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 <asm/unaligned.h>
  33 #include "compat.h"
  34 #include "ctree.h"
  35 #include "disk-io.h"
  36 #include "transaction.h"
  37 #include "btrfs_inode.h"
  38 #include "volumes.h"
  39 #include "print-tree.h"
  40 #include "async-thread.h"
  41 #include "locking.h"
  42 #include "tree-log.h"
  43 #include "free-space-cache.h"
  44 #include "inode-map.h"
  45
  46 static struct extent_io_ops btree_extent_io_ops;
  47 static void end_workqueue_fn(struct btrfs_work *work);
  48 static void free_fs_root(struct btrfs_root *root);
  49 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  50                                     int read_only);
  51 static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
  52 static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
  53 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  54                                       struct btrfs_root *root);
  55 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
  56 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  57 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
  58                                         struct extent_io_tree *dirty_pages,
  59                                         int mark);
  60 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
  61                                        struct extent_io_tree *pinned_extents);
  62 static int btrfs_cleanup_transaction(struct btrfs_root *root);
  63
  64 /*
  65  * end_io_wq structs are used to do processing in task context when an IO is
  66  * complete.  This is used during reads to verify checksums, and it is used
  67  * by writes to insert metadata for new file extents after IO is complete.
  68  */
  69 struct end_io_wq {
  70         struct bio *bio;
  71         bio_end_io_t *end_io;
  72         void *private;
  73         struct btrfs_fs_info *info;
  74         int error;
  75         int metadata;
  76         struct list_head list;
  77         struct btrfs_work work;
  78 };
  79
  80 /*
  81  * async submit bios are used to offload expensive checksumming
  82  * onto the worker threads.  They checksum file and metadata bios
  83  * just before they are sent down the IO stack.
  84  */
  85 struct async_submit_bio {
  86         struct inode *inode;
  87         struct bio *bio;
  88         struct list_head list;
  89         extent_submit_bio_hook_t *submit_bio_start;
  90         extent_submit_bio_hook_t *submit_bio_done;
  91         int rw;
  92         int mirror_num;
  93         unsigned long bio_flags;
  94         /*
  95          * bio_offset is optional, can be used if the pages in the bio
  96          * can't tell us where in the file the bio should go
  97          */
  98         u64 bio_offset;
  99         struct btrfs_work work;
 100 };
 101
 102 /* These are used to set the lockdep class on the extent buffer locks.
 103  * The class is set by the readpage_end_io_hook after the buffer has
 104  * passed csum validation but before the pages are unlocked.
 105  *
 106  * The lockdep class is also set by btrfs_init_new_buffer on freshly
 107  * allocated blocks.
 108  *
 109  * The class is based on the level in the tree block, which allows lockdep
 110  * to know that lower nodes nest inside the locks of higher nodes.
 111  *
 112  * We also add a check to make sure the highest level of the tree is
 113  * the same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this
 114  * code needs update as well.
 115  */
 116 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 117 # if BTRFS_MAX_LEVEL != 8
 118 #  error
 119 # endif
 120 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
 121 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
 122         /* leaf */
 123         "btrfs-extent-00",
 124         "btrfs-extent-01",
 125         "btrfs-extent-02",
 126         "btrfs-extent-03",
 127         "btrfs-extent-04",
 128         "btrfs-extent-05",
 129         "btrfs-extent-06",
 130         "btrfs-extent-07",
 131         /* highest possible level */
 132         "btrfs-extent-08",
 133 };
 134 #endif
 135
 136 /*
 137  * extents on the btree inode are pretty simple, there's one extent
 138  * that covers the entire device
 139  */
 140 static struct extent_map *btree_get_extent(struct inode *inode,
 141                 struct page *page, size_t page_offset, u64 start, u64 len,
 142                 int create)
 143 {
 144         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 145         struct extent_map *em;
 146         int ret;
 147
 148         read_lock(&em_tree->lock);
 149         em = lookup_extent_mapping(em_tree, start, len);
 150         if (em) {
 151                 em->bdev =
 152                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 153                 read_unlock(&em_tree->lock);
 154                 goto out;
 155         }
 156         read_unlock(&em_tree->lock);
 157
 158         em = alloc_extent_map(GFP_NOFS);
 159         if (!em) {
 160                 em = ERR_PTR(-ENOMEM);
 161                 goto out;
 162         }
 163         em->start = 0;
 164         em->len = (u64)-1;
 165         em->block_len = (u64)-1;
 166         em->block_start = 0;
 167         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 168
 169         write_lock(&em_tree->lock);
 170         ret = add_extent_mapping(em_tree, em);
 171         if (ret == -EEXIST) {
 172                 u64 failed_start = em->start;
 173                 u64 failed_len = em->len;
 174
 175                 free_extent_map(em);
 176                 em = lookup_extent_mapping(em_tree, start, len);
 177                 if (em) {
 178                         ret = 0;
 179                 } else {
 180                         em = lookup_extent_mapping(em_tree, failed_start,
 181                                                    failed_len);
 182                         ret = -EIO;
 183                 }
 184         } else if (ret) {
 185                 free_extent_map(em);
 186                 em = NULL;
 187         }
 188         write_unlock(&em_tree->lock);
 189
 190         if (ret)
 191                 em = ERR_PTR(ret);
 192 out:
 193         return em;
 194 }
 195
 196 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 197 {
 198         return crc32c(seed, data, len);
 199 }
 200
 201 void btrfs_csum_final(u32 crc, char *result)
 202 {
 203         put_unaligned_le32(~crc, result);
 204 }
 205
 206 /*
 207  * compute the csum for a btree block, and either verify it or write it
 208  * into the csum field of the block.
 209  */
 210 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 211                            int verify)
 212 {
 213         u16 csum_size =
 214                 btrfs_super_csum_size(&root->fs_info->super_copy);
 215         char *result = NULL;
 216         unsigned long len;
 217         unsigned long cur_len;
 218         unsigned long offset = BTRFS_CSUM_SIZE;
 219         char *map_token = NULL;
 220         char *kaddr;
 221         unsigned long map_start;
 222         unsigned long map_len;
 223         int err;
 224         u32 crc = ~(u32)0;
 225         unsigned long inline_result;
 226
 227         len = buf->len - offset;
 228         while (len > 0) {
 229                 err = map_private_extent_buffer(buf, offset, 32,
 230                                         &map_token, &kaddr,
 231                                         &map_start, &map_len, KM_USER0);
 232                 if (err)
 233                         return 1;
 234                 cur_len = min(len, map_len - (offset - map_start));
 235                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 236                                       crc, cur_len);
 237                 len -= cur_len;
 238                 offset += cur_len;
 239                 unmap_extent_buffer(buf, map_token, KM_USER0);
 240         }
 241         if (csum_size > sizeof(inline_result)) {
 242                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
 243                 if (!result)
 244                         return 1;
 245         } else {
 246                 result = (char *)&inline_result;
 247         }
 248
 249         btrfs_csum_final(crc, result);
 250
 251         if (verify) {
 252                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 253                         u32 val;
 254                         u32 found = 0;
 255                         memcpy(&found, result, csum_size);
 256
 257                         read_extent_buffer(buf, &val, 0, csum_size);
 258                         if (printk_ratelimit()) {
 259                                 printk(KERN_INFO "btrfs: %s checksum verify "
 260                                        "failed on %llu wanted %X found %X "
 261                                        "level %d\n",
 262                                        root->fs_info->sb->s_id,
 263                                        (unsigned long long)buf->start, val, found,
 264                                        btrfs_header_level(buf));
 265                         }
 266                         if (result != (char *)&inline_result)
 267                                 kfree(result);
 268                         return 1;
 269                 }
 270         } else {
 271                 write_extent_buffer(buf, result, 0, csum_size);
 272         }
 273         if (result != (char *)&inline_result)
 274                 kfree(result);
 275         return 0;
 276 }
 277
 278 /*
 279  * we can't consider a given block up to date unless the transid of the
 280  * block matches the transid in the parent node's pointer.  This is how we
 281  * detect blocks that either didn't get written at all or got written
 282  * in the wrong place.
 283  */
 284 static int verify_parent_transid(struct extent_io_tree *io_tree,
 285                                  struct extent_buffer *eb, u64 parent_transid)
 286 {
 287         struct extent_state *cached_state = NULL;
 288         int ret;
 289
 290         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 291                 return 0;
 292
 293         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 294                          0, &cached_state, GFP_NOFS);
 295         if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
 296             btrfs_header_generation(eb) == parent_transid) {
 297                 ret = 0;
 298                 goto out;
 299         }
 300         if (printk_ratelimit()) {
 301                 printk("parent transid verify failed on %llu wanted %llu "
 302                        "found %llu\n",
 303                        (unsigned long long)eb->start,
 304                        (unsigned long long)parent_transid,
 305                        (unsigned long long)btrfs_header_generation(eb));
 306         }
 307         ret = 1;
 308         clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
 309 out:
 310         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 311                              &cached_state, GFP_NOFS);
 312         return ret;
 313 }
 314
 315 /*
 316  * helper to read a given tree block, doing retries as required when
 317  * the checksums don't match and we have alternate mirrors to try.
 318  */
 319 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 320                                           struct extent_buffer *eb,
 321                                           u64 start, u64 parent_transid)
 322 {
 323         struct extent_io_tree *io_tree;
 324         int ret;
 325         int num_copies = 0;
 326         int mirror_num = 0;
 327
 328         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 329         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 330         while (1) {
 331                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 332                                                btree_get_extent, mirror_num);
 333                 if (!ret &&
 334                     !verify_parent_transid(io_tree, eb, parent_transid))
 335                         return ret;
 336
 337                 /*
 338                  * This buffer's crc is fine, but its contents are corrupted, so
 339                  * there is no reason to read the other copies, they won't be
 340                  * any less wrong.
 341                  */
 342                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 343                         return ret;
 344
 345                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 346                                               eb->start, eb->len);
 347                 if (num_copies == 1)
 348                         return ret;
 349
 350                 mirror_num++;
 351                 if (mirror_num > num_copies)
 352                         return ret;
 353         }
 354         return -EIO;
 355 }
 356
 357 /*
 358  * checksum a dirty tree block before IO.  This has extra checks to make sure
 359  * we only fill in the checksum field in the first page of a multi-page block
 360  */
 361
 362 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 363 {
 364         struct extent_io_tree *tree;
 365         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 366         u64 found_start;
 367         unsigned long len;
 368         struct extent_buffer *eb;
 369         int ret;
 370
 371         tree = &BTRFS_I(page->mapping->host)->io_tree;
 372
 373         if (page->private == EXTENT_PAGE_PRIVATE) {
 374                 WARN_ON(1);
 375                 goto out;
 376         }
 377         if (!page->private) {
 378                 WARN_ON(1);
 379                 goto out;
 380         }
 381         len = page->private >> 2;
 382         WARN_ON(len == 0);
 383
 384         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 385         if (eb == NULL) {
 386                 WARN_ON(1);
 387                 goto out;
 388         }
 389         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 390                                              btrfs_header_generation(eb));
 391         BUG_ON(ret);
 392         WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
 393
 394         found_start = btrfs_header_bytenr(eb);
 395         if (found_start != start) {
 396                 WARN_ON(1);
 397                 goto err;
 398         }
 399         if (eb->first_page != page) {
 400                 WARN_ON(1);
 401                 goto err;
 402         }
 403         if (!PageUptodate(page)) {
 404                 WARN_ON(1);
 405                 goto err;
 406         }
 407         csum_tree_block(root, eb, 0);
 408 err:
 409         free_extent_buffer(eb);
 410 out:
 411         return 0;
 412 }
 413
 414 static int check_tree_block_fsid(struct btrfs_root *root,
 415                                  struct extent_buffer *eb)
 416 {
 417         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 418         u8 fsid[BTRFS_UUID_SIZE];
 419         int ret = 1;
 420
 421         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
 422                            BTRFS_FSID_SIZE);
 423         while (fs_devices) {
 424                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 425                         ret = 0;
 426                         break;
 427                 }
 428                 fs_devices = fs_devices->seed;
 429         }
 430         return ret;
 431 }
 432
 433 #define CORRUPT(reason, eb, root, slot)                         \
 434         printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
 435                "root=%llu, slot=%d\n", reason,                  \
 436                (unsigned long long)btrfs_header_bytenr(eb),     \
 437                (unsigned long long)root->objectid, slot)
 438
 439 static noinline int check_leaf(struct btrfs_root *root,
 440                                struct extent_buffer *leaf)
 441 {
 442         struct btrfs_key key;
 443         struct btrfs_key leaf_key;
 444         u32 nritems = btrfs_header_nritems(leaf);
 445         int slot;
 446
 447         if (nritems == 0)
 448                 return 0;
 449
 450         /* Check the 0 item */
 451         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 452             BTRFS_LEAF_DATA_SIZE(root)) {
 453                 CORRUPT("invalid item offset size pair", leaf, root, 0);
 454                 return -EIO;
 455         }
 456
 457         /*
 458          * Check to make sure each items keys are in the correct order and their
 459          * offsets make sense.  We only have to loop through nritems-1 because
 460          * we check the current slot against the next slot, which verifies the
 461          * next slot's offset+size makes sense and that the current's slot
 462          * offset is correct.
