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 "compat.h"
  40 #include "ctree.h"
  41 #include "disk-io.h"
  42 #include "transaction.h"
  43 #include "btrfs_inode.h"
  44 #include "ioctl.h"
  45 #include "print-tree.h"
  46 #include "volumes.h"
  47 #include "ordered-data.h"
  48 #include "xattr.h"
  49 #include "tree-log.h"
  50 #include "compression.h"
  51 #include "locking.h"
  52
  53 struct btrfs_iget_args {
  54         u64 ino;
  55         struct btrfs_root *root;
  56 };
  57
  58 static struct inode_operations btrfs_dir_inode_operations;
  59 static struct inode_operations btrfs_symlink_inode_operations;
  60 static struct inode_operations btrfs_dir_ro_inode_operations;
  61 static struct inode_operations btrfs_special_inode_operations;
  62 static struct inode_operations btrfs_file_inode_operations;
  63 static struct address_space_operations btrfs_aops;
  64 static struct address_space_operations btrfs_symlink_aops;
  65 static struct file_operations btrfs_dir_file_operations;
  66 static struct extent_io_ops btrfs_extent_io_ops;
  67
  68 static struct kmem_cache *btrfs_inode_cachep;
  69 struct kmem_cache *btrfs_trans_handle_cachep;
  70 struct kmem_cache *btrfs_transaction_cachep;
  71 struct kmem_cache *btrfs_path_cachep;
  72
  73 #define S_SHIFT 12
  74 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  75         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  76         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  77         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  78         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  79         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  80         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  81         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  82 };
  83
  84 static void btrfs_truncate(struct inode *inode);
  85 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  86 static noinline int cow_file_range(struct inode *inode,
  87                                    struct page *locked_page,
  88                                    u64 start, u64 end, int *page_started,
  89                                    unsigned long *nr_written, int unlock);
  90
  91 static int btrfs_init_inode_security(struct inode *inode,  struct inode *dir)
  92 {
  93         int err;
  94
  95         err = btrfs_init_acl(inode, dir);
  96         if (!err)
  97                 err = btrfs_xattr_security_init(inode, dir);
  98         return err;
  99 }
 100
 101 /*
 102  * this does all the hard work for inserting an inline extent into
 103  * the btree.  The caller should have done a btrfs_drop_extents so that
 104  * no overlapping inline items exist in the btree
 105  */
 106 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 107                                 struct btrfs_root *root, struct inode *inode,
 108                                 u64 start, size_t size, size_t compressed_size,
 109                                 struct page **compressed_pages)
 110 {
 111         struct btrfs_key key;
 112         struct btrfs_path *path;
 113         struct extent_buffer *leaf;
 114         struct page *page = NULL;
 115         char *kaddr;
 116         unsigned long ptr;
 117         struct btrfs_file_extent_item *ei;
 118         int err = 0;
 119         int ret;
 120         size_t cur_size = size;
 121         size_t datasize;
 122         unsigned long offset;
 123         int use_compress = 0;
 124
 125         if (compressed_size && compressed_pages) {
 126                 use_compress = 1;
 127                 cur_size = compressed_size;
 128         }
 129
 130         path = btrfs_alloc_path();
 131         if (!path)
 132                 return -ENOMEM;
 133
 134         path->leave_spinning = 1;
 135         btrfs_set_trans_block_group(trans, inode);
 136
 137         key.objectid = inode->i_ino;
 138         key.offset = start;
 139         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 140         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 141
 142         inode_add_bytes(inode, size);
 143         ret = btrfs_insert_empty_item(trans, root, path, &key,
 144                                       datasize);
 145         BUG_ON(ret);
 146         if (ret) {
 147                 err = ret;
 148                 goto fail;
 149         }
 150         leaf = path->nodes[0];
 151         ei = btrfs_item_ptr(leaf, path->slots[0],
 152                             struct btrfs_file_extent_item);
 153         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 154         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 155         btrfs_set_file_extent_encryption(leaf, ei, 0);
 156         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 157         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 158         ptr = btrfs_file_extent_inline_start(ei);
 159
 160         if (use_compress) {
 161                 struct page *cpage;
 162                 int i = 0;
 163                 while (compressed_size > 0) {
 164                         cpage = compressed_pages[i];
 165                         cur_size = min_t(unsigned long, compressed_size,
 166                                        PAGE_CACHE_SIZE);
 167
 168                         kaddr = kmap_atomic(cpage, KM_USER0);
 169                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 170                         kunmap_atomic(kaddr, KM_USER0);
 171
 172                         i++;
 173                         ptr += cur_size;
 174                         compressed_size -= cur_size;
 175                 }
 176                 btrfs_set_file_extent_compression(leaf, ei,
 177                                                   BTRFS_COMPRESS_ZLIB);
 178         } else {
 179                 page = find_get_page(inode->i_mapping,
 180                                      start >> PAGE_CACHE_SHIFT);
 181                 btrfs_set_file_extent_compression(leaf, ei, 0);
 182                 kaddr = kmap_atomic(page, KM_USER0);
 183                 offset = start & (PAGE_CACHE_SIZE - 1);
 184                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 185                 kunmap_atomic(kaddr, KM_USER0);
 186                 page_cache_release(page);
 187         }
 188         btrfs_mark_buffer_dirty(leaf);
 189         btrfs_free_path(path);
 190
 191         BTRFS_I(inode)->disk_i_size = inode->i_size;
 192         btrfs_update_inode(trans, root, inode);
 193         return 0;
 194 fail:
 195         btrfs_free_path(path);
 196         return err;
 197 }
 198
 199
 200 /*
 201  * conditionally insert an inline extent into the file.  This
 202  * does the checks required to make sure the data is small enough
 203  * to fit as an inline extent.
 204  */
 205 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 206                                  struct btrfs_root *root,
 207                                  struct inode *inode, u64 start, u64 end,
 208                                  size_t compressed_size,
 209                                  struct page **compressed_pages)
 210 {
 211         u64 isize = i_size_read(inode);
 212         u64 actual_end = min(end + 1, isize);
 213         u64 inline_len = actual_end - start;
 214         u64 aligned_end = (end + root->sectorsize - 1) &
 215                         ~((u64)root->sectorsize - 1);
 216         u64 hint_byte;
 217         u64 data_len = inline_len;
 218         int ret;
 219
 220         if (compressed_size)
 221                 data_len = compressed_size;
 222
 223         if (start > 0 ||
 224             actual_end >= PAGE_CACHE_SIZE ||
 225             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 226             (!compressed_size &&
 227             (actual_end & (root->sectorsize - 1)) == 0) ||
 228             end + 1 < isize ||
 229             data_len > root->fs_info->max_inline) {
 230                 return 1;
 231         }
 232
 233         ret = btrfs_drop_extents(trans, root, inode, start,
 234                                  aligned_end, aligned_end, start,
 235                                  &hint_byte, 1);
 236         BUG_ON(ret);
 237
 238         if (isize > actual_end)
 239                 inline_len = min_t(u64, isize, actual_end);
 240         ret = insert_inline_extent(trans, root, inode, start,
 241                                    inline_len, compressed_size,
 242                                    compressed_pages);
 243         BUG_ON(ret);
 244         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 245         return 0;
 246 }
 247
 248 struct async_extent {
 249         u64 start;
 250         u64 ram_size;
 251         u64 compressed_size;
 252         struct page **pages;
 253         unsigned long nr_pages;
 254         struct list_head list;
 255 };
 256
 257 struct async_cow {
 258         struct inode *inode;
 259         struct btrfs_root *root;
 260         struct page *locked_page;
 261         u64 start;
 262         u64 end;
 263         struct list_head extents;
 264         struct btrfs_work work;
 265 };
 266
 267 static noinline int add_async_extent(struct async_cow *cow,
 268                                      u64 start, u64 ram_size,
 269                                      u64 compressed_size,
 270                                      struct page **pages,
 271                                      unsigned long nr_pages)
 272 {
 273         struct async_extent *async_extent;
 274
 275         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 276         async_extent->start = start;
 277         async_extent->ram_size = ram_size;
 278         async_extent->compressed_size = compressed_size;
 279         async_extent->pages = pages;
 280         async_extent->nr_pages = nr_pages;
 281         list_add_tail(&async_extent->list, &cow->extents);
 282         return 0;
 283 }
 284
 285 /*
 286  * we create compressed extents in two phases.  The first
 287  * phase compresses a range of pages that have already been
 288  * locked (both pages and state bits are locked).
 289  *
 290  * This is done inside an ordered work queue, and the compression
 291  * is spread across many cpus.  The actual IO submission is step
 292  * two, and the ordered work queue takes care of making sure that
 293  * happens in the same order things were put onto the queue by
 294  * writepages and friends.
 295  *
 296  * If this code finds it can't get good compression, it puts an
 297  * entry onto the work queue to write the uncompressed bytes.  This
 298  * makes sure that both compressed inodes and uncompressed inodes
 299  * are written in the same order that pdflush sent them down.
 300  */
 301 static noinline int compress_file_range(struct inode *inode,
 302                                         struct page *locked_page,
 303                                         u64 start, u64 end,
 304                                         struct async_cow *async_cow,
 305                                         int *num_added)
 306 {
 307         struct btrfs_root *root = BTRFS_I(inode)->root;
 308         struct btrfs_trans_handle *trans;
 309         u64 num_bytes;
 310         u64 orig_start;
 311         u64 disk_num_bytes;
 312         u64 blocksize = root->sectorsize;
 313         u64 actual_end;
 314         u64 isize = i_size_read(inode);
 315         int ret = 0;
 316         struct page **pages = NULL;
 317         unsigned long nr_pages;
 318         unsigned long nr_pages_ret = 0;
 319         unsigned long total_compressed = 0;
 320         unsigned long total_in = 0;
 321         unsigned long max_compressed = 128 * 1024;
 322         unsigned long max_uncompressed = 128 * 1024;
 323         int i;
 324         int will_compress;
 325
 326         orig_start = start;
 327
 328         actual_end = min_t(u64, isize, end + 1);
 329 again:
 330         will_compress = 0;
 331         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 332         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 333
 334         /*
 335          * we don't want to send crud past the end of i_size through
 336          * compression, that's just a waste of CPU time.  So, if the
 337          * end of the file is before the start of our current
 338          * requested range of bytes, we bail out to the uncompressed
 339          * cleanup code that can deal with all of this.
 340          *
 341          * It isn't really the fastest way to fix things, but this is a
 342          * very uncommon corner.
 343          */
 344         if (actual_end <= start)
 345                 goto cleanup_and_bail_uncompressed;
 346
 347         total_compressed = actual_end - start;
 348
 349         /* we want to make sure that amount of ram required to uncompress
 350          * an extent is reasonable, so we limit the total size in ram
 351          * of a compressed extent to 128k.  This is a crucial number
 352          * because it also controls how easily we can spread reads across
 353          * cpus for decompression.
 354          *
 355          * We also want to make sure the amount of IO required to do
 356          * a random read is reasonably small, so we limit the size of
 357          * a compressed extent to 128k.
 358          */
 359         total_compressed = min(total_compressed, max_uncompressed);
 360         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 361         num_bytes = max(blocksize,  num_bytes);
 362         disk_num_bytes = num_bytes;
 363         total_in = 0;
 364         ret = 0;
 365
 366         /*
 367          * we do compression for mount -o compress and when the
 368          * inode has not been flagged as nocompress.  This flag can
 369          * change at any time if we discover bad compression ratios.
 370          */
 371         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 372             btrfs_test_opt(root, COMPRESS)) {
 373                 WARN_ON(pages);
 374                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 375
 376                 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
 377                                                 total_compressed, pages,
 378                                                 nr_pages, &nr_pages_ret,
 379                                                 &total_in,
 380                                                 &total_compressed,
 381                                                 max_compressed);
 382
 383                 if (!ret) {
 384                         unsigned long offset = total_compressed &
 385                                 (PAGE_CACHE_SIZE - 1);
 386                         struct page *page = pages[nr_pages_ret - 1];
 387                         char *kaddr;
 388
 389                         /* zero the tail end of the last page, we might be
 390                          * sending it down to disk
 391                          */
 392                         if (offset) {
 393                                 kaddr = kmap_atomic(page, KM_USER0);
 394                                 memset(kaddr + offset, 0,
 395                                        PAGE_CACHE_SIZE - offset);
 396                                 kunmap_atomic(kaddr, KM_USER0);
 397                         }
 398                         will_compress = 1;
 399                 }
 400         }
 401         if (start == 0) {
 402                 trans = btrfs_join_transaction(root, 1);
 403                 BUG_ON(!trans);
 404                 btrfs_set_trans_block_group(trans, inode);
 405
 406                 /* lets try to make an inline extent */
 407                 if (ret || total_in < (actual_end - start)) {
 408                         /* we didn't compress the entire range, try
 409                          * to make an uncompressed inline extent.
