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         .sync_page      = block_sync_page,
 941 #ifdef CONFIG_MIGRATION
 942         .migratepage    = btree_migratepage,
 943 #endif
 944 };
 945
 946 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 947                          u64 parent_transid)
 948 {
 949         struct extent_buffer *buf = NULL;
 950         struct inode *btree_inode = root->fs_info->btree_inode;
 951         int ret = 0;
 952
 953         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 954         if (!buf)
 955                 return 0;
 956         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 957                                  buf, 0, 0, btree_get_extent, 0);
 958         free_extent_buffer(buf);
 959         return ret;
 960 }
 961
 962 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 963                                             u64 bytenr, u32 blocksize)
 964 {
 965         struct inode *btree_inode = root->fs_info->btree_inode;
 966         struct extent_buffer *eb;
 967         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 968                                 bytenr, blocksize, GFP_NOFS);
 969         return eb;
 970 }
 971
 972 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 973                                                  u64 bytenr, u32 blocksize)
 974 {
 975         struct inode *btree_inode = root->fs_info->btree_inode;
 976         struct extent_buffer *eb;
 977
 978         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 979                                  bytenr, blocksize, NULL, GFP_NOFS);
 980         return eb;
 981 }
 982
 983
 984 int btrfs_write_tree_block(struct extent_buffer *buf)
 985 {
 986         return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
 987                                         buf->start + buf->len - 1);
 988 }
 989
 990 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 991 {
 992         return filemap_fdatawait_range(buf->first_page->mapping,
 993                                        buf->start, buf->start + buf->len - 1);
 994 }
 995
 996 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 997                                       u32 blocksize, u64 parent_transid)
 998 {
 999         struct extent_buffer *buf = NULL;
1000         int ret;
1001
1002         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1003         if (!buf)
1004                 return NULL;
1005
1006         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1007
1008         if (ret == 0)
1009                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1010         return buf;
1011
1012 }
1013
1014 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1015                      struct extent_buffer *buf)
1016 {
1017         struct inode *btree_inode = root->fs_info->btree_inode;
1018         if (btrfs_header_generation(buf) ==
1019             root->fs_info->running_transaction->transid) {
1020                 btrfs_assert_tree_locked(buf);
1021
1022                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1023                         spin_lock(&root->fs_info->delalloc_lock);
1024                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
1025                                 root->fs_info->dirty_metadata_bytes -= buf->len;
1026                         else
1027                                 WARN_ON(1);
1028                         spin_unlock(&root->fs_info->delalloc_lock);
1029                 }
1030
1031                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1032                 btrfs_set_lock_blocking(buf);
1033                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1034                                           buf);
1035         }
1036         return 0;
1037 }
1038
1039 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1040                         u32 stripesize, struct btrfs_root *root,
1041                         struct btrfs_fs_info *fs_info,
1042                         u64 objectid)
1043 {
1044         root->node = NULL;
1045         root->commit_root = NULL;
1046         root->sectorsize = sectorsize;
1047         root->nodesize = nodesize;
1048         root->leafsize = leafsize;
1049         root->stripesize = stripesize;
1050         root->ref_cows = 0;
1051         root->track_dirty = 0;
1052         root->in_radix = 0;
1053         root->orphan_item_inserted = 0;
1054         root->orphan_cleanup_state = 0;
1055
1056         root->fs_info = fs_info;
1057         root->objectid = objectid;
1058         root->last_trans = 0;
1059         root->highest_objectid = 0;
1060         root->name = NULL;
1061         root->in_sysfs = 0;
1062         root->inode_tree = RB_ROOT;
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  * this unplugs every device on the box, and it is only used when page
1443  * is null
1444  */
1445 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1446 {
1447         struct btrfs_device *device;
1448         struct btrfs_fs_info *info;
1449
1450         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1451         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1452                 if (!device->bdev)
1453                         continue;
1454
1455                 bdi = blk_get_backing_dev_info(device->bdev);
1456                 if (bdi->unplug_io_fn)
1457                         bdi->unplug_io_fn(bdi, page);
1458         }
1459 }
1460
1461 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1462 {
1463         struct inode *inode;
1464         struct extent_map_tree *em_tree;
1465         struct extent_map *em;
1466         struct address_space *mapping;
1467         u64 offset;
1468
1469         /* the generic O_DIRECT read code does this */
1470         if (1 || !page) {
1471                 __unplug_io_fn(bdi, page);
1472                 return;
1473         }
1474
1475         /*
1476          * page->mapping may change at any time.  Get a consistent copy
1477          * and use that for everything below
1478          */
1479         smp_mb();
1480         mapping = page->mapping;
1481         if (!mapping)
1482                 return;
1483
1484         inode = mapping->host;
1485
1486         /*
1487          * don't do the expensive searching for a small number of
1488          * devices
1489          */
1490         if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1491                 __unplug_io_fn(bdi, page);
1492                 return;
1493         }
1494
1495         offset = page_offset(page);
1496
1497         em_tree = &BTRFS_I(inode)->extent_tree;
1498         read_lock(&em_tree->lock);
1499         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1500         read_unlock(&em_tree->lock);
1501         if (!em) {
1502                 __unplug_io_fn(bdi, page);
1503                 return;
1504         }
1505
1506         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1507                 free_extent_map(em);
1508                 __unplug_io_fn(bdi, page);
1509                 return;
1510         }
1511         offset = offset - em->start;
1512         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1513                           em->block_start + offset, page);
1514         free_extent_map(em);
1515 }
1516
1517 /*
1518  * If this fails, caller must call bdi_destroy() to get rid of the
1519  * bdi again.
