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         spin_lock(&BTRFS_I(inode)->lock);
1302         BTRFS_I(inode)->outstanding_extents++;
1303         spin_unlock(&BTRFS_I(inode)->lock);
1304         return 0;
1305 }
1306
1307 /*
1308  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1309  * extents so we can keep track of new extents that are just merged onto old
1310  * extents, such as when we are doing sequential writes, so we can properly
1311  * account for the metadata space we'll need.
1312  */
1313 static int btrfs_merge_extent_hook(struct inode *inode,
1314                                    struct extent_state *new,
1315                                    struct extent_state *other)
1316 {
1317         /* not delalloc, ignore it */
1318         if (!(other->state & EXTENT_DELALLOC))
1319                 return 0;
1320
1321         spin_lock(&BTRFS_I(inode)->lock);
1322         BTRFS_I(inode)->outstanding_extents--;
1323         spin_unlock(&BTRFS_I(inode)->lock);
1324         return 0;
1325 }
1326
1327 /*
1328  * extent_io.c set_bit_hook, used to track delayed allocation
1329  * bytes in this file, and to maintain the list of inodes that
1330  * have pending delalloc work to be done.
1331  */
1332 static int btrfs_set_bit_hook(struct inode *inode,
1333                               struct extent_state *state, int *bits)
1334 {
1335
1336         /*
1337          * set_bit and clear bit hooks normally require _irqsave/restore
1338          * but in this case, we are only testing for the DELALLOC
1339          * bit, which is only set or cleared with irqs on
1340          */
1341         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1342                 struct btrfs_root *root = BTRFS_I(inode)->root;
1343                 u64 len = state->end + 1 - state->start;
1344                 bool do_list = !is_free_space_inode(root, inode);
1345
1346                 if (*bits & EXTENT_FIRST_DELALLOC) {
1347                         *bits &= ~EXTENT_FIRST_DELALLOC;
1348                 } else {
1349                         spin_lock(&BTRFS_I(inode)->lock);
1350                         BTRFS_I(inode)->outstanding_extents++;
1351                         spin_unlock(&BTRFS_I(inode)->lock);
1352                 }
1353
1354                 spin_lock(&root->fs_info->delalloc_lock);
1355                 BTRFS_I(inode)->delalloc_bytes += len;
1356                 root->fs_info->delalloc_bytes += len;
1357                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1358                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1359                                       &root->fs_info->delalloc_inodes);
1360                 }
1361                 spin_unlock(&root->fs_info->delalloc_lock);
1362         }
1363         return 0;
1364 }
1365
1366 /*
1367  * extent_io.c clear_bit_hook, see set_bit_hook for why
1368  */
1369 static int btrfs_clear_bit_hook(struct inode *inode,
1370                                 struct extent_state *state, int *bits)
1371 {
1372         /*
1373          * set_bit and clear bit hooks normally require _irqsave/restore
1374          * but in this case, we are only testing for the DELALLOC
1375          * bit, which is only set or cleared with irqs on
1376          */
1377         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1378                 struct btrfs_root *root = BTRFS_I(inode)->root;
1379                 u64 len = state->end + 1 - state->start;
1380                 bool do_list = !is_free_space_inode(root, inode);
1381
1382                 if (*bits & EXTENT_FIRST_DELALLOC) {
1383                         *bits &= ~EXTENT_FIRST_DELALLOC;
1384                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1385                         spin_lock(&BTRFS_I(inode)->lock);
1386                         BTRFS_I(inode)->outstanding_extents--;
1387                         spin_unlock(&BTRFS_I(inode)->lock);
1388                 }
1389
1390                 if (*bits & EXTENT_DO_ACCOUNTING)
1391                         btrfs_delalloc_release_metadata(inode, len);
1392
1393                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1394                     && do_list)
1395                         btrfs_free_reserved_data_space(inode, len);
1396
1397                 spin_lock(&root->fs_info->delalloc_lock);
1398                 root->fs_info->delalloc_bytes -= len;
1399                 BTRFS_I(inode)->delalloc_bytes -= len;
1400
1401                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1402                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1403                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1404                 }
1405                 spin_unlock(&root->fs_info->delalloc_lock);
1406         }
1407         return 0;
1408 }
1409
1410 /*
1411  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1412  * we don't create bios that span stripes or chunks
1413  */
1414 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1415                          size_t size, struct bio *bio,
1416                          unsigned long bio_flags)
1417 {
1418         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1419         struct btrfs_mapping_tree *map_tree;
1420         u64 logical = (u64)bio->bi_sector << 9;
1421         u64 length = 0;
1422         u64 map_length;
1423         int ret;
1424
1425         if (bio_flags & EXTENT_BIO_COMPRESSED)
1426                 return 0;
1427
1428         length = bio->bi_size;
1429         map_tree = &root->fs_info->mapping_tree;
1430         map_length = length;
1431         ret = btrfs_map_block(map_tree, READ, logical,
1432                               &map_length, NULL, 0);
1433
1434         if (map_length < length + size)
1435                 return 1;
1436         return ret;
1437 }
1438
1439 /*
1440  * in order to insert checksums into the metadata in large chunks,
1441  * we wait until bio submission time.   All the pages in the bio are
1442  * checksummed and sums are attached onto the ordered extent record.
1443  *
1444  * At IO completion time the cums attached on the ordered extent record
1445  * are inserted into the btree
1446  */
1447 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1448                                     struct bio *bio, int mirror_num,
1449                                     unsigned long bio_flags,
1450                                     u64 bio_offset)
1451 {
1452         struct btrfs_root *root = BTRFS_I(inode)->root;
1453         int ret = 0;
1454
1455         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1456         BUG_ON(ret);
1457         return 0;
1458 }
1459
1460 /*
1461  * in order to insert checksums into the metadata in large chunks,
1462  * we wait until bio submission time.   All the pages in the bio are
1463  * checksummed and sums are attached onto the ordered extent record.
1464  *
1465  * At IO completion time the cums attached on the ordered extent record
1466  * are inserted into the btree
1467  */
1468 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1469                           int mirror_num, unsigned long bio_flags,
1470                           u64 bio_offset)
1471 {
1472         struct btrfs_root *root = BTRFS_I(inode)->root;
1473         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1474 }
1475
1476 /*
1477  * extent_io.c submission hook. This does the right thing for csum calculation
1478  * on write, or reading the csums from the tree before a read
1479  */
1480 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1481                           int mirror_num, unsigned long bio_flags,
1482                           u64 bio_offset)
1483 {
1484         struct btrfs_root *root = BTRFS_I(inode)->root;
1485         int ret = 0;
1486         int skip_sum;
1487
1488         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1489
1490         if (is_free_space_inode(root, inode))
1491                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1492         else
1493                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1494         BUG_ON(ret);
1495
1496         if (!(rw & REQ_WRITE)) {
1497                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1498                         return btrfs_submit_compressed_read(inode, bio,
1499                                                     mirror_num, bio_flags);
1500                 } else if (!skip_sum) {
1501                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1502                         if (ret)
1503                                 return ret;
1504                 }
1505                 goto mapit;
1506         } else if (!skip_sum) {
1507                 /* csum items have already been cloned */
1508                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1509                         goto mapit;
1510                 /* we're doing a write, do the async checksumming */
1511                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1512                                    inode, rw, bio, mirror_num,
1513                                    bio_flags, bio_offset,
1514                                    __btrfs_submit_bio_start,
1515                                    __btrfs_submit_bio_done);
1516         }
1517
1518 mapit:
1519         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1520 }
1521
1522 /*
1523  * given a list of ordered sums record them in the inode.  This happens
1524  * at IO completion time based on sums calculated at bio submission time.
1525  */
1526 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1527                              struct inode *inode, u64 file_offset,
1528                              struct list_head *list)
1529 {
1530         struct btrfs_ordered_sum *sum;
1531
1532         list_for_each_entry(sum, list, list) {
1533                 btrfs_csum_file_blocks(trans,
1534                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1535         }
1536         return 0;
1537 }
1538
1539 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1540                               struct extent_state **cached_state)
1541 {
1542         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1543                 WARN_ON(1);
1544         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1545                                    cached_state, GFP_NOFS);
1546 }
1547
1548 /* see btrfs_writepage_start_hook for details on why this is required */
1549 struct btrfs_writepage_fixup {
1550         struct page *page;
1551         struct btrfs_work work;
1552 };
1553
1554 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1555 {
1556         struct btrfs_writepage_fixup *fixup;
1557         struct btrfs_ordered_extent *ordered;
1558         struct extent_state *cached_state = NULL;
1559         struct page *page;
1560         struct inode *inode;
1561         u64 page_start;
1562         u64 page_end;
1563
1564         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1565         page = fixup->page;
1566 again:
1567         lock_page(page);
1568         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1569                 ClearPageChecked(page);
1570                 goto out_page;
1571         }
1572
1573         inode = page->mapping->host;
1574         page_start = page_offset(page);
1575         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1576
1577         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1578                          &cached_state, GFP_NOFS);
1579
1580         /* already ordered? We're done */
1581         if (PagePrivate2(page))
1582                 goto out;
1583
1584         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1585         if (ordered) {
1586                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1587                                      page_end, &cached_state, GFP_NOFS);
1588                 unlock_page(page);
1589                 btrfs_start_ordered_extent(inode, ordered, 1);
1590                 goto again;
1591         }
1592
1593         BUG();
1594         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1595         ClearPageChecked(page);
1596 out:
1597         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1598                              &cached_state, GFP_NOFS);
1599 out_page:
1600         unlock_page(page);
1601         page_cache_release(page);
1602         kfree(fixup);
1603 }
1604
1605 /*
1606  * There are a few paths in the higher layers of the kernel that directly
1607  * set the page dirty bit without asking the filesystem if it is a
1608  * good idea.  This causes problems because we want to make sure COW
1609  * properly happens and the data=ordered rules are followed.
1610  *
1611  * In our case any range that doesn't have the ORDERED bit set
1612  * hasn't been properly setup for IO.  We kick off an async process
1613  * to fix it up.  The async helper will wait for ordered extents, set
1614  * the delalloc bit and make it safe to write the page.
1615  */
1616 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1617 {
1618         struct inode *inode = page->mapping->host;
1619         struct btrfs_writepage_fixup *fixup;
1620         struct btrfs_root *root = BTRFS_I(inode)->root;
1621
1622         /* this page is properly in the ordered list */
1623         if (TestClearPagePrivate2(page))
1624                 return 0;
1625
1626         if (PageChecked(page))
1627                 return -EAGAIN;
1628
1629         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1630         if (!fixup)
1631                 return -EAGAIN;
1632
1633         SetPageChecked(page);
1634         page_cache_get(page);
1635         fixup->work.func = btrfs_writepage_fixup_worker;
1636         fixup->page = page;
1637         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1638         return -EAGAIN;
1639 }
1640
1641 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1642                                        struct inode *inode, u64 file_pos,
1643                                        u64 disk_bytenr, u64 disk_num_bytes,
1644                                        u64 num_bytes, u64 ram_bytes,
1645                                        u8 compression, u8 encryption,
1646                                        u16 other_encoding, int extent_type)
1647 {
1648         struct btrfs_root *root = BTRFS_I(inode)->root;
1649         struct btrfs_file_extent_item *fi;
1650         struct btrfs_path *path;
1651         struct extent_buffer *leaf;
1652         struct btrfs_key ins;
1653         u64 hint;
1654         int ret;
1655
1656         path = btrfs_alloc_path();
1657         BUG_ON(!path);
1658
1659         path->leave_spinning = 1;
1660
1661         /*
1662          * we may be replacing one extent in the tree with another.
1663          * The new extent is pinned in the extent map, and we don't want
1664          * to drop it from the cache until it is completely in the btree.
1665          *
1666          * So, tell btrfs_drop_extents to leave this extent in the cache.
1667          * the caller is expected to unpin it and allow it to be merged
1668          * with the others.
1669          */
1670         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1671                                  &hint, 0);
1672         BUG_ON(ret);
1673
1674         ins.objectid = btrfs_ino(inode);
1675         ins.offset = file_pos;
1676         ins.type = BTRFS_EXTENT_DATA_KEY;
1677         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1678         BUG_ON(ret);
1679         leaf = path->nodes[0];
1680         fi = btrfs_item_ptr(leaf, path->slots[0],
1681                             struct btrfs_file_extent_item);
1682         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1683         btrfs_set_file_extent_type(leaf, fi, extent_type);
1684         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1685         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1686         btrfs_set_file_extent_offset(leaf, fi, 0);
1687         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1688         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1689         btrfs_set_file_extent_compression(leaf, fi, compression);
1690         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1691         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1692
1693         btrfs_unlock_up_safe(path, 1);
1694         btrfs_set_lock_blocking(leaf);
1695
1696         btrfs_mark_buffer_dirty(leaf);
1697
1698         inode_add_bytes(inode, num_bytes);
1699
1700         ins.objectid = disk_bytenr;
1701         ins.offset = disk_num_bytes;
1702         ins.type = BTRFS_EXTENT_ITEM_KEY;
1703         ret = btrfs_alloc_reserved_file_extent(trans, root,
1704                                         root->root_key.objectid,
1705                                         btrfs_ino(inode), file_pos, &ins);
1706         BUG_ON(ret);
1707         btrfs_free_path(path);
1708
1709         return 0;
1710 }
1711
1712 /*
1713  * helper function for btrfs_finish_ordered_io, this
1714  * just reads in some of the csum leaves to prime them into ram
1715  * before we start the transaction.  It limits the amount of btree
1716  * reads required while inside the transaction.
1717  */
1718 /* as ordered data IO finishes, this gets called so we can finish
1719  * an ordered extent if the range of bytes in the file it covers are
1720  * fully written.
1721  */
1722 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1723 {
1724         struct btrfs_root *root = BTRFS_I(inode)->root;
1725         struct btrfs_trans_handle *trans = NULL;
1726         struct btrfs_ordered_extent *ordered_extent = NULL;
1727         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1728         struct extent_state *cached_state = NULL;
1729         int compress_type = 0;
1730         int ret;
1731         bool nolock;
1732
1733         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1734                                              end - start + 1);
1735         if (!ret)
1736                 return 0;
1737         BUG_ON(!ordered_extent);
1738
1739         nolock = is_free_space_inode(root, inode);
1740
1741         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1742                 BUG_ON(!list_empty(&ordered_extent->list));
1743                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1744                 if (!ret) {
1745                         if (nolock)
1746                                 trans = btrfs_join_transaction_nolock(root);
1747                         else
1748                                 trans = btrfs_join_transaction(root);
1749                         BUG_ON(IS_ERR(trans));
1750                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1751                         ret = btrfs_update_inode(trans, root, inode);
1752                         BUG_ON(ret);
1753                 }
1754                 goto out;
1755         }
1756
1757         lock_extent_bits(io_tree, ordered_extent->file_offset,
1758                          ordered_extent->file_offset + ordered_extent->len - 1,
1759                          0, &cached_state, GFP_NOFS);
1760
1761         if (nolock)
1762                 trans = btrfs_join_transaction_nolock(root);
1763         else
1764                 trans = btrfs_join_transaction(root);
1765         BUG_ON(IS_ERR(trans));
1766         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1767
1768         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1769                 compress_type = ordered_extent->compress_type;
1770         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1771                 BUG_ON(compress_type);
1772                 ret = btrfs_mark_extent_written(trans, inode,
1773                                                 ordered_extent->file_offset,
1774                                                 ordered_extent->file_offset +
1775                                                 ordered_extent->len);
1776                 BUG_ON(ret);
1777         } else {
1778                 BUG_ON(root == root->fs_info->tree_root);
1779                 ret = insert_reserved_file_extent(trans, inode,
1780                                                 ordered_extent->file_offset,
1781                                                 ordered_extent->start,
1782                                                 ordered_extent->disk_len,
1783                                                 ordered_extent->len,
1784                                                 ordered_extent->len,
1785                                                 compress_type, 0, 0,
1786                                                 BTRFS_FILE_EXTENT_REG);
1787                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1788                                    ordered_extent->file_offset,
1789                                    ordered_extent->len);
1790                 BUG_ON(ret);
1791         }
1792         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1793                              ordered_extent->file_offset +
1794                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1795
1796         add_pending_csums(trans, inode, ordered_extent->file_offset,
1797                           &ordered_extent->list);
1798
1799         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1800         if (!ret) {
1801                 ret = btrfs_update_inode(trans, root, inode);
1802                 BUG_ON(ret);
1803         }
1804         ret = 0;
1805 out:
1806         if (nolock) {
1807                 if (trans)
1808                         btrfs_end_transaction_nolock(trans, root);
1809         } else {
1810                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1811                 if (trans)
1812                         btrfs_end_transaction(trans, root);
1813         }
1814
1815         /* once for us */
1816         btrfs_put_ordered_extent(ordered_extent);
1817         /* once for the tree */
1818         btrfs_put_ordered_extent(ordered_extent);
1819
1820         return 0;
1821 }
1822
1823 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1824                                 struct extent_state *state, int uptodate)
1825 {
1826         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1827
1828         ClearPagePrivate2(page);
1829         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1830 }
1831
1832 /*
1833  * When IO fails, either with EIO or csum verification fails, we
1834  * try other mirrors that might have a good copy of the data.  This
1835  * io_failure_record is used to record state as we go through all the
1836  * mirrors.  If another mirror has good data, the page is set up to date
1837  * and things continue.  If a good mirror can't be found, the original
1838  * bio end_io callback is called to indicate things have failed.
1839  */
1840 struct io_failure_record {
1841         struct page *page;
1842         u64 start;
1843         u64 len;
1844         u64 logical;
1845         unsigned long bio_flags;
1846         int last_mirror;
1847 };
1848
1849 static int btrfs_io_failed_hook(struct bio *failed_bio,
1850                          struct page *page, u64 start, u64 end,
1851                          struct extent_state *state)
1852 {
1853         struct io_failure_record *failrec = NULL;
1854         u64 private;
1855         struct extent_map *em;
1856         struct inode *inode = page->mapping->host;
1857         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1858         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1859         struct bio *bio;
1860         int num_copies;
1861         int ret;
1862         int rw;
1863         u64 logical;
1864
1865         ret = get_state_private(failure_tree, start, &private);
1866         if (ret) {
1867                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1868                 if (!failrec)
1869                         return -ENOMEM;
1870                 failrec->start = start;
1871                 failrec->len = end - start + 1;
1872                 failrec->last_mirror = 0;
1873                 failrec->bio_flags = 0;
1874
1875                 read_lock(&em_tree->lock);
1876                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1877                 if (em->start > start || em->start + em->len < start) {
1878                         free_extent_map(em);
1879                         em = NULL;
1880                 }
1881                 read_unlock(&em_tree->lock);
1882
1883                 if (IS_ERR_OR_NULL(em)) {
1884                         kfree(failrec);
1885                         return -EIO;
1886                 }
1887                 logical = start - em->start;
1888                 logical = em->block_start + logical;
1889                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1890                         logical = em->block_start;
1891                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1892                         extent_set_compress_type(&failrec->bio_flags,
1893                                                  em->compress_type);
1894                 }
1895                 failrec->logical = logical;
1896                 free_extent_map(em);
1897                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1898                                 EXTENT_DIRTY, GFP_NOFS);
1899                 set_state_private(failure_tree, start,
1900                                  (u64)(unsigned long)failrec);
1901         } else {
1902                 failrec = (struct io_failure_record *)(unsigned long)private;
1903         }
1904         num_copies = btrfs_num_copies(
1905                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1906                               failrec->logical, failrec->len);
1907         failrec->last_mirror++;
1908         if (!state) {
1909                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1910                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1911                                                     failrec->start,
1912                                                     EXTENT_LOCKED);
1913                 if (state && state->start != failrec->start)
1914                         state = NULL;
1915                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1916         }
1917         if (!state || failrec->last_mirror > num_copies) {
1918                 set_state_private(failure_tree, failrec->start, 0);
1919                 clear_extent_bits(failure_tree, failrec->start,
1920                                   failrec->start + failrec->len - 1,
1921                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1922                 kfree(failrec);
1923                 return -EIO;
1924         }
1925         bio = bio_alloc(GFP_NOFS, 1);
1926         bio->bi_private = state;
1927         bio->bi_end_io = failed_bio->bi_end_io;
1928         bio->bi_sector = failrec->logical >> 9;
1929         bio->bi_bdev = failed_bio->bi_bdev;
1930         bio->bi_size = 0;
1931
1932         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1933         if (failed_bio->bi_rw & REQ_WRITE)
1934                 rw = WRITE;
1935         else
1936                 rw = READ;
1937
1938         ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1939                                                       failrec->last_mirror,
1940                                                       failrec->bio_flags, 0);
1941         return ret;
1942 }
1943
1944 /*
1945  * each time an IO finishes, we do a fast check in the IO failure tree
1946  * to see if we need to process or clean up an io_failure_record
1947  */
1948 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1949 {
1950         u64 private;
1951         u64 private_failure;
1952         struct io_failure_record *failure;
1953         int ret;
1954
1955         private = 0;
1956         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1957                              (u64)-1, 1, EXTENT_DIRTY, 0)) {
1958                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1959                                         start, &private_failure);
1960                 if (ret == 0) {
1961                         failure = (struct io_failure_record *)(unsigned long)
1962                                    private_failure;
1963                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1964                                           failure->start, 0);
1965                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1966                                           failure->start,
1967                                           failure->start + failure->len - 1,
1968                                           EXTENT_DIRTY | EXTENT_LOCKED,
1969                                           GFP_NOFS);
1970                         kfree(failure);
1971                 }
1972         }
1973         return 0;
1974 }
1975
1976 /*
1977  * when reads are done, we need to check csums to verify the data is correct
1978  * if there's a match, we allow the bio to finish.  If not, we go through
1979  * the io_failure_record routines to find good copies
1980  */
1981 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1982                                struct extent_state *state)
1983 {
1984         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1985         struct inode *inode = page->mapping->host;
1986         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1987         char *kaddr;
1988         u64 private = ~(u32)0;
1989         int ret;
1990         struct btrfs_root *root = BTRFS_I(inode)->root;
1991         u32 csum = ~(u32)0;
1992
1993         if (PageChecked(page)) {
1994                 ClearPageChecked(page);
1995                 goto good;
1996         }
1997
1998         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1999                 goto good;
2000
2001         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2002             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2003                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2004                                   GFP_NOFS);
2005                 return 0;
2006         }
2007
2008         if (state && state->start == start) {
2009                 private = state->private;
2010                 ret = 0;
2011         } else {
2012                 ret = get_state_private(io_tree, start, &private);
2013         }
2014         kaddr = kmap_atomic(page, KM_USER0);
2015         if (ret)
2016                 goto zeroit;
2017
2018         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2019         btrfs_csum_final(csum, (char *)&csum);
2020         if (csum != private)
2021                 goto zeroit;
2022
2023         kunmap_atomic(kaddr, KM_USER0);
2024 good:
2025         /* if the io failure tree for this inode is non-empty,
2026          * check to see if we've recovered from a failed IO
2027          */
2028         btrfs_clean_io_failures(inode, start);
2029         return 0;
2030
2031 zeroit:
2032         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2033                        "private %llu\n",
2034                        (unsigned long long)btrfs_ino(page->mapping->host),
2035                        (unsigned long long)start, csum,
2036                        (unsigned long long)private);
2037         memset(kaddr + offset, 1, end - start + 1);
2038         flush_dcache_page(page);
2039         kunmap_atomic(kaddr, KM_USER0);
2040         if (private == 0)
2041                 return 0;
2042         return -EIO;
2043 }
2044
2045 struct delayed_iput {
2046         struct list_head list;
2047         struct inode *inode;
2048 };
2049
2050 void btrfs_add_delayed_iput(struct inode *inode)
2051 {
2052         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2053         struct delayed_iput *delayed;
2054
2055         if (atomic_add_unless(&inode->i_count, -1, 1))
2056                 return;
2057
2058         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2059         delayed->inode = inode;
2060
2061         spin_lock(&fs_info->delayed_iput_lock);
2062         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2063         spin_unlock(&fs_info->delayed_iput_lock);
2064 }
2065
2066 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2067 {
2068         LIST_HEAD(list);
2069         struct btrfs_fs_info *fs_info = root->fs_info;
2070         struct delayed_iput *delayed;
2071         int empty;
2072
2073         spin_lock(&fs_info->delayed_iput_lock);
2074         empty = list_empty(&fs_info->delayed_iputs);
2075         spin_unlock(&fs_info->delayed_iput_lock);
2076         if (empty)
2077                 return;
2078
2079         down_read(&root->fs_info->cleanup_work_sem);
2080         spin_lock(&fs_info->delayed_iput_lock);
2081         list_splice_init(&fs_info->delayed_iputs, &list);
2082         spin_unlock(&fs_info->delayed_iput_lock);
2083
2084         while (!list_empty(&list)) {
2085                 delayed = list_entry(list.next, struct delayed_iput, list);
2086                 list_del(&delayed->list);
2087                 iput(delayed->inode);
2088                 kfree(delayed);
2089         }
2090         up_read(&root->fs_info->cleanup_work_sem);
2091 }
2092
2093 /*
2094  * calculate extra metadata reservation when snapshotting a subvolume
2095  * contains orphan files.