 410                          */
 411                         ret = cow_file_range_inline(trans, root, inode,
 412                                                     start, end, 0, NULL);
 413                 } else {
 414                         /* try making a compressed inline extent */
 415                         ret = cow_file_range_inline(trans, root, inode,
 416                                                     start, end,
 417                                                     total_compressed, pages);
 418                 }
 419                 btrfs_end_transaction(trans, root);
 420                 if (ret == 0) {
 421                         /*
 422                          * inline extent creation worked, we don't need
 423                          * to create any more async work items.  Unlock
 424                          * and free up our temp pages.
 425                          */
 426                         extent_clear_unlock_delalloc(inode,
 427                                                      &BTRFS_I(inode)->io_tree,
 428                                                      start, end, NULL, 1, 0,
 429                                                      0, 1, 1, 1, 0);
 430                         ret = 0;
 431                         goto free_pages_out;
 432                 }
 433         }
 434
 435         if (will_compress) {
 436                 /*
 437                  * we aren't doing an inline extent round the compressed size
 438                  * up to a block size boundary so the allocator does sane
 439                  * things
 440                  */
 441                 total_compressed = (total_compressed + blocksize - 1) &
 442                         ~(blocksize - 1);
 443
 444                 /*
 445                  * one last check to make sure the compression is really a
 446                  * win, compare the page count read with the blocks on disk
 447                  */
 448                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 449                         ~(PAGE_CACHE_SIZE - 1);
 450                 if (total_compressed >= total_in) {
 451                         will_compress = 0;
 452                 } else {
 453                         disk_num_bytes = total_compressed;
 454                         num_bytes = total_in;
 455                 }
 456         }
 457         if (!will_compress && pages) {
 458                 /*
 459                  * the compression code ran but failed to make things smaller,
 460                  * free any pages it allocated and our page pointer array
 461                  */
 462                 for (i = 0; i < nr_pages_ret; i++) {
 463                         WARN_ON(pages[i]->mapping);
 464                         page_cache_release(pages[i]);
 465                 }
 466                 kfree(pages);
 467                 pages = NULL;
 468                 total_compressed = 0;
 469                 nr_pages_ret = 0;
 470
 471                 /* flag the file so we don't compress in the future */
 472                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 473         }
 474         if (will_compress) {
 475                 *num_added += 1;
 476
 477                 /* the async work queues will take care of doing actual
 478                  * allocation on disk for these compressed pages,
 479                  * and will submit them to the elevator.
 480                  */
 481                 add_async_extent(async_cow, start, num_bytes,
 482                                  total_compressed, pages, nr_pages_ret);
 483
 484                 if (start + num_bytes < end && start + num_bytes < actual_end) {
 485                         start += num_bytes;
 486                         pages = NULL;
 487                         cond_resched();
 488                         goto again;
 489                 }
 490         } else {
 491 cleanup_and_bail_uncompressed:
 492                 /*
 493                  * No compression, but we still need to write the pages in
 494                  * the file we've been given so far.  redirty the locked
 495                  * page if it corresponds to our extent and set things up
 496                  * for the async work queue to run cow_file_range to do
 497                  * the normal delalloc dance
 498                  */
 499                 if (page_offset(locked_page) >= start &&
 500                     page_offset(locked_page) <= end) {
 501                         __set_page_dirty_nobuffers(locked_page);
 502                         /* unlocked later on in the async handlers */
 503                 }
 504                 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
 505                 *num_added += 1;
 506         }
 507
 508 out:
 509         return 0;
 510
 511 free_pages_out:
 512         for (i = 0; i < nr_pages_ret; i++) {
 513                 WARN_ON(pages[i]->mapping);
 514                 page_cache_release(pages[i]);
 515         }
 516         kfree(pages);
 517
 518         goto out;
 519 }
 520
 521 /*
 522  * phase two of compressed writeback.  This is the ordered portion
 523  * of the code, which only gets called in the order the work was
 524  * queued.  We walk all the async extents created by compress_file_range
 525  * and send them down to the disk.
 526  */
 527 static noinline int submit_compressed_extents(struct inode *inode,
 528                                               struct async_cow *async_cow)
 529 {
 530         struct async_extent *async_extent;
 531         u64 alloc_hint = 0;
 532         struct btrfs_trans_handle *trans;
 533         struct btrfs_key ins;
 534         struct extent_map *em;
 535         struct btrfs_root *root = BTRFS_I(inode)->root;
 536         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 537         struct extent_io_tree *io_tree;
 538         int ret;
 539
 540         if (list_empty(&async_cow->extents))
 541                 return 0;
 542
 543         trans = btrfs_join_transaction(root, 1);
 544
 545         while (!list_empty(&async_cow->extents)) {
 546                 async_extent = list_entry(async_cow->extents.next,
 547                                           struct async_extent, list);
 548                 list_del(&async_extent->list);
 549
 550                 io_tree = &BTRFS_I(inode)->io_tree;
 551
 552                 /* did the compression code fall back to uncompressed IO? */
 553                 if (!async_extent->pages) {
 554                         int page_started = 0;
 555                         unsigned long nr_written = 0;
 556
 557                         lock_extent(io_tree, async_extent->start,
 558                                     async_extent->start +
 559                                     async_extent->ram_size - 1, GFP_NOFS);
 560
 561                         /* allocate blocks */
 562                         cow_file_range(inode, async_cow->locked_page,
 563                                        async_extent->start,
 564                                        async_extent->start +
 565                                        async_extent->ram_size - 1,
 566                                        &page_started, &nr_written, 0);
 567
 568                         /*
 569                          * if page_started, cow_file_range inserted an
 570                          * inline extent and took care of all the unlocking
 571                          * and IO for us.  Otherwise, we need to submit
 572                          * all those pages down to the drive.
 573                          */
 574                         if (!page_started)
 575                                 extent_write_locked_range(io_tree,
 576                                                   inode, async_extent->start,
 577                                                   async_extent->start +
 578                                                   async_extent->ram_size - 1,
 579                                                   btrfs_get_extent,
 580                                                   WB_SYNC_ALL);
 581                         kfree(async_extent);
 582                         cond_resched();
 583                         continue;
 584                 }
 585
 586                 lock_extent(io_tree, async_extent->start,
 587                             async_extent->start + async_extent->ram_size - 1,
 588                             GFP_NOFS);
 589                 /*
 590                  * here we're doing allocation and writeback of the
 591                  * compressed pages
 592                  */
 593                 btrfs_drop_extent_cache(inode, async_extent->start,
 594                                         async_extent->start +
 595                                         async_extent->ram_size - 1, 0);
 596
 597                 ret = btrfs_reserve_extent(trans, root,
 598                                            async_extent->compressed_size,
 599                                            async_extent->compressed_size,
 600                                            0, alloc_hint,
 601                                            (u64)-1, &ins, 1);
 602                 BUG_ON(ret);
 603                 em = alloc_extent_map(GFP_NOFS);
 604                 em->start = async_extent->start;
 605                 em->len = async_extent->ram_size;
 606                 em->orig_start = em->start;
 607
 608                 em->block_start = ins.objectid;
 609                 em->block_len = ins.offset;
 610                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 611                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 612                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 613
 614                 while (1) {
 615                         write_lock(&em_tree->lock);
 616                         ret = add_extent_mapping(em_tree, em);
 617                         write_unlock(&em_tree->lock);
 618                         if (ret != -EEXIST) {
 619                                 free_extent_map(em);
 620                                 break;
 621                         }
 622                         btrfs_drop_extent_cache(inode, async_extent->start,
 623                                                 async_extent->start +
 624                                                 async_extent->ram_size - 1, 0);
 625                 }
 626
 627                 ret = btrfs_add_ordered_extent(inode, async_extent->start,
 628                                                ins.objectid,
 629                                                async_extent->ram_size,
 630                                                ins.offset,
 631                                                BTRFS_ORDERED_COMPRESSED);
 632                 BUG_ON(ret);
 633
 634                 btrfs_end_transaction(trans, root);
 635
 636                 /*
 637                  * clear dirty, set writeback and unlock the pages.
 638                  */
 639                 extent_clear_unlock_delalloc(inode,
 640                                              &BTRFS_I(inode)->io_tree,
 641                                              async_extent->start,
 642                                              async_extent->start +
 643                                              async_extent->ram_size - 1,
 644                                              NULL, 1, 1, 0, 1, 1, 0, 0);
 645
 646                 ret = btrfs_submit_compressed_write(inode,
 647                                     async_extent->start,
 648                                     async_extent->ram_size,
 649                                     ins.objectid,
 650                                     ins.offset, async_extent->pages,
 651                                     async_extent->nr_pages);
 652
 653                 BUG_ON(ret);
 654                 trans = btrfs_join_transaction(root, 1);
 655                 alloc_hint = ins.objectid + ins.offset;
 656                 kfree(async_extent);
 657                 cond_resched();
 658         }
 659
 660         btrfs_end_transaction(trans, root);
 661         return 0;
 662 }
 663
 664 /*
 665  * when extent_io.c finds a delayed allocation range in the file,
 666  * the call backs end up in this code.  The basic idea is to
 667  * allocate extents on disk for the range, and create ordered data structs
 668  * in ram to track those extents.
 669  *
 670  * locked_page is the page that writepage had locked already.  We use
 671  * it to make sure we don't do extra locks or unlocks.
 672  *
 673  * *page_started is set to one if we unlock locked_page and do everything
 674  * required to start IO on it.  It may be clean and already done with
 675  * IO when we return.