1520  */
1521 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1522 {
1523         int err;
1524
1525         bdi->capabilities = BDI_CAP_MAP_COPY;
1526         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1527         if (err)
1528                 return err;
1529
1530         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1531         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1532         bdi->unplug_io_data     = info;
1533         bdi->congested_fn       = btrfs_congested_fn;
1534         bdi->congested_data     = info;
1535         return 0;
1536 }
1537
1538 static int bio_ready_for_csum(struct bio *bio)
1539 {
1540         u64 length = 0;
1541         u64 buf_len = 0;
1542         u64 start = 0;
1543         struct page *page;
1544         struct extent_io_tree *io_tree = NULL;
1545         struct bio_vec *bvec;
1546         int i;
1547         int ret;
1548
1549         bio_for_each_segment(bvec, bio, i) {
1550                 page = bvec->bv_page;
1551                 if (page->private == EXTENT_PAGE_PRIVATE) {
1552                         length += bvec->bv_len;
1553                         continue;
1554                 }
1555                 if (!page->private) {
1556                         length += bvec->bv_len;
1557                         continue;
1558                 }
1559                 length = bvec->bv_len;
1560                 buf_len = page->private >> 2;
1561                 start = page_offset(page) + bvec->bv_offset;
1562                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1563         }
1564         /* are we fully contained in this bio? */
1565         if (buf_len <= length)
1566                 return 1;
1567
1568         ret = extent_range_uptodate(io_tree, start + length,
1569                                     start + buf_len - 1);
1570         return ret;
1571 }
1572
1573 /*
1574  * called by the kthread helper functions to finally call the bio end_io
1575  * functions.  This is where read checksum verification actually happens
1576  */
1577 static void end_workqueue_fn(struct btrfs_work *work)
1578 {
1579         struct bio *bio;
1580         struct end_io_wq *end_io_wq;
1581         struct btrfs_fs_info *fs_info;
1582         int error;
1583
1584         end_io_wq = container_of(work, struct end_io_wq, work);
1585         bio = end_io_wq->bio;
1586         fs_info = end_io_wq->info;
1587
1588         /* metadata bio reads are special because the whole tree block must
1589          * be checksummed at once.  This makes sure the entire block is in
1590          * ram and up to date before trying to verify things.  For
1591          * blocksize <= pagesize, it is basically a noop
1592          */
1593         if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1594             !bio_ready_for_csum(bio)) {
1595                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1596                                    &end_io_wq->work);
1597                 return;
1598         }
1599         error = end_io_wq->error;
1600         bio->bi_private = end_io_wq->private;
1601         bio->bi_end_io = end_io_wq->end_io;
1602         kfree(end_io_wq);
1603         bio_endio(bio, error);
1604 }
1605
1606 static int cleaner_kthread(void *arg)
1607 {
1608         struct btrfs_root *root = arg;
1609
1610         do {
1611                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1612
1613                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1614                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1615                         btrfs_run_delayed_iputs(root);
1616                         btrfs_clean_old_snapshots(root);
1617                         mutex_unlock(&root->fs_info->cleaner_mutex);
1618                 }
1619
1620                 if (freezing(current)) {
1621                         refrigerator();
1622                 } else {
1623                         set_current_state(TASK_INTERRUPTIBLE);
1624                         if (!kthread_should_stop())
1625                                 schedule();
1626                         __set_current_state(TASK_RUNNING);
1627                 }
1628         } while (!kthread_should_stop());
1629         return 0;
1630 }
1631
1632 static int transaction_kthread(void *arg)
1633 {
1634         struct btrfs_root *root = arg;
1635         struct btrfs_trans_handle *trans;
1636         struct btrfs_transaction *cur;
1637         u64 transid;
1638         unsigned long now;
1639         unsigned long delay;
1640         int ret;
1641
1642         do {
1643                 delay = HZ * 30;
1644                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1645                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1646
1647                 spin_lock(&root->fs_info->new_trans_lock);
1648                 cur = root->fs_info->running_transaction;
1649                 if (!cur) {
1650                         spin_unlock(&root->fs_info->new_trans_lock);
1651                         goto sleep;
1652                 }
1653
1654                 now = get_seconds();
1655                 if (!cur->blocked &&
1656                     (now < cur->start_time || now - cur->start_time < 30)) {
1657                         spin_unlock(&root->fs_info->new_trans_lock);
1658                         delay = HZ * 5;
1659                         goto sleep;
1660                 }
1661                 transid = cur->transid;
1662                 spin_unlock(&root->fs_info->new_trans_lock);
1663
1664                 trans = btrfs_join_transaction(root, 1);
1665                 BUG_ON(IS_ERR(trans));
1666                 if (transid == trans->transid) {
1667                         ret = btrfs_commit_transaction(trans, root);
1668                         BUG_ON(ret);
1669                 } else {
1670                         btrfs_end_transaction(trans, root);
1671                 }
1672 sleep:
1673                 wake_up_process(root->fs_info->cleaner_kthread);
1674                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1675
1676                 if (freezing(current)) {
1677                         refrigerator();
1678                 } else {
1679                         set_current_state(TASK_INTERRUPTIBLE);
1680                         if (!kthread_should_stop() &&
1681                             !btrfs_transaction_blocked(root->fs_info))
1682                                 schedule_timeout(delay);
1683                         __set_current_state(TASK_RUNNING);
1684                 }
1685         } while (!kthread_should_stop());
1686         return 0;
1687 }
1688
1689 struct btrfs_root *open_ctree(struct super_block *sb,
1690                               struct btrfs_fs_devices *fs_devices,
1691                               char *options)
1692 {
1693         u32 sectorsize;
1694         u32 nodesize;
1695         u32 leafsize;
1696         u32 blocksize;
1697         u32 stripesize;
1698         u64 generation;
1699         u64 features;
1700         struct btrfs_key location;
1701         struct buffer_head *bh;
1702         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1703                                                  GFP_NOFS);
1704         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1705                                                  GFP_NOFS);
1706         struct btrfs_root *tree_root = btrfs_sb(sb);
1707         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1708         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1709                                                 GFP_NOFS);
1710         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1711                                               GFP_NOFS);
1712         struct btrfs_root *log_tree_root;
1713
1714         int ret;
1715         int err = -EINVAL;
1716
1717         struct btrfs_super_block *disk_super;
1718
1719         if (!extent_root || !tree_root || !fs_info ||
1720             !chunk_root || !dev_root || !csum_root) {
1721                 err = -ENOMEM;
1722                 goto fail;
1723         }
1724
1725         ret = init_srcu_struct(&fs_info->subvol_srcu);
1726         if (ret) {
1727                 err = ret;
1728                 goto fail;
1729         }
1730
1731         ret = setup_bdi(fs_info, &fs_info->bdi);
1732         if (ret) {
1733                 err = ret;
1734                 goto fail_srcu;
1735         }
1736
1737         fs_info->btree_inode = new_inode(sb);
1738         if (!fs_info->btree_inode) {
1739                 err = -ENOMEM;
1740                 goto fail_bdi;
1741         }
1742
1743         fs_info->btree_inode->i_mapping->flags &= ~__GFP_FS;
1744
1745         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1746         INIT_LIST_HEAD(&fs_info->trans_list);
1747         INIT_LIST_HEAD(&fs_info->dead_roots);
1748         INIT_LIST_HEAD(&fs_info->delayed_iputs);
1749         INIT_LIST_HEAD(&fs_info->hashers);
1750         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1751         INIT_LIST_HEAD(&fs_info->ordered_operations);
1752         INIT_LIST_HEAD(&fs_info->caching_block_groups);
1753         spin_lock_init(&fs_info->delalloc_lock);
1754         spin_lock_init(&fs_info->new_trans_lock);
1755         spin_lock_init(&fs_info->ref_cache_lock);
1756         spin_lock_init(&fs_info->fs_roots_radix_lock);
1757         spin_lock_init(&fs_info->delayed_iput_lock);
1758
1759         init_completion(&fs_info->kobj_unregister);
1760         fs_info->tree_root = tree_root;
1761         fs_info->extent_root = extent_root;
1762         fs_info->csum_root = csum_root;
1763         fs_info->chunk_root = chunk_root;
1764         fs_info->dev_root = dev_root;
1765         fs_info->fs_devices = fs_devices;
1766         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1767         INIT_LIST_HEAD(&fs_info->space_info);
1768         btrfs_mapping_init(&fs_info->mapping_tree);
1769         btrfs_init_block_rsv(&fs_info->global_block_rsv);
1770         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1771         btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1772         btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1773         btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1774         INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1775         mutex_init(&fs_info->durable_block_rsv_mutex);
1776         atomic_set(&fs_info->nr_async_submits, 0);
1777         atomic_set(&fs_info->async_delalloc_pages, 0);
1778         atomic_set(&fs_info->async_submit_draining, 0);
1779         atomic_set(&fs_info->nr_async_bios, 0);
1780         fs_info->sb = sb;
1781         fs_info->max_inline = 8192 * 1024;
1782         fs_info->metadata_ratio = 0;
1783
1784         fs_info->thread_pool_size = min_t(unsigned long,
1785                                           num_online_cpus() + 2, 8);
1786
1787         INIT_LIST_HEAD(&fs_info->ordered_extents);
1788         spin_lock_init(&fs_info->ordered_extent_lock);
1789
1790         sb->s_blocksize = 4096;
1791         sb->s_blocksize_bits = blksize_bits(4096);
1792         sb->s_bdi = &fs_info->bdi;
1793
1794         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1795         fs_info->btree_inode->i_nlink = 1;
1796         /*
1797          * we set the i_size on the btree inode to the max possible int.
