0;
4906 }
4907
4908 /*
4909  * a bit scary, this does extent mapping from logical file offset to the disk.
4910  * the ugly parts come from merging extents from the disk with the in-ram
4911  * representation.  This gets more complex because of the data=ordered code,
4912  * where the in-ram extents might be locked pending data=ordered completion.
4913  *
4914  * This also copies inline extents directly into the page.
4915  */
4916
4917 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4918                                     size_t pg_offset, u64 start, u64 len,
4919                                     int create)
4920 {
4921         int ret;
4922         int err = 0;
4923         u64 bytenr;
4924         u64 extent_start = 0;
4925         u64 extent_end = 0;
4926         u64 objectid = btrfs_ino(inode);
4927         u32 found_type;
4928         struct btrfs_path *path = NULL;
4929         struct btrfs_root *root = BTRFS_I(inode)->root;
4930         struct btrfs_file_extent_item *item;
4931         struct extent_buffer *leaf;
4932         struct btrfs_key found_key;
4933         struct extent_map *em = NULL;
4934         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4935         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4936         struct btrfs_trans_handle *trans = NULL;
4937         int compress_type;
4938
4939 again:
4940         read_lock(&em_tree->lock);
4941         em = lookup_extent_mapping(em_tree, start, len);
4942         if (em)
4943                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4944         read_unlock(&em_tree->lock);
4945
4946         if (em) {
4947                 if (em->start > start || em->start + em->len <= start)
4948                         free_extent_map(em);
4949                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4950                         free_extent_map(em);
4951                 else
4952                         goto out;
4953         }
4954         em = alloc_extent_map();
4955         if (!em) {
4956                 err = -ENOMEM;
4957                 goto out;
4958         }
4959         em->bdev = root->fs_info->fs_devices->latest_bdev;
4960         em->start = EXTENT_MAP_HOLE;
4961         em->orig_start = EXTENT_MAP_HOLE;
4962         em->len = (u64)-1;
4963         em->block_len = (u64)-1;
4964
4965         if (!path) {
4966                 path = btrfs_alloc_path();
4967                 if (!path) {
4968                         err = -ENOMEM;
4969                         goto out;
4970                 }
4971                 /*
4972                  * Chances are we'll be called again, so go ahead and do
4973                  * readahead
4974                  */
4975                 path->reada = 1;
4976         }
4977
4978         ret = btrfs_lookup_file_extent(trans, root, path,
4979                                        objectid, start, trans != NULL);
4980         if (ret < 0) {
4981                 err = ret;
4982                 goto out;
4983         }
4984
4985         if (ret != 0) {
4986                 if (path->slots[0] == 0)
4987                         goto not_found;
4988                 path->slots[0]--;
4989         }
4990
4991         leaf = path->nodes[0];
4992         item = btrfs_item_ptr(leaf, path->slots[0],
4993                               struct btrfs_file_extent_item);
4994         /* are we inside the extent that was found? */
4995         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4996         found_type = btrfs_key_type(&found_key);
4997         if (found_key.objectid != objectid ||
4998             found_type != BTRFS_EXTENT_DATA_KEY) {
4999                 goto not_found;
5000         }
5001
5002         found_type = btrfs_file_extent_type(leaf, item);
5003         extent_start = found_key.offset;
5004         compress_type = btrfs_file_extent_compression(leaf, item);
5005         if (found_type == BTRFS_FILE_EXTENT_REG ||
5006             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5007                 extent_end = extent_start +
5008                        btrfs_file_extent_num_bytes(leaf, item);
5009         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5010                 size_t size;
5011                 size = btrfs_file_extent_inline_len(leaf, item);
5012                 extent_end = (extent_start + size + root->sectorsize - 1) &
5013                         ~((u64)root->sectorsize - 1);
5014         }
5015
5016         if (start >= extent_end) {
5017                 path->slots[0]++;
5018                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5019                         ret = btrfs_next_leaf(root, path);
5020                         if (ret < 0) {
5021                                 err = ret;
5022                                 goto out;
5023                         }
5024                         if (ret > 0)
5025                                 goto not_found;
5026                         leaf = path->nodes[0];
5027                 }
5028                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5029                 if (found_key.objectid != objectid ||
5030                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5031                         goto not_found;
5032                 if (start + len <= found_key.offset)
5033                         goto not_found;
5034                 em->start = start;
5035                 em->len = found_key.offset - start;
5036                 goto not_found_em;
5037         }
5038
5039         if (found_type == BTRFS_FILE_EXTENT_REG ||
5040             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5041                 em->start = extent_start;
5042                 em->len = extent_end - extent_start;
5043                 em->orig_start = extent_start -
5044                                  btrfs_file_extent_offset(leaf, item);
5045                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5046                 if (bytenr == 0) {
5047                         em->block_start = EXTENT_MAP_HOLE;
5048                         goto insert;
5049                 }
5050                 if (compress_type != BTRFS_COMPRESS_NONE) {
5051                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5052                         em->compress_type = compress_type;
5053                         em->block_start = bytenr;
5054                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5055                                                                          item);
5056                 } else {
5057                         bytenr += btrfs_file_extent_offset(leaf, item);
5058                         em->block_start = bytenr;
5059                         em->block_len = em->len;
5060                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5061                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5062                 }
5063                 goto insert;
5064         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5065                 unsigned long ptr;
5066                 char *map;
5067                 size_t size;
5068                 size_t extent_offset;
5069                 size_t copy_size;
5070
5071                 em->block_start = EXTENT_MAP_INLINE;
5072                 if (!page || create) {
5073                         em->start = extent_start;
5074                         em->len = extent_end - extent_start;
5075                         goto out;
5076                 }
5077
5078                 size = btrfs_file_extent_inline_len(leaf, item);
5079                 extent_offset = page_offset(page) + pg_offset - extent_start;
5080                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5081                                 size - extent_offset);
5082                 em->start = extent_start + extent_offset;
5083                 em->len = (copy_size + root->sectorsize - 1) &
5084                         ~((u64)root->sectorsize - 1);
5085                 em->orig_start = EXTENT_MAP_INLINE;
5086                 if (compress_type) {
5087                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5088                         em->compress_type = compress_type;
5089                 }
5090                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5091                 if (create == 0 && !PageUptodate(page)) {
5092                         if (btrfs_file_extent_compression(leaf, item) !=
5093                             BTRFS_COMPRESS_NONE) {
5094                                 ret = uncompress_inline(path, inode, page,
5095                                                         pg_offset,
5096                                                         extent_offset, item);
5097                                 BUG_ON(ret);
5098                         } else {
5099                                 map = kmap(page);
5100                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5101                                                    copy_size);
5102                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5103                                         memset(map + pg_offset + copy_size, 0,
5104                                                PAGE_CACHE_SIZE - pg_offset -
5105                                                copy_size);
5106                                 }
5107                                 kunmap(page);
5108                         }
5109                         flush_dcache_page(page);
5110                 } else if (create && PageUptodate(page)) {
5111                         WARN_ON(1);
5112                         if (!trans) {
5113                                 kunmap(page);
5114                                 free_extent_map(em);
5115                                 em = NULL;
5116
5117                                 btrfs_release_path(path);
5118                                 trans = btrfs_join_transaction(root);
5119
5120                                 if (IS_ERR(trans))
5121                                         return ERR_CAST(trans);
5122                                 goto again;
5123                         }
5124                         map = kmap(page);
5125                         write_extent_buffer(leaf, map + pg_offset, ptr,
5126                                             copy_size);
5127                         kunmap(page);
5128                         btrfs_mark_buffer_dirty(leaf);
5129                 }
5130                 set_extent_uptodate(io_tree, em->start,
5131                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5132                 goto insert;
5133         } else {
5134                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5135                 WARN_ON(1);
5136         }
5137 not_found:
5138         em->start = start;
5139         em->len = len;
5140 not_found_em:
5141         em->block_start = EXTENT_MAP_HOLE;
5142         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5143 insert:
5144         btrfs_release_path(path);
5145         if (em->start > start || extent_map_end(em) <= start) {
5146                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5147                        "[%llu %llu]\n", (unsigned long long)em->start,
5148                        (unsigned long long)em->len,
5149                        (unsigned long long)start,
5150                        (unsigned long long)len);
5151                 err = -EIO;
5152                 goto out;
5153         }
5154
5155         err = 0;
5156         write_lock(&em_tree->lock);
5157         ret = add_extent_mapping(em_tree, em);
5158         /* it is possible that someone inserted the extent into the tree
5159          * while we had the lock dropped.  It is also possible that
5160          * an overlapping map exists in the tree
5161          */
5162         if (ret == -EEXIST) {
5163                 struct extent_map *existing;
5164
5165                 ret = 0;
5166
5167                 existing = lookup_extent_mapping(em_tree, start, len);
5168                 if (existing && (existing->start > start ||
5169                     existing->start + existing->len <= start)) {
5170                         free_extent_map(existing);
5171                         existing = NULL;
5172                 }
5173                 if (!existing) {
5174                         existing = lookup_extent_mapping(em_tree, em->start,
5175                                                          em->len);
5176                         if (existing) {
5177                                 err = merge_extent_mapping(em_tree, existing,
5178                                                            em, start,
5179                                                            root->sectorsize);
5180                                 free_extent_map(existing);
5181                                 if (err) {
5182                                         free_extent_map(em);
5183                                         em = NULL;
5184                                 }
5185                         } else {
5186                                 err = -EIO;
5187                                 free_extent_map(em);
5188                                 em = NULL;
5189                         }
5190                 } else {
5191                         free_extent_map(em);
5192                         em = existing;
5193                         err = 0;
5194                 }
5195         }
5196         write_unlock(&em_tree->lock);
5197 out:
5198
5199         trace_btrfs_get_extent(root, em);
5200
5201         if (path)
5202                 btrfs_free_path(path);
5203         if (trans) {
5204                 ret = btrfs_end_transaction(trans, root);
5205                 if (!err)
5206                         err = ret;
5207         }
5208         if (err) {
5209                 free_extent_map(em);
5210                 return ERR_PTR(err);
5211         }
5212         return em;
5213 }
5214