2096  */
2097 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2098                                 struct btrfs_pending_snapshot *pending,
2099                                 u64 *bytes_to_reserve)
2100 {
2101         struct btrfs_root *root;
2102         struct btrfs_block_rsv *block_rsv;
2103         u64 num_bytes;
2104         int index;
2105
2106         root = pending->root;
2107         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2108                 return;
2109
2110         block_rsv = root->orphan_block_rsv;
2111
2112         /* orphan block reservation for the snapshot */
2113         num_bytes = block_rsv->size;
2114
2115         /*
2116          * after the snapshot is created, COWing tree blocks may use more
2117          * space than it frees. So we should make sure there is enough
2118          * reserved space.
2119          */
2120         index = trans->transid & 0x1;
2121         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2122                 num_bytes += block_rsv->size -
2123                              (block_rsv->reserved + block_rsv->freed[index]);
2124         }
2125
2126         *bytes_to_reserve += num_bytes;
2127 }
2128
2129 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2130                                 struct btrfs_pending_snapshot *pending)
2131 {
2132         struct btrfs_root *root = pending->root;
2133         struct btrfs_root *snap = pending->snap;
2134         struct btrfs_block_rsv *block_rsv;
2135         u64 num_bytes;
2136         int index;
2137         int ret;
2138
2139         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2140                 return;
2141
2142         /* refill source subvolume's orphan block reservation */
2143         block_rsv = root->orphan_block_rsv;
2144         index = trans->transid & 0x1;
2145         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2146                 num_bytes = block_rsv->size -
2147                             (block_rsv->reserved + block_rsv->freed[index]);
2148                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2149                                               root->orphan_block_rsv,
2150                                               num_bytes);
2151                 BUG_ON(ret);
2152         }
2153
2154         /* setup orphan block reservation for the snapshot */
2155         block_rsv = btrfs_alloc_block_rsv(snap);
2156         BUG_ON(!block_rsv);
2157
2158         btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2159         snap->orphan_block_rsv = block_rsv;
2160
2161         num_bytes = root->orphan_block_rsv->size;
2162         ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2163                                       block_rsv, num_bytes);
2164         BUG_ON(ret);
2165
2166 #if 0
2167         /* insert orphan item for the snapshot */
2168         WARN_ON(!root->orphan_item_inserted);
2169         ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2170                                        snap->root_key.objectid);
2171         BUG_ON(ret);
2172         snap->orphan_item_inserted = 1;
2173 #endif
2174 }
2175
2176 enum btrfs_orphan_cleanup_state {
2177         ORPHAN_CLEANUP_STARTED  = 1,
2178         ORPHAN_CLEANUP_DONE     = 2,
2179 };
2180
2181 /*
2182  * This is called in transaction commmit time. If there are no orphan
2183  * files in the subvolume, it removes orphan item and frees block_rsv
2184  * structure.
2185  */
2186 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2187                               struct btrfs_root *root)
2188 {
2189         int ret;
2190
2191         if (!list_empty(&root->orphan_list) ||
2192             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2193                 return;
2194
2195         if (root->orphan_item_inserted &&
2196             btrfs_root_refs(&root->root_item) > 0) {
2197                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2198                                             root->root_key.objectid);
2199                 BUG_ON(ret);
2200                 root->orphan_item_inserted = 0;
2201         }
2202
2203         if (root->orphan_block_rsv) {
2204                 WARN_ON(root->orphan_block_rsv->size > 0);
2205                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2206                 root->orphan_block_rsv = NULL;
2207         }
2208 }
2209
2210 /*
2211  * This creates an orphan entry for the given inode in case something goes
2212  * wrong in the middle of an unlink/truncate.
2213  *
2214  * NOTE: caller of this function should reserve 5 units of metadata for
2215  *       this function.
2216  */
2217 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2218 {
2219         struct btrfs_root *root = BTRFS_I(inode)->root;
2220         struct btrfs_block_rsv *block_rsv = NULL;
2221         int reserve = 0;
2222         int insert = 0;
2223         int ret;
2224
2225         if (!root->orphan_block_rsv) {
2226                 block_rsv = btrfs_alloc_block_rsv(root);
2227                 BUG_ON(!block_rsv);
2228         }
2229
2230         spin_lock(&root->orphan_lock);
2231         if (!root->orphan_block_rsv) {
2232                 root->orphan_block_rsv = block_rsv;
2233         } else if (block_rsv) {
2234                 btrfs_free_block_rsv(root, block_rsv);
2235                 block_rsv = NULL;
2236         }
2237
2238         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2239                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2240 #if 0
2241                 /*
2242                  * For proper ENOSPC handling, we should do orphan
2243                  * cleanup when mounting. But this introduces backward
2244                  * compatibility issue.
2245                  */
2246                 if (!xchg(&root->orphan_item_inserted, 1))
2247                         insert = 2;
2248                 else
2249                         insert = 1;
2250 #endif
2251                 insert = 1;
2252         }
2253
2254         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2255                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2256                 reserve = 1;
2257         }
2258         spin_unlock(&root->orphan_lock);
2259
2260         if (block_rsv)
2261                 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2262
2263         /* grab metadata reservation from transaction handle */
2264         if (reserve) {
2265                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2266                 BUG_ON(ret);
2267         }
2268
2269         /* insert an orphan item to track this unlinked/truncated file */
2270         if (insert >= 1) {
2271                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2272                 BUG_ON(ret);
2273         }
2274
2275         /* insert an orphan item to track subvolume contains orphan files */
2276         if (insert >= 2) {
2277                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2278                                                root->root_key.objectid);
2279                 BUG_ON(ret);
2280         }
2281         return 0;
2282 }
2283
2284 /*
2285  * We have done the truncate/delete so we can go ahead and remove the orphan
2286  * item for this particular inode.
2287  */
2288 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2289 {
2290         struct btrfs_root *root = BTRFS_I(inode)->root;
2291         int delete_item = 0;
2292         int release_rsv = 0;
2293         int ret = 0;
2294
2295         spin_lock(&root->orphan_lock);
2296         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2297                 list_del_init(&BTRFS_I(inode)->i_orphan);
2298                 delete_item = 1;
2299         }
2300
2301         if (BTRFS_I(inode)->orphan_meta_reserved) {
2302                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2303                 release_rsv = 1;
2304         }
2305         spin_unlock(&root->orphan_lock);
2306
2307         if (trans && delete_item) {
2308                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2309                 BUG_ON(ret);
2310         }
2311
2312         if (release_rsv)
2313                 btrfs_orphan_release_metadata(inode);
2314
2315         return 0;
2316 }
2317
2318 /*
2319  * this cleans up any orphans that may be left on the list from the last use
2320  * of this root.
2321  */
2322 int btrfs_orphan_cleanup(struct btrfs_root *root)
2323 {
2324         struct btrfs_path *path;
2325         struct extent_buffer *leaf;
2326         struct btrfs_key key, found_key;
2327         struct btrfs_trans_handle *trans;
2328         struct inode *inode;
2329         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2330
2331         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2332                 return 0;
2333
2334         path = btrfs_alloc_path();
2335         if (!path) {
2336                 ret = -ENOMEM;
2337                 goto out;
2338         }
2339         path->reada = -1;
2340
2341         key.objectid = BTRFS_ORPHAN_OBJECTID;
2342         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2343         key.offset = (u64)-1;
2344
2345         while (1) {
2346                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2347                 if (ret < 0)
2348                         goto out;
2349
2350                 /*
2351                  * if ret == 0 means we found what we were searching for, which
2352                  * is weird, but possible, so only screw with path if we didn't
2353                  * find the key and see if we have stuff that matches
2354                  */
2355                 if (ret > 0) {
2356                         ret = 0;
2357                         if (path->slots[0] == 0)
2358                                 break;
2359                         path->slots[0]--;
2360                 }
2361
2362                 /* pull out the item */
2363                 leaf = path->nodes[0];
2364                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2365
2366                 /* make sure the item matches what we want */
2367                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2368                         break;
2369                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2370                         break;
2371
2372                 /* release the path since we're done with it */
2373                 btrfs_release_path(path);
2374
2375                 /*
2376                  * this is where we are basically btrfs_lookup, without the
2377                  * crossing root thing.  we store the inode number in the
2378                  * offset of the orphan item.
2379                  */
2380                 found_key.objectid = found_key.offset;
2381                 found_key.type = BTRFS_INODE_ITEM_KEY;
2382                 found_key.offset = 0;
2383                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2384                 if (IS_ERR(inode)) {
2385                         ret = PTR_ERR(inode);
2386                         goto out;
2387                 }
2388
2389                 /*
2390                  * add this inode to the orphan list so btrfs_orphan_del does
2391                  * the proper thing when we hit it
2392                  */
2393                 spin_lock(&root->orphan_lock);
2394                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2395                 spin_unlock(&root->orphan_lock);
2396
2397                 /*
2398                  * if this is a bad inode, means we actually succeeded in
2399                  * removing the inode, but not the orphan record, which means
2400                  * we need to manually delete the orphan since iput will just
2401                  * do a destroy_inode
2402                  */
2403                 if (is_bad_inode(inode)) {
2404                         trans = btrfs_start_transaction(root, 0);
2405                         if (IS_ERR(trans)) {
2406                                 ret = PTR_ERR(trans);
2407                                 goto out;
2408                         }
2409                         btrfs_orphan_del(trans, inode);
2410                         btrfs_end_transaction(trans, root);
2411                         iput(inode);
2412                         continue;
2413                 }
2414
2415                 /* if we have links, this was a truncate, lets do that */
2416                 if (inode->i_nlink) {
2417                         if (!S_ISREG(inode->i_mode)) {
2418                                 WARN_ON(1);
2419                                 iput(inode);
2420                                 continue;
2421                         }
2422                         nr_truncate++;
2423                         ret = btrfs_truncate(inode);
2424                 } else {
2425                         nr_unlink++;
2426                 }
2427
2428                 /* this will do delete_inode and everything for us */
2429                 iput(inode);
2430                 if (ret)
2431                         goto out;
2432         }
2433         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2434
2435         if (root->orphan_block_rsv)
2436                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2437                                         (u64)-1);
2438
2439         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2440                 trans = btrfs_join_transaction(root);
2441                 if (!IS_ERR(trans))
2442                         btrfs_end_transaction(trans, root);
2443         }
2444
2445         if (nr_unlink)
2446                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2447         if (nr_truncate)
2448                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2449
2450 out:
2451         if (ret)
2452                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2453         btrfs_free_path(path);
2454         return ret;
2455 }
2456
2457 /*
2458  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2459  * don't find any xattrs, we know there can't be any acls.
2460  *
2461  * slot is the slot the inode is in, objectid is the objectid of the inode
2462  */
2463 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2464                                           int slot, u64 objectid)
2465 {
2466         u32 nritems = btrfs_header_nritems(leaf);
2467         struct btrfs_key found_key;
2468         int scanned = 0;
2469
2470         slot++;
2471         while (slot < nritems) {
2472                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2473
2474                 /* we found a different objectid, there must not be acls */
2475                 if (found_key.objectid != objectid)
2476                         return 0;
2477
2478                 /* we found an xattr, assume we've got an acl */
2479                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2480                         return 1;
2481
2482                 /*
2483                  * we found a key greater than an xattr key, there can't
2484                  * be any acls later on
2485                  */
2486                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2487                         return 0;
2488
2489                 slot++;
2490                 scanned++;
2491
2492                 /*
2493                  * it goes inode, inode backrefs, xattrs, extents,
2494                  * so if there are a ton of hard links to an inode there can
2495                  * be a lot of backrefs.  Don't waste time searching too hard,
2496                  * this is just an optimization
2497                  */
2498                 if (scanned >= 8)
2499                         break;
2500         }
2501         /* we hit the end of the leaf before we found an xattr or
2502          * something larger than an xattr.  We have to assume the inode
2503          * has acls
2504          */
2505         return 1;
2506 }
2507
2508 /*
2509  * read an inode from the btree into the in-memory inode
2510  */
2511 static void btrfs_read_locked_inode(struct inode *inode)
2512 {
2513         struct btrfs_path *path;
2514         struct extent_buffer *leaf;
2515         struct btrfs_inode_item *inode_item;
2516         struct btrfs_timespec *tspec;
2517         struct btrfs_root *root = BTRFS_I(inode)->root;
2518         struct btrfs_key location;
2519         int maybe_acls;
2520         u32 rdev;
2521         int ret;
2522         bool filled = false;
2523
2524         ret = btrfs_fill_inode(inode, &rdev);
2525         if (!ret)
2526                 filled = true;
2527
2528         path = btrfs_alloc_path();
2529         BUG_ON(!path);
2530         path->leave_spinning = 1;
2531         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2532
2533         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2534         if (ret)
2535                 goto make_bad;
2536
2537         leaf = path->nodes[0];
2538
2539         if (filled)
2540                 goto cache_acl;
2541
2542         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2543                                     struct btrfs_inode_item);
2544         if (!leaf->map_token)
2545                 map_private_extent_buffer(leaf, (unsigned long)inode_item,
2546                                           sizeof(struct btrfs_inode_item),
2547                                           &leaf->map_token, &leaf->kaddr,
2548                                           &leaf->map_start, &leaf->map_len,
2549                                           KM_USER1);
2550
2551         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2552         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2553         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2554         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2555         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2556
2557         tspec = btrfs_inode_atime(inode_item);
2558         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2559         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2560
2561         tspec = btrfs_inode_mtime(inode_item);
2562         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2563         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2564
2565         tspec = btrfs_inode_ctime(inode_item);
2566         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2567         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2568
2569         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2570         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2571         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2572         inode->i_generation = BTRFS_I(inode)->generation;
2573         inode->i_rdev = 0;
2574         rdev = btrfs_inode_rdev(leaf, inode_item);
2575
2576         BTRFS_I(inode)->index_cnt = (u64)-1;
2577         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2578 cache_acl:
2579         /*
2580          * try to precache a NULL acl entry for files that don't have
2581          * any xattrs or acls
2582          */
2583         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2584                                            btrfs_ino(inode));
2585         if (!maybe_acls)
2586                 cache_no_acl(inode);
2587
2588         if (leaf->map_token) {
2589                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2590                 leaf->map_token = NULL;
2591         }
2592
2593         btrfs_free_path(path);
2594
2595         switch (inode->i_mode & S_IFMT) {
2596         case S_IFREG:
2597                 inode->i_mapping->a_ops = &btrfs_aops;
2598                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2599                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2600                 inode->i_fop = &btrfs_file_operations;
2601                 inode->i_op = &btrfs_file_inode_operations;
2602                 break;
2603         case S_IFDIR:
2604                 inode->i_fop = &btrfs_dir_file_operations;
2605                 if (root == root->fs_info->tree_root)
2606                         inode->i_op = &btrfs_dir_ro_inode_operations;
2607                 else
2608                         inode->i_op = &btrfs_dir_inode_operations;
2609                 break;
2610         case S_IFLNK:
2611                 inode->i_op = &btrfs_symlink_inode_operations;
2612                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2613                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2614                 break;
2615         default:
2616                 inode->i_op = &btrfs_special_inode_operations;
2617                 init_special_inode(inode, inode->i_mode, rdev);
2618                 break;
2619         }
2620
2621         btrfs_update_iflags(inode);
2622         return;
2623
2624 make_bad:
2625         btrfs_free_path(path);
2626         make_bad_inode(inode);
2627 }
2628
2629 /*
2630  * given a leaf and an inode, copy the inode fields into the leaf
2631  */
2632 static void fill_inode_item(struct btrfs_trans_handle *trans,
2633                             struct extent_buffer *leaf,
2634                             struct btrfs_inode_item *item,
2635                             struct inode *inode)
2636 {
2637         if (!leaf->map_token)
2638                 map_private_extent_buffer(leaf, (unsigned long)item,
2639                                           sizeof(struct btrfs_inode_item),
2640                                           &leaf->map_token, &leaf->kaddr,
2641                                           &leaf->map_start, &leaf->map_len,
2642                                           KM_USER1);
2643
2644         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2645         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2646         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2647         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2648         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2649
2650         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2651                                inode->i_atime.tv_sec);
2652         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2653                                 inode->i_atime.tv_nsec);
2654
2655         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2656                                inode->i_mtime.tv_sec);
2657         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2658                                 inode->i_mtime.tv_nsec);
2659
2660         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2661                                inode->i_ctime.tv_sec);
2662         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2663                                 inode->i_ctime.tv_nsec);
2664
2665         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2666         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2667         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2668         btrfs_set_inode_transid(leaf, item, trans->transid);
2669         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2670         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2671         btrfs_set_inode_block_group(leaf, item, 0);
2672
2673         if (leaf->map_token) {
2674                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2675                 leaf->map_token = NULL;
2676         }
2677 }
2678
2679 /*
2680  * copy everything in the in-memory inode into the btree.
2681  */
2682 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2683                                 struct btrfs_root *root, struct inode *inode)
2684 {
2685         struct btrfs_inode_item *inode_item;
2686         struct btrfs_path *path;
2687         struct extent_buffer *leaf;
2688         int ret;
2689
2690         /*
2691          * If the inode is a free space inode, we can deadlock during commit
2692          * if we put it into the delayed code.
2693          *
2694          * The data relocation inode should also be directly updated
2695          * without delay
2696          */
2697         if (!is_free_space_inode(root, inode)
2698             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2699                 ret = btrfs_delayed_update_inode(trans, root, inode);
2700                 if (!ret)
2701                         btrfs_set_inode_last_trans(trans, inode);
2702                 return ret;
2703         }
2704
2705         path = btrfs_alloc_path();
2706         if (!path)
2707                 return -ENOMEM;
2708
2709         path->leave_spinning = 1;
2710         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2711                                  1);
2712         if (ret) {
2713                 if (ret > 0)
2714                         ret = -ENOENT;
2715                 goto failed;
2716         }
2717
2718         btrfs_unlock_up_safe(path, 1);
2719         leaf = path->nodes[0];
2720         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2721                                     struct btrfs_inode_item);
2722
2723         fill_inode_item(trans, leaf, inode_item, inode);
2724         btrfs_mark_buffer_dirty(leaf);
2725         btrfs_set_inode_last_trans(trans, inode);
2726         ret = 0;
2727 failed:
2728         btrfs_free_path(path);
2729         return ret;
2730 }
2731
2732 /*
2733  * unlink helper that gets used here in inode.c and in the tree logging
2734  * recovery code.  It remove a link in a directory with a given name, and
2735  * also drops the back refs in the inode to the directory
2736  */
2737 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2738                                 struct btrfs_root *root,
2739                                 struct inode *dir, struct inode *inode,
2740                                 const char *name, int name_len)
2741 {
2742         struct btrfs_path *path;
2743         int ret = 0;
2744         struct extent_buffer *leaf;
2745         struct btrfs_dir_item *di;
2746         struct btrfs_key key;
2747         u64 index;
2748         u64 ino = btrfs_ino(inode);
2749         u64 dir_ino = btrfs_ino(dir);
2750
2751         path = btrfs_alloc_path();
2752         if (!path) {
2753                 ret = -ENOMEM;
2754                 goto out;
2755         }
2756
2757         path->leave_spinning = 1;
2758         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2759                                     name, name_len, -1);
2760         if (IS_ERR(di)) {
2761                 ret = PTR_ERR(di);
2762                 goto err;
2763         }
2764         if (!di) {
2765                 ret = -ENOENT;
2766                 goto err;
2767         }
2768         leaf = path->nodes[0];
2769         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2770         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2771         if (ret)
2772                 goto err;
2773         btrfs_release_path(path);
2774
2775         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2776                                   dir_ino, &index);
2777         if (ret) {
2778                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2779                        "inode %llu parent %llu\n", name_len, name,
2780                        (unsigned long long)ino, (unsigned long long)dir_ino);
2781                 goto err;
2782         }
2783
2784         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2785         if (ret)
2786                 goto err;
2787
2788         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2789                                          inode, dir_ino);
2790         BUG_ON(ret != 0 && ret != -ENOENT);
2791
2792         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2793                                            dir, index);
2794         if (ret == -ENOENT)
2795                 ret = 0;
2796 err:
2797         btrfs_free_path(path);
2798         if (ret)
2799                 goto out;
2800
2801         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2802         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2803         btrfs_update_inode(trans, root, dir);
2804 out:
2805         return ret;
2806 }
2807
2808 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2809                        struct btrfs_root *root,
2810                        struct inode *dir, struct inode *inode,
2811                        const char *name, int name_len)
2812 {
2813         int ret;
2814         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2815         if (!ret) {
2816                 btrfs_drop_nlink(inode);
2817                 ret = btrfs_update_inode(trans, root, inode);
2818         }
2819         return ret;
2820 }
2821
2822
2823 /* helper to check if there is any shared block in the path */
2824 static int check_path_shared(struct btrfs_root *root,
2825                              struct btrfs_path *path)
2826 {
2827         struct extent_buffer *eb;
2828         int level;
2829         u64 refs = 1;
2830
2831         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2832                 int ret;
2833
2834                 if (!path->nodes[level])
2835                         break;
2836                 eb = path->nodes[level];
2837                 if (!btrfs_block_can_be_shared(root, eb))
2838                         continue;
2839                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2840                                                &refs, NULL);
2841                 if (refs > 1)
2842                         return 1;
2843         }
2844         return 0;
2845 }
2846
2847 /*
2848  * helper to start transaction for unlink and rmdir.
2849  *
2850  * unlink and rmdir are special in btrfs, they do not always free space.
2851  * so in enospc case, we should make sure they will free space before
2852  * allowing them to use the global metadata reservation.
