fs/btrfs/disk-io.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/fs.h>
  20 #include <linux/blkdev.h>
  21 #include <linux/scatterlist.h>
  22 #include <linux/swap.h>
  23 #include <linux/radix-tree.h>
  24 #include <linux/writeback.h>
  25 #include <linux/buffer_head.h>
  26 #include <linux/workqueue.h>
  27 #include <linux/kthread.h>
  28 #include <linux/freezer.h>
  29 #include <linux/crc32c.h>
  30 #include <linux/slab.h>
  31 #include <linux/migrate.h>
  32 #include <linux/ratelimit.h>
  33 #include <asm/unaligned.h>
  34 #include "compat.h"
  35 #include "ctree.h"
  36 #include "disk-io.h"
  37 #include "transaction.h"
  38 #include "btrfs_inode.h"
  39 #include "volumes.h"
  40 #include "print-tree.h"
  41 #include "async-thread.h"
  42 #include "locking.h"
  43 #include "tree-log.h"
  44 #include "free-space-cache.h"
  45 #include "inode-map.h"
  46
  47 static struct extent_io_ops btree_extent_io_ops;
  48 static void end_workqueue_fn(struct btrfs_work *work);
  49 static void free_fs_root(struct btrfs_root *root);
  50 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  51                                     int read_only);
  52 static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
  53 static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
  54 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  55                                       struct btrfs_root *root);
  56 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
  57 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  58 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
  59                                         struct extent_io_tree *dirty_pages,
  60                                         int mark);
  61 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
  62                                        struct extent_io_tree *pinned_extents);
  63 static int btrfs_cleanup_transaction(struct btrfs_root *root);
  64
  65 /*
  66  * end_io_wq structs are used to do processing in task context when an IO is
  67  * complete.  This is used during reads to verify checksums, and it is used
  68  * by writes to insert metadata for new file extents after IO is complete.
  69  */
  70 struct end_io_wq {
  71         struct bio *bio;
  72         bio_end_io_t *end_io;
  73         void *private;
  74         struct btrfs_fs_info *info;
  75         int error;
  76         int metadata;
  77         struct list_head list;
  78         struct btrfs_work work;
  79 };
  80
  81 /*
  82  * async submit bios are used to offload expensive checksumming
  83  * onto the worker threads.  They checksum file and metadata bios
  84  * just before they are sent down the IO stack.
  85  */
  86 struct async_submit_bio {
  87         struct inode *inode;
  88         struct bio *bio;
  89         struct list_head list;
  90         extent_submit_bio_hook_t *submit_bio_start;
  91         extent_submit_bio_hook_t *submit_bio_done;
  92         int rw;
  93         int mirror_num;
  94         unsigned long bio_flags;
  95         /*
  96          * bio_offset is optional, can be used if the pages in the bio
  97          * can't tell us where in the file the bio should go
  98          */
  99         u64 bio_offset;
 100         struct btrfs_work work;
 101 };
 102
 103 /*
 104  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 105  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 106  * the level the eb occupies in the tree.
 107  *
 108  * Different roots are used for different purposes and may nest inside each
 109  * other and they require separate keysets.  As lockdep keys should be
 110  * static, assign keysets according to the purpose of the root as indicated
 111  * by btrfs_root->objectid.  This ensures that all special purpose roots
 112  * have separate keysets.
 113  *
 114  * Lock-nesting across peer nodes is always done with the immediate parent
 115  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 116  * subclass to avoid triggering lockdep warning in such cases.
 117  *
 118  * The key is set by the readpage_end_io_hook after the buffer has passed
 119  * csum validation but before the pages are unlocked.  It is also set by
 120  * btrfs_init_new_buffer on freshly allocated blocks.
 121  *
 122  * We also add a check to make sure the highest level of the tree is the
 123  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 124  * needs update as well.
 125  */
 126 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 127 # if BTRFS_MAX_LEVEL != 8
 128 #  error
 129 # endif
 130
 131 static struct btrfs_lockdep_keyset {
 132         u64                     id;             /* root objectid */
 133         const char              *name_stem;     /* lock name stem */
 134         char                    names[BTRFS_MAX_LEVEL + 1][20];
 135         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 136 } btrfs_lockdep_keysets[] = {
 137         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 138         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 139         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 140         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 141         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 142         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 143         { .id = BTRFS_ORPHAN_OBJECTID,          .name_stem = "orphan"   },
 144         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 145         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 146         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 147         { .id = 0,                              .name_stem = "tree"     },
 148 };
 149
 150 void __init btrfs_init_lockdep(void)
 151 {
 152         int i, j;
 153
 154         /* initialize lockdep class names */
 155         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 156                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 157
 158                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 159                         snprintf(ks->names[j], sizeof(ks->names[j]),
 160                                  "btrfs-%s-%02d", ks->name_stem, j);
 161         }
 162 }
 163
 164 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 165                                     int level)
 166 {
 167         struct btrfs_lockdep_keyset *ks;
 168
 169         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 170
 171         /* find the matching keyset, id 0 is the default entry */
 172         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 173                 if (ks->id == objectid)
 174                         break;
 175
 176         lockdep_set_class_and_name(&eb->lock,
 177                                    &ks->keys[level], ks->names[level]);
 178 }
 179
 180 #endif
 181
 182 /*
 183  * extents on the btree inode are pretty simple, there's one extent
 184  * that covers the entire device
 185  */
 186 static struct extent_map *btree_get_extent(struct inode *inode,
 187                 struct page *page, size_t pg_offset, u64 start, u64 len,
 188                 int create)
 189 {
 190         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 191         struct extent_map *em;
 192         int ret;
 193
 194         read_lock(&em_tree->lock);
 195         em = lookup_extent_mapping(em_tree, start, len);
 196         if (em) {
 197                 em->bdev =
 198                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 199                 read_unlock(&em_tree->lock);
 200                 goto out;
 201         }
 202         read_unlock(&em_tree->lock);
 203
 204         em = alloc_extent_map();
 205         if (!em) {
 206                 em = ERR_PTR(-ENOMEM);
 207                 goto out;
 208         }
 209         em->start = 0;
 210         em->len = (u64)-1;
 211         em->block_len = (u64)-1;
 212         em->block_start = 0;
 213         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 214
 215         write_lock(&em_tree->lock);
 216         ret = add_extent_mapping(em_tree, em);
 217         if (ret == -EEXIST) {
 218                 u64 failed_start = em->start;
 219                 u64 failed_len = em->len;
 220
 221                 free_extent_map(em);
 222                 em = lookup_extent_mapping(em_tree, start, len);
 223                 if (em) {
 224                         ret = 0;
 225                 } else {
 226                         em = lookup_extent_mapping(em_tree, failed_start,
 227                                                    failed_len);
 228                         ret = -EIO;
 229                 }
 230         } else if (ret) {
 231                 free_extent_map(em);
 232                 em = NULL;
 233         }
 234         write_unlock(&em_tree->lock);
 235
 236         if (ret)
 237                 em = ERR_PTR(ret);
 238 out:
 239         return em;
 240 }
 241
 242 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 243 {
 244         return crc32c(seed, data, len);
 245 }
 246
 247 void btrfs_csum_final(u32 crc, char *result)
 248 {
 249         put_unaligned_le32(~crc, result);
 250 }
 251
 252 /*
 253  * compute the csum for a btree block, and either verify it or write it
 254  * into the csum field of the block.
 255  */
 256 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 257                            int verify)
 258 {
 259         u16 csum_size =
 260                 btrfs_super_csum_size(&root->fs_info->super_copy);
 261         char *result = NULL;
 262         unsigned long len;
 263         unsigned long cur_len;
 264         unsigned long offset = BTRFS_CSUM_SIZE;
 265         char *kaddr;
 266         unsigned long map_start;
 267         unsigned long map_len;
 268         int err;
 269         u32 crc = ~(u32)0;
 270         unsigned long inline_result;
 271
 272         len = buf->len - offset;
 273         while (len > 0) {
 274                 err = map_private_extent_buffer(buf, offset, 32,
 275                                         &kaddr, &map_start, &map_len);
 276                 if (err)
 277                         return 1;
 278                 cur_len = min(len, map_len - (offset - map_start));
 279                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 280                                       crc, cur_len);
 281                 len -= cur_len;
 282                 offset += cur_len;
 283         }
 284         if (csum_size > sizeof(inline_result)) {
 285                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
 286                 if (!result)
 287                         return 1;
 288         } else {
 289                 result = (char *)&inline_result;
 290         }
 291
 292         btrfs_csum_final(crc, result);
 293
 294         if (verify) {
 295                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 296                         u32 val;
 297                         u32 found = 0;
 298                         memcpy(&found, result, csum_size);
 299
 300                         read_extent_buffer(buf, &val, 0, csum_size);
 301                         printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
 302                                        "failed on %llu wanted %X found %X "
 303                                        "level %d\n",
 304                                        root->fs_info->sb->s_id,
 305                                        (unsigned long long)buf->start, val, found,
 306                                        btrfs_header_level(buf));
 307                         if (result != (char *)&inline_result)
 308                                 kfree(result);
 309                         return 1;
 310                 }
 311         } else {
 312                 write_extent_buffer(buf, result, 0, csum_size);
 313         }
 314         if (result != (char *)&inline_result)
 315                 kfree(result);
 316         return 0;
 317 }
 318
 319 /*
 320  * we can't consider a given block up to date unless the transid of the
 321  * block matches the transid in the parent node's pointer.  This is how we
 322  * detect blocks that either didn't get written at all or got written
 323  * in the wrong place.
 324  */
 325 static int verify_parent_transid(struct extent_io_tree *io_tree,
 326                                  struct extent_buffer *eb, u64 parent_transid)
 327 {
 328         struct extent_state *cached_state = NULL;
 329         int ret;
 330
 331         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 332                 return 0;
 333
 334         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 335                          0, &cached_state, GFP_NOFS);
 336         if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
 337             btrfs_header_generation(eb) == parent_transid) {
 338                 ret = 0;
 339                 goto out;
 340         }
 341         printk_ratelimited("parent transid verify failed on %llu wanted %llu "
 342                        "found %llu\n",
 343                        (unsigned long long)eb->start,
 344                        (unsigned long long)parent_transid,
 345                        (unsigned long long)btrfs_header_generation(eb));
 346         ret = 1;
 347         clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
 348 out:
 349         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 350                              &cached_state, GFP_NOFS);
 351         return ret;
 352 }
 353
 354 /*
 355  * helper to read a given tree block, doing retries as required when
 356  * the checksums don't match and we have alternate mirrors to try.
 357  */
 358 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 359                                           struct extent_buffer *eb,
 360                                           u64 start, u64 parent_transid)
 361 {
 362         struct extent_io_tree *io_tree;
 363         int ret;
 364         int num_copies = 0;
 365         int mirror_num = 0;
 366
 367         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 368         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 369         while (1) {
 370                 ret = read_extent_buffer_pages(io_tree, eb, start,
 371                                                WAIT_COMPLETE,
 372                                                btree_get_extent, mirror_num);
 373                 if (!ret &&
 374                     !verify_parent_transid(io_tree, eb, parent_transid))
 375                         return ret;
 376
 377                 /*
 378                  * This buffer's crc is fine, but its contents are corrupted, so
 379                  * there is no reason to read the other copies, they won't be
 380                  * any less wrong.