5215 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5216                                            size_t pg_offset, u64 start, u64 len,
5217                                            int create)
5218 {
5219         struct extent_map *em;
5220         struct extent_map *hole_em = NULL;
5221         u64 range_start = start;
5222         u64 end;
5223         u64 found;
5224         u64 found_end;
5225         int err = 0;
5226
5227         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5228         if (IS_ERR(em))
5229                 return em;
5230         if (em) {
5231                 /*
5232                  * if our em maps to a hole, there might
5233                  * actually be delalloc bytes behind it
5234                  */
5235                 if (em->block_start != EXTENT_MAP_HOLE)
5236                         return em;
5237                 else
5238                         hole_em = em;
5239         }
5240
5241         /* check to see if we've wrapped (len == -1 or similar) */
5242         end = start + len;
5243         if (end < start)
5244                 end = (u64)-1;
5245         else
5246                 end -= 1;
5247
5248         em = NULL;
5249
5250         /* ok, we didn't find anything, lets look for delalloc */
5251         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5252                                  end, len, EXTENT_DELALLOC, 1);
5253         found_end = range_start + found;
5254         if (found_end < range_start)
5255                 found_end = (u64)-1;
5256
5257         /*
5258          * we didn't find anything useful, return
5259          * the original results from get_extent()
5260          */
5261         if (range_start > end || found_end <= start) {
5262                 em = hole_em;
5263                 hole_em = NULL;
5264                 goto out;
5265         }
5266
5267         /* adjust the range_start to make sure it doesn't
5268          * go backwards from the start they passed in
5269          */
5270         range_start = max(start,range_start);
5271         found = found_end - range_start;
5272
5273         if (found > 0) {
5274                 u64 hole_start = start;
5275                 u64 hole_len = len;
5276
5277                 em = alloc_extent_map();
5278                 if (!em) {
5279                         err = -ENOMEM;
5280                         goto out;
5281                 }
5282                 /*
5283                  * when btrfs_get_extent can't find anything it
5284                  * returns one huge hole
5285                  *
5286                  * make sure what it found really fits our range, and
5287                  * adjust to make sure it is based on the start from
5288                  * the caller
5289                  */
5290                 if (hole_em) {
5291                         u64 calc_end = extent_map_end(hole_em);
5292
5293                         if (calc_end <= start || (hole_em->start > end)) {
5294                                 free_extent_map(hole_em);
5295                                 hole_em = NULL;
5296                         } else {
5297                                 hole_start = max(hole_em->start, start);
5298                                 hole_len = calc_end - hole_start;
5299                         }
5300                 }
5301                 em->bdev = NULL;
5302                 if (hole_em && range_start > hole_start) {
5303                         /* our hole starts before our delalloc, so we
5304                          * have to return just the parts of the hole
5305                          * that go until  the delalloc starts
5306                          */
5307                         em->len = min(hole_len,
5308                                       range_start - hole_start);
5309                         em->start = hole_start;
5310                         em->orig_start = hole_start;
5311                         /*
5312                          * don't adjust block start at all,
5313                          * it is fixed at EXTENT_MAP_HOLE
5314                          */
5315                         em->block_start = hole_em->block_start;
5316                         em->block_len = hole_len;
5317                 } else {
5318                         em->start = range_start;
5319                         em->len = found;
5320                         em->orig_start = range_start;
5321                         em->block_start = EXTENT_MAP_DELALLOC;
5322                         em->block_len = found;
5323                 }
5324         } else if (hole_em) {
5325                 return hole_em;
5326         }
5327 out:
5328
5329         free_extent_map(hole_em);
5330         if (err) {
5331                 free_extent_map(em);
5332                 return ERR_PTR(err);
5333         }
5334         return em;
5335 }
5336
5337 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5338                                                   struct extent_map *em,
5339                                                   u64 start, u64 len)
5340 {
5341         struct btrfs_root *root = BTRFS_I(inode)->root;
5342         struct btrfs_trans_handle *trans;
5343         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5344         struct btrfs_key ins;
5345         u64 alloc_hint;
5346         int ret;
5347         bool insert = false;
5348
5349         /*
5350          * Ok if the extent map we looked up is a hole and is for the exact
5351          * range we want, there is no reason to allocate a new one, however if
5352          * it is not right then we need to free this one and drop the cache for
5353          * our range.
5354          */
5355         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5356             em->len != len) {
5357                 free_extent_map(em);
5358                 em = NULL;
5359                 insert = true;
5360                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5361         }
5362
5363         trans = btrfs_join_transaction(root);
5364         if (IS_ERR(trans))
5365                 return ERR_CAST(trans);
5366
5367         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5368                 btrfs_add_inode_defrag(trans, inode);
5369
5370         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5371
5372         alloc_hint = get_extent_allocation_hint(inode, start, len);
5373         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5374                                    alloc_hint, (u64)-1, &ins, 1);
5375         if (ret) {
5376                 em = ERR_PTR(ret);
5377                 goto out;
5378         }
5379
5380         if (!em) {
5381                 em = alloc_extent_map();
5382                 if (!em) {
5383                         em = ERR_PTR(-ENOMEM);
5384                         goto out;
5385                 }
5386         }
5387
5388         em->start = start;
5389         em->orig_start = em->start;
5390         em->len = ins.offset;
5391
5392         em->block_start = ins.objectid;
5393         em->block_len = ins.offset;
5394         em->bdev = root->fs_info->fs_devices->latest_bdev;
5395
5396         /*
5397          * We need to do this because if we're using the original em we searched
5398          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5399          */
5400         em->flags = 0;
5401         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5402
5403         while (insert) {
5404                 write_lock(&em_tree->lock);
5405                 ret = add_extent_mapping(em_tree, em);
5406                 write_unlock(&em_tree->lock);
5407                 if (ret != -EEXIST)
5408                         break;
5409                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5410         }
5411
5412         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5413                                            ins.offset, ins.offset, 0);
5414         if (ret) {
5415                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5416                 em = ERR_PTR(ret);
5417         }
5418 out:
5419         btrfs_end_transaction(trans, root);
5420         return em;
5421 }
5422
5423 /*
5424  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5425  * block must be cow'd
5426  */
5427 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5428                                       struct inode *inode, u64 offset, u64 len)
5429 {
5430         struct btrfs_path *path;
5431         int ret;
5432         struct extent_buffer *leaf;
5433         struct btrfs_root *root = BTRFS_I(inode)->root;
5434         struct btrfs_file_extent_item *fi;
5435         struct btrfs_key key;
5436         u64 disk_bytenr;
5437         u64 backref_offset;
5438         u64 extent_end;
5439         u64 num_bytes;
5440         int slot;
5441         int found_type;
5442
5443         path = btrfs_alloc_path();
5444         if (!path)
5445                 return -ENOMEM;
5446
5447         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5448                                        offset, 0);
5449         if (ret < 0)
5450                 goto out;
5451
5452         slot = path->slots[0];
5453         if (ret == 1) {
5454                 if (slot == 0) {
5455                         /* can't find the item, must cow */
5456                         ret = 0;
5457                         goto out;
5458                 }
5459                 slot--;
5460         }
5461         ret = 0;
5462         leaf = path->nodes[0];
5463         btrfs_item_key_to_cpu(leaf, &key, slot);
5464         if (key.objectid != btrfs_ino(inode) ||
5465             key.type != BTRFS_EXTENT_DATA_KEY) {
5466                 /* not our file or wrong item type, must cow */
5467                 goto out;
5468         }
5469
5470         if (key.offset > offset) {
5471                 /* Wrong offset, must cow */
5472                 goto out;
5473         }
5474
5475         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5476         found_type = btrfs_file_extent_type(leaf, fi);
5477         if (found_type != BTRFS_FILE_EXTENT_REG &&
5478             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5479                 /* not a regular extent, must cow */
5480                 goto out;
5481         }
5482         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5483         backref_offset = btrfs_file_extent_offset(leaf, fi);
5484
5485         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5486         if (extent_end < offset + len) {
5487                 /* extent doesn't include our full range, must cow */
5488                 goto out;
5489         }
5490
5491         if (btrfs_extent_readonly(root, disk_bytenr))
5492                 goto out;
5493
5494         /*
5495          * look for other files referencing this extent, if we
5496          * find any we must cow
5497          */
5498         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5499                                   key.offset - backref_offset, disk_bytenr))
5500                 goto out;
5501
5502         /*
5503          * adjust disk_bytenr and num_bytes to cover just the bytes
5504          * in this extent we are about to write.  If there
5505          * are any csums in that range we have to cow in order
5506          * to keep the csums correct
5507          */
5508         disk_bytenr += backref_offset;
5509         disk_bytenr += offset - key.offset;
5510         num_bytes = min(offset + len, extent_end) - offset;
5511         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5512                                 goto out;
5513         /*
5514          * all of the above have passed, it is safe to overwrite this extent
5515          * without cow
5516          */
5517         ret = 1;
5518 out:
5519         btrfs_free_path(path);
5520         return ret;
5521 }
5522
5523 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5524                                    struct buffer_head *bh_result, int create)
5525 {
5526         struct extent_map *em;
5527         struct btrfs_root *root = BTRFS_I(inode)->root;
5528         u64 start = iblock << inode->i_blkbits;
5529         u64 len = bh_result->b_size;
5530         struct btrfs_trans_handle *trans;
5531
5532         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5533         if (IS_ERR(em))
5534                 return PTR_ERR(em);
5535
5536         /*
5537          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5538          * io.  INLINE is special, and we could probably kludge it in here, but
5539          * it's still buffered so for safety lets just fall back to the generic
5540          * buffered path.
5541          *
5542          * For COMPRESSED we _have_ to read the entire extent in so we can
5543          * decompress it, so there will be buffering required no matter what we
5544          * do, so go ahead and fallback to buffered.
5545          *
5546          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5547          * to buffered IO.  Don't blame me, this is the price we pay for using
5548          * the generic code.
5549          */
5550         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5551             em->block_start == EXTENT_MAP_INLINE) {
5552                 free_extent_map(em);
5553                 return -ENOTBLK;
5554         }
5555
5556         /* Just a good old fashioned hole, return */
5557         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5558                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5559                 free_extent_map(em);
5560                 /* DIO will do one hole at a time, so just unlock a sector */
5561                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5562                               start + root->sectorsize - 1, GFP_NOFS);
5563                 return 0;
5564         }
5565
5566         /*
5567          * We don't allocate a new extent in the following cases
5568          *
5569          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5570          * existing extent.