 676  */
 677 static noinline int cow_file_range(struct inode *inode,
 678                                    struct page *locked_page,
 679                                    u64 start, u64 end, int *page_started,
 680                                    unsigned long *nr_written,
 681                                    int unlock)
 682 {
 683         struct btrfs_root *root = BTRFS_I(inode)->root;
 684         struct btrfs_trans_handle *trans;
 685         u64 alloc_hint = 0;
 686         u64 num_bytes;
 687         unsigned long ram_size;
 688         u64 disk_num_bytes;
 689         u64 cur_alloc_size;
 690         u64 blocksize = root->sectorsize;
 691         u64 actual_end;
 692         u64 isize = i_size_read(inode);
 693         struct btrfs_key ins;
 694         struct extent_map *em;
 695         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 696         int ret = 0;
 697
 698         trans = btrfs_join_transaction(root, 1);
 699         BUG_ON(!trans);
 700         btrfs_set_trans_block_group(trans, inode);
 701
 702         actual_end = min_t(u64, isize, end + 1);
 703
 704         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 705         num_bytes = max(blocksize,  num_bytes);
 706         disk_num_bytes = num_bytes;
 707         ret = 0;
 708
 709         if (start == 0) {
 710                 /* lets try to make an inline extent */
 711                 ret = cow_file_range_inline(trans, root, inode,
 712                                             start, end, 0, NULL);
 713                 if (ret == 0) {
 714                         extent_clear_unlock_delalloc(inode,
 715                                                      &BTRFS_I(inode)->io_tree,
 716                                                      start, end, NULL, 1, 1,
 717                                                      1, 1, 1, 1, 0);
 718                         *nr_written = *nr_written +
 719                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 720                         *page_started = 1;
 721                         ret = 0;
 722                         goto out;
 723                 }
 724         }
 725
 726         BUG_ON(disk_num_bytes >
 727                btrfs_super_total_bytes(&root->fs_info->super_copy));
 728
 729         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 730
 731         while (disk_num_bytes > 0) {
 732                 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
 733                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 734                                            root->sectorsize, 0, alloc_hint,
 735                                            (u64)-1, &ins, 1);
 736                 BUG_ON(ret);
 737
 738                 em = alloc_extent_map(GFP_NOFS);
 739                 em->start = start;
 740                 em->orig_start = em->start;
 741
 742                 ram_size = ins.offset;
 743                 em->len = ins.offset;
 744
 745                 em->block_start = ins.objectid;
 746                 em->block_len = ins.offset;
 747                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 748                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 749
 750                 while (1) {
 751                         write_lock(&em_tree->lock);
 752                         ret = add_extent_mapping(em_tree, em);
 753                         write_unlock(&em_tree->lock);
 754                         if (ret != -EEXIST) {
 755                                 free_extent_map(em);
 756                                 break;
 757                         }
 758                         btrfs_drop_extent_cache(inode, start,
 759                                                 start + ram_size - 1, 0);
 760                 }
 761
 762                 cur_alloc_size = ins.offset;
 763                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 764                                                ram_size, cur_alloc_size, 0);
 765                 BUG_ON(ret);
 766
 767                 if (root->root_key.objectid ==
 768                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 769                         ret = btrfs_reloc_clone_csums(inode, start,
 770                                                       cur_alloc_size);
 771                         BUG_ON(ret);
 772                 }
 773
 774                 if (disk_num_bytes < cur_alloc_size)
 775                         break;
 776
 777                 /* we're not doing compressed IO, don't unlock the first
 778                  * page (which the caller expects to stay locked), don't
 779                  * clear any dirty bits and don't set any writeback bits
 780                  *
 781                  * Do set the Private2 bit so we know this page was properly
 782                  * setup for writepage
 783                  */
 784                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 785                                              start, start + ram_size - 1,
 786                                              locked_page, unlock, 1,
 787                                              1, 0, 0, 0, 1);
 788                 disk_num_bytes -= cur_alloc_size;
 789                 num_bytes -= cur_alloc_size;
 790                 alloc_hint = ins.objectid + ins.offset;
 791                 start += cur_alloc_size;
 792         }
 793 out:
 794         ret = 0;
 795         btrfs_end_transaction(trans, root);
 796
 797         return ret;
 798 }
 799
 800 /*
 801  * work queue call back to started compression on a file and pages
 802  */
 803 static noinline void async_cow_start(struct btrfs_work *work)
 804 {
 805         struct async_cow *async_cow;
 806         int num_added = 0;
 807         async_cow = container_of(work, struct async_cow, work);
 808
 809         compress_file_range(async_cow->inode, async_cow->locked_page,
 810                             async_cow->start, async_cow->end, async_cow,
 811                             &num_added);
 812         if (num_added == 0)
 813                 async_cow->inode = NULL;
 814 }
 815
 816 /*
 817  * work queue call back to submit previously compressed pages
 818  */
 819 static noinline void async_cow_submit(struct btrfs_work *work)
 820 {
 821         struct async_cow *async_cow;
 822         struct btrfs_root *root;
 823         unsigned long nr_pages;
 824
 825         async_cow = container_of(work, struct async_cow, work);
 826
 827         root = async_cow->root;
 828         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 829                 PAGE_CACHE_SHIFT;
 830
 831         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 832
 833         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 834             5 * 1042 * 1024 &&
 835             waitqueue_active(&root->fs_info->async_submit_wait))
 836                 wake_up(&root->fs_info->async_submit_wait);
 837
 838         if (async_cow->inode)
 839                 submit_compressed_extents(async_cow->inode, async_cow);
 840 }
 841
 842 static noinline void async_cow_free(struct btrfs_work *work)
 843 {
 844         struct async_cow *async_cow;
 845         async_cow = container_of(work, struct async_cow, work);
 846         kfree(async_cow);
 847 }
 848
 849 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 850                                 u64 start, u64 end, int *page_started,
 851                                 unsigned long *nr_written)
 852 {
 853         struct async_cow *async_cow;
 854         struct btrfs_root *root = BTRFS_I(inode)->root;
 855         unsigned long nr_pages;
 856         u64 cur_end;
 857         int limit = 10 * 1024 * 1042;
 858
 859         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
 860                          EXTENT_DELALLOC, 1, 0, NULL, GFP_NOFS);
 861         while (start < end) {
 862                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 863                 async_cow->inode = inode;
 864                 async_cow->root = root;
 865                 async_cow->locked_page = locked_page;
 866                 async_cow->start = start;
 867
 868                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 869                         cur_end = end;
 870                 else
 871                         cur_end = min(end, start + 512 * 1024 - 1);
 872
 873                 async_cow->end = cur_end;
 874                 INIT_LIST_HEAD(&async_cow->extents);
 875
 876                 async_cow->work.func = async_cow_start;
 877                 async_cow->work.ordered_func = async_cow_submit;
 878                 async_cow->work.ordered_free = async_cow_free;
 879                 async_cow->work.flags = 0;
 880
 881                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 882                         PAGE_CACHE_SHIFT;
 883                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 884
 885                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 886                                    &async_cow->work);
 887
 888                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 889                         wait_event(root->fs_info->async_submit_wait,
 890                            (atomic_read(&root->fs_info->async_delalloc_pages) <
 891                             limit));
 892                 }
 893
 894                 while (atomic_read(&root->fs_info->async_submit_draining) &&
 895                       atomic_read(&root->fs_info->async_delalloc_pages)) {
 896                         wait_event(root->fs_info->async_submit_wait,
 897                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
 898                            0));
 899                 }
 900
 901                 *nr_written += nr_pages;
 902                 start = cur_end + 1;
 903         }
 904         *page_started = 1;
 905         return 0;
 906 }
 907
 908 static noinline int csum_exist_in_range(struct btrfs_root *root,
 909                                         u64 bytenr, u64 num_bytes)
 910 {
 911         int ret;
 912         struct btrfs_ordered_sum *sums;
 913         LIST_HEAD(list);
 914
 915         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
 916                                        bytenr + num_bytes - 1, &list);
 917         if (ret == 0 && list_empty(&list))
 918                 return 0;
 919
 920         while (!list_empty(&list)) {
 921                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
 922                 list_del(&sums->list);
 923                 kfree(sums);
 924         }
 925         return 1;
 926 }
 927
 928 /*
 929  * when nowcow writeback call back.  This checks for snapshots or COW copies
 930  * of the extents that exist in the file, and COWs the file as required.
 931  *
 932  * If no cow copies or snapshots exist, we write directly to the existing
 933  * blocks on disk
 934  */
 935 static noinline int run_delalloc_nocow(struct inode *inode,
 936                                        struct page *locked_page,
 937                               u64 start, u64 end, int *page_started, int force,
 938                               unsigned long *nr_written)
 939 {
 940         struct btrfs_root *root = BTRFS_I(inode)->root;
 941         struct btrfs_trans_handle *trans;
 942         struct extent_buffer *leaf;
 943         struct btrfs_path *path;
 944         struct btrfs_file_extent_item *fi;
 945         struct btrfs_key found_key;
 946         u64 cow_start;
 947         u64 cur_offset;
 948         u64 extent_end;
 949         u64 extent_offset;
 950         u64 disk_bytenr;
 951         u64 num_bytes;
 952         int extent_type;
 953         int ret;
 954         int type;
 955         int nocow;
 956         int check_prev = 1;
 957
 958         path = btrfs_alloc_path();
 959         BUG_ON(!path);
 960         trans = btrfs_join_transaction(root, 1);
 961         BUG_ON(!trans);
 962
 963         cow_start = (u64)-1;
 964         cur_offset = start;
 965         while (1) {
 966                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 967                                                cur_offset, 0);
 968                 BUG_ON(ret < 0);
 969                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
 970                         leaf = path->nodes[0];
 971                         btrfs_item_key_to_cpu(leaf, &found_key,
 972                                               path->slots[0] - 1);
 973                         if (found_key.objectid == inode->i_ino &&
 974                             found_key.type == BTRFS_EXTENT_DATA_KEY)
 975                                 path->slots[0]--;
 976                 }
 977                 check_prev = 0;
 978 next_slot:
 979                 leaf = path->nodes[0];
 980                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 981                         ret = btrfs_next_leaf(root, path);
 982                         if (ret < 0)
 983                                 BUG_ON(1);
 984                         if (ret > 0)
 985                                 break;
 986                         leaf = path->nodes[0];
 987                 }
 988
 989                 nocow = 0;
 990                 disk_bytenr = 0;
 991                 num_bytes = 0;
 992                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 993
 994                 if (found_key.objectid > inode->i_ino ||
 995                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
 996                     found_key.offset > end)
 997                         break;
 998
 999                 if (found_key.offset > cur_offset) {
1000                         extent_end = found_key.offset;
1001                         goto out_check;
1002                 }
1003
1004                 fi = btrfs_item_ptr(leaf, path->slots[0],
1005                                     struct btrfs_file_extent_item);
1006                 extent_type = btrfs_file_extent_type(leaf, fi);
1007
1008                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1009                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1010                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1011                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1012                         extent_end = found_key.offset +
1013                                 btrfs_file_extent_num_bytes(leaf, fi);
1014                         if (extent_end <= start) {
1015                                 path->slots[0]++;
1016                                 goto next_slot;
1017                         }
1018                         if (disk_bytenr == 0)
1019                                 goto out_check;
1020                         if (btrfs_file_extent_compression(leaf, fi) ||
1021                             btrfs_file_extent_encryption(leaf, fi) ||
1022                             btrfs_file_extent_other_encoding(leaf, fi))
1023                                 goto out_check;
1024                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1025                                 goto out_check;
1026                         if (btrfs_extent_readonly(root, disk_bytenr))
1027                                 goto out_check;
1028                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1029                                                   found_key.offset -
1030                                                   extent_offset, disk_bytenr))
1031                                 goto out_check;
1032                         disk_bytenr += extent_offset;
1033                         disk_bytenr += cur_offset - found_key.offset;
1034                         num_bytes = min(end + 1, extent_end) - cur_offset;
1035                         /*
1036                          * force cow if csum exists in the range.
1037                          * this ensure that csum for a given extent are
1038                          * either valid or do not exist.
1039                          */
1040                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1041                                 goto out_check;
1042                         nocow = 1;
1043                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1044                         extent_end = found_key.offset +
1045                                 btrfs_file_extent_inline_len(leaf, fi);
1046                         extent_end = ALIGN(extent_end, root->sectorsize);
1047                 } else {
1048                         BUG_ON(1);
1049                 }
1050 out_check:
1051                 if (extent_end <= start) {
1052                         path->slots[0]++;
1053                         goto next_slot;
1054                 }
1055                 if (!nocow) {
1056                         if (cow_start == (u64)-1)
1057                                 cow_start = cur_offset;
1058                         cur_offset = extent_end;
1059                         if (cur_offset > end)
1060                                 break;
1061                         path->slots[0]++;
1062                         goto next_slot;
1063                 }
1064
1065                 btrfs_release_path(root, path);
1066                 if (cow_start != (u64)-1) {
1067                         ret = cow_file_range(inode, locked_page, cow_start,
1068                                         found_key.offset - 1, page_started,
1069                                         nr_written, 1);
1070                         BUG_ON(ret);
1071                         cow_start = (u64)-1;
1072                 }
1073
1074                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1075                         struct extent_map *em;
1076                         struct extent_map_tree *em_tree;
1077                         em_tree = &BTRFS_I(inode)->extent_tree;
1078                         em = alloc_extent_map(GFP_NOFS);
1079                         em->start = cur_offset;
1080                         em->orig_start = em->start;
1081                         em->len = num_bytes;
1082                         em->block_len = num_bytes;
1083                         em->block_start = disk_bytenr;
1084                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1085                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1086                         while (1) {
1087                                 write_lock(&em_tree->lock);
1088                                 ret = add_extent_mapping(em_tree, em);
1089                                 write_unlock(&em_tree->lock);
1090                                 if (ret != -EEXIST) {
1091                                         free_extent_map(em);
1092                                         break;
1093                                 }
1094                                 btrfs_drop_extent_cache(inode, em->start,
1095                                                 em->start + em->len - 1, 0);
1096                         }
1097                         type = BTRFS_ORDERED_PREALLOC;
1098                 } else {
1099                         type = BTRFS_ORDERED_NOCOW;
1100                 }
1101
1102                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1103                                                num_bytes, num_bytes, type);
1104                 BUG_ON(ret);
1105
1106                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1107                                         cur_offset, cur_offset + num_bytes - 1,
1108                                         locked_page, 1, 1, 1, 0, 0, 0, 1);
1109                 cur_offset = extent_end;
1110                 if (cur_offset > end)
1111                         break;
1112         }
1113         btrfs_release_path(root, path);
1114
1115         if (cur_offset <= end && cow_start == (u64)-1)
1116                 cow_start = cur_offset;
1117         if (cow_start != (u64)-1) {
1118                 ret = cow_file_range(inode, locked_page, cow_start, end,
1119                                      page_started, nr_written, 1);
1120                 BUG_ON(ret);
1121         }
1122
1123         ret = btrfs_end_transaction(trans, root);
1124         BUG_ON(ret);
1125         btrfs_free_path(path);
1126         return 0;
1127 }
1128
1129 /*
1130  * extent_io.c call back to do delayed allocation processing
1131  */
1132 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1133                               u64 start, u64 end, int *page_started,
1134                               unsigned long *nr_written)
1135 {
1136         int ret;
1137         struct btrfs_root *root = BTRFS_I(inode)->root;
1138
1139         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1140                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1141                                          page_started, 1, nr_written);
1142         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1143                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1144                                          page_started, 0, nr_written);
1145         else if (!btrfs_test_opt(root, COMPRESS))
1146                 ret = cow_file_range(inode, locked_page, start, end,
1147                                       page_started, nr_written, 1);
1148         else
1149                 ret = cow_file_range_async(inode, locked_page, start, end,
1150                                            page_started, nr_written);
1151         return ret;
1152 }
1153
1154 /*
1155  * extent_io.c set_bit_hook, used to track delayed allocation
1156  * bytes in this file, and to maintain the list of inodes that
1157  * have pending delalloc work to be done.