1798          * the real end of the address space is determined by all of
1799          * the devices in the system
1800          */
1801         fs_info->btree_inode->i_size = OFFSET_MAX;
1802         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1803         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1804
1805         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1806         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1807                              fs_info->btree_inode->i_mapping,
1808                              GFP_NOFS);
1809         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1810                              GFP_NOFS);
1811
1812         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1813
1814         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1815         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1816                sizeof(struct btrfs_key));
1817         BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1818         insert_inode_hash(fs_info->btree_inode);
1819
1820         spin_lock_init(&fs_info->block_group_cache_lock);
1821         fs_info->block_group_cache_tree = RB_ROOT;
1822
1823         extent_io_tree_init(&fs_info->freed_extents[0],
1824                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1825         extent_io_tree_init(&fs_info->freed_extents[1],
1826                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1827         fs_info->pinned_extents = &fs_info->freed_extents[0];
1828         fs_info->do_barriers = 1;
1829
1830
1831         mutex_init(&fs_info->trans_mutex);
1832         mutex_init(&fs_info->ordered_operations_mutex);
1833         mutex_init(&fs_info->tree_log_mutex);
1834         mutex_init(&fs_info->chunk_mutex);
1835         mutex_init(&fs_info->transaction_kthread_mutex);
1836         mutex_init(&fs_info->cleaner_mutex);
1837         mutex_init(&fs_info->volume_mutex);
1838         init_rwsem(&fs_info->extent_commit_sem);
1839         init_rwsem(&fs_info->cleanup_work_sem);
1840         init_rwsem(&fs_info->subvol_sem);
1841
1842         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1843         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1844
1845         init_waitqueue_head(&fs_info->transaction_throttle);
1846         init_waitqueue_head(&fs_info->transaction_wait);
1847         init_waitqueue_head(&fs_info->transaction_blocked_wait);
1848         init_waitqueue_head(&fs_info->async_submit_wait);
1849
1850         __setup_root(4096, 4096, 4096, 4096, tree_root,
1851                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1852
1853         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1854         if (!bh) {
1855                 err = -EINVAL;
1856                 goto fail_iput;
1857         }
1858
1859         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1860         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1861                sizeof(fs_info->super_for_commit));
1862         brelse(bh);
1863
1864         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1865
1866         disk_super = &fs_info->super_copy;
1867         if (!btrfs_super_root(disk_super))
1868                 goto fail_iput;
1869
1870         /* check FS state, whether FS is broken. */
1871         fs_info->fs_state |= btrfs_super_flags(disk_super);
1872
1873         btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
1874
1875         /*
1876          * In the long term, we'll store the compression type in the super
1877          * block, and it'll be used for per file compression control.
1878          */
1879         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
1880
1881         ret = btrfs_parse_options(tree_root, options);
1882         if (ret) {
1883                 err = ret;
1884                 goto fail_iput;
1885         }
1886
1887         features = btrfs_super_incompat_flags(disk_super) &
1888                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1889         if (features) {
1890                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1891                        "unsupported optional features (%Lx).\n",
1892                        (unsigned long long)features);
1893                 err = -EINVAL;
1894                 goto fail_iput;
1895         }
1896
1897         features = btrfs_super_incompat_flags(disk_super);
1898         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1899         if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1900                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1901         btrfs_set_super_incompat_flags(disk_super, features);
1902
1903         features = btrfs_super_compat_ro_flags(disk_super) &
1904                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1905         if (!(sb->s_flags & MS_RDONLY) && features) {
1906                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1907                        "unsupported option features (%Lx).\n",
1908                        (unsigned long long)features);
1909                 err = -EINVAL;
1910                 goto fail_iput;
1911         }
1912
1913         btrfs_init_workers(&fs_info->generic_worker,
1914                            "genwork", 1, NULL);
1915
1916         btrfs_init_workers(&fs_info->workers, "worker",
1917                            fs_info->thread_pool_size,
1918                            &fs_info->generic_worker);
1919
1920         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1921                            fs_info->thread_pool_size,
1922                            &fs_info->generic_worker);
1923
1924         btrfs_init_workers(&fs_info->submit_workers, "submit",
1925                            min_t(u64, fs_devices->num_devices,
1926                            fs_info->thread_pool_size),
1927                            &fs_info->generic_worker);
1928
1929         /* a higher idle thresh on the submit workers makes it much more
1930          * likely that bios will be send down in a sane order to the
1931          * devices
1932          */
1933         fs_info->submit_workers.idle_thresh = 64;
1934
1935         fs_info->workers.idle_thresh = 16;
1936         fs_info->workers.ordered = 1;
1937
1938         fs_info->delalloc_workers.idle_thresh = 2;
1939         fs_info->delalloc_workers.ordered = 1;
1940
1941         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1942                            &fs_info->generic_worker);
1943         btrfs_init_workers(&fs_info->endio_workers, "endio",
1944                            fs_info->thread_pool_size,
1945                            &fs_info->generic_worker);
1946         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1947                            fs_info->thread_pool_size,
1948                            &fs_info->generic_worker);
1949         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1950                            "endio-meta-write", fs_info->thread_pool_size,
1951                            &fs_info->generic_worker);
1952         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1953                            fs_info->thread_pool_size,
1954                            &fs_info->generic_worker);
1955         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1956                            1, &fs_info->generic_worker);
1957
1958         /*
1959          * endios are largely parallel and should have a very
1960          * low idle thresh
1961          */
1962         fs_info->endio_workers.idle_thresh = 4;
1963         fs_info->endio_meta_workers.idle_thresh = 4;
1964
1965         fs_info->endio_write_workers.idle_thresh = 2;
1966         fs_info->endio_meta_write_workers.idle_thresh = 2;
1967
1968         btrfs_start_workers(&fs_info->workers, 1);
1969         btrfs_start_workers(&fs_info->generic_worker, 1);
1970         btrfs_start_workers(&fs_info->submit_workers, 1);
1971         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1972         btrfs_start_workers(&fs_info->fixup_workers, 1);
1973         btrfs_start_workers(&fs_info->endio_workers, 1);
1974         btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1975         btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1976         btrfs_start_workers(&fs_info->endio_write_workers, 1);
1977         btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1978
1979         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1980         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1981                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1982
1983         nodesize = btrfs_super_nodesize(disk_super);
1984         leafsize = btrfs_super_leafsize(disk_super);
1985         sectorsize = btrfs_super_sectorsize(disk_super);
1986         stripesize = btrfs_super_stripesize(disk_super);
1987         tree_root->nodesize = nodesize;
1988         tree_root->leafsize = leafsize;
1989         tree_root->sectorsize = sectorsize;
1990         tree_root->stripesize = stripesize;
1991
1992         sb->s_blocksize = sectorsize;
1993         sb->s_blocksize_bits = blksize_bits(sectorsize);
1994
1995         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1996                     sizeof(disk_super->magic))) {
1997                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1998                 goto fail_sb_buffer;
1999         }
2000
2001         mutex_lock(&fs_info->chunk_mutex);
2002         ret = btrfs_read_sys_array(tree_root);
2003         mutex_unlock(&fs_info->chunk_mutex);
2004         if (ret) {
2005                 printk(KERN_WARNING "btrfs: failed to read the system "
2006                        "array on %s\n", sb->s_id);
2007                 goto fail_sb_buffer;
2008         }
2009
2010         blocksize = btrfs_level_size(tree_root,
2011                                      btrfs_super_chunk_root_level(disk_super));
2012         generation = btrfs_super_chunk_root_generation(disk_super);
2013
2014         __setup_root(nodesize, leafsize, sectorsize, stripesize,
2015                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2016
2017         chunk_root->node = read_tree_block(chunk_root,
2018                                            btrfs_super_chunk_root(disk_super),
2019                                            blocksize, generation);
2020         BUG_ON(!chunk_root->node);
2021         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2022                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2023                        sb->s_id);
2024                 goto fail_chunk_root;
2025         }