2853  */
2854 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2855                                                        struct dentry *dentry)
2856 {
2857         struct btrfs_trans_handle *trans;
2858         struct btrfs_root *root = BTRFS_I(dir)->root;
2859         struct btrfs_path *path;
2860         struct btrfs_inode_ref *ref;
2861         struct btrfs_dir_item *di;
2862         struct inode *inode = dentry->d_inode;
2863         u64 index;
2864         int check_link = 1;
2865         int err = -ENOSPC;
2866         int ret;
2867         u64 ino = btrfs_ino(inode);
2868         u64 dir_ino = btrfs_ino(dir);
2869
2870         trans = btrfs_start_transaction(root, 10);
2871         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2872                 return trans;
2873
2874         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2875                 return ERR_PTR(-ENOSPC);
2876
2877         /* check if there is someone else holds reference */
2878         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2879                 return ERR_PTR(-ENOSPC);
2880
2881         if (atomic_read(&inode->i_count) > 2)
2882                 return ERR_PTR(-ENOSPC);
2883
2884         if (xchg(&root->fs_info->enospc_unlink, 1))
2885                 return ERR_PTR(-ENOSPC);
2886
2887         path = btrfs_alloc_path();
2888         if (!path) {
2889                 root->fs_info->enospc_unlink = 0;
2890                 return ERR_PTR(-ENOMEM);
2891         }
2892
2893         trans = btrfs_start_transaction(root, 0);
2894         if (IS_ERR(trans)) {
2895                 btrfs_free_path(path);
2896                 root->fs_info->enospc_unlink = 0;
2897                 return trans;
2898         }
2899
2900         path->skip_locking = 1;
2901         path->search_commit_root = 1;
2902
2903         ret = btrfs_lookup_inode(trans, root, path,
2904                                 &BTRFS_I(dir)->location, 0);
2905         if (ret < 0) {
2906                 err = ret;
2907                 goto out;
2908         }
2909         if (ret == 0) {
2910                 if (check_path_shared(root, path))
2911                         goto out;
2912         } else {
2913                 check_link = 0;
2914         }
2915         btrfs_release_path(path);
2916
2917         ret = btrfs_lookup_inode(trans, root, path,
2918                                 &BTRFS_I(inode)->location, 0);
2919         if (ret < 0) {
2920                 err = ret;
2921                 goto out;
2922         }
2923         if (ret == 0) {
2924                 if (check_path_shared(root, path))
2925                         goto out;
2926         } else {
2927                 check_link = 0;
2928         }
2929         btrfs_release_path(path);
2930
2931         if (ret == 0 && S_ISREG(inode->i_mode)) {
2932                 ret = btrfs_lookup_file_extent(trans, root, path,
2933                                                ino, (u64)-1, 0);
2934                 if (ret < 0) {
2935                         err = ret;
2936                         goto out;
2937                 }
2938                 BUG_ON(ret == 0);
2939                 if (check_path_shared(root, path))
2940                         goto out;
2941                 btrfs_release_path(path);
2942         }
2943
2944         if (!check_link) {
2945                 err = 0;
2946                 goto out;
2947         }
2948
2949         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2950                                 dentry->d_name.name, dentry->d_name.len, 0);
2951         if (IS_ERR(di)) {
2952                 err = PTR_ERR(di);
2953                 goto out;
2954         }
2955         if (di) {
2956                 if (check_path_shared(root, path))
2957                         goto out;
2958         } else {
2959                 err = 0;
2960                 goto out;
2961         }
2962         btrfs_release_path(path);
2963
2964         ref = btrfs_lookup_inode_ref(trans, root, path,
2965                                 dentry->d_name.name, dentry->d_name.len,
2966                                 ino, dir_ino, 0);
2967         if (IS_ERR(ref)) {
2968                 err = PTR_ERR(ref);
2969                 goto out;
2970         }
2971         BUG_ON(!ref);
2972         if (check_path_shared(root, path))
2973                 goto out;
2974         index = btrfs_inode_ref_index(path->nodes[0], ref);
2975         btrfs_release_path(path);
2976
2977         /*
2978          * This is a commit root search, if we can lookup inode item and other
2979          * relative items in the commit root, it means the transaction of
2980          * dir/file creation has been committed, and the dir index item that we
2981          * delay to insert has also been inserted into the commit root. So
2982          * we needn't worry about the delayed insertion of the dir index item
2983          * here.
2984          */
2985         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2986                                 dentry->d_name.name, dentry->d_name.len, 0);
2987         if (IS_ERR(di)) {
2988                 err = PTR_ERR(di);
2989                 goto out;
2990         }
2991         BUG_ON(ret == -ENOENT);
2992         if (check_path_shared(root, path))
2993                 goto out;
2994
2995         err = 0;
2996 out:
2997         btrfs_free_path(path);
2998         if (err) {
2999                 btrfs_end_transaction(trans, root);
3000                 root->fs_info->enospc_unlink = 0;
3001                 return ERR_PTR(err);
3002         }
3003
3004         trans->block_rsv = &root->fs_info->global_block_rsv;
3005         return trans;
3006 }
3007
3008 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3009                                struct btrfs_root *root)
3010 {
3011         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
3012                 BUG_ON(!root->fs_info->enospc_unlink);
3013                 root->fs_info->enospc_unlink = 0;
3014         }
3015         btrfs_end_transaction_throttle(trans, root);
3016 }
3017
3018 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3019 {
3020         struct btrfs_root *root = BTRFS_I(dir)->root;
3021         struct btrfs_trans_handle *trans;
3022         struct inode *inode = dentry->d_inode;
3023         int ret;
3024         unsigned long nr = 0;
3025
3026         trans = __unlink_start_trans(dir, dentry);
3027         if (IS_ERR(trans))
3028                 return PTR_ERR(trans);
3029
3030         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3031
3032         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3033                                  dentry->d_name.name, dentry->d_name.len);
3034         BUG_ON(ret);
3035
3036         if (inode->i_nlink == 0) {
3037                 ret = btrfs_orphan_add(trans, inode);
3038                 BUG_ON(ret);
3039         }
3040
3041         nr = trans->blocks_used;
3042         __unlink_end_trans(trans, root);
3043         btrfs_btree_balance_dirty(root, nr);
3044         return ret;
3045 }
3046
3047 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3048                         struct btrfs_root *root,
3049                         struct inode *dir, u64 objectid,
3050                         const char *name, int name_len)
3051 {
3052         struct btrfs_path *path;
3053         struct extent_buffer *leaf;
3054         struct btrfs_dir_item *di;
3055         struct btrfs_key key;
3056         u64 index;
3057         int ret;
3058         u64 dir_ino = btrfs_ino(dir);
3059
3060         path = btrfs_alloc_path();
3061         if (!path)
3062                 return -ENOMEM;
3063
3064         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3065                                    name, name_len, -1);
3066         BUG_ON(IS_ERR_OR_NULL(di));
3067
3068         leaf = path->nodes[0];
3069         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3070         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3071         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3072         BUG_ON(ret);
3073         btrfs_release_path(path);
3074
3075         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3076                                  objectid, root->root_key.objectid,
3077                                  dir_ino, &index, name, name_len);
3078         if (ret < 0) {
3079                 BUG_ON(ret != -ENOENT);
3080                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3081                                                  name, name_len);
3082                 BUG_ON(IS_ERR_OR_NULL(di));
3083
3084                 leaf = path->nodes[0];
3085                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3086                 btrfs_release_path(path);
3087                 index = key.offset;
3088         }
3089         btrfs_release_path(path);
3090
3091         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3092         BUG_ON(ret);
3093
3094         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3095         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3096         ret = btrfs_update_inode(trans, root, dir);
3097         BUG_ON(ret);
3098
3099         btrfs_free_path(path);
3100         return 0;
3101 }
3102
3103 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3104 {
3105         struct inode *inode = dentry->d_inode;
3106         int err = 0;
3107         struct btrfs_root *root = BTRFS_I(dir)->root;
3108         struct btrfs_trans_handle *trans;
3109         unsigned long nr = 0;
3110
3111         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3112             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3113                 return -ENOTEMPTY;
3114
3115         trans = __unlink_start_trans(dir, dentry);
3116         if (IS_ERR(trans))
3117                 return PTR_ERR(trans);
3118
3119         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3120                 err = btrfs_unlink_subvol(trans, root, dir,
3121                                           BTRFS_I(inode)->location.objectid,
3122                                           dentry->d_name.name,
3123                                           dentry->d_name.len);
3124                 goto out;
3125         }
3126
3127         err = btrfs_orphan_add(trans, inode);
3128         if (err)
3129                 goto out;
3130
3131         /* now the directory is empty */
3132         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3133                                  dentry->d_name.name, dentry->d_name.len);
3134         if (!err)
3135                 btrfs_i_size_write(inode, 0);
3136 out:
3137         nr = trans->blocks_used;
3138         __unlink_end_trans(trans, root);
3139         btrfs_btree_balance_dirty(root, nr);
3140
3141         return err;
3142 }
3143
3144 /*
3145  * this can truncate away extent items, csum items and directory items.
3146  * It starts at a high offset and removes keys until it can't find
3147  * any higher than new_size
3148  *
3149  * csum items that cross the new i_size are truncated to the new size
3150  * as well.
3151  *
3152  * min_type is the minimum key type to truncate down to.  If set to 0, this
3153  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3154  */
3155 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3156                                struct btrfs_root *root,
3157                                struct inode *inode,
3158                                u64 new_size, u32 min_type)
3159 {
3160         struct btrfs_path *path;
3161         struct extent_buffer *leaf;
3162         struct btrfs_file_extent_item *fi;
3163         struct btrfs_key key;
3164         struct btrfs_key found_key;
3165         u64 extent_start = 0;
3166         u64 extent_num_bytes = 0;
3167         u64 extent_offset = 0;
3168         u64 item_end = 0;
3169         u64 mask = root->sectorsize - 1;
3170         u32 found_type = (u8)-1;
3171         int found_extent;
3172         int del_item;
3173         int pending_del_nr = 0;
3174         int pending_del_slot = 0;
3175         int extent_type = -1;
3176         int encoding;
3177         int ret;
3178         int err = 0;
3179         u64 ino = btrfs_ino(inode);
3180
3181         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3182
3183         if (root->ref_cows || root == root->fs_info->tree_root)
3184                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3185
3186         /*
3187          * This function is also used to drop the items in the log tree before
3188          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3189          * it is used to drop the loged items. So we shouldn't kill the delayed
3190          * items.
3191          */
3192         if (min_type == 0 && root == BTRFS_I(inode)->root)
3193                 btrfs_kill_delayed_inode_items(inode);
3194
3195         path = btrfs_alloc_path();
3196         BUG_ON(!path);
3197         path->reada = -1;
3198
3199         key.objectid = ino;
3200         key.offset = (u64)-1;
3201         key.type = (u8)-1;
3202
3203 search_again:
3204         path->leave_spinning = 1;
3205         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3206         if (ret < 0) {
3207                 err = ret;
3208                 goto out;
3209         }
3210
3211         if (ret > 0) {
3212                 /* there are no items in the tree for us to truncate, we're
3213                  * done
3214                  */
3215                 if (path->slots[0] == 0)
3216                         goto out;
3217                 path->slots[0]--;
3218         }
3219
3220         while (1) {
3221                 fi = NULL;
3222                 leaf = path->nodes[0];
3223                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3224                 found_type = btrfs_key_type(&found_key);
3225                 encoding = 0;
3226
3227                 if (found_key.objectid != ino)
3228                         break;
3229
3230                 if (found_type < min_type)
3231                         break;
3232
3233                 item_end = found_key.offset;
3234                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3235                         fi = btrfs_item_ptr(leaf, path->slots[0],
3236                                             struct btrfs_file_extent_item);
3237                         extent_type = btrfs_file_extent_type(leaf, fi);
3238                         encoding = btrfs_file_extent_compression(leaf, fi);
3239                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3240                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3241
3242                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3243                                 item_end +=
3244                                     btrfs_file_extent_num_bytes(leaf, fi);
3245                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3246                                 item_end += btrfs_file_extent_inline_len(leaf,
3247                                                                          fi);
3248                         }
3249                         item_end--;
3250                 }
3251                 if (found_type > min_type) {
3252                         del_item = 1;
3253                 } else {
3254                         if (item_end < new_size)
3255                                 break;
3256                         if (found_key.offset >= new_size)
3257                                 del_item = 1;
3258                         else
3259                                 del_item = 0;
3260                 }
3261                 found_extent = 0;
3262                 /* FIXME, shrink the extent if the ref count is only 1 */
3263                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3264                         goto delete;
3265
3266                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3267                         u64 num_dec;
3268                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3269                         if (!del_item && !encoding) {
3270                                 u64 orig_num_bytes =
3271                                         btrfs_file_extent_num_bytes(leaf, fi);
3272                                 extent_num_bytes = new_size -
3273                                         found_key.offset + root->sectorsize - 1;
3274                                 extent_num_bytes = extent_num_bytes &
3275                                         ~((u64)root->sectorsize - 1);
3276                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3277                                                          extent_num_bytes);
3278                                 num_dec = (orig_num_bytes -
3279                                            extent_num_bytes);
3280                                 if (root->ref_cows && extent_start != 0)
3281                                         inode_sub_bytes(inode, num_dec);
3282                                 btrfs_mark_buffer_dirty(leaf);
3283                         } else {
3284                                 extent_num_bytes =
3285                                         btrfs_file_extent_disk_num_bytes(leaf,
3286                                                                          fi);
3287                                 extent_offset = found_key.offset -
3288                                         btrfs_file_extent_offset(leaf, fi);
3289
3290                                 /* FIXME blocksize != 4096 */
3291                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3292                                 if (extent_start != 0) {
3293                                         found_extent = 1;
3294                                         if (root->ref_cows)
3295                                                 inode_sub_bytes(inode, num_dec);
3296                                 }
3297                         }
3298                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3299                         /*
3300                          * we can't truncate inline items that have had
3301                          * special encodings
3302                          */
3303                         if (!del_item &&
3304                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3305                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3306                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3307                                 u32 size = new_size - found_key.offset;
3308
3309                                 if (root->ref_cows) {
3310                                         inode_sub_bytes(inode, item_end + 1 -
3311                                                         new_size);
3312                                 }
3313                                 size =
3314                                     btrfs_file_extent_calc_inline_size(size);
3315                                 ret = btrfs_truncate_item(trans, root, path,
3316                                                           size, 1);
3317                         } else if (root->ref_cows) {
3318                                 inode_sub_bytes(inode, item_end + 1 -
3319                                                 found_key.offset);
3320                         }
3321                 }
3322 delete:
3323                 if (del_item) {
3324                         if (!pending_del_nr) {
3325                                 /* no pending yet, add ourselves */
3326                                 pending_del_slot = path->slots[0];
3327                                 pending_del_nr = 1;
3328                         } else if (pending_del_nr &&
3329                                    path->slots[0] + 1 == pending_del_slot) {
3330                                 /* hop on the pending chunk */
3331                                 pending_del_nr++;
3332                                 pending_del_slot = path->slots[0];
3333                         } else {
3334                                 BUG();
3335                         }
3336                 } else {
3337                         break;
3338                 }
3339                 if (found_extent && (root->ref_cows ||
3340                                      root == root->fs_info->tree_root)) {
3341                         btrfs_set_path_blocking(path);
3342                         ret = btrfs_free_extent(trans, root, extent_start,
3343                                                 extent_num_bytes, 0,
3344                                                 btrfs_header_owner(leaf),
3345                                                 ino, extent_offset);
3346                         BUG_ON(ret);
3347                 }
3348
3349                 if (found_type == BTRFS_INODE_ITEM_KEY)
3350                         break;
3351
3352                 if (path->slots[0] == 0 ||
3353                     path->slots[0] != pending_del_slot) {
3354                         if (root->ref_cows &&
3355                             BTRFS_I(inode)->location.objectid !=
3356                                                 BTRFS_FREE_INO_OBJECTID) {
3357                                 err = -EAGAIN;
3358                                 goto out;
3359                         }
3360                         if (pending_del_nr) {
3361                                 ret = btrfs_del_items(trans, root, path,
3362                                                 pending_del_slot,
3363                                                 pending_del_nr);
3364                                 BUG_ON(ret);
3365                                 pending_del_nr = 0;
3366                         }
3367                         btrfs_release_path(path);
3368                         goto search_again;
3369                 } else {
3370                         path->slots[0]--;
3371                 }
3372         }
3373 out:
3374         if (pending_del_nr) {
3375                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3376                                       pending_del_nr);
3377                 BUG_ON(ret);
3378         }
3379         btrfs_free_path(path);
3380         return err;
3381 }
3382
3383 /*
3384  * taken from block_truncate_page, but does cow as it zeros out
3385  * any bytes left in the last page in the file.
3386  */
3387 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3388 {
3389         struct inode *inode = mapping->host;
3390         struct btrfs_root *root = BTRFS_I(inode)->root;
3391         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3392         struct btrfs_ordered_extent *ordered;
3393         struct extent_state *cached_state = NULL;
3394         char *kaddr;
3395         u32 blocksize = root->sectorsize;
3396         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3397         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3398         struct page *page;
3399         int ret = 0;
3400         u64 page_start;
3401         u64 page_end;
3402
3403         if ((offset & (blocksize - 1)) == 0)
3404                 goto out;
3405         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3406         if (ret)
3407                 goto out;
3408
3409         ret = -ENOMEM;
3410 again:
3411         page = find_or_create_page(mapping, index, GFP_NOFS);
3412         if (!page) {
3413                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3414                 goto out;
3415         }
3416
3417         page_start = page_offset(page);
3418         page_end = page_start + PAGE_CACHE_SIZE - 1;
3419
3420         if (!PageUptodate(page)) {
3421                 ret = btrfs_readpage(NULL, page);
3422                 lock_page(page);
3423                 if (page->mapping != mapping) {
3424                         unlock_page(page);
3425                         page_cache_release(page);
3426                         goto again;
3427                 }
3428                 if (!PageUptodate(page)) {
3429                         ret = -EIO;
3430                         goto out_unlock;
3431                 }
3432         }
3433         wait_on_page_writeback(page);
3434
3435         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3436                          GFP_NOFS);
3437         set_page_extent_mapped(page);
3438
3439         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3440         if (ordered) {
3441                 unlock_extent_cached(io_tree, page_start, page_end,
3442                                      &cached_state, GFP_NOFS);
3443                 unlock_page(page);
3444                 page_cache_release(page);
3445                 btrfs_start_ordered_extent(inode, ordered, 1);
3446                 btrfs_put_ordered_extent(ordered);
3447                 goto again;
3448         }
3449
3450         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3451                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3452                           0, 0, &cached_state, GFP_NOFS);
3453
3454         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3455                                         &cached_state);
3456         if (ret) {
3457                 unlock_extent_cached(io_tree, page_start, page_end,
3458                                      &cached_state, GFP_NOFS);
3459                 goto out_unlock;
3460         }
3461
3462         ret = 0;
3463         if (offset != PAGE_CACHE_SIZE) {
3464                 kaddr = kmap(page);
3465                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3466                 flush_dcache_page(page);
3467                 kunmap(page);
3468         }
3469         ClearPageChecked(page);
3470         set_page_dirty(page);
3471         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3472                              GFP_NOFS);
3473
3474 out_unlock:
3475         if (ret)
3476                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3477         unlock_page(page);
3478         page_cache_release(page);
3479 out:
3480         return ret;
3481 }
3482
3483 /*
3484  * This function puts in dummy file extents for the area we're creating a hole
3485  * for.  So if we are truncating this file to a larger size we need to insert
3486  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3487  * the range between oldsize and size
3488  */
3489 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3490 {
3491         struct btrfs_trans_handle *trans;
3492         struct btrfs_root *root = BTRFS_I(inode)->root;
3493         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3494         struct extent_map *em = NULL;
3495         struct extent_state *cached_state = NULL;
3496         u64 mask = root->sectorsize - 1;
3497         u64 hole_start = (oldsize + mask) & ~mask;
3498         u64 block_end = (size + mask) & ~mask;
3499         u64 last_byte;
3500         u64 cur_offset;
3501         u64 hole_size;
3502         int err = 0;
3503
3504         if (size <= hole_start)
3505                 return 0;
3506
3507         while (1) {
3508                 struct btrfs_ordered_extent *ordered;
3509                 btrfs_wait_ordered_range(inode, hole_start,
3510                                          block_end - hole_start);
3511                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3512                                  &cached_state, GFP_NOFS);
3513                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3514                 if (!ordered)
3515                         break;
3516                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3517                                      &cached_state, GFP_NOFS);
3518                 btrfs_put_ordered_extent(ordered);
3519         }
3520
3521         cur_offset = hole_start;
3522         while (1) {
3523                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3524                                 block_end - cur_offset, 0);
3525                 BUG_ON(IS_ERR_OR_NULL(em));
3526                 last_byte = min(extent_map_end(em), block_end);
3527                 last_byte = (last_byte + mask) & ~mask;
3528                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3529                         u64 hint_byte = 0;
3530                         hole_size = last_byte - cur_offset;
3531
3532                         trans = btrfs_start_transaction(root, 2);
3533                         if (IS_ERR(trans)) {
3534                                 err = PTR_ERR(trans);
3535                                 break;
3536                         }
3537
3538                         err = btrfs_drop_extents(trans, inode, cur_offset,
3539                                                  cur_offset + hole_size,
3540                                                  &hint_byte, 1);
3541                         if (err)
3542                                 break;
3543
3544                         err = btrfs_insert_file_extent(trans, root,
3545                                         btrfs_ino(inode), cur_offset, 0,
3546                                         0, hole_size, 0, hole_size,
3547                                         0, 0, 0);
3548                         if (err)
3549                                 break;
3550
3551                         btrfs_drop_extent_cache(inode, hole_start,
3552                                         last_byte - 1, 0);
3553
3554                         btrfs_end_transaction(trans, root);
3555                 }
3556                 free_extent_map(em);
3557                 em = NULL;
3558                 cur_offset = last_byte;
3559                 if (cur_offset >= block_end)
3560                         break;
3561         }
3562
3563         free_extent_map(em);
3564         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3565                              GFP_NOFS);
3566         return err;
3567 }
3568
3569 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3570 {
3571         loff_t oldsize = i_size_read(inode);
3572         int ret;
3573
3574         if (newsize == oldsize)
3575                 return 0;
3576
3577         if (newsize > oldsize) {
3578                 i_size_write(inode, newsize);
3579                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3580                 truncate_pagecache(inode, oldsize, newsize);
3581                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3582                 if (ret) {
3583                         btrfs_setsize(inode, oldsize);
3584                         return ret;
3585                 }
3586
3587                 mark_inode_dirty(inode);
3588         } else {
3589
3590                 /*
3591                  * We're truncating a file that used to have good data down to
3592                  * zero. Make sure it gets into the ordered flush list so that
3593                  * any new writes get down to disk quickly.
3594                  */
3595                 if (newsize == 0)
3596                         BTRFS_I(inode)->ordered_data_close = 1;
3597
3598                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3599                 truncate_setsize(inode, newsize);
3600                 ret = btrfs_truncate(inode);
3601         }
3602
3603         return ret;
3604 }
3605
3606 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3607 {
3608         struct inode *inode = dentry->d_inode;
3609         struct btrfs_root *root = BTRFS_I(inode)->root;
3610         int err;
3611
3612         if (btrfs_root_readonly(root))
3613                 return -EROFS;
3614
3615         err = inode_change_ok(inode, attr);
3616         if (err)
3617                 return err;
3618
3619         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3620                 err = btrfs_setsize(inode, attr->ia_size);
3621                 if (err)
3622                         return err;
3623         }
3624
3625         if (attr->ia_valid) {
3626                 setattr_copy(inode, attr);
3627                 mark_inode_dirty(inode);
3628
3629                 if (attr->ia_valid & ATTR_MODE)
3630                         err = btrfs_acl_chmod(inode);
3631         }
3632
3633         return err;
3634 }
3635
3636 void btrfs_evict_inode(struct inode *inode)
3637 {
3638         struct btrfs_trans_handle *trans;
3639         struct btrfs_root *root = BTRFS_I(inode)->root;
3640         unsigned long nr;
3641         int ret;
3642
3643         trace_btrfs_inode_evict(inode);
3644
3645         truncate_inode_pages(&inode->i_data, 0);
3646         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3647                                is_free_space_inode(root, inode)))
3648                 goto no_delete;
3649
3650         if (is_bad_inode(inode)) {
3651                 btrfs_orphan_del(NULL, inode);
3652                 goto no_delete;
3653         }
3654         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3655         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3656
3657         if (root->fs_info->log_root_recovering) {
3658                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3659                 goto no_delete;
3660         }
3661
3662         if (inode->i_nlink > 0) {
3663                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3664                 goto no_delete;
3665         }
3666
3667         btrfs_i_size_write(inode, 0);
3668
3669         while (1) {
3670                 trans = btrfs_join_transaction(root);
3671                 BUG_ON(IS_ERR(trans));
3672                 trans->block_rsv = root->orphan_block_rsv;
3673
3674                 ret = btrfs_block_rsv_check(trans, root,
3675                                             root->orphan_block_rsv, 0, 5);
3676                 if (ret) {
3677                         BUG_ON(ret != -EAGAIN);
3678                         ret = btrfs_commit_transaction(trans, root);
3679                         BUG_ON(ret);
3680                         continue;
3681                 }
3682
3683                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3684                 if (ret != -EAGAIN)
3685                         break;
3686
3687                 nr = trans->blocks_used;
3688                 btrfs_end_transaction(trans, root);
3689                 trans = NULL;
3690                 btrfs_btree_balance_dirty(root, nr);
3691
3692         }
3693
3694         if (ret == 0) {
3695                 ret = btrfs_orphan_del(trans, inode);
3696                 BUG_ON(ret);
3697         }
3698
3699         if (!(root == root->fs_info->tree_root ||
3700               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3701                 btrfs_return_ino(root, btrfs_ino(inode));
3702
3703         nr = trans->blocks_used;
3704         btrfs_end_transaction(trans, root);
3705         btrfs_btree_balance_dirty(root, nr);
3706 no_delete:
3707         end_writeback(inode);
3708         return;
3709 }
3710
3711 /*
3712  * this returns the key found in the dir entry in the location pointer.