5571          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5572          * just use the extent.
5573          *
5574          */
5575         if (!create) {
5576                 len = em->len - (start - em->start);
5577                 goto map;
5578         }
5579
5580         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5581             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5582              em->block_start != EXTENT_MAP_HOLE)) {
5583                 int type;
5584                 int ret;
5585                 u64 block_start;
5586
5587                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5588                         type = BTRFS_ORDERED_PREALLOC;
5589                 else
5590                         type = BTRFS_ORDERED_NOCOW;
5591                 len = min(len, em->len - (start - em->start));
5592                 block_start = em->block_start + (start - em->start);
5593
5594                 /*
5595                  * we're not going to log anything, but we do need
5596                  * to make sure the current transaction stays open
5597                  * while we look for nocow cross refs
5598                  */
5599                 trans = btrfs_join_transaction(root);
5600                 if (IS_ERR(trans))
5601                         goto must_cow;
5602
5603                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5604                         ret = btrfs_add_ordered_extent_dio(inode, start,
5605                                            block_start, len, len, type);
5606                         btrfs_end_transaction(trans, root);
5607                         if (ret) {
5608                                 free_extent_map(em);
5609                                 return ret;
5610                         }
5611                         goto unlock;
5612                 }
5613                 btrfs_end_transaction(trans, root);
5614         }
5615 must_cow:
5616         /*
5617          * this will cow the extent, reset the len in case we changed
5618          * it above
5619          */
5620         len = bh_result->b_size;
5621         em = btrfs_new_extent_direct(inode, em, start, len);
5622         if (IS_ERR(em))
5623                 return PTR_ERR(em);
5624         len = min(len, em->len - (start - em->start));
5625 unlock:
5626         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5627                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5628                           0, NULL, GFP_NOFS);
5629 map:
5630         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5631                 inode->i_blkbits;
5632         bh_result->b_size = len;
5633         bh_result->b_bdev = em->bdev;
5634         set_buffer_mapped(bh_result);
5635         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5636                 set_buffer_new(bh_result);
5637
5638         free_extent_map(em);
5639
5640         return 0;
5641 }
5642
5643 struct btrfs_dio_private {
5644         struct inode *inode;
5645         u64 logical_offset;
5646         u64 disk_bytenr;
5647         u64 bytes;
5648         u32 *csums;
5649         void *private;
5650
5651         /* number of bios pending for this dio */
5652         atomic_t pending_bios;
5653
5654         /* IO errors */
5655         int errors;
5656
5657         struct bio *orig_bio;
5658 };
5659
5660 static void btrfs_endio_direct_read(struct bio *bio, int err)
5661 {
5662         struct btrfs_dio_private *dip = bio->bi_private;
5663         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5664         struct bio_vec *bvec = bio->bi_io_vec;
5665         struct inode *inode = dip->inode;
5666         struct btrfs_root *root = BTRFS_I(inode)->root;
5667         u64 start;
5668         u32 *private = dip->csums;
5669
5670         start = dip->logical_offset;
5671         do {
5672                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5673                         struct page *page = bvec->bv_page;
5674                         char *kaddr;
5675                         u32 csum = ~(u32)0;
5676                         unsigned long flags;
5677
5678                         local_irq_save(flags);
5679                         kaddr = kmap_atomic(page, KM_IRQ0);
5680                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5681                                                csum, bvec->bv_len);
5682                         btrfs_csum_final(csum, (char *)&csum);
5683                         kunmap_atomic(kaddr, KM_IRQ0);
5684                         local_irq_restore(flags);
5685
5686                         flush_dcache_page(bvec->bv_page);
5687                         if (csum != *private) {
5688                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
5689                                       " %llu csum %u private %u\n",
5690                                       (unsigned long long)btrfs_ino(inode),
5691                                       (unsigned long long)start,
5692                                       csum, *private);
5693                                 err = -EIO;
5694                         }
5695                 }
5696
5697                 start += bvec->bv_len;
5698                 private++;
5699                 bvec++;
5700         } while (bvec <= bvec_end);
5701
5702         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5703                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5704         bio->bi_private = dip->private;
5705
5706         kfree(dip->csums);
5707         kfree(dip);
5708
5709         /* If we had a csum failure make sure to clear the uptodate flag */
5710         if (err)
5711                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5712         dio_end_io(bio, err);
5713 }
5714
5715 static void btrfs_endio_direct_write(struct bio *bio, int err)
5716 {
5717         struct btrfs_dio_private *dip = bio->bi_private;
5718         struct inode *inode = dip->inode;
5719         struct btrfs_root *root = BTRFS_I(inode)->root;
5720         struct btrfs_trans_handle *trans;
5721         struct btrfs_ordered_extent *ordered = NULL;
5722         struct extent_state *cached_state = NULL;
5723         u64 ordered_offset = dip->logical_offset;
5724         u64 ordered_bytes = dip->bytes;
5725         int ret;
5726
5727         if (err)
5728                 goto out_done;
5729 again:
5730         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5731                                                    &ordered_offset,
5732                                                    ordered_bytes);
5733         if (!ret)
5734                 goto out_test;
5735
5736         BUG_ON(!ordered);
5737
5738         trans = btrfs_join_transaction(root);
5739         if (IS_ERR(trans)) {
5740                 err = -ENOMEM;
5741                 goto out;
5742         }
5743         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5744
5745         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5746                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5747                 if (!ret)
5748                         ret = btrfs_update_inode(trans, root, inode);
5749                 err = ret;
5750                 goto out;
5751         }
5752
5753         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5754                          ordered->file_offset + ordered->len - 1, 0,
5755                          &cached_state, GFP_NOFS);
5756
5757         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5758                 ret = btrfs_mark_extent_written(trans, inode,
5759                                                 ordered->file_offset,
5760                                                 ordered->file_offset +
5761                                                 ordered->len);
5762                 if (ret) {
5763                         err = ret;
5764                         goto out_unlock;
5765                 }
5766         } else {
5767                 ret = insert_reserved_file_extent(trans, inode,
5768                                                   ordered->file_offset,
5769                                                   ordered->start,
5770                                                   ordered->disk_len,
5771                                                   ordered->len,
5772                                                   ordered->len,
5773                                                   0, 0, 0,
5774                                                   BTRFS_FILE_EXTENT_REG);
5775                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5776                                    ordered->file_offset, ordered->len);
5777                 if (ret) {
5778                         err = ret;
5779                         WARN_ON(1);
5780                         goto out_unlock;
5781                 }
5782         }
5783
5784         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5785         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5786         if (!ret)
5787                 btrfs_update_inode(trans, root, inode);
5788         ret = 0;
5789 out_unlock:
5790         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5791                              ordered->file_offset + ordered->len - 1,
5792                              &cached_state, GFP_NOFS);
5793 out:
5794         btrfs_delalloc_release_metadata(inode, ordered->len);
5795         btrfs_end_transaction(trans, root);
5796         ordered_offset = ordered->file_offset + ordered->len;
5797         btrfs_put_ordered_extent(ordered);
5798         btrfs_put_ordered_extent(ordered);
5799
5800 out_test:
5801         /*
5802          * our bio might span multiple ordered extents.  If we haven't
5803          * completed the accounting for the whole dio, go back and try again
5804          */
5805         if (ordered_offset < dip->logical_offset + dip->bytes) {
5806                 ordered_bytes = dip->logical_offset + dip->bytes -
5807                         ordered_offset;
5808                 goto again;
5809         }
5810 out_done:
5811         bio->bi_private = dip->private;
5812
5813         kfree(dip->csums);
5814         kfree(dip);
5815
5816         /* If we had an error make sure to clear the uptodate flag */
5817         if (err)
5818                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5819         dio_end_io(bio, err);
5820 }
5821
5822 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5823                                     struct bio *bio, int mirror_num,
5824                                     unsigned long bio_flags, u64 offset)
5825 {
5826         int ret;
5827         struct btrfs_root *root = BTRFS_I(inode)->root;
5828         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5829         BUG_ON(ret);
5830         return 0;
5831 }
5832
5833 static void btrfs_end_dio_bio(struct bio *bio, int err)
5834 {
5835         struct btrfs_dio_private *dip = bio->bi_private;
5836
5837         if (err) {
5838                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5839                       "sector %#Lx len %u err no %d\n",
5840                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5841                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5842                 dip->errors = 1;
5843
5844                 /*
5845                  * before atomic variable goto zero, we must make sure
5846                  * dip->errors is perceived to be set.
5847                  */
5848                 smp_mb__before_atomic_dec();
5849         }
5850
5851         /* if there are more bios still pending for this dio, just exit */
5852         if (!atomic_dec_and_test(&dip->pending_bios))
5853                 goto out;
5854
5855         if (dip->errors)
5856                 bio_io_error(dip->orig_bio);
5857         else {
5858                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5859                 bio_endio(dip->orig_bio, 0);
5860         }
5861 out:
5862         bio_put(bio);
5863 }
5864
5865 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5866                                        u64 first_sector, gfp_t gfp_flags)
5867 {
5868         int nr_vecs = bio_get_nr_vecs(bdev);
5869         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5870 }
5871
5872 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5873                                          int rw, u64 file_offset, int skip_sum,
5874                                          u32 *csums, int async_submit)
5875 {
5876         int write = rw & REQ_WRITE;
5877         struct btrfs_root *root = BTRFS_I(inode)->root;
5878         int ret;
5879
5880         bio_get(bio);
5881         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5882         if (ret)
5883                 goto err;
5884
5885         if (skip_sum)
5886                 goto map;
5887
5888         if (write && async_submit) {
5889                 ret = btrfs_wq_submit_bio(root->fs_info,
5890                                    inode, rw, bio, 0, 0,
5891                                    file_offset,
5892                                    __btrfs_submit_bio_start_direct_io,
5893                                    __btrfs_submit_bio_done);
5894                 goto err;
5895         } else if (write) {
5896                 /*
5897                  * If we aren't doing async submit, calculate the csum of the
5898                  * bio now.