 381                  */
 382                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 383                         return ret;
 384
 385                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 386                                               eb->start, eb->len);
 387                 if (num_copies == 1)
 388                         return ret;
 389
 390                 mirror_num++;
 391                 if (mirror_num > num_copies)
 392                         return ret;
 393         }
 394         return -EIO;
 395 }
 396
 397 /*
 398  * checksum a dirty tree block before IO.  This has extra checks to make sure
 399  * we only fill in the checksum field in the first page of a multi-page block
 400  */
 401
 402 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 403 {
 404         struct extent_io_tree *tree;
 405         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 406         u64 found_start;
 407         unsigned long len;
 408         struct extent_buffer *eb;
 409         int ret;
 410
 411         tree = &BTRFS_I(page->mapping->host)->io_tree;
 412
 413         if (page->private == EXTENT_PAGE_PRIVATE) {
 414                 WARN_ON(1);
 415                 goto out;
 416         }
 417         if (!page->private) {
 418                 WARN_ON(1);
 419                 goto out;
 420         }
 421         len = page->private >> 2;
 422         WARN_ON(len == 0);
 423
 424         eb = alloc_extent_buffer(tree, start, len, page);
 425         if (eb == NULL) {
 426                 WARN_ON(1);
 427                 goto out;
 428         }
 429         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 430                                              btrfs_header_generation(eb));
 431         BUG_ON(ret);
 432         WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
 433
 434         found_start = btrfs_header_bytenr(eb);
 435         if (found_start != start) {
 436                 WARN_ON(1);
 437                 goto err;
 438         }
 439         if (eb->first_page != page) {
 440                 WARN_ON(1);
 441                 goto err;
 442         }
 443         if (!PageUptodate(page)) {
 444                 WARN_ON(1);
 445                 goto err;
 446         }
 447         csum_tree_block(root, eb, 0);
 448 err:
 449         free_extent_buffer(eb);
 450 out:
 451         return 0;
 452 }
 453
 454 static int check_tree_block_fsid(struct btrfs_root *root,
 455                                  struct extent_buffer *eb)
 456 {
 457         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 458         u8 fsid[BTRFS_UUID_SIZE];
 459         int ret = 1;
 460
 461         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
 462                            BTRFS_FSID_SIZE);
 463         while (fs_devices) {
 464                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 465                         ret = 0;
 466                         break;
 467                 }
 468                 fs_devices = fs_devices->seed;
 469         }
 470         return ret;
 471 }
 472
 473 #define CORRUPT(reason, eb, root, slot)                         \
 474         printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
 475                "root=%llu, slot=%d\n", reason,                  \
 476                (unsigned long long)btrfs_header_bytenr(eb),     \
 477                (unsigned long long)root->objectid, slot)
 478
 479 static noinline int check_leaf(struct btrfs_root *root,
 480                                struct extent_buffer *leaf)
 481 {
 482         struct btrfs_key key;
 483         struct btrfs_key leaf_key;
 484         u32 nritems = btrfs_header_nritems(leaf);
 485         int slot;
 486
 487         if (nritems == 0)
 488                 return 0;
 489
 490         /* Check the 0 item */
 491         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 492             BTRFS_LEAF_DATA_SIZE(root)) {
 493                 CORRUPT("invalid item offset size pair", leaf, root, 0);
 494                 return -EIO;
 495         }
 496
 497         /*
 498          * Check to make sure each items keys are in the correct order and their
 499          * offsets make sense.  We only have to loop through nritems-1 because
 500          * we check the current slot against the next slot, which verifies the
 501          * next slot's offset+size makes sense and that the current's slot
 502          * offset is correct.
 503          */
 504         for (slot = 0; slot < nritems - 1; slot++) {
 505                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 506                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 507
 508                 /* Make sure the keys are in the right order */
 509                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 510                         CORRUPT("bad key order", leaf, root, slot);
 511                         return -EIO;
 512                 }
 513
 514                 /*
 515                  * Make sure the offset and ends are right, remember that the
 516                  * item data starts at the end of the leaf and grows towards the
 517                  * front.
 518                  */
 519                 if (btrfs_item_offset_nr(leaf, slot) !=
 520                         btrfs_item_end_nr(leaf, slot + 1)) {
 521                         CORRUPT("slot offset bad", leaf, root, slot);
 522                         return -EIO;
 523                 }
 524
 525                 /*
 526                  * Check to make sure that we don't point outside of the leaf,
 527                  * just incase all the items are consistent to eachother, but
 528                  * all point outside of the leaf.
 529                  */
 530                 if (btrfs_item_end_nr(leaf, slot) >
 531                     BTRFS_LEAF_DATA_SIZE(root)) {
 532                         CORRUPT("slot end outside of leaf", leaf, root, slot);
 533                         return -EIO;
 534                 }
 535         }
 536
 537         return 0;
 538 }
 539
 540 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 541                                struct extent_state *state)
 542 {
 543         struct extent_io_tree *tree;
 544         u64 found_start;
 545         int found_level;
 546         unsigned long len;
 547         struct extent_buffer *eb;
 548         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 549         int ret = 0;
 550
 551         tree = &BTRFS_I(page->mapping->host)->io_tree;
 552         if (page->private == EXTENT_PAGE_PRIVATE)
 553                 goto out;
 554         if (!page->private)
 555                 goto out;
 556
 557         len = page->private >> 2;
 558         WARN_ON(len == 0);
 559
 560         eb = alloc_extent_buffer(tree, start, len, page);
 561         if (eb == NULL) {
 562                 ret = -EIO;
 563                 goto out;
 564         }
 565
 566         found_start = btrfs_header_bytenr(eb);
 567         if (found_start != start) {
 568                 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
 569                                "%llu %llu\n",
 570                                (unsigned long long)found_start,
 571                                (unsigned long long)eb->start);
 572                 ret = -EIO;
 573                 goto err;
 574         }
 575         if (eb->first_page != page) {
 576                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
 577                        eb->first_page->index, page->index);
 578                 WARN_ON(1);
 579                 ret = -EIO;
 580                 goto err;
 581         }
 582         if (check_tree_block_fsid(root, eb)) {
 583                 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
 584                                (unsigned long long)eb->start);
 585                 ret = -EIO;
 586                 goto err;
 587         }
 588         found_level = btrfs_header_level(eb);
 589
 590         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 591                                        eb, found_level);
 592
 593         ret = csum_tree_block(root, eb, 1);
 594         if (ret) {
 595                 ret = -EIO;
 596                 goto err;
 597         }
 598
 599         /*
 600          * If this is a leaf block and it is corrupt, set the corrupt bit so
 601          * that we don't try and read the other copies of this block, just
 602          * return -EIO.
 603          */
 604         if (found_level == 0 && check_leaf(root, eb)) {
 605                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 606                 ret = -EIO;
 607         }
 608
 609         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 610         end = eb->start + end - 1;
 611 err:
 612         if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
 613                 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
 614                 btree_readahead_hook(root, eb, eb->start, ret);
 615         }
 616
 617         free_extent_buffer(eb);
 618 out:
 619         return ret;
 620 }
 621
 622 static int btree_io_failed_hook(struct bio *failed_bio,
 623                          struct page *page, u64 start, u64 end,
 624                          struct extent_state *state)
 625 {
 626         struct extent_io_tree *tree;
 627         unsigned long len;
 628         struct extent_buffer *eb;
 629         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 630
 631         tree = &BTRFS_I(page->mapping->host)->io_tree;
 632         if (page->private == EXTENT_PAGE_PRIVATE)
 633                 goto out;
 634         if (!page->private)
 635                 goto out;
 636
 637         len = page->private >> 2;
 638         WARN_ON(len == 0);
 639
 640         eb = alloc_extent_buffer(tree, start, len, page);
 641         if (eb == NULL)
 642                 goto out;
 643
 644         if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
 645                 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
 646                 btree_readahead_hook(root, eb, eb->start, -EIO);
 647         }
 648
 649 out:
 650         return -EIO;    /* we fixed nothing */
 651 }
 652
 653 static void end_workqueue_bio(struct bio *bio, int err)
 654 {
 655         struct end_io_wq *end_io_wq = bio->bi_private;
 656         struct btrfs_fs_info *fs_info;
 657
 658         fs_info = end_io_wq->info;
 659         end_io_wq->error = err;
 660         end_io_wq->work.func = end_workqueue_fn;
 661         end_io_wq->work.flags = 0;
 662
 663         if (bio->bi_rw & REQ_WRITE) {
 664                 if (end_io_wq->metadata == 1)
 665                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
 666                                            &end_io_wq->work);
 667                 else if (end_io_wq->metadata == 2)
 668                         btrfs_queue_worker(&fs_info->endio_freespace_worker,
 669                                            &end_io_wq->work);
 670                 else
 671                         btrfs_queue_worker(&fs_info->endio_write_workers,
 672                                            &end_io_wq->work);
 673         } else {
 674                 if (end_io_wq->metadata)
 675                         btrfs_queue_worker(&fs_info->endio_meta_workers,
 676                                            &end_io_wq->work);
 677                 else
 678                         btrfs_queue_worker(&fs_info->endio_workers,
 679                                            &end_io_wq->work);
 680         }
 681 }
 682
 683 /*
 684  * For the metadata arg you want
 685  *
 686  * 0 - if data
 687  * 1 - if normal metadta
 688  * 2 - if writing to the free space cache area
 689  */
 690 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 691                         int metadata)
 692 {
 693         struct end_io_wq *end_io_wq;
 694         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 695         if (!end_io_wq)
 696                 return -ENOMEM;
 697
 698         end_io_wq->private = bio->bi_private;
 699         end_io_wq->end_io = bio->bi_end_io;
 700         end_io_wq->info = info;
 701         end_io_wq->error = 0;
 702         end_io_wq->bio = bio;
 703         end_io_wq->metadata = metadata;
 704
 705         bio->bi_private = end_io_wq;
 706         bio->bi_end_io = end_workqueue_bio;
 707         return 0;
 708 }
 709
 710 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 711 {
 712         unsigned long limit = min_t(unsigned long,
 713                                     info->workers.max_workers,
 714                                     info->fs_devices->open_devices);
 715         return 256 * limit;
 716 }
 717
 718 static void run_one_async_start(struct btrfs_work *work)
 719 {
 720         struct async_submit_bio *async;
 721
 722         async = container_of(work, struct  async_submit_bio, work);
 723         async->submit_bio_start(async->inode, async->rw, async->bio,
 724                                async->mirror_num, async->bio_flags,
 725                                async->bio_offset);
 726 }
 727
 728 static void run_one_async_done(struct btrfs_work *work)
 729 {
 730         struct btrfs_fs_info *fs_info;
 731         struct async_submit_bio *async;
 732         int limit;
 733
 734         async = container_of(work, struct  async_submit_bio, work);
 735         fs_info = BTRFS_I(async->inode)->root->fs_info;
 736
 737         limit = btrfs_async_submit_limit(fs_info);
 738         limit = limit * 2 / 3;
 739
 740         atomic_dec(&fs_info->nr_async_submits);
 741
 742         if (atomic_read(&fs_info->nr_async_submits) < limit &&
 743             waitqueue_active(&fs_info->async_submit_wait))
 744                 wake_up(&fs_info->async_submit_wait);
 745
 746         async->submit_bio_done(async->inode, async->rw, async->bio,
 747                                async->mirror_num, async->bio_flags,
 748                                async->bio_offset);
 749 }
 750
 751 static void run_one_async_free(struct btrfs_work *work)
 752 {
 753         struct async_submit_bio *async;
 754
 755         async = container_of(work, struct  async_submit_bio, work);
 756         kfree(async);
 757 }
 758
 759 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 760                         int rw, struct bio *bio, int mirror_num,
 761                         unsigned long bio_flags,
 762                         u64 bio_offset,
 763                         extent_submit_bio_hook_t *submit_bio_start,
 764                         extent_submit_bio_hook_t *submit_bio_done)
 765 {
 766         struct async_submit_bio *async;
 767
 768         async = kmalloc(sizeof(*async), GFP_NOFS);
 769         if (!async)
 770                 return -ENOMEM;
 771
 772         async->inode = inode;
 773         async->rw = rw;
 774         async->bio = bio;
 775         async->mirror_num = mirror_num;
 776         async->submit_bio_start = submit_bio_start;
 777         async->submit_bio_done = submit_bio_done;
 778
 779         async->work.func = run_one_async_start;
 780         async->work.ordered_func = run_one_async_done;
 781         async->work.ordered_free = run_one_async_free;
 782
 783         async->work.flags = 0;
 784         async->bio_flags = bio_flags;
 785         async->bio_offset = bio_offset;
 786
 787         atomic_inc(&fs_info->nr_async_submits);
 788
 789         if (rw & REQ_SYNC)
 790                 btrfs_set_work_high_prio(&async->work);
 791
 792         btrfs_queue_worker(&fs_info->workers, &async->work);
 793
 794         while (atomic_read(&fs_info->async_submit_draining) &&
 795               atomic_read(&fs_info->nr_async_submits)) {
 796                 wait_event(fs_info->async_submit_wait,
 797                            (atomic_read(&fs_info->nr_async_submits) == 0));
 798         }
 799
 800         return 0;
 801 }
 802
 803 static int btree_csum_one_bio(struct bio *bio)
 804 {
 805         struct bio_vec *bvec = bio->bi_io_vec;
 806         int bio_index = 0;
 807         struct btrfs_root *root;
 808
 809         WARN_ON(bio->bi_vcnt <= 0);
 810         while (bio_index < bio->bi_vcnt) {
 811                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 812                 csum_dirty_buffer(root, bvec->bv_page);
 813                 bio_index++;
 814                 bvec++;
 815         }
 816         return 0;
 817 }
 818
 819 static int __btree_submit_bio_start(struct inode *inode, int rw,
 820                                     struct bio *bio, int mirror_num,
 821                                     unsigned long bio_flags,
 822                                     u64 bio_offset)
 823 {
 824         /*
 825          * when we're called for a write, we're already in the async
 826          * submission context.  Just jump into btrfs_map_bio
 827          */
 828         btree_csum_one_bio(bio);
 829         return 0;
 830 }
 831
 832 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 833                                  int mirror_num, unsigned long bio_flags,
 834                                  u64 bio_offset)
 835 {
 836         /*
 837          * when we're called for a write, we're already in the async
 838          * submission context.  Just jump into btrfs_map_bio
 839          */
 840         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 841 }
 842
 843 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 844                                  int mirror_num, unsigned long bio_flags,
 845                                  u64 bio_offset)
 846 {
 847         int ret;
 848
 849         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 850                                           bio, 1);
 851         BUG_ON(ret);
 852
 853         if (!(rw & REQ_WRITE)) {
 854                 /*
 855                  * called for a read, do the setup so that checksum validation
 856                  * can happen in the async kernel threads
 857                  */
 858                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 859                                      mirror_num, 0);
 860         }
 861
 862         /*
 863          * kthread helpers are used to submit writes so that checksumming
 864          * can happen in parallel across all CPUs
 865          */
 866         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 867                                    inode, rw, bio, mirror_num, 0,
 868                                    bio_offset,
 869                                    __btree_submit_bio_start,
 870                                    __btree_submit_bio_done);
 871 }
 872
 873 #ifdef CONFIG_MIGRATION
 874 static int btree_migratepage(struct address_space *mapping,
 875                         struct page *newpage, struct page *page)
 876 {
 877         /*
 878          * we can't safely write a btree page from here,
 879          * we haven't done the locking hook
 880          */
 881         if (PageDirty(page))
 882                 return -EAGAIN;
 883         /*
 884          * Buffers may be managed in a filesystem specific way.