2026         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2027         chunk_root->commit_root = btrfs_root_node(chunk_root);
2028
2029         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2030            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2031            BTRFS_UUID_SIZE);
2032
2033         mutex_lock(&fs_info->chunk_mutex);
2034         ret = btrfs_read_chunk_tree(chunk_root);
2035         mutex_unlock(&fs_info->chunk_mutex);
2036         if (ret) {
2037                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2038                        sb->s_id);
2039                 goto fail_chunk_root;
2040         }
2041
2042         btrfs_close_extra_devices(fs_devices);
2043
2044         blocksize = btrfs_level_size(tree_root,
2045                                      btrfs_super_root_level(disk_super));
2046         generation = btrfs_super_generation(disk_super);
2047
2048         tree_root->node = read_tree_block(tree_root,
2049                                           btrfs_super_root(disk_super),
2050                                           blocksize, generation);
2051         if (!tree_root->node)
2052                 goto fail_chunk_root;
2053         if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2054                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2055                        sb->s_id);
2056                 goto fail_tree_root;
2057         }
2058         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2059         tree_root->commit_root = btrfs_root_node(tree_root);
2060
2061         ret = find_and_setup_root(tree_root, fs_info,
2062                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2063         if (ret)
2064                 goto fail_tree_root;
2065         extent_root->track_dirty = 1;
2066
2067         ret = find_and_setup_root(tree_root, fs_info,
2068                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
2069         if (ret)
2070                 goto fail_extent_root;
2071         dev_root->track_dirty = 1;
2072
2073         ret = find_and_setup_root(tree_root, fs_info,
2074                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
2075         if (ret)
2076                 goto fail_dev_root;
2077
2078         csum_root->track_dirty = 1;
2079
2080         fs_info->generation = generation;
2081         fs_info->last_trans_committed = generation;
2082         fs_info->data_alloc_profile = (u64)-1;
2083         fs_info->metadata_alloc_profile = (u64)-1;
2084         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
2085
2086         ret = btrfs_init_space_info(fs_info);
2087         if (ret) {
2088                 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2089                 goto fail_block_groups;
2090         }
2091
2092         ret = btrfs_read_block_groups(extent_root);
2093         if (ret) {
2094                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2095                 goto fail_block_groups;
2096         }
2097
2098         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2099                                                "btrfs-cleaner");
2100         if (IS_ERR(fs_info->cleaner_kthread))
2101                 goto fail_block_groups;
2102
2103         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2104                                                    tree_root,
2105                                                    "btrfs-transaction");
2106         if (IS_ERR(fs_info->transaction_kthread))
2107                 goto fail_cleaner;
2108
2109         if (!btrfs_test_opt(tree_root, SSD) &&
2110             !btrfs_test_opt(tree_root, NOSSD) &&
2111             !fs_info->fs_devices->rotating) {
2112                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2113                        "mode\n");
2114                 btrfs_set_opt(fs_info->mount_opt, SSD);
2115         }
2116
2117         /* do not make disk changes in broken FS */
2118         if (btrfs_super_log_root(disk_super) != 0 &&
2119             !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2120                 u64 bytenr = btrfs_super_log_root(disk_super);
2121
2122                 if (fs_devices->rw_devices == 0) {
2123                         printk(KERN_WARNING "Btrfs log replay required "
2124                                "on RO media\n");
2125                         err = -EIO;
2126                         goto fail_trans_kthread;
2127                 }
2128                 blocksize =
2129                      btrfs_level_size(tree_root,
2130                                       btrfs_super_log_root_level(disk_super));
2131
2132                 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2133                 if (!log_tree_root) {
2134                         err = -ENOMEM;
2135                         goto fail_trans_kthread;
2136                 }
2137
2138                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2139                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2140
2141                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2142                                                       blocksize,
2143                                                       generation + 1);
2144                 ret = btrfs_recover_log_trees(log_tree_root);
2145                 BUG_ON(ret);
2146
2147                 if (sb->s_flags & MS_RDONLY) {
2148                         ret =  btrfs_commit_super(tree_root);
2149                         BUG_ON(ret);
2150                 }
2151         }
2152
2153         ret = btrfs_find_orphan_roots(tree_root);
2154         BUG_ON(ret);
2155
2156         if (!(sb->s_flags & MS_RDONLY)) {
2157                 ret = btrfs_cleanup_fs_roots(fs_info);
2158                 BUG_ON(ret);
2159
2160                 ret = btrfs_recover_relocation(tree_root);
2161                 if (ret < 0) {
2162                         printk(KERN_WARNING
2163                                "btrfs: failed to recover relocation\n");
2164                         err = -EINVAL;
2165                         goto fail_trans_kthread;
2166                 }
2167         }
2168
2169         location.objectid = BTRFS_FS_TREE_OBJECTID;
2170         location.type = BTRFS_ROOT_ITEM_KEY;
2171         location.offset = (u64)-1;
2172
2173         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2174         if (!fs_info->fs_root)
2175                 goto fail_trans_kthread;
2176         if (IS_ERR(fs_info->fs_root)) {
2177                 err = PTR_ERR(fs_info->fs_root);
2178                 goto fail_trans_kthread;
2179         }
2180
2181         if (!(sb->s_flags & MS_RDONLY)) {
2182                 down_read(&fs_info->cleanup_work_sem);
2183                 err = btrfs_orphan_cleanup(fs_info->fs_root);
2184                 if (!err)
2185                         err = btrfs_orphan_cleanup(fs_info->tree_root);
2186                 up_read(&fs_info->cleanup_work_sem);
2187                 if (err) {
2188                         close_ctree(tree_root);
2189                         return ERR_PTR(err);
2190                 }
2191         }
2192
2193         return tree_root;
2194
2195 fail_trans_kthread:
2196         kthread_stop(fs_info->transaction_kthread);
2197 fail_cleaner:
2198         kthread_stop(fs_info->cleaner_kthread);
2199
2200         /*
2201          * make sure we're done with the btree inode before we stop our
2202          * kthreads
2203          */
2204         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2205         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2206
2207 fail_block_groups:
2208         btrfs_free_block_groups(fs_info);
2209         free_extent_buffer(csum_root->node);
2210         free_extent_buffer(csum_root->commit_root);
2211 fail_dev_root:
2212         free_extent_buffer(dev_root->node);
2213         free_extent_buffer(dev_root->commit_root);
2214 fail_extent_root:
2215         free_extent_buffer(extent_root->node);
2216         free_extent_buffer(extent_root->commit_root);
2217 fail_tree_root:
2218         free_extent_buffer(tree_root->node);
2219         free_extent_buffer(tree_root->commit_root);
2220 fail_chunk_root:
2221         free_extent_buffer(chunk_root->node);
2222         free_extent_buffer(chunk_root->commit_root);
2223 fail_sb_buffer:
2224         btrfs_stop_workers(&fs_info->generic_worker);
2225         btrfs_stop_workers(&fs_info->fixup_workers);
2226         btrfs_stop_workers(&fs_info->delalloc_workers);
2227         btrfs_stop_workers(&fs_info->workers);
2228         btrfs_stop_workers(&fs_info->endio_workers);
2229         btrfs_stop_workers(&fs_info->endio_meta_workers);
2230         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2231         btrfs_stop_workers(&fs_info->endio_write_workers);
2232         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2233         btrfs_stop_workers(&fs_info->submit_workers);
2234 fail_iput:
2235         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2236         iput(fs_info->btree_inode);
2237
2238         btrfs_close_devices(fs_info->fs_devices);
2239         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2240 fail_bdi:
2241         bdi_destroy(&fs_info->bdi);
2242 fail_srcu:
2243         cleanup_srcu_struct(&fs_info->subvol_srcu);
2244 fail:
2245         kfree(extent_root);
2246         kfree(tree_root);
2247         kfree(fs_info);
2248         kfree(chunk_root);
2249         kfree(dev_root);
2250         kfree(csum_root);
2251         return ERR_PTR(err);
2252 }
2253
2254 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2255 {
2256         char b[BDEVNAME_SIZE];
2257
2258         if (uptodate) {
2259                 set_buffer_uptodate(bh);
2260         } else {
2261                 if (printk_ratelimit()) {
2262                         printk(KERN_WARNING "lost page write due to "
2263                                         "I/O error on %s\n",
2264                                        bdevname(bh->b_bdev, b));
2265                 }
2266                 /* note, we dont' set_buffer_write_io_error because we have
2267                  * our own ways of dealing with the IO errors
2268                  */
2269                 clear_buffer_uptodate(bh);
2270         }
2271         unlock_buffer(bh);
2272         put_bh(bh);
2273 }
2274
2275 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2276 {
2277         struct buffer_head *bh;
2278         struct buffer_head *latest = NULL;
2279         struct btrfs_super_block *super;
2280         int i;
2281         u64 transid = 0;
2282         u64 bytenr;
2283
2284         /* we would like to check all the supers, but that would make
2285          * a btrfs mount succeed after a mkfs from a different FS.