5899                  */
5900                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5901                 if (ret)
5902                         goto err;
5903         } else if (!skip_sum) {
5904                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5905                                           file_offset, csums);
5906                 if (ret)
5907                         goto err;
5908         }
5909
5910 map:
5911         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5912 err:
5913         bio_put(bio);
5914         return ret;
5915 }
5916
5917 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5918                                     int skip_sum)
5919 {
5920         struct inode *inode = dip->inode;
5921         struct btrfs_root *root = BTRFS_I(inode)->root;
5922         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5923         struct bio *bio;
5924         struct bio *orig_bio = dip->orig_bio;
5925         struct bio_vec *bvec = orig_bio->bi_io_vec;
5926         u64 start_sector = orig_bio->bi_sector;
5927         u64 file_offset = dip->logical_offset;
5928         u64 submit_len = 0;
5929         u64 map_length;
5930         int nr_pages = 0;
5931         u32 *csums = dip->csums;
5932         int ret = 0;
5933         int async_submit = 0;
5934         int write = rw & REQ_WRITE;
5935
5936         map_length = orig_bio->bi_size;
5937         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5938                               &map_length, NULL, 0);
5939         if (ret) {
5940                 bio_put(orig_bio);
5941                 return -EIO;
5942         }
5943
5944         if (map_length >= orig_bio->bi_size) {
5945                 bio = orig_bio;
5946                 goto submit;
5947         }
5948
5949         async_submit = 1;
5950         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5951         if (!bio)
5952                 return -ENOMEM;
5953         bio->bi_private = dip;
5954         bio->bi_end_io = btrfs_end_dio_bio;
5955         atomic_inc(&dip->pending_bios);
5956
5957         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5958                 if (unlikely(map_length < submit_len + bvec->bv_len ||
5959                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5960                                  bvec->bv_offset) < bvec->bv_len)) {
5961                         /*
5962                          * inc the count before we submit the bio so
5963                          * we know the end IO handler won't happen before
5964                          * we inc the count. Otherwise, the dip might get freed
5965                          * before we're done setting it up
5966                          */
5967                         atomic_inc(&dip->pending_bios);
5968                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
5969                                                      file_offset, skip_sum,
5970                                                      csums, async_submit);
5971                         if (ret) {
5972                                 bio_put(bio);
5973                                 atomic_dec(&dip->pending_bios);
5974                                 goto out_err;
5975                         }
5976
5977                         /* Write's use the ordered csums */
5978                         if (!write && !skip_sum)
5979                                 csums = csums + nr_pages;
5980                         start_sector += submit_len >> 9;
5981                         file_offset += submit_len;
5982
5983                         submit_len = 0;
5984                         nr_pages = 0;
5985
5986                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
5987                                                   start_sector, GFP_NOFS);
5988                         if (!bio)
5989                                 goto out_err;
5990                         bio->bi_private = dip;
5991                         bio->bi_end_io = btrfs_end_dio_bio;
5992
5993                         map_length = orig_bio->bi_size;
5994                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5995                                               &map_length, NULL, 0);
5996                         if (ret) {
5997                                 bio_put(bio);
5998                                 goto out_err;
5999                         }
6000                 } else {
6001                         submit_len += bvec->bv_len;
6002                         nr_pages ++;
6003                         bvec++;
6004                 }
6005         }
6006
6007 submit:
6008         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6009                                      csums, async_submit);
6010         if (!ret)
6011                 return 0;
6012
6013         bio_put(bio);
6014 out_err:
6015         dip->errors = 1;
6016         /*
6017          * before atomic variable goto zero, we must
6018          * make sure dip->errors is perceived to be set.
6019          */
6020         smp_mb__before_atomic_dec();
6021         if (atomic_dec_and_test(&dip->pending_bios))
6022                 bio_io_error(dip->orig_bio);
6023
6024         /* bio_end_io() will handle error, so we needn't return it */
6025         return 0;
6026 }
6027
6028 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6029                                 loff_t file_offset)
6030 {
6031         struct btrfs_root *root = BTRFS_I(inode)->root;
6032         struct btrfs_dio_private *dip;
6033         struct bio_vec *bvec = bio->bi_io_vec;
6034         int skip_sum;
6035         int write = rw & REQ_WRITE;
6036         int ret = 0;
6037
6038         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6039
6040         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6041         if (!dip) {
6042                 ret = -ENOMEM;
6043                 goto free_ordered;
6044         }
6045         dip->csums = NULL;
6046
6047         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6048         if (!write && !skip_sum) {
6049                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6050                 if (!dip->csums) {
6051                         kfree(dip);
6052                         ret = -ENOMEM;
6053                         goto free_ordered;
6054                 }
6055         }
6056
6057         dip->private = bio->bi_private;
6058         dip->inode = inode;
6059         dip->logical_offset = file_offset;
6060
6061         dip->bytes = 0;
6062         do {
6063                 dip->bytes += bvec->bv_len;
6064                 bvec++;
6065         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6066
6067         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6068         bio->bi_private = dip;
6069         dip->errors = 0;
6070         dip->orig_bio = bio;
6071         atomic_set(&dip->pending_bios, 0);
6072
6073         if (write)
6074                 bio->bi_end_io = btrfs_endio_direct_write;
6075         else
6076                 bio->bi_end_io = btrfs_endio_direct_read;
6077
6078         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6079         if (!ret)
6080                 return;
6081 free_ordered:
6082         /*
6083          * If this is a write, we need to clean up the reserved space and kill
6084          * the ordered extent.
6085          */
6086         if (write) {
6087                 struct btrfs_ordered_extent *ordered;
6088                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6089                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6090                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6091                         btrfs_free_reserved_extent(root, ordered->start,
6092                                                    ordered->disk_len);
6093                 btrfs_put_ordered_extent(ordered);
6094                 btrfs_put_ordered_extent(ordered);
6095         }
6096         bio_endio(bio, ret);
6097 }
6098
6099 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6100                         const struct iovec *iov, loff_t offset,
6101                         unsigned long nr_segs)
6102 {
6103         int seg;
6104         int i;
6105         size_t size;
6106         unsigned long addr;
6107         unsigned blocksize_mask = root->sectorsize - 1;
6108         ssize_t retval = -EINVAL;
6109         loff_t end = offset;
6110
6111         if (offset & blocksize_mask)
6112                 goto out;
6113
6114         /* Check the memory alignment.  Blocks cannot straddle pages */
6115         for (seg = 0; seg < nr_segs; seg++) {
6116                 addr = (unsigned long)iov[seg].iov_base;
6117                 size = iov[seg].iov_len;
6118                 end += size;
6119                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6120                         goto out;
6121
6122                 /* If this is a write we don't need to check anymore */
6123                 if (rw & WRITE)
6124                         continue;
6125
6126                 /*
6127                  * Check to make sure we don't have duplicate iov_base's in this
6128                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6129                  * when reading back.
6130                  */
6131                 for (i = seg + 1; i < nr_segs; i++) {
6132                         if (iov[seg].iov_base == iov[i].iov_base)
6133                                 goto out;
6134                 }
6135         }
6136         retval = 0;
6137 out:
6138         return retval;
6139 }
6140 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6141                         const struct iovec *iov, loff_t offset,
6142                         unsigned long nr_segs)
6143 {
6144         struct file *file = iocb->ki_filp;
6145         struct inode *inode = file->f_mapping->host;
6146         struct btrfs_ordered_extent *ordered;
6147         struct extent_state *cached_state = NULL;
6148         u64 lockstart, lockend;
6149         ssize_t ret;
6150         int writing = rw & WRITE;
6151         int write_bits = 0;
6152         size_t count = iov_length(iov, nr_segs);
6153
6154         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6155                             offset, nr_segs)) {
6156                 return 0;
6157         }
6158
6159         lockstart = offset;
6160         lockend = offset + count - 1;
6161
6162         if (writing) {
6163                 ret = btrfs_delalloc_reserve_space(inode, count);
6164                 if (ret)
6165                         goto out;
6166         }
6167
6168         while (1) {
6169                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6170                                  0, &cached_state, GFP_NOFS);
6171                 /*
6172                  * We're concerned with the entire range that we're going to be
6173                  * doing DIO to, so we need to make sure theres no ordered
6174                  * extents in this range.
6175                  */
6176                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6177                                                      lockend - lockstart + 1);
6178                 if (!ordered)
6179                         break;
6180                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6181                                      &cached_state, GFP_NOFS);
6182                 btrfs_start_ordered_extent(inode, ordered, 1);
6183                 btrfs_put_ordered_extent(ordered);
6184                 cond_resched();
6185         }
6186
6187         /*
6188          * we don't use btrfs_set_extent_delalloc because we don't want
6189          * the dirty or uptodate bits
6190          */
6191         if (writing) {
6192                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6193                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6194                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6195                                      GFP_NOFS);
6196                 if (ret) {
6197                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6198                                          lockend, EXTENT_LOCKED | write_bits,
6199                                          1, 0, &cached_state, GFP_NOFS);
6200                         goto out;
6201                 }
6202         }
6203
6204         free_extent_state(cached_state);
6205         cached_state = NULL;
6206
6207         ret = __blockdev_direct_IO(rw, iocb, inode,
6208                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6209                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6210                    btrfs_submit_direct, 0);
6211
6212         if (ret < 0 && ret != -EIOCBQUEUED) {
6213                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6214                               offset + iov_length(iov, nr_segs) - 1,
6215                               EXTENT_LOCKED | write_bits, 1, 0,
6216                               &cached_state, GFP_NOFS);
6217         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6218                 /*
6219                  * We're falling back to buffered, unlock the section we didn't
6220                  * do IO on.