3713  * If no dir entries were found, location->objectid is 0.
3714  */
3715 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3716                                struct btrfs_key *location)
3717 {
3718         const char *name = dentry->d_name.name;
3719         int namelen = dentry->d_name.len;
3720         struct btrfs_dir_item *di;
3721         struct btrfs_path *path;
3722         struct btrfs_root *root = BTRFS_I(dir)->root;
3723         int ret = 0;
3724
3725         path = btrfs_alloc_path();
3726         BUG_ON(!path);
3727
3728         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3729                                     namelen, 0);
3730         if (IS_ERR(di))
3731                 ret = PTR_ERR(di);
3732
3733         if (IS_ERR_OR_NULL(di))
3734                 goto out_err;
3735
3736         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3737 out:
3738         btrfs_free_path(path);
3739         return ret;
3740 out_err:
3741         location->objectid = 0;
3742         goto out;
3743 }
3744
3745 /*
3746  * when we hit a tree root in a directory, the btrfs part of the inode
3747  * needs to be changed to reflect the root directory of the tree root.  This
3748  * is kind of like crossing a mount point.
3749  */
3750 static int fixup_tree_root_location(struct btrfs_root *root,
3751                                     struct inode *dir,
3752                                     struct dentry *dentry,
3753                                     struct btrfs_key *location,
3754                                     struct btrfs_root **sub_root)
3755 {
3756         struct btrfs_path *path;
3757         struct btrfs_root *new_root;
3758         struct btrfs_root_ref *ref;
3759         struct extent_buffer *leaf;
3760         int ret;
3761         int err = 0;
3762
3763         path = btrfs_alloc_path();
3764         if (!path) {
3765                 err = -ENOMEM;
3766                 goto out;
3767         }
3768
3769         err = -ENOENT;
3770         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3771                                   BTRFS_I(dir)->root->root_key.objectid,
3772                                   location->objectid);
3773         if (ret) {
3774                 if (ret < 0)
3775                         err = ret;
3776                 goto out;
3777         }
3778
3779         leaf = path->nodes[0];
3780         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3781         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3782             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3783                 goto out;
3784
3785         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3786                                    (unsigned long)(ref + 1),
3787                                    dentry->d_name.len);
3788         if (ret)
3789                 goto out;
3790
3791         btrfs_release_path(path);
3792
3793         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3794         if (IS_ERR(new_root)) {
3795                 err = PTR_ERR(new_root);
3796                 goto out;
3797         }
3798
3799         if (btrfs_root_refs(&new_root->root_item) == 0) {
3800                 err = -ENOENT;
3801                 goto out;
3802         }
3803
3804         *sub_root = new_root;
3805         location->objectid = btrfs_root_dirid(&new_root->root_item);
3806         location->type = BTRFS_INODE_ITEM_KEY;
3807         location->offset = 0;
3808         err = 0;
3809 out:
3810         btrfs_free_path(path);
3811         return err;
3812 }
3813
3814 static void inode_tree_add(struct inode *inode)
3815 {
3816         struct btrfs_root *root = BTRFS_I(inode)->root;
3817         struct btrfs_inode *entry;
3818         struct rb_node **p;
3819         struct rb_node *parent;
3820         u64 ino = btrfs_ino(inode);
3821 again:
3822         p = &root->inode_tree.rb_node;
3823         parent = NULL;
3824
3825         if (inode_unhashed(inode))
3826                 return;
3827
3828         spin_lock(&root->inode_lock);
3829         while (*p) {
3830                 parent = *p;
3831                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3832
3833                 if (ino < btrfs_ino(&entry->vfs_inode))
3834                         p = &parent->rb_left;
3835                 else if (ino > btrfs_ino(&entry->vfs_inode))
3836                         p = &parent->rb_right;
3837                 else {
3838                         WARN_ON(!(entry->vfs_inode.i_state &
3839                                   (I_WILL_FREE | I_FREEING)));
3840                         rb_erase(parent, &root->inode_tree);
3841                         RB_CLEAR_NODE(parent);
3842                         spin_unlock(&root->inode_lock);
3843                         goto again;
3844                 }
3845         }
3846         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3847         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3848         spin_unlock(&root->inode_lock);
3849 }
3850
3851 static void inode_tree_del(struct inode *inode)
3852 {
3853         struct btrfs_root *root = BTRFS_I(inode)->root;
3854         int empty = 0;
3855
3856         spin_lock(&root->inode_lock);
3857         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3858                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3859                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3860                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3861         }
3862         spin_unlock(&root->inode_lock);
3863
3864         /*
3865          * Free space cache has inodes in the tree root, but the tree root has a
3866          * root_refs of 0, so this could end up dropping the tree root as a
3867          * snapshot, so we need the extra !root->fs_info->tree_root check to
3868          * make sure we don't drop it.
3869          */
3870         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3871             root != root->fs_info->tree_root) {
3872                 synchronize_srcu(&root->fs_info->subvol_srcu);
3873                 spin_lock(&root->inode_lock);
3874                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3875                 spin_unlock(&root->inode_lock);
3876                 if (empty)
3877                         btrfs_add_dead_root(root);
3878         }
3879 }
3880
3881 int btrfs_invalidate_inodes(struct btrfs_root *root)
3882 {
3883         struct rb_node *node;
3884         struct rb_node *prev;
3885         struct btrfs_inode *entry;
3886         struct inode *inode;
3887         u64 objectid = 0;
3888
3889         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3890
3891         spin_lock(&root->inode_lock);
3892 again:
3893         node = root->inode_tree.rb_node;
3894         prev = NULL;
3895         while (node) {
3896                 prev = node;
3897                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3898
3899                 if (objectid < btrfs_ino(&entry->vfs_inode))
3900                         node = node->rb_left;
3901                 else if (objectid > btrfs_ino(&entry->vfs_inode))
3902                         node = node->rb_right;
3903                 else
3904                         break;
3905         }
3906         if (!node) {
3907                 while (prev) {
3908                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3909                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3910                                 node = prev;
3911                                 break;
3912                         }
3913                         prev = rb_next(prev);
3914                 }
3915         }
3916         while (node) {
3917                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3918                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3919                 inode = igrab(&entry->vfs_inode);
3920                 if (inode) {
3921                         spin_unlock(&root->inode_lock);
3922                         if (atomic_read(&inode->i_count) > 1)
3923                                 d_prune_aliases(inode);
3924                         /*
3925                          * btrfs_drop_inode will have it removed from
3926                          * the inode cache when its usage count
3927                          * hits zero.
3928                          */
3929                         iput(inode);
3930                         cond_resched();
3931                         spin_lock(&root->inode_lock);
3932                         goto again;
3933                 }
3934
3935                 if (cond_resched_lock(&root->inode_lock))
3936                         goto again;
3937
3938                 node = rb_next(node);
3939         }
3940         spin_unlock(&root->inode_lock);
3941         return 0;
3942 }
3943
3944 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3945 {
3946         struct btrfs_iget_args *args = p;
3947         inode->i_ino = args->ino;
3948         BTRFS_I(inode)->root = args->root;
3949         btrfs_set_inode_space_info(args->root, inode);
3950         return 0;
3951 }
3952
3953 static int btrfs_find_actor(struct inode *inode, void *opaque)
3954 {
3955         struct btrfs_iget_args *args = opaque;
3956         return args->ino == btrfs_ino(inode) &&
3957                 args->root == BTRFS_I(inode)->root;
3958 }
3959
3960 static struct inode *btrfs_iget_locked(struct super_block *s,
3961                                        u64 objectid,
3962                                        struct btrfs_root *root)
3963 {
3964         struct inode *inode;
3965         struct btrfs_iget_args args;
3966         args.ino = objectid;
3967         args.root = root;
3968
3969         inode = iget5_locked(s, objectid, btrfs_find_actor,
3970                              btrfs_init_locked_inode,
3971                              (void *)&args);
3972         return inode;
3973 }
3974
3975 /* Get an inode object given its location and corresponding root.
3976  * Returns in *is_new if the inode was read from disk
3977  */
3978 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3979                          struct btrfs_root *root, int *new)
3980 {
3981         struct inode *inode;
3982
3983         inode = btrfs_iget_locked(s, location->objectid, root);
3984         if (!inode)
3985                 return ERR_PTR(-ENOMEM);
3986
3987         if (inode->i_state & I_NEW) {
3988                 BTRFS_I(inode)->root = root;
3989                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3990                 btrfs_read_locked_inode(inode);
3991                 inode_tree_add(inode);
3992                 unlock_new_inode(inode);
3993                 if (new)
3994                         *new = 1;
3995         }
3996
3997         return inode;
3998 }
3999
4000 static struct inode *new_simple_dir(struct super_block *s,
4001                                     struct btrfs_key *key,
4002                                     struct btrfs_root *root)
4003 {
4004         struct inode *inode = new_inode(s);
4005
4006         if (!inode)
4007                 return ERR_PTR(-ENOMEM);
4008
4009         BTRFS_I(inode)->root = root;
4010         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4011         BTRFS_I(inode)->dummy_inode = 1;
4012
4013         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4014         inode->i_op = &simple_dir_inode_operations;
4015         inode->i_fop = &simple_dir_operations;
4016         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4017         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4018
4019         return inode;
4020 }
4021
4022 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4023 {
4024         struct inode *inode;
4025         struct btrfs_root *root = BTRFS_I(dir)->root;
4026         struct btrfs_root *sub_root = root;
4027         struct btrfs_key location;
4028         int index;
4029         int ret;
4030
4031         if (dentry->d_name.len > BTRFS_NAME_LEN)
4032                 return ERR_PTR(-ENAMETOOLONG);
4033
4034         ret = btrfs_inode_by_name(dir, dentry, &location);
4035
4036         if (ret < 0)
4037                 return ERR_PTR(ret);
4038
4039         if (location.objectid == 0)
4040                 return NULL;
4041
4042         if (location.type == BTRFS_INODE_ITEM_KEY) {
4043                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4044                 return inode;
4045         }
4046
4047         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4048
4049         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4050         ret = fixup_tree_root_location(root, dir, dentry,
4051                                        &location, &sub_root);
4052         if (ret < 0) {
4053                 if (ret != -ENOENT)
4054                         inode = ERR_PTR(ret);
4055                 else
4056                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4057         } else {
4058                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4059         }
4060         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4061
4062         if (!IS_ERR(inode) && root != sub_root) {
4063                 down_read(&root->fs_info->cleanup_work_sem);
4064                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4065                         ret = btrfs_orphan_cleanup(sub_root);
4066                 up_read(&root->fs_info->cleanup_work_sem);
4067                 if (ret)
4068                         inode = ERR_PTR(ret);
4069         }
4070
4071         return inode;
4072 }
4073
4074 static int btrfs_dentry_delete(const struct dentry *dentry)
4075 {
4076         struct btrfs_root *root;
4077
4078         if (!dentry->d_inode && !IS_ROOT(dentry))
4079                 dentry = dentry->d_parent;
4080
4081         if (dentry->d_inode) {
4082                 root = BTRFS_I(dentry->d_inode)->root;
4083                 if (btrfs_root_refs(&root->root_item) == 0)
4084                         return 1;
4085         }
4086         return 0;
4087 }
4088
4089 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4090                                    struct nameidata *nd)
4091 {
4092         struct inode *inode;
4093
4094         inode = btrfs_lookup_dentry(dir, dentry);
4095         if (IS_ERR(inode))
4096                 return ERR_CAST(inode);
4097
4098         return d_splice_alias(inode, dentry);
4099 }
4100
4101 unsigned char btrfs_filetype_table[] = {
4102         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4103 };
4104
4105 static int btrfs_real_readdir(struct file *filp, void *dirent,
4106                               filldir_t filldir)
4107 {
4108         struct inode *inode = filp->f_dentry->d_inode;
4109         struct btrfs_root *root = BTRFS_I(inode)->root;
4110         struct btrfs_item *item;
4111         struct btrfs_dir_item *di;
4112         struct btrfs_key key;
4113         struct btrfs_key found_key;
4114         struct btrfs_path *path;
4115         struct list_head ins_list;
4116         struct list_head del_list;
4117         int ret;
4118         struct extent_buffer *leaf;
4119         int slot;
4120         unsigned char d_type;
4121         int over = 0;
4122         u32 di_cur;
4123         u32 di_total;
4124         u32 di_len;
4125         int key_type = BTRFS_DIR_INDEX_KEY;
4126         char tmp_name[32];
4127         char *name_ptr;
4128         int name_len;
4129         int is_curr = 0;        /* filp->f_pos points to the current index? */
4130
4131         /* FIXME, use a real flag for deciding about the key type */
4132         if (root->fs_info->tree_root == root)
4133                 key_type = BTRFS_DIR_ITEM_KEY;
4134
4135         /* special case for "." */
4136         if (filp->f_pos == 0) {
4137                 over = filldir(dirent, ".", 1, 1, btrfs_ino(inode), DT_DIR);
4138                 if (over)
4139                         return 0;
4140                 filp->f_pos = 1;
4141         }
4142         /* special case for .., just use the back ref */
4143         if (filp->f_pos == 1) {
4144                 u64 pino = parent_ino(filp->f_path.dentry);
4145                 over = filldir(dirent, "..", 2,
4146                                2, pino, DT_DIR);
4147                 if (over)
4148                         return 0;
4149                 filp->f_pos = 2;
4150         }
4151         path = btrfs_alloc_path();
4152         if (!path)
4153                 return -ENOMEM;
4154
4155         path->reada = 1;
4156
4157         if (key_type == BTRFS_DIR_INDEX_KEY) {
4158                 INIT_LIST_HEAD(&ins_list);
4159                 INIT_LIST_HEAD(&del_list);
4160                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4161         }
4162
4163         btrfs_set_key_type(&key, key_type);
4164         key.offset = filp->f_pos;
4165         key.objectid = btrfs_ino(inode);
4166
4167         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4168         if (ret < 0)
4169                 goto err;
4170
4171         while (1) {
4172                 leaf = path->nodes[0];
4173                 slot = path->slots[0];
4174                 if (slot >= btrfs_header_nritems(leaf)) {
4175                         ret = btrfs_next_leaf(root, path);
4176                         if (ret < 0)
4177                                 goto err;
4178                         else if (ret > 0)
4179                                 break;
4180                         continue;
4181                 }
4182
4183                 item = btrfs_item_nr(leaf, slot);
4184                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4185
4186                 if (found_key.objectid != key.objectid)
4187                         break;
4188                 if (btrfs_key_type(&found_key) != key_type)
4189                         break;
4190                 if (found_key.offset < filp->f_pos)
4191                         goto next;
4192                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4193                     btrfs_should_delete_dir_index(&del_list,
4194                                                   found_key.offset))
4195                         goto next;
4196
4197                 filp->f_pos = found_key.offset;
4198                 is_curr = 1;
4199
4200                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4201                 di_cur = 0;
4202                 di_total = btrfs_item_size(leaf, item);
4203
4204                 while (di_cur < di_total) {
4205                         struct btrfs_key location;
4206
4207                         if (verify_dir_item(root, leaf, di))
4208                                 break;
4209
4210                         name_len = btrfs_dir_name_len(leaf, di);
4211                         if (name_len <= sizeof(tmp_name)) {
4212                                 name_ptr = tmp_name;
4213                         } else {
4214                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4215                                 if (!name_ptr) {
4216                                         ret = -ENOMEM;
4217                                         goto err;
4218                                 }
4219                         }
4220                         read_extent_buffer(leaf, name_ptr,
4221                                            (unsigned long)(di + 1), name_len);
4222
4223                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4224                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4225
4226                         /* is this a reference to our own snapshot? If so
4227                          * skip it
4228                          */
4229                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4230                             location.objectid == root->root_key.objectid) {
4231                                 over = 0;
4232                                 goto skip;
4233                         }
4234                         over = filldir(dirent, name_ptr, name_len,
4235                                        found_key.offset, location.objectid,
4236                                        d_type);
4237
4238 skip:
4239                         if (name_ptr != tmp_name)
4240                                 kfree(name_ptr);
4241
4242                         if (over)
4243                                 goto nopos;
4244                         di_len = btrfs_dir_name_len(leaf, di) +
4245                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4246                         di_cur += di_len;
4247                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4248                 }
4249 next:
4250                 path->slots[0]++;
4251         }
4252
4253         if (key_type == BTRFS_DIR_INDEX_KEY) {
4254                 if (is_curr)
4255                         filp->f_pos++;
4256                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4257                                                       &ins_list);
4258                 if (ret)
4259                         goto nopos;
4260         }
4261
4262         /* Reached end of directory/root. Bump pos past the last item. */
4263         if (key_type == BTRFS_DIR_INDEX_KEY)
4264                 /*
4265                  * 32-bit glibc will use getdents64, but then strtol -
4266                  * so the last number we can serve is this.
4267                  */
4268                 filp->f_pos = 0x7fffffff;
4269         else
4270                 filp->f_pos++;
4271 nopos:
4272         ret = 0;
4273 err:
4274         if (key_type == BTRFS_DIR_INDEX_KEY)
4275                 btrfs_put_delayed_items(&ins_list, &del_list);
4276         btrfs_free_path(path);
4277         return ret;
4278 }
4279
4280 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4281 {
4282         struct btrfs_root *root = BTRFS_I(inode)->root;
4283         struct btrfs_trans_handle *trans;
4284         int ret = 0;
4285         bool nolock = false;
4286
4287         if (BTRFS_I(inode)->dummy_inode)
4288                 return 0;
4289
4290         if (btrfs_fs_closing(root->fs_info) && is_free_space_inode(root, inode))
4291                 nolock = true;
4292
4293         if (wbc->sync_mode == WB_SYNC_ALL) {
4294                 if (nolock)
4295                         trans = btrfs_join_transaction_nolock(root);
4296                 else
4297                         trans = btrfs_join_transaction(root);
4298                 if (IS_ERR(trans))
4299                         return PTR_ERR(trans);
4300                 if (nolock)
4301                         ret = btrfs_end_transaction_nolock(trans, root);
4302                 else
4303                         ret = btrfs_commit_transaction(trans, root);
4304         }
4305         return ret;
4306 }
4307
4308 /*
4309  * This is somewhat expensive, updating the tree every time the
4310  * inode changes.  But, it is most likely to find the inode in cache.
4311  * FIXME, needs more benchmarking...there are no reasons other than performance
4312  * to keep or drop this code.
4313  */
4314 void btrfs_dirty_inode(struct inode *inode)
4315 {
4316         struct btrfs_root *root = BTRFS_I(inode)->root;
4317         struct btrfs_trans_handle *trans;
4318         int ret;
4319
4320         if (BTRFS_I(inode)->dummy_inode)
4321                 return;
4322
4323         trans = btrfs_join_transaction(root);
4324         BUG_ON(IS_ERR(trans));
4325
4326         ret = btrfs_update_inode(trans, root, inode);
4327         if (ret && ret == -ENOSPC) {
4328                 /* whoops, lets try again with the full transaction */
4329                 btrfs_end_transaction(trans, root);
4330                 trans = btrfs_start_transaction(root, 1);
4331                 if (IS_ERR(trans)) {
4332                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4333                                        "dirty  inode %llu error %ld\n",
4334                                        (unsigned long long)btrfs_ino(inode),
4335                                        PTR_ERR(trans));
4336                         return;
4337                 }
4338
4339                 ret = btrfs_update_inode(trans, root, inode);
4340                 if (ret) {
4341                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4342                                        "dirty  inode %llu error %d\n",
4343                                        (unsigned long long)btrfs_ino(inode),
4344                                        ret);
4345                 }
4346         }
4347         btrfs_end_transaction(trans, root);
4348         if (BTRFS_I(inode)->delayed_node)
4349                 btrfs_balance_delayed_items(root);
4350 }
4351
4352 /*
4353  * find the highest existing sequence number in a directory
4354  * and then set the in-memory index_cnt variable to reflect
4355  * free sequence numbers
4356  */
4357 static int btrfs_set_inode_index_count(struct inode *inode)
4358 {
4359         struct btrfs_root *root = BTRFS_I(inode)->root;
4360         struct btrfs_key key, found_key;
4361         struct btrfs_path *path;
4362         struct extent_buffer *leaf;
4363         int ret;
4364
4365         key.objectid = btrfs_ino(inode);
4366         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4367         key.offset = (u64)-1;
4368
4369         path = btrfs_alloc_path();
4370         if (!path)
4371                 return -ENOMEM;
4372
4373         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4374         if (ret < 0)
4375                 goto out;
4376         /* FIXME: we should be able to handle this */
4377         if (ret == 0)
4378                 goto out;
4379         ret = 0;
4380
4381         /*
4382          * MAGIC NUMBER EXPLANATION:
4383          * since we search a directory based on f_pos we have to start at 2
4384          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4385          * else has to start at 2
4386          */
4387         if (path->slots[0] == 0) {
4388                 BTRFS_I(inode)->index_cnt = 2;
4389                 goto out;
4390         }
4391
4392         path->slots[0]--;
4393
4394         leaf = path->nodes[0];
4395         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4396
4397         if (found_key.objectid != btrfs_ino(inode) ||
4398             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4399                 BTRFS_I(inode)->index_cnt = 2;
4400                 goto out;
4401         }
4402
4403         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4404 out:
4405         btrfs_free_path(path);
4406         return ret;
4407 }
4408
4409 /*
4410  * helper to find a free sequence number in a given directory.  This current
4411  * code is very simple, later versions will do smarter things in the btree
4412  */
4413 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4414 {
4415         int ret = 0;
4416
4417         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4418                 ret = btrfs_inode_delayed_dir_index_count(dir);
4419                 if (ret) {
4420                         ret = btrfs_set_inode_index_count(dir);
4421                         if (ret)
4422                                 return ret;
4423                 }
4424         }
4425
4426         *index = BTRFS_I(dir)->index_cnt;
4427         BTRFS_I(dir)->index_cnt++;
4428
4429         return ret;
4430 }
4431
4432 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4433                                      struct btrfs_root *root,
4434                                      struct inode *dir,
4435                                      const char *name, int name_len,
4436                                      u64 ref_objectid, u64 objectid, int mode,
4437                                      u64 *index)
4438 {
4439         struct inode *inode;
4440         struct btrfs_inode_item *inode_item;
4441         struct btrfs_key *location;
4442         struct btrfs_path *path;
4443         struct btrfs_inode_ref *ref;
4444         struct btrfs_key key[2];
4445         u32 sizes[2];
4446         unsigned long ptr;
4447         int ret;
4448         int owner;
4449
4450         path = btrfs_alloc_path();
4451         BUG_ON(!path);
4452
4453         inode = new_inode(root->fs_info->sb);
4454         if (!inode) {
4455                 btrfs_free_path(path);
4456                 return ERR_PTR(-ENOMEM);
4457         }
4458
4459         /*
4460          * we have to initialize this early, so we can reclaim the inode
4461          * number if we fail afterwards in this function.
