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
  45 static struct extent_io_ops btree_extent_io_ops;
  46 static void end_workqueue_fn(struct btrfs_work *work);
  47 static void free_fs_root(struct btrfs_root *root);
  48 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  49                                     int read_only);
  50 static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
  51 static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
  52 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  53                                       struct btrfs_root *root);
  54 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
  55 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  56 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
  57                                         struct extent_io_tree *dirty_pages,
  58                                         int mark);
  59 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
  60                                        struct extent_io_tree *pinned_extents);
  61 static int btrfs_cleanup_transaction(struct btrfs_root *root);
  62
  63 /*
  64  * end_io_wq structs are used to do processing in task context when an IO is
  65  * complete.  This is used during reads to verify checksums, and it is used
  66  * by writes to insert metadata for new file extents after IO is complete.
  67  */
  68 struct end_io_wq {
  69         struct bio *bio;
  70         bio_end_io_t *end_io;
  71         void *private;
  72         struct btrfs_fs_info *info;
  73         int error;
  74         int metadata;
  75         struct list_head list;
  76         struct btrfs_work work;
  77 };
  78
  79 /*
  80  * async submit bios are used to offload expensive checksumming
  81  * onto the worker threads.  They checksum file and metadata bios
  82  * just before they are sent down the IO stack.
  83  */
  84 struct async_submit_bio {
  85         struct inode *inode;
  86         struct bio *bio;
  87         struct list_head list;
  88         extent_submit_bio_hook_t *submit_bio_start;
  89         extent_submit_bio_hook_t *submit_bio_done;
  90         int rw;
  91         int mirror_num;
  92         unsigned long bio_flags;
  93         /*
  94          * bio_offset is optional, can be used if the pages in the bio
  95          * can't tell us where in the file the bio should go
  96          */
  97         u64 bio_offset;
  98         struct btrfs_work work;
  99 };
 100
 101 /* These are used to set the lockdep class on the extent buffer locks.
 102  * The class is set by the readpage_end_io_hook after the buffer has
 103  * passed csum validation but before the pages are unlocked.
 104  *
 105  * The lockdep class is also set by btrfs_init_new_buffer on freshly
 106  * allocated blocks.
 107  *
 108  * The class is based on the level in the tree block, which allows lockdep
 109  * to know that lower nodes nest inside the locks of higher nodes.
 110  *
 111  * We also add a check to make sure the highest level of the tree is
 112  * the same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this
 113  * code needs update as well.
 114  */
 115 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 116 # if BTRFS_MAX_LEVEL != 8
 117 #  error
 118 # endif
 119 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
 120 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
 121         /* leaf */
 122         "btrfs-extent-00",
 123         "btrfs-extent-01",
 124         "btrfs-extent-02",
 125         "btrfs-extent-03",
 126         "btrfs-extent-04",
 127         "btrfs-extent-05",
 128         "btrfs-extent-06",
 129         "btrfs-extent-07",
 130         /* highest possible level */
 131         "btrfs-extent-08",
 132 };
 133 #endif
 134
 135 /*
 136  * extents on the btree inode are pretty simple, there's one extent
 137  * that covers the entire device
 138  */
 139 static struct extent_map *btree_get_extent(struct inode *inode,
 140                 struct page *page, size_t page_offset, u64 start, u64 len,
 141                 int create)
 142 {
 143         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 144         struct extent_map *em;
 145         int ret;
 146
 147         read_lock(&em_tree->lock);
 148         em = lookup_extent_mapping(em_tree, start, len);
 149         if (em) {
 150                 em->bdev =
 151                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 152                 read_unlock(&em_tree->lock);
 153                 goto out;
 154         }
 155         read_unlock(&em_tree->lock);
 156
 157         em = alloc_extent_map(GFP_NOFS);
 158         if (!em) {
 159                 em = ERR_PTR(-ENOMEM);
 160                 goto out;
 161         }
 162         em->start = 0;
 163         em->len = (u64)-1;
 164         em->block_len = (u64)-1;
 165         em->block_start = 0;
 166         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 167
 168         write_lock(&em_tree->lock);
 169         ret = add_extent_mapping(em_tree, em);
 170         if (ret == -EEXIST) {
 171                 u64 failed_start = em->start;
 172                 u64 failed_len = em->len;
 173
 174                 free_extent_map(em);
 175                 em = lookup_extent_mapping(em_tree, start, len);
 176                 if (em) {
 177                         ret = 0;
 178                 } else {
 179                         em = lookup_extent_mapping(em_tree, failed_start,
 180                                                    failed_len);
 181                         ret = -EIO;
 182                 }
 183         } else if (ret) {
 184                 free_extent_map(em);
 185                 em = NULL;
 186         }
 187         write_unlock(&em_tree->lock);
 188
 189         if (ret)
 190                 em = ERR_PTR(ret);