6221                  */
6222                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6223                               offset + iov_length(iov, nr_segs) - 1,
6224                               EXTENT_LOCKED | write_bits, 1, 0,
6225                               &cached_state, GFP_NOFS);
6226         }
6227 out:
6228         free_extent_state(cached_state);
6229         return ret;
6230 }
6231
6232 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6233                 __u64 start, __u64 len)
6234 {
6235         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6236 }
6237
6238 int btrfs_readpage(struct file *file, struct page *page)
6239 {
6240         struct extent_io_tree *tree;
6241         tree = &BTRFS_I(page->mapping->host)->io_tree;
6242         return extent_read_full_page(tree, page, btrfs_get_extent);
6243 }
6244
6245 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6246 {
6247         struct extent_io_tree *tree;
6248
6249
6250         if (current->flags & PF_MEMALLOC) {
6251                 redirty_page_for_writepage(wbc, page);
6252                 unlock_page(page);
6253                 return 0;
6254         }
6255         tree = &BTRFS_I(page->mapping->host)->io_tree;
6256         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6257 }
6258
6259 int btrfs_writepages(struct address_space *mapping,
6260                      struct writeback_control *wbc)
6261 {
6262         struct extent_io_tree *tree;
6263
6264         tree = &BTRFS_I(mapping->host)->io_tree;
6265         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6266 }
6267
6268 static int
6269 btrfs_readpages(struct file *file, struct address_space *mapping,
6270                 struct list_head *pages, unsigned nr_pages)
6271 {
6272         struct extent_io_tree *tree;
6273         tree = &BTRFS_I(mapping->host)->io_tree;
6274         return extent_readpages(tree, mapping, pages, nr_pages,
6275                                 btrfs_get_extent);
6276 }
6277 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6278 {
6279         struct extent_io_tree *tree;
6280         struct extent_map_tree *map;
6281         int ret;
6282
6283         tree = &BTRFS_I(page->mapping->host)->io_tree;
6284         map = &BTRFS_I(page->mapping->host)->extent_tree;
6285         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6286         if (ret == 1) {
6287                 ClearPagePrivate(page);
6288                 set_page_private(page, 0);
6289                 page_cache_release(page);
6290         }
6291         return ret;
6292 }
6293
6294 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6295 {
6296         if (PageWriteback(page) || PageDirty(page))
6297                 return 0;
6298         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6299 }
6300
6301 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6302 {
6303         struct extent_io_tree *tree;
6304         struct btrfs_ordered_extent *ordered;
6305         struct extent_state *cached_state = NULL;
6306         u64 page_start = page_offset(page);
6307         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6308
6309
6310         /*
6311          * we have the page locked, so new writeback can't start,
6312          * and the dirty bit won't be cleared while we are here.
6313          *
6314          * Wait for IO on this page so that we can safely clear
6315          * the PagePrivate2 bit and do ordered accounting
6316          */
6317         wait_on_page_writeback(page);
6318
6319         tree = &BTRFS_I(page->mapping->host)->io_tree;
6320         if (offset) {
6321                 btrfs_releasepage(page, GFP_NOFS);
6322                 return;
6323         }
6324         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6325                          GFP_NOFS);
6326         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6327                                            page_offset(page));
6328         if (ordered) {
6329                 /*
6330                  * IO on this page will never be started, so we need
6331                  * to account for any ordered extents now
6332                  */
6333                 clear_extent_bit(tree, page_start, page_end,
6334                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6335                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6336                                  &cached_state, GFP_NOFS);
6337                 /*
6338                  * whoever cleared the private bit is responsible
6339                  * for the finish_ordered_io
6340                  */
6341                 if (TestClearPagePrivate2(page)) {
6342                         btrfs_finish_ordered_io(page->mapping->host,
6343                                                 page_start, page_end);
6344                 }
6345                 btrfs_put_ordered_extent(ordered);
6346                 cached_state = NULL;
6347                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6348                                  GFP_NOFS);
6349         }
6350         clear_extent_bit(tree, page_start, page_end,
6351                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6352                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6353         __btrfs_releasepage(page, GFP_NOFS);
6354
6355         ClearPageChecked(page);
6356         if (PagePrivate(page)) {
6357                 ClearPagePrivate(page);
6358                 set_page_private(page, 0);
6359                 page_cache_release(page);
6360         }
6361 }
6362
6363 /*
6364  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6365  * called from a page fault handler when a page is first dirtied. Hence we must
6366  * be careful to check for EOF conditions here. We set the page up correctly
6367  * for a written page which means we get ENOSPC checking when writing into
6368  * holes and correct delalloc and unwritten extent mapping on filesystems that
6369  * support these features.
6370  *
6371  * We are not allowed to take the i_mutex here so we have to play games to
6372  * protect against truncate races as the page could now be beyond EOF.  Because
6373  * vmtruncate() writes the inode size before removing pages, once we have the
6374  * page lock we can determine safely if the page is beyond EOF. If it is not
6375  * beyond EOF, then the page is guaranteed safe against truncation until we
6376  * unlock the page.
6377  */
6378 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6379 {
6380         struct page *page = vmf->page;
6381         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6382         struct btrfs_root *root = BTRFS_I(inode)->root;
6383         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6384         struct btrfs_ordered_extent *ordered;
6385         struct extent_state *cached_state = NULL;
6386         char *kaddr;
6387         unsigned long zero_start;
6388         loff_t size;
6389         int ret;
6390         u64 page_start;
6391         u64 page_end;
6392
6393         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6394         if (ret) {
6395                 if (ret == -ENOMEM)
6396                         ret = VM_FAULT_OOM;
6397                 else /* -ENOSPC, -EIO, etc */
6398                         ret = VM_FAULT_SIGBUS;
6399                 goto out;
6400         }
6401
6402         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6403 again:
6404         lock_page(page);
6405         size = i_size_read(inode);
6406         page_start = page_offset(page);
6407         page_end = page_start + PAGE_CACHE_SIZE - 1;
6408
6409         if ((page->mapping != inode->i_mapping) ||
6410             (page_start >= size)) {
6411                 /* page got truncated out from underneath us */
6412                 goto out_unlock;
6413         }
6414         wait_on_page_writeback(page);
6415
6416         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6417                          GFP_NOFS);
6418         set_page_extent_mapped(page);
6419
6420         /*
6421          * we can't set the delalloc bits if there are pending ordered
6422          * extents.  Drop our locks and wait for them to finish
6423          */
6424         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6425         if (ordered) {
6426                 unlock_extent_cached(io_tree, page_start, page_end,
6427                                      &cached_state, GFP_NOFS);
6428                 unlock_page(page);
6429                 btrfs_start_ordered_extent(inode, ordered, 1);
6430                 btrfs_put_ordered_extent(ordered);
6431                 goto again;
6432         }
6433
6434         /*
6435          * XXX - page_mkwrite gets called every time the page is dirtied, even
6436          * if it was already dirty, so for space accounting reasons we need to
6437          * clear any delalloc bits for the range we are fixing to save.  There
6438          * is probably a better way to do this, but for now keep consistent with
6439          * prepare_pages in the normal write path.
6440          */
6441         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6442                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6443                           0, 0, &cached_state, GFP_NOFS);
6444
6445         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6446                                         &cached_state);
6447         if (ret) {
6448                 unlock_extent_cached(io_tree, page_start, page_end,
6449                                      &cached_state, GFP_NOFS);
6450                 ret = VM_FAULT_SIGBUS;
6451                 goto out_unlock;
6452         }
6453         ret = 0;
6454
6455         /* page is wholly or partially inside EOF */
6456         if (page_start + PAGE_CACHE_SIZE > size)
6457                 zero_start = size & ~PAGE_CACHE_MASK;
6458         else
6459                 zero_start = PAGE_CACHE_SIZE;
6460
6461         if (zero_start != PAGE_CACHE_SIZE) {
6462                 kaddr = kmap(page);
6463                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6464                 flush_dcache_page(page);
6465                 kunmap(page);
6466         }
6467         ClearPageChecked(page);
6468         set_page_dirty(page);
6469         SetPageUptodate(page);
6470
6471         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6472         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6473
6474         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6475
6476 out_unlock:
6477         if (!ret)
6478                 return VM_FAULT_LOCKED;
6479         unlock_page(page);
6480         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6481 out:
6482         return ret;
6483 }
6484
6485 static int btrfs_truncate(struct inode *inode)
6486 {
6487         struct btrfs_root *root = BTRFS_I(inode)->root;
6488         struct btrfs_block_rsv *rsv;
6489         int ret;
6490         int err = 0;
6491         struct btrfs_trans_handle *trans;
6492         unsigned long nr;
6493         u64 mask = root->sectorsize - 1;
6494
6495         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6496         if (ret)
6497                 return ret;
6498
6499         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6500         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6501
6502         /*
6503          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6504          * 3 things going on here
6505          *
6506          * 1) We need to reserve space for our orphan item and the space to
6507          * delete our orphan item.  Lord knows we don't want to have a dangling
6508          * orphan item because we didn't reserve space to remove it.
6509          *
6510          * 2) We need to reserve space to update our inode.
6511          *
6512          * 3) We need to have something to cache all the space that is going to
6513          * be free'd up by the truncate operation, but also have some slack
6514          * space reserved in case it uses space during the truncate (thank you
6515          * very much snapshotting).
6516          *
6517          * And we need these to all be seperate.  The fact is we can use alot of
6518          * space doing the truncate, and we have no earthly idea how much space
6519          * we will use, so we need the truncate reservation to be seperate so it
6520          * doesn't end up using space reserved for updating the inode or
6521          * removing the orphan item.  We also need to be able to stop the
6522          * transaction and start a new one, which means we need to be able to
6523          * update the inode several times, and we have no idea of knowing how
6524          * many times that will be, so we can't just reserve 1 item for the
6525          * entirety of the opration, so that has to be done seperately as well.
6526          * Then there is the orphan item, which does indeed need to be held on
6527          * to for the whole operation, and we need nobody to touch this reserved
6528          * space except the orphan code.
6529          *
6530          * So that leaves us with
6531          *
6532          * 1) root->orphan_block_rsv - for the orphan deletion.
6533          * 2) rsv - for the truncate reservation, which we will steal from the
6534          * transaction reservation.
6535          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6536          * updating the inode.
6537          */
6538         rsv = btrfs_alloc_block_rsv(root);
6539         if (!rsv)
6540                 return -ENOMEM;
6541         btrfs_add_durable_block_rsv(root->fs_info, rsv);
6542
6543         trans = btrfs_start_transaction(root, 4);
6544         if (IS_ERR(trans)) {
6545                 err = PTR_ERR(trans);
6546                 goto out;
6547         }
6548
6549         /*
6550          * Reserve space for the truncate process.  Truncate should be adding
6551          * space, but if there are snapshots it may end up using space.