4462          */
4463         inode->i_ino = objectid;
4464
4465         if (dir) {
4466                 trace_btrfs_inode_request(dir);
4467
4468                 ret = btrfs_set_inode_index(dir, index);
4469                 if (ret) {
4470                         btrfs_free_path(path);
4471                         iput(inode);
4472                         return ERR_PTR(ret);
4473                 }
4474         }
4475         /*
4476          * index_cnt is ignored for everything but a dir,
4477          * btrfs_get_inode_index_count has an explanation for the magic
4478          * number
4479          */
4480         BTRFS_I(inode)->index_cnt = 2;
4481         BTRFS_I(inode)->root = root;
4482         BTRFS_I(inode)->generation = trans->transid;
4483         inode->i_generation = BTRFS_I(inode)->generation;
4484         btrfs_set_inode_space_info(root, inode);
4485
4486         if (mode & S_IFDIR)
4487                 owner = 0;
4488         else
4489                 owner = 1;
4490
4491         key[0].objectid = objectid;
4492         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4493         key[0].offset = 0;
4494
4495         key[1].objectid = objectid;
4496         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4497         key[1].offset = ref_objectid;
4498
4499         sizes[0] = sizeof(struct btrfs_inode_item);
4500         sizes[1] = name_len + sizeof(*ref);
4501
4502         path->leave_spinning = 1;
4503         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4504         if (ret != 0)
4505                 goto fail;
4506
4507         inode_init_owner(inode, dir, mode);
4508         inode_set_bytes(inode, 0);
4509         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4510         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4511                                   struct btrfs_inode_item);
4512         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4513
4514         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4515                              struct btrfs_inode_ref);
4516         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4517         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4518         ptr = (unsigned long)(ref + 1);
4519         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4520
4521         btrfs_mark_buffer_dirty(path->nodes[0]);
4522         btrfs_free_path(path);
4523
4524         location = &BTRFS_I(inode)->location;
4525         location->objectid = objectid;
4526         location->offset = 0;
4527         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4528
4529         btrfs_inherit_iflags(inode, dir);
4530
4531         if ((mode & S_IFREG)) {
4532                 if (btrfs_test_opt(root, NODATASUM))
4533                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4534                 if (btrfs_test_opt(root, NODATACOW) ||
4535                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4536                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4537         }
4538
4539         insert_inode_hash(inode);
4540         inode_tree_add(inode);
4541
4542         trace_btrfs_inode_new(inode);
4543         btrfs_set_inode_last_trans(trans, inode);
4544
4545         return inode;
4546 fail:
4547         if (dir)
4548                 BTRFS_I(dir)->index_cnt--;
4549         btrfs_free_path(path);
4550         iput(inode);
4551         return ERR_PTR(ret);
4552 }
4553
4554 static inline u8 btrfs_inode_type(struct inode *inode)
4555 {
4556         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4557 }
4558
4559 /*
4560  * utility function to add 'inode' into 'parent_inode' with
4561  * a give name and a given sequence number.
4562  * if 'add_backref' is true, also insert a backref from the
4563  * inode to the parent directory.
4564  */
4565 int btrfs_add_link(struct btrfs_trans_handle *trans,
4566                    struct inode *parent_inode, struct inode *inode,
4567                    const char *name, int name_len, int add_backref, u64 index)
4568 {
4569         int ret = 0;
4570         struct btrfs_key key;
4571         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4572         u64 ino = btrfs_ino(inode);
4573         u64 parent_ino = btrfs_ino(parent_inode);
4574
4575         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4576                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4577         } else {
4578                 key.objectid = ino;
4579                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4580                 key.offset = 0;
4581         }
4582
4583         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4584                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4585                                          key.objectid, root->root_key.objectid,
4586                                          parent_ino, index, name, name_len);
4587         } else if (add_backref) {
4588                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4589                                              parent_ino, index);
4590         }
4591
4592         if (ret == 0) {
4593                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4594                                             parent_inode, &key,
4595                                             btrfs_inode_type(inode), index);
4596                 BUG_ON(ret);
4597
4598                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4599                                    name_len * 2);
4600                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4601                 ret = btrfs_update_inode(trans, root, parent_inode);
4602         }
4603         return ret;
4604 }
4605
4606 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4607                             struct inode *dir, struct dentry *dentry,
4608                             struct inode *inode, int backref, u64 index)
4609 {
4610         int err = btrfs_add_link(trans, dir, inode,
4611                                  dentry->d_name.name, dentry->d_name.len,
4612                                  backref, index);
4613         if (!err) {
4614                 d_instantiate(dentry, inode);
4615                 return 0;
4616         }
4617         if (err > 0)
4618                 err = -EEXIST;
4619         return err;
4620 }
4621
4622 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4623                         int mode, dev_t rdev)
4624 {
4625         struct btrfs_trans_handle *trans;
4626         struct btrfs_root *root = BTRFS_I(dir)->root;
4627         struct inode *inode = NULL;
4628         int err;
4629         int drop_inode = 0;
4630         u64 objectid;
4631         unsigned long nr = 0;
4632         u64 index = 0;
4633
4634         if (!new_valid_dev(rdev))
4635                 return -EINVAL;
4636
4637         /*
4638          * 2 for inode item and ref
4639          * 2 for dir items
4640          * 1 for xattr if selinux is on
4641          */
4642         trans = btrfs_start_transaction(root, 5);
4643         if (IS_ERR(trans))
4644                 return PTR_ERR(trans);
4645
4646         err = btrfs_find_free_ino(root, &objectid);
4647         if (err)
4648                 goto out_unlock;
4649
4650         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4651                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4652                                 mode, &index);
4653         if (IS_ERR(inode)) {
4654                 err = PTR_ERR(inode);
4655                 goto out_unlock;
4656         }
4657
4658         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4659         if (err) {
4660                 drop_inode = 1;
4661                 goto out_unlock;
4662         }
4663
4664         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4665         if (err)
4666                 drop_inode = 1;
4667         else {
4668                 inode->i_op = &btrfs_special_inode_operations;
4669                 init_special_inode(inode, inode->i_mode, rdev);
4670                 btrfs_update_inode(trans, root, inode);
4671         }
4672 out_unlock:
4673         nr = trans->blocks_used;
4674         btrfs_end_transaction_throttle(trans, root);
4675         btrfs_btree_balance_dirty(root, nr);
4676         if (drop_inode) {
4677                 inode_dec_link_count(inode);
4678                 iput(inode);
4679         }
4680         return err;
4681 }
4682
4683 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4684                         int mode, struct nameidata *nd)
4685 {
4686         struct btrfs_trans_handle *trans;
4687         struct btrfs_root *root = BTRFS_I(dir)->root;
4688         struct inode *inode = NULL;
4689         int drop_inode = 0;
4690         int err;
4691         unsigned long nr = 0;
4692         u64 objectid;
4693         u64 index = 0;
4694
4695         /*
4696          * 2 for inode item and ref
4697          * 2 for dir items
4698          * 1 for xattr if selinux is on
4699          */
4700         trans = btrfs_start_transaction(root, 5);
4701         if (IS_ERR(trans))
4702                 return PTR_ERR(trans);
4703
4704         err = btrfs_find_free_ino(root, &objectid);
4705         if (err)
4706                 goto out_unlock;
4707
4708         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4709                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4710                                 mode, &index);
4711         if (IS_ERR(inode)) {
4712                 err = PTR_ERR(inode);
4713                 goto out_unlock;
4714         }
4715
4716         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4717         if (err) {
4718                 drop_inode = 1;
4719                 goto out_unlock;
4720         }
4721
4722         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4723         if (err)
4724                 drop_inode = 1;
4725         else {
4726                 inode->i_mapping->a_ops = &btrfs_aops;
4727                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4728                 inode->i_fop = &btrfs_file_operations;
4729                 inode->i_op = &btrfs_file_inode_operations;
4730                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4731         }
4732 out_unlock:
4733         nr = trans->blocks_used;
4734         btrfs_end_transaction_throttle(trans, root);
4735         if (drop_inode) {
4736                 inode_dec_link_count(inode);
4737                 iput(inode);
4738         }
4739         btrfs_btree_balance_dirty(root, nr);
4740         return err;
4741 }
4742
4743 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4744                       struct dentry *dentry)
4745 {
4746         struct btrfs_trans_handle *trans;
4747         struct btrfs_root *root = BTRFS_I(dir)->root;
4748         struct inode *inode = old_dentry->d_inode;
4749         u64 index;
4750         unsigned long nr = 0;
4751         int err;
4752         int drop_inode = 0;
4753
4754         /* do not allow sys_link's with other subvols of the same device */
4755         if (root->objectid != BTRFS_I(inode)->root->objectid)
4756                 return -EXDEV;
4757
4758         if (inode->i_nlink == ~0U)
4759                 return -EMLINK;
4760
4761         err = btrfs_set_inode_index(dir, &index);
4762         if (err)
4763                 goto fail;
4764
4765         /*
4766          * 2 items for inode and inode ref
4767          * 2 items for dir items
4768          * 1 item for parent inode
4769          */
4770         trans = btrfs_start_transaction(root, 5);
4771         if (IS_ERR(trans)) {
4772                 err = PTR_ERR(trans);
4773                 goto fail;
4774         }
4775
4776         btrfs_inc_nlink(inode);
4777         inode->i_ctime = CURRENT_TIME;
4778         ihold(inode);
4779
4780         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4781
4782         if (err) {
4783                 drop_inode = 1;
4784         } else {
4785                 struct dentry *parent = dget_parent(dentry);
4786                 err = btrfs_update_inode(trans, root, inode);
4787                 BUG_ON(err);
4788                 btrfs_log_new_name(trans, inode, NULL, parent);
4789                 dput(parent);
4790         }
4791
4792         nr = trans->blocks_used;
4793         btrfs_end_transaction_throttle(trans, root);
4794 fail:
4795         if (drop_inode) {
4796                 inode_dec_link_count(inode);
4797                 iput(inode);
4798         }
4799         btrfs_btree_balance_dirty(root, nr);
4800         return err;
4801 }
4802
4803 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4804 {
4805         struct inode *inode = NULL;
4806         struct btrfs_trans_handle *trans;
4807         struct btrfs_root *root = BTRFS_I(dir)->root;
4808         int err = 0;
4809         int drop_on_err = 0;
4810         u64 objectid = 0;
4811         u64 index = 0;
4812         unsigned long nr = 1;
4813
4814         /*
4815          * 2 items for inode and ref
4816          * 2 items for dir items
4817          * 1 for xattr if selinux is on
4818          */
4819         trans = btrfs_start_transaction(root, 5);
4820         if (IS_ERR(trans))
4821                 return PTR_ERR(trans);
4822
4823         err = btrfs_find_free_ino(root, &objectid);
4824         if (err)
4825                 goto out_fail;
4826
4827         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4828                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4829                                 S_IFDIR | mode, &index);
4830         if (IS_ERR(inode)) {
4831                 err = PTR_ERR(inode);
4832                 goto out_fail;
4833         }
4834
4835         drop_on_err = 1;
4836
4837         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4838         if (err)
4839                 goto out_fail;
4840
4841         inode->i_op = &btrfs_dir_inode_operations;
4842         inode->i_fop = &btrfs_dir_file_operations;
4843
4844         btrfs_i_size_write(inode, 0);
4845         err = btrfs_update_inode(trans, root, inode);
4846         if (err)
4847                 goto out_fail;
4848
4849         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4850                              dentry->d_name.len, 0, index);
4851         if (err)
4852                 goto out_fail;
4853
4854         d_instantiate(dentry, inode);
4855         drop_on_err = 0;
4856
4857 out_fail:
4858         nr = trans->blocks_used;
4859         btrfs_end_transaction_throttle(trans, root);
4860         if (drop_on_err)
4861                 iput(inode);
4862         btrfs_btree_balance_dirty(root, nr);
4863         return err;
4864 }
4865
4866 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4867  * and an extent that you want to insert, deal with overlap and insert
4868  * the new extent into the tree.
4869  */
4870 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4871                                 struct extent_map *existing,
4872                                 struct extent_map *em,
4873                                 u64 map_start, u64 map_len)
4874 {
4875         u64 start_diff;
4876
4877         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4878         start_diff = map_start - em->start;
4879         em->start = map_start;
4880         em->len = map_len;
4881         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4882             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4883                 em->block_start += start_diff;
4884                 em->block_len -= start_diff;
4885         }
4886         return add_extent_mapping(em_tree, em);
4887 }
4888
4889 static noinline int uncompress_inline(struct btrfs_path *path,
4890                                       struct inode *inode, struct page *page,
4891                                       size_t pg_offset, u64 extent_offset,
4892                                       struct btrfs_file_extent_item *item)
4893 {
4894         int ret;
4895         struct extent_buffer *leaf = path->nodes[0];
4896         char *tmp;
4897         size_t max_size;
4898         unsigned long inline_size;
4899         unsigned long ptr;
4900         int compress_type;
4901
4902         WARN_ON(pg_offset != 0);
4903         compress_type = btrfs_file_extent_compression(leaf, item);
4904         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4905         inline_size = btrfs_file_extent_inline_item_len(leaf,
4906                                         btrfs_item_nr(leaf, path->slots[0]));
4907         tmp = kmalloc(inline_size, GFP_NOFS);
4908         if (!tmp)
4909                 return -ENOMEM;
4910         ptr = btrfs_file_extent_inline_start(item);
4911
4912         read_extent_buffer(leaf, tmp, ptr, inline_size);
4913
4914         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4915         ret = btrfs_decompress(compress_type, tmp, page,
4916                                extent_offset, inline_size, max_size);
4917         if (ret) {
4918                 char *kaddr = kmap_atomic(page, KM_USER0);
4919                 unsigned long copy_size = min_t(u64,
4920                                   PAGE_CACHE_SIZE - pg_offset,
4921                                   max_size - extent_offset);
4922                 memset(kaddr + pg_offset, 0, copy_size);
4923                 kunmap_atomic(kaddr, KM_USER0);
4924         }
4925         kfree(tmp);
4926         return 0;
4927 }
4928
4929 /*
4930  * a bit scary, this does extent mapping from logical file offset to the disk.
4931  * the ugly parts come from merging extents from the disk with the in-ram
4932  * representation.  This gets more complex because of the data=ordered code,
4933  * where the in-ram extents might be locked pending data=ordered completion.
4934  *
4935  * This also copies inline extents directly into the page.
4936  */
4937
4938 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4939                                     size_t pg_offset, u64 start, u64 len,
4940                                     int create)
4941 {
4942         int ret;
4943         int err = 0;
4944         u64 bytenr;
4945         u64 extent_start = 0;
4946         u64 extent_end = 0;
4947         u64 objectid = btrfs_ino(inode);
4948         u32 found_type;
4949         struct btrfs_path *path = NULL;
4950         struct btrfs_root *root = BTRFS_I(inode)->root;
4951         struct btrfs_file_extent_item *item;
4952         struct extent_buffer *leaf;
4953         struct btrfs_key found_key;
4954         struct extent_map *em = NULL;
4955         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4956         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4957         struct btrfs_trans_handle *trans = NULL;
4958         int compress_type;
4959
4960 again:
4961         read_lock(&em_tree->lock);
4962         em = lookup_extent_mapping(em_tree, start, len);
4963         if (em)
4964                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4965         read_unlock(&em_tree->lock);
4966
4967         if (em) {
4968                 if (em->start > start || em->start + em->len <= start)
4969                         free_extent_map(em);
4970                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4971                         free_extent_map(em);
4972                 else
4973                         goto out;
4974         }
4975         em = alloc_extent_map();
4976         if (!em) {
4977                 err = -ENOMEM;
4978                 goto out;
4979         }
4980         em->bdev = root->fs_info->fs_devices->latest_bdev;
4981         em->start = EXTENT_MAP_HOLE;
4982         em->orig_start = EXTENT_MAP_HOLE;
4983         em->len = (u64)-1;
4984         em->block_len = (u64)-1;
4985
4986         if (!path) {
4987                 path = btrfs_alloc_path();
4988                 if (!path) {
4989                         err = -ENOMEM;
4990                         goto out;
4991                 }
4992                 /*
4993                  * Chances are we'll be called again, so go ahead and do
4994                  * readahead
4995                  */
4996                 path->reada = 1;
4997         }
4998
4999         ret = btrfs_lookup_file_extent(trans, root, path,
5000                                        objectid, start, trans != NULL);
5001         if (ret < 0) {
5002                 err = ret;
5003                 goto out;
5004         }
5005
5006         if (ret != 0) {
5007                 if (path->slots[0] == 0)
5008                         goto not_found;
5009                 path->slots[0]--;
5010         }
5011
5012         leaf = path->nodes[0];
5013         item = btrfs_item_ptr(leaf, path->slots[0],
5014                               struct btrfs_file_extent_item);
5015         /* are we inside the extent that was found? */
5016         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5017         found_type = btrfs_key_type(&found_key);
5018         if (found_key.objectid != objectid ||
5019             found_type != BTRFS_EXTENT_DATA_KEY) {
5020                 goto not_found;
5021         }
5022
5023         found_type = btrfs_file_extent_type(leaf, item);
5024         extent_start = found_key.offset;
5025         compress_type = btrfs_file_extent_compression(leaf, item);
5026         if (found_type == BTRFS_FILE_EXTENT_REG ||
5027             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5028                 extent_end = extent_start +
5029                        btrfs_file_extent_num_bytes(leaf, item);
5030         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5031                 size_t size;
5032                 size = btrfs_file_extent_inline_len(leaf, item);
5033                 extent_end = (extent_start + size + root->sectorsize - 1) &
5034                         ~((u64)root->sectorsize - 1);
5035         }
5036
5037         if (start >= extent_end) {
5038                 path->slots[0]++;
5039                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5040                         ret = btrfs_next_leaf(root, path);
5041                         if (ret < 0) {
5042                                 err = ret;
5043                                 goto out;
5044                         }
5045                         if (ret > 0)
5046                                 goto not_found;
5047                         leaf = path->nodes[0];
5048                 }
5049                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5050                 if (found_key.objectid != objectid ||
5051                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5052                         goto not_found;
5053                 if (start + len <= found_key.offset)
5054                         goto not_found;
5055                 em->start = start;
5056                 em->len = found_key.offset - start;
5057                 goto not_found_em;
5058         }
5059
5060         if (found_type == BTRFS_FILE_EXTENT_REG ||
5061             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5062                 em->start = extent_start;
5063                 em->len = extent_end - extent_start;
5064                 em->orig_start = extent_start -
5065                                  btrfs_file_extent_offset(leaf, item);
5066                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5067                 if (bytenr == 0) {
5068                         em->block_start = EXTENT_MAP_HOLE;
5069                         goto insert;
5070                 }
5071                 if (compress_type != BTRFS_COMPRESS_NONE) {
5072                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5073                         em->compress_type = compress_type;
5074                         em->block_start = bytenr;
5075                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5076                                                                          item);
5077                 } else {
5078                         bytenr += btrfs_file_extent_offset(leaf, item);
5079                         em->block_start = bytenr;
5080                         em->block_len = em->len;
5081                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5082                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5083                 }
5084                 goto insert;
5085         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5086                 unsigned long ptr;
5087                 char *map;
5088                 size_t size;
5089                 size_t extent_offset;
5090                 size_t copy_size;
5091
5092                 em->block_start = EXTENT_MAP_INLINE;
5093                 if (!page || create) {
5094                         em->start = extent_start;
5095                         em->len = extent_end - extent_start;
5096                         goto out;
5097                 }
5098
5099                 size = btrfs_file_extent_inline_len(leaf, item);
5100                 extent_offset = page_offset(page) + pg_offset - extent_start;
5101                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5102                                 size - extent_offset);
5103                 em->start = extent_start + extent_offset;
5104                 em->len = (copy_size + root->sectorsize - 1) &
5105                         ~((u64)root->sectorsize - 1);
5106                 em->orig_start = EXTENT_MAP_INLINE;
5107                 if (compress_type) {
5108                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5109                         em->compress_type = compress_type;
5110                 }
5111                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5112                 if (create == 0 && !PageUptodate(page)) {
5113                         if (btrfs_file_extent_compression(leaf, item) !=
5114                             BTRFS_COMPRESS_NONE) {
5115                                 ret = uncompress_inline(path, inode, page,
5116                                                         pg_offset,
5117                                                         extent_offset, item);
5118                                 BUG_ON(ret);
5119                         } else {
5120                                 map = kmap(page);
5121                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5122                                                    copy_size);
5123                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5124                                         memset(map + pg_offset + copy_size, 0,
5125                                                PAGE_CACHE_SIZE - pg_offset -
5126                                                copy_size);
5127                                 }
5128                                 kunmap(page);
5129                         }
5130                         flush_dcache_page(page);
5131                 } else if (create && PageUptodate(page)) {
5132                         WARN_ON(1);
5133                         if (!trans) {
5134                                 kunmap(page);
5135                                 free_extent_map(em);
5136                                 em = NULL;
5137
5138                                 btrfs_release_path(path);
5139                                 trans = btrfs_join_transaction(root);
5140
5141                                 if (IS_ERR(trans))
5142                                         return ERR_CAST(trans);
5143                                 goto again;
5144                         }
5145                         map = kmap(page);
5146                         write_extent_buffer(leaf, map + pg_offset, ptr,
5147                                             copy_size);
5148                         kunmap(page);
5149                         btrfs_mark_buffer_dirty(leaf);
5150                 }
5151                 set_extent_uptodate(io_tree, em->start,
5152                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5153                 goto insert;
5154         } else {
5155                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5156                 WARN_ON(1);
5157         }
5158 not_found:
5159         em->start = start;
5160         em->len = len;
5161 not_found_em:
5162         em->block_start = EXTENT_MAP_HOLE;
5163         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5164 insert:
5165         btrfs_release_path(path);
5166         if (em->start > start || extent_map_end(em) <= start) {
5167                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5168                        "[%llu %llu]\n", (unsigned long long)em->start,
5169                        (unsigned long long)em->len,
5170                        (unsigned long long)start,
5171                        (unsigned long long)len);
5172                 err = -EIO;
5173                 goto out;
5174         }
5175
5176         err = 0;
5177         write_lock(&em_tree->lock);
5178         ret = add_extent_mapping(em_tree, em);
5179         /* it is possible that someone inserted the extent into the tree
5180          * while we had the lock dropped.  It is also possible that
5181          * an overlapping map exists in the tree
5182          */
5183         if (ret == -EEXIST) {
5184                 struct extent_map *existing;
5185
5186                 ret = 0;
5187
5188                 existing = lookup_extent_mapping(em_tree, start, len);
5189                 if (existing && (existing->start > start ||
5190                     existing->start + existing->len <= start)) {
5191                         free_extent_map(existing);
5192                         existing = NULL;
5193                 }
5194                 if (!existing) {
5195                         existing = lookup_extent_mapping(em_tree, em->start,
5196                                                          em->len);
5197                         if (existing) {
5198                                 err = merge_extent_mapping(em_tree, existing,
5199                                                            em, start,
5200                                                            root->sectorsize);
5201                                 free_extent_map(existing);
5202                                 if (err) {
5203                                         free_extent_map(em);
5204                                         em = NULL;
5205                                 }
5206                         } else {
5207                                 err = -EIO;
5208                                 free_extent_map(em);
5209                                 em = NULL;
5210                         }
5211                 } else {
5212                         free_extent_map(em);
5213                         em = existing;
5214                         err = 0;
5215                 }
5216         }
5217         write_unlock(&em_tree->lock);
5218 out:
5219
5220         trace_btrfs_get_extent(root, em);
5221
5222         if (path)
5223                 btrfs_free_path(path);
5224         if (trans) {
5225                 ret = btrfs_end_transaction(trans, root);
5226                 if (!err)
5227                         err = ret;
5228         }
5229         if (err) {
5230                 free_extent_map(em);
5231                 return ERR_PTR(err);
5232         }
5233         return em;
5234 }
5235
5236 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5237                                            size_t pg_offset, u64 start, u64 len,
5238                                            int create)
5239 {
5240         struct extent_map *em;
5241         struct extent_map *hole_em = NULL;
5242         u64 range_start = start;
5243         u64 end;
5244         u64 found;
5245         u64 found_end;
5246         int err = 0;
5247
5248         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5249         if (IS_ERR(em))
5250                 return em;
5251         if (em) {
5252                 /*
5253                  * if our em maps to a hole, there might
5254                  * actually be delalloc bytes behind it
5255                  */
5256                 if (em->block_start != EXTENT_MAP_HOLE)
5257                         return em;
5258                 else
5259                         hole_em = em;
5260         }
5261
5262         /* check to see if we've wrapped (len == -1 or similar) */
5263         end = start + len;
5264         if (end < start)
5265                 end = (u64)-1;
5266         else
5267                 end -= 1;
5268
5269         em = NULL;
5270
5271         /* ok, we didn't find anything, lets look for delalloc */
5272         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5273                                  end, len, EXTENT_DELALLOC, 1);
5274         found_end = range_start + found;
5275         if (found_end < range_start)
5276                 found_end = (u64)-1;
5277
5278         /*
5279          * we didn't find anything useful, return
5280          * the original results from get_extent()
5281          */
5282         if (range_start > end || found_end <= start) {
5283                 em = hole_em;
5284                 hole_em = NULL;
5285                 goto out;
5286         }
5287
5288         /* adjust the range_start to make sure it doesn't
5289          * go backwards from the start they passed in
5290          */
5291         range_start = max(start,range_start);
5292         found = found_end - range_start;
5293
5294         if (found > 0) {
5295                 u64 hole_start = start;
5296                 u64 hole_len = len;
5297
5298                 em = alloc_extent_map();
5299                 if (!em) {
5300                         err = -ENOMEM;
5301                         goto out;
5302                 }
5303                 /*
5304                  * when btrfs_get_extent can't find anything it
5305                  * returns one huge hole
5306                  *
5307                  * make sure what it found really fits our range, and
5308                  * adjust to make sure it is based on the start from
5309                  * the caller
5310                  */
5311                 if (hole_em) {
5312                         u64 calc_end = extent_map_end(hole_em);
5313
5314                         if (calc_end <= start || (hole_em->start > end)) {
5315                                 free_extent_map(hole_em);
5316                                 hole_em = NULL;
5317                         } else {
5318                                 hole_start = max(hole_em->start, start);
5319                                 hole_len = calc_end - hole_start;
5320                         }
5321                 }
5322                 em->bdev = NULL;
5323                 if (hole_em && range_start > hole_start) {
5324                         /* our hole starts before our delalloc, so we
5325                          * have to return just the parts of the hole
5326                          * that go until  the delalloc starts
5327                          */
5328                         em->len = min(hole_len,
5329                                       range_start - hole_start);
5330                         em->start = hole_start;
5331                         em->orig_start = hole_start;
5332                         /*
5333                          * don't adjust block start at all,
5334                          * it is fixed at EXTENT_MAP_HOLE
5335                          */
5336                         em->block_start = hole_em->block_start;
5337                         em->block_len = hole_len;
5338                 } else {
5339                         em->start = range_start;
5340                         em->len = found;
5341                         em->orig_start = range_start;
5342                         em->block_start = EXTENT_MAP_DELALLOC;
5343                         em->block_len = found;
5344                 }
5345         } else if (hole_em) {
5346                 return hole_em;
5347         }
5348 out:
5349
5350         free_extent_map(hole_em);
5351         if (err) {
5352                 free_extent_map(em);
5353                 return ERR_PTR(err);
5354         }
5355         return em;
5356 }
5357
5358 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5359                                                   struct extent_map *em,
5360                                                   u64 start, u64 len)
5361 {
5362         struct btrfs_root *root = BTRFS_I(inode)->root;
5363         struct btrfs_trans_handle *trans;
5364         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5365         struct btrfs_key ins;
5366         u64 alloc_hint;
5367         int ret;
5368         bool insert = false;
5369
5370         /*
5371          * Ok if the extent map we looked up is a hole and is for the exact
5372          * range we want, there is no reason to allocate a new one, however if
5373          * it is not right then we need to free this one and drop the cache for
5374          * our range.