 885          * We must have no buffers or drop them.
 886          */
 887         if (page_has_private(page) &&
 888             !try_to_release_page(page, GFP_KERNEL))
 889                 return -EAGAIN;
 890         return migrate_page(mapping, newpage, page);
 891 }
 892 #endif
 893
 894 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 895 {
 896         struct extent_io_tree *tree;
 897         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 898         struct extent_buffer *eb;
 899         int was_dirty;
 900
 901         tree = &BTRFS_I(page->mapping->host)->io_tree;
 902         if (!(current->flags & PF_MEMALLOC)) {
 903                 return extent_write_full_page(tree, page,
 904                                               btree_get_extent, wbc);
 905         }
 906
 907         redirty_page_for_writepage(wbc, page);
 908         eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
 909         WARN_ON(!eb);
 910
 911         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
 912         if (!was_dirty) {
 913                 spin_lock(&root->fs_info->delalloc_lock);
 914                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
 915                 spin_unlock(&root->fs_info->delalloc_lock);
 916         }
 917         free_extent_buffer(eb);
 918
 919         unlock_page(page);
 920         return 0;
 921 }
 922
 923 static int btree_writepages(struct address_space *mapping,
 924                             struct writeback_control *wbc)
 925 {
 926         struct extent_io_tree *tree;
 927         tree = &BTRFS_I(mapping->host)->io_tree;
 928         if (wbc->sync_mode == WB_SYNC_NONE) {
 929                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
 930                 u64 num_dirty;
 931                 unsigned long thresh = 32 * 1024 * 1024;
 932
 933                 if (wbc->for_kupdate)
 934                         return 0;
 935
 936                 /* this is a bit racy, but that's ok */
 937                 num_dirty = root->fs_info->dirty_metadata_bytes;
 938                 if (num_dirty < thresh)
 939                         return 0;
 940         }
 941         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 942 }
 943
 944 static int btree_readpage(struct file *file, struct page *page)
 945 {
 946         struct extent_io_tree *tree;
 947         tree = &BTRFS_I(page->mapping->host)->io_tree;
 948         return extent_read_full_page(tree, page, btree_get_extent);
 949 }
 950
 951 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 952 {
 953         struct extent_io_tree *tree;
 954         struct extent_map_tree *map;
 955         int ret;
 956
 957         if (PageWriteback(page) || PageDirty(page))
 958                 return 0;
 959
 960         tree = &BTRFS_I(page->mapping->host)->io_tree;
 961         map = &BTRFS_I(page->mapping->host)->extent_tree;
 962
 963         ret = try_release_extent_state(map, tree, page, gfp_flags);
 964         if (!ret)
 965                 return 0;
 966
 967         ret = try_release_extent_buffer(tree, page);
 968         if (ret == 1) {
 969                 ClearPagePrivate(page);
 970                 set_page_private(page, 0);
 971                 page_cache_release(page);
 972         }
 973
 974         return ret;
 975 }
 976
 977 static void btree_invalidatepage(struct page *page, unsigned long offset)
 978 {
 979         struct extent_io_tree *tree;
 980         tree = &BTRFS_I(page->mapping->host)->io_tree;
 981         extent_invalidatepage(tree, page, offset);
 982         btree_releasepage(page, GFP_NOFS);
 983         if (PagePrivate(page)) {
 984                 printk(KERN_WARNING "btrfs warning page private not zero "
 985                        "on page %llu\n", (unsigned long long)page_offset(page));
 986                 ClearPagePrivate(page);
 987                 set_page_private(page, 0);
 988                 page_cache_release(page);
 989         }
 990 }
 991
 992 static const struct address_space_operations btree_aops = {
 993         .readpage       = btree_readpage,
 994         .writepage      = btree_writepage,
 995         .writepages     = btree_writepages,
 996         .releasepage    = btree_releasepage,
 997         .invalidatepage = btree_invalidatepage,
 998 #ifdef CONFIG_MIGRATION
 999         .migratepage    = btree_migratepage,
1000 #endif
1001 };
1002
1003 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1004                          u64 parent_transid)
1005 {
1006         struct extent_buffer *buf = NULL;
1007         struct inode *btree_inode = root->fs_info->btree_inode;
1008         int ret = 0;
1009
1010         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1011         if (!buf)
1012                 return 0;
1013         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1014                                  buf, 0, WAIT_NONE, btree_get_extent, 0);
1015         free_extent_buffer(buf);
1016         return ret;
1017 }
1018
1019 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1020                          int mirror_num, struct extent_buffer **eb)
1021 {
1022         struct extent_buffer *buf = NULL;
1023         struct inode *btree_inode = root->fs_info->btree_inode;
1024         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1025         int ret;
1026
1027         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1028         if (!buf)
1029                 return 0;
1030
1031         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1032
1033         ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1034                                        btree_get_extent, mirror_num);
1035         if (ret) {
1036                 free_extent_buffer(buf);
1037                 return ret;
1038         }
1039
1040         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1041                 free_extent_buffer(buf);
1042                 return -EIO;
1043         } else if (extent_buffer_uptodate(io_tree, buf, NULL)) {
1044                 *eb = buf;
1045         } else {
1046                 free_extent_buffer(buf);
1047         }
1048         return 0;
1049 }
1050
1051 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1052                                             u64 bytenr, u32 blocksize)
1053 {
1054         struct inode *btree_inode = root->fs_info->btree_inode;
1055         struct extent_buffer *eb;
1056         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1057                                 bytenr, blocksize);
1058         return eb;
1059 }
1060
1061 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1062                                                  u64 bytenr, u32 blocksize)
1063 {
1064         struct inode *btree_inode = root->fs_info->btree_inode;
1065         struct extent_buffer *eb;
1066
1067         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1068                                  bytenr, blocksize, NULL);
1069         return eb;
1070 }
1071
1072
1073 int btrfs_write_tree_block(struct extent_buffer *buf)
1074 {
1075         return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
1076                                         buf->start + buf->len - 1);
1077 }
1078
1079 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1080 {
1081         return filemap_fdatawait_range(buf->first_page->mapping,
1082                                        buf->start, buf->start + buf->len - 1);
1083 }
1084
1085 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1086                                       u32 blocksize, u64 parent_transid)
1087 {
1088         struct extent_buffer *buf = NULL;
1089         int ret;
1090
1091         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1092         if (!buf)
1093                 return NULL;
1094
1095         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1096
1097         if (ret == 0)
1098                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1099         return buf;
1100
1101 }
1102
1103 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1104                      struct extent_buffer *buf)
1105 {
1106         struct inode *btree_inode = root->fs_info->btree_inode;
1107         if (btrfs_header_generation(buf) ==
1108             root->fs_info->running_transaction->transid) {
1109                 btrfs_assert_tree_locked(buf);
1110
1111                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1112                         spin_lock(&root->fs_info->delalloc_lock);
1113                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
1114                                 root->fs_info->dirty_metadata_bytes -= buf->len;
1115                         else
1116                                 WARN_ON(1);
1117                         spin_unlock(&root->fs_info->delalloc_lock);
1118                 }
1119
1120                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1121                 btrfs_set_lock_blocking(buf);
1122                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1123                                           buf);
1124         }
1125         return 0;
1126 }
1127
1128 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1129                         u32 stripesize, struct btrfs_root *root,
1130                         struct btrfs_fs_info *fs_info,
1131                         u64 objectid)
1132 {
1133         root->node = NULL;
1134         root->commit_root = NULL;
1135         root->sectorsize = sectorsize;
1136         root->nodesize = nodesize;
1137         root->leafsize = leafsize;
1138         root->stripesize = stripesize;
1139         root->ref_cows = 0;
1140         root->track_dirty = 0;
1141         root->in_radix = 0;
1142         root->orphan_item_inserted = 0;
1143         root->orphan_cleanup_state = 0;
1144
1145         root->fs_info = fs_info;
1146         root->objectid = objectid;
1147         root->last_trans = 0;
1148         root->highest_objectid = 0;
1149         root->name = NULL;
1150         root->inode_tree = RB_ROOT;
1151         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1152         root->block_rsv = NULL;
1153         root->orphan_block_rsv = NULL;
1154
1155         INIT_LIST_HEAD(&root->dirty_list);
1156         INIT_LIST_HEAD(&root->orphan_list);
1157         INIT_LIST_HEAD(&root->root_list);
1158         spin_lock_init(&root->orphan_lock);
1159         spin_lock_init(&root->inode_lock);
1160         spin_lock_init(&root->accounting_lock);
1161         mutex_init(&root->objectid_mutex);
1162         mutex_init(&root->log_mutex);
1163         init_waitqueue_head(&root->log_writer_wait);
1164         init_waitqueue_head(&root->log_commit_wait[0]);
1165         init_waitqueue_head(&root->log_commit_wait[1]);
1166         atomic_set(&root->log_commit[0], 0);
1167         atomic_set(&root->log_commit[1], 0);
1168         atomic_set(&root->log_writers, 0);
1169         root->log_batch = 0;
1170         root->log_transid = 0;
1171         root->last_log_commit = 0;
1172         extent_io_tree_init(&root->dirty_log_pages,
1173                              fs_info->btree_inode->i_mapping);
1174
1175         memset(&root->root_key, 0, sizeof(root->root_key));
1176         memset(&root->root_item, 0, sizeof(root->root_item));
1177         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1178         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1179         root->defrag_trans_start = fs_info->generation;
1180         init_completion(&root->kobj_unregister);
1181         root->defrag_running = 0;
1182         root->root_key.objectid = objectid;
1183         root->anon_dev = 0;
1184         return 0;
1185 }
1186
1187 static int find_and_setup_root(struct btrfs_root *tree_root,
1188                                struct btrfs_fs_info *fs_info,
1189                                u64 objectid,
1190                                struct btrfs_root *root)
1191 {
1192         int ret;
1193         u32 blocksize;
1194         u64 generation;
1195
1196         __setup_root(tree_root->nodesize, tree_root->leafsize,
1197                      tree_root->sectorsize, tree_root->stripesize,
1198                      root, fs_info, objectid);
1199         ret = btrfs_find_last_root(tree_root, objectid,
1200                                    &root->root_item, &root->root_key);
1201         if (ret > 0)
1202                 return -ENOENT;
1203         BUG_ON(ret);
1204
1205         generation = btrfs_root_generation(&root->root_item);
1206         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1207         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1208                                      blocksize, generation);
1209         if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1210                 free_extent_buffer(root->node);
1211                 return -EIO;
1212         }
1213         root->commit_root = btrfs_root_node(root);
1214         return 0;
1215 }
1216
1217 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1218                                          struct btrfs_fs_info *fs_info)
1219 {
1220         struct btrfs_root *root;
1221         struct btrfs_root *tree_root = fs_info->tree_root;
1222         struct extent_buffer *leaf;
1223
1224         root = kzalloc(sizeof(*root), GFP_NOFS);
1225         if (!root)
1226                 return ERR_PTR(-ENOMEM);
1227
1228         __setup_root(tree_root->nodesize, tree_root->leafsize,
1229                      tree_root->sectorsize, tree_root->stripesize,
1230                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1231
1232         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1233         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1234         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1235         /*
1236          * log trees do not get reference counted because they go away
1237          * before a real commit is actually done.  They do store pointers
1238          * to file data extents, and those reference counts still get
1239          * updated (along with back refs to the log tree).