 463          */
 464         for (slot = 0; slot < nritems - 1; slot++) {
 465                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 466                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 467
 468                 /* Make sure the keys are in the right order */
 469                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 470                         CORRUPT("bad key order", leaf, root, slot);
 471                         return -EIO;
 472                 }
 473
 474                 /*
 475                  * Make sure the offset and ends are right, remember that the
 476                  * item data starts at the end of the leaf and grows towards the
 477                  * front.
 478                  */
 479                 if (btrfs_item_offset_nr(leaf, slot) !=
 480                         btrfs_item_end_nr(leaf, slot + 1)) {
 481                         CORRUPT("slot offset bad", leaf, root, slot);
 482                         return -EIO;
 483                 }
 484
 485                 /*
 486                  * Check to make sure that we don't point outside of the leaf,
 487                  * just incase all the items are consistent to eachother, but
 488                  * all point outside of the leaf.
 489                  */
 490                 if (btrfs_item_end_nr(leaf, slot) >
 491                     BTRFS_LEAF_DATA_SIZE(root)) {
 492                         CORRUPT("slot end outside of leaf", leaf, root, slot);
 493                         return -EIO;
 494                 }
 495         }
 496
 497         return 0;
 498 }
 499
 500 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 501 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
 502 {
 503         lockdep_set_class_and_name(&eb->lock,
 504                            &btrfs_eb_class[level],
 505                            btrfs_eb_name[level]);
 506 }
 507 #endif
 508
 509 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 510                                struct extent_state *state)
 511 {
 512         struct extent_io_tree *tree;
 513         u64 found_start;
 514         int found_level;
 515         unsigned long len;
 516         struct extent_buffer *eb;
 517         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 518         int ret = 0;
 519
 520         tree = &BTRFS_I(page->mapping->host)->io_tree;
 521         if (page->private == EXTENT_PAGE_PRIVATE)
 522                 goto out;
 523         if (!page->private)
 524                 goto out;
 525
 526         len = page->private >> 2;
 527         WARN_ON(len == 0);
 528
 529         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 530         if (eb == NULL) {
 531                 ret = -EIO;
 532                 goto out;
 533         }
 534
 535         found_start = btrfs_header_bytenr(eb);
 536         if (found_start != start) {
 537                 if (printk_ratelimit()) {
 538                         printk(KERN_INFO "btrfs bad tree block start "
 539                                "%llu %llu\n",
 540                                (unsigned long long)found_start,
 541                                (unsigned long long)eb->start);
 542                 }
 543                 ret = -EIO;
 544                 goto err;
 545         }
 546         if (eb->first_page != page) {
 547                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
 548                        eb->first_page->index, page->index);
 549                 WARN_ON(1);
 550                 ret = -EIO;
 551                 goto err;
 552         }
 553         if (check_tree_block_fsid(root, eb)) {
 554                 if (printk_ratelimit()) {
 555                         printk(KERN_INFO "btrfs bad fsid on block %llu\n",
 556                                (unsigned long long)eb->start);
 557                 }
 558                 ret = -EIO;
 559                 goto err;
 560         }
 561         found_level = btrfs_header_level(eb);
 562
 563         btrfs_set_buffer_lockdep_class(eb, found_level);
 564
 565         ret = csum_tree_block(root, eb, 1);
 566         if (ret) {
 567                 ret = -EIO;
 568                 goto err;
 569         }
 570
 571         /*
 572          * If this is a leaf block and it is corrupt, set the corrupt bit so
 573          * that we don't try and read the other copies of this block, just
 574          * return -EIO.
 575          */
 576         if (found_level == 0 && check_leaf(root, eb)) {
 577                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 578                 ret = -EIO;
 579         }
 580
 581         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 582         end = eb->start + end - 1;
 583 err:
 584         free_extent_buffer(eb);
 585 out:
 586         return ret;
 587 }
 588
 589 static void end_workqueue_bio(struct bio *bio, int err)
 590 {
 591         struct end_io_wq *end_io_wq = bio->bi_private;
 592         struct btrfs_fs_info *fs_info;
 593
 594         fs_info = end_io_wq->info;
 595         end_io_wq->error = err;
 596         end_io_wq->work.func = end_workqueue_fn;
 597         end_io_wq->work.flags = 0;
 598
 599         if (bio->bi_rw & REQ_WRITE) {
 600                 if (end_io_wq->metadata == 1)
 601                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
 602                                            &end_io_wq->work);
 603                 else if (end_io_wq->metadata == 2)
 604                         btrfs_queue_worker(&fs_info->endio_freespace_worker,
 605                                            &end_io_wq->work);
 606                 else
 607                         btrfs_queue_worker(&fs_info->endio_write_workers,
 608                                            &end_io_wq->work);
 609         } else {
 610                 if (end_io_wq->metadata)
 611                         btrfs_queue_worker(&fs_info->endio_meta_workers,
 612                                            &end_io_wq->work);
 613                 else
 614                         btrfs_queue_worker(&fs_info->endio_workers,
 615                                            &end_io_wq->work);
 616         }
 617 }
 618
 619 /*
 620  * For the metadata arg you want
 621  *
 622  * 0 - if data
 623  * 1 - if normal metadta
 624  * 2 - if writing to the free space cache area
 625  */
 626 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 627                         int metadata)
 628 {
 629         struct end_io_wq *end_io_wq;
 630         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 631         if (!end_io_wq)
 632                 return -ENOMEM;
 633
 634         end_io_wq->private = bio->bi_private;
 635         end_io_wq->end_io = bio->bi_end_io;
 636         end_io_wq->info = info;
 637         end_io_wq->error = 0;
 638         end_io_wq->bio = bio;
 639         end_io_wq->metadata = metadata;
 640
 641         bio->bi_private = end_io_wq;
 642         bio->bi_end_io = end_workqueue_bio;
 643         return 0;
 644 }
 645
 646 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 647 {
 648         unsigned long limit = min_t(unsigned long,
 649                                     info->workers.max_workers,
 650                                     info->fs_devices->open_devices);
 651         return 256 * limit;
 652 }
 653
 654 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
 655 {
 656         return atomic_read(&info->nr_async_bios) >
 657                 btrfs_async_submit_limit(info);
 658 }
 659
 660 static void run_one_async_start(struct btrfs_work *work)
 661 {
 662         struct async_submit_bio *async;
 663
 664         async = container_of(work, struct  async_submit_bio, work);
 665         async->submit_bio_start(async->inode, async->rw, async->bio,
 666                                async->mirror_num, async->bio_flags,
 667                                async->bio_offset);
 668 }
 669
 670 static void run_one_async_done(struct btrfs_work *work)
 671 {
 672         struct btrfs_fs_info *fs_info;
 673         struct async_submit_bio *async;
 674         int limit;
 675
 676         async = container_of(work, struct  async_submit_bio, work);
 677         fs_info = BTRFS_I(async->inode)->root->fs_info;
 678
 679         limit = btrfs_async_submit_limit(fs_info);
 680         limit = limit * 2 / 3;
 681
 682         atomic_dec(&fs_info->nr_async_submits);
 683
 684         if (atomic_read(&fs_info->nr_async_submits) < limit &&
 685             waitqueue_active(&fs_info->async_submit_wait))
 686                 wake_up(&fs_info->async_submit_wait);
 687
 688         async->submit_bio_done(async->inode, async->rw, async->bio,
 689                                async->mirror_num, async->bio_flags,
 690                                async->bio_offset);
 691 }
 692
 693 static void run_one_async_free(struct btrfs_work *work)
 694 {
 695         struct async_submit_bio *async;
 696
 697         async = container_of(work, struct  async_submit_bio, work);
 698         kfree(async);
 699 }
 700
 701 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 702                         int rw, struct bio *bio, int mirror_num,
 703                         unsigned long bio_flags,
 704                         u64 bio_offset,
 705                         extent_submit_bio_hook_t *submit_bio_start,
 706                         extent_submit_bio_hook_t *submit_bio_done)
 707 {
 708         struct async_submit_bio *async;
 709
 710         async = kmalloc(sizeof(*async), GFP_NOFS);
 711         if (!async)
 712                 return -ENOMEM;
 713
 714         async->inode = inode;
 715         async->rw = rw;
 716         async->bio = bio;
 717         async->mirror_num = mirror_num;
 718         async->submit_bio_start = submit_bio_start;
 719         async->submit_bio_done = submit_bio_done;
 720
 721         async->work.func = run_one_async_start;
 722         async->work.ordered_func = run_one_async_done;
 723         async->work.ordered_free = run_one_async_free;
 724
 725         async->work.flags = 0;
 726         async->bio_flags = bio_flags;
 727         async->bio_offset = bio_offset;
 728
 729         atomic_inc(&fs_info->nr_async_submits);
 730
 731         if (rw & REQ_SYNC)
 732                 btrfs_set_work_high_prio(&async->work);
 733
 734         btrfs_queue_worker(&fs_info->workers, &async->work);
 735
 736         while (atomic_read(&fs_info->async_submit_draining) &&
 737               atomic_read(&fs_info->nr_async_submits)) {
 738                 wait_event(fs_info->async_submit_wait,
 739                            (atomic_read(&fs_info->nr_async_submits) == 0));
 740         }
 741
 742         return 0;
 743 }
 744
 745 static int btree_csum_one_bio(struct bio *bio)
 746 {
 747         struct bio_vec *bvec = bio->bi_io_vec;
 748         int bio_index = 0;
 749         struct btrfs_root *root;
 750
 751         WARN_ON(bio->bi_vcnt <= 0);
 752         while (bio_index < bio->bi_vcnt) {
 753                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 754                 csum_dirty_buffer(root, bvec->bv_page);
 755                 bio_index++;
 756                 bvec++;
 757         }
 758         return 0;
 759 }
 760
 761 static int __btree_submit_bio_start(struct inode *inode, int rw,
 762                                     struct bio *bio, int mirror_num,
 763                                     unsigned long bio_flags,
 764                                     u64 bio_offset)
 765 {
 766         /*
 767          * when we're called for a write, we're already in the async
 768          * submission context.  Just jump into btrfs_map_bio
 769          */
 770         btree_csum_one_bio(bio);
 771         return 0;
 772 }
 773
 774 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 775                                  int mirror_num, unsigned long bio_flags,
 776                                  u64 bio_offset)
 777 {
 778         /*
 779          * when we're called for a write, we're already in the async
 780          * submission context.  Just jump into btrfs_map_bio
 781          */
 782         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 783 }
 784
 785 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 786                                  int mirror_num, unsigned long bio_flags,
 787                                  u64 bio_offset)
 788 {
 789         int ret;
 790
 791         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 792                                           bio, 1);
 793         BUG_ON(ret);
 794
 795         if (!(rw & REQ_WRITE)) {
 796                 /*
 797                  * called for a read, do the setup so that checksum validation
 798                  * can happen in the async kernel threads
 799                  */
 800                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 801                                      mirror_num, 0);
 802         }
 803
 804         /*
 805          * kthread helpers are used to submit writes so that checksumming
 806          * can happen in parallel across all CPUs
 807          */
 808         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 809                                    inode, rw, bio, mirror_num, 0,
 810                                    bio_offset,
 811                                    __btree_submit_bio_start,
 812                                    __btree_submit_bio_done);
 813 }
 814
 815 #ifdef CONFIG_MIGRATION
 816 static int btree_migratepage(struct address_space *mapping,
 817                         struct page *newpage, struct page *page)
 818 {
 819         /*
 820          * we can't safely write a btree page from here,
 821          * we haven't done the locking hook
 822          */
 823         if (PageDirty(page))
 824                 return -EAGAIN;
 825         /*
 826          * Buffers may be managed in a filesystem specific way.
 827          * We must have no buffers or drop them.