5375          */
5376         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5377             em->len != len) {
5378                 free_extent_map(em);
5379                 em = NULL;
5380                 insert = true;
5381                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5382         }
5383
5384         trans = btrfs_join_transaction(root);
5385         if (IS_ERR(trans))
5386                 return ERR_CAST(trans);
5387
5388         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5389                 btrfs_add_inode_defrag(trans, inode);
5390
5391         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5392
5393         alloc_hint = get_extent_allocation_hint(inode, start, len);
5394         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5395                                    alloc_hint, (u64)-1, &ins, 1);
5396         if (ret) {
5397                 em = ERR_PTR(ret);
5398                 goto out;
5399         }
5400
5401         if (!em) {
5402                 em = alloc_extent_map();
5403                 if (!em) {
5404                         em = ERR_PTR(-ENOMEM);
5405                         goto out;
5406                 }
5407         }
5408
5409         em->start = start;
5410         em->orig_start = em->start;
5411         em->len = ins.offset;
5412
5413         em->block_start = ins.objectid;
5414         em->block_len = ins.offset;
5415         em->bdev = root->fs_info->fs_devices->latest_bdev;
5416
5417         /*
5418          * We need to do this because if we're using the original em we searched
5419          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5420          */
5421         em->flags = 0;
5422         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5423
5424         while (insert) {
5425                 write_lock(&em_tree->lock);
5426                 ret = add_extent_mapping(em_tree, em);
5427                 write_unlock(&em_tree->lock);
5428                 if (ret != -EEXIST)
5429                         break;
5430                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5431         }
5432
5433         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5434                                            ins.offset, ins.offset, 0);
5435         if (ret) {
5436                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5437                 em = ERR_PTR(ret);
5438         }
5439 out:
5440         btrfs_end_transaction(trans, root);
5441         return em;
5442 }
5443
5444 /*
5445  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5446  * block must be cow'd
5447  */
5448 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5449                                       struct inode *inode, u64 offset, u64 len)
5450 {
5451         struct btrfs_path *path;
5452         int ret;
5453         struct extent_buffer *leaf;
5454         struct btrfs_root *root = BTRFS_I(inode)->root;
5455         struct btrfs_file_extent_item *fi;
5456         struct btrfs_key key;
5457         u64 disk_bytenr;
5458         u64 backref_offset;
5459         u64 extent_end;
5460         u64 num_bytes;
5461         int slot;
5462         int found_type;
5463
5464         path = btrfs_alloc_path();
5465         if (!path)
5466                 return -ENOMEM;
5467
5468         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5469                                        offset, 0);
5470         if (ret < 0)
5471                 goto out;
5472
5473         slot = path->slots[0];
5474         if (ret == 1) {
5475                 if (slot == 0) {
5476                         /* can't find the item, must cow */
5477                         ret = 0;
5478                         goto out;
5479                 }
5480                 slot--;
5481         }
5482         ret = 0;
5483         leaf = path->nodes[0];
5484         btrfs_item_key_to_cpu(leaf, &key, slot);
5485         if (key.objectid != btrfs_ino(inode) ||
5486             key.type != BTRFS_EXTENT_DATA_KEY) {
5487                 /* not our file or wrong item type, must cow */
5488                 goto out;
5489         }
5490
5491         if (key.offset > offset) {
5492                 /* Wrong offset, must cow */
5493                 goto out;
5494         }
5495
5496         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5497         found_type = btrfs_file_extent_type(leaf, fi);
5498         if (found_type != BTRFS_FILE_EXTENT_REG &&
5499             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5500                 /* not a regular extent, must cow */
5501                 goto out;
5502         }
5503         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5504         backref_offset = btrfs_file_extent_offset(leaf, fi);
5505
5506         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5507         if (extent_end < offset + len) {
5508                 /* extent doesn't include our full range, must cow */
5509                 goto out;
5510         }
5511
5512         if (btrfs_extent_readonly(root, disk_bytenr))
5513                 goto out;
5514
5515         /*
5516          * look for other files referencing this extent, if we
5517          * find any we must cow
5518          */
5519         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5520                                   key.offset - backref_offset, disk_bytenr))
5521                 goto out;
5522
5523         /*
5524          * adjust disk_bytenr and num_bytes to cover just the bytes
5525          * in this extent we are about to write.  If there
5526          * are any csums in that range we have to cow in order
5527          * to keep the csums correct
5528          */
5529         disk_bytenr += backref_offset;
5530         disk_bytenr += offset - key.offset;
5531         num_bytes = min(offset + len, extent_end) - offset;
5532         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5533                                 goto out;
5534         /*
5535          * all of the above have passed, it is safe to overwrite this extent
5536          * without cow
5537          */
5538         ret = 1;
5539 out:
5540         btrfs_free_path(path);
5541         return ret;
5542 }
5543
5544 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5545                                    struct buffer_head *bh_result, int create)
5546 {
5547         struct extent_map *em;
5548         struct btrfs_root *root = BTRFS_I(inode)->root;
5549         u64 start = iblock << inode->i_blkbits;
5550         u64 len = bh_result->b_size;
5551         struct btrfs_trans_handle *trans;
5552
5553         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5554         if (IS_ERR(em))
5555                 return PTR_ERR(em);
5556
5557         /*
5558          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5559          * io.  INLINE is special, and we could probably kludge it in here, but
5560          * it's still buffered so for safety lets just fall back to the generic
5561          * buffered path.
5562          *
5563          * For COMPRESSED we _have_ to read the entire extent in so we can
5564          * decompress it, so there will be buffering required no matter what we
5565          * do, so go ahead and fallback to buffered.
5566          *
5567          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5568          * to buffered IO.  Don't blame me, this is the price we pay for using
5569          * the generic code.
5570          */
5571         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5572             em->block_start == EXTENT_MAP_INLINE) {
5573                 free_extent_map(em);
5574                 return -ENOTBLK;
5575         }
5576
5577         /* Just a good old fashioned hole, return */
5578         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5579                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5580                 free_extent_map(em);
5581                 /* DIO will do one hole at a time, so just unlock a sector */
5582                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5583                               start + root->sectorsize - 1, GFP_NOFS);
5584                 return 0;
5585         }
5586
5587         /*
5588          * We don't allocate a new extent in the following cases
5589          *
5590          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5591          * existing extent.
5592          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5593          * just use the extent.
5594          *
5595          */
5596         if (!create) {
5597                 len = em->len - (start - em->start);
5598                 goto map;
5599         }
5600
5601         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5602             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5603              em->block_start != EXTENT_MAP_HOLE)) {
5604                 int type;
5605                 int ret;
5606                 u64 block_start;
5607
5608                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5609                         type = BTRFS_ORDERED_PREALLOC;
5610                 else
5611                         type = BTRFS_ORDERED_NOCOW;
5612                 len = min(len, em->len - (start - em->start));
5613                 block_start = em->block_start + (start - em->start);
5614
5615                 /*
5616                  * we're not going to log anything, but we do need
5617                  * to make sure the current transaction stays open
5618                  * while we look for nocow cross refs
5619                  */
5620                 trans = btrfs_join_transaction(root);
5621                 if (IS_ERR(trans))
5622                         goto must_cow;
5623
5624                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5625                         ret = btrfs_add_ordered_extent_dio(inode, start,
5626                                            block_start, len, len, type);
5627                         btrfs_end_transaction(trans, root);
5628                         if (ret) {
5629                                 free_extent_map(em);
5630                                 return ret;
5631                         }
5632                         goto unlock;
5633                 }
5634                 btrfs_end_transaction(trans, root);
5635         }
5636 must_cow:
5637         /*
5638          * this will cow the extent, reset the len in case we changed
5639          * it above
5640          */
5641         len = bh_result->b_size;
5642         em = btrfs_new_extent_direct(inode, em, start, len);
5643         if (IS_ERR(em))
5644                 return PTR_ERR(em);
5645         len = min(len, em->len - (start - em->start));
5646 unlock:
5647         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5648                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5649                           0, NULL, GFP_NOFS);
5650 map:
5651         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5652                 inode->i_blkbits;
5653         bh_result->b_size = len;
5654         bh_result->b_bdev = em->bdev;
5655         set_buffer_mapped(bh_result);
5656         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5657                 set_buffer_new(bh_result);
5658
5659         free_extent_map(em);
5660
5661         return 0;
5662 }
5663
5664 struct btrfs_dio_private {
5665         struct inode *inode;
5666         u64 logical_offset;
5667         u64 disk_bytenr;
5668         u64 bytes;
5669         u32 *csums;
5670         void *private;
5671
5672         /* number of bios pending for this dio */
5673         atomic_t pending_bios;
5674
5675         /* IO errors */
5676         int errors;
5677
5678         struct bio *orig_bio;
5679 };
5680
5681 static void btrfs_endio_direct_read(struct bio *bio, int err)
5682 {
5683         struct btrfs_dio_private *dip = bio->bi_private;
5684         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5685         struct bio_vec *bvec = bio->bi_io_vec;
5686         struct inode *inode = dip->inode;
5687         struct btrfs_root *root = BTRFS_I(inode)->root;
5688         u64 start;
5689         u32 *private = dip->csums;
5690
5691         start = dip->logical_offset;
5692         do {
5693                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5694                         struct page *page = bvec->bv_page;
5695                         char *kaddr;
5696                         u32 csum = ~(u32)0;
5697                         unsigned long flags;
5698
5699                         local_irq_save(flags);
5700                         kaddr = kmap_atomic(page, KM_IRQ0);
5701                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5702                                                csum, bvec->bv_len);
5703                         btrfs_csum_final(csum, (char *)&csum);
5704                         kunmap_atomic(kaddr, KM_IRQ0);
5705                         local_irq_restore(flags);
5706
5707                         flush_dcache_page(bvec->bv_page);
5708                         if (csum != *private) {
5709                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
5710                                       " %llu csum %u private %u\n",
5711                                       (unsigned long long)btrfs_ino(inode),
5712                                       (unsigned long long)start,
5713                                       csum, *private);
5714                                 err = -EIO;
5715                         }
5716                 }
5717
5718                 start += bvec->bv_len;
5719                 private++;
5720                 bvec++;
5721         } while (bvec <= bvec_end);
5722
5723         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5724                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5725         bio->bi_private = dip->private;
5726
5727         kfree(dip->csums);
5728         kfree(dip);
5729
5730         /* If we had a csum failure make sure to clear the uptodate flag */
5731         if (err)
5732                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5733         dio_end_io(bio, err);
5734 }
5735
5736 static void btrfs_endio_direct_write(struct bio *bio, int err)
5737 {
5738         struct btrfs_dio_private *dip = bio->bi_private;
5739         struct inode *inode = dip->inode;
5740         struct btrfs_root *root = BTRFS_I(inode)->root;
5741         struct btrfs_trans_handle *trans;
5742         struct btrfs_ordered_extent *ordered = NULL;
5743         struct extent_state *cached_state = NULL;
5744         u64 ordered_offset = dip->logical_offset;
5745         u64 ordered_bytes = dip->bytes;
5746         int ret;
5747
5748         if (err)
5749                 goto out_done;
5750 again:
5751         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5752                                                    &ordered_offset,
5753                                                    ordered_bytes);
5754         if (!ret)
5755                 goto out_test;
5756
5757         BUG_ON(!ordered);
5758
5759         trans = btrfs_join_transaction(root);
5760         if (IS_ERR(trans)) {
5761                 err = -ENOMEM;
5762                 goto out;
5763         }
5764         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5765
5766         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5767                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5768                 if (!ret)
5769                         ret = btrfs_update_inode(trans, root, inode);
5770                 err = ret;
5771                 goto out;
5772         }
5773
5774         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5775                          ordered->file_offset + ordered->len - 1, 0,
5776                          &cached_state, GFP_NOFS);
5777
5778         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5779                 ret = btrfs_mark_extent_written(trans, inode,
5780                                                 ordered->file_offset,
5781                                                 ordered->file_offset +
5782                                                 ordered->len);
5783                 if (ret) {
5784                         err = ret;
5785                         goto out_unlock;
5786                 }
5787         } else {
5788                 ret = insert_reserved_file_extent(trans, inode,
5789                                                   ordered->file_offset,
5790                                                   ordered->start,
5791                                                   ordered->disk_len,
5792                                                   ordered->len,
5793                                                   ordered->len,
5794                                                   0, 0, 0,
5795                                                   BTRFS_FILE_EXTENT_REG);
5796                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5797                                    ordered->file_offset, ordered->len);
5798                 if (ret) {
5799                         err = ret;
5800                         WARN_ON(1);
5801                         goto out_unlock;
5802                 }
5803         }
5804
5805         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5806         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5807         if (!ret)
5808                 btrfs_update_inode(trans, root, inode);
5809         ret = 0;
5810 out_unlock:
5811         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5812                              ordered->file_offset + ordered->len - 1,
5813                              &cached_state, GFP_NOFS);
5814 out:
5815         btrfs_delalloc_release_metadata(inode, ordered->len);
5816         btrfs_end_transaction(trans, root);
5817         ordered_offset = ordered->file_offset + ordered->len;
5818         btrfs_put_ordered_extent(ordered);
5819         btrfs_put_ordered_extent(ordered);
5820
5821 out_test:
5822         /*
5823          * our bio might span multiple ordered extents.  If we haven't
5824          * completed the accounting for the whole dio, go back and try again
5825          */
5826         if (ordered_offset < dip->logical_offset + dip->bytes) {
5827                 ordered_bytes = dip->logical_offset + dip->bytes -
5828                         ordered_offset;
5829                 goto again;
5830         }
5831 out_done:
5832         bio->bi_private = dip->private;
5833
5834         kfree(dip->csums);
5835         kfree(dip);
5836
5837         /* If we had an error make sure to clear the uptodate flag */
5838         if (err)
5839                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5840         dio_end_io(bio, err);
5841 }
5842
5843 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5844                                     struct bio *bio, int mirror_num,
5845                                     unsigned long bio_flags, u64 offset)
5846 {
5847         int ret;
5848         struct btrfs_root *root = BTRFS_I(inode)->root;
5849         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5850         BUG_ON(ret);
5851         return 0;
5852 }
5853
5854 static void btrfs_end_dio_bio(struct bio *bio, int err)
5855 {
5856         struct btrfs_dio_private *dip = bio->bi_private;
5857
5858         if (err) {
5859                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5860                       "sector %#Lx len %u err no %d\n",
5861                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5862                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5863                 dip->errors = 1;
5864
5865                 /*
5866                  * before atomic variable goto zero, we must make sure
5867                  * dip->errors is perceived to be set.
5868                  */
5869                 smp_mb__before_atomic_dec();
5870         }
5871
5872         /* if there are more bios still pending for this dio, just exit */
5873         if (!atomic_dec_and_test(&dip->pending_bios))
5874                 goto out;
5875
5876         if (dip->errors)
5877                 bio_io_error(dip->orig_bio);
5878         else {
5879                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5880                 bio_endio(dip->orig_bio, 0);
5881         }
5882 out:
5883         bio_put(bio);
5884 }
5885
5886 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5887                                        u64 first_sector, gfp_t gfp_flags)
5888 {
5889         int nr_vecs = bio_get_nr_vecs(bdev);
5890         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5891 }
5892
5893 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5894                                          int rw, u64 file_offset, int skip_sum,
5895                                          u32 *csums, int async_submit)
5896 {
5897         int write = rw & REQ_WRITE;
5898         struct btrfs_root *root = BTRFS_I(inode)->root;
5899         int ret;
5900
5901         bio_get(bio);
5902         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5903         if (ret)
5904                 goto err;
5905
5906         if (skip_sum)
5907                 goto map;
5908
5909         if (write && async_submit) {
5910                 ret = btrfs_wq_submit_bio(root->fs_info,
5911                                    inode, rw, bio, 0, 0,
5912                                    file_offset,
5913                                    __btrfs_submit_bio_start_direct_io,
5914                                    __btrfs_submit_bio_done);
5915                 goto err;
5916         } else if (write) {
5917                 /*
5918                  * If we aren't doing async submit, calculate the csum of the
5919                  * bio now.
5920                  */
5921                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5922                 if (ret)
5923                         goto err;
5924         } else if (!skip_sum) {
5925                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5926                                           file_offset, csums);
5927                 if (ret)
5928                         goto err;
5929         }
5930
5931 map:
5932         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5933 err:
5934         bio_put(bio);
5935         return ret;
5936 }
5937
5938 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5939                                     int skip_sum)
5940 {
5941         struct inode *inode = dip->inode;
5942         struct btrfs_root *root = BTRFS_I(inode)->root;
5943         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5944         struct bio *bio;
5945         struct bio *orig_bio = dip->orig_bio;
5946         struct bio_vec *bvec = orig_bio->bi_io_vec;
5947         u64 start_sector = orig_bio->bi_sector;
5948         u64 file_offset = dip->logical_offset;
5949         u64 submit_len = 0;
5950         u64 map_length;
5951         int nr_pages = 0;
5952         u32 *csums = dip->csums;
5953         int ret = 0;
5954         int async_submit = 0;
5955         int write = rw & REQ_WRITE;
5956
5957         map_length = orig_bio->bi_size;
5958         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5959                               &map_length, NULL, 0);
5960         if (ret) {
5961                 bio_put(orig_bio);
5962                 return -EIO;
5963         }
5964
5965         if (map_length >= orig_bio->bi_size) {
5966                 bio = orig_bio;
5967                 goto submit;
5968         }
5969
5970         async_submit = 1;
5971         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5972         if (!bio)
5973                 return -ENOMEM;
5974         bio->bi_private = dip;
5975         bio->bi_end_io = btrfs_end_dio_bio;
5976         atomic_inc(&dip->pending_bios);
5977
5978         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5979                 if (unlikely(map_length < submit_len + bvec->bv_len ||
5980                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5981                                  bvec->bv_offset) < bvec->bv_len)) {
5982                         /*
5983                          * inc the count before we submit the bio so
5984                          * we know the end IO handler won't happen before
5985                          * we inc the count. Otherwise, the dip might get freed
5986                          * before we're done setting it up
5987                          */
5988                         atomic_inc(&dip->pending_bios);
5989                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
5990                                                      file_offset, skip_sum,
5991                                                      csums, async_submit);
5992                         if (ret) {
5993                                 bio_put(bio);
5994                                 atomic_dec(&dip->pending_bios);
5995                                 goto out_err;
5996                         }
5997
5998                         /* Write's use the ordered csums */
5999                         if (!write && !skip_sum)
6000                                 csums = csums + nr_pages;
6001                         start_sector += submit_len >> 9;
6002                         file_offset += submit_len;
6003
6004                         submit_len = 0;
6005                         nr_pages = 0;
6006
6007                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6008                                                   start_sector, GFP_NOFS);
6009                         if (!bio)
6010                                 goto out_err;
6011                         bio->bi_private = dip;
6012                         bio->bi_end_io = btrfs_end_dio_bio;
6013
6014                         map_length = orig_bio->bi_size;
6015                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6016                                               &map_length, NULL, 0);
6017                         if (ret) {
6018                                 bio_put(bio);
6019                                 goto out_err;
6020                         }
6021                 } else {
6022                         submit_len += bvec->bv_len;
6023                         nr_pages ++;
6024                         bvec++;
6025                 }
6026         }
6027
6028 submit:
6029         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6030                                      csums, async_submit);
6031         if (!ret)
6032                 return 0;
6033
6034         bio_put(bio);
6035 out_err:
6036         dip->errors = 1;
6037         /*
6038          * before atomic variable goto zero, we must
6039          * make sure dip->errors is perceived to be set.