1158  */
1159 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1160                        unsigned long old, unsigned long bits)
1161 {
1162         /*
1163          * set_bit and clear bit hooks normally require _irqsave/restore
1164          * but in this case, we are only testeing for the DELALLOC
1165          * bit, which is only set or cleared with irqs on
1166          */
1167         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1168                 struct btrfs_root *root = BTRFS_I(inode)->root;
1169                 btrfs_delalloc_reserve_space(root, inode, end - start + 1);
1170                 spin_lock(&root->fs_info->delalloc_lock);
1171                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1172                 root->fs_info->delalloc_bytes += end - start + 1;
1173                 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1174                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1175                                       &root->fs_info->delalloc_inodes);
1176                 }
1177                 spin_unlock(&root->fs_info->delalloc_lock);
1178         }
1179         return 0;
1180 }
1181
1182 /*
1183  * extent_io.c clear_bit_hook, see set_bit_hook for why
1184  */
1185 static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
1186                          unsigned long old, unsigned long bits)
1187 {
1188         /*
1189          * set_bit and clear bit hooks normally require _irqsave/restore
1190          * but in this case, we are only testeing for the DELALLOC
1191          * bit, which is only set or cleared with irqs on
1192          */
1193         if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1194                 struct btrfs_root *root = BTRFS_I(inode)->root;
1195
1196                 spin_lock(&root->fs_info->delalloc_lock);
1197                 if (end - start + 1 > root->fs_info->delalloc_bytes) {
1198                         printk(KERN_INFO "btrfs warning: delalloc account "
1199                                "%llu %llu\n",
1200                                (unsigned long long)end - start + 1,
1201                                (unsigned long long)
1202                                root->fs_info->delalloc_bytes);
1203                         btrfs_delalloc_free_space(root, inode, (u64)-1);
1204                         root->fs_info->delalloc_bytes = 0;
1205                         BTRFS_I(inode)->delalloc_bytes = 0;
1206                 } else {
1207                         btrfs_delalloc_free_space(root, inode,
1208                                                   end - start + 1);
1209                         root->fs_info->delalloc_bytes -= end - start + 1;
1210                         BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
1211                 }
1212                 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1213                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1214                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1215                 }
1216                 spin_unlock(&root->fs_info->delalloc_lock);
1217         }
1218         return 0;
1219 }
1220
1221 /*
1222  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1223  * we don't create bios that span stripes or chunks
1224  */
1225 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1226                          size_t size, struct bio *bio,
1227                          unsigned long bio_flags)
1228 {
1229         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1230         struct btrfs_mapping_tree *map_tree;
1231         u64 logical = (u64)bio->bi_sector << 9;
1232         u64 length = 0;
1233         u64 map_length;
1234         int ret;
1235
1236         if (bio_flags & EXTENT_BIO_COMPRESSED)
1237                 return 0;
1238
1239         length = bio->bi_size;
1240         map_tree = &root->fs_info->mapping_tree;
1241         map_length = length;
1242         ret = btrfs_map_block(map_tree, READ, logical,
1243                               &map_length, NULL, 0);
1244
1245         if (map_length < length + size)
1246                 return 1;
1247         return 0;
1248 }
1249
1250 /*
1251  * in order to insert checksums into the metadata in large chunks,
1252  * we wait until bio submission time.   All the pages in the bio are
1253  * checksummed and sums are attached onto the ordered extent record.
1254  *
1255  * At IO completion time the cums attached on the ordered extent record
1256  * are inserted into the btree
1257  */
1258 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1259                                     struct bio *bio, int mirror_num,
1260                                     unsigned long bio_flags)
1261 {
1262         struct btrfs_root *root = BTRFS_I(inode)->root;
1263         int ret = 0;
1264
1265         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1266         BUG_ON(ret);
1267         return 0;
1268 }
1269
1270 /*
1271  * in order to insert checksums into the metadata in large chunks,
1272  * we wait until bio submission time.   All the pages in the bio are
1273  * checksummed and sums are attached onto the ordered extent record.
1274  *
1275  * At IO completion time the cums attached on the ordered extent record
1276  * are inserted into the btree
1277  */
1278 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1279                           int mirror_num, unsigned long bio_flags)
1280 {
1281         struct btrfs_root *root = BTRFS_I(inode)->root;
1282         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1283 }
1284
1285 /*
1286  * extent_io.c submission hook. This does the right thing for csum calculation
1287  * on write, or reading the csums from the tree before a read
1288  */
1289 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1290                           int mirror_num, unsigned long bio_flags)
1291 {
1292         struct btrfs_root *root = BTRFS_I(inode)->root;
1293         int ret = 0;
1294         int skip_sum;
1295
1296         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1297
1298         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1299         BUG_ON(ret);
1300
1301         if (!(rw & (1 << BIO_RW))) {
1302                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1303                         return btrfs_submit_compressed_read(inode, bio,
1304                                                     mirror_num, bio_flags);
1305                 } else if (!skip_sum)
1306                         btrfs_lookup_bio_sums(root, inode, bio, NULL);
1307                 goto mapit;
1308         } else if (!skip_sum) {
1309                 /* csum items have already been cloned */
1310                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1311                         goto mapit;
1312                 /* we're doing a write, do the async checksumming */
1313                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1314                                    inode, rw, bio, mirror_num,
1315                                    bio_flags, __btrfs_submit_bio_start,
1316                                    __btrfs_submit_bio_done);
1317         }
1318
1319 mapit:
1320         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1321 }
1322
1323 /*
1324  * given a list of ordered sums record them in the inode.  This happens
1325  * at IO completion time based on sums calculated at bio submission time.
1326  */
1327 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1328                              struct inode *inode, u64 file_offset,
1329                              struct list_head *list)
1330 {
1331         struct btrfs_ordered_sum *sum;
1332
1333         btrfs_set_trans_block_group(trans, inode);
1334
1335         list_for_each_entry(sum, list, list) {
1336                 btrfs_csum_file_blocks(trans,
1337                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1338         }
1339         return 0;
1340 }
1341
1342 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1343 {
1344         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1345                 WARN_ON(1);
1346         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1347                                    GFP_NOFS);
1348 }
1349
1350 /* see btrfs_writepage_start_hook for details on why this is required */
1351 struct btrfs_writepage_fixup {
1352         struct page *page;
1353         struct btrfs_work work;
1354 };
1355
1356 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1357 {
1358         struct btrfs_writepage_fixup *fixup;
1359         struct btrfs_ordered_extent *ordered;
1360         struct page *page;
1361         struct inode *inode;
1362         u64 page_start;
1363         u64 page_end;
1364
1365         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1366         page = fixup->page;
1367 again:
1368         lock_page(page);
1369         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1370                 ClearPageChecked(page);
1371                 goto out_page;
1372         }
1373
1374         inode = page->mapping->host;
1375         page_start = page_offset(page);
1376         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1377
1378         lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1379
1380         /* already ordered? We're done */
1381         if (PagePrivate2(page))
1382                 goto out;
1383
1384         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1385         if (ordered) {
1386                 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1387                               page_end, GFP_NOFS);
1388                 unlock_page(page);
1389                 btrfs_start_ordered_extent(inode, ordered, 1);
1390                 goto again;
1391         }
1392
1393         btrfs_set_extent_delalloc(inode, page_start, page_end);
1394         ClearPageChecked(page);
1395 out:
1396         unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1397 out_page:
1398         unlock_page(page);
1399         page_cache_release(page);
1400 }
1401
1402 /*
1403  * There are a few paths in the higher layers of the kernel that directly
1404  * set the page dirty bit without asking the filesystem if it is a
1405  * good idea.  This causes problems because we want to make sure COW
1406  * properly happens and the data=ordered rules are followed.
1407  *
1408  * In our case any range that doesn't have the ORDERED bit set
1409  * hasn't been properly setup for IO.  We kick off an async process
1410  * to fix it up.  The async helper will wait for ordered extents, set
1411  * the delalloc bit and make it safe to write the page.
1412  */
1413 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1414 {
1415         struct inode *inode = page->mapping->host;
1416         struct btrfs_writepage_fixup *fixup;
1417         struct btrfs_root *root = BTRFS_I(inode)->root;
1418
1419         /* this page is properly in the ordered list */
1420         if (TestClearPagePrivate2(page))
1421                 return 0;
1422
1423         if (PageChecked(page))
1424                 return -EAGAIN;
1425
1426         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1427         if (!fixup)
1428                 return -EAGAIN;
1429
1430         SetPageChecked(page);
1431         page_cache_get(page);
1432         fixup->work.func = btrfs_writepage_fixup_worker;
1433         fixup->page = page;
1434         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1435         return -EAGAIN;
1436 }
1437
1438 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1439                                        struct inode *inode, u64 file_pos,
1440                                        u64 disk_bytenr, u64 disk_num_bytes,
1441                                        u64 num_bytes, u64 ram_bytes,
1442                                        u64 locked_end,
1443                                        u8 compression, u8 encryption,
1444                                        u16 other_encoding, int extent_type)
1445 {
1446         struct btrfs_root *root = BTRFS_I(inode)->root;
1447         struct btrfs_file_extent_item *fi;
1448         struct btrfs_path *path;
1449         struct extent_buffer *leaf;
1450         struct btrfs_key ins;
1451         u64 hint;
1452         int ret;
1453
1454         path = btrfs_alloc_path();
1455         BUG_ON(!path);
1456
1457         path->leave_spinning = 1;
1458
1459         /*
1460          * we may be replacing one extent in the tree with another.
1461          * The new extent is pinned in the extent map, and we don't want
1462          * to drop it from the cache until it is completely in the btree.
1463          *
1464          * So, tell btrfs_drop_extents to leave this extent in the cache.
1465          * the caller is expected to unpin it and allow it to be merged
1466          * with the others.
1467          */
1468         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1469                                  file_pos + num_bytes, locked_end,
1470                                  file_pos, &hint, 0);
1471         BUG_ON(ret);
1472
1473         ins.objectid = inode->i_ino;
1474         ins.offset = file_pos;
1475         ins.type = BTRFS_EXTENT_DATA_KEY;
1476         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1477         BUG_ON(ret);
1478         leaf = path->nodes[0];
1479         fi = btrfs_item_ptr(leaf, path->slots[0],
1480                             struct btrfs_file_extent_item);
1481         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1482         btrfs_set_file_extent_type(leaf, fi, extent_type);
1483         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1484         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1485         btrfs_set_file_extent_offset(leaf, fi, 0);
1486         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1487         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1488         btrfs_set_file_extent_compression(leaf, fi, compression);
1489         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1490         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1491
1492         btrfs_unlock_up_safe(path, 1);
1493         btrfs_set_lock_blocking(leaf);
1494
1495         btrfs_mark_buffer_dirty(leaf);
1496
1497         inode_add_bytes(inode, num_bytes);
1498
1499         ins.objectid = disk_bytenr;
1500         ins.offset = disk_num_bytes;
1501         ins.type = BTRFS_EXTENT_ITEM_KEY;
1502         ret = btrfs_alloc_reserved_file_extent(trans, root,
1503                                         root->root_key.objectid,
1504                                         inode->i_ino, file_pos, &ins);
1505         BUG_ON(ret);
1506         btrfs_free_path(path);
1507
1508         return 0;
1509 }
1510
1511 /*
1512  * helper function for btrfs_finish_ordered_io, this
1513  * just reads in some of the csum leaves to prime them into ram
1514  * before we start the transaction.  It limits the amount of btree
1515  * reads required while inside the transaction.
1516  */
1517 static noinline void reada_csum(struct btrfs_root *root,
1518                                 struct btrfs_path *path,
1519                                 struct btrfs_ordered_extent *ordered_extent)
1520 {
1521         struct btrfs_ordered_sum *sum;
1522         u64 bytenr;
1523
1524         sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum,
1525                          list);
1526         bytenr = sum->sums[0].bytenr;
1527
1528         /*
1529          * we don't care about the results, the point of this search is
1530          * just to get the btree leaves into ram
1531          */
1532         btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0);
1533 }
1534
1535 /* as ordered data IO finishes, this gets called so we can finish
1536  * an ordered extent if the range of bytes in the file it covers are
1537  * fully written.
1538  */
1539 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1540 {
1541         struct btrfs_root *root = BTRFS_I(inode)->root;
1542         struct btrfs_trans_handle *trans;
1543         struct btrfs_ordered_extent *ordered_extent = NULL;
1544         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1545         struct btrfs_path *path;
1546         int compressed = 0;
1547         int ret;
1548
1549         ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1550         if (!ret)
1551                 return 0;
1552
1553         /*
1554          * before we join the transaction, try to do some of our IO.
1555          * This will limit the amount of IO that we have to do with
1556          * the transaction running.  We're unlikely to need to do any
1557          * IO if the file extents are new, the disk_i_size checks
1558          * covers the most common case.
