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