1240          */
1241         root->ref_cows = 0;
1242
1243         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1244                                       BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1245         if (IS_ERR(leaf)) {
1246                 kfree(root);
1247                 return ERR_CAST(leaf);
1248         }
1249
1250         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1251         btrfs_set_header_bytenr(leaf, leaf->start);
1252         btrfs_set_header_generation(leaf, trans->transid);
1253         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1254         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1255         root->node = leaf;
1256
1257         write_extent_buffer(root->node, root->fs_info->fsid,
1258                             (unsigned long)btrfs_header_fsid(root->node),
1259                             BTRFS_FSID_SIZE);
1260         btrfs_mark_buffer_dirty(root->node);
1261         btrfs_tree_unlock(root->node);
1262         return root;
1263 }
1264
1265 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1266                              struct btrfs_fs_info *fs_info)
1267 {
1268         struct btrfs_root *log_root;
1269
1270         log_root = alloc_log_tree(trans, fs_info);
1271         if (IS_ERR(log_root))
1272                 return PTR_ERR(log_root);
1273         WARN_ON(fs_info->log_root_tree);
1274         fs_info->log_root_tree = log_root;
1275         return 0;
1276 }
1277
1278 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1279                        struct btrfs_root *root)
1280 {
1281         struct btrfs_root *log_root;
1282         struct btrfs_inode_item *inode_item;
1283
1284         log_root = alloc_log_tree(trans, root->fs_info);
1285         if (IS_ERR(log_root))
1286                 return PTR_ERR(log_root);
1287
1288         log_root->last_trans = trans->transid;
1289         log_root->root_key.offset = root->root_key.objectid;
1290
1291         inode_item = &log_root->root_item.inode;
1292         inode_item->generation = cpu_to_le64(1);
1293         inode_item->size = cpu_to_le64(3);
1294         inode_item->nlink = cpu_to_le32(1);
1295         inode_item->nbytes = cpu_to_le64(root->leafsize);
1296         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1297
1298         btrfs_set_root_node(&log_root->root_item, log_root->node);
1299
1300         WARN_ON(root->log_root);
1301         root->log_root = log_root;
1302         root->log_transid = 0;
1303         root->last_log_commit = 0;
1304         return 0;
1305 }
1306
1307 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1308                                                struct btrfs_key *location)
1309 {
1310         struct btrfs_root *root;
1311         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1312         struct btrfs_path *path;
1313         struct extent_buffer *l;
1314         u64 generation;
1315         u32 blocksize;
1316         int ret = 0;
1317
1318         root = kzalloc(sizeof(*root), GFP_NOFS);
1319         if (!root)
1320                 return ERR_PTR(-ENOMEM);
1321         if (location->offset == (u64)-1) {
1322                 ret = find_and_setup_root(tree_root, fs_info,
1323                                           location->objectid, root);
1324                 if (ret) {
1325                         kfree(root);
1326                         return ERR_PTR(ret);
1327                 }
1328                 goto out;
1329         }
1330
1331         __setup_root(tree_root->nodesize, tree_root->leafsize,
1332                      tree_root->sectorsize, tree_root->stripesize,
1333                      root, fs_info, location->objectid);
1334
1335         path = btrfs_alloc_path();
1336         if (!path) {
1337                 kfree(root);
1338                 return ERR_PTR(-ENOMEM);
1339         }
1340         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1341         if (ret == 0) {
1342                 l = path->nodes[0];
1343                 read_extent_buffer(l, &root->root_item,
1344                                 btrfs_item_ptr_offset(l, path->slots[0]),
1345                                 sizeof(root->root_item));
1346                 memcpy(&root->root_key, location, sizeof(*location));
1347         }
1348         btrfs_free_path(path);
1349         if (ret) {
1350                 kfree(root);
1351                 if (ret > 0)
1352                         ret = -ENOENT;
1353                 return ERR_PTR(ret);
1354         }
1355
1356         generation = btrfs_root_generation(&root->root_item);
1357         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1358         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1359                                      blocksize, generation);
1360         root->commit_root = btrfs_root_node(root);
1361         BUG_ON(!root->node);
1362 out:
1363         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1364                 root->ref_cows = 1;
1365                 btrfs_check_and_init_root_item(&root->root_item);
1366         }
1367
1368         return root;
1369 }
1370
1371 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1372                                               struct btrfs_key *location)
1373 {
1374         struct btrfs_root *root;
1375         int ret;
1376
1377         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1378                 return fs_info->tree_root;
1379         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1380                 return fs_info->extent_root;
1381         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1382                 return fs_info->chunk_root;
1383         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1384                 return fs_info->dev_root;
1385         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1386                 return fs_info->csum_root;
1387 again:
1388         spin_lock(&fs_info->fs_roots_radix_lock);
1389         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1390                                  (unsigned long)location->objectid);
1391         spin_unlock(&fs_info->fs_roots_radix_lock);
1392         if (root)
1393                 return root;
1394
1395         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1396         if (IS_ERR(root))
1397                 return root;
1398
1399         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1400         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1401                                         GFP_NOFS);
1402         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1403                 ret = -ENOMEM;
1404                 goto fail;
1405         }
1406
1407         btrfs_init_free_ino_ctl(root);
1408         mutex_init(&root->fs_commit_mutex);
1409         spin_lock_init(&root->cache_lock);
1410         init_waitqueue_head(&root->cache_wait);
1411
1412         ret = get_anon_bdev(&root->anon_dev);
1413         if (ret)
1414                 goto fail;
1415
1416         if (btrfs_root_refs(&root->root_item) == 0) {
1417                 ret = -ENOENT;
1418                 goto fail;
1419         }
1420
1421         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1422         if (ret < 0)
1423                 goto fail;
1424         if (ret == 0)
1425                 root->orphan_item_inserted = 1;
1426
1427         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1428         if (ret)
1429                 goto fail;
1430
1431         spin_lock(&fs_info->fs_roots_radix_lock);
1432         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1433                                 (unsigned long)root->root_key.objectid,
1434                                 root);
1435         if (ret == 0)
1436                 root->in_radix = 1;
1437
1438         spin_unlock(&fs_info->fs_roots_radix_lock);
1439         radix_tree_preload_end();
1440         if (ret) {
1441                 if (ret == -EEXIST) {
1442                         free_fs_root(root);
1443                         goto again;
1444                 }
1445                 goto fail;
1446         }
1447
1448         ret = btrfs_find_dead_roots(fs_info->tree_root,
1449                                     root->root_key.objectid);
1450         WARN_ON(ret);
1451         return root;
1452 fail:
1453         free_fs_root(root);
1454         return ERR_PTR(ret);
1455 }
1456
1457 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1458 {
1459         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1460         int ret = 0;
1461         struct btrfs_device *device;
1462         struct backing_dev_info *bdi;
1463
1464         rcu_read_lock();
1465         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1466                 if (!device->bdev)
1467                         continue;
1468                 bdi = blk_get_backing_dev_info(device->bdev);
1469                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1470                         ret = 1;
1471                         break;
1472                 }
1473         }
1474         rcu_read_unlock();
1475         return ret;
1476 }
1477
1478 /*
1479  * If this fails, caller must call bdi_destroy() to get rid of the
1480  * bdi again.
1481  */
1482 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1483 {
1484         int err;
1485
1486         bdi->capabilities = BDI_CAP_MAP_COPY;
1487         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1488         if (err)
1489                 return err;
1490
1491         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1492         bdi->congested_fn       = btrfs_congested_fn;
1493         bdi->congested_data     = info;
1494         return 0;
1495 }
1496
1497 static int bio_ready_for_csum(struct bio *bio)
1498 {
1499         u64 length = 0;
1500         u64 buf_len = 0;
1501         u64 start = 0;
1502         struct page *page;
1503         struct extent_io_tree *io_tree = NULL;
1504         struct bio_vec *bvec;
1505         int i;
1506         int ret;
1507
1508         bio_for_each_segment(bvec, bio, i) {
1509                 page = bvec->bv_page;
1510                 if (page->private == EXTENT_PAGE_PRIVATE) {
1511                         length += bvec->bv_len;
1512                         continue;
1513                 }
1514                 if (!page->private) {
1515                         length += bvec->bv_len;
1516                         continue;
1517                 }
1518                 length = bvec->bv_len;
1519                 buf_len = page->private >> 2;
1520                 start = page_offset(page) + bvec->bv_offset;
1521                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1522         }
1523         /* are we fully contained in this bio? */
1524         if (buf_len <= length)
1525                 return 1;
1526
1527         ret = extent_range_uptodate(io_tree, start + length,
1528                                     start + buf_len - 1);
1529         return ret;
1530 }
1531
1532 /*
1533  * called by the kthread helper functions to finally call the bio end_io
1534  * functions.  This is where read checksum verification actually happens
1535  */
1536 static void end_workqueue_fn(struct btrfs_work *work)
1537 {
1538         struct bio *bio;
1539         struct end_io_wq *end_io_wq;
1540         struct btrfs_fs_info *fs_info;
1541         int error;
1542
1543         end_io_wq = container_of(work, struct end_io_wq, work);
1544         bio = end_io_wq->bio;
1545         fs_info = end_io_wq->info;
1546
1547         /* metadata bio reads are special because the whole tree block must
1548          * be checksummed at once.  This makes sure the entire block is in
1549          * ram and up to date before trying to verify things.  For
1550          * blocksize <= pagesize, it is basically a noop
1551          */
1552         if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1553             !bio_ready_for_csum(bio)) {
1554                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1555                                    &end_io_wq->work);
1556                 return;
1557         }
1558         error = end_io_wq->error;
1559         bio->bi_private = end_io_wq->private;
1560         bio->bi_end_io = end_io_wq->end_io;
1561         kfree(end_io_wq);
1562         bio_endio(bio, error);
1563 }
1564
1565 static int cleaner_kthread(void *arg)
1566 {
1567         struct btrfs_root *root = arg;
1568
1569         do {
1570                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1571
1572                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1573                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1574                         btrfs_run_delayed_iputs(root);
1575                         btrfs_clean_old_snapshots(root);
1576                         mutex_unlock(&root->fs_info->cleaner_mutex);
1577                         btrfs_run_defrag_inodes(root->fs_info);
1578                 }
1579
1580                 if (freezing(current)) {
1581                         refrigerator();
1582                 } else {
1583                         set_current_state(TASK_INTERRUPTIBLE);
1584                         if (!kthread_should_stop())
1585                                 schedule();
1586                         __set_current_state(TASK_RUNNING);
1587                 }
1588         } while (!kthread_should_stop());
1589         return 0;
1590 }
1591
1592 static int transaction_kthread(void *arg)
1593 {
1594         struct btrfs_root *root = arg;
1595         struct btrfs_trans_handle *trans;
1596         struct btrfs_transaction *cur;
1597         u64 transid;
1598         unsigned long now;
1599         unsigned long delay;
1600         int ret;
1601
1602         do {
1603                 delay = HZ * 30;
1604                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1605                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1606
1607                 spin_lock(&root->fs_info->trans_lock);
1608                 cur = root->fs_info->running_transaction;
1609                 if (!cur) {
1610                         spin_unlock(&root->fs_info->trans_lock);
1611                         goto sleep;
1612                 }
1613
1614                 now = get_seconds();
1615                 if (!cur->blocked &&
1616                     (now < cur->start_time || now - cur->start_time < 30)) {
1617                         spin_unlock(&root->fs_info->trans_lock);
1618                         delay = HZ * 5;
1619                         goto sleep;
1620                 }
1621                 transid = cur->transid;
1622                 spin_unlock(&root->fs_info->trans_lock);
1623
1624                 trans = btrfs_join_transaction(root);
1625                 BUG_ON(IS_ERR(trans));
1626                 if (transid == trans->transid) {
1627                         ret = btrfs_commit_transaction(trans, root);
1628                         BUG_ON(ret);
1629                 } else {
1630                         btrfs_end_transaction(trans, root);
1631                 }
1632 sleep:
1633                 wake_up_process(root->fs_info->cleaner_kthread);