1559          */
1560         if (start < BTRFS_I(inode)->disk_i_size) {
1561                 path = btrfs_alloc_path();
1562                 if (path) {
1563                         ret = btrfs_lookup_file_extent(NULL, root, path,
1564                                                        inode->i_ino,
1565                                                        start, 0);
1566                         ordered_extent = btrfs_lookup_ordered_extent(inode,
1567                                                                      start);
1568                         if (!list_empty(&ordered_extent->list)) {
1569                                 btrfs_release_path(root, path);
1570                                 reada_csum(root, path, ordered_extent);
1571                         }
1572                         btrfs_free_path(path);
1573                 }
1574         }
1575
1576         trans = btrfs_join_transaction(root, 1);
1577
1578         if (!ordered_extent)
1579                 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1580         BUG_ON(!ordered_extent);
1581         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1582                 goto nocow;
1583
1584         lock_extent(io_tree, ordered_extent->file_offset,
1585                     ordered_extent->file_offset + ordered_extent->len - 1,
1586                     GFP_NOFS);
1587
1588         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1589                 compressed = 1;
1590         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1591                 BUG_ON(compressed);
1592                 ret = btrfs_mark_extent_written(trans, root, inode,
1593                                                 ordered_extent->file_offset,
1594                                                 ordered_extent->file_offset +
1595                                                 ordered_extent->len);
1596                 BUG_ON(ret);
1597         } else {
1598                 ret = insert_reserved_file_extent(trans, inode,
1599                                                 ordered_extent->file_offset,
1600                                                 ordered_extent->start,
1601                                                 ordered_extent->disk_len,
1602                                                 ordered_extent->len,
1603                                                 ordered_extent->len,
1604                                                 ordered_extent->file_offset +
1605                                                 ordered_extent->len,
1606                                                 compressed, 0, 0,
1607                                                 BTRFS_FILE_EXTENT_REG);
1608                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1609                                    ordered_extent->file_offset,
1610                                    ordered_extent->len);
1611                 BUG_ON(ret);
1612         }
1613         unlock_extent(io_tree, ordered_extent->file_offset,
1614                     ordered_extent->file_offset + ordered_extent->len - 1,
1615                     GFP_NOFS);
1616 nocow:
1617         add_pending_csums(trans, inode, ordered_extent->file_offset,
1618                           &ordered_extent->list);
1619
1620         mutex_lock(&BTRFS_I(inode)->extent_mutex);
1621         btrfs_ordered_update_i_size(inode, ordered_extent);
1622         btrfs_update_inode(trans, root, inode);
1623         btrfs_remove_ordered_extent(inode, ordered_extent);
1624         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1625
1626         /* once for us */
1627         btrfs_put_ordered_extent(ordered_extent);
1628         /* once for the tree */
1629         btrfs_put_ordered_extent(ordered_extent);
1630
1631         btrfs_end_transaction(trans, root);
1632         return 0;
1633 }
1634
1635 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1636                                 struct extent_state *state, int uptodate)
1637 {
1638         ClearPagePrivate2(page);
1639         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1640 }
1641
1642 /*
1643  * When IO fails, either with EIO or csum verification fails, we
1644  * try other mirrors that might have a good copy of the data.  This
1645  * io_failure_record is used to record state as we go through all the
1646  * mirrors.  If another mirror has good data, the page is set up to date
1647  * and things continue.  If a good mirror can't be found, the original
1648  * bio end_io callback is called to indicate things have failed.
1649  */
1650 struct io_failure_record {
1651         struct page *page;
1652         u64 start;
1653         u64 len;
1654         u64 logical;
1655         unsigned long bio_flags;
1656         int last_mirror;
1657 };
1658
1659 static int btrfs_io_failed_hook(struct bio *failed_bio,
1660                          struct page *page, u64 start, u64 end,
1661                          struct extent_state *state)
1662 {
1663         struct io_failure_record *failrec = NULL;
1664         u64 private;
1665         struct extent_map *em;
1666         struct inode *inode = page->mapping->host;
1667         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1668         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1669         struct bio *bio;
1670         int num_copies;
1671         int ret;
1672         int rw;
1673         u64 logical;
1674
1675         ret = get_state_private(failure_tree, start, &private);
1676         if (ret) {
1677                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1678                 if (!failrec)
1679                         return -ENOMEM;
1680                 failrec->start = start;
1681                 failrec->len = end - start + 1;
1682                 failrec->last_mirror = 0;
1683                 failrec->bio_flags = 0;
1684
1685                 read_lock(&em_tree->lock);
1686                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1687                 if (em->start > start || em->start + em->len < start) {
1688                         free_extent_map(em);
1689                         em = NULL;
1690                 }
1691                 read_unlock(&em_tree->lock);
1692
1693                 if (!em || IS_ERR(em)) {
1694                         kfree(failrec);
1695                         return -EIO;
1696                 }
1697                 logical = start - em->start;
1698                 logical = em->block_start + logical;
1699                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1700                         logical = em->block_start;
1701                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1702                 }
1703                 failrec->logical = logical;
1704                 free_extent_map(em);
1705                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1706                                 EXTENT_DIRTY, GFP_NOFS);
1707                 set_state_private(failure_tree, start,
1708                                  (u64)(unsigned long)failrec);
1709         } else {
1710                 failrec = (struct io_failure_record *)(unsigned long)private;
1711         }
1712         num_copies = btrfs_num_copies(
1713                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1714                               failrec->logical, failrec->len);
1715         failrec->last_mirror++;
1716         if (!state) {
1717                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1718                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1719                                                     failrec->start,
1720                                                     EXTENT_LOCKED);
1721                 if (state && state->start != failrec->start)
1722                         state = NULL;
1723                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1724         }
1725         if (!state || failrec->last_mirror > num_copies) {
1726                 set_state_private(failure_tree, failrec->start, 0);
1727                 clear_extent_bits(failure_tree, failrec->start,
1728                                   failrec->start + failrec->len - 1,
1729                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1730                 kfree(failrec);
1731                 return -EIO;
1732         }
1733         bio = bio_alloc(GFP_NOFS, 1);
1734         bio->bi_private = state;
1735         bio->bi_end_io = failed_bio->bi_end_io;
1736         bio->bi_sector = failrec->logical >> 9;
1737         bio->bi_bdev = failed_bio->bi_bdev;
1738         bio->bi_size = 0;
1739
1740         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1741         if (failed_bio->bi_rw & (1 << BIO_RW))
1742                 rw = WRITE;
1743         else
1744                 rw = READ;
1745
1746         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1747                                                       failrec->last_mirror,
1748                                                       failrec->bio_flags);
1749         return 0;
1750 }
1751
1752 /*
1753  * each time an IO finishes, we do a fast check in the IO failure tree
1754  * to see if we need to process or clean up an io_failure_record
1755  */
1756 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1757 {
1758         u64 private;
1759         u64 private_failure;
1760         struct io_failure_record *failure;
1761         int ret;
1762
1763         private = 0;
1764         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1765                              (u64)-1, 1, EXTENT_DIRTY)) {
1766                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1767                                         start, &private_failure);
1768                 if (ret == 0) {
1769                         failure = (struct io_failure_record *)(unsigned long)
1770                                    private_failure;
1771                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1772                                           failure->start, 0);
1773                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1774                                           failure->start,
1775                                           failure->start + failure->len - 1,
1776                                           EXTENT_DIRTY | EXTENT_LOCKED,
1777                                           GFP_NOFS);
1778                         kfree(failure);
1779                 }
1780         }
1781         return 0;
1782 }
1783
1784 /*
1785  * when reads are done, we need to check csums to verify the data is correct
1786  * if there's a match, we allow the bio to finish.  If not, we go through
1787  * the io_failure_record routines to find good copies
1788  */
1789 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1790                                struct extent_state *state)
1791 {
1792         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1793         struct inode *inode = page->mapping->host;
1794         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1795         char *kaddr;
1796         u64 private = ~(u32)0;
1797         int ret;
1798         struct btrfs_root *root = BTRFS_I(inode)->root;
1799         u32 csum = ~(u32)0;
1800
1801         if (PageChecked(page)) {
1802                 ClearPageChecked(page);
1803                 goto good;
1804         }
1805
1806         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1807                 return 0;
1808
1809         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1810             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1811                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1812                                   GFP_NOFS);
1813                 return 0;
1814         }
1815
1816         if (state && state->start == start) {
1817                 private = state->private;
1818                 ret = 0;
1819         } else {
1820                 ret = get_state_private(io_tree, start, &private);
1821         }
1822         kaddr = kmap_atomic(page, KM_USER0);
1823         if (ret)
1824                 goto zeroit;
1825
1826         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1827         btrfs_csum_final(csum, (char *)&csum);
1828         if (csum != private)
1829                 goto zeroit;
1830
1831         kunmap_atomic(kaddr, KM_USER0);
1832 good:
1833         /* if the io failure tree for this inode is non-empty,
1834          * check to see if we've recovered from a failed IO
1835          */
1836         btrfs_clean_io_failures(inode, start);
1837         return 0;
1838
1839 zeroit:
1840         if (printk_ratelimit()) {
1841                 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1842                        "private %llu\n", page->mapping->host->i_ino,
1843                        (unsigned long long)start, csum,
1844                        (unsigned long long)private);
1845         }
1846         memset(kaddr + offset, 1, end - start + 1);
1847         flush_dcache_page(page);
1848         kunmap_atomic(kaddr, KM_USER0);
1849         if (private == 0)
1850                 return 0;
1851         return -EIO;
1852 }
1853
1854 /*
1855  * This creates an orphan entry for the given inode in case something goes
1856  * wrong in the middle of an unlink/truncate.
1857  */
1858 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1859 {
1860         struct btrfs_root *root = BTRFS_I(inode)->root;
1861         int ret = 0;
1862
1863         spin_lock(&root->list_lock);
1864
1865         /* already on the orphan list, we're good */
1866         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
1867                 spin_unlock(&root->list_lock);
1868                 return 0;
1869         }
1870
1871         list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1872
1873         spin_unlock(&root->list_lock);
1874
1875         /*
1876          * insert an orphan item to track this unlinked/truncated file
1877          */
1878         ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1879
1880         return ret;
1881 }
1882
1883 /*
1884  * We have done the truncate/delete so we can go ahead and remove the orphan
1885  * item for this particular inode.
1886  */
1887 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1888 {
1889         struct btrfs_root *root = BTRFS_I(inode)->root;
1890         int ret = 0;
1891
1892         spin_lock(&root->list_lock);
1893
1894         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
1895                 spin_unlock(&root->list_lock);
1896                 return 0;
1897         }
1898
1899         list_del_init(&BTRFS_I(inode)->i_orphan);
1900         if (!trans) {
1901                 spin_unlock(&root->list_lock);
1902                 return 0;
1903         }
1904
1905         spin_unlock(&root->list_lock);
1906
1907         ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1908
1909         return ret;
1910 }
1911
1912 /*
1913  * this cleans up any orphans that may be left on the list from the last use
1914  * of this root.
1915  */
1916 void btrfs_orphan_cleanup(struct btrfs_root *root)
1917 {
1918         struct btrfs_path *path;
1919         struct extent_buffer *leaf;
1920         struct btrfs_item *item;
1921         struct btrfs_key key, found_key;
1922         struct btrfs_trans_handle *trans;
1923         struct inode *inode;
1924         int ret = 0, nr_unlink = 0, nr_truncate = 0;
1925
1926         path = btrfs_alloc_path();
1927         if (!path)
1928                 return;
1929         path->reada = -1;
1930
1931         key.objectid = BTRFS_ORPHAN_OBJECTID;
1932         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1933         key.offset = (u64)-1;
1934
1935
1936         while (1) {
1937                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1938                 if (ret < 0) {
1939                         printk(KERN_ERR "Error searching slot for orphan: %d"
1940                                "\n", ret);
1941                         break;
1942                 }
1943
1944                 /*
1945                  * if ret == 0 means we found what we were searching for, which
1946                  * is weird, but possible, so only screw with path if we didnt
1947                  * find the key and see if we have stuff that matches
1948                  */
1949                 if (ret > 0) {
1950                         if (path->slots[0] == 0)
1951                                 break;
1952                         path->slots[0]--;
1953                 }
1954
1955                 /* pull out the item */
1956                 leaf = path->nodes[0];
1957                 item = btrfs_item_nr(leaf, path->slots[0]);
1958                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1959
1960                 /* make sure the item matches what we want */
1961                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1962                         break;
1963                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1964                         break;
1965
1966                 /* release the path since we're done with it */
1967                 btrfs_release_path(root, path);
1968
1969                 /*
1970                  * this is where we are basically btrfs_lookup, without the
1971                  * crossing root thing.  we store the inode number in the
1972                  * offset of the orphan item.