 828          */
 829         if (page_has_private(page) &&
 830             !try_to_release_page(page, GFP_KERNEL))
 831                 return -EAGAIN;
 832         return migrate_page(mapping, newpage, page);
 833 }
 834 #endif
 835
 836 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 837 {
 838         struct extent_io_tree *tree;
 839         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 840         struct extent_buffer *eb;
 841         int was_dirty;
 842
 843         tree = &BTRFS_I(page->mapping->host)->io_tree;
 844         if (!(current->flags & PF_MEMALLOC)) {
 845                 return extent_write_full_page(tree, page,
 846                                               btree_get_extent, wbc);
 847         }
 848
 849         redirty_page_for_writepage(wbc, page);
 850         eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
 851         WARN_ON(!eb);
 852
 853         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
 854         if (!was_dirty) {
 855                 spin_lock(&root->fs_info->delalloc_lock);
 856                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
 857                 spin_unlock(&root->fs_info->delalloc_lock);
 858         }
 859         free_extent_buffer(eb);
 860
 861         unlock_page(page);
 862         return 0;
 863 }
 864
 865 static int btree_writepages(struct address_space *mapping,
 866                             struct writeback_control *wbc)
 867 {
 868         struct extent_io_tree *tree;
 869         tree = &BTRFS_I(mapping->host)->io_tree;
 870         if (wbc->sync_mode == WB_SYNC_NONE) {
 871                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
 872                 u64 num_dirty;
 873                 unsigned long thresh = 32 * 1024 * 1024;
 874
 875                 if (wbc->for_kupdate)
 876                         return 0;
 877
 878                 /* this is a bit racy, but that's ok */
 879                 num_dirty = root->fs_info->dirty_metadata_bytes;
 880                 if (num_dirty < thresh)
 881                         return 0;
 882         }
 883         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 884 }
 885
 886 static int btree_readpage(struct file *file, struct page *page)
 887 {
 888         struct extent_io_tree *tree;
 889         tree = &BTRFS_I(page->mapping->host)->io_tree;
 890         return extent_read_full_page(tree, page, btree_get_extent);
 891 }
 892
 893 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 894 {
 895         struct extent_io_tree *tree;
 896         struct extent_map_tree *map;
 897         int ret;
 898
 899         if (PageWriteback(page) || PageDirty(page))
 900                 return 0;
 901
 902         tree = &BTRFS_I(page->mapping->host)->io_tree;
 903         map = &BTRFS_I(page->mapping->host)->extent_tree;
 904
 905         ret = try_release_extent_state(map, tree, page, gfp_flags);
 906         if (!ret)
 907                 return 0;
 908
 909         ret = try_release_extent_buffer(tree, page);
 910         if (ret == 1) {
 911                 ClearPagePrivate(page);
 912                 set_page_private(page, 0);
 913                 page_cache_release(page);
 914         }
 915
 916         return ret;
 917 }
 918
 919 static void btree_invalidatepage(struct page *page, unsigned long offset)
 920 {
 921         struct extent_io_tree *tree;
 922         tree = &BTRFS_I(page->mapping->host)->io_tree;
 923         extent_invalidatepage(tree, page, offset);
 924         btree_releasepage(page, GFP_NOFS);
 925         if (PagePrivate(page)) {
 926                 printk(KERN_WARNING "btrfs warning page private not zero "
 927                        "on page %llu\n", (unsigned long long)page_offset(page));
 928                 ClearPagePrivate(page);
 929                 set_page_private(page, 0);
 930                 page_cache_release(page);
 931         }
 932 }
 933
 934 static const struct address_space_operations btree_aops = {
 935         .readpage       = btree_readpage,
 936         .writepage      = btree_writepage,
 937         .writepages     = btree_writepages,
 938         .releasepage    = btree_releasepage,
 939         .invalidatepage = btree_invalidatepage,
 940 #ifdef CONFIG_MIGRATION
 941         .migratepage    = btree_migratepage,
 942 #endif
 943 };
 944
 945 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 946                          u64 parent_transid)
 947 {
 948         struct extent_buffer *buf = NULL;
 949         struct inode *btree_inode = root->fs_info->btree_inode;
 950         int ret = 0;
 951
 952         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 953         if (!buf)
 954                 return 0;
 955         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 956                                  buf, 0, 0, btree_get_extent, 0);
 957         free_extent_buffer(buf);
 958         return ret;
 959 }
 960
 961 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 962                                             u64 bytenr, u32 blocksize)
 963 {
 964         struct inode *btree_inode = root->fs_info->btree_inode;
 965         struct extent_buffer *eb;
 966         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 967                                 bytenr, blocksize, GFP_NOFS);
 968         return eb;
 969 }
 970
 971 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 972                                                  u64 bytenr, u32 blocksize)
 973 {
 974         struct inode *btree_inode = root->fs_info->btree_inode;
 975         struct extent_buffer *eb;
 976
 977         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 978                                  bytenr, blocksize, NULL, GFP_NOFS);
 979         return eb;
 980 }
 981
 982
 983 int btrfs_write_tree_block(struct extent_buffer *buf)
 984 {
 985         return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
 986                                         buf->start + buf->len - 1);
 987 }
 988
 989 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 990 {
 991         return filemap_fdatawait_range(buf->first_page->mapping,
 992                                        buf->start, buf->start + buf->len - 1);
 993 }
 994
 995 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 996                                       u32 blocksize, u64 parent_transid)
 997 {
 998         struct extent_buffer *buf = NULL;
 999         int ret;
1000
1001         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1002         if (!buf)
1003                 return NULL;
1004
1005         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1006
1007         if (ret == 0)
1008                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1009         return buf;
1010
1011 }
1012
1013 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1014                      struct extent_buffer *buf)
1015 {
1016         struct inode *btree_inode = root->fs_info->btree_inode;
1017         if (btrfs_header_generation(buf) ==
1018             root->fs_info->running_transaction->transid) {
1019                 btrfs_assert_tree_locked(buf);
1020
1021                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1022                         spin_lock(&root->fs_info->delalloc_lock);
1023                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
1024                                 root->fs_info->dirty_metadata_bytes -= buf->len;
1025                         else
1026                                 WARN_ON(1);
1027                         spin_unlock(&root->fs_info->delalloc_lock);
1028                 }
1029
1030                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1031                 btrfs_set_lock_blocking(buf);
1032                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1033                                           buf);
1034         }
1035         return 0;
1036 }
1037
1038 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1039                         u32 stripesize, struct btrfs_root *root,
1040                         struct btrfs_fs_info *fs_info,
1041                         u64 objectid)
1042 {
1043         root->node = NULL;
1044         root->commit_root = NULL;
1045         root->sectorsize = sectorsize;
1046         root->nodesize = nodesize;
1047         root->leafsize = leafsize;
1048         root->stripesize = stripesize;
1049         root->ref_cows = 0;
1050         root->track_dirty = 0;
1051         root->in_radix = 0;
1052         root->orphan_item_inserted = 0;
1053         root->orphan_cleanup_state = 0;
1054
1055         root->fs_info = fs_info;
1056         root->objectid = objectid;
1057         root->last_trans = 0;
1058         root->highest_objectid = 0;
1059         root->name = NULL;
1060         root->in_sysfs = 0;
1061         root->inode_tree = RB_ROOT;
1062         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1063         root->block_rsv = NULL;
1064         root->orphan_block_rsv = NULL;
1065
1066         INIT_LIST_HEAD(&root->dirty_list);
1067         INIT_LIST_HEAD(&root->orphan_list);
1068         INIT_LIST_HEAD(&root->root_list);
1069         spin_lock_init(&root->node_lock);
1070         spin_lock_init(&root->orphan_lock);
1071         spin_lock_init(&root->inode_lock);
1072         spin_lock_init(&root->accounting_lock);
1073         mutex_init(&root->objectid_mutex);
1074         mutex_init(&root->log_mutex);
1075         init_waitqueue_head(&root->log_writer_wait);
1076         init_waitqueue_head(&root->log_commit_wait[0]);
1077         init_waitqueue_head(&root->log_commit_wait[1]);
1078         atomic_set(&root->log_commit[0], 0);
1079         atomic_set(&root->log_commit[1], 0);
1080         atomic_set(&root->log_writers, 0);
1081         root->log_batch = 0;
1082         root->log_transid = 0;
1083         root->last_log_commit = 0;
1084         extent_io_tree_init(&root->dirty_log_pages,
1085                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1086
1087         memset(&root->root_key, 0, sizeof(root->root_key));
1088         memset(&root->root_item, 0, sizeof(root->root_item));
1089         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1090         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1091         root->defrag_trans_start = fs_info->generation;
1092         init_completion(&root->kobj_unregister);
1093         root->defrag_running = 0;
1094         root->root_key.objectid = objectid;
1095         root->anon_super.s_root = NULL;
1096         root->anon_super.s_dev = 0;
1097         INIT_LIST_HEAD(&root->anon_super.s_list);
1098         INIT_LIST_HEAD(&root->anon_super.s_instances);
1099         init_rwsem(&root->anon_super.s_umount);
1100
1101         return 0;
1102 }
1103
1104 static int find_and_setup_root(struct btrfs_root *tree_root,
1105                                struct btrfs_fs_info *fs_info,
1106                                u64 objectid,
1107                                struct btrfs_root *root)
1108 {
1109         int ret;
1110         u32 blocksize;
1111         u64 generation;
1112
1113         __setup_root(tree_root->nodesize, tree_root->leafsize,
1114                      tree_root->sectorsize, tree_root->stripesize,
1115                      root, fs_info, objectid);
1116         ret = btrfs_find_last_root(tree_root, objectid,
1117                                    &root->root_item, &root->root_key);
1118         if (ret > 0)
1119                 return -ENOENT;
1120         BUG_ON(ret);
1121
1122         generation = btrfs_root_generation(&root->root_item);
1123         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1124         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1125                                      blocksize, generation);
1126         if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1127                 free_extent_buffer(root->node);
1128                 return -EIO;
1129         }
1130         root->commit_root = btrfs_root_node(root);
1131         return 0;
1132 }
1133
1134 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1135                                          struct btrfs_fs_info *fs_info)
1136 {
1137         struct btrfs_root *root;
1138         struct btrfs_root *tree_root = fs_info->tree_root;
1139         struct extent_buffer *leaf;
1140
1141         root = kzalloc(sizeof(*root), GFP_NOFS);
1142         if (!root)
1143                 return ERR_PTR(-ENOMEM);
1144
1145         __setup_root(tree_root->nodesize, tree_root->leafsize,
1146                      tree_root->sectorsize, tree_root->stripesize,
1147                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1148
1149         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1150         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1151         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1152         /*
1153          * log trees do not get reference counted because they go away
1154          * before a real commit is actually done.  They do store pointers
1155          * to file data extents, and those reference counts still get
1156          * updated (along with back refs to the log tree).