1634                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1635
1636                 if (freezing(current)) {
1637                         refrigerator();
1638                 } else {
1639                         set_current_state(TASK_INTERRUPTIBLE);
1640                         if (!kthread_should_stop() &&
1641                             !btrfs_transaction_blocked(root->fs_info))
1642                                 schedule_timeout(delay);
1643                         __set_current_state(TASK_RUNNING);
1644                 }
1645         } while (!kthread_should_stop());
1646         return 0;
1647 }
1648
1649 struct btrfs_root *open_ctree(struct super_block *sb,
1650                               struct btrfs_fs_devices *fs_devices,
1651                               char *options)
1652 {
1653         u32 sectorsize;
1654         u32 nodesize;
1655         u32 leafsize;
1656         u32 blocksize;
1657         u32 stripesize;
1658         u64 generation;
1659         u64 features;
1660         struct btrfs_key location;
1661         struct buffer_head *bh;
1662         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1663                                                  GFP_NOFS);
1664         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1665                                                  GFP_NOFS);
1666         struct btrfs_root *tree_root = btrfs_sb(sb);
1667         struct btrfs_fs_info *fs_info = NULL;
1668         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1669                                                 GFP_NOFS);
1670         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1671                                               GFP_NOFS);
1672         struct btrfs_root *log_tree_root;
1673
1674         int ret;
1675         int err = -EINVAL;
1676
1677         struct btrfs_super_block *disk_super;
1678
1679         if (!extent_root || !tree_root || !tree_root->fs_info ||
1680             !chunk_root || !dev_root || !csum_root) {
1681                 err = -ENOMEM;
1682                 goto fail;
1683         }
1684         fs_info = tree_root->fs_info;
1685
1686         ret = init_srcu_struct(&fs_info->subvol_srcu);
1687         if (ret) {
1688                 err = ret;
1689                 goto fail;
1690         }
1691
1692         ret = setup_bdi(fs_info, &fs_info->bdi);
1693         if (ret) {
1694                 err = ret;
1695                 goto fail_srcu;
1696         }
1697
1698         fs_info->btree_inode = new_inode(sb);
1699         if (!fs_info->btree_inode) {
1700                 err = -ENOMEM;
1701                 goto fail_bdi;
1702         }
1703
1704         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1705
1706         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1707         INIT_LIST_HEAD(&fs_info->trans_list);
1708         INIT_LIST_HEAD(&fs_info->dead_roots);
1709         INIT_LIST_HEAD(&fs_info->delayed_iputs);
1710         INIT_LIST_HEAD(&fs_info->hashers);
1711         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1712         INIT_LIST_HEAD(&fs_info->ordered_operations);
1713         INIT_LIST_HEAD(&fs_info->caching_block_groups);
1714         spin_lock_init(&fs_info->delalloc_lock);
1715         spin_lock_init(&fs_info->trans_lock);
1716         spin_lock_init(&fs_info->ref_cache_lock);
1717         spin_lock_init(&fs_info->fs_roots_radix_lock);
1718         spin_lock_init(&fs_info->delayed_iput_lock);
1719         spin_lock_init(&fs_info->defrag_inodes_lock);
1720         mutex_init(&fs_info->reloc_mutex);
1721
1722         init_completion(&fs_info->kobj_unregister);
1723         fs_info->tree_root = tree_root;
1724         fs_info->extent_root = extent_root;
1725         fs_info->csum_root = csum_root;
1726         fs_info->chunk_root = chunk_root;
1727         fs_info->dev_root = dev_root;
1728         fs_info->fs_devices = fs_devices;
1729         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1730         INIT_LIST_HEAD(&fs_info->space_info);
1731         btrfs_mapping_init(&fs_info->mapping_tree);
1732         btrfs_init_block_rsv(&fs_info->global_block_rsv);
1733         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1734         btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1735         btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1736         btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1737         INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1738         mutex_init(&fs_info->durable_block_rsv_mutex);
1739         atomic_set(&fs_info->nr_async_submits, 0);
1740         atomic_set(&fs_info->async_delalloc_pages, 0);
1741         atomic_set(&fs_info->async_submit_draining, 0);
1742         atomic_set(&fs_info->nr_async_bios, 0);
1743         atomic_set(&fs_info->defrag_running, 0);
1744         fs_info->sb = sb;
1745         fs_info->max_inline = 8192 * 1024;
1746         fs_info->metadata_ratio = 0;
1747         fs_info->defrag_inodes = RB_ROOT;
1748         fs_info->trans_no_join = 0;
1749
1750         /* readahead state */
1751         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
1752         spin_lock_init(&fs_info->reada_lock);
1753
1754         fs_info->thread_pool_size = min_t(unsigned long,
1755                                           num_online_cpus() + 2, 8);
1756
1757         INIT_LIST_HEAD(&fs_info->ordered_extents);
1758         spin_lock_init(&fs_info->ordered_extent_lock);
1759         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1760                                         GFP_NOFS);
1761         if (!fs_info->delayed_root) {
1762                 err = -ENOMEM;
1763                 goto fail_iput;
1764         }
1765         btrfs_init_delayed_root(fs_info->delayed_root);
1766
1767         mutex_init(&fs_info->scrub_lock);
1768         atomic_set(&fs_info->scrubs_running, 0);
1769         atomic_set(&fs_info->scrub_pause_req, 0);
1770         atomic_set(&fs_info->scrubs_paused, 0);
1771         atomic_set(&fs_info->scrub_cancel_req, 0);
1772         init_waitqueue_head(&fs_info->scrub_pause_wait);
1773         init_rwsem(&fs_info->scrub_super_lock);
1774         fs_info->scrub_workers_refcnt = 0;
1775
1776         sb->s_blocksize = 4096;
1777         sb->s_blocksize_bits = blksize_bits(4096);
1778         sb->s_bdi = &fs_info->bdi;
1779
1780         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1781         fs_info->btree_inode->i_nlink = 1;
1782         /*
1783          * we set the i_size on the btree inode to the max possible int.
1784          * the real end of the address space is determined by all of
1785          * the devices in the system
1786          */
1787         fs_info->btree_inode->i_size = OFFSET_MAX;
1788         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1789         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1790
1791         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1792         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1793                              fs_info->btree_inode->i_mapping);
1794         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
1795
1796         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1797
1798         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1799         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1800                sizeof(struct btrfs_key));
1801         BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1802         insert_inode_hash(fs_info->btree_inode);
1803
1804         spin_lock_init(&fs_info->block_group_cache_lock);
1805         fs_info->block_group_cache_tree = RB_ROOT;
1806
1807         extent_io_tree_init(&fs_info->freed_extents[0],
1808                              fs_info->btree_inode->i_mapping);
1809         extent_io_tree_init(&fs_info->freed_extents[1],
1810                              fs_info->btree_inode->i_mapping);
1811         fs_info->pinned_extents = &fs_info->freed_extents[0];
1812         fs_info->do_barriers = 1;
1813
1814
1815         mutex_init(&fs_info->ordered_operations_mutex);
1816         mutex_init(&fs_info->tree_log_mutex);
1817         mutex_init(&fs_info->chunk_mutex);
1818         mutex_init(&fs_info->transaction_kthread_mutex);
1819         mutex_init(&fs_info->cleaner_mutex);
1820         mutex_init(&fs_info->volume_mutex);
1821         init_rwsem(&fs_info->extent_commit_sem);
1822         init_rwsem(&fs_info->cleanup_work_sem);
1823         init_rwsem(&fs_info->subvol_sem);
1824
1825         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1826         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1827
1828         init_waitqueue_head(&fs_info->transaction_throttle);
1829         init_waitqueue_head(&fs_info->transaction_wait);
1830         init_waitqueue_head(&fs_info->transaction_blocked_wait);
1831         init_waitqueue_head(&fs_info->async_submit_wait);
1832
1833         __setup_root(4096, 4096, 4096, 4096, tree_root,
1834                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1835
1836         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1837         if (!bh) {
1838                 err = -EINVAL;
1839                 goto fail_alloc;
1840         }
1841
1842         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1843         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1844                sizeof(fs_info->super_for_commit));
1845         brelse(bh);
1846
1847         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1848
1849         disk_super = &fs_info->super_copy;
1850         if (!btrfs_super_root(disk_super))
1851                 goto fail_alloc;
1852
1853         /* check FS state, whether FS is broken. */
1854         fs_info->fs_state |= btrfs_super_flags(disk_super);
1855
1856         btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
1857
1858         /*
1859          * In the long term, we'll store the compression type in the super
1860          * block, and it'll be used for per file compression control.
1861          */
1862         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
1863
1864         ret = btrfs_parse_options(tree_root, options);
1865         if (ret) {
1866                 err = ret;
1867                 goto fail_alloc;
1868         }
1869
1870         features = btrfs_super_incompat_flags(disk_super) &
1871                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1872         if (features) {
1873                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1874                        "unsupported optional features (%Lx).\n",
1875                        (unsigned long long)features);
1876                 err = -EINVAL;
1877                 goto fail_alloc;
1878         }
1879
1880         features = btrfs_super_incompat_flags(disk_super);
1881         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1882         if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1883                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1884         btrfs_set_super_incompat_flags(disk_super, features);
1885
1886         features = btrfs_super_compat_ro_flags(disk_super) &
1887                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1888         if (!(sb->s_flags & MS_RDONLY) && features) {
1889                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1890                        "unsupported option features (%Lx).\n",
1891                        (unsigned long long)features);
1892                 err = -EINVAL;
1893                 goto fail_alloc;
1894         }
1895
1896         btrfs_init_workers(&fs_info->generic_worker,
1897                            "genwork", 1, NULL);
1898
1899         btrfs_init_workers(&fs_info->workers, "worker",
1900                            fs_info->thread_pool_size,
1901                            &fs_info->generic_worker);
1902
1903         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1904                            fs_info->thread_pool_size,
1905                            &fs_info->generic_worker);
1906
1907         btrfs_init_workers(&fs_info->submit_workers, "submit",
1908                            min_t(u64, fs_devices->num_devices,
1909                            fs_info->thread_pool_size),
1910                            &fs_info->generic_worker);
1911
1912         btrfs_init_workers(&fs_info->caching_workers, "cache",
1913                            2, &fs_info->generic_worker);
1914
1915         /* a higher idle thresh on the submit workers makes it much more
1916          * likely that bios will be send down in a sane order to the
1917          * devices
1918          */
1919         fs_info->submit_workers.idle_thresh = 64;
1920
1921         fs_info->workers.idle_thresh = 16;
1922         fs_info->workers.ordered = 1;
1923
1924         fs_info->delalloc_workers.idle_thresh = 2;
1925         fs_info->delalloc_workers.ordered = 1;
1926
1927         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1928                            &fs_info->generic_worker);
1929         btrfs_init_workers(&fs_info->endio_workers, "endio",
1930                            fs_info->thread_pool_size,
1931                            &fs_info->generic_worker);
1932         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1933                            fs_info->thread_pool_size,
1934                            &fs_info->generic_worker);
1935         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1936                            "endio-meta-write", fs_info->thread_pool_size,
1937                            &fs_info->generic_worker);
1938         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1939                            fs_info->thread_pool_size,
1940                            &fs_info->generic_worker);
1941         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1942                            1, &fs_info->generic_worker);
1943         btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
1944                            fs_info->thread_pool_size,
1945                            &fs_info->generic_worker);
1946         btrfs_init_workers(&fs_info->readahead_workers, "readahead",
1947                            fs_info->thread_pool_size,
1948                            &fs_info->generic_worker);
1949
1950         /*
1951          * endios are largely parallel and should have a very
1952          * low idle thresh
1953          */
1954         fs_info->endio_workers.idle_thresh = 4;
1955         fs_info->endio_meta_workers.idle_thresh = 4;
1956
1957         fs_info->endio_write_workers.idle_thresh = 2;
1958         fs_info->endio_meta_write_workers.idle_thresh = 2;
1959         fs_info->readahead_workers.idle_thresh = 2;
1960
1961         btrfs_start_workers(&fs_info->workers, 1);
1962         btrfs_start_workers(&fs_info->generic_worker, 1);
1963         btrfs_start_workers(&fs_info->submit_workers, 1);
1964         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1965         btrfs_start_workers(&fs_info->fixup_workers, 1);
1966         btrfs_start_workers(&fs_info->endio_workers, 1);