6040          */
6041         smp_mb__before_atomic_dec();
6042         if (atomic_dec_and_test(&dip->pending_bios))
6043                 bio_io_error(dip->orig_bio);
6044
6045         /* bio_end_io() will handle error, so we needn't return it */
6046         return 0;
6047 }
6048
6049 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6050                                 loff_t file_offset)
6051 {
6052         struct btrfs_root *root = BTRFS_I(inode)->root;
6053         struct btrfs_dio_private *dip;
6054         struct bio_vec *bvec = bio->bi_io_vec;
6055         int skip_sum;
6056         int write = rw & REQ_WRITE;
6057         int ret = 0;
6058
6059         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6060
6061         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6062         if (!dip) {
6063                 ret = -ENOMEM;
6064                 goto free_ordered;
6065         }
6066         dip->csums = NULL;
6067
6068         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6069         if (!write && !skip_sum) {
6070                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6071                 if (!dip->csums) {
6072                         kfree(dip);
6073                         ret = -ENOMEM;
6074                         goto free_ordered;
6075                 }
6076         }
6077
6078         dip->private = bio->bi_private;
6079         dip->inode = inode;
6080         dip->logical_offset = file_offset;
6081
6082         dip->bytes = 0;
6083         do {
6084                 dip->bytes += bvec->bv_len;
6085                 bvec++;
6086         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6087
6088         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6089         bio->bi_private = dip;
6090         dip->errors = 0;
6091         dip->orig_bio = bio;
6092         atomic_set(&dip->pending_bios, 0);
6093
6094         if (write)
6095                 bio->bi_end_io = btrfs_endio_direct_write;
6096         else
6097                 bio->bi_end_io = btrfs_endio_direct_read;
6098
6099         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6100         if (!ret)
6101                 return;
6102 free_ordered:
6103         /*
6104          * If this is a write, we need to clean up the reserved space and kill
6105          * the ordered extent.
6106          */
6107         if (write) {
6108                 struct btrfs_ordered_extent *ordered;
6109                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6110                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6111                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6112                         btrfs_free_reserved_extent(root, ordered->start,
6113                                                    ordered->disk_len);
6114                 btrfs_put_ordered_extent(ordered);
6115                 btrfs_put_ordered_extent(ordered);
6116         }
6117         bio_endio(bio, ret);
6118 }
6119
6120 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6121                         const struct iovec *iov, loff_t offset,
6122                         unsigned long nr_segs)
6123 {
6124         int seg;
6125         int i;
6126         size_t size;
6127         unsigned long addr;
6128         unsigned blocksize_mask = root->sectorsize - 1;
6129         ssize_t retval = -EINVAL;
6130         loff_t end = offset;
6131
6132         if (offset & blocksize_mask)
6133                 goto out;
6134
6135         /* Check the memory alignment.  Blocks cannot straddle pages */
6136         for (seg = 0; seg < nr_segs; seg++) {
6137                 addr = (unsigned long)iov[seg].iov_base;
6138                 size = iov[seg].iov_len;
6139                 end += size;
6140                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6141                         goto out;
6142
6143                 /* If this is a write we don't need to check anymore */
6144                 if (rw & WRITE)
6145                         continue;
6146
6147                 /*
6148                  * Check to make sure we don't have duplicate iov_base's in this
6149                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6150                  * when reading back.
6151                  */
6152                 for (i = seg + 1; i < nr_segs; i++) {
6153                         if (iov[seg].iov_base == iov[i].iov_base)
6154                                 goto out;
6155                 }
6156         }
6157         retval = 0;
6158 out:
6159         return retval;
6160 }
6161 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6162                         const struct iovec *iov, loff_t offset,
6163                         unsigned long nr_segs)
6164 {
6165         struct file *file = iocb->ki_filp;
6166         struct inode *inode = file->f_mapping->host;
6167         struct btrfs_ordered_extent *ordered;
6168         struct extent_state *cached_state = NULL;
6169         u64 lockstart, lockend;
6170         ssize_t ret;
6171         int writing = rw & WRITE;
6172         int write_bits = 0;
6173         size_t count = iov_length(iov, nr_segs);
6174
6175         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6176                             offset, nr_segs)) {
6177                 return 0;
6178         }
6179
6180         lockstart = offset;
6181         lockend = offset + count - 1;
6182
6183         if (writing) {
6184                 ret = btrfs_delalloc_reserve_space(inode, count);
6185                 if (ret)
6186                         goto out;
6187         }
6188
6189         while (1) {
6190                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6191                                  0, &cached_state, GFP_NOFS);
6192                 /*
6193                  * We're concerned with the entire range that we're going to be
6194                  * doing DIO to, so we need to make sure theres no ordered
6195                  * extents in this range.
6196                  */
6197                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6198                                                      lockend - lockstart + 1);
6199                 if (!ordered)
6200                         break;
6201                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6202                                      &cached_state, GFP_NOFS);
6203                 btrfs_start_ordered_extent(inode, ordered, 1);
6204                 btrfs_put_ordered_extent(ordered);
6205                 cond_resched();
6206         }
6207
6208         /*
6209          * we don't use btrfs_set_extent_delalloc because we don't want
6210          * the dirty or uptodate bits
6211          */
6212         if (writing) {
6213                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6214                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6215                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6216                                      GFP_NOFS);
6217                 if (ret) {
6218                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6219                                          lockend, EXTENT_LOCKED | write_bits,
6220                                          1, 0, &cached_state, GFP_NOFS);
6221                         goto out;
6222                 }
6223         }
6224
6225         free_extent_state(cached_state);
6226         cached_state = NULL;
6227
6228         ret = __blockdev_direct_IO(rw, iocb, inode,
6229                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6230                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6231                    btrfs_submit_direct, 0);
6232
6233         if (ret < 0 && ret != -EIOCBQUEUED) {
6234                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6235                               offset + iov_length(iov, nr_segs) - 1,
6236                               EXTENT_LOCKED | write_bits, 1, 0,
6237                               &cached_state, GFP_NOFS);
6238         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6239                 /*
6240                  * We're falling back to buffered, unlock the section we didn't
6241                  * do IO on.
6242                  */
6243                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6244                               offset + iov_length(iov, nr_segs) - 1,
6245                               EXTENT_LOCKED | write_bits, 1, 0,
6246                               &cached_state, GFP_NOFS);
6247         }
6248 out:
6249         free_extent_state(cached_state);
6250         return ret;
6251 }
6252
6253 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6254                 __u64 start, __u64 len)
6255 {
6256         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6257 }
6258
6259 int btrfs_readpage(struct file *file, struct page *page)
6260 {
6261         struct extent_io_tree *tree;
6262         tree = &BTRFS_I(page->mapping->host)->io_tree;
6263         return extent_read_full_page(tree, page, btrfs_get_extent);
6264 }
6265
6266 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6267 {
6268         struct extent_io_tree *tree;
6269
6270
6271         if (current->flags & PF_MEMALLOC) {
6272                 redirty_page_for_writepage(wbc, page);
6273                 unlock_page(page);
6274                 return 0;
6275         }
6276         tree = &BTRFS_I(page->mapping->host)->io_tree;
6277         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6278 }
6279
6280 int btrfs_writepages(struct address_space *mapping,
6281                      struct writeback_control *wbc)
6282 {
6283         struct extent_io_tree *tree;
6284
6285         tree = &BTRFS_I(mapping->host)->io_tree;
6286         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6287 }
6288
6289 static int
6290 btrfs_readpages(struct file *file, struct address_space *mapping,
6291                 struct list_head *pages, unsigned nr_pages)
6292 {
6293         struct extent_io_tree *tree;
6294         tree = &BTRFS_I(mapping->host)->io_tree;
6295         return extent_readpages(tree, mapping, pages, nr_pages,
6296                                 btrfs_get_extent);
6297 }
6298 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6299 {
6300         struct extent_io_tree *tree;
6301         struct extent_map_tree *map;
6302         int ret;
6303
6304         tree = &BTRFS_I(page->mapping->host)->io_tree;
6305         map = &BTRFS_I(page->mapping->host)->extent_tree;
6306         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6307         if (ret == 1) {
6308                 ClearPagePrivate(page);
6309                 set_page_private(page, 0);
6310                 page_cache_release(page);
6311         }
6312         return ret;
6313 }
6314
6315 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6316 {
6317         if (PageWriteback(page) || PageDirty(page))
6318                 return 0;
6319         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6320 }
6321
6322 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6323 {
6324         struct extent_io_tree *tree;
6325         struct btrfs_ordered_extent *ordered;
6326         struct extent_state *cached_state = NULL;
6327         u64 page_start = page_offset(page);
6328         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6329
6330
6331         /*
6332          * we have the page locked, so new writeback can't start,
6333          * and the dirty bit won't be cleared while we are here.
6334          *
6335          * Wait for IO on this page so that we can safely clear
6336          * the PagePrivate2 bit and do ordered accounting
6337          */
6338         wait_on_page_writeback(page);
6339
6340         tree = &BTRFS_I(page->mapping->host)->io_tree;
6341         if (offset) {
6342                 btrfs_releasepage(page, GFP_NOFS);
6343                 return;
6344         }
6345         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6346                          GFP_NOFS);
6347         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6348                                            page_offset(page));
6349         if (ordered) {
6350                 /*
6351                  * IO on this page will never be started, so we need
6352                  * to account for any ordered extents now
6353                  */
6354                 clear_extent_bit(tree, page_start, page_end,
6355                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6356                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6357                                  &cached_state, GFP_NOFS);
6358                 /*
6359                  * whoever cleared the private bit is responsible
6360                  * for the finish_ordered_io
6361                  */
6362                 if (TestClearPagePrivate2(page)) {
6363                         btrfs_finish_ordered_io(page->mapping->host,
6364                                                 page_start, page_end);
6365                 }
6366                 btrfs_put_ordered_extent(ordered);
6367                 cached_state = NULL;
6368                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6369                                  GFP_NOFS);
6370         }
6371         clear_extent_bit(tree, page_start, page_end,
6372                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6373                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6374         __btrfs_releasepage(page, GFP_NOFS);
6375
6376         ClearPageChecked(page);
6377         if (PagePrivate(page)) {
6378                 ClearPagePrivate(page);
6379                 set_page_private(page, 0);
6380                 page_cache_release(page);
6381         }
6382 }
6383
6384 /*
6385  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6386  * called from a page fault handler when a page is first dirtied. Hence we must
6387  * be careful to check for EOF conditions here. We set the page up correctly
6388  * for a written page which means we get ENOSPC checking when writing into
6389  * holes and correct delalloc and unwritten extent mapping on filesystems that
6390  * support these features.
6391  *
6392  * We are not allowed to take the i_mutex here so we have to play games to
6393  * protect against truncate races as the page could now be beyond EOF.  Because
6394  * vmtruncate() writes the inode size before removing pages, once we have the
6395  * page lock we can determine safely if the page is beyond EOF. If it is not
6396  * beyond EOF, then the page is guaranteed safe against truncation until we
6397  * unlock the page.
6398  */
6399 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6400 {
6401         struct page *page = vmf->page;
6402         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6403         struct btrfs_root *root = BTRFS_I(inode)->root;
6404         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6405         struct btrfs_ordered_extent *ordered;
6406         struct extent_state *cached_state = NULL;
6407         char *kaddr;
6408         unsigned long zero_start;
6409         loff_t size;
6410         int ret;
6411         u64 page_start;
6412         u64 page_end;
6413
6414         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6415         if (ret) {
6416                 if (ret == -ENOMEM)
6417                         ret = VM_FAULT_OOM;
6418                 else /* -ENOSPC, -EIO, etc */
6419                         ret = VM_FAULT_SIGBUS;
6420                 goto out;
6421         }
6422
6423         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6424 again:
6425         lock_page(page);
6426         size = i_size_read(inode);
6427         page_start = page_offset(page);
6428         page_end = page_start + PAGE_CACHE_SIZE - 1;
6429
6430         if ((page->mapping != inode->i_mapping) ||
6431             (page_start >= size)) {
6432                 /* page got truncated out from underneath us */
6433                 goto out_unlock;
6434         }
6435         wait_on_page_writeback(page);
6436
6437         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6438                          GFP_NOFS);
6439         set_page_extent_mapped(page);
6440
6441         /*
6442          * we can't set the delalloc bits if there are pending ordered
6443          * extents.  Drop our locks and wait for them to finish
6444          */
6445         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6446         if (ordered) {
6447                 unlock_extent_cached(io_tree, page_start, page_end,
6448                                      &cached_state, GFP_NOFS);
6449                 unlock_page(page);
6450                 btrfs_start_ordered_extent(inode, ordered, 1);
6451                 btrfs_put_ordered_extent(ordered);
6452                 goto again;
6453         }
6454
6455         /*
6456          * XXX - page_mkwrite gets called every time the page is dirtied, even
6457          * if it was already dirty, so for space accounting reasons we need to
6458          * clear any delalloc bits for the range we are fixing to save.  There
6459          * is probably a better way to do this, but for now keep consistent with
6460          * prepare_pages in the normal write path.
6461          */
6462         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6463                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6464                           0, 0, &cached_state, GFP_NOFS);
6465
6466         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6467                                         &cached_state);
6468         if (ret) {
6469                 unlock_extent_cached(io_tree, page_start, page_end,
6470                                      &cached_state, GFP_NOFS);
6471                 ret = VM_FAULT_SIGBUS;
6472                 goto out_unlock;
6473         }
6474         ret = 0;
6475
6476         /* page is wholly or partially inside EOF */
6477         if (page_start + PAGE_CACHE_SIZE > size)
6478                 zero_start = size & ~PAGE_CACHE_MASK;
6479         else
6480                 zero_start = PAGE_CACHE_SIZE;
6481
6482         if (zero_start != PAGE_CACHE_SIZE) {
6483                 kaddr = kmap(page);
6484                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6485                 flush_dcache_page(page);
6486                 kunmap(page);
6487         }
6488         ClearPageChecked(page);
6489         set_page_dirty(page);
6490         SetPageUptodate(page);
6491
6492         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6493         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6494
6495         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6496
6497 out_unlock:
6498         if (!ret)
6499                 return VM_FAULT_LOCKED;
6500         unlock_page(page);
6501         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6502 out:
6503         return ret;
6504 }
6505
6506 static int btrfs_truncate(struct inode *inode)
6507 {
6508         struct btrfs_root *root = BTRFS_I(inode)->root;
6509         struct btrfs_block_rsv *rsv;
6510         int ret;
6511         int err = 0;
6512         struct btrfs_trans_handle *trans;
6513         unsigned long nr;
6514         u64 mask = root->sectorsize - 1;
6515
6516         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6517         if (ret)
6518                 return ret;
6519
6520         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6521         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6522
6523         /*
6524          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6525          * 3 things going on here
6526          *
6527          * 1) We need to reserve space for our orphan item and the space to
6528          * delete our orphan item.  Lord knows we don't want to have a dangling
6529          * orphan item because we didn't reserve space to remove it.
6530          *
6531          * 2) We need to reserve space to update our inode.
6532          *
6533          * 3) We need to have something to cache all the space that is going to
6534          * be free'd up by the truncate operation, but also have some slack
6535          * space reserved in case it uses space during the truncate (thank you
6536          * very much snapshotting).
6537          *
6538          * And we need these to all be seperate.  The fact is we can use alot of
6539          * space doing the truncate, and we have no earthly idea how much space
6540          * we will use, so we need the truncate reservation to be seperate so it
6541          * doesn't end up using space reserved for updating the inode or
6542          * removing the orphan item.  We also need to be able to stop the
6543          * transaction and start a new one, which means we need to be able to
6544          * update the inode several times, and we have no idea of knowing how
6545          * many times that will be, so we can't just reserve 1 item for the
6546          * entirety of the opration, so that has to be done seperately as well.
6547          * Then there is the orphan item, which does indeed need to be held on
6548          * to for the whole operation, and we need nobody to touch this reserved
6549          * space except the orphan code.
6550          *
6551          * So that leaves us with
6552          *
6553          * 1) root->orphan_block_rsv - for the orphan deletion.
6554          * 2) rsv - for the truncate reservation, which we will steal from the
6555          * transaction reservation.
6556          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6557          * updating the inode.
6558          */
6559         rsv = btrfs_alloc_block_rsv(root);
6560         if (!rsv)
6561                 return -ENOMEM;
6562         btrfs_add_durable_block_rsv(root->fs_info, rsv);
6563
6564         trans = btrfs_start_transaction(root, 4);
6565         if (IS_ERR(trans)) {
6566                 err = PTR_ERR(trans);
6567                 goto out;
6568         }
6569
6570         /*
6571          * Reserve space for the truncate process.  Truncate should be adding
6572          * space, but if there are snapshots it may end up using space.
6573          */
6574         ret = btrfs_truncate_reserve_metadata(trans, root, rsv);
6575         BUG_ON(ret);
6576
6577         ret = btrfs_orphan_add(trans, inode);
6578         if (ret) {
6579                 btrfs_end_transaction(trans, root);
6580                 goto out;
6581         }
6582
6583         nr = trans->blocks_used;
6584         btrfs_end_transaction(trans, root);
6585         btrfs_btree_balance_dirty(root, nr);
6586
6587         /*
6588          * Ok so we've already migrated our bytes over for the truncate, so here
6589          * just reserve the one slot we need for updating the inode.
6590          */
6591         trans = btrfs_start_transaction(root, 1);
6592         if (IS_ERR(trans)) {
6593                 err = PTR_ERR(trans);
6594                 goto out;
6595         }
6596         trans->block_rsv = rsv;
6597
6598         /*
6599          * setattr is responsible for setting the ordered_data_close flag,
6600          * but that is only tested during the last file release.  That
6601          * could happen well after the next commit, leaving a great big
6602          * window where new writes may get lost if someone chooses to write
6603          * to this file after truncating to zero
6604          *
6605          * The inode doesn't have any dirty data here, and so if we commit
6606          * this is a noop.  If someone immediately starts writing to the inode
6607          * it is very likely we'll catch some of their writes in this
6608          * transaction, and the commit will find this file on the ordered
6609          * data list with good things to send down.
6610          *
6611          * This is a best effort solution, there is still a window where
6612          * using truncate to replace the contents of the file will
6613          * end up with a zero length file after a crash.
6614          */
6615         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6616                 btrfs_add_ordered_operation(trans, root, inode);
6617
6618         while (1) {
6619                 if (!trans) {
6620                         trans = btrfs_start_transaction(root, 3);
6621                         if (IS_ERR(trans)) {
6622                                 err = PTR_ERR(trans);
6623                                 goto out;
6624                         }
6625
6626                         ret = btrfs_truncate_reserve_metadata(trans, root,
6627                                                               rsv);
6628                         BUG_ON(ret);
6629
6630                         trans->block_rsv = rsv;
6631                 }
6632
6633                 ret = btrfs_truncate_inode_items(trans, root, inode,
6634                                                  inode->i_size,
6635                                                  BTRFS_EXTENT_DATA_KEY);
6636                 if (ret != -EAGAIN) {
6637                         err = ret;
6638                         break;
6639                 }
6640
6641                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6642                 ret = btrfs_update_inode(trans, root, inode);
6643                 if (ret) {
6644                         err = ret;
6645                         break;
6646                 }
6647
6648                 nr = trans->blocks_used;
6649                 btrfs_end_transaction(trans, root);
6650                 trans = NULL;
6651                 btrfs_btree_balance_dirty(root, nr);
6652         }
6653
6654         if (ret == 0 && inode->i_nlink > 0) {
6655                 trans->block_rsv = root->orphan_block_rsv;
6656                 ret = btrfs_orphan_del(trans, inode);
6657                 if (ret)
6658                         err = ret;
6659         } else if (ret && inode->i_nlink > 0) {
6660                 /*
6661                  * Failed to do the truncate, remove us from the in memory
6662                  * orphan list.
6663                  */
6664                 ret = btrfs_orphan_del(NULL, inode);
6665         }
6666
6667         trans->block_rsv = &root->fs_info->trans_block_rsv;
6668         ret = btrfs_update_inode(trans, root, inode);
6669         if (ret && !err)
6670                 err = ret;
6671
6672         nr = trans->blocks_used;
6673         ret = btrfs_end_transaction_throttle(trans, root);
6674         btrfs_btree_balance_dirty(root, nr);
6675
6676 out:
6677         btrfs_free_block_rsv(root, rsv);
6678
6679         if (ret && !err)
6680                 err = ret;
6681
6682         return err;
6683 }
6684
6685 /*
6686  * create a new subvolume directory/inode (helper for the ioctl).
6687  */
6688 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6689                              struct btrfs_root *new_root, u64 new_dirid)
6690 {
6691         struct inode *inode;
6692         int err;
6693         u64 index = 0;
6694
6695         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6696                                 new_dirid, S_IFDIR | 0700, &index);
6697         if (IS_ERR(inode))
6698                 return PTR_ERR(inode);
6699         inode->i_op = &btrfs_dir_inode_operations;
6700         inode->i_fop = &btrfs_dir_file_operations;
6701
6702         inode->i_nlink = 1;
6703         btrfs_i_size_write(inode, 0);
6704
6705         err = btrfs_update_inode(trans, new_root, inode);
6706         BUG_ON(err);
6707
6708         iput(inode);
6709         return 0;
6710 }
6711
6712 /* helper function for file defrag and space balancing.  This
6713  * forces readahead on a given range of bytes in an inode
6714  */
6715 unsigned long btrfs_force_ra(struct address_space *mapping,
6716                               struct file_ra_state *ra, struct file *file,
6717                               pgoff_t offset, pgoff_t last_index)
6718 {
6719         pgoff_t req_size = last_index - offset + 1;
6720
6721         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6722         return offset + req_size;
6723 }
6724
6725 struct inode *btrfs_alloc_inode(struct super_block *sb)
6726 {
6727         struct btrfs_inode *ei;
6728         struct inode *inode;
6729
6730         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6731         if (!ei)
6732                 return NULL;
6733
6734         ei->root = NULL;
6735         ei->space_info = NULL;
6736         ei->generation = 0;
6737         ei->sequence = 0;
6738         ei->last_trans = 0;
6739         ei->last_sub_trans = 0;
6740         ei->logged_trans = 0;
6741         ei->delalloc_bytes = 0;
6742         ei->reserved_bytes = 0;
6743         ei->disk_i_size = 0;
6744         ei->flags = 0;
6745         ei->index_cnt = (u64)-1;
6746         ei->last_unlink_trans = 0;
6747
6748         spin_lock_init(&ei->lock);
6749         ei->outstanding_extents = 0;
6750         ei->reserved_extents = 0;
6751
6752         ei->ordered_data_close = 0;
6753         ei->orphan_meta_reserved = 0;
6754         ei->dummy_inode = 0;
6755         ei->in_defrag = 0;
6756         ei->force_compress = BTRFS_COMPRESS_NONE;
6757
6758         ei->delayed_node = NULL;
6759
6760         inode = &ei->vfs_inode;
6761         extent_map_tree_init(&ei->extent_tree);
6762         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6763         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6764         mutex_init(&ei->log_mutex);
6765         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6766         INIT_LIST_HEAD(&ei->i_orphan);
6767         INIT_LIST_HEAD(&ei->delalloc_inodes);
6768         INIT_LIST_HEAD(&ei->ordered_operations);
6769         RB_CLEAR_NODE(&ei->rb_node);
6770
6771         return inode;
6772 }
6773
6774 static void btrfs_i_callback(struct rcu_head *head)
6775 {
6776         struct inode *inode = container_of(head, struct inode, i_rcu);
6777         INIT_LIST_HEAD(&inode->i_dentry);
6778         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6779 }
6780
6781 void btrfs_destroy_inode(struct inode *inode)
6782 {
6783         struct btrfs_ordered_extent *ordered;
6784         struct btrfs_root *root = BTRFS_I(inode)->root;
6785
6786         WARN_ON(!list_empty(&inode->i_dentry));
6787         WARN_ON(inode->i_data.nrpages);
6788         WARN_ON(BTRFS_I(inode)->outstanding_extents);
6789         WARN_ON(BTRFS_I(inode)->reserved_extents);
6790
6791         /*
6792          * This can happen where we create an inode, but somebody else also
6793          * created the same inode and we need to destroy the one we already
6794          * created.