1157          */
1158         root->ref_cows = 0;
1159
1160         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1161                                       BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1162         if (IS_ERR(leaf)) {
1163                 kfree(root);
1164                 return ERR_CAST(leaf);
1165         }
1166
1167         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1168         btrfs_set_header_bytenr(leaf, leaf->start);
1169         btrfs_set_header_generation(leaf, trans->transid);
1170         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1171         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1172         root->node = leaf;
1173
1174         write_extent_buffer(root->node, root->fs_info->fsid,
1175                             (unsigned long)btrfs_header_fsid(root->node),
1176                             BTRFS_FSID_SIZE);
1177         btrfs_mark_buffer_dirty(root->node);
1178         btrfs_tree_unlock(root->node);
1179         return root;
1180 }
1181
1182 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1183                              struct btrfs_fs_info *fs_info)
1184 {
1185         struct btrfs_root *log_root;
1186
1187         log_root = alloc_log_tree(trans, fs_info);
1188         if (IS_ERR(log_root))
1189                 return PTR_ERR(log_root);
1190         WARN_ON(fs_info->log_root_tree);
1191         fs_info->log_root_tree = log_root;
1192         return 0;
1193 }
1194
1195 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1196                        struct btrfs_root *root)
1197 {
1198         struct btrfs_root *log_root;
1199         struct btrfs_inode_item *inode_item;
1200
1201         log_root = alloc_log_tree(trans, root->fs_info);
1202         if (IS_ERR(log_root))
1203                 return PTR_ERR(log_root);
1204
1205         log_root->last_trans = trans->transid;
1206         log_root->root_key.offset = root->root_key.objectid;
1207
1208         inode_item = &log_root->root_item.inode;
1209         inode_item->generation = cpu_to_le64(1);
1210         inode_item->size = cpu_to_le64(3);
1211         inode_item->nlink = cpu_to_le32(1);
1212         inode_item->nbytes = cpu_to_le64(root->leafsize);
1213         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1214
1215         btrfs_set_root_node(&log_root->root_item, log_root->node);
1216
1217         WARN_ON(root->log_root);
1218         root->log_root = log_root;
1219         root->log_transid = 0;
1220         root->last_log_commit = 0;
1221         return 0;
1222 }
1223
1224 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1225                                                struct btrfs_key *location)
1226 {
1227         struct btrfs_root *root;
1228         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1229         struct btrfs_path *path;
1230         struct extent_buffer *l;
1231         u64 generation;
1232         u32 blocksize;
1233         int ret = 0;
1234
1235         root = kzalloc(sizeof(*root), GFP_NOFS);
1236         if (!root)
1237                 return ERR_PTR(-ENOMEM);
1238         if (location->offset == (u64)-1) {
1239                 ret = find_and_setup_root(tree_root, fs_info,
1240                                           location->objectid, root);
1241                 if (ret) {
1242                         kfree(root);
1243                         return ERR_PTR(ret);
1244                 }
1245                 goto out;
1246         }
1247
1248         __setup_root(tree_root->nodesize, tree_root->leafsize,
1249                      tree_root->sectorsize, tree_root->stripesize,
1250                      root, fs_info, location->objectid);
1251
1252         path = btrfs_alloc_path();
1253         if (!path) {
1254                 kfree(root);
1255                 return ERR_PTR(-ENOMEM);
1256         }
1257         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1258         if (ret == 0) {
1259                 l = path->nodes[0];
1260                 read_extent_buffer(l, &root->root_item,
1261                                 btrfs_item_ptr_offset(l, path->slots[0]),
1262                                 sizeof(root->root_item));
1263                 memcpy(&root->root_key, location, sizeof(*location));
1264         }
1265         btrfs_free_path(path);
1266         if (ret) {
1267                 kfree(root);
1268                 if (ret > 0)
1269                         ret = -ENOENT;
1270                 return ERR_PTR(ret);
1271         }
1272
1273         generation = btrfs_root_generation(&root->root_item);
1274         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1275         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1276                                      blocksize, generation);
1277         root->commit_root = btrfs_root_node(root);
1278         BUG_ON(!root->node);
1279 out:
1280         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1281                 root->ref_cows = 1;
1282                 btrfs_check_and_init_root_item(&root->root_item);
1283         }
1284
1285         return root;
1286 }
1287
1288 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1289                                         u64 root_objectid)
1290 {
1291         struct btrfs_root *root;
1292
1293         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1294                 return fs_info->tree_root;
1295         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1296                 return fs_info->extent_root;
1297
1298         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1299                                  (unsigned long)root_objectid);
1300         return root;
1301 }
1302
1303 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1304                                               struct btrfs_key *location)
1305 {
1306         struct btrfs_root *root;
1307         int ret;
1308
1309         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1310                 return fs_info->tree_root;
1311         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1312                 return fs_info->extent_root;
1313         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1314                 return fs_info->chunk_root;
1315         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1316                 return fs_info->dev_root;
1317         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1318                 return fs_info->csum_root;
1319 again:
1320         spin_lock(&fs_info->fs_roots_radix_lock);
1321         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1322                                  (unsigned long)location->objectid);
1323         spin_unlock(&fs_info->fs_roots_radix_lock);
1324         if (root)
1325                 return root;
1326
1327         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1328         if (IS_ERR(root))
1329                 return root;
1330
1331         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1332         if (!root->free_ino_ctl)
1333                 goto fail;
1334         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1335                                         GFP_NOFS);
1336         if (!root->free_ino_pinned)
1337                 goto fail;
1338
1339         btrfs_init_free_ino_ctl(root);
1340         mutex_init(&root->fs_commit_mutex);
1341         spin_lock_init(&root->cache_lock);
1342         init_waitqueue_head(&root->cache_wait);
1343
1344         set_anon_super(&root->anon_super, NULL);
1345
1346         if (btrfs_root_refs(&root->root_item) == 0) {
1347                 ret = -ENOENT;
1348                 goto fail;
1349         }
1350
1351         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1352         if (ret < 0)
1353                 goto fail;
1354         if (ret == 0)
1355                 root->orphan_item_inserted = 1;
1356
1357         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1358         if (ret)
1359                 goto fail;
1360
1361         spin_lock(&fs_info->fs_roots_radix_lock);
1362         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1363                                 (unsigned long)root->root_key.objectid,
1364                                 root);
1365         if (ret == 0)
1366                 root->in_radix = 1;
1367
1368         spin_unlock(&fs_info->fs_roots_radix_lock);
1369         radix_tree_preload_end();
1370         if (ret) {
1371                 if (ret == -EEXIST) {
1372                         free_fs_root(root);
1373                         goto again;
1374                 }
1375                 goto fail;
1376         }
1377
1378         ret = btrfs_find_dead_roots(fs_info->tree_root,
1379                                     root->root_key.objectid);
1380         WARN_ON(ret);
1381         return root;
1382 fail:
1383         free_fs_root(root);
1384         return ERR_PTR(ret);
1385 }
1386
1387 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1388                                       struct btrfs_key *location,
1389                                       const char *name, int namelen)
1390 {
1391         return btrfs_read_fs_root_no_name(fs_info, location);
1392 #if 0
1393         struct btrfs_root *root;
1394         int ret;
1395
1396         root = btrfs_read_fs_root_no_name(fs_info, location);
1397         if (!root)
1398                 return NULL;
1399
1400         if (root->in_sysfs)
1401                 return root;
1402
1403         ret = btrfs_set_root_name(root, name, namelen);
1404         if (ret) {
1405                 free_extent_buffer(root->node);
1406                 kfree(root);
1407                 return ERR_PTR(ret);
1408         }
1409
1410         ret = btrfs_sysfs_add_root(root);
1411         if (ret) {
1412                 free_extent_buffer(root->node);
1413                 kfree(root->name);
1414                 kfree(root);
1415                 return ERR_PTR(ret);
1416         }
1417         root->in_sysfs = 1;
1418         return root;
1419 #endif
1420 }
1421
1422 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1423 {
1424         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1425         int ret = 0;
1426         struct btrfs_device *device;
1427         struct backing_dev_info *bdi;
1428
1429         list_for_each_entry(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         return ret;
1439 }
1440
1441 /*
1442  * If this fails, caller must call bdi_destroy() to get rid of the
1443  * bdi again.
1444  */
1445 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1446 {
1447         int err;
1448
1449         bdi->capabilities = BDI_CAP_MAP_COPY;
1450         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1451         if (err)
1452                 return err;
1453
1454         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1455         bdi->congested_fn       = btrfs_congested_fn;
1456         bdi->congested_data     = info;
1457         return 0;
1458 }
1459
1460 static int bio_ready_for_csum(struct bio *bio)
1461 {
1462         u64 length = 0;
1463         u64 buf_len = 0;
1464         u64 start = 0;
1465         struct page *page;
1466         struct extent_io_tree *io_tree = NULL;
1467         struct bio_vec *bvec;
1468         int i;
1469         int ret;
1470
1471         bio_for_each_segment(bvec, bio, i) {
1472                 page = bvec->bv_page;
1473                 if (page->private == EXTENT_PAGE_PRIVATE) {
1474                         length += bvec->bv_len;
1475                         continue;
1476                 }
1477                 if (!page->private) {
1478                         length += bvec->bv_len;
1479                         continue;
1480                 }
1481                 length = bvec->bv_len;
1482                 buf_len = page->private >> 2;
1483                 start = page_offset(page) + bvec->bv_offset;
1484                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1485         }
1486         /* are we fully contained in this bio? */
1487         if (buf_len <= length)
1488                 return 1;
1489
1490         ret = extent_range_uptodate(io_tree, start + length,
1491                                     start + buf_len - 1);
1492         return ret;
1493 }
1494
1495 /*
1496  * called by the kthread helper functions to finally call the bio end_io
1497  * functions.  This is where read checksum verification actually happens
1498  */
1499 static void end_workqueue_fn(struct btrfs_work *work)
1500 {
1501         struct bio *bio;
1502         struct end_io_wq *end_io_wq;
1503         struct btrfs_fs_info *fs_info;
1504         int error;
1505
1506         end_io_wq = container_of(work, struct end_io_wq, work);
1507         bio = end_io_wq->bio;
1508         fs_info = end_io_wq->info;
1509
1510         /* metadata bio reads are special because the whole tree block must
1511          * be checksummed at once.  This makes sure the entire block is in
1512          * ram and up to date before trying to verify things.  For
1513          * blocksize <= pagesize, it is basically a noop
1514          */
1515         if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1516             !bio_ready_for_csum(bio)) {
1517                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1518                                    &end_io_wq->work);
1519                 return;
1520         }
1521         error = end_io_wq->error;
1522         bio->bi_private = end_io_wq->private;
1523         bio->bi_end_io = end_io_wq->end_io;
1524         kfree(end_io_wq);
1525         bio_endio(bio, error);
1526 }
1527
1528 static int cleaner_kthread(void *arg)
1529 {
1530         struct btrfs_root *root = arg;
1531
1532         do {
1533                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1534
1535                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1536                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1537                         btrfs_run_delayed_iputs(root);
1538                         btrfs_clean_old_snapshots(root);
1539                         mutex_unlock(&root->fs_info->cleaner_mutex);
1540                 }
1541
1542                 if (freezing(current)) {
1543                         refrigerator();
1544                 } else {
1545                         set_current_state(TASK_INTERRUPTIBLE);
1546                         if (!kthread_should_stop())
1547                                 schedule();
1548                         __set_current_state(TASK_RUNNING);
1549                 }
1550         } while (!kthread_should_stop());
1551         return 0;
1552 }
1553
1554 static int transaction_kthread(void *arg)
1555 {
1556         struct btrfs_root *root = arg;
1557         struct btrfs_trans_handle *trans;
1558         struct btrfs_transaction *cur;
1559         u64 transid;
1560         unsigned long now;
1561         unsigned long delay;
1562         int ret;
1563
1564         do {
1565                 delay = HZ * 30;
1566                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1567                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1568