1973                  */
1974                 found_key.objectid = found_key.offset;
1975                 found_key.type = BTRFS_INODE_ITEM_KEY;
1976                 found_key.offset = 0;
1977                 inode = btrfs_iget(root->fs_info->sb, &found_key, root);
1978                 if (IS_ERR(inode))
1979                         break;
1980
1981                 /*
1982                  * add this inode to the orphan list so btrfs_orphan_del does
1983                  * the proper thing when we hit it
1984                  */
1985                 spin_lock(&root->list_lock);
1986                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1987                 spin_unlock(&root->list_lock);
1988
1989                 /*
1990                  * if this is a bad inode, means we actually succeeded in
1991                  * removing the inode, but not the orphan record, which means
1992                  * we need to manually delete the orphan since iput will just
1993                  * do a destroy_inode
1994                  */
1995                 if (is_bad_inode(inode)) {
1996                         trans = btrfs_start_transaction(root, 1);
1997                         btrfs_orphan_del(trans, inode);
1998                         btrfs_end_transaction(trans, root);
1999                         iput(inode);
2000                         continue;
2001                 }
2002
2003                 /* if we have links, this was a truncate, lets do that */
2004                 if (inode->i_nlink) {
2005                         nr_truncate++;
2006                         btrfs_truncate(inode);
2007                 } else {
2008                         nr_unlink++;
2009                 }
2010
2011                 /* this will do delete_inode and everything for us */
2012                 iput(inode);
2013         }
2014
2015         if (nr_unlink)
2016                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2017         if (nr_truncate)
2018                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2019
2020         btrfs_free_path(path);
2021 }
2022
2023 /*
2024  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2025  * don't find any xattrs, we know there can't be any acls.
2026  *
2027  * slot is the slot the inode is in, objectid is the objectid of the inode
2028  */
2029 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2030                                           int slot, u64 objectid)
2031 {
2032         u32 nritems = btrfs_header_nritems(leaf);
2033         struct btrfs_key found_key;
2034         int scanned = 0;
2035
2036         slot++;
2037         while (slot < nritems) {
2038                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2039
2040                 /* we found a different objectid, there must not be acls */
2041                 if (found_key.objectid != objectid)
2042                         return 0;
2043
2044                 /* we found an xattr, assume we've got an acl */
2045                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2046                         return 1;
2047
2048                 /*
2049                  * we found a key greater than an xattr key, there can't
2050                  * be any acls later on
2051                  */
2052                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2053                         return 0;
2054
2055                 slot++;
2056                 scanned++;
2057
2058                 /*
2059                  * it goes inode, inode backrefs, xattrs, extents,
2060                  * so if there are a ton of hard links to an inode there can
2061                  * be a lot of backrefs.  Don't waste time searching too hard,
2062                  * this is just an optimization
2063                  */
2064                 if (scanned >= 8)
2065                         break;
2066         }
2067         /* we hit the end of the leaf before we found an xattr or
2068          * something larger than an xattr.  We have to assume the inode
2069          * has acls
2070          */
2071         return 1;
2072 }
2073
2074 /*
2075  * read an inode from the btree into the in-memory inode
2076  */
2077 static void btrfs_read_locked_inode(struct inode *inode)
2078 {
2079         struct btrfs_path *path;
2080         struct extent_buffer *leaf;
2081         struct btrfs_inode_item *inode_item;
2082         struct btrfs_timespec *tspec;
2083         struct btrfs_root *root = BTRFS_I(inode)->root;
2084         struct btrfs_key location;
2085         int maybe_acls;
2086         u64 alloc_group_block;
2087         u32 rdev;
2088         int ret;
2089
2090         path = btrfs_alloc_path();
2091         BUG_ON(!path);
2092         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2093
2094         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2095         if (ret)
2096                 goto make_bad;
2097
2098         leaf = path->nodes[0];
2099         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2100                                     struct btrfs_inode_item);
2101
2102         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2103         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2104         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2105         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2106         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2107
2108         tspec = btrfs_inode_atime(inode_item);
2109         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2110         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2111
2112         tspec = btrfs_inode_mtime(inode_item);
2113         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2114         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2115
2116         tspec = btrfs_inode_ctime(inode_item);
2117         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2118         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2119
2120         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2121         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2122         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2123         inode->i_generation = BTRFS_I(inode)->generation;
2124         inode->i_rdev = 0;
2125         rdev = btrfs_inode_rdev(leaf, inode_item);
2126
2127         BTRFS_I(inode)->index_cnt = (u64)-1;
2128         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2129
2130         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2131
2132         /*
2133          * try to precache a NULL acl entry for files that don't have
2134          * any xattrs or acls
2135          */
2136         maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2137         if (!maybe_acls)
2138                 cache_no_acl(inode);
2139
2140         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2141                                                 alloc_group_block, 0);
2142         btrfs_free_path(path);
2143         inode_item = NULL;
2144
2145         switch (inode->i_mode & S_IFMT) {
2146         case S_IFREG:
2147                 inode->i_mapping->a_ops = &btrfs_aops;
2148                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2149                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2150                 inode->i_fop = &btrfs_file_operations;
2151                 inode->i_op = &btrfs_file_inode_operations;
2152                 break;
2153         case S_IFDIR:
2154                 inode->i_fop = &btrfs_dir_file_operations;
2155                 if (root == root->fs_info->tree_root)
2156                         inode->i_op = &btrfs_dir_ro_inode_operations;
2157                 else
2158                         inode->i_op = &btrfs_dir_inode_operations;
2159                 break;
2160         case S_IFLNK:
2161                 inode->i_op = &btrfs_symlink_inode_operations;
2162                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2163                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2164                 break;
2165         default:
2166                 inode->i_op = &btrfs_special_inode_operations;
2167                 init_special_inode(inode, inode->i_mode, rdev);
2168                 break;
2169         }
2170
2171         btrfs_update_iflags(inode);
2172         return;
2173
2174 make_bad:
2175         btrfs_free_path(path);
2176         make_bad_inode(inode);
2177 }
2178
2179 /*
2180  * given a leaf and an inode, copy the inode fields into the leaf
2181  */
2182 static void fill_inode_item(struct btrfs_trans_handle *trans,
2183                             struct extent_buffer *leaf,
2184                             struct btrfs_inode_item *item,
2185                             struct inode *inode)
2186 {
2187         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2188         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2189         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2190         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2191         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2192
2193         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2194                                inode->i_atime.tv_sec);
2195         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2196                                 inode->i_atime.tv_nsec);
2197
2198         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2199                                inode->i_mtime.tv_sec);
2200         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2201                                 inode->i_mtime.tv_nsec);
2202
2203         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2204                                inode->i_ctime.tv_sec);
2205         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2206                                 inode->i_ctime.tv_nsec);
2207
2208         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2209         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2210         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2211         btrfs_set_inode_transid(leaf, item, trans->transid);
2212         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2213         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2214         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2215 }
2216
2217 /*
2218  * copy everything in the in-memory inode into the btree.
2219  */
2220 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2221                                 struct btrfs_root *root, struct inode *inode)
2222 {
2223         struct btrfs_inode_item *inode_item;
2224         struct btrfs_path *path;
2225         struct extent_buffer *leaf;
2226         int ret;
2227
2228         path = btrfs_alloc_path();
2229         BUG_ON(!path);
2230         path->leave_spinning = 1;
2231         ret = btrfs_lookup_inode(trans, root, path,
2232                                  &BTRFS_I(inode)->location, 1);
2233         if (ret) {
2234                 if (ret > 0)
2235                         ret = -ENOENT;
2236                 goto failed;
2237         }
2238
2239         btrfs_unlock_up_safe(path, 1);
2240         leaf = path->nodes[0];
2241         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2242                                   struct btrfs_inode_item);
2243
2244         fill_inode_item(trans, leaf, inode_item, inode);
2245         btrfs_mark_buffer_dirty(leaf);
2246         btrfs_set_inode_last_trans(trans, inode);
2247         ret = 0;
2248 failed:
2249         btrfs_free_path(path);
2250         return ret;
2251 }
2252
2253
2254 /*
2255  * unlink helper that gets used here in inode.c and in the tree logging
2256  * recovery code.  It remove a link in a directory with a given name, and
2257  * also drops the back refs in the inode to the directory
2258  */
2259 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2260                        struct btrfs_root *root,
2261                        struct inode *dir, struct inode *inode,
2262                        const char *name, int name_len)
2263 {
2264         struct btrfs_path *path;
2265         int ret = 0;
2266         struct extent_buffer *leaf;
2267         struct btrfs_dir_item *di;
2268         struct btrfs_key key;
2269         u64 index;
2270
2271         path = btrfs_alloc_path();
2272         if (!path) {
2273                 ret = -ENOMEM;
2274                 goto err;
2275         }
2276
2277         path->leave_spinning = 1;
2278         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2279                                     name, name_len, -1);
2280         if (IS_ERR(di)) {
2281                 ret = PTR_ERR(di);
2282                 goto err;
2283         }
2284         if (!di) {
2285                 ret = -ENOENT;
2286                 goto err;
2287         }
2288         leaf = path->nodes[0];
2289         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2290         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2291         if (ret)
2292                 goto err;
2293         btrfs_release_path(root, path);
2294
2295         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2296                                   inode->i_ino,
2297                                   dir->i_ino, &index);
2298         if (ret) {
2299                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2300                        "inode %lu parent %lu\n", name_len, name,
2301                        inode->i_ino, dir->i_ino);
2302                 goto err;
2303         }
2304
2305         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2306                                          index, name, name_len, -1);
2307         if (IS_ERR(di)) {
2308                 ret = PTR_ERR(di);
2309                 goto err;
2310         }
2311         if (!di) {
2312                 ret = -ENOENT;
2313                 goto err;
2314         }
2315         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2316         btrfs_release_path(root, path);
2317
2318         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2319                                          inode, dir->i_ino);
2320         BUG_ON(ret != 0 && ret != -ENOENT);
2321
2322         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2323                                            dir, index);
2324         BUG_ON(ret);
2325 err:
2326         btrfs_free_path(path);
2327         if (ret)
2328                 goto out;
2329
2330         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2331         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2332         btrfs_update_inode(trans, root, dir);
2333         btrfs_drop_nlink(inode);
2334         ret = btrfs_update_inode(trans, root, inode);
2335 out:
2336         return ret;
2337 }
2338
2339 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2340 {
2341         struct btrfs_root *root;
2342         struct btrfs_trans_handle *trans;
2343         struct inode *inode = dentry->d_inode;
2344         int ret;
2345         unsigned long nr = 0;
2346
2347         root = BTRFS_I(dir)->root;
2348
2349         trans = btrfs_start_transaction(root, 1);
2350
2351         btrfs_set_trans_block_group(trans, dir);
2352
2353         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2354
2355         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2356                                  dentry->d_name.name, dentry->d_name.len);
2357
2358         if (inode->i_nlink == 0)
2359                 ret = btrfs_orphan_add(trans, inode);
2360
2361         nr = trans->blocks_used;
2362
2363         btrfs_end_transaction_throttle(trans, root);
2364         btrfs_btree_balance_dirty(root, nr);
2365         return ret;
2366 }
2367
2368 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2369 {
2370         struct inode *inode = dentry->d_inode;
2371         int err = 0;
2372         int ret;
2373         struct btrfs_root *root = BTRFS_I(dir)->root;
2374         struct btrfs_trans_handle *trans;
2375         unsigned long nr = 0;
2376
2377         /*
2378          * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2379          * the root of a subvolume or snapshot
2380          */
2381         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2382             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
2383                 return -ENOTEMPTY;
2384         }
2385
2386         trans = btrfs_start_transaction(root, 1);
2387         btrfs_set_trans_block_group(trans, dir);
2388
2389         err = btrfs_orphan_add(trans, inode);
2390         if (err)
2391                 goto fail_trans;
2392
2393         /* now the directory is empty */
2394         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2395                                  dentry->d_name.name, dentry->d_name.len);
2396         if (!err)
2397                 btrfs_i_size_write(inode, 0);
2398
2399 fail_trans:
2400         nr = trans->blocks_used;
2401         ret = btrfs_end_transaction_throttle(trans, root);
2402         btrfs_btree_balance_dirty(root, nr);
2403
2404         if (ret && !err)
2405                 err = ret;
2406         return err;
2407 }
2408
2409 #if 0
2410 /*
2411  * when truncating bytes in a file, it is possible to avoid reading
2412  * the leaves that contain only checksum items.  This can be the
2413  * majority of the IO required to delete a large file, but it must
2414  * be done carefully.
2415  *
2416  * The keys in the level just above the leaves are checked to make sure
2417  * the lowest key in a given leaf is a csum key, and starts at an offset
2418  * after the new  size.
2419  *
2420  * Then the key for the next leaf is checked to make sure it also has
2421  * a checksum item for the same file.  If it does, we know our target leaf
2422  * contains only checksum items, and it can be safely freed without reading
2423  * it.
2424  *
2425  * This is just an optimization targeted at large files.  It may do
2426  * nothing.  It will return 0 unless things went badly.