2286          * So, we need to add a special mount option to scan for
2287          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2288          */
2289         for (i = 0; i < 1; i++) {
2290                 bytenr = btrfs_sb_offset(i);
2291                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2292                         break;
2293                 bh = __bread(bdev, bytenr / 4096, 4096);
2294                 if (!bh)
2295                         continue;
2296
2297                 super = (struct btrfs_super_block *)bh->b_data;
2298                 if (btrfs_super_bytenr(super) != bytenr ||
2299                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2300                             sizeof(super->magic))) {
2301                         brelse(bh);
2302                         continue;
2303                 }
2304
2305                 if (!latest || btrfs_super_generation(super) > transid) {
2306                         brelse(latest);
2307                         latest = bh;
2308                         transid = btrfs_super_generation(super);
2309                 } else {
2310                         brelse(bh);
2311                 }
2312         }
2313         return latest;
2314 }
2315
2316 /*
2317  * this should be called twice, once with wait == 0 and
2318  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2319  * we write are pinned.
2320  *
2321  * They are released when wait == 1 is done.
2322  * max_mirrors must be the same for both runs, and it indicates how
2323  * many supers on this one device should be written.
2324  *
2325  * max_mirrors == 0 means to write them all.
2326  */
2327 static int write_dev_supers(struct btrfs_device *device,
2328                             struct btrfs_super_block *sb,
2329                             int do_barriers, int wait, int max_mirrors)
2330 {
2331         struct buffer_head *bh;
2332         int i;
2333         int ret;
2334         int errors = 0;
2335         u32 crc;
2336         u64 bytenr;
2337         int last_barrier = 0;
2338
2339         if (max_mirrors == 0)
2340                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2341
2342         /* make sure only the last submit_bh does a barrier */
2343         if (do_barriers) {
2344                 for (i = 0; i < max_mirrors; i++) {
2345                         bytenr = btrfs_sb_offset(i);
2346                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2347                             device->total_bytes)
2348                                 break;
2349                         last_barrier = i;
2350                 }
2351         }
2352
2353         for (i = 0; i < max_mirrors; i++) {
2354                 bytenr = btrfs_sb_offset(i);
2355                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2356                         break;
2357
2358                 if (wait) {
2359                         bh = __find_get_block(device->bdev, bytenr / 4096,
2360                                               BTRFS_SUPER_INFO_SIZE);
2361                         BUG_ON(!bh);
2362                         wait_on_buffer(bh);
2363                         if (!buffer_uptodate(bh))
2364                                 errors++;
2365
2366                         /* drop our reference */
2367                         brelse(bh);
2368
2369                         /* drop the reference from the wait == 0 run */
2370                         brelse(bh);
2371                         continue;
2372                 } else {
2373                         btrfs_set_super_bytenr(sb, bytenr);
2374
2375                         crc = ~(u32)0;
2376                         crc = btrfs_csum_data(NULL, (char *)sb +
2377                                               BTRFS_CSUM_SIZE, crc,
2378                                               BTRFS_SUPER_INFO_SIZE -
2379                                               BTRFS_CSUM_SIZE);
2380                         btrfs_csum_final(crc, sb->csum);
2381
2382                         /*
2383                          * one reference for us, and we leave it for the
2384                          * caller
2385                          */
2386                         bh = __getblk(device->bdev, bytenr / 4096,
2387                                       BTRFS_SUPER_INFO_SIZE);
2388                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2389
2390                         /* one reference for submit_bh */
2391                         get_bh(bh);
2392
2393                         set_buffer_uptodate(bh);
2394                         lock_buffer(bh);
2395                         bh->b_end_io = btrfs_end_buffer_write_sync;
2396                 }
2397
2398                 if (i == last_barrier && do_barriers)
2399                         ret = submit_bh(WRITE_FLUSH_FUA, bh);
2400                 else
2401                         ret = submit_bh(WRITE_SYNC, bh);
2402
2403                 if (ret)
2404                         errors++;
2405         }
2406         return errors < i ? 0 : -1;
2407 }
2408
2409 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2410 {
2411         struct list_head *head;
2412         struct btrfs_device *dev;
2413         struct btrfs_super_block *sb;
2414         struct btrfs_dev_item *dev_item;
2415         int ret;
2416         int do_barriers;
2417         int max_errors;
2418         int total_errors = 0;
2419         u64 flags;
2420
2421         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2422         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2423
2424         sb = &root->fs_info->super_for_commit;
2425         dev_item = &sb->dev_item;
2426
2427         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2428         head = &root->fs_info->fs_devices->devices;
2429         list_for_each_entry(dev, head, dev_list) {
2430                 if (!dev->bdev) {
2431                         total_errors++;
2432                         continue;
2433                 }
2434                 if (!dev->in_fs_metadata || !dev->writeable)
2435                         continue;
2436
2437                 btrfs_set_stack_device_generation(dev_item, 0);
2438                 btrfs_set_stack_device_type(dev_item, dev->type);
2439                 btrfs_set_stack_device_id(dev_item, dev->devid);
2440                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2441                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2442                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2443                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2444                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2445                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2446                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2447
2448                 flags = btrfs_super_flags(sb);
2449                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2450
2451                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2452                 if (ret)
2453                         total_errors++;
2454         }
2455         if (total_errors > max_errors) {
2456                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2457                        total_errors);
2458                 BUG();
2459         }
2460
2461         total_errors = 0;
2462         list_for_each_entry(dev, head, dev_list) {
2463                 if (!dev->bdev)
2464                         continue;
2465                 if (!dev->in_fs_metadata || !dev->writeable)
2466                         continue;
2467
2468                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2469                 if (ret)
2470                         total_errors++;
2471         }
2472         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2473         if (total_errors > max_errors) {
2474                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2475                        total_errors);
2476                 BUG();
2477         }
2478         return 0;
2479 }
2480
2481 int write_ctree_super(struct btrfs_trans_handle *trans,
2482                       struct btrfs_root *root, int max_mirrors)
2483 {
2484         int ret;
2485
2486         ret = write_all_supers(root, max_mirrors);
2487         return ret;
2488 }
2489
2490 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2491 {
2492         spin_lock(&fs_info->fs_roots_radix_lock);
2493         radix_tree_delete(&fs_info->fs_roots_radix,
2494                           (unsigned long)root->root_key.objectid);
2495         spin_unlock(&fs_info->fs_roots_radix_lock);
2496
2497         if (btrfs_root_refs(&root->root_item) == 0)