6552          */
6553         ret = btrfs_truncate_reserve_metadata(trans, root, rsv);
6554         BUG_ON(ret);
6555
6556         ret = btrfs_orphan_add(trans, inode);
6557         if (ret) {
6558                 btrfs_end_transaction(trans, root);
6559                 goto out;
6560         }
6561
6562         nr = trans->blocks_used;
6563         btrfs_end_transaction(trans, root);
6564         btrfs_btree_balance_dirty(root, nr);
6565
6566         /*
6567          * Ok so we've already migrated our bytes over for the truncate, so here
6568          * just reserve the one slot we need for updating the inode.
6569          */
6570         trans = btrfs_start_transaction(root, 1);
6571         if (IS_ERR(trans)) {
6572                 err = PTR_ERR(trans);
6573                 goto out;
6574         }
6575         trans->block_rsv = rsv;
6576
6577         /*
6578          * setattr is responsible for setting the ordered_data_close flag,
6579          * but that is only tested during the last file release.  That
6580          * could happen well after the next commit, leaving a great big
6581          * window where new writes may get lost if someone chooses to write
6582          * to this file after truncating to zero
6583          *
6584          * The inode doesn't have any dirty data here, and so if we commit
6585          * this is a noop.  If someone immediately starts writing to the inode
6586          * it is very likely we'll catch some of their writes in this
6587          * transaction, and the commit will find this file on the ordered
6588          * data list with good things to send down.
6589          *
6590          * This is a best effort solution, there is still a window where
6591          * using truncate to replace the contents of the file will
6592          * end up with a zero length file after a crash.
6593          */
6594         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6595                 btrfs_add_ordered_operation(trans, root, inode);
6596
6597         while (1) {
6598                 if (!trans) {
6599                         trans = btrfs_start_transaction(root, 3);
6600                         if (IS_ERR(trans)) {
6601                                 err = PTR_ERR(trans);
6602                                 goto out;
6603                         }
6604
6605                         ret = btrfs_truncate_reserve_metadata(trans, root,
6606                                                               rsv);
6607                         BUG_ON(ret);
6608
6609                         trans->block_rsv = rsv;
6610                 }
6611
6612                 ret = btrfs_truncate_inode_items(trans, root, inode,
6613                                                  inode->i_size,
6614                                                  BTRFS_EXTENT_DATA_KEY);
6615                 if (ret != -EAGAIN) {
6616                         err = ret;
6617                         break;
6618                 }
6619
6620                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6621                 ret = btrfs_update_inode(trans, root, inode);
6622                 if (ret) {
6623                         err = ret;
6624                         break;
6625                 }
6626
6627                 nr = trans->blocks_used;
6628                 btrfs_end_transaction(trans, root);
6629                 trans = NULL;
6630                 btrfs_btree_balance_dirty(root, nr);
6631         }
6632
6633         if (ret == 0 && inode->i_nlink > 0) {
6634                 trans->block_rsv = root->orphan_block_rsv;
6635                 ret = btrfs_orphan_del(trans, inode);
6636                 if (ret)
6637                         err = ret;
6638         } else if (ret && inode->i_nlink > 0) {
6639                 /*
6640                  * Failed to do the truncate, remove us from the in memory
6641                  * orphan list.
6642                  */
6643                 ret = btrfs_orphan_del(NULL, inode);
6644         }
6645
6646         trans->block_rsv = &root->fs_info->trans_block_rsv;
6647         ret = btrfs_update_inode(trans, root, inode);
6648         if (ret && !err)
6649                 err = ret;
6650
6651         nr = trans->blocks_used;
6652         ret = btrfs_end_transaction_throttle(trans, root);
6653         btrfs_btree_balance_dirty(root, nr);
6654
6655 out:
6656         btrfs_free_block_rsv(root, rsv);
6657
6658         if (ret && !err)
6659                 err = ret;
6660
6661         return err;
6662 }
6663
6664 /*
6665  * create a new subvolume directory/inode (helper for the ioctl).
6666  */
6667 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6668                              struct btrfs_root *new_root, u64 new_dirid)
6669 {
6670         struct inode *inode;
6671         int err;
6672         u64 index = 0;
6673
6674         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6675                                 new_dirid, S_IFDIR | 0700, &index);
6676         if (IS_ERR(inode))
6677                 return PTR_ERR(inode);
6678         inode->i_op = &btrfs_dir_inode_operations;
6679         inode->i_fop = &btrfs_dir_file_operations;
6680
6681         inode->i_nlink = 1;
6682         btrfs_i_size_write(inode, 0);
6683
6684         err = btrfs_update_inode(trans, new_root, inode);
6685         BUG_ON(err);
6686
6687         iput(inode);
6688         return 0;
6689 }
6690
6691 /* helper function for file defrag and space balancing.  This
6692  * forces readahead on a given range of bytes in an inode
6693  */
6694 unsigned long btrfs_force_ra(struct address_space *mapping,
6695                               struct file_ra_state *ra, struct file *file,
6696                               pgoff_t offset, pgoff_t last_index)
6697 {
6698         pgoff_t req_size = last_index - offset + 1;
6699
6700         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6701         return offset + req_size;
6702 }
6703
6704 struct inode *btrfs_alloc_inode(struct super_block *sb)
6705 {
6706         struct btrfs_inode *ei;
6707         struct inode *inode;
6708
6709         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6710         if (!ei)
6711                 return NULL;
6712
6713         ei->root = NULL;
6714         ei->space_info = NULL;
6715         ei->generation = 0;
6716         ei->sequence = 0;
6717         ei->last_trans = 0;
6718         ei->last_sub_trans = 0;
6719         ei->logged_trans = 0;
6720         ei->delalloc_bytes = 0;
6721         ei->reserved_bytes = 0;
6722         ei->disk_i_size = 0;
6723         ei->flags = 0;
6724         ei->index_cnt = (u64)-1;
6725         ei->last_unlink_trans = 0;
6726
6727         atomic_set(&ei->outstanding_extents, 0);
6728         atomic_set(&ei->reserved_extents, 0);
6729
6730         ei->ordered_data_close = 0;
6731         ei->orphan_meta_reserved = 0;
6732         ei->dummy_inode = 0;
6733         ei->in_defrag = 0;
6734         ei->force_compress = BTRFS_COMPRESS_NONE;
6735
6736         ei->delayed_node = NULL;
6737
6738         inode = &ei->vfs_inode;
6739         extent_map_tree_init(&ei->extent_tree);
6740         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6741         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6742         mutex_init(&ei->log_mutex);
6743         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6744         INIT_LIST_HEAD(&ei->i_orphan);
6745         INIT_LIST_HEAD(&ei->delalloc_inodes);
6746         INIT_LIST_HEAD(&ei->ordered_operations);
6747         RB_CLEAR_NODE(&ei->rb_node);
6748
6749         return inode;
6750 }
6751
6752 static void btrfs_i_callback(struct rcu_head *head)
6753 {
6754         struct inode *inode = container_of(head, struct inode, i_rcu);
6755         INIT_LIST_HEAD(&inode->i_dentry);
6756         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6757 }
6758
6759 void btrfs_destroy_inode(struct inode *inode)
6760 {
6761         struct btrfs_ordered_extent *ordered;
6762         struct btrfs_root *root = BTRFS_I(inode)->root;
6763
6764         WARN_ON(!list_empty(&inode->i_dentry));
6765         WARN_ON(inode->i_data.nrpages);
6766         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6767         WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
6768
6769         /*
6770          * This can happen where we create an inode, but somebody else also
6771          * created the same inode and we need to destroy the one we already
6772          * created.
6773          */
6774         if (!root)
6775                 goto free;
6776
6777         /*
6778          * Make sure we're properly removed from the ordered operation
6779          * lists.
6780          */
6781         smp_mb();
6782         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6783                 spin_lock(&root->fs_info->ordered_extent_lock);
6784                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6785                 spin_unlock(&root->fs_info->ordered_extent_lock);
6786         }
6787
6788         spin_lock(&root->orphan_lock);
6789         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6790                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6791                        (unsigned long long)btrfs_ino(inode));
6792                 list_del_init(&BTRFS_I(inode)->i_orphan);
6793         }
6794         spin_unlock(&root->orphan_lock);
6795
6796         while (1) {
6797                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6798                 if (!ordered)
6799                         break;
6800                 else {
6801                         printk(KERN_ERR "btrfs found ordered "
6802                                "extent %llu %llu on inode cleanup\n",
6803                                (unsigned long long)ordered->file_offset,
6804                                (unsigned long long)ordered->len);
6805                         btrfs_remove_ordered_extent(inode, ordered);
6806                         btrfs_put_ordered_extent(ordered);
6807                         btrfs_put_ordered_extent(ordered);
6808                 }
6809         }
6810         inode_tree_del(inode);
6811         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6812 free:
6813         btrfs_remove_delayed_node(inode);
6814         call_rcu(&inode->i_rcu, btrfs_i_callback);
6815 }
6816
6817 int btrfs_drop_inode(struct inode *inode)
6818 {
6819         struct btrfs_root *root = BTRFS_I(inode)->root;
6820
6821         if (btrfs_root_refs(&root->root_item) == 0 &&
6822             !is_free_space_inode(root, inode))
6823                 return 1;
6824         else
6825                 return generic_drop_inode(inode);
6826 }
6827
6828 static void init_once(void *foo)
6829 {
6830         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6831
6832         inode_init_once(&ei->vfs_inode);
6833 }
6834
6835 void btrfs_destroy_cachep(void)
6836 {
6837         if (btrfs_inode_cachep)
6838                 kmem_cache_destroy(btrfs_inode_cachep);
6839         if (btrfs_trans_handle_cachep)
6840                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6841         if (btrfs_transaction_cachep)
6842                 kmem_cache_destroy(btrfs_transaction_cachep);
6843         if (btrfs_path_cachep)
6844                 kmem_cache_destroy(btrfs_path_cachep);
6845         if (btrfs_free_space_cachep)
6846                 kmem_cache_destroy(btrfs_free_space_cachep);
6847 }
6848
6849 int btrfs_init_cachep(void)
6850 {
6851         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6852                         sizeof(struct btrfs_inode), 0,
6853                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6854         if (!btrfs_inode_cachep)
6855                 goto fail;
6856
6857         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6858                         sizeof(struct btrfs_trans_handle), 0,
6859                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6860         if (!btrfs_trans_handle_cachep)
6861                 goto fail;
6862
6863         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6864                         sizeof(struct btrfs_transaction), 0,
6865                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6866         if (!btrfs_transaction_cachep)
6867                 goto fail;
6868
6869         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6870                         sizeof(struct btrfs_path), 0,
6871                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6872         if (!btrfs_path_cachep)
6873                 goto fail;
6874
6875         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6876                         sizeof(struct btrfs_free_space), 0,
6877                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6878         if (!btrfs_free_space_cachep)
6879                 goto fail;
6880
6881         return 0;
6882 fail:
6883         btrfs_destroy_cachep();
6884         return -ENOMEM;
6885 }
6886
6887 static int btrfs_getattr(struct vfsmount *mnt,
6888                          struct dentry *dentry, struct kstat *stat)
6889 {
6890         struct inode *inode = dentry->d_inode;
6891         generic_fillattr(inode, stat);
6892         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6893         stat->blksize = PAGE_CACHE_SIZE;
6894         stat->blocks = (inode_get_bytes(inode) +
6895                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6896         return 0;
6897 }
6898
6899 /*
6900  * If a file is moved, it will inherit the cow and compression flags of the new
6901  * directory.