1967         btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1968         btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1969         btrfs_start_workers(&fs_info->endio_write_workers, 1);
1970         btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1971         btrfs_start_workers(&fs_info->delayed_workers, 1);
1972         btrfs_start_workers(&fs_info->caching_workers, 1);
1973         btrfs_start_workers(&fs_info->readahead_workers, 1);
1974
1975         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1976         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1977                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1978
1979         nodesize = btrfs_super_nodesize(disk_super);
1980         leafsize = btrfs_super_leafsize(disk_super);
1981         sectorsize = btrfs_super_sectorsize(disk_super);
1982         stripesize = btrfs_super_stripesize(disk_super);
1983         tree_root->nodesize = nodesize;
1984         tree_root->leafsize = leafsize;
1985         tree_root->sectorsize = sectorsize;
1986         tree_root->stripesize = stripesize;
1987
1988         sb->s_blocksize = sectorsize;
1989         sb->s_blocksize_bits = blksize_bits(sectorsize);
1990
1991         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1992                     sizeof(disk_super->magic))) {
1993                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1994                 goto fail_sb_buffer;
1995         }
1996
1997         mutex_lock(&fs_info->chunk_mutex);
1998         ret = btrfs_read_sys_array(tree_root);
1999         mutex_unlock(&fs_info->chunk_mutex);
2000         if (ret) {
2001                 printk(KERN_WARNING "btrfs: failed to read the system "
2002                        "array on %s\n", sb->s_id);
2003                 goto fail_sb_buffer;
2004         }
2005
2006         blocksize = btrfs_level_size(tree_root,
2007                                      btrfs_super_chunk_root_level(disk_super));
2008         generation = btrfs_super_chunk_root_generation(disk_super);
2009
2010         __setup_root(nodesize, leafsize, sectorsize, stripesize,
2011                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2012
2013         chunk_root->node = read_tree_block(chunk_root,
2014                                            btrfs_super_chunk_root(disk_super),
2015                                            blocksize, generation);
2016         BUG_ON(!chunk_root->node);
2017         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2018                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2019                        sb->s_id);
2020                 goto fail_chunk_root;
2021         }
2022         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2023         chunk_root->commit_root = btrfs_root_node(chunk_root);
2024
2025         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2026            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2027            BTRFS_UUID_SIZE);
2028
2029         mutex_lock(&fs_info->chunk_mutex);
2030         ret = btrfs_read_chunk_tree(chunk_root);
2031         mutex_unlock(&fs_info->chunk_mutex);
2032         if (ret) {
2033                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2034                        sb->s_id);
2035                 goto fail_chunk_root;
2036         }
2037
2038         btrfs_close_extra_devices(fs_devices);
2039
2040         blocksize = btrfs_level_size(tree_root,
2041                                      btrfs_super_root_level(disk_super));
2042         generation = btrfs_super_generation(disk_super);
2043
2044         tree_root->node = read_tree_block(tree_root,
2045                                           btrfs_super_root(disk_super),
2046                                           blocksize, generation);
2047         if (!tree_root->node)
2048                 goto fail_chunk_root;
2049         if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2050                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2051                        sb->s_id);
2052                 goto fail_tree_root;
2053         }
2054         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2055         tree_root->commit_root = btrfs_root_node(tree_root);
2056
2057         ret = find_and_setup_root(tree_root, fs_info,
2058                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2059         if (ret)
2060                 goto fail_tree_root;
2061         extent_root->track_dirty = 1;
2062
2063         ret = find_and_setup_root(tree_root, fs_info,
2064                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
2065         if (ret)
2066                 goto fail_extent_root;
2067         dev_root->track_dirty = 1;
2068
2069         ret = find_and_setup_root(tree_root, fs_info,
2070                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
2071         if (ret)
2072                 goto fail_dev_root;
2073
2074         csum_root->track_dirty = 1;
2075
2076         fs_info->generation = generation;
2077         fs_info->last_trans_committed = generation;
2078         fs_info->data_alloc_profile = (u64)-1;
2079         fs_info->metadata_alloc_profile = (u64)-1;
2080         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
2081
2082         ret = btrfs_init_space_info(fs_info);
2083         if (ret) {
2084                 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2085                 goto fail_block_groups;
2086         }
2087
2088         ret = btrfs_read_block_groups(extent_root);
2089         if (ret) {
2090                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2091                 goto fail_block_groups;
2092         }
2093
2094         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2095                                                "btrfs-cleaner");
2096         if (IS_ERR(fs_info->cleaner_kthread))
2097                 goto fail_block_groups;
2098
2099         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2100                                                    tree_root,
2101                                                    "btrfs-transaction");
2102         if (IS_ERR(fs_info->transaction_kthread))
2103                 goto fail_cleaner;
2104
2105         if (!btrfs_test_opt(tree_root, SSD) &&
2106             !btrfs_test_opt(tree_root, NOSSD) &&
2107             !fs_info->fs_devices->rotating) {
2108                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2109                        "mode\n");
2110                 btrfs_set_opt(fs_info->mount_opt, SSD);
2111         }
2112
2113         /* do not make disk changes in broken FS */
2114         if (btrfs_super_log_root(disk_super) != 0 &&
2115             !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2116                 u64 bytenr = btrfs_super_log_root(disk_super);
2117
2118                 if (fs_devices->rw_devices == 0) {
2119                         printk(KERN_WARNING "Btrfs log replay required "
2120                                "on RO media\n");
2121                         err = -EIO;
2122                         goto fail_trans_kthread;
2123                 }
2124                 blocksize =
2125                      btrfs_level_size(tree_root,
2126                                       btrfs_super_log_root_level(disk_super));
2127
2128                 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2129                 if (!log_tree_root) {
2130                         err = -ENOMEM;
2131                         goto fail_trans_kthread;
2132                 }
2133
2134                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2135                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2136
2137                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2138                                                       blocksize,
2139                                                       generation + 1);
2140                 ret = btrfs_recover_log_trees(log_tree_root);
2141                 BUG_ON(ret);
2142
2143                 if (sb->s_flags & MS_RDONLY) {
2144                         ret =  btrfs_commit_super(tree_root);
2145                         BUG_ON(ret);
2146                 }
2147         }
2148
2149         ret = btrfs_find_orphan_roots(tree_root);
2150         BUG_ON(ret);
2151
2152         if (!(sb->s_flags & MS_RDONLY)) {
2153                 ret = btrfs_cleanup_fs_roots(fs_info);
2154                 BUG_ON(ret);
2155
2156                 ret = btrfs_recover_relocation(tree_root);
2157                 if (ret < 0) {
2158                         printk(KERN_WARNING
2159                                "btrfs: failed to recover relocation\n");
2160                         err = -EINVAL;
2161                         goto fail_trans_kthread;
2162                 }
2163         }
2164
2165         location.objectid = BTRFS_FS_TREE_OBJECTID;
2166         location.type = BTRFS_ROOT_ITEM_KEY;
2167         location.offset = (u64)-1;
2168
2169         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2170         if (!fs_info->fs_root)
2171                 goto fail_trans_kthread;
2172         if (IS_ERR(fs_info->fs_root)) {
2173                 err = PTR_ERR(fs_info->fs_root);
2174                 goto fail_trans_kthread;
2175         }
2176
2177         if (!(sb->s_flags & MS_RDONLY)) {
2178                 down_read(&fs_info->cleanup_work_sem);
2179                 err = btrfs_orphan_cleanup(fs_info->fs_root);
2180                 if (!err)
2181                         err = btrfs_orphan_cleanup(fs_info->tree_root);
2182                 up_read(&fs_info->cleanup_work_sem);
2183                 if (err) {
2184                         close_ctree(tree_root);
2185                         return ERR_PTR(err);
2186                 }
2187         }
2188
2189         return tree_root;
2190
2191 fail_trans_kthread:
2192         kthread_stop(fs_info->transaction_kthread);
2193 fail_cleaner:
2194         kthread_stop(fs_info->cleaner_kthread);
2195
2196         /*
2197          * make sure we're done with the btree inode before we stop our
2198          * kthreads
2199          */
2200         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2201         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2202
2203 fail_block_groups:
2204         btrfs_free_block_groups(fs_info);
2205         free_extent_buffer(csum_root->node);
2206         free_extent_buffer(csum_root->commit_root);
2207 fail_dev_root:
2208         free_extent_buffer(dev_root->node);
2209         free_extent_buffer(dev_root->commit_root);
2210 fail_extent_root:
2211         free_extent_buffer(extent_root->node);
2212         free_extent_buffer(extent_root->commit_root);
2213 fail_tree_root:
2214         free_extent_buffer(tree_root->node);
2215         free_extent_buffer(tree_root->commit_root);
2216 fail_chunk_root:
2217         free_extent_buffer(chunk_root->node);
2218         free_extent_buffer(chunk_root->commit_root);
2219 fail_sb_buffer:
2220         btrfs_stop_workers(&fs_info->generic_worker);
2221         btrfs_stop_workers(&fs_info->fixup_workers);
2222         btrfs_stop_workers(&fs_info->delalloc_workers);
2223         btrfs_stop_workers(&fs_info->workers);
2224         btrfs_stop_workers(&fs_info->endio_workers);
2225         btrfs_stop_workers(&fs_info->endio_meta_workers);
2226         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2227         btrfs_stop_workers(&fs_info->endio_write_workers);
2228         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2229         btrfs_stop_workers(&fs_info->submit_workers);
2230         btrfs_stop_workers(&fs_info->delayed_workers);
2231         btrfs_stop_workers(&fs_info->caching_workers);
2232 fail_alloc:
2233         kfree(fs_info->delayed_root);
2234 fail_iput:
2235         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2236         iput(fs_info->btree_inode);
2237
2238         btrfs_close_devices(fs_info->fs_devices);
2239         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2240 fail_bdi:
2241         bdi_destroy(&fs_info->bdi);
2242 fail_srcu:
2243         cleanup_srcu_struct(&fs_info->subvol_srcu);
2244 fail:
2245         kfree(extent_root);
2246         kfree(tree_root);
2247         kfree(fs_info);
2248         kfree(chunk_root);
2249         kfree(dev_root);
2250         kfree(csum_root);
2251         return ERR_PTR(err);
2252 }
2253
2254 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2255 {
2256         char b[BDEVNAME_SIZE];
2257
2258         if (uptodate) {
2259                 set_buffer_uptodate(bh);
2260         } else {
2261                 printk_ratelimited(KERN_WARNING "lost page write due to "
2262                                         "I/O error on %s\n",
2263                                        bdevname(bh->b_bdev, b));
2264                 /* note, we dont' set_buffer_write_io_error because we have
2265                  * our own ways of dealing with the IO errors
2266                  */
2267                 clear_buffer_uptodate(bh);
2268         }
2269         unlock_buffer(bh);
2270         put_bh(bh);
2271 }
2272
2273 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2274 {
2275         struct buffer_head *bh;
2276         struct buffer_head *latest = NULL;
2277         struct btrfs_super_block *super;
2278         int i;
2279         u64 transid = 0;
2280         u64 bytenr;
2281
2282         /* we would like to check all the supers, but that would make
2283          * a btrfs mount succeed after a mkfs from a different FS.
2284          * So, we need to add a special mount option to scan for
2285          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2286          */
2287         for (i = 0; i < 1; i++) {
2288                 bytenr = btrfs_sb_offset(i);
2289                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2290                         break;
2291                 bh = __bread(bdev, bytenr / 4096, 4096);
2292                 if (!bh)
2293                         continue;
2294
2295                 super = (struct btrfs_super_block *)bh->b_data;
2296                 if (btrfs_super_bytenr(super) != bytenr ||
2297                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2298                             sizeof(super->magic))) {
2299                         brelse(bh);
2300                         continue;
2301                 }
2302
2303                 if (!latest || btrfs_super_generation(super) > transid) {
2304                         brelse(latest);
2305                         latest = bh;
2306                         transid = btrfs_super_generation(super);
2307                 } else {
2308                         brelse(bh);
2309                 }
2310         }
2311         return latest;
2312 }
2313
2314 /*
2315  * this should be called twice, once with wait == 0 and
2316  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2317  * we write are pinned.
2318  *
2319  * They are released when wait == 1 is done.
2320  * max_mirrors must be the same for both runs, and it indicates how
2321  * many supers on this one device should be written.
2322  *
2323  * max_mirrors == 0 means to write them all.