2498                 synchronize_srcu(&fs_info->subvol_srcu);
2499
2500         __btrfs_remove_free_space_cache(root->free_ino_pinned);
2501         __btrfs_remove_free_space_cache(root->free_ino_ctl);
2502         free_fs_root(root);
2503         return 0;
2504 }
2505
2506 static void free_fs_root(struct btrfs_root *root)
2507 {
2508         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2509         if (root->anon_super.s_dev) {
2510                 down_write(&root->anon_super.s_umount);
2511                 kill_anon_super(&root->anon_super);
2512         }
2513         free_extent_buffer(root->node);
2514         free_extent_buffer(root->commit_root);
2515         kfree(root->free_ino_ctl);
2516         kfree(root->free_ino_pinned);
2517         kfree(root->name);
2518         kfree(root);
2519 }
2520
2521 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2522 {
2523         int ret;
2524         struct btrfs_root *gang[8];
2525         int i;
2526
2527         while (!list_empty(&fs_info->dead_roots)) {
2528                 gang[0] = list_entry(fs_info->dead_roots.next,
2529                                      struct btrfs_root, root_list);
2530                 list_del(&gang[0]->root_list);
2531
2532                 if (gang[0]->in_radix) {
2533                         btrfs_free_fs_root(fs_info, gang[0]);
2534                 } else {
2535                         free_extent_buffer(gang[0]->node);
2536                         free_extent_buffer(gang[0]->commit_root);
2537                         kfree(gang[0]);
2538                 }
2539         }
2540
2541         while (1) {
2542                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2543                                              (void **)gang, 0,
2544                                              ARRAY_SIZE(gang));
2545                 if (!ret)
2546                         break;
2547                 for (i = 0; i < ret; i++)
2548                         btrfs_free_fs_root(fs_info, gang[i]);
2549         }
2550         return 0;
2551 }
2552
2553 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2554 {
2555         u64 root_objectid = 0;
2556         struct btrfs_root *gang[8];
2557         int i;
2558         int ret;
2559
2560         while (1) {
2561                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2562                                              (void **)gang, root_objectid,
2563                                              ARRAY_SIZE(gang));
2564                 if (!ret)
2565                         break;
2566
2567                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2568                 for (i = 0; i < ret; i++) {
2569                         int err;
2570
2571                         root_objectid = gang[i]->root_key.objectid;
2572                         err = btrfs_orphan_cleanup(gang[i]);
2573                         if (err)
2574                                 return err;
2575                 }
2576                 root_objectid++;
2577         }
2578         return 0;
2579 }
2580
2581 int btrfs_commit_super(struct btrfs_root *root)
2582 {
2583         struct btrfs_trans_handle *trans;
2584         int ret;
2585
2586         mutex_lock(&root->fs_info->cleaner_mutex);
2587         btrfs_run_delayed_iputs(root);
2588         btrfs_clean_old_snapshots(root);
2589         mutex_unlock(&root->fs_info->cleaner_mutex);
2590
2591         /* wait until ongoing cleanup work done */
2592         down_write(&root->fs_info->cleanup_work_sem);
2593         up_write(&root->fs_info->cleanup_work_sem);
2594
2595         trans = btrfs_join_transaction(root, 1);
2596         if (IS_ERR(trans))
2597                 return PTR_ERR(trans);
2598         ret = btrfs_commit_transaction(trans, root);
2599         BUG_ON(ret);
2600         /* run commit again to drop the original snapshot */
2601         trans = btrfs_join_transaction(root, 1);
2602         if (IS_ERR(trans))
2603                 return PTR_ERR(trans);
2604         btrfs_commit_transaction(trans, root);
2605         ret = btrfs_write_and_wait_transaction(NULL, root);
2606         BUG_ON(ret);
2607
2608         ret = write_ctree_super(NULL, root, 0);
2609         return ret;
2610 }
2611
2612 int close_ctree(struct btrfs_root *root)
2613 {
2614         struct btrfs_fs_info *fs_info = root->fs_info;
2615         int ret;
2616
2617         fs_info->closing = 1;
2618         smp_mb();
2619
2620         btrfs_put_block_group_cache(fs_info);
2621
2622         /*
2623          * Here come 2 situations when btrfs is broken to flip readonly:
2624          *
2625          * 1. when btrfs flips readonly somewhere else before
2626          * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2627          * and btrfs will skip to write sb directly to keep
2628          * ERROR state on disk.
2629          *
2630          * 2. when btrfs flips readonly just in btrfs_commit_super,
2631          * and in such case, btrfs cannnot write sb via btrfs_commit_super,
2632          * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2633          * btrfs will cleanup all FS resources first and write sb then.
2634          */
2635         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2636                 ret = btrfs_commit_super(root);
2637                 if (ret)
2638                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2639         }
2640
2641         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2642                 ret = btrfs_error_commit_super(root);
2643                 if (ret)
2644                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2645         }
2646
2647         kthread_stop(root->fs_info->transaction_kthread);
2648         kthread_stop(root->fs_info->cleaner_kthread);
2649
2650         fs_info->closing = 2;
2651         smp_mb();
2652
2653         if (fs_info->delalloc_bytes) {
2654                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2655                        (unsigned long long)fs_info->delalloc_bytes);
2656         }
2657         if (fs_info->total_ref_cache_size) {
2658                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2659                        (unsigned long long)fs_info->total_ref_cache_size);
2660         }
2661
2662         free_extent_buffer(fs_info->extent_root->node);
2663         free_extent_buffer(fs_info->extent_root->commit_root);
2664         free_extent_buffer(fs_info->tree_root->node);
2665         free_extent_buffer(fs_info->tree_root->commit_root);
2666         free_extent_buffer(root->fs_info->chunk_root->node);
2667         free_extent_buffer(root->fs_info->chunk_root->commit_root);
2668         free_extent_buffer(root->fs_info->dev_root->node);
2669         free_extent_buffer(root->fs_info->dev_root->commit_root);
2670         free_extent_buffer(root->fs_info->csum_root->node);
2671         free_extent_buffer(root->fs_info->csum_root->commit_root);
2672
2673         btrfs_free_block_groups(root->fs_info);
2674
2675         del_fs_roots(fs_info);
2676
2677         iput(fs_info->btree_inode);
2678
2679         btrfs_stop_workers(&fs_info->generic_worker);
2680         btrfs_stop_workers(&fs_info->fixup_workers);
2681         btrfs_stop_workers(&fs_info->delalloc_workers);
2682         btrfs_stop_workers(&fs_info->workers);
2683         btrfs_stop_workers(&fs_info->endio_workers);
2684         btrfs_stop_workers(&fs_info->endio_meta_workers);
2685         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2686         btrfs_stop_workers(&fs_info->endio_write_workers);
2687         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2688         btrfs_stop_workers(&fs_info->submit_workers);
2689
2690         btrfs_close_devices(fs_info->fs_devices);
2691         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2692
2693         bdi_destroy(&fs_info->bdi);
2694         cleanup_srcu_struct(&fs_info->subvol_srcu);
2695
2696         kfree(fs_info->extent_root);
2697         kfree(fs_info->tree_root);
2698         kfree(fs_info->chunk_root);
2699         kfree(fs_info->dev_root);
2700         kfree(fs_info->csum_root);
2701         kfree(fs_info);
2702
2703         return 0;
2704 }
2705
2706 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2707 {
2708         int ret;
2709         struct inode *btree_inode = buf->first_page->mapping->host;
2710
2711         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2712                                      NULL);
2713         if (!ret)
2714                 return ret;
2715
2716         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2717                                     parent_transid);
2718         return !ret;