6795          */
6796         if (!root)
6797                 goto free;
6798
6799         /*
6800          * Make sure we're properly removed from the ordered operation
6801          * lists.
6802          */
6803         smp_mb();
6804         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6805                 spin_lock(&root->fs_info->ordered_extent_lock);
6806                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6807                 spin_unlock(&root->fs_info->ordered_extent_lock);
6808         }
6809
6810         spin_lock(&root->orphan_lock);
6811         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6812                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6813                        (unsigned long long)btrfs_ino(inode));
6814                 list_del_init(&BTRFS_I(inode)->i_orphan);
6815         }
6816         spin_unlock(&root->orphan_lock);
6817
6818         while (1) {
6819                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6820                 if (!ordered)
6821                         break;
6822                 else {
6823                         printk(KERN_ERR "btrfs found ordered "
6824                                "extent %llu %llu on inode cleanup\n",
6825                                (unsigned long long)ordered->file_offset,
6826                                (unsigned long long)ordered->len);
6827                         btrfs_remove_ordered_extent(inode, ordered);
6828                         btrfs_put_ordered_extent(ordered);
6829                         btrfs_put_ordered_extent(ordered);
6830                 }
6831         }
6832         inode_tree_del(inode);
6833         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6834 free:
6835         btrfs_remove_delayed_node(inode);
6836         call_rcu(&inode->i_rcu, btrfs_i_callback);
6837 }
6838
6839 int btrfs_drop_inode(struct inode *inode)
6840 {
6841         struct btrfs_root *root = BTRFS_I(inode)->root;
6842
6843         if (btrfs_root_refs(&root->root_item) == 0 &&
6844             !is_free_space_inode(root, inode))
6845                 return 1;
6846         else
6847                 return generic_drop_inode(inode);
6848 }
6849
6850 static void init_once(void *foo)
6851 {
6852         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6853
6854         inode_init_once(&ei->vfs_inode);
6855 }
6856
6857 void btrfs_destroy_cachep(void)
6858 {
6859         if (btrfs_inode_cachep)
6860                 kmem_cache_destroy(btrfs_inode_cachep);
6861         if (btrfs_trans_handle_cachep)
6862                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6863         if (btrfs_transaction_cachep)
6864                 kmem_cache_destroy(btrfs_transaction_cachep);
6865         if (btrfs_path_cachep)
6866                 kmem_cache_destroy(btrfs_path_cachep);
6867         if (btrfs_free_space_cachep)
6868                 kmem_cache_destroy(btrfs_free_space_cachep);
6869 }
6870
6871 int btrfs_init_cachep(void)
6872 {
6873         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6874                         sizeof(struct btrfs_inode), 0,
6875                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6876         if (!btrfs_inode_cachep)
6877                 goto fail;
6878
6879         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6880                         sizeof(struct btrfs_trans_handle), 0,
6881                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6882         if (!btrfs_trans_handle_cachep)
6883                 goto fail;
6884
6885         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6886                         sizeof(struct btrfs_transaction), 0,
6887                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6888         if (!btrfs_transaction_cachep)
6889                 goto fail;
6890
6891         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6892                         sizeof(struct btrfs_path), 0,
6893                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6894         if (!btrfs_path_cachep)
6895                 goto fail;
6896
6897         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6898                         sizeof(struct btrfs_free_space), 0,
6899                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6900         if (!btrfs_free_space_cachep)
6901                 goto fail;
6902
6903         return 0;
6904 fail:
6905         btrfs_destroy_cachep();
6906         return -ENOMEM;
6907 }
6908
6909 static int btrfs_getattr(struct vfsmount *mnt,
6910                          struct dentry *dentry, struct kstat *stat)
6911 {
6912         struct inode *inode = dentry->d_inode;
6913         generic_fillattr(inode, stat);
6914         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6915         stat->blksize = PAGE_CACHE_SIZE;
6916         stat->blocks = (inode_get_bytes(inode) +
6917                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6918         return 0;
6919 }
6920
6921 /*
6922  * If a file is moved, it will inherit the cow and compression flags of the new
6923  * directory.
6924  */
6925 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6926 {
6927         struct btrfs_inode *b_dir = BTRFS_I(dir);
6928         struct btrfs_inode *b_inode = BTRFS_I(inode);
6929
6930         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6931                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6932         else
6933                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6934
6935         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6936                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6937         else
6938                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6939 }
6940
6941 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6942                            struct inode *new_dir, struct dentry *new_dentry)
6943 {
6944         struct btrfs_trans_handle *trans;
6945         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6946         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6947         struct inode *new_inode = new_dentry->d_inode;
6948         struct inode *old_inode = old_dentry->d_inode;
6949         struct timespec ctime = CURRENT_TIME;
6950         u64 index = 0;
6951         u64 root_objectid;
6952         int ret;
6953         u64 old_ino = btrfs_ino(old_inode);
6954
6955         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6956                 return -EPERM;
6957
6958         /* we only allow rename subvolume link between subvolumes */
6959         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6960                 return -EXDEV;
6961
6962         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6963             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6964                 return -ENOTEMPTY;
6965
6966         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6967             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6968                 return -ENOTEMPTY;
6969         /*
6970          * we're using rename to replace one file with another.
6971          * and the replacement file is large.  Start IO on it now so
6972          * we don't add too much work to the end of the transaction
6973          */
6974         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6975             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6976                 filemap_flush(old_inode->i_mapping);
6977
6978         /* close the racy window with snapshot create/destroy ioctl */
6979         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6980                 down_read(&root->fs_info->subvol_sem);
6981         /*
6982          * We want to reserve the absolute worst case amount of items.  So if
6983          * both inodes are subvols and we need to unlink them then that would
6984          * require 4 item modifications, but if they are both normal inodes it
6985          * would require 5 item modifications, so we'll assume their normal
6986          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6987          * should cover the worst case number of items we'll modify.
6988          */
6989         trans = btrfs_start_transaction(root, 20);
6990         if (IS_ERR(trans)) {
6991                 ret = PTR_ERR(trans);
6992                 goto out_notrans;
6993         }
6994
6995         if (dest != root)
6996                 btrfs_record_root_in_trans(trans, dest);
6997
6998         ret = btrfs_set_inode_index(new_dir, &index);
6999         if (ret)
7000                 goto out_fail;
7001
7002         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7003                 /* force full log commit if subvolume involved. */
7004                 root->fs_info->last_trans_log_full_commit = trans->transid;
7005         } else {
7006                 ret = btrfs_insert_inode_ref(trans, dest,
7007                                              new_dentry->d_name.name,
7008                                              new_dentry->d_name.len,
7009                                              old_ino,
7010                                              btrfs_ino(new_dir), index);
7011                 if (ret)
7012                         goto out_fail;
7013                 /*
7014                  * this is an ugly little race, but the rename is required
7015                  * to make sure that if we crash, the inode is either at the
7016                  * old name or the new one.  pinning the log transaction lets
7017                  * us make sure we don't allow a log commit to come in after
7018                  * we unlink the name but before we add the new name back in.
7019                  */
7020                 btrfs_pin_log_trans(root);
7021         }
7022         /*
7023          * make sure the inode gets flushed if it is replacing
7024          * something.
7025          */
7026         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7027                 btrfs_add_ordered_operation(trans, root, old_inode);
7028
7029         old_dir->i_ctime = old_dir->i_mtime = ctime;
7030         new_dir->i_ctime = new_dir->i_mtime = ctime;
7031         old_inode->i_ctime = ctime;
7032
7033         if (old_dentry->d_parent != new_dentry->d_parent)
7034                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7035
7036         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7037                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7038                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7039                                         old_dentry->d_name.name,
7040                                         old_dentry->d_name.len);
7041         } else {
7042                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7043                                         old_dentry->d_inode,
7044                                         old_dentry->d_name.name,
7045                                         old_dentry->d_name.len);
7046                 if (!ret)
7047                         ret = btrfs_update_inode(trans, root, old_inode);
7048         }
7049         BUG_ON(ret);
7050
7051         if (new_inode) {
7052                 new_inode->i_ctime = CURRENT_TIME;
7053                 if (unlikely(btrfs_ino(new_inode) ==
7054                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7055                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7056                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7057                                                 root_objectid,
7058                                                 new_dentry->d_name.name,
7059                                                 new_dentry->d_name.len);
7060                         BUG_ON(new_inode->i_nlink == 0);
7061                 } else {
7062                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7063                                                  new_dentry->d_inode,
7064                                                  new_dentry->d_name.name,
7065                                                  new_dentry->d_name.len);
7066                 }
7067                 BUG_ON(ret);
7068                 if (new_inode->i_nlink == 0) {
7069                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7070                         BUG_ON(ret);
7071                 }
7072         }
7073
7074         fixup_inode_flags(new_dir, old_inode);
7075
7076         ret = btrfs_add_link(trans, new_dir, old_inode,
7077                              new_dentry->d_name.name,
7078                              new_dentry->d_name.len, 0, index);
7079         BUG_ON(ret);
7080
7081         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7082                 struct dentry *parent = dget_parent(new_dentry);
7083                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7084                 dput(parent);
7085                 btrfs_end_log_trans(root);
7086         }
7087 out_fail:
7088         btrfs_end_transaction_throttle(trans, root);
7089 out_notrans:
7090         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7091                 up_read(&root->fs_info->subvol_sem);
7092
7093         return ret;
7094 }
7095
7096 /*
7097  * some fairly slow code that needs optimization. This walks the list
7098  * of all the inodes with pending delalloc and forces them to disk.
7099  */
7100 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7101 {
7102         struct list_head *head = &root->fs_info->delalloc_inodes;
7103         struct btrfs_inode *binode;
7104         struct inode *inode;
7105
7106         if (root->fs_info->sb->s_flags & MS_RDONLY)
7107                 return -EROFS;
7108
7109         spin_lock(&root->fs_info->delalloc_lock);
7110         while (!list_empty(head)) {
7111                 binode = list_entry(head->next, struct btrfs_inode,
7112                                     delalloc_inodes);
7113                 inode = igrab(&binode->vfs_inode);
7114                 if (!inode)
7115                         list_del_init(&binode->delalloc_inodes);
7116                 spin_unlock(&root->fs_info->delalloc_lock);
7117                 if (inode) {
7118                         filemap_flush(inode->i_mapping);
7119                         if (delay_iput)
7120                                 btrfs_add_delayed_iput(inode);
7121                         else
7122                                 iput(inode);
7123                 }
7124                 cond_resched();
7125                 spin_lock(&root->fs_info->delalloc_lock);
7126         }
7127         spin_unlock(&root->fs_info->delalloc_lock);
7128
7129         /* the filemap_flush will queue IO into the worker threads, but
7130          * we have to make sure the IO is actually started and that
7131          * ordered extents get created before we return
7132          */
7133         atomic_inc(&root->fs_info->async_submit_draining);
7134         while (atomic_read(&root->fs_info->nr_async_submits) ||
7135               atomic_read(&root->fs_info->async_delalloc_pages)) {
7136                 wait_event(root->fs_info->async_submit_wait,
7137                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7138                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7139         }
7140         atomic_dec(&root->fs_info->async_submit_draining);
7141         return 0;
7142 }
7143
7144 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7145                          const char *symname)
7146 {
7147         struct btrfs_trans_handle *trans;
7148         struct btrfs_root *root = BTRFS_I(dir)->root;
7149         struct btrfs_path *path;
7150         struct btrfs_key key;
7151         struct inode *inode = NULL;
7152         int err;
7153         int drop_inode = 0;
7154         u64 objectid;
7155         u64 index = 0 ;
7156         int name_len;
7157         int datasize;
7158         unsigned long ptr;
7159         struct btrfs_file_extent_item *ei;
7160         struct extent_buffer *leaf;
7161         unsigned long nr = 0;
7162
7163         name_len = strlen(symname) + 1;
7164         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7165                 return -ENAMETOOLONG;
7166
7167         /*
7168          * 2 items for inode item and ref
7169          * 2 items for dir items
7170          * 1 item for xattr if selinux is on
7171          */
7172         trans = btrfs_start_transaction(root, 5);
7173         if (IS_ERR(trans))
7174                 return PTR_ERR(trans);
7175
7176         err = btrfs_find_free_ino(root, &objectid);
7177         if (err)
7178                 goto out_unlock;
7179
7180         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7181                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7182                                 S_IFLNK|S_IRWXUGO, &index);
7183         if (IS_ERR(inode)) {
7184                 err = PTR_ERR(inode);
7185                 goto out_unlock;
7186         }
7187
7188         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7189         if (err) {
7190                 drop_inode = 1;
7191                 goto out_unlock;
7192         }
7193
7194         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7195         if (err)
7196                 drop_inode = 1;
7197         else {
7198                 inode->i_mapping->a_ops = &btrfs_aops;
7199                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7200                 inode->i_fop = &btrfs_file_operations;
7201                 inode->i_op = &btrfs_file_inode_operations;
7202                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7203         }
7204         if (drop_inode)
7205                 goto out_unlock;
7206
7207         path = btrfs_alloc_path();
7208         BUG_ON(!path);
7209         key.objectid = btrfs_ino(inode);
7210         key.offset = 0;
7211         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7212         datasize = btrfs_file_extent_calc_inline_size(name_len);
7213         err = btrfs_insert_empty_item(trans, root, path, &key,
7214                                       datasize);
7215         if (err) {
7216                 drop_inode = 1;
7217                 btrfs_free_path(path);
7218                 goto out_unlock;
7219         }
7220         leaf = path->nodes[0];
7221         ei = btrfs_item_ptr(leaf, path->slots[0],
7222                             struct btrfs_file_extent_item);
7223         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7224         btrfs_set_file_extent_type(leaf, ei,
7225                                    BTRFS_FILE_EXTENT_INLINE);
7226         btrfs_set_file_extent_encryption(leaf, ei, 0);
7227         btrfs_set_file_extent_compression(leaf, ei, 0);
7228         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7229         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7230
7231         ptr = btrfs_file_extent_inline_start(ei);
7232         write_extent_buffer(leaf, symname, ptr, name_len);
7233         btrfs_mark_buffer_dirty(leaf);
7234         btrfs_free_path(path);
7235
7236         inode->i_op = &btrfs_symlink_inode_operations;
7237         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7238         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7239         inode_set_bytes(inode, name_len);
7240         btrfs_i_size_write(inode, name_len - 1);
7241         err = btrfs_update_inode(trans, root, inode);
7242         if (err)
7243                 drop_inode = 1;
7244
7245 out_unlock:
7246         nr = trans->blocks_used;
7247         btrfs_end_transaction_throttle(trans, root);
7248         if (drop_inode) {
7249                 inode_dec_link_count(inode);
7250                 iput(inode);
7251         }
7252         btrfs_btree_balance_dirty(root, nr);
7253         return err;
7254 }
7255
7256 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7257                                        u64 start, u64 num_bytes, u64 min_size,
7258                                        loff_t actual_len, u64 *alloc_hint,
7259                                        struct btrfs_trans_handle *trans)
7260 {
7261         struct btrfs_root *root = BTRFS_I(inode)->root;
7262         struct btrfs_key ins;
7263         u64 cur_offset = start;
7264         u64 i_size;
7265         int ret = 0;
7266         bool own_trans = true;
7267
7268         if (trans)
7269                 own_trans = false;
7270         while (num_bytes > 0) {
7271                 if (own_trans) {
7272                         trans = btrfs_start_transaction(root, 3);
7273                         if (IS_ERR(trans)) {
7274                                 ret = PTR_ERR(trans);
7275                                 break;
7276                         }
7277                 }
7278
7279                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7280                                            0, *alloc_hint, (u64)-1, &ins, 1);
7281                 if (ret) {
7282                         if (own_trans)
7283                                 btrfs_end_transaction(trans, root);
7284                         break;
7285                 }
7286
7287                 ret = insert_reserved_file_extent(trans, inode,
7288                                                   cur_offset, ins.objectid,
7289                                                   ins.offset, ins.offset,
7290                                                   ins.offset, 0, 0, 0,
7291                                                   BTRFS_FILE_EXTENT_PREALLOC);
7292                 BUG_ON(ret);
7293                 btrfs_drop_extent_cache(inode, cur_offset,
7294                                         cur_offset + ins.offset -1, 0);
7295
7296                 num_bytes -= ins.offset;
7297                 cur_offset += ins.offset;
7298                 *alloc_hint = ins.objectid + ins.offset;
7299
7300                 inode->i_ctime = CURRENT_TIME;
7301                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7302                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7303                     (actual_len > inode->i_size) &&
7304                     (cur_offset > inode->i_size)) {
7305                         if (cur_offset > actual_len)
7306                                 i_size = actual_len;
7307                         else
7308                                 i_size = cur_offset;
7309                         i_size_write(inode, i_size);
7310                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7311                 }
7312
7313                 ret = btrfs_update_inode(trans, root, inode);
7314                 BUG_ON(ret);
7315
7316                 if (own_trans)
7317                         btrfs_end_transaction(trans, root);
7318         }
7319         return ret;
7320 }
7321
7322 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7323                               u64 start, u64 num_bytes, u64 min_size,
7324                               loff_t actual_len, u64 *alloc_hint)
7325 {
7326         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7327                                            min_size, actual_len, alloc_hint,
7328                                            NULL);
7329 }
7330
7331 int btrfs_prealloc_file_range_trans(struct inode *inode,
7332                                     struct btrfs_trans_handle *trans, int mode,
7333                                     u64 start, u64 num_bytes, u64 min_size,
7334                                     loff_t actual_len, u64 *alloc_hint)
7335 {
7336         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7337                                            min_size, actual_len, alloc_hint, trans);
7338 }
7339
7340 static int btrfs_set_page_dirty(struct page *page)
7341 {
7342         return __set_page_dirty_nobuffers(page);
7343 }
7344
7345 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7346 {
7347         struct btrfs_root *root = BTRFS_I(inode)->root;
7348
7349         if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7350                 return -EROFS;
7351         if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7352                 return -EACCES;
7353         return generic_permission(inode, mask, flags, btrfs_check_acl);
7354 }
7355
7356 static const struct inode_operations btrfs_dir_inode_operations = {
7357         .getattr        = btrfs_getattr,
7358         .lookup         = btrfs_lookup,
7359         .create         = btrfs_create,
7360         .unlink         = btrfs_unlink,
7361         .link           = btrfs_link,
7362         .mkdir          = btrfs_mkdir,
7363         .rmdir          = btrfs_rmdir,
7364         .rename         = btrfs_rename,
7365         .symlink        = btrfs_symlink,
7366         .setattr        = btrfs_setattr,
7367         .mknod          = btrfs_mknod,
7368         .setxattr       = btrfs_setxattr,
7369         .getxattr       = btrfs_getxattr,
7370         .listxattr      = btrfs_listxattr,
7371         .removexattr    = btrfs_removexattr,
7372         .permission     = btrfs_permission,
7373 };
7374 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7375         .lookup         = btrfs_lookup,
7376         .permission     = btrfs_permission,
7377 };
7378
7379 static const struct file_operations btrfs_dir_file_operations = {
7380         .llseek         = generic_file_llseek,
7381         .read           = generic_read_dir,
7382         .readdir        = btrfs_real_readdir,
7383         .unlocked_ioctl = btrfs_ioctl,
7384 #ifdef CONFIG_COMPAT
7385         .compat_ioctl   = btrfs_ioctl,
7386 #endif
7387         .release        = btrfs_release_file,
7388         .fsync          = btrfs_sync_file,
7389 };
7390
7391 static struct extent_io_ops btrfs_extent_io_ops = {
7392         .fill_delalloc = run_delalloc_range,
7393         .submit_bio_hook = btrfs_submit_bio_hook,
7394         .merge_bio_hook = btrfs_merge_bio_hook,
7395         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7396         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7397         .writepage_start_hook = btrfs_writepage_start_hook,
7398         .readpage_io_failed_hook = btrfs_io_failed_hook,
7399         .set_bit_hook = btrfs_set_bit_hook,
7400         .clear_bit_hook = btrfs_clear_bit_hook,
7401         .merge_extent_hook = btrfs_merge_extent_hook,
7402         .split_extent_hook = btrfs_split_extent_hook,
7403 };
7404
7405 /*
7406  * btrfs doesn't support the bmap operation because swapfiles
7407  * use bmap to make a mapping of extents in the file.  They assume
7408  * these extents won't change over the life of the file and they
7409  * use the bmap result to do IO directly to the drive.
7410  *
7411  * the btrfs bmap call would return logical addresses that aren't
7412  * suitable for IO and they also will change frequently as COW
7413  * operations happen.  So, swapfile + btrfs == corruption.
7414  *
7415  * For now we're avoiding this by dropping bmap.
7416  */
7417 static const struct address_space_operations btrfs_aops = {
7418         .readpage       = btrfs_readpage,
7419         .writepage      = btrfs_writepage,
7420         .writepages     = btrfs_writepages,
7421         .readpages      = btrfs_readpages,
7422         .direct_IO      = btrfs_direct_IO,
7423         .invalidatepage = btrfs_invalidatepage,
7424         .releasepage    = btrfs_releasepage,
7425         .set_page_dirty = btrfs_set_page_dirty,
7426         .error_remove_page = generic_error_remove_page,
7427 };
7428
7429 static const struct address_space_operations btrfs_symlink_aops = {
7430         .readpage       = btrfs_readpage,
7431         .writepage      = btrfs_writepage,
7432         .invalidatepage = btrfs_invalidatepage,
7433         .releasepage    = btrfs_releasepage,
7434 };
7435
7436 static const struct inode_operations btrfs_file_inode_operations = {
7437         .getattr        = btrfs_getattr,
7438         .setattr        = btrfs_setattr,
7439         .setxattr       = btrfs_setxattr,
7440         .getxattr       = btrfs_getxattr,
7441         .listxattr      = btrfs_listxattr,
7442         .removexattr    = btrfs_removexattr,
7443         .permission     = btrfs_permission,
7444         .fiemap         = btrfs_fiemap,
7445 };
7446 static const struct inode_operations btrfs_special_inode_operations = {
7447         .getattr        = btrfs_getattr,
7448         .setattr        = btrfs_setattr,
7449         .permission     = btrfs_permission,
7450         .setxattr       = btrfs_setxattr,
7451         .getxattr       = btrfs_getxattr,
7452         .listxattr      = btrfs_listxattr,
7453         .removexattr    = btrfs_removexattr,
7454 };
7455 static const struct inode_operations btrfs_symlink_inode_operations = {
7456         .readlink       = generic_readlink,
7457         .follow_link    = page_follow_link_light,
7458         .put_link       = page_put_link,
7459         .getattr        = btrfs_getattr,
7460         .permission     = btrfs_permission,
7461         .setxattr       = btrfs_setxattr,
7462         .getxattr       = btrfs_getxattr,
7463         .listxattr      = btrfs_listxattr,
7464         .removexattr    = btrfs_removexattr,
7465 };
7466
7467 const struct dentry_operations btrfs_dentry_operations = {
7468         .d_delete       = btrfs_dentry_delete,
7469 };