2324  */
2325 static int write_dev_supers(struct btrfs_device *device,
2326                             struct btrfs_super_block *sb,
2327                             int do_barriers, int wait, int max_mirrors)
2328 {
2329         struct buffer_head *bh;
2330         int i;
2331         int ret;
2332         int errors = 0;
2333         u32 crc;
2334         u64 bytenr;
2335         int last_barrier = 0;
2336
2337         if (max_mirrors == 0)
2338                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2339
2340         /* make sure only the last submit_bh does a barrier */
2341         if (do_barriers) {
2342                 for (i = 0; i < max_mirrors; i++) {
2343                         bytenr = btrfs_sb_offset(i);
2344                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2345                             device->total_bytes)
2346                                 break;
2347                         last_barrier = i;
2348                 }
2349         }
2350
2351         for (i = 0; i < max_mirrors; i++) {
2352                 bytenr = btrfs_sb_offset(i);
2353                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2354                         break;
2355
2356                 if (wait) {
2357                         bh = __find_get_block(device->bdev, bytenr / 4096,
2358                                               BTRFS_SUPER_INFO_SIZE);
2359                         BUG_ON(!bh);
2360                         wait_on_buffer(bh);
2361                         if (!buffer_uptodate(bh))
2362                                 errors++;
2363
2364                         /* drop our reference */
2365                         brelse(bh);
2366
2367                         /* drop the reference from the wait == 0 run */
2368                         brelse(bh);
2369                         continue;
2370                 } else {
2371                         btrfs_set_super_bytenr(sb, bytenr);
2372
2373                         crc = ~(u32)0;
2374                         crc = btrfs_csum_data(NULL, (char *)sb +
2375                                               BTRFS_CSUM_SIZE, crc,
2376                                               BTRFS_SUPER_INFO_SIZE -
2377                                               BTRFS_CSUM_SIZE);
2378                         btrfs_csum_final(crc, sb->csum);
2379
2380                         /*
2381                          * one reference for us, and we leave it for the
2382                          * caller
2383                          */
2384                         bh = __getblk(device->bdev, bytenr / 4096,
2385                                       BTRFS_SUPER_INFO_SIZE);
2386                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2387
2388                         /* one reference for submit_bh */
2389                         get_bh(bh);
2390
2391                         set_buffer_uptodate(bh);
2392                         lock_buffer(bh);
2393                         bh->b_end_io = btrfs_end_buffer_write_sync;
2394                 }
2395
2396                 if (i == last_barrier && do_barriers)
2397                         ret = submit_bh(WRITE_FLUSH_FUA, bh);
2398                 else
2399                         ret = submit_bh(WRITE_SYNC, bh);
2400
2401                 if (ret)
2402                         errors++;
2403         }
2404         return errors < i ? 0 : -1;
2405 }
2406
2407 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2408 {
2409         struct list_head *head;
2410         struct btrfs_device *dev;
2411         struct btrfs_super_block *sb;
2412         struct btrfs_dev_item *dev_item;
2413         int ret;
2414         int do_barriers;
2415         int max_errors;
2416         int total_errors = 0;
2417         u64 flags;
2418
2419         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2420         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2421
2422         sb = &root->fs_info->super_for_commit;
2423         dev_item = &sb->dev_item;
2424
2425         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2426         head = &root->fs_info->fs_devices->devices;
2427         list_for_each_entry_rcu(dev, head, dev_list) {
2428                 if (!dev->bdev) {
2429                         total_errors++;
2430                         continue;
2431                 }
2432                 if (!dev->in_fs_metadata || !dev->writeable)
2433                         continue;
2434
2435                 btrfs_set_stack_device_generation(dev_item, 0);
2436                 btrfs_set_stack_device_type(dev_item, dev->type);
2437                 btrfs_set_stack_device_id(dev_item, dev->devid);
2438                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2439                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2440                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2441                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2442                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2443                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2444                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2445
2446                 flags = btrfs_super_flags(sb);
2447                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2448
2449                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2450                 if (ret)
2451                         total_errors++;
2452         }
2453         if (total_errors > max_errors) {
2454                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2455                        total_errors);
2456                 BUG();
2457         }
2458
2459         total_errors = 0;
2460         list_for_each_entry_rcu(dev, head, dev_list) {
2461                 if (!dev->bdev)
2462                         continue;
2463                 if (!dev->in_fs_metadata || !dev->writeable)
2464                         continue;
2465
2466                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2467                 if (ret)
2468                         total_errors++;
2469         }
2470         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2471         if (total_errors > max_errors) {
2472                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2473                        total_errors);
2474                 BUG();
2475         }
2476         return 0;
2477 }
2478
2479 int write_ctree_super(struct btrfs_trans_handle *trans,
2480                       struct btrfs_root *root, int max_mirrors)
2481 {
2482         int ret;
2483
2484         ret = write_all_supers(root, max_mirrors);
2485         return ret;
2486 }
2487
2488 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2489 {
2490         spin_lock(&fs_info->fs_roots_radix_lock);
2491         radix_tree_delete(&fs_info->fs_roots_radix,
2492                           (unsigned long)root->root_key.objectid);
2493         spin_unlock(&fs_info->fs_roots_radix_lock);
2494
2495         if (btrfs_root_refs(&root->root_item) == 0)
2496                 synchronize_srcu(&fs_info->subvol_srcu);
2497
2498         __btrfs_remove_free_space_cache(root->free_ino_pinned);
2499         __btrfs_remove_free_space_cache(root->free_ino_ctl);
2500         free_fs_root(root);
2501         return 0;
2502 }
2503
2504 static void free_fs_root(struct btrfs_root *root)
2505 {
2506         iput(root->cache_inode);
2507         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2508         if (root->anon_dev)
2509                 free_anon_bdev(root->anon_dev);
2510         free_extent_buffer(root->node);
2511         free_extent_buffer(root->commit_root);
2512         kfree(root->free_ino_ctl);
2513         kfree(root->free_ino_pinned);
2514         kfree(root->name);
2515         kfree(root);
2516 }
2517
2518 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2519 {
2520         int ret;
2521         struct btrfs_root *gang[8];
2522         int i;
2523
2524         while (!list_empty(&fs_info->dead_roots)) {
2525                 gang[0] = list_entry(fs_info->dead_roots.next,
2526                                      struct btrfs_root, root_list);
2527                 list_del(&gang[0]->root_list);
2528
2529                 if (gang[0]->in_radix) {
2530                         btrfs_free_fs_root(fs_info, gang[0]);
2531                 } else {
2532                         free_extent_buffer(gang[0]->node);
2533                         free_extent_buffer(gang[0]->commit_root);
2534                         kfree(gang[0]);
2535                 }
2536         }
2537
2538         while (1) {
2539                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2540                                              (void **)gang, 0,
2541                                              ARRAY_SIZE(gang));
2542                 if (!ret)
2543                         break;
2544                 for (i = 0; i < ret; i++)
2545                         btrfs_free_fs_root(fs_info, gang[i]);
2546         }
2547         return 0;
2548 }
2549
2550 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2551 {
2552         u64 root_objectid = 0;
2553         struct btrfs_root *gang[8];
2554         int i;
2555         int ret;
2556
2557         while (1) {
2558                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2559                                              (void **)gang, root_objectid,
2560                                              ARRAY_SIZE(gang));
2561                 if (!ret)
2562                         break;
2563
2564                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2565                 for (i = 0; i < ret; i++) {
2566                         int err;
2567
2568                         root_objectid = gang[i]->root_key.objectid;
2569                         err = btrfs_orphan_cleanup(gang[i]);
2570                         if (err)
2571                                 return err;
2572                 }
2573                 root_objectid++;
2574         }
2575         return 0;
2576 }
2577
2578 int btrfs_commit_super(struct btrfs_root *root)
2579 {
2580         struct btrfs_trans_handle *trans;
2581         int ret;
2582
2583         mutex_lock(&root->fs_info->cleaner_mutex);
2584         btrfs_run_delayed_iputs(root);
2585         btrfs_clean_old_snapshots(root);
2586         mutex_unlock(&root->fs_info->cleaner_mutex);
2587
2588         /* wait until ongoing cleanup work done */
2589         down_write(&root->fs_info->cleanup_work_sem);
2590         up_write(&root->fs_info->cleanup_work_sem);
2591
2592         trans = btrfs_join_transaction(root);
2593         if (IS_ERR(trans))
2594                 return PTR_ERR(trans);
2595         ret = btrfs_commit_transaction(trans, root);
2596         BUG_ON(ret);
2597         /* run commit again to drop the original snapshot */
2598         trans = btrfs_join_transaction(root);
2599         if (IS_ERR(trans))
2600                 return PTR_ERR(trans);
2601         btrfs_commit_transaction(trans, root);
2602         ret = btrfs_write_and_wait_transaction(NULL, root);
2603         BUG_ON(ret);
2604
2605         ret = write_ctree_super(NULL, root, 0);
2606         return ret;
2607 }
2608
2609 int close_ctree(struct btrfs_root *root)
2610 {
2611         struct btrfs_fs_info *fs_info = root->fs_info;
2612         int ret;
2613
2614         fs_info->closing = 1;
2615         smp_mb();
2616
2617         btrfs_scrub_cancel(root);
2618
2619         /* wait for any defraggers to finish */
2620         wait_event(fs_info->transaction_wait,
2621                    (atomic_read(&fs_info->defrag_running) == 0));
2622
2623         /* clear out the rbtree of defraggable inodes */
2624         btrfs_run_defrag_inodes(root->fs_info);
2625
2626         btrfs_put_block_group_cache(fs_info);
2627
2628         /*
2629          * Here come 2 situations when btrfs is broken to flip readonly:
2630          *
2631          * 1. when btrfs flips readonly somewhere else before
2632          * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2633          * and btrfs will skip to write sb directly to keep
2634          * ERROR state on disk.
2635          *
2636          * 2. when btrfs flips readonly just in btrfs_commit_super,
2637          * and in such case, btrfs cannot write sb via btrfs_commit_super,
2638          * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2639          * btrfs will cleanup all FS resources first and write sb then.
2640          */
2641         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2642                 ret = btrfs_commit_super(root);
2643                 if (ret)
2644                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2645         }
2646
2647         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2648                 ret = btrfs_error_commit_super(root);
2649                 if (ret)
2650                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2651         }
2652
2653         kthread_stop(root->fs_info->transaction_kthread);
2654         kthread_stop(root->fs_info->cleaner_kthread);
2655
2656         fs_info->closing = 2;
2657         smp_mb();
2658
2659         if (fs_info->delalloc_bytes) {
2660                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2661                        (unsigned long long)fs_info->delalloc_bytes);
2662         }
2663         if (fs_info->total_ref_cache_size) {
2664                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2665                        (unsigned long long)fs_info->total_ref_cache_size);
2666         }
2667
2668         free_extent_buffer(fs_info->extent_root->node);
2669         free_extent_buffer(fs_info->extent_root->commit_root);
2670         free_extent_buffer(fs_info->tree_root->node);
2671         free_extent_buffer(fs_info->tree_root->commit_root);
2672         free_extent_buffer(root->fs_info->chunk_root->node);
2673         free_extent_buffer(root->fs_info->chunk_root->commit_root);
2674         free_extent_buffer(root->fs_info->dev_root->node);
2675         free_extent_buffer(root->fs_info->dev_root->commit_root);
2676         free_extent_buffer(root->fs_info->csum_root->node);
2677         free_extent_buffer(root->fs_info->csum_root->commit_root);
2678
2679         btrfs_free_block_groups(root->fs_info);
2680
2681         del_fs_roots(fs_info);
2682
2683         iput(fs_info->btree_inode);
2684         kfree(fs_info->delayed_root);
2685
2686         btrfs_stop_workers(&fs_info->generic_worker);
2687         btrfs_stop_workers(&fs_info->fixup_workers);
2688         btrfs_stop_workers(&fs_info->delalloc_workers);
2689         btrfs_stop_workers(&fs_info->workers);
2690         btrfs_stop_workers(&fs_info->endio_workers);
2691         btrfs_stop_workers(&fs_info->endio_meta_workers);
2692         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2693         btrfs_stop_workers(&fs_info->endio_write_workers);
2694         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2695         btrfs_stop_workers(&fs_info->submit_workers);
2696         btrfs_stop_workers(&fs_info->delayed_workers);
2697         btrfs_stop_workers(&fs_info->caching_workers);
2698         btrfs_stop_workers(&fs_info->readahead_workers);
2699
2700         btrfs_close_devices(fs_info->fs_devices);
2701         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2702
2703         bdi_destroy(&fs_info->bdi);
2704         cleanup_srcu_struct(&fs_info->subvol_srcu);
2705
2706         kfree(fs_info->extent_root);
2707         kfree(fs_info->tree_root);
2708         kfree(fs_info->chunk_root);
2709         kfree(fs_info->dev_root);
2710         kfree(fs_info->csum_root);
2711         kfree(fs_info);
2712
2713         return 0;
2714 }
2715
2716 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2717 {
2718         int ret;
2719         struct inode *btree_inode = buf->first_page->mapping->host;
2720
2721         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2722                                      NULL);
2723         if (!ret)
2724                 return ret;
2725
2726         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2727                                     parent_transid);
2728         return !ret;
2729 }
2730
2731 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2732 {
2733         struct inode *btree_inode = buf->first_page->mapping->host;
2734         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2735                                           buf);
2736 }
2737
2738 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2739 {
2740         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2741         u64 transid = btrfs_header_generation(buf);
2742         struct inode *btree_inode = root->fs_info->btree_inode;
2743         int was_dirty;
2744
2745         btrfs_assert_tree_locked(buf);
2746         if (transid != root->fs_info->generation) {
2747                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2748                        "found %llu running %llu\n",
2749                         (unsigned long long)buf->start,
2750                         (unsigned long long)transid,
2751                         (unsigned long long)root->fs_info->generation);