2427  */
2428 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2429                                      struct btrfs_root *root,
2430                                      struct btrfs_path *path,
2431                                      struct inode *inode, u64 new_size)
2432 {
2433         struct btrfs_key key;
2434         int ret;
2435         int nritems;
2436         struct btrfs_key found_key;
2437         struct btrfs_key other_key;
2438         struct btrfs_leaf_ref *ref;
2439         u64 leaf_gen;
2440         u64 leaf_start;
2441
2442         path->lowest_level = 1;
2443         key.objectid = inode->i_ino;
2444         key.type = BTRFS_CSUM_ITEM_KEY;
2445         key.offset = new_size;
2446 again:
2447         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2448         if (ret < 0)
2449                 goto out;
2450
2451         if (path->nodes[1] == NULL) {
2452                 ret = 0;
2453                 goto out;
2454         }
2455         ret = 0;
2456         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2457         nritems = btrfs_header_nritems(path->nodes[1]);
2458
2459         if (!nritems)
2460                 goto out;
2461
2462         if (path->slots[1] >= nritems)
2463                 goto next_node;
2464
2465         /* did we find a key greater than anything we want to delete? */
2466         if (found_key.objectid > inode->i_ino ||
2467            (found_key.objectid == inode->i_ino && found_key.type > key.type))
2468                 goto out;
2469
2470         /* we check the next key in the node to make sure the leave contains
2471          * only checksum items.  This comparison doesn't work if our
2472          * leaf is the last one in the node
2473          */
2474         if (path->slots[1] + 1 >= nritems) {
2475 next_node:
2476                 /* search forward from the last key in the node, this
2477                  * will bring us into the next node in the tree
2478                  */
2479                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2480
2481                 /* unlikely, but we inc below, so check to be safe */
2482                 if (found_key.offset == (u64)-1)
2483                         goto out;
2484
2485                 /* search_forward needs a path with locks held, do the
2486                  * search again for the original key.  It is possible
2487                  * this will race with a balance and return a path that
2488                  * we could modify, but this drop is just an optimization
2489                  * and is allowed to miss some leaves.
2490                  */
2491                 btrfs_release_path(root, path);
2492                 found_key.offset++;
2493
2494                 /* setup a max key for search_forward */
2495                 other_key.offset = (u64)-1;
2496                 other_key.type = key.type;
2497                 other_key.objectid = key.objectid;
2498
2499                 path->keep_locks = 1;
2500                 ret = btrfs_search_forward(root, &found_key, &other_key,
2501                                            path, 0, 0);
2502                 path->keep_locks = 0;
2503                 if (ret || found_key.objectid != key.objectid ||
2504                     found_key.type != key.type) {
2505                         ret = 0;
2506                         goto out;
2507                 }
2508
2509                 key.offset = found_key.offset;
2510                 btrfs_release_path(root, path);
2511                 cond_resched();
2512                 goto again;
2513         }
2514
2515         /* we know there's one more slot after us in the tree,
2516          * read that key so we can verify it is also a checksum item
2517          */
2518         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2519
2520         if (found_key.objectid < inode->i_ino)
2521                 goto next_key;
2522
2523         if (found_key.type != key.type || found_key.offset < new_size)
2524                 goto next_key;
2525
2526         /*
2527          * if the key for the next leaf isn't a csum key from this objectid,
2528          * we can't be sure there aren't good items inside this leaf.
2529          * Bail out
2530          */
2531         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2532                 goto out;
2533
2534         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2535         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2536         /*
2537          * it is safe to delete this leaf, it contains only
2538          * csum items from this inode at an offset >= new_size
2539          */
2540         ret = btrfs_del_leaf(trans, root, path, leaf_start);
2541         BUG_ON(ret);
2542
2543         if (root->ref_cows && leaf_gen < trans->transid) {
2544                 ref = btrfs_alloc_leaf_ref(root, 0);
2545                 if (ref) {
2546                         ref->root_gen = root->root_key.offset;
2547                         ref->bytenr = leaf_start;
2548                         ref->owner = 0;
2549                         ref->generation = leaf_gen;
2550                         ref->nritems = 0;
2551
2552                         btrfs_sort_leaf_ref(ref);
2553
2554                         ret = btrfs_add_leaf_ref(root, ref, 0);
2555                         WARN_ON(ret);
2556                         btrfs_free_leaf_ref(root, ref);
2557                 } else {
2558                         WARN_ON(1);
2559                 }
2560         }
2561 next_key:
2562         btrfs_release_path(root, path);
2563
2564         if (other_key.objectid == inode->i_ino &&
2565             other_key.type == key.type && other_key.offset > key.offset) {
2566                 key.offset = other_key.offset;
2567                 cond_resched();
2568                 goto again;
2569         }
2570         ret = 0;
2571 out:
2572         /* fixup any changes we've made to the path */
2573         path->lowest_level = 0;
2574         path->keep_locks = 0;
2575         btrfs_release_path(root, path);
2576         return ret;
2577 }
2578
2579 #endif
2580
2581 /*
2582  * this can truncate away extent items, csum items and directory items.
2583  * It starts at a high offset and removes keys until it can't find
2584  * any higher than new_size
2585  *
2586  * csum items that cross the new i_size are truncated to the new size
2587  * as well.
2588  *
2589  * min_type is the minimum key type to truncate down to.  If set to 0, this
2590  * will kill all the items on this inode, including the INODE_ITEM_KEY.
2591  */
2592 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2593                                         struct btrfs_root *root,
2594                                         struct inode *inode,
2595                                         u64 new_size, u32 min_type)
2596 {
2597         int ret;
2598         struct btrfs_path *path;
2599         struct btrfs_key key;
2600         struct btrfs_key found_key;
2601         u32 found_type = (u8)-1;
2602         struct extent_buffer *leaf;
2603         struct btrfs_file_extent_item *fi;
2604         u64 extent_start = 0;
2605         u64 extent_num_bytes = 0;
2606         u64 extent_offset = 0;
2607         u64 item_end = 0;
2608         int found_extent;
2609         int del_item;
2610         int pending_del_nr = 0;
2611         int pending_del_slot = 0;
2612         int extent_type = -1;
2613         int encoding;
2614         u64 mask = root->sectorsize - 1;
2615
2616         if (root->ref_cows)
2617                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2618         path = btrfs_alloc_path();
2619         BUG_ON(!path);
2620         path->reada = -1;
2621
2622         /* FIXME, add redo link to tree so we don't leak on crash */
2623         key.objectid = inode->i_ino;
2624         key.offset = (u64)-1;
2625         key.type = (u8)-1;
2626
2627 search_again:
2628         path->leave_spinning = 1;
2629         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2630         if (ret < 0)
2631                 goto error;
2632
2633         if (ret > 0) {
2634                 /* there are no items in the tree for us to truncate, we're
2635                  * done
2636                  */
2637                 if (path->slots[0] == 0) {
2638                         ret = 0;
2639                         goto error;
2640                 }
2641                 path->slots[0]--;
2642         }
2643
2644         while (1) {
2645                 fi = NULL;
2646                 leaf = path->nodes[0];
2647                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2648                 found_type = btrfs_key_type(&found_key);
2649                 encoding = 0;
2650
2651                 if (found_key.objectid != inode->i_ino)
2652                         break;
2653
2654                 if (found_type < min_type)
2655                         break;
2656
2657                 item_end = found_key.offset;
2658                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2659                         fi = btrfs_item_ptr(leaf, path->slots[0],
2660                                             struct btrfs_file_extent_item);
2661                         extent_type = btrfs_file_extent_type(leaf, fi);
2662                         encoding = btrfs_file_extent_compression(leaf, fi);
2663                         encoding |= btrfs_file_extent_encryption(leaf, fi);
2664                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2665
2666                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2667                                 item_end +=
2668                                     btrfs_file_extent_num_bytes(leaf, fi);
2669                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2670                                 item_end += btrfs_file_extent_inline_len(leaf,
2671                                                                          fi);
2672                         }
2673                         item_end--;
2674                 }
2675                 if (item_end < new_size) {
2676                         if (found_type == BTRFS_DIR_ITEM_KEY)
2677                                 found_type = BTRFS_INODE_ITEM_KEY;
2678                         else if (found_type == BTRFS_EXTENT_ITEM_KEY)
2679                                 found_type = BTRFS_EXTENT_DATA_KEY;
2680                         else if (found_type == BTRFS_EXTENT_DATA_KEY)
2681                                 found_type = BTRFS_XATTR_ITEM_KEY;
2682                         else if (found_type == BTRFS_XATTR_ITEM_KEY)
2683                                 found_type = BTRFS_INODE_REF_KEY;
2684                         else if (found_type)
2685                                 found_type--;
2686                         else
2687                                 break;
2688                         btrfs_set_key_type(&key, found_type);
2689                         goto next;
2690                 }
2691                 if (found_key.offset >= new_size)
2692                         del_item = 1;
2693                 else
2694                         del_item = 0;
2695                 found_extent = 0;
2696
2697                 /* FIXME, shrink the extent if the ref count is only 1 */
2698                 if (found_type != BTRFS_EXTENT_DATA_KEY)
2699                         goto delete;
2700
2701                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2702                         u64 num_dec;
2703                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2704                         if (!del_item && !encoding) {
2705                                 u64 orig_num_bytes =
2706                                         btrfs_file_extent_num_bytes(leaf, fi);
2707                                 extent_num_bytes = new_size -
2708                                         found_key.offset + root->sectorsize - 1;
2709                                 extent_num_bytes = extent_num_bytes &
2710                                         ~((u64)root->sectorsize - 1);
2711                                 btrfs_set_file_extent_num_bytes(leaf, fi,
2712                                                          extent_num_bytes);
2713                                 num_dec = (orig_num_bytes -
2714                                            extent_num_bytes);
2715                                 if (root->ref_cows && extent_start != 0)
2716                                         inode_sub_bytes(inode, num_dec);
2717                                 btrfs_mark_buffer_dirty(leaf);
2718                         } else {
2719                                 extent_num_bytes =
2720                                         btrfs_file_extent_disk_num_bytes(leaf,
2721                                                                          fi);
2722                                 extent_offset = found_key.offset -
2723                                         btrfs_file_extent_offset(leaf, fi);
2724
2725                                 /* FIXME blocksize != 4096 */
2726                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2727                                 if (extent_start != 0) {
2728                                         found_extent = 1;
2729                                         if (root->ref_cows)
2730                                                 inode_sub_bytes(inode, num_dec);
2731                                 }
2732                         }
2733                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2734                         /*
2735                          * we can't truncate inline items that have had
2736                          * special encodings
2737                          */
2738                         if (!del_item &&
2739                             btrfs_file_extent_compression(leaf, fi) == 0 &&
2740                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
2741                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2742                                 u32 size = new_size - found_key.offset;
2743
2744                                 if (root->ref_cows) {
2745                                         inode_sub_bytes(inode, item_end + 1 -
2746                                                         new_size);
2747                                 }
2748                                 size =
2749                                     btrfs_file_extent_calc_inline_size(size);
2750                                 ret = btrfs_truncate_item(trans, root, path,
2751                                                           size, 1);
2752                                 BUG_ON(ret);
2753                         } else if (root->ref_cows) {
2754                                 inode_sub_bytes(inode, item_end + 1 -
2755                                                 found_key.offset);
2756                         }
2757                 }
2758 delete:
2759                 if (del_item) {
2760                         if (!pending_del_nr) {
2761                                 /* no pending yet, add ourselves */
2762                                 pending_del_slot = path->slots[0];
2763                                 pending_del_nr = 1;
2764                         } else if (pending_del_nr &&
2765                                    path->slots[0] + 1 == pending_del_slot) {
2766                                 /* hop on the pending chunk */
2767                                 pending_del_nr++;
2768                                 pending_del_slot = path->slots[0];
2769                         } else {
2770                                 BUG();
2771                         }
2772                 } else {
2773                         break;
2774                 }
2775                 if (found_extent && root->ref_cows) {
2776                         btrfs_set_path_blocking(path);
2777                         ret = btrfs_free_extent(trans, root, extent_start,
2778                                                 extent_num_bytes, 0,
2779                                                 btrfs_header_owner(leaf),
2780                                                 inode->i_ino, extent_offset);
2781                         BUG_ON(ret);
2782                 }
2783 next:
2784                 if (path->slots[0] == 0) {
2785                         if (pending_del_nr)
2786                                 goto del_pending;
2787                         btrfs_release_path(root, path);
2788                         if (found_type == BTRFS_INODE_ITEM_KEY)
2789                                 break;
2790                         goto search_again;
2791                 }
2792
2793                 path->slots[0]--;
2794                 if (pending_del_nr &&
2795                     path->slots[0] + 1 != pending_del_slot) {
2796                         struct btrfs_key debug;
2797 del_pending:
2798                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
2799                                               pending_del_slot);
2800                         ret = btrfs_del_items(trans, root, path,
2801                                               pending_del_slot,
2802                                               pending_del_nr);
2803                         BUG_ON(ret);
2804                         pending_del_nr = 0;
2805                         btrfs_release_path(root, path);
2806                         if (found_type == BTRFS_INODE_ITEM_KEY)
2807                                 break;
2808                         goto search_again;
2809                 }
2810         }
2811         ret = 0;
2812 error:
2813         if (pending_del_nr) {
2814                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2815                                       pending_del_nr);
2816         }
2817         btrfs_free_path(path);
2818         return ret;
2819 }
2820
2821 /*
2822  * taken from block_truncate_page, but does cow as it zeros out
2823  * any bytes left in the last page in the file.