6902  */
6903 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6904 {
6905         struct btrfs_inode *b_dir = BTRFS_I(dir);
6906         struct btrfs_inode *b_inode = BTRFS_I(inode);
6907
6908         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6909                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6910         else
6911                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6912
6913         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6914                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6915         else
6916                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6917 }
6918
6919 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6920                            struct inode *new_dir, struct dentry *new_dentry)
6921 {
6922         struct btrfs_trans_handle *trans;
6923         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6924         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6925         struct inode *new_inode = new_dentry->d_inode;
6926         struct inode *old_inode = old_dentry->d_inode;
6927         struct timespec ctime = CURRENT_TIME;
6928         u64 index = 0;
6929         u64 root_objectid;
6930         int ret;
6931         u64 old_ino = btrfs_ino(old_inode);
6932
6933         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6934                 return -EPERM;
6935
6936         /* we only allow rename subvolume link between subvolumes */
6937         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6938                 return -EXDEV;
6939
6940         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6941             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6942                 return -ENOTEMPTY;
6943
6944         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6945             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6946                 return -ENOTEMPTY;
6947         /*
6948          * we're using rename to replace one file with another.
6949          * and the replacement file is large.  Start IO on it now so
6950          * we don't add too much work to the end of the transaction
6951          */
6952         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6953             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6954                 filemap_flush(old_inode->i_mapping);
6955
6956         /* close the racy window with snapshot create/destroy ioctl */
6957         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6958                 down_read(&root->fs_info->subvol_sem);
6959         /*
6960          * We want to reserve the absolute worst case amount of items.  So if
6961          * both inodes are subvols and we need to unlink them then that would
6962          * require 4 item modifications, but if they are both normal inodes it
6963          * would require 5 item modifications, so we'll assume their normal
6964          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6965          * should cover the worst case number of items we'll modify.
6966          */
6967         trans = btrfs_start_transaction(root, 20);
6968         if (IS_ERR(trans)) {
6969                 ret = PTR_ERR(trans);
6970                 goto out_notrans;
6971         }
6972
6973         if (dest != root)
6974                 btrfs_record_root_in_trans(trans, dest);
6975
6976         ret = btrfs_set_inode_index(new_dir, &index);
6977         if (ret)
6978                 goto out_fail;
6979
6980         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6981                 /* force full log commit if subvolume involved. */
6982                 root->fs_info->last_trans_log_full_commit = trans->transid;
6983         } else {
6984                 ret = btrfs_insert_inode_ref(trans, dest,
6985                                              new_dentry->d_name.name,
6986                                              new_dentry->d_name.len,
6987                                              old_ino,
6988                                              btrfs_ino(new_dir), index);
6989                 if (ret)
6990                         goto out_fail;
6991                 /*
6992                  * this is an ugly little race, but the rename is required
6993                  * to make sure that if we crash, the inode is either at the
6994                  * old name or the new one.  pinning the log transaction lets
6995                  * us make sure we don't allow a log commit to come in after
6996                  * we unlink the name but before we add the new name back in.
6997                  */
6998                 btrfs_pin_log_trans(root);
6999         }
7000         /*
7001          * make sure the inode gets flushed if it is replacing
7002          * something.
7003          */
7004         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7005                 btrfs_add_ordered_operation(trans, root, old_inode);
7006
7007         old_dir->i_ctime = old_dir->i_mtime = ctime;
7008         new_dir->i_ctime = new_dir->i_mtime = ctime;
7009         old_inode->i_ctime = ctime;
7010
7011         if (old_dentry->d_parent != new_dentry->d_parent)
7012                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7013
7014         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7015                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7016                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7017                                         old_dentry->d_name.name,
7018                                         old_dentry->d_name.len);
7019         } else {
7020                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7021                                         old_dentry->d_inode,
7022                                         old_dentry->d_name.name,
7023                                         old_dentry->d_name.len);
7024                 if (!ret)
7025                         ret = btrfs_update_inode(trans, root, old_inode);
7026         }
7027         BUG_ON(ret);
7028
7029         if (new_inode) {
7030                 new_inode->i_ctime = CURRENT_TIME;
7031                 if (unlikely(btrfs_ino(new_inode) ==
7032                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7033                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7034                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7035                                                 root_objectid,
7036                                                 new_dentry->d_name.name,
7037                                                 new_dentry->d_name.len);
7038                         BUG_ON(new_inode->i_nlink == 0);
7039                 } else {
7040                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7041                                                  new_dentry->d_inode,
7042                                                  new_dentry->d_name.name,
7043                                                  new_dentry->d_name.len);
7044                 }
7045                 BUG_ON(ret);
7046                 if (new_inode->i_nlink == 0) {
7047                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7048                         BUG_ON(ret);
7049                 }
7050         }
7051
7052         fixup_inode_flags(new_dir, old_inode);
7053
7054         ret = btrfs_add_link(trans, new_dir, old_inode,
7055                              new_dentry->d_name.name,
7056                              new_dentry->d_name.len, 0, index);
7057         BUG_ON(ret);
7058
7059         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7060                 struct dentry *parent = dget_parent(new_dentry);
7061                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7062                 dput(parent);
7063                 btrfs_end_log_trans(root);
7064         }
7065 out_fail:
7066         btrfs_end_transaction_throttle(trans, root);
7067 out_notrans:
7068         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7069                 up_read(&root->fs_info->subvol_sem);
7070
7071         return ret;
7072 }
7073
7074 /*
7075  * some fairly slow code that needs optimization. This walks the list
7076  * of all the inodes with pending delalloc and forces them to disk.
7077  */
7078 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7079 {
7080         struct list_head *head = &root->fs_info->delalloc_inodes;
7081         struct btrfs_inode *binode;
7082         struct inode *inode;
7083
7084         if (root->fs_info->sb->s_flags & MS_RDONLY)
7085                 return -EROFS;
7086
7087         spin_lock(&root->fs_info->delalloc_lock);
7088         while (!list_empty(head)) {
7089                 binode = list_entry(head->next, struct btrfs_inode,
7090                                     delalloc_inodes);
7091                 inode = igrab(&binode->vfs_inode);
7092                 if (!inode)
7093                         list_del_init(&binode->delalloc_inodes);
7094                 spin_unlock(&root->fs_info->delalloc_lock);
7095                 if (inode) {
7096                         filemap_flush(inode->i_mapping);
7097                         if (delay_iput)
7098                                 btrfs_add_delayed_iput(inode);
7099                         else
7100                                 iput(inode);
7101                 }
7102                 cond_resched();
7103                 spin_lock(&root->fs_info->delalloc_lock);
7104         }
7105         spin_unlock(&root->fs_info->delalloc_lock);
7106
7107         /* the filemap_flush will queue IO into the worker threads, but
7108          * we have to make sure the IO is actually started and that
7109          * ordered extents get created before we return
7110          */
7111         atomic_inc(&root->fs_info->async_submit_draining);
7112         while (atomic_read(&root->fs_info->nr_async_submits) ||
7113               atomic_read(&root->fs_info->async_delalloc_pages)) {
7114                 wait_event(root->fs_info->async_submit_wait,
7115                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7116                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7117         }
7118         atomic_dec(&root->fs_info->async_submit_draining);
7119         return 0;
7120 }
7121
7122 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7123                          const char *symname)
7124 {
7125         struct btrfs_trans_handle *trans;
7126         struct btrfs_root *root = BTRFS_I(dir)->root;
7127         struct btrfs_path *path;
7128         struct btrfs_key key;
7129         struct inode *inode = NULL;
7130         int err;
7131         int drop_inode = 0;
7132         u64 objectid;
7133         u64 index = 0 ;
7134         int name_len;
7135         int datasize;
7136         unsigned long ptr;
7137         struct btrfs_file_extent_item *ei;
7138         struct extent_buffer *leaf;
7139         unsigned long nr = 0;
7140
7141         name_len = strlen(symname) + 1;
7142         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7143                 return -ENAMETOOLONG;
7144
7145         /*
7146          * 2 items for inode item and ref
7147          * 2 items for dir items
7148          * 1 item for xattr if selinux is on
7149          */
7150         trans = btrfs_start_transaction(root, 5);
7151         if (IS_ERR(trans))
7152                 return PTR_ERR(trans);
7153
7154         err = btrfs_find_free_ino(root, &objectid);
7155         if (err)
7156                 goto out_unlock;
7157
7158         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7159                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7160                                 S_IFLNK|S_IRWXUGO, &index);
7161         if (IS_ERR(inode)) {
7162                 err = PTR_ERR(inode);
7163                 goto out_unlock;
7164         }
7165
7166         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7167         if (err) {
7168                 drop_inode = 1;
7169                 goto out_unlock;
7170         }
7171