1569                 spin_lock(&root->fs_info->new_trans_lock);
1570                 cur = root->fs_info->running_transaction;
1571                 if (!cur) {
1572                         spin_unlock(&root->fs_info->new_trans_lock);
1573                         goto sleep;
1574                 }
1575
1576                 now = get_seconds();
1577                 if (!cur->blocked &&
1578                     (now < cur->start_time || now - cur->start_time < 30)) {
1579                         spin_unlock(&root->fs_info->new_trans_lock);
1580                         delay = HZ * 5;
1581                         goto sleep;
1582                 }
1583                 transid = cur->transid;
1584                 spin_unlock(&root->fs_info->new_trans_lock);
1585
1586                 trans = btrfs_join_transaction(root, 1);
1587                 BUG_ON(IS_ERR(trans));
1588                 if (transid == trans->transid) {
1589                         ret = btrfs_commit_transaction(trans, root);
1590                         BUG_ON(ret);
1591                 } else {
1592                         btrfs_end_transaction(trans, root);
1593                 }
1594 sleep:
1595                 wake_up_process(root->fs_info->cleaner_kthread);
1596                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1597
1598                 if (freezing(current)) {
1599                         refrigerator();
1600                 } else {
1601                         set_current_state(TASK_INTERRUPTIBLE);
1602                         if (!kthread_should_stop() &&
1603                             !btrfs_transaction_blocked(root->fs_info))
1604                                 schedule_timeout(delay);
1605                         __set_current_state(TASK_RUNNING);
1606                 }
1607         } while (!kthread_should_stop());
1608         return 0;
1609 }
1610
1611 struct btrfs_root *open_ctree(struct super_block *sb,
1612                               struct btrfs_fs_devices *fs_devices,
1613                               char *options)
1614 {
1615         u32 sectorsize;
1616         u32 nodesize;
1617         u32 leafsize;
1618         u32 blocksize;
1619         u32 stripesize;
1620         u64 generation;
1621         u64 features;
1622         struct btrfs_key location;
1623         struct buffer_head *bh;
1624         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1625                                                  GFP_NOFS);
1626         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1627                                                  GFP_NOFS);
1628         struct btrfs_root *tree_root = btrfs_sb(sb);
1629         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1630         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1631                                                 GFP_NOFS);
1632         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1633                                               GFP_NOFS);
1634         struct btrfs_root *log_tree_root;
1635
1636         int ret;
1637         int err = -EINVAL;
1638
1639         struct btrfs_super_block *disk_super;
1640
1641         if (!extent_root || !tree_root || !fs_info ||
1642             !chunk_root || !dev_root || !csum_root) {
1643                 err = -ENOMEM;
1644                 goto fail;
1645         }
1646
1647         ret = init_srcu_struct(&fs_info->subvol_srcu);
1648         if (ret) {
1649                 err = ret;
1650                 goto fail;
1651         }
1652
1653         ret = setup_bdi(fs_info, &fs_info->bdi);
1654         if (ret) {
1655                 err = ret;
1656                 goto fail_srcu;
1657         }
1658
1659         fs_info->btree_inode = new_inode(sb);
1660         if (!fs_info->btree_inode) {
1661                 err = -ENOMEM;
1662                 goto fail_bdi;
1663         }
1664
1665         fs_info->btree_inode->i_mapping->flags &= ~__GFP_FS;
1666
1667         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1668         INIT_LIST_HEAD(&fs_info->trans_list);
1669         INIT_LIST_HEAD(&fs_info->dead_roots);
1670         INIT_LIST_HEAD(&fs_info->delayed_iputs);
1671         INIT_LIST_HEAD(&fs_info->hashers);
1672         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1673         INIT_LIST_HEAD(&fs_info->ordered_operations);
1674         INIT_LIST_HEAD(&fs_info->caching_block_groups);
1675         spin_lock_init(&fs_info->delalloc_lock);
1676         spin_lock_init(&fs_info->new_trans_lock);
1677         spin_lock_init(&fs_info->ref_cache_lock);
1678         spin_lock_init(&fs_info->fs_roots_radix_lock);
1679         spin_lock_init(&fs_info->delayed_iput_lock);
1680
1681         init_completion(&fs_info->kobj_unregister);
1682         fs_info->tree_root = tree_root;
1683         fs_info->extent_root = extent_root;
1684         fs_info->csum_root = csum_root;
1685         fs_info->chunk_root = chunk_root;
1686         fs_info->dev_root = dev_root;
1687         fs_info->fs_devices = fs_devices;
1688         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1689         INIT_LIST_HEAD(&fs_info->space_info);
1690         btrfs_mapping_init(&fs_info->mapping_tree);
1691         btrfs_init_block_rsv(&fs_info->global_block_rsv);
1692         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1693         btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1694         btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1695         btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1696         INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1697         mutex_init(&fs_info->durable_block_rsv_mutex);
1698         atomic_set(&fs_info->nr_async_submits, 0);
1699         atomic_set(&fs_info->async_delalloc_pages, 0);
1700         atomic_set(&fs_info->async_submit_draining, 0);
1701         atomic_set(&fs_info->nr_async_bios, 0);
1702         fs_info->sb = sb;
1703         fs_info->max_inline = 8192 * 1024;
1704         fs_info->metadata_ratio = 0;
1705
1706         fs_info->thread_pool_size = min_t(unsigned long,
1707                                           num_online_cpus() + 2, 8);
1708
1709         INIT_LIST_HEAD(&fs_info->ordered_extents);
1710         spin_lock_init(&fs_info->ordered_extent_lock);
1711         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1712                                         GFP_NOFS);
1713         if (!fs_info->delayed_root) {
1714                 err = -ENOMEM;
1715                 goto fail_iput;
1716         }
1717         btrfs_init_delayed_root(fs_info->delayed_root);
1718
1719         sb->s_blocksize = 4096;
1720         sb->s_blocksize_bits = blksize_bits(4096);
1721         sb->s_bdi = &fs_info->bdi;
1722
1723         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1724         fs_info->btree_inode->i_nlink = 1;
1725         /*
1726          * we set the i_size on the btree inode to the max possible int.
1727          * the real end of the address space is determined by all of
1728          * the devices in the system
1729          */
1730         fs_info->btree_inode->i_size = OFFSET_MAX;
1731         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1732         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1733
1734         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1735         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1736                              fs_info->btree_inode->i_mapping,
1737                              GFP_NOFS);
1738         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1739                              GFP_NOFS);
1740
1741         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1742
1743         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1744         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1745                sizeof(struct btrfs_key));
1746         BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1747         insert_inode_hash(fs_info->btree_inode);
1748
1749         spin_lock_init(&fs_info->block_group_cache_lock);
1750         fs_info->block_group_cache_tree = RB_ROOT;
1751
1752         extent_io_tree_init(&fs_info->freed_extents[0],
1753                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1754         extent_io_tree_init(&fs_info->freed_extents[1],
1755                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1756         fs_info->pinned_extents = &fs_info->freed_extents[0];
1757         fs_info->do_barriers = 1;
1758
1759
1760         mutex_init(&fs_info->trans_mutex);
1761         mutex_init(&fs_info->ordered_operations_mutex);
1762         mutex_init(&fs_info->tree_log_mutex);
1763         mutex_init(&fs_info->chunk_mutex);
1764         mutex_init(&fs_info->transaction_kthread_mutex);
1765         mutex_init(&fs_info->cleaner_mutex);
1766         mutex_init(&fs_info->volume_mutex);
1767         init_rwsem(&fs_info->extent_commit_sem);
1768         init_rwsem(&fs_info->cleanup_work_sem);
1769         init_rwsem(&fs_info->subvol_sem);
1770
1771         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1772         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1773
1774         init_waitqueue_head(&fs_info->transaction_throttle);
1775         init_waitqueue_head(&fs_info->transaction_wait);
1776         init_waitqueue_head(&fs_info->transaction_blocked_wait);
1777         init_waitqueue_head(&fs_info->async_submit_wait);
1778
1779         __setup_root(4096, 4096, 4096, 4096, tree_root,
1780                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1781
1782         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1783         if (!bh) {
1784                 err = -EINVAL;
1785                 goto fail_alloc;
1786         }
1787
1788         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1789         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1790                sizeof(fs_info->super_for_commit));
1791         brelse(bh);
1792
1793         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1794
1795         disk_super = &fs_info->super_copy;
1796         if (!btrfs_super_root(disk_super))
1797                 goto fail_alloc;
1798
1799         /* check FS state, whether FS is broken. */
1800         fs_info->fs_state |= btrfs_super_flags(disk_super);
1801
1802         btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
1803
1804         /*
1805          * In the long term, we'll store the compression type in the super
1806          * block, and it'll be used for per file compression control.
1807          */
1808         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
1809
1810         ret = btrfs_parse_options(tree_root, options);
1811         if (ret) {
1812                 err = ret;
1813                 goto fail_alloc;
1814         }
1815
1816         features = btrfs_super_incompat_flags(disk_super) &
1817                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1818         if (features) {
1819                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1820                        "unsupported optional features (%Lx).\n",
1821                        (unsigned long long)features);
1822                 err = -EINVAL;
1823                 goto fail_alloc;
1824         }
1825
1826         features = btrfs_super_incompat_flags(disk_super);
1827         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1828         if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1829                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1830         btrfs_set_super_incompat_flags(disk_super, features);
1831
1832         features = btrfs_super_compat_ro_flags(disk_super) &
1833                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1834         if (!(sb->s_flags & MS_RDONLY) && features) {
1835                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1836                        "unsupported option features (%Lx).\n",
1837                        (unsigned long long)features);
1838                 err = -EINVAL;
1839                 goto fail_alloc;
1840         }
1841
1842         btrfs_init_workers(&fs_info->generic_worker,
1843                            "genwork", 1, NULL);
1844
1845         btrfs_init_workers(&fs_info->workers, "worker",
1846                            fs_info->thread_pool_size,
1847                            &fs_info->generic_worker);
1848
1849         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1850                            fs_info->thread_pool_size,
1851                            &fs_info->generic_worker);
1852
1853         btrfs_init_workers(&fs_info->submit_workers, "submit",
1854                            min_t(u64, fs_devices->num_devices,
1855                            fs_info->thread_pool_size),
1856                            &fs_info->generic_worker);
1857
1858         /* a higher idle thresh on the submit workers makes it much more
1859          * likely that bios will be send down in a sane order to the
1860          * devices
1861          */
1862         fs_info->submit_workers.idle_thresh = 64;
1863
1864         fs_info->workers.idle_thresh = 16;
1865         fs_info->workers.ordered = 1;
1866
1867         fs_info->delalloc_workers.idle_thresh = 2;
1868         fs_info->delalloc_workers.ordered = 1;
1869
1870         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1871                            &fs_info->generic_worker);
1872         btrfs_init_workers(&fs_info->endio_workers, "endio",
1873                            fs_info->thread_pool_size,
1874                            &fs_info->generic_worker);
1875         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1876                            fs_info->thread_pool_size,
1877                            &fs_info->generic_worker);
1878         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1879                            "endio-meta-write", fs_info->thread_pool_size,
1880                            &fs_info->generic_worker);
1881         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1882                            fs_info->thread_pool_size,
1883                            &fs_info->generic_worker);
1884         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1885                            1, &fs_info->generic_worker);
1886         btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
1887                            fs_info->thread_pool_size,
1888                            &fs_info->generic_worker);
1889
1890         /*
1891          * endios are largely parallel and should have a very
1892          * low idle thresh
1893          */
1894         fs_info->endio_workers.idle_thresh = 4;
1895         fs_info->endio_meta_workers.idle_thresh = 4;
1896
1897         fs_info->endio_write_workers.idle_thresh = 2;
1898         fs_info->endio_meta_write_workers.idle_thresh = 2;
1899
1900         btrfs_start_workers(&fs_info->workers, 1);
1901         btrfs_start_workers(&fs_info->generic_worker, 1);
1902         btrfs_start_workers(&fs_info->submit_workers, 1);
1903         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1904         btrfs_start_workers(&fs_info->fixup_workers, 1);