2752                 WARN_ON(1);
2753         }
2754         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2755                                             buf);
2756         if (!was_dirty) {
2757                 spin_lock(&root->fs_info->delalloc_lock);
2758                 root->fs_info->dirty_metadata_bytes += buf->len;
2759                 spin_unlock(&root->fs_info->delalloc_lock);
2760         }
2761 }
2762
2763 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2764 {
2765         /*
2766          * looks as though older kernels can get into trouble with
2767          * this code, they end up stuck in balance_dirty_pages forever
2768          */
2769         u64 num_dirty;
2770         unsigned long thresh = 32 * 1024 * 1024;
2771
2772         if (current->flags & PF_MEMALLOC)
2773                 return;
2774
2775         btrfs_balance_delayed_items(root);
2776
2777         num_dirty = root->fs_info->dirty_metadata_bytes;
2778
2779         if (num_dirty > thresh) {
2780                 balance_dirty_pages_ratelimited_nr(
2781                                    root->fs_info->btree_inode->i_mapping, 1);
2782         }
2783         return;
2784 }
2785
2786 void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2787 {
2788         /*
2789          * looks as though older kernels can get into trouble with
2790          * this code, they end up stuck in balance_dirty_pages forever
2791          */
2792         u64 num_dirty;
2793         unsigned long thresh = 32 * 1024 * 1024;
2794
2795         if (current->flags & PF_MEMALLOC)
2796                 return;
2797
2798         num_dirty = root->fs_info->dirty_metadata_bytes;
2799
2800         if (num_dirty > thresh) {
2801                 balance_dirty_pages_ratelimited_nr(
2802                                    root->fs_info->btree_inode->i_mapping, 1);
2803         }
2804         return;
2805 }
2806
2807 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2808 {
2809         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2810         int ret;
2811         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2812         if (ret == 0)
2813                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2814         return ret;
2815 }
2816
2817 int btree_lock_page_hook(struct page *page)
2818 {
2819         struct inode *inode = page->mapping->host;
2820         struct btrfs_root *root = BTRFS_I(inode)->root;
2821         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2822         struct extent_buffer *eb;
2823         unsigned long len;
2824         u64 bytenr = page_offset(page);
2825
2826         if (page->private == EXTENT_PAGE_PRIVATE)
2827                 goto out;
2828
2829         len = page->private >> 2;
2830         eb = find_extent_buffer(io_tree, bytenr, len);
2831         if (!eb)
2832                 goto out;
2833
2834         btrfs_tree_lock(eb);
2835         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2836
2837         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2838                 spin_lock(&root->fs_info->delalloc_lock);
2839                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2840                         root->fs_info->dirty_metadata_bytes -= eb->len;
2841                 else
2842                         WARN_ON(1);
2843                 spin_unlock(&root->fs_info->delalloc_lock);
2844         }
2845
2846         btrfs_tree_unlock(eb);
2847         free_extent_buffer(eb);
2848 out:
2849         lock_page(page);
2850         return 0;
2851 }
2852
2853 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
2854                               int read_only)
2855 {
2856         if (read_only)
2857                 return;
2858
2859         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
2860                 printk(KERN_WARNING "warning: mount fs with errors, "
2861                        "running btrfsck is recommended\n");
2862 }
2863
2864 int btrfs_error_commit_super(struct btrfs_root *root)
2865 {
2866         int ret;
2867
2868         mutex_lock(&root->fs_info->cleaner_mutex);
2869         btrfs_run_delayed_iputs(root);
2870         mutex_unlock(&root->fs_info->cleaner_mutex);
2871
2872         down_write(&root->fs_info->cleanup_work_sem);
2873         up_write(&root->fs_info->cleanup_work_sem);
2874
2875         /* cleanup FS via transaction */
2876         btrfs_cleanup_transaction(root);
2877
2878         ret = write_ctree_super(NULL, root, 0);
2879
2880         return ret;
2881 }
2882
2883 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
2884 {
2885         struct btrfs_inode *btrfs_inode;
2886         struct list_head splice;
2887
2888         INIT_LIST_HEAD(&splice);
2889
2890         mutex_lock(&root->fs_info->ordered_operations_mutex);
2891         spin_lock(&root->fs_info->ordered_extent_lock);
2892
2893         list_splice_init(&root->fs_info->ordered_operations, &splice);
2894         while (!list_empty(&splice)) {
2895                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2896                                          ordered_operations);
2897
2898                 list_del_init(&btrfs_inode->ordered_operations);
2899
2900                 btrfs_invalidate_inodes(btrfs_inode->root);
2901         }
2902
2903         spin_unlock(&root->fs_info->ordered_extent_lock);
2904         mutex_unlock(&root->fs_info->ordered_operations_mutex);
2905
2906         return 0;
2907 }
2908
2909 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
2910 {
2911         struct list_head splice;
2912         struct btrfs_ordered_extent *ordered;
2913         struct inode *inode;
2914
2915         INIT_LIST_HEAD(&splice);
2916
2917         spin_lock(&root->fs_info->ordered_extent_lock);
2918
2919         list_splice_init(&root->fs_info->ordered_extents, &splice);
2920         while (!list_empty(&splice)) {
2921                 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
2922                                      root_extent_list);
2923
2924                 list_del_init(&ordered->root_extent_list);
2925                 atomic_inc(&ordered->refs);
2926
2927                 /* the inode may be getting freed (in sys_unlink path). */
2928                 inode = igrab(ordered->inode);
2929
2930                 spin_unlock(&root->fs_info->ordered_extent_lock);
2931                 if (inode)
2932                         iput(inode);
2933
2934                 atomic_set(&ordered->refs, 1);
2935                 btrfs_put_ordered_extent(ordered);
2936
2937                 spin_lock(&root->fs_info->ordered_extent_lock);
2938         }
2939
2940         spin_unlock(&root->fs_info->ordered_extent_lock);
2941
2942         return 0;
2943 }
2944
2945 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
2946                                       struct btrfs_root *root)
2947 {
2948         struct rb_node *node;
2949         struct btrfs_delayed_ref_root *delayed_refs;
2950         struct btrfs_delayed_ref_node *ref;
2951         int ret = 0;
2952
2953         delayed_refs = &trans->delayed_refs;
2954
2955         spin_lock(&delayed_refs->lock);
2956         if (delayed_refs->num_entries == 0) {
2957                 spin_unlock(&delayed_refs->lock);
2958                 printk(KERN_INFO "delayed_refs has NO entry\n");
2959                 return ret;
2960         }
2961
2962         node = rb_first(&delayed_refs->root);
2963         while (node) {
2964                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2965                 node = rb_next(node);
2966
2967                 ref->in_tree = 0;
2968                 rb_erase(&ref->rb_node, &delayed_refs->root);
2969                 delayed_refs->num_entries--;
2970
2971                 atomic_set(&ref->refs, 1);
2972                 if (btrfs_delayed_ref_is_head(ref)) {
2973                         struct btrfs_delayed_ref_head *head;
2974
2975                         head = btrfs_delayed_node_to_head(ref);
2976                         mutex_lock(&head->mutex);
2977                         kfree(head->extent_op);
2978                         delayed_refs->num_heads--;
2979                         if (list_empty(&head->cluster))
2980                                 delayed_refs->num_heads_ready--;
2981                         list_del_init(&head->cluster);
2982                         mutex_unlock(&head->mutex);
2983                 }
2984
2985                 spin_unlock(&delayed_refs->lock);
2986                 btrfs_put_delayed_ref(ref);
2987
2988                 cond_resched();
2989                 spin_lock(&delayed_refs->lock);
2990         }
2991
2992         spin_unlock(&delayed_refs->lock);
2993
2994         return ret;
2995 }
2996
2997 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
2998 {
2999         struct btrfs_pending_snapshot *snapshot;
3000         struct list_head splice;
3001
3002         INIT_LIST_HEAD(&splice);
3003
3004         list_splice_init(&t->pending_snapshots, &splice);
3005
3006         while (!list_empty(&splice)) {
3007                 snapshot = list_entry(splice.next,
3008                                       struct btrfs_pending_snapshot,
3009                                       list);
3010
3011                 list_del_init(&snapshot->list);
3012
3013                 kfree(snapshot);
3014         }
3015
3016         return 0;
3017 }
3018
3019 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3020 {
3021         struct btrfs_inode *btrfs_inode;
3022         struct list_head splice;
3023
3024         INIT_LIST_HEAD(&splice);
3025
3026         spin_lock(&root->fs_info->delalloc_lock);
3027         list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3028
3029         while (!list_empty(&splice)) {
3030                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3031                                     delalloc_inodes);
3032
3033                 list_del_init(&btrfs_inode->delalloc_inodes);
3034
3035                 btrfs_invalidate_inodes(btrfs_inode->root);
3036         }
3037
3038         spin_unlock(&root->fs_info->delalloc_lock);
3039
3040         return 0;
3041 }
3042
3043 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3044                                         struct extent_io_tree *dirty_pages,
3045                                         int mark)
3046 {
3047         int ret;
3048         struct page *page;
3049         struct inode *btree_inode = root->fs_info->btree_inode;
3050         struct extent_buffer *eb;
3051         u64 start = 0;
3052         u64 end;
3053         u64 offset;
3054         unsigned long index;
3055
3056         while (1) {
3057                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3058                                             mark);
3059                 if (ret)
3060                         break;
3061
3062                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3063                 while (start <= end) {
3064                         index = start >> PAGE_CACHE_SHIFT;
3065                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3066                         page = find_get_page(btree_inode->i_mapping, index);
3067                         if (!page)
3068                                 continue;
3069                         offset = page_offset(page);
3070
3071                         spin_lock(&dirty_pages->buffer_lock);
3072                         eb = radix_tree_lookup(
3073                              &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3074                                                offset >> PAGE_CACHE_SHIFT);
3075                         spin_unlock(&dirty_pages->buffer_lock);
3076                         if (eb) {
3077                                 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3078                                                          &eb->bflags);
3079                                 atomic_set(&eb->refs, 1);
3080                         }
3081                         if (PageWriteback(page))
3082                                 end_page_writeback(page);
3083
3084                         lock_page(page);
3085                         if (PageDirty(page)) {
3086                                 clear_page_dirty_for_io(page);
3087                                 spin_lock_irq(&page->mapping->tree_lock);
3088                                 radix_tree_tag_clear(&page->mapping->page_tree,
3089                                                         page_index(page),
3090                                                         PAGECACHE_TAG_DIRTY);
3091                                 spin_unlock_irq(&page->mapping->tree_lock);
3092                         }
3093
3094                         page->mapping->a_ops->invalidatepage(page, 0);
3095                         unlock_page(page);
3096                 }
3097         }
3098
3099         return ret;
3100 }
3101
3102 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3103                                        struct extent_io_tree *pinned_extents)
3104 {
3105         struct extent_io_tree *unpin;
3106         u64 start;
3107         u64 end;
3108         int ret;
3109
3110         unpin = pinned_extents;
3111         while (1) {
3112                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3113                                             EXTENT_DIRTY);
3114                 if (ret)
3115                         break;
3116
3117                 /* opt_discard */
3118                 if (btrfs_test_opt(root, DISCARD))
3119                         ret = btrfs_error_discard_extent(root, start,
3120                                                          end + 1 - start,
3121                                                          NULL);
3122
3123                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3124                 btrfs_error_unpin_extent_range(root, start, end);
3125                 cond_resched();
3126         }
3127
3128         return 0;
3129 }
3130
3131 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3132 {
3133         struct btrfs_transaction *t;
3134         LIST_HEAD(list);
3135
3136         WARN_ON(1);
3137
3138         mutex_lock(&root->fs_info->transaction_kthread_mutex);
3139
3140         spin_lock(&root->fs_info->trans_lock);
3141         list_splice_init(&root->fs_info->trans_list, &list);
3142         root->fs_info->trans_no_join = 1;
3143         spin_unlock(&root->fs_info->trans_lock);
3144
3145         while (!list_empty(&list)) {
3146                 t = list_entry(list.next, struct btrfs_transaction, list);
3147                 if (!t)
3148                         break;
3149
3150                 btrfs_destroy_ordered_operations(root);
3151
3152                 btrfs_destroy_ordered_extents(root);
3153
3154                 btrfs_destroy_delayed_refs(t, root);
3155
3156                 btrfs_block_rsv_release(root,
3157                                         &root->fs_info->trans_block_rsv,
3158                                         t->dirty_pages.dirty_bytes);
3159
3160                 /* FIXME: cleanup wait for commit */
3161                 t->in_commit = 1;
3162                 t->blocked = 1;
3163                 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3164                         wake_up(&root->fs_info->transaction_blocked_wait);
3165
3166                 t->blocked = 0;
3167                 if (waitqueue_active(&root->fs_info->transaction_wait))
3168                         wake_up(&root->fs_info->transaction_wait);
3169
3170                 t->commit_done = 1;
3171                 if (waitqueue_active(&t->commit_wait))
3172                         wake_up(&t->commit_wait);
3173
3174                 btrfs_destroy_pending_snapshots(t);
3175
3176                 btrfs_destroy_delalloc_inodes(root);
3177
3178                 spin_lock(&root->fs_info->trans_lock);
3179                 root->fs_info->running_transaction = NULL;
3180                 spin_unlock(&root->fs_info->trans_lock);
3181
3182                 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3183                                              EXTENT_DIRTY);
3184
3185                 btrfs_destroy_pinned_extent(root,
3186                                             root->fs_info->pinned_extents);
3187
3188                 atomic_set(&t->use_count, 0);
3189                 list_del_init(&t->list);
3190                 memset(t, 0, sizeof(*t));
3191                 kmem_cache_free(btrfs_transaction_cachep, t);
3192         }
3193
3194         spin_lock(&root->fs_info->trans_lock);
3195         root->fs_info->trans_no_join = 0;
3196         spin_unlock(&root->fs_info->trans_lock);
3197         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3198
3199         return 0;
3200 }
3201
3202 static struct extent_io_ops btree_extent_io_ops = {
3203         .write_cache_pages_lock_hook = btree_lock_page_hook,
3204         .readpage_end_io_hook = btree_readpage_end_io_hook,
3205         .readpage_io_failed_hook = btree_io_failed_hook,
3206         .submit_bio_hook = btree_submit_bio_hook,
3207         /* note we're sharing with inode.c for the merge bio hook */
3208         .merge_bio_hook = btrfs_merge_bio_hook,
3209 };