2824  */
2825 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2826 {
2827         struct inode *inode = mapping->host;
2828         struct btrfs_root *root = BTRFS_I(inode)->root;
2829         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2830         struct btrfs_ordered_extent *ordered;
2831         char *kaddr;
2832         u32 blocksize = root->sectorsize;
2833         pgoff_t index = from >> PAGE_CACHE_SHIFT;
2834         unsigned offset = from & (PAGE_CACHE_SIZE-1);
2835         struct page *page;
2836         int ret = 0;
2837         u64 page_start;
2838         u64 page_end;
2839
2840         if ((offset & (blocksize - 1)) == 0)
2841                 goto out;
2842
2843         ret = -ENOMEM;
2844 again:
2845         page = grab_cache_page(mapping, index);
2846         if (!page)
2847                 goto out;
2848
2849         page_start = page_offset(page);
2850         page_end = page_start + PAGE_CACHE_SIZE - 1;
2851
2852         if (!PageUptodate(page)) {
2853                 ret = btrfs_readpage(NULL, page);
2854                 lock_page(page);
2855                 if (page->mapping != mapping) {
2856                         unlock_page(page);
2857                         page_cache_release(page);
2858                         goto again;
2859                 }
2860                 if (!PageUptodate(page)) {
2861                         ret = -EIO;
2862                         goto out_unlock;
2863                 }
2864         }
2865         wait_on_page_writeback(page);
2866
2867         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2868         set_page_extent_mapped(page);
2869
2870         ordered = btrfs_lookup_ordered_extent(inode, page_start);
2871         if (ordered) {
2872                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2873                 unlock_page(page);
2874                 page_cache_release(page);
2875                 btrfs_start_ordered_extent(inode, ordered, 1);
2876                 btrfs_put_ordered_extent(ordered);
2877                 goto again;
2878         }
2879
2880         btrfs_set_extent_delalloc(inode, page_start, page_end);
2881         ret = 0;
2882         if (offset != PAGE_CACHE_SIZE) {
2883                 kaddr = kmap(page);
2884                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2885                 flush_dcache_page(page);
2886                 kunmap(page);
2887         }
2888         ClearPageChecked(page);
2889         set_page_dirty(page);
2890         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2891
2892 out_unlock:
2893         unlock_page(page);
2894         page_cache_release(page);
2895 out:
2896         return ret;
2897 }
2898
2899 int btrfs_cont_expand(struct inode *inode, loff_t size)
2900 {
2901         struct btrfs_trans_handle *trans;
2902         struct btrfs_root *root = BTRFS_I(inode)->root;
2903         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2904         struct extent_map *em;
2905         u64 mask = root->sectorsize - 1;
2906         u64 hole_start = (inode->i_size + mask) & ~mask;
2907         u64 block_end = (size + mask) & ~mask;
2908         u64 last_byte;
2909         u64 cur_offset;
2910         u64 hole_size;
2911         int err;
2912
2913         if (size <= hole_start)
2914                 return 0;
2915
2916         err = btrfs_check_metadata_free_space(root);
2917         if (err)
2918                 return err;
2919
2920         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2921
2922         while (1) {
2923                 struct btrfs_ordered_extent *ordered;
2924                 btrfs_wait_ordered_range(inode, hole_start,
2925                                          block_end - hole_start);
2926                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2927                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2928                 if (!ordered)
2929                         break;
2930                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2931                 btrfs_put_ordered_extent(ordered);
2932         }
2933
2934         trans = btrfs_start_transaction(root, 1);
2935         btrfs_set_trans_block_group(trans, inode);
2936
2937         cur_offset = hole_start;
2938         while (1) {
2939                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2940                                 block_end - cur_offset, 0);
2941                 BUG_ON(IS_ERR(em) || !em);
2942                 last_byte = min(extent_map_end(em), block_end);
2943                 last_byte = (last_byte + mask) & ~mask;
2944                 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2945                         u64 hint_byte = 0;
2946                         hole_size = last_byte - cur_offset;
2947                         err = btrfs_drop_extents(trans, root, inode,
2948                                                  cur_offset,
2949                                                  cur_offset + hole_size,
2950                                                  block_end,
2951                                                  cur_offset, &hint_byte, 1);
2952                         if (err)
2953                                 break;
2954                         err = btrfs_insert_file_extent(trans, root,
2955                                         inode->i_ino, cur_offset, 0,
2956                                         0, hole_size, 0, hole_size,
2957                                         0, 0, 0);
2958                         btrfs_drop_extent_cache(inode, hole_start,
2959                                         last_byte - 1, 0);
2960                 }
2961                 free_extent_map(em);
2962                 cur_offset = last_byte;
2963                 if (err || cur_offset >= block_end)
2964                         break;
2965         }
2966
2967         btrfs_end_transaction(trans, root);
2968         unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2969         return err;
2970 }
2971
2972 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2973 {
2974         struct inode *inode = dentry->d_inode;
2975         int err;
2976
2977         err = inode_change_ok(inode, attr);
2978         if (err)
2979                 return err;
2980
2981         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
2982                 if (attr->ia_size > inode->i_size) {
2983                         err = btrfs_cont_expand(inode, attr->ia_size);
2984                         if (err)
2985                                 return err;
2986                 } else if (inode->i_size > 0 &&
2987                            attr->ia_size == 0) {
2988
2989                         /* we're truncating a file that used to have good
2990                          * data down to zero.  Make sure it gets into
2991                          * the ordered flush list so that any new writes
2992                          * get down to disk quickly.
2993                          */
2994                         BTRFS_I(inode)->ordered_data_close = 1;
2995                 }
2996         }
2997
2998         err = inode_setattr(inode, attr);
2999
3000         if (!err && ((attr->ia_valid & ATTR_MODE)))
3001                 err = btrfs_acl_chmod(inode);
3002         return err;
3003 }
3004
3005 void btrfs_delete_inode(struct inode *inode)
3006 {
3007         struct btrfs_trans_handle *trans;
3008         struct btrfs_root *root = BTRFS_I(inode)->root;
3009         unsigned long nr;
3010         int ret;
3011
3012         truncate_inode_pages(&inode->i_data, 0);
3013         if (is_bad_inode(inode)) {
3014                 btrfs_orphan_del(NULL, inode);
3015                 goto no_delete;
3016         }
3017         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3018
3019         btrfs_i_size_write(inode, 0);
3020         trans = btrfs_join_transaction(root, 1);
3021
3022         btrfs_set_trans_block_group(trans, inode);
3023         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
3024         if (ret) {
3025                 btrfs_orphan_del(NULL, inode);
3026                 goto no_delete_lock;
3027         }
3028
3029         btrfs_orphan_del(trans, inode);
3030
3031         nr = trans->blocks_used;
3032         clear_inode(inode);
3033
3034         btrfs_end_transaction(trans, root);
3035         btrfs_btree_balance_dirty(root, nr);
3036         return;
3037
3038 no_delete_lock:
3039         nr = trans->blocks_used;
3040         btrfs_end_transaction(trans, root);
3041         btrfs_btree_balance_dirty(root, nr);
3042 no_delete:
3043         clear_inode(inode);
3044 }
3045
3046 /*
3047  * this returns the key found in the dir entry in the location pointer.
3048  * If no dir entries were found, location->objectid is 0.
3049  */
3050 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3051                                struct btrfs_key *location)
3052 {
3053         const char *name = dentry->d_name.name;
3054         int namelen = dentry->d_name.len;
3055         struct btrfs_dir_item *di;
3056         struct btrfs_path *path;
3057         struct btrfs_root *root = BTRFS_I(dir)->root;
3058         int ret = 0;
3059
3060         path = btrfs_alloc_path();
3061         BUG_ON(!path);
3062
3063         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3064                                     namelen, 0);
3065         if (IS_ERR(di))
3066                 ret = PTR_ERR(di);
3067
3068         if (!di || IS_ERR(di))
3069                 goto out_err;
3070
3071         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3072 out:
3073         btrfs_free_path(path);
3074         return ret;
3075 out_err:
3076         location->objectid = 0;
3077         goto out;
3078 }
3079
3080 /*
3081  * when we hit a tree root in a directory, the btrfs part of the inode
3082  * needs to be changed to reflect the root directory of the tree root.  This
3083  * is kind of like crossing a mount point.
3084  */
3085 static int fixup_tree_root_location(struct btrfs_root *root,
3086                              struct btrfs_key *location,
3087                              struct btrfs_root **sub_root,
3088                              struct dentry *dentry)
3089 {
3090         struct btrfs_root_item *ri;
3091
3092         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
3093                 return 0;
3094         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
3095                 return 0;
3096
3097         *sub_root = btrfs_read_fs_root(root->fs_info, location,
3098                                         dentry->d_name.name,
3099                                         dentry->d_name.len);
3100         if (IS_ERR(*sub_root))
3101                 return PTR_ERR(*sub_root);
3102
3103         ri = &(*sub_root)->root_item;
3104         location->objectid = btrfs_root_dirid(ri);
3105         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3106         location->offset = 0;
3107
3108         return 0;
3109 }
3110
3111 static void inode_tree_add(struct inode *inode)
3112 {
3113         struct btrfs_root *root = BTRFS_I(inode)->root;
3114         struct btrfs_inode *entry;
3115         struct rb_node **p;
3116         struct rb_node *parent;
3117
3118 again:
3119         p = &root->inode_tree.rb_node;
3120         parent = NULL;
3121
3122         spin_lock(&root->inode_lock);
3123         while (*p) {
3124                 parent = *p;
3125                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3126
3127                 if (inode->i_ino < entry->vfs_inode.i_ino)
3128                         p = &parent->rb_left;
3129                 else if (inode->i_ino > entry->vfs_inode.i_ino)
3130                         p = &parent->rb_right;
3131                 else {
3132                         WARN_ON(!(entry->vfs_inode.i_state &
3133                                   (I_WILL_FREE | I_FREEING | I_CLEAR)));
3134                         rb_erase(parent, &root->inode_tree);
3135                         RB_CLEAR_NODE(parent);
3136                         spin_unlock(&root->inode_lock);
3137                         goto again;
3138                 }
3139         }
3140         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3141         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3142         spin_unlock(&root->inode_lock);
3143 }
3144
3145 static void inode_tree_del(struct inode *inode)
3146 {
3147         struct btrfs_root *root = BTRFS_I(inode)->root;
3148
3149         spin_lock(&root->inode_lock);
3150         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3151                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3152                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3153         }
3154         spin_unlock(&root->inode_lock);
3155 }
3156
3157 static noinline void init_btrfs_i(struct inode *inode)
3158 {
3159         struct btrfs_inode *bi = BTRFS_I(inode);
3160
3161         bi->generation = 0;
3162         bi->sequence = 0;
3163         bi->last_trans = 0;
3164         bi->logged_trans = 0;
3165         bi->delalloc_bytes = 0;
3166         bi->reserved_bytes = 0;
3167         bi->disk_i_size = 0;
3168         bi->flags = 0;
3169         bi->index_cnt = (u64)-1;
3170         bi->last_unlink_trans = 0;
3171         bi->ordered_data_close = 0;
3172         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3173         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3174                              inode->i_mapping, GFP_NOFS);
3175         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3176                              inode->i_mapping, GFP_NOFS);
3177         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3178         INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations);
3179         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3180         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3181         mutex_init(&BTRFS_I(inode)->extent_mutex);
3182         mutex_init(&BTRFS_I(inode)->log_mutex);
3183 }
3184
3185 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3186 {
3187         struct btrfs_iget_args *args = p;
3188         inode->i_ino = args->ino;
3189         init_btrfs_i(inode);
3190         BTRFS_I(inode)->root = args->root;
3191         btrfs_set_inode_space_info(args->root, inode);
3192         return 0;
3193 }
3194
3195 static int btrfs_find_actor(struct inode *inode, void *opaque)
3196 {
3197         struct btrfs_iget_args *args = opaque;
3198         return args->ino == inode->i_ino &&
3199                 args->root == BTRFS_I(inode)->root;
3200 }
3201
3202 static struct inode *btrfs_iget_locked(struct super_block *s,
3203                                        u64 objectid,
3204                                        struct btrfs_root *root)
3205 {
3206         struct inode *inode;
3207         struct btrfs_iget_args args;
3208         args.ino = objectid;
3209         args.root = root;
3210
3211         inode = iget5_locked(s, objectid, btrfs_find_actor,
3212                              btrfs_init_locked_inode,
3213                              (void *)&args);
3214         return inode;
3215 }