1905         btrfs_start_workers(&fs_info->endio_workers, 1);
1906         btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1907         btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1908         btrfs_start_workers(&fs_info->endio_write_workers, 1);
1909         btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1910         btrfs_start_workers(&fs_info->delayed_workers, 1);
1911
1912         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1913         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1914                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1915
1916         nodesize = btrfs_super_nodesize(disk_super);
1917         leafsize = btrfs_super_leafsize(disk_super);
1918         sectorsize = btrfs_super_sectorsize(disk_super);
1919         stripesize = btrfs_super_stripesize(disk_super);
1920         tree_root->nodesize = nodesize;
1921         tree_root->leafsize = leafsize;
1922         tree_root->sectorsize = sectorsize;
1923         tree_root->stripesize = stripesize;
1924
1925         sb->s_blocksize = sectorsize;
1926         sb->s_blocksize_bits = blksize_bits(sectorsize);
1927
1928         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1929                     sizeof(disk_super->magic))) {
1930                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1931                 goto fail_sb_buffer;
1932         }
1933
1934         mutex_lock(&fs_info->chunk_mutex);
1935         ret = btrfs_read_sys_array(tree_root);
1936         mutex_unlock(&fs_info->chunk_mutex);
1937         if (ret) {
1938                 printk(KERN_WARNING "btrfs: failed to read the system "
1939                        "array on %s\n", sb->s_id);
1940                 goto fail_sb_buffer;
1941         }
1942
1943         blocksize = btrfs_level_size(tree_root,
1944                                      btrfs_super_chunk_root_level(disk_super));
1945         generation = btrfs_super_chunk_root_generation(disk_super);
1946
1947         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1948                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1949
1950         chunk_root->node = read_tree_block(chunk_root,
1951                                            btrfs_super_chunk_root(disk_super),
1952                                            blocksize, generation);
1953         BUG_ON(!chunk_root->node);
1954         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1955                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1956                        sb->s_id);
1957                 goto fail_chunk_root;
1958         }
1959         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1960         chunk_root->commit_root = btrfs_root_node(chunk_root);
1961
1962         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1963            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1964            BTRFS_UUID_SIZE);
1965
1966         mutex_lock(&fs_info->chunk_mutex);
1967         ret = btrfs_read_chunk_tree(chunk_root);
1968         mutex_unlock(&fs_info->chunk_mutex);
1969         if (ret) {
1970                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1971                        sb->s_id);
1972                 goto fail_chunk_root;
1973         }
1974
1975         btrfs_close_extra_devices(fs_devices);
1976
1977         blocksize = btrfs_level_size(tree_root,
1978                                      btrfs_super_root_level(disk_super));
1979         generation = btrfs_super_generation(disk_super);
1980
1981         tree_root->node = read_tree_block(tree_root,
1982                                           btrfs_super_root(disk_super),
1983                                           blocksize, generation);
1984         if (!tree_root->node)
1985                 goto fail_chunk_root;
1986         if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1987                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1988                        sb->s_id);
1989                 goto fail_tree_root;
1990         }
1991         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1992         tree_root->commit_root = btrfs_root_node(tree_root);
1993
1994         ret = find_and_setup_root(tree_root, fs_info,
1995                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1996         if (ret)
1997                 goto fail_tree_root;
1998         extent_root->track_dirty = 1;
1999
2000         ret = find_and_setup_root(tree_root, fs_info,
2001                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
2002         if (ret)
2003                 goto fail_extent_root;
2004         dev_root->track_dirty = 1;
2005
2006         ret = find_and_setup_root(tree_root, fs_info,
2007                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
2008         if (ret)
2009                 goto fail_dev_root;
2010
2011         csum_root->track_dirty = 1;
2012
2013         fs_info->generation = generation;
2014         fs_info->last_trans_committed = generation;
2015         fs_info->data_alloc_profile = (u64)-1;
2016         fs_info->metadata_alloc_profile = (u64)-1;
2017         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
2018
2019         ret = btrfs_init_space_info(fs_info);
2020         if (ret) {
2021                 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2022                 goto fail_block_groups;
2023         }
2024
2025         ret = btrfs_read_block_groups(extent_root);
2026         if (ret) {
2027                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2028                 goto fail_block_groups;
2029         }
2030
2031         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2032                                                "btrfs-cleaner");
2033         if (IS_ERR(fs_info->cleaner_kthread))
2034                 goto fail_block_groups;
2035
2036         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2037                                                    tree_root,
2038                                                    "btrfs-transaction");
2039         if (IS_ERR(fs_info->transaction_kthread))
2040                 goto fail_cleaner;
2041
2042         if (!btrfs_test_opt(tree_root, SSD) &&
2043             !btrfs_test_opt(tree_root, NOSSD) &&
2044             !fs_info->fs_devices->rotating) {
2045                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2046                        "mode\n");
2047                 btrfs_set_opt(fs_info->mount_opt, SSD);
2048         }
2049
2050         /* do not make disk changes in broken FS */
2051         if (btrfs_super_log_root(disk_super) != 0 &&
2052             !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2053                 u64 bytenr = btrfs_super_log_root(disk_super);
2054
2055                 if (fs_devices->rw_devices == 0) {
2056                         printk(KERN_WARNING "Btrfs log replay required "
2057                                "on RO media\n");
2058                         err = -EIO;
2059                         goto fail_trans_kthread;
2060                 }
2061                 blocksize =
2062                      btrfs_level_size(tree_root,
2063                                       btrfs_super_log_root_level(disk_super));
2064
2065                 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2066                 if (!log_tree_root) {
2067                         err = -ENOMEM;
2068                         goto fail_trans_kthread;
2069                 }
2070
2071                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2072                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2073
2074                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2075                                                       blocksize,
2076                                                       generation + 1);
2077                 ret = btrfs_recover_log_trees(log_tree_root);
2078                 BUG_ON(ret);
2079
2080                 if (sb->s_flags & MS_RDONLY) {
2081                         ret =  btrfs_commit_super(tree_root);
2082                         BUG_ON(ret);
2083                 }
2084         }
2085
2086         ret = btrfs_find_orphan_roots(tree_root);
2087         BUG_ON(ret);
2088
2089         if (!(sb->s_flags & MS_RDONLY)) {
2090                 ret = btrfs_cleanup_fs_roots(fs_info);
2091                 BUG_ON(ret);
2092
2093                 ret = btrfs_recover_relocation(tree_root);
2094                 if (ret < 0) {
2095                         printk(KERN_WARNING
2096                                "btrfs: failed to recover relocation\n");
2097                         err = -EINVAL;
2098                         goto fail_trans_kthread;
2099                 }
2100         }
2101
2102         location.objectid = BTRFS_FS_TREE_OBJECTID;
2103         location.type = BTRFS_ROOT_ITEM_KEY;
2104         location.offset = (u64)-1;
2105
2106         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2107         if (!fs_info->fs_root)
2108                 goto fail_trans_kthread;
2109         if (IS_ERR(fs_info->fs_root)) {
2110                 err = PTR_ERR(fs_info->fs_root);
2111                 goto fail_trans_kthread;
2112         }
2113
2114         if (!(sb->s_flags & MS_RDONLY)) {
2115                 down_read(&fs_info->cleanup_work_sem);
2116                 err = btrfs_orphan_cleanup(fs_info->fs_root);
2117                 if (!err)
2118                         err = btrfs_orphan_cleanup(fs_info->tree_root);
2119                 up_read(&fs_info->cleanup_work_sem);
2120                 if (err) {
2121                         close_ctree(tree_root);
2122                         return ERR_PTR(err);
2123                 }
2124         }
2125
2126         return tree_root;
2127
2128 fail_trans_kthread:
2129         kthread_stop(fs_info->transaction_kthread);
2130 fail_cleaner:
2131         kthread_stop(fs_info->cleaner_kthread);
2132
2133         /*
2134          * make sure we're done with the btree inode before we stop our
2135          * kthreads
2136          */
2137         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2138         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2139
2140 fail_block_groups:
2141         btrfs_free_block_groups(fs_info);
2142         free_extent_buffer(csum_root->node);
2143         free_extent_buffer(csum_root->commit_root);
2144 fail_dev_root:
2145         free_extent_buffer(dev_root->node);
2146         free_extent_buffer(dev_root->commit_root);
2147 fail_extent_root:
2148         free_extent_buffer(extent_root->node);
2149         free_extent_buffer(extent_root->commit_root);
2150 fail_tree_root:
2151         free_extent_buffer(tree_root->node);
2152         free_extent_buffer(tree_root->commit_root);
2153 fail_chunk_root:
2154         free_extent_buffer(chunk_root->node);
2155         free_extent_buffer(chunk_root->commit_root);
2156 fail_sb_buffer:
2157         btrfs_stop_workers(&fs_info->generic_worker);
2158         btrfs_stop_workers(&fs_info->fixup_workers);
2159         btrfs_stop_workers(&fs_info->delalloc_workers);
2160         btrfs_stop_workers(&fs_info->workers);
2161         btrfs_stop_workers(&fs_info->endio_workers);
2162         btrfs_stop_workers(&fs_info->endio_meta_workers);
2163         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2164         btrfs_stop_workers(&fs_info->endio_write_workers);
2165         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2166         btrfs_stop_workers(&fs_info->submit_workers);
2167         btrfs_stop_workers(&fs_info->delayed_workers);
2168 fail_alloc:
2169         kfree(fs_info->delayed_root);
2170 fail_iput:
2171         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2172         iput(fs_info->btree_inode);
2173
2174         btrfs_close_devices(fs_info->fs_devices);
2175         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2176 fail_bdi:
2177         bdi_destroy(&fs_info->bdi);
2178 fail_srcu:
2179         cleanup_srcu_struct(&fs_info->subvol_srcu);
2180 fail:
2181         kfree(extent_root);
2182         kfree(tree_root);
2183         kfree(fs_info);
2184         kfree(chunk_root);
2185         kfree(dev_root);
2186         kfree(csum_root);
2187         return ERR_PTR(err);
2188 }
2189
2190 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2191 {
2192         char b[BDEVNAME_SIZE];
2193
2194         if (uptodate) {
2195                 set_buffer_uptodate(bh);
2196         } else {
2197                 if (printk_ratelimit()) {
2198                         printk(KERN_WARNING "lost page write due to "
2199                                         "I/O error on %s\n",
2200                                        bdevname(bh->b_bdev, b));
2201                 }
2202                 /* note, we dont' set_buffer_write_io_error because we have
2203                  * our own ways of dealing with the IO errors
2204                  */
2205                 clear_buffer_uptodate(bh);
2206         }
2207         unlock_buffer(bh);
2208         put_bh(bh);
2209 }
2210
2211 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2212 {
2213         struct buffer_head *bh;
2214         struct buffer_head *latest = NULL;
2215         struct btrfs_super_block *super;
2216         int i;
2217         u64 transid = 0;
2218         u64 bytenr;
2219
2220         /* we would like to check all the supers, but that would make
2221          * a btrfs mount succeed after a mkfs from a different FS.
2222          * So, we need to add a special mount option to scan for
2223          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2224          */
2225         for (i = 0; i < 1; i++) {
2226                 bytenr = btrfs_sb_offset(i);
2227                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2228                         break;
2229                 bh = __bread(bdev, bytenr / 4096, 4096);
2230                 if (!bh)
2231                         continue;
2232
2233                 super = (struct btrfs_super_block *)bh->b_data;
2234                 if (btrfs_super_bytenr(super) != bytenr ||
2235                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2236                             sizeof(super->magic))) {
2237                         brelse(bh);
2238                         continue;
2239                 }
2240
2241                 if (!latest || btrfs_super_generation(super) > transid) {
2242                         brelse(latest);
2243                         latest = bh;
2244                         transid = btrfs_super_generation(super);
2245                 } else {
2246                         brelse(bh);
2247                 }
2248         }
2249         return latest;
2250 }
2251
2252 /*
2253  * this should be called twice, once with wait == 0 and
2254  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2255  * we write are pinned.
2256  *
2257  * They are released when wait == 1 is done.
2258  * max_mirrors must be the same for both runs, and it indicates how
2259  * many supers on this one device should be written.
2260  *
2261  * max_mirrors == 0 means to write them all.