7172         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7173         if (err)
7174                 drop_inode = 1;
7175         else {
7176                 inode->i_mapping->a_ops = &btrfs_aops;
7177                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7178                 inode->i_fop = &btrfs_file_operations;
7179                 inode->i_op = &btrfs_file_inode_operations;
7180                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7181         }
7182         if (drop_inode)
7183                 goto out_unlock;
7184
7185         path = btrfs_alloc_path();
7186         BUG_ON(!path);
7187         key.objectid = btrfs_ino(inode);
7188         key.offset = 0;
7189         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7190         datasize = btrfs_file_extent_calc_inline_size(name_len);
7191         err = btrfs_insert_empty_item(trans, root, path, &key,
7192                                       datasize);
7193         if (err) {
7194                 drop_inode = 1;
7195                 btrfs_free_path(path);
7196                 goto out_unlock;
7197         }
7198         leaf = path->nodes[0];
7199         ei = btrfs_item_ptr(leaf, path->slots[0],
7200                             struct btrfs_file_extent_item);
7201         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7202         btrfs_set_file_extent_type(leaf, ei,
7203                                    BTRFS_FILE_EXTENT_INLINE);
7204         btrfs_set_file_extent_encryption(leaf, ei, 0);
7205         btrfs_set_file_extent_compression(leaf, ei, 0);
7206         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7207         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7208
7209         ptr = btrfs_file_extent_inline_start(ei);
7210         write_extent_buffer(leaf, symname, ptr, name_len);
7211         btrfs_mark_buffer_dirty(leaf);
7212         btrfs_free_path(path);
7213
7214         inode->i_op = &btrfs_symlink_inode_operations;
7215         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7216         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7217         inode_set_bytes(inode, name_len);
7218         btrfs_i_size_write(inode, name_len - 1);
7219         err = btrfs_update_inode(trans, root, inode);
7220         if (err)
7221                 drop_inode = 1;
7222
7223 out_unlock:
7224         nr = trans->blocks_used;
7225         btrfs_end_transaction_throttle(trans, root);
7226         if (drop_inode) {
7227                 inode_dec_link_count(inode);
7228                 iput(inode);
7229         }
7230         btrfs_btree_balance_dirty(root, nr);
7231         return err;
7232 }
7233
7234 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7235                                        u64 start, u64 num_bytes, u64 min_size,
7236                                        loff_t actual_len, u64 *alloc_hint,
7237                                        struct btrfs_trans_handle *trans)
7238 {
7239         struct btrfs_root *root = BTRFS_I(inode)->root;
7240         struct btrfs_key ins;
7241         u64 cur_offset = start;
7242         u64 i_size;
7243         int ret = 0;
7244         bool own_trans = true;
7245
7246         if (trans)
7247                 own_trans = false;
7248         while (num_bytes > 0) {
7249                 if (own_trans) {
7250                         trans = btrfs_start_transaction(root, 3);
7251                         if (IS_ERR(trans)) {
7252                                 ret = PTR_ERR(trans);
7253                                 break;
7254                         }
7255                 }
7256
7257                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7258                                            0, *alloc_hint, (u64)-1, &ins, 1);
7259                 if (ret) {
7260                         if (own_trans)
7261                                 btrfs_end_transaction(trans, root);
7262                         break;
7263                 }
7264
7265                 ret = insert_reserved_file_extent(trans, inode,
7266                                                   cur_offset, ins.objectid,
7267                                                   ins.offset, ins.offset,
7268                                                   ins.offset, 0, 0, 0,
7269                                                   BTRFS_FILE_EXTENT_PREALLOC);
7270                 BUG_ON(ret);
7271                 btrfs_drop_extent_cache(inode, cur_offset,
7272                                         cur_offset + ins.offset -1, 0);
7273
7274                 num_bytes -= ins.offset;
7275                 cur_offset += ins.offset;
7276                 *alloc_hint = ins.objectid + ins.offset;
7277
7278                 inode->i_ctime = CURRENT_TIME;
7279                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7280                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7281                     (actual_len > inode->i_size) &&
7282                     (cur_offset > inode->i_size)) {
7283                         if (cur_offset > actual_len)
7284                                 i_size = actual_len;
7285                         else
7286                                 i_size = cur_offset;
7287                         i_size_write(inode, i_size);
7288                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7289                 }
7290
7291                 ret = btrfs_update_inode(trans, root, inode);
7292                 BUG_ON(ret);
7293
7294                 if (own_trans)
7295                         btrfs_end_transaction(trans, root);
7296         }
7297         return ret;
7298 }
7299
7300 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7301                               u64 start, u64 num_bytes, u64 min_size,
7302                               loff_t actual_len, u64 *alloc_hint)
7303 {
7304         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7305                                            min_size, actual_len, alloc_hint,
7306                                            NULL);
7307 }
7308
7309 int btrfs_prealloc_file_range_trans(struct inode *inode,
7310                                     struct btrfs_trans_handle *trans, int mode,
7311                                     u64 start, u64 num_bytes, u64 min_size,
7312                                     loff_t actual_len, u64 *alloc_hint)
7313 {
7314         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7315                                            min_size, actual_len, alloc_hint, trans);
7316 }
7317
7318 static int btrfs_set_page_dirty(struct page *page)
7319 {
7320         return __set_page_dirty_nobuffers(page);
7321 }
7322
7323 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7324 {
7325         struct btrfs_root *root = BTRFS_I(inode)->root;
7326
7327         if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7328                 return -EROFS;
7329         if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7330                 return -EACCES;
7331         return generic_permission(inode, mask, flags, btrfs_check_acl);
7332 }
7333
7334 static const struct inode_operations btrfs_dir_inode_operations = {
7335         .getattr        = btrfs_getattr,
7336         .lookup         = btrfs_lookup,
7337         .create         = btrfs_create,
7338         .unlink         = btrfs_unlink,
7339         .link           = btrfs_link,
7340         .mkdir          = btrfs_mkdir,
7341         .rmdir          = btrfs_rmdir,
7342         .rename         = btrfs_rename,
7343         .symlink        = btrfs_symlink,
7344         .setattr        = btrfs_setattr,
7345         .mknod          = btrfs_mknod,
7346         .setxattr       = btrfs_setxattr,
7347         .getxattr       = btrfs_getxattr,
7348         .listxattr      = btrfs_listxattr,
7349         .removexattr    = btrfs_removexattr,
7350         .permission     = btrfs_permission,
7351 };
7352 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7353         .lookup         = btrfs_lookup,
7354         .permission     = btrfs_permission,
7355 };
7356
7357 static const struct file_operations btrfs_dir_file_operations = {
7358         .llseek         = generic_file_llseek,
7359         .read           = generic_read_dir,
7360         .readdir        = btrfs_real_readdir,
7361         .unlocked_ioctl = btrfs_ioctl,
7362 #ifdef CONFIG_COMPAT
7363         .compat_ioctl   = btrfs_ioctl,
7364 #endif
7365         .release        = btrfs_release_file,
7366         .fsync          = btrfs_sync_file,
7367 };
7368
7369 static struct extent_io_ops btrfs_extent_io_ops = {
7370         .fill_delalloc = run_delalloc_range,
7371         .submit_bio_hook = btrfs_submit_bio_hook,
7372         .merge_bio_hook = btrfs_merge_bio_hook,
7373         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7374         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7375         .writepage_start_hook = btrfs_writepage_start_hook,
7376         .readpage_io_failed_hook = btrfs_io_failed_hook,
7377         .set_bit_hook = btrfs_set_bit_hook,
7378         .clear_bit_hook = btrfs_clear_bit_hook,
7379         .merge_extent_hook = btrfs_merge_extent_hook,
7380         .split_extent_hook = btrfs_split_extent_hook,
7381 };
7382
7383 /*
7384  * btrfs doesn't support the bmap operation because swapfiles
7385  * use bmap to make a mapping of extents in the file.  They assume
7386  * these extents won't change over the life of the file and they
7387  * use the bmap result to do IO directly to the drive.
7388  *
7389  * the btrfs bmap call would return logical addresses that aren't
7390  * suitable for IO and they also will change frequently as COW
7391  * operations happen.  So, swapfile + btrfs == corruption.
7392  *
7393  * For now we're avoiding this by dropping bmap.
7394  */
7395 static const struct address_space_operations btrfs_aops = {
7396         .readpage       = btrfs_readpage,
7397         .writepage      = btrfs_writepage,
7398         .writepages     = btrfs_writepages,
7399         .readpages      = btrfs_readpages,
7400         .direct_IO      = btrfs_direct_IO,
7401         .invalidatepage = btrfs_invalidatepage,
7402         .releasepage    = btrfs_releasepage,
7403         .set_page_dirty = btrfs_set_page_dirty,
7404         .error_remove_page = generic_error_remove_page,
7405 };
7406
7407 static const struct address_space_operations btrfs_symlink_aops = {
7408         .readpage       = btrfs_readpage,
7409         .writepage      = btrfs_writepage,
7410         .invalidatepage = btrfs_invalidatepage,
7411         .releasepage    = btrfs_releasepage,
7412 };
7413
7414 static const struct inode_operations btrfs_file_inode_operations = {
7415         .getattr        = btrfs_getattr,
7416         .setattr        = btrfs_setattr,
7417         .setxattr       = btrfs_setxattr,
7418         .getxattr       = btrfs_getxattr,
7419         .listxattr      = btrfs_listxattr,
7420         .removexattr    = btrfs_removexattr,
7421         .permission     = btrfs_permission,
7422         .fiemap         = btrfs_fiemap,
7423 };
7424 static const struct inode_operations btrfs_special_inode_operations = {
7425         .getattr        = btrfs_getattr,
7426         .setattr        = btrfs_setattr,
7427         .permission     = btrfs_permission,
7428         .setxattr       = btrfs_setxattr,
7429         .getxattr       = btrfs_getxattr,
7430         .listxattr      = btrfs_listxattr,
7431         .removexattr    = btrfs_removexattr,
7432 };
7433 static const struct inode_operations btrfs_symlink_inode_operations = {
7434         .readlink       = generic_readlink,
7435         .follow_link    = page_follow_link_light,
7436         .put_link       = page_put_link,
7437         .getattr        = btrfs_getattr,
7438         .permission     = btrfs_permission,
7439         .setxattr       = btrfs_setxattr,
7440         .getxattr       = btrfs_getxattr,
7441         .listxattr      = btrfs_listxattr,
7442         .removexattr    = btrfs_removexattr,
7443 };
7444
7445 const struct dentry_operations btrfs_dentry_operations = {
7446         .d_delete       = btrfs_dentry_delete,
7447 };