2719 }
2720
2721 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2722 {
2723         struct inode *btree_inode = buf->first_page->mapping->host;
2724         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2725                                           buf);
2726 }
2727
2728 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2729 {
2730         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2731         u64 transid = btrfs_header_generation(buf);
2732         struct inode *btree_inode = root->fs_info->btree_inode;
2733         int was_dirty;
2734
2735         btrfs_assert_tree_locked(buf);
2736         if (transid != root->fs_info->generation) {
2737                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2738                        "found %llu running %llu\n",
2739                         (unsigned long long)buf->start,
2740                         (unsigned long long)transid,
2741                         (unsigned long long)root->fs_info->generation);
2742                 WARN_ON(1);
2743         }
2744         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2745                                             buf);
2746         if (!was_dirty) {
2747                 spin_lock(&root->fs_info->delalloc_lock);
2748                 root->fs_info->dirty_metadata_bytes += buf->len;
2749                 spin_unlock(&root->fs_info->delalloc_lock);
2750         }
2751 }
2752
2753 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2754 {
2755         /*
2756          * looks as though older kernels can get into trouble with
2757          * this code, they end up stuck in balance_dirty_pages forever
2758          */
2759         u64 num_dirty;
2760         unsigned long thresh = 32 * 1024 * 1024;
2761
2762         if (current->flags & PF_MEMALLOC)
2763                 return;
2764
2765         num_dirty = root->fs_info->dirty_metadata_bytes;
2766
2767         if (num_dirty > thresh) {
2768                 balance_dirty_pages_ratelimited_nr(
2769                                    root->fs_info->btree_inode->i_mapping, 1);
2770         }
2771         return;
2772 }
2773
2774 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2775 {
2776         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2777         int ret;
2778         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2779         if (ret == 0)
2780                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2781         return ret;
2782 }
2783
2784 int btree_lock_page_hook(struct page *page)
2785 {
2786         struct inode *inode = page->mapping->host;
2787         struct btrfs_root *root = BTRFS_I(inode)->root;
2788         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2789         struct extent_buffer *eb;
2790         unsigned long len;
2791         u64 bytenr = page_offset(page);
2792
2793         if (page->private == EXTENT_PAGE_PRIVATE)
2794                 goto out;
2795
2796         len = page->private >> 2;
2797         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2798         if (!eb)
2799                 goto out;
2800
2801         btrfs_tree_lock(eb);
2802         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2803
2804         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2805                 spin_lock(&root->fs_info->delalloc_lock);
2806                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2807                         root->fs_info->dirty_metadata_bytes -= eb->len;
2808                 else
2809                         WARN_ON(1);
2810                 spin_unlock(&root->fs_info->delalloc_lock);
2811         }
2812
2813         btrfs_tree_unlock(eb);
2814         free_extent_buffer(eb);
2815 out:
2816         lock_page(page);
2817         return 0;
2818 }
2819
2820 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
2821                               int read_only)
2822 {
2823         if (read_only)
2824                 return;
2825
2826         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
2827                 printk(KERN_WARNING "warning: mount fs with errors, "
2828                        "running btrfsck is recommended\n");
2829 }
2830
2831 int btrfs_error_commit_super(struct btrfs_root *root)
2832 {
2833         int ret;
2834
2835         mutex_lock(&root->fs_info->cleaner_mutex);
2836         btrfs_run_delayed_iputs(root);
2837         mutex_unlock(&root->fs_info->cleaner_mutex);
2838
2839         down_write(&root->fs_info->cleanup_work_sem);
2840         up_write(&root->fs_info->cleanup_work_sem);
2841
2842         /* cleanup FS via transaction */
2843         btrfs_cleanup_transaction(root);
2844
2845         ret = write_ctree_super(NULL, root, 0);
2846
2847         return ret;
2848 }
2849
2850 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
2851 {
2852         struct btrfs_inode *btrfs_inode;
2853         struct list_head splice;
2854
2855         INIT_LIST_HEAD(&splice);
2856
2857         mutex_lock(&root->fs_info->ordered_operations_mutex);
2858         spin_lock(&root->fs_info->ordered_extent_lock);
2859
2860         list_splice_init(&root->fs_info->ordered_operations, &splice);
2861         while (!list_empty(&splice)) {
2862                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2863                                          ordered_operations);
2864
2865                 list_del_init(&btrfs_inode->ordered_operations);
2866
2867                 btrfs_invalidate_inodes(btrfs_inode->root);
2868         }
2869
2870         spin_unlock(&root->fs_info->ordered_extent_lock);
2871         mutex_unlock(&root->fs_info->ordered_operations_mutex);
2872
2873         return 0;
2874 }
2875
2876 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
2877 {
2878         struct list_head splice;
2879         struct btrfs_ordered_extent *ordered;
2880         struct inode *inode;
2881
2882         INIT_LIST_HEAD(&splice);
2883
2884         spin_lock(&root->fs_info->ordered_extent_lock);
2885
2886         list_splice_init(&root->fs_info->ordered_extents, &splice);
2887         while (!list_empty(&splice)) {
2888                 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
2889                                      root_extent_list);
2890
2891                 list_del_init(&ordered->root_extent_list);
2892                 atomic_inc(&ordered->refs);
2893
2894                 /* the inode may be getting freed (in sys_unlink path). */
2895                 inode = igrab(ordered->inode);
2896
2897                 spin_unlock(&root->fs_info->ordered_extent_lock);
2898                 if (inode)
2899                         iput(inode);
2900
2901                 atomic_set(&ordered->refs, 1);
2902                 btrfs_put_ordered_extent(ordered);
2903
2904                 spin_lock(&root->fs_info->ordered_extent_lock);
2905         }
2906
2907         spin_unlock(&root->fs_info->ordered_extent_lock);
2908
2909         return 0;
2910 }
2911
2912 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
2913                                       struct btrfs_root *root)
2914 {
2915         struct rb_node *node;
2916         struct btrfs_delayed_ref_root *delayed_refs;
2917         struct btrfs_delayed_ref_node *ref;
2918         int ret = 0;
2919
2920         delayed_refs = &trans->delayed_refs;
2921
2922         spin_lock(&delayed_refs->lock);
2923         if (delayed_refs->num_entries == 0) {
2924                 printk(KERN_INFO "delayed_refs has NO entry\n");
2925                 return ret;
2926         }
2927
2928         node = rb_first(&delayed_refs->root);
2929         while (node) {
2930                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2931                 node = rb_next(node);
2932
2933                 ref->in_tree = 0;
2934                 rb_erase(&ref->rb_node, &delayed_refs->root);
2935                 delayed_refs->num_entries--;
2936
2937                 atomic_set(&ref->refs, 1);
2938                 if (btrfs_delayed_ref_is_head(ref)) {
2939                         struct btrfs_delayed_ref_head *head;
2940
2941                         head = btrfs_delayed_node_to_head(ref);
2942                         mutex_lock(&head->mutex);
2943                         kfree(head->extent_op);
2944                         delayed_refs->num_heads--;
2945                         if (list_empty(&head->cluster))
2946                                 delayed_refs->num_heads_ready--;
2947                         list_del_init(&head->cluster);
2948                         mutex_unlock(&head->mutex);
2949                 }
2950
2951                 spin_unlock(&delayed_refs->lock);
2952                 btrfs_put_delayed_ref(ref);
2953
2954                 cond_resched();