2262  */
2263 static int write_dev_supers(struct btrfs_device *device,
2264                             struct btrfs_super_block *sb,
2265                             int do_barriers, int wait, int max_mirrors)
2266 {
2267         struct buffer_head *bh;
2268         int i;
2269         int ret;
2270         int errors = 0;
2271         u32 crc;
2272         u64 bytenr;
2273         int last_barrier = 0;
2274
2275         if (max_mirrors == 0)
2276                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2277
2278         /* make sure only the last submit_bh does a barrier */
2279         if (do_barriers) {
2280                 for (i = 0; i < max_mirrors; i++) {
2281                         bytenr = btrfs_sb_offset(i);
2282                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2283                             device->total_bytes)
2284                                 break;
2285                         last_barrier = i;
2286                 }
2287         }
2288
2289         for (i = 0; i < max_mirrors; i++) {
2290                 bytenr = btrfs_sb_offset(i);
2291                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2292                         break;
2293
2294                 if (wait) {
2295                         bh = __find_get_block(device->bdev, bytenr / 4096,
2296                                               BTRFS_SUPER_INFO_SIZE);
2297                         BUG_ON(!bh);
2298                         wait_on_buffer(bh);
2299                         if (!buffer_uptodate(bh))
2300                                 errors++;
2301
2302                         /* drop our reference */
2303                         brelse(bh);
2304
2305                         /* drop the reference from the wait == 0 run */
2306                         brelse(bh);
2307                         continue;
2308                 } else {
2309                         btrfs_set_super_bytenr(sb, bytenr);
2310
2311                         crc = ~(u32)0;
2312                         crc = btrfs_csum_data(NULL, (char *)sb +
2313                                               BTRFS_CSUM_SIZE, crc,
2314                                               BTRFS_SUPER_INFO_SIZE -
2315                                               BTRFS_CSUM_SIZE);
2316                         btrfs_csum_final(crc, sb->csum);
2317
2318                         /*
2319                          * one reference for us, and we leave it for the
2320                          * caller
2321                          */
2322                         bh = __getblk(device->bdev, bytenr / 4096,
2323                                       BTRFS_SUPER_INFO_SIZE);
2324                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2325
2326                         /* one reference for submit_bh */
2327                         get_bh(bh);
2328
2329                         set_buffer_uptodate(bh);
2330                         lock_buffer(bh);
2331                         bh->b_end_io = btrfs_end_buffer_write_sync;
2332                 }
2333
2334                 if (i == last_barrier && do_barriers)
2335                         ret = submit_bh(WRITE_FLUSH_FUA, bh);
2336                 else
2337                         ret = submit_bh(WRITE_SYNC, bh);
2338
2339                 if (ret)
2340                         errors++;
2341         }
2342         return errors < i ? 0 : -1;
2343 }
2344
2345 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2346 {
2347         struct list_head *head;
2348         struct btrfs_device *dev;
2349         struct btrfs_super_block *sb;
2350         struct btrfs_dev_item *dev_item;
2351         int ret;
2352         int do_barriers;
2353         int max_errors;
2354         int total_errors = 0;
2355         u64 flags;
2356
2357         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2358         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2359
2360         sb = &root->fs_info->super_for_commit;
2361         dev_item = &sb->dev_item;
2362
2363         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2364         head = &root->fs_info->fs_devices->devices;
2365         list_for_each_entry(dev, head, dev_list) {
2366                 if (!dev->bdev) {
2367                         total_errors++;
2368                         continue;
2369                 }
2370                 if (!dev->in_fs_metadata || !dev->writeable)
2371                         continue;
2372
2373                 btrfs_set_stack_device_generation(dev_item, 0);
2374                 btrfs_set_stack_device_type(dev_item, dev->type);
2375                 btrfs_set_stack_device_id(dev_item, dev->devid);
2376                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2377                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2378                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2379                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2380                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2381                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2382                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2383
2384                 flags = btrfs_super_flags(sb);
2385                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2386
2387                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2388                 if (ret)
2389                         total_errors++;
2390         }
2391         if (total_errors > max_errors) {
2392                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2393                        total_errors);
2394                 BUG();
2395         }
2396
2397         total_errors = 0;
2398         list_for_each_entry(dev, head, dev_list) {
2399                 if (!dev->bdev)
2400                         continue;
2401                 if (!dev->in_fs_metadata || !dev->writeable)
2402                         continue;
2403
2404                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2405                 if (ret)
2406                         total_errors++;
2407         }
2408         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2409         if (total_errors > max_errors) {
2410                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2411                        total_errors);
2412                 BUG();
2413         }
2414         return 0;
2415 }
2416
2417 int write_ctree_super(struct btrfs_trans_handle *trans,
2418                       struct btrfs_root *root, int max_mirrors)
2419 {
2420         int ret;
2421
2422         ret = write_all_supers(root, max_mirrors);
2423         return ret;
2424 }
2425
2426 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2427 {
2428         spin_lock(&fs_info->fs_roots_radix_lock);
2429         radix_tree_delete(&fs_info->fs_roots_radix,
2430                           (unsigned long)root->root_key.objectid);
2431         spin_unlock(&fs_info->fs_roots_radix_lock);
2432
2433         if (btrfs_root_refs(&root->root_item) == 0)
2434                 synchronize_srcu(&fs_info->subvol_srcu);
2435
2436         __btrfs_remove_free_space_cache(root->free_ino_pinned);
2437         __btrfs_remove_free_space_cache(root->free_ino_ctl);
2438         free_fs_root(root);
2439         return 0;
2440 }
2441
2442 static void free_fs_root(struct btrfs_root *root)
2443 {
2444         iput(root->cache_inode);
2445         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2446         if (root->anon_super.s_dev) {
2447                 down_write(&root->anon_super.s_umount);
2448                 kill_anon_super(&root->anon_super);
2449         }
2450         free_extent_buffer(root->node);
2451         free_extent_buffer(root->commit_root);
2452         kfree(root->free_ino_ctl);
2453         kfree(root->free_ino_pinned);
2454         kfree(root->name);
2455         kfree(root);
2456 }
2457
2458 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2459 {
2460         int ret;
2461         struct btrfs_root *gang[8];
2462         int i;
2463
2464         while (!list_empty(&fs_info->dead_roots)) {
2465                 gang[0] = list_entry(fs_info->dead_roots.next,
2466                                      struct btrfs_root, root_list);
2467                 list_del(&gang[0]->root_list);
2468
2469                 if (gang[0]->in_radix) {
2470                         btrfs_free_fs_root(fs_info, gang[0]);
2471                 } else {
2472                         free_extent_buffer(gang[0]->node);
2473                         free_extent_buffer(gang[0]->commit_root);
2474                         kfree(gang[0]);
2475                 }
2476         }
2477
2478         while (1) {
2479                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2480                                              (void **)gang, 0,
2481                                              ARRAY_SIZE(gang));
2482                 if (!ret)
2483                         break;
2484                 for (i = 0; i < ret; i++)
2485                         btrfs_free_fs_root(fs_info, gang[i]);
2486         }
2487         return 0;
2488 }
2489
2490 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2491 {
2492         u64 root_objectid = 0;
2493         struct btrfs_root *gang[8];
2494         int i;
2495         int ret;
2496
2497         while (1) {
2498                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2499                                              (void **)gang, root_objectid,
2500                                              ARRAY_SIZE(gang));
2501                 if (!ret)
2502                         break;
2503
2504                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2505                 for (i = 0; i < ret; i++) {
2506                         int err;
2507
2508                         root_objectid = gang[i]->root_key.objectid;
2509                         err = btrfs_orphan_cleanup(gang[i]);
2510                         if (err)
2511                                 return err;
2512                 }
2513                 root_objectid++;
2514         }
2515         return 0;
2516 }
2517
2518 int btrfs_commit_super(struct btrfs_root *root)
2519 {
2520         struct btrfs_trans_handle *trans;
2521         int ret;
2522
2523         mutex_lock(&root->fs_info->cleaner_mutex);
2524         btrfs_run_delayed_iputs(root);
2525         btrfs_clean_old_snapshots(root);
2526         mutex_unlock(&root->fs_info->cleaner_mutex);
2527
2528         /* wait until ongoing cleanup work done */
2529         down_write(&root->fs_info->cleanup_work_sem);
2530         up_write(&root->fs_info->cleanup_work_sem);
2531
2532         trans = btrfs_join_transaction(root, 1);
2533         if (IS_ERR(trans))
2534                 return PTR_ERR(trans);
2535         ret = btrfs_commit_transaction(trans, root);
2536         BUG_ON(ret);
2537         /* run commit again to drop the original snapshot */
2538         trans = btrfs_join_transaction(root, 1);
2539         if (IS_ERR(trans))
2540                 return PTR_ERR(trans);
2541         btrfs_commit_transaction(trans, root);
2542         ret = btrfs_write_and_wait_transaction(NULL, root);
2543         BUG_ON(ret);
2544
2545         ret = write_ctree_super(NULL, root, 0);
2546         return ret;
2547 }
2548
2549 int close_ctree(struct btrfs_root *root)
2550 {
2551         struct btrfs_fs_info *fs_info = root->fs_info;
2552         int ret;
2553
2554         fs_info->closing = 1;
2555         smp_mb();
2556
2557         btrfs_put_block_group_cache(fs_info);
2558
2559         /*
2560          * Here come 2 situations when btrfs is broken to flip readonly:
2561          *
2562          * 1. when btrfs flips readonly somewhere else before
2563          * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2564          * and btrfs will skip to write sb directly to keep
2565          * ERROR state on disk.
2566          *
2567          * 2. when btrfs flips readonly just in btrfs_commit_super,
2568          * and in such case, btrfs cannot write sb via btrfs_commit_super,
2569          * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2570          * btrfs will cleanup all FS resources first and write sb then.
2571          */
2572         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2573                 ret = btrfs_commit_super(root);
2574                 if (ret)
2575                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2576         }
2577
2578         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2579                 ret = btrfs_error_commit_super(root);
2580                 if (ret)
2581                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2582         }
2583
2584         kthread_stop(root->fs_info->transaction_kthread);
2585         kthread_stop(root->fs_info->cleaner_kthread);
2586
2587         fs_info->closing = 2;
2588         smp_mb();
2589
2590         if (fs_info->delalloc_bytes) {
2591                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2592                        (unsigned long long)fs_info->delalloc_bytes);
2593         }
2594         if (fs_info->total_ref_cache_size) {
2595                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2596                        (unsigned long long)fs_info->total_ref_cache_size);
2597         }
2598
2599         free_extent_buffer(fs_info->extent_root->node);
2600         free_extent_buffer(fs_info->extent_root->commit_root);
2601         free_extent_buffer(fs_info->tree_root->node);
2602         free_extent_buffer(fs_info->tree_root->commit_root);
2603         free_extent_buffer(root->fs_info->chunk_root->node);
2604         free_extent_buffer(root->fs_info->chunk_root->commit_root);
2605         free_extent_buffer(root->fs_info->dev_root->node);
2606         free_extent_buffer(root->fs_info->dev_root->commit_root);
2607         free_extent_buffer(root->fs_info->csum_root->node);
2608         free_extent_buffer(root->fs_info->csum_root->commit_root);
2609
2610         btrfs_free_block_groups(root->fs_info);
2611
2612         del_fs_roots(fs_info);
2613
2614         iput(fs_info->btree_inode);
2615         kfree(fs_info->delayed_root);
2616
2617         btrfs_stop_workers(&fs_info->generic_worker);
2618         btrfs_stop_workers(&fs_info->fixup_workers);
2619         btrfs_stop_workers(&fs_info->delalloc_workers);
2620         btrfs_stop_workers(&fs_info->workers);
2621         btrfs_stop_workers(&fs_info->endio_workers);
2622         btrfs_stop_workers(&fs_info->endio_meta_workers);
2623         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2624         btrfs_stop_workers(&fs_info->endio_write_workers);
2625         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2626         btrfs_stop_workers(&fs_info->submit_workers);
2627         btrfs_stop_workers(&fs_info->delayed_workers);
2628
2629         btrfs_close_devices(fs_info->fs_devices);
2630         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2631
2632         bdi_destroy(&fs_info->bdi);
2633         cleanup_srcu_struct(&fs_info->subvol_srcu);
2634
2635         kfree(fs_info->extent_root);
2636         kfree(fs_info->tree_root);
2637         kfree(fs_info->chunk_root);
2638         kfree(fs_info->dev_root);
2639         kfree(fs_info->csum_root);
2640         kfree(fs_info);
2641
2642         return 0;
2643 }
2644
2645 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2646 {
2647         int ret;
2648         struct inode *btree_inode = buf->first_page->mapping->host;
2649
2650         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2651                                      NULL);
2652         if (!ret)
2653                 return ret;
2654
2655         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2656                                     parent_transid);
2657         return !ret;
2658 }
2659
2660 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2661 {
2662         struct inode *btree_inode = buf->first_page->mapping->host;
2663         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2664                                           buf);
2665 }
2666
2667 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2668 {
2669         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2670         u64 transid = btrfs_header_generation(buf);
2671         struct inode *btree_inode = root->fs_info->btree_inode;
2672         int was_dirty;
2673
2674         btrfs_assert_tree_locked(buf);
2675         if (transid != root->fs_info->generation) {
2676                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2677                        "found %llu running %llu\n",
2678                         (unsigned long long)buf->start,
2679                         (unsigned long long)transid,
2680                         (unsigned long long)root->fs_info->generation);
2681                 WARN_ON(1);
2682         }
2683         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2684                                             buf);
2685         if (!was_dirty) {
2686                 spin_lock(&root->fs_info->delalloc_lock);
2687                 root->fs_info->dirty_metadata_bytes += buf->len;
2688                 spin_unlock(&root->fs_info->delalloc_lock);
2689         }
2690 }
2691
2692 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2693 {
2694         /*
2695          * looks as though older kernels can get into trouble with
2696          * this code, they end up stuck in balance_dirty_pages forever
2697          */
2698         u64 num_dirty;
2699         unsigned long thresh = 32 * 1024 * 1024;
2700
2701         if (current->flags & PF_MEMALLOC)
2702                 return;
2703
2704         btrfs_balance_delayed_items(root);
2705
2706         num_dirty = root->fs_info->dirty_metadata_bytes;
2707
2708         if (num_dirty > thresh) {
2709                 balance_dirty_pages_ratelimited_nr(
2710                                    root->fs_info->btree_inode->i_mapping, 1);
2711         }
2712         return;
2713 }
2714
2715 void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2716 {
2717         /*
2718          * looks as though older kernels can get into trouble with
2719          * this code, they end up stuck in balance_dirty_pages forever
2720          */
2721         u64 num_dirty;
2722         unsigned long thresh = 32 * 1024 * 1024;
2723
2724         if (current->flags & PF_MEMALLOC)
2725                 return;
2726
2727         num_dirty = root->fs_info->dirty_metadata_bytes;
2728
2729         if (num_dirty > thresh) {
2730                 balance_dirty_pages_ratelimited_nr(
2731                                    root->fs_info->btree_inode->i_mapping, 1);
2732         }
2733         return;
2734 }
2735
2736 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2737 {
2738         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2739         int ret;
2740         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2741         if (ret == 0)
2742                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2743         return ret;
2744 }
2745
2746 int btree_lock_page_hook(struct page *page)
2747 {
2748         struct inode *inode = page->mapping->host;
2749         struct btrfs_root *root = BTRFS_I(inode)->root;
2750         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2751         struct extent_buffer *eb;
2752         unsigned long len;
2753         u64 bytenr = page_offset(page);
2754
2755         if (page->private == EXTENT_PAGE_PRIVATE)
2756                 goto out;
2757
2758         len = page->private >> 2;
2759         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2760         if (!eb)
2761                 goto out;
2762
2763         btrfs_tree_lock(eb);
2764         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2765
2766         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2767                 spin_lock(&root->fs_info->delalloc_lock);
2768                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2769                         root->fs_info->dirty_metadata_bytes -= eb->len;
2770                 else
2771                         WARN_ON(1);
2772                 spin_unlock(&root->fs_info->delalloc_lock);
2773         }
2774
2775         btrfs_tree_unlock(eb);
2776         free_extent_buffer(eb);
2777 out:
2778         lock_page(page);
2779         return 0;
2780 }
2781
2782 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
2783                               int read_only)
2784 {
2785         if (read_only)
2786                 return;
2787
2788         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
2789                 printk(KERN_WARNING "warning: mount fs with errors, "
2790                        "running btrfsck is recommended\n");
2791 }
2792
2793 int btrfs_error_commit_super(struct btrfs_root *root)
2794 {
2795         int ret;
2796
2797         mutex_lock(&root->fs_info->cleaner_mutex);
2798         btrfs_run_delayed_iputs(root);
2799         mutex_unlock(&root->fs_info->cleaner_mutex);
2800
2801         down_write(&root->fs_info->cleanup_work_sem);
2802         up_write(&root->fs_info->cleanup_work_sem);
2803
2804         /* cleanup FS via transaction */
2805         btrfs_cleanup_transaction(root);
2806
2807         ret = write_ctree_super(NULL, root, 0);
2808
2809         return ret;
2810 }
2811
2812 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
2813 {
2814         struct btrfs_inode *btrfs_inode;
2815         struct list_head splice;
2816
2817         INIT_LIST_HEAD(&splice);
2818
2819         mutex_lock(&root->fs_info->ordered_operations_mutex);
2820         spin_lock(&root->fs_info->ordered_extent_lock);
2821
2822         list_splice_init(&root->fs_info->ordered_operations, &splice);
2823         while (!list_empty(&splice)) {
2824                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2825                                          ordered_operations);
2826
2827                 list_del_init(&btrfs_inode->ordered_operations);
2828
2829                 btrfs_invalidate_inodes(btrfs_inode->root);
2830         }
2831
2832         spin_unlock(&root->fs_info->ordered_extent_lock);
2833         mutex_unlock(&root->fs_info->ordered_operations_mutex);
2834
2835         return 0;
2836 }
2837
2838 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
2839 {
2840         struct list_head splice;
2841         struct btrfs_ordered_extent *ordered;
2842         struct inode *inode;
2843
2844         INIT_LIST_HEAD(&splice);
2845
2846         spin_lock(&root->fs_info->ordered_extent_lock);
2847
2848         list_splice_init(&root->fs_info->ordered_extents, &splice);
2849         while (!list_empty(&splice)) {
2850                 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
2851                                      root_extent_list);
2852
2853                 list_del_init(&ordered->root_extent_list);
2854                 atomic_inc(&ordered->refs);
2855
2856                 /* the inode may be getting freed (in sys_unlink path). */
2857                 inode = igrab(ordered->inode);
2858
2859                 spin_unlock(&root->fs_info->ordered_extent_lock);
2860                 if (inode)
2861                         iput(inode);
2862
2863                 atomic_set(&ordered->refs, 1);
2864                 btrfs_put_ordered_extent(ordered);
2865
2866                 spin_lock(&root->fs_info->ordered_extent_lock);
2867         }
2868
2869         spin_unlock(&root->fs_info->ordered_extent_lock);
2870
2871         return 0;
2872 }
2873
2874 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
2875                                       struct btrfs_root *root)
2876 {
2877         struct rb_node *node;
2878         struct btrfs_delayed_ref_root *delayed_refs;
2879         struct btrfs_delayed_ref_node *ref;
2880         int ret = 0;
2881
2882         delayed_refs = &trans->delayed_refs;
2883
2884         spin_lock(&delayed_refs->lock);
2885         if (delayed_refs->num_entries == 0) {
2886                 spin_unlock(&delayed_refs->lock);
2887                 printk(KERN_INFO "delayed_refs has NO entry\n");
2888                 return ret;
2889         }
2890
2891         node = rb_first(&delayed_refs->root);
2892         while (node) {
2893                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2894                 node = rb_next(node);
2895
2896                 ref->in_tree = 0;
2897                 rb_erase(&ref->rb_node, &delayed_refs->root);
2898                 delayed_refs->num_entries--;
2899
2900                 atomic_set(&ref->refs, 1);
2901                 if (btrfs_delayed_ref_is_head(ref)) {
2902                         struct btrfs_delayed_ref_head *head;
2903
2904                         head = btrfs_delayed_node_to_head(ref);
2905                         mutex_lock(&head->mutex);
2906                         kfree(head->extent_op);
2907                         delayed_refs->num_heads--;
2908                         if (list_empty(&head->cluster))
2909                                 delayed_refs->num_heads_ready--;
2910                         list_del_init(&head->cluster);
2911                         mutex_unlock(&head->mutex);
2912                 }
2913
2914                 spin_unlock(&delayed_refs->lock);
2915                 btrfs_put_delayed_ref(ref);
2916
2917                 cond_resched();
2918                 spin_lock(&delayed_refs->lock);
2919         }
2920
2921         spin_unlock(&delayed_refs->lock);
2922
2923         return ret;
2924 }
2925
2926 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
2927 {
2928         struct btrfs_pending_snapshot *snapshot;
2929         struct list_head splice;
2930
2931         INIT_LIST_HEAD(&splice);
2932
2933         list_splice_init(&t->pending_snapshots, &splice);
2934
2935         while (!list_empty(&splice)) {
2936                 snapshot = list_entry(splice.next,
2937                                       struct btrfs_pending_snapshot,
2938                                       list);
2939
2940                 list_del_init(&snapshot->list);
2941
2942                 kfree(snapshot);
2943         }
2944
2945         return 0;
2946 }
2947
2948 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
2949 {
2950         struct btrfs_inode *btrfs_inode;
2951         struct list_head splice;
2952
2953         INIT_LIST_HEAD(&splice);
2954
2955         list_splice_init(&root->fs_info->delalloc_inodes, &splice);
2956
2957         spin_lock(&root->fs_info->delalloc_lock);
2958
2959         while (!list_empty(&splice)) {
2960                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2961                                     delalloc_inodes);
2962
2963                 list_del_init(&btrfs_inode->delalloc_inodes);
2964
2965                 btrfs_invalidate_inodes(btrfs_inode->root);
2966         }
2967
2968         spin_unlock(&root->fs_info->delalloc_lock);
2969
2970         return 0;
2971 }
2972
2973 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
2974                                         struct extent_io_tree *dirty_pages,
2975                                         int mark)
2976 {
2977         int ret;
2978         struct page *page;
2979         struct inode *btree_inode = root->fs_info->btree_inode;
2980         struct extent_buffer *eb;
2981         u64 start = 0;
2982         u64 end;
2983         u64 offset;
2984         unsigned long index;
2985
2986         while (1) {
2987                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
2988                                             mark);
2989                 if (ret)
2990                         break;
2991
2992                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
2993                 while (start <= end) {
2994                         index = start >> PAGE_CACHE_SHIFT;
2995                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
2996                         page = find_get_page(btree_inode->i_mapping, index);
2997                         if (!page)
2998                                 continue;
2999                         offset = page_offset(page);
3000
3001                         spin_lock(&dirty_pages->buffer_lock);
3002                         eb = radix_tree_lookup(
3003                              &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3004                                                offset >> PAGE_CACHE_SHIFT);
3005                         spin_unlock(&dirty_pages->buffer_lock);
3006                         if (eb) {
3007                                 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3008                                                          &eb->bflags);
3009                                 atomic_set(&eb->refs, 1);
3010                         }
3011                         if (PageWriteback(page))
3012                                 end_page_writeback(page);
3013
3014                         lock_page(page);
3015                         if (PageDirty(page)) {
3016                                 clear_page_dirty_for_io(page);
3017                                 spin_lock_irq(&page->mapping->tree_lock);
3018                                 radix_tree_tag_clear(&page->mapping->page_tree,
3019                                                         page_index(page),
3020                                                         PAGECACHE_TAG_DIRTY);
3021                                 spin_unlock_irq(&page->mapping->tree_lock);
3022                         }
3023
3024                         page->mapping->a_ops->invalidatepage(page, 0);
3025                         unlock_page(page);
3026                 }
3027         }
3028
3029         return ret;
3030 }
3031
3032 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3033                                        struct extent_io_tree *pinned_extents)
3034 {
3035         struct extent_io_tree *unpin;
3036         u64 start;
3037         u64 end;
3038         int ret;
3039
3040         unpin = pinned_extents;
3041         while (1) {
3042                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3043                                             EXTENT_DIRTY);
3044                 if (ret)
3045                         break;
3046
3047                 /* opt_discard */
3048                 if (btrfs_test_opt(root, DISCARD))
3049                         ret = btrfs_error_discard_extent(root, start,
3050                                                          end + 1 - start,
3051                                                          NULL);
3052
3053                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3054                 btrfs_error_unpin_extent_range(root, start, end);
3055                 cond_resched();
3056         }
3057
3058         return 0;
3059 }
3060
3061 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3062 {
3063         struct btrfs_transaction *t;
3064         LIST_HEAD(list);
3065
3066         WARN_ON(1);
3067
3068         mutex_lock(&root->fs_info->trans_mutex);
3069         mutex_lock(&root->fs_info->transaction_kthread_mutex);
3070
3071         list_splice_init(&root->fs_info->trans_list, &list);
3072         while (!list_empty(&list)) {
3073                 t = list_entry(list.next, struct btrfs_transaction, list);
3074                 if (!t)
3075                         break;
3076
3077                 btrfs_destroy_ordered_operations(root);
3078
3079                 btrfs_destroy_ordered_extents(root);
3080
3081                 btrfs_destroy_delayed_refs(t, root);
3082
3083                 btrfs_block_rsv_release(root,
3084                                         &root->fs_info->trans_block_rsv,
3085                                         t->dirty_pages.dirty_bytes);
3086
3087                 /* FIXME: cleanup wait for commit */
3088                 t->in_commit = 1;
3089                 t->blocked = 1;
3090                 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3091                         wake_up(&root->fs_info->transaction_blocked_wait);
3092
3093                 t->blocked = 0;
3094                 if (waitqueue_active(&root->fs_info->transaction_wait))
3095                         wake_up(&root->fs_info->transaction_wait);
3096                 mutex_unlock(&root->fs_info->trans_mutex);
3097
3098                 mutex_lock(&root->fs_info->trans_mutex);
3099                 t->commit_done = 1;
3100                 if (waitqueue_active(&t->commit_wait))
3101                         wake_up(&t->commit_wait);
3102                 mutex_unlock(&root->fs_info->trans_mutex);
3103
3104                 mutex_lock(&root->fs_info->trans_mutex);
3105
3106                 btrfs_destroy_pending_snapshots(t);
3107
3108                 btrfs_destroy_delalloc_inodes(root);
3109
3110                 spin_lock(&root->fs_info->new_trans_lock);
3111                 root->fs_info->running_transaction = NULL;
3112                 spin_unlock(&root->fs_info->new_trans_lock);
3113
3114                 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3115                                              EXTENT_DIRTY);
3116
3117                 btrfs_destroy_pinned_extent(root,
3118                                             root->fs_info->pinned_extents);
3119
3120                 atomic_set(&t->use_count, 0);
3121                 list_del_init(&t->list);
3122                 memset(t, 0, sizeof(*t));
3123                 kmem_cache_free(btrfs_transaction_cachep, t);
3124         }
3125
3126         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3127         mutex_unlock(&root->fs_info->trans_mutex);
3128
3129         return 0;
3130 }
3131
3132 static struct extent_io_ops btree_extent_io_ops = {
3133         .write_cache_pages_lock_hook = btree_lock_page_hook,
3134         .readpage_end_io_hook = btree_readpage_end_io_hook,
3135         .submit_bio_hook = btree_submit_bio_hook,
3136         /* note we're sharing with inode.c for the merge bio hook */
3137         .merge_bio_hook = btrfs_merge_bio_hook,
3138 };