2955                 spin_lock(&delayed_refs->lock);
2956         }
2957
2958         spin_unlock(&delayed_refs->lock);
2959
2960         return ret;
2961 }
2962
2963 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
2964 {
2965         struct btrfs_pending_snapshot *snapshot;
2966         struct list_head splice;
2967
2968         INIT_LIST_HEAD(&splice);
2969
2970         list_splice_init(&t->pending_snapshots, &splice);
2971
2972         while (!list_empty(&splice)) {
2973                 snapshot = list_entry(splice.next,
2974                                       struct btrfs_pending_snapshot,
2975                                       list);
2976
2977                 list_del_init(&snapshot->list);
2978
2979                 kfree(snapshot);
2980         }
2981
2982         return 0;
2983 }
2984
2985 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
2986 {
2987         struct btrfs_inode *btrfs_inode;
2988         struct list_head splice;
2989
2990         INIT_LIST_HEAD(&splice);
2991
2992         list_splice_init(&root->fs_info->delalloc_inodes, &splice);
2993
2994         spin_lock(&root->fs_info->delalloc_lock);
2995
2996         while (!list_empty(&splice)) {
2997                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2998                                     delalloc_inodes);
2999
3000                 list_del_init(&btrfs_inode->delalloc_inodes);
3001
3002                 btrfs_invalidate_inodes(btrfs_inode->root);
3003         }
3004
3005         spin_unlock(&root->fs_info->delalloc_lock);
3006
3007         return 0;
3008 }
3009
3010 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3011                                         struct extent_io_tree *dirty_pages,
3012                                         int mark)
3013 {
3014         int ret;
3015         struct page *page;
3016         struct inode *btree_inode = root->fs_info->btree_inode;
3017         struct extent_buffer *eb;
3018         u64 start = 0;
3019         u64 end;
3020         u64 offset;
3021         unsigned long index;
3022
3023         while (1) {
3024                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3025                                             mark);
3026                 if (ret)
3027                         break;
3028
3029                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3030                 while (start <= end) {
3031                         index = start >> PAGE_CACHE_SHIFT;
3032                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3033                         page = find_get_page(btree_inode->i_mapping, index);
3034                         if (!page)
3035                                 continue;
3036                         offset = page_offset(page);
3037
3038                         spin_lock(&dirty_pages->buffer_lock);
3039                         eb = radix_tree_lookup(
3040                              &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3041                                                offset >> PAGE_CACHE_SHIFT);
3042                         spin_unlock(&dirty_pages->buffer_lock);
3043                         if (eb) {
3044                                 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3045                                                          &eb->bflags);
3046                                 atomic_set(&eb->refs, 1);
3047                         }
3048                         if (PageWriteback(page))
3049                                 end_page_writeback(page);
3050
3051                         lock_page(page);
3052                         if (PageDirty(page)) {
3053                                 clear_page_dirty_for_io(page);
3054                                 spin_lock_irq(&page->mapping->tree_lock);
3055                                 radix_tree_tag_clear(&page->mapping->page_tree,
3056                                                         page_index(page),
3057                                                         PAGECACHE_TAG_DIRTY);
3058                                 spin_unlock_irq(&page->mapping->tree_lock);
3059                         }
3060
3061                         page->mapping->a_ops->invalidatepage(page, 0);
3062                         unlock_page(page);
3063                 }
3064         }
3065
3066         return ret;
3067 }
3068
3069 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3070                                        struct extent_io_tree *pinned_extents)
3071 {
3072         struct extent_io_tree *unpin;
3073         u64 start;
3074         u64 end;
3075         int ret;
3076
3077         unpin = pinned_extents;
3078         while (1) {
3079                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3080                                             EXTENT_DIRTY);
3081                 if (ret)
3082                         break;
3083
3084                 /* opt_discard */
3085                 if (btrfs_test_opt(root, DISCARD))
3086                         ret = btrfs_error_discard_extent(root, start,
3087                                                          end + 1 - start,
3088                                                          NULL);
3089
3090                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3091                 btrfs_error_unpin_extent_range(root, start, end);
3092                 cond_resched();
3093         }
3094
3095         return 0;
3096 }
3097
3098 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3099 {
3100         struct btrfs_transaction *t;
3101         LIST_HEAD(list);
3102
3103         WARN_ON(1);
3104
3105         mutex_lock(&root->fs_info->trans_mutex);
3106         mutex_lock(&root->fs_info->transaction_kthread_mutex);
3107
3108         list_splice_init(&root->fs_info->trans_list, &list);
3109         while (!list_empty(&list)) {
3110                 t = list_entry(list.next, struct btrfs_transaction, list);
3111                 if (!t)
3112                         break;
3113
3114                 btrfs_destroy_ordered_operations(root);
3115
3116                 btrfs_destroy_ordered_extents(root);
3117
3118                 btrfs_destroy_delayed_refs(t, root);
3119
3120                 btrfs_block_rsv_release(root,
3121                                         &root->fs_info->trans_block_rsv,
3122                                         t->dirty_pages.dirty_bytes);
3123
3124                 /* FIXME: cleanup wait for commit */
3125                 t->in_commit = 1;
3126                 t->blocked = 1;
3127                 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3128                         wake_up(&root->fs_info->transaction_blocked_wait);
3129
3130                 t->blocked = 0;
3131                 if (waitqueue_active(&root->fs_info->transaction_wait))
3132                         wake_up(&root->fs_info->transaction_wait);
3133                 mutex_unlock(&root->fs_info->trans_mutex);
3134
3135                 mutex_lock(&root->fs_info->trans_mutex);
3136                 t->commit_done = 1;
3137                 if (waitqueue_active(&t->commit_wait))
3138                         wake_up(&t->commit_wait);
3139                 mutex_unlock(&root->fs_info->trans_mutex);
3140
3141                 mutex_lock(&root->fs_info->trans_mutex);
3142
3143                 btrfs_destroy_pending_snapshots(t);
3144
3145                 btrfs_destroy_delalloc_inodes(root);
3146
3147                 spin_lock(&root->fs_info->new_trans_lock);
3148                 root->fs_info->running_transaction = NULL;
3149                 spin_unlock(&root->fs_info->new_trans_lock);
3150
3151                 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3152                                              EXTENT_DIRTY);
3153
3154                 btrfs_destroy_pinned_extent(root,
3155                                             root->fs_info->pinned_extents);
3156
3157                 atomic_set(&t->use_count, 0);
3158                 list_del_init(&t->list);
3159                 memset(t, 0, sizeof(*t));
3160                 kmem_cache_free(btrfs_transaction_cachep, t);
3161         }
3162
3163         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3164         mutex_unlock(&root->fs_info->trans_mutex);
3165
3166         return 0;
3167 }
3168
3169 static struct extent_io_ops btree_extent_io_ops = {
3170         .write_cache_pages_lock_hook = btree_lock_page_hook,
3171         .readpage_end_io_hook = btree_readpage_end_io_hook,
3172         .submit_bio_hook = btree_submit_bio_hook,
3173         /* note we're sharing with inode.c for the merge bio hook */
3174         .merge_bio_hook = btrfs_merge_bio_hook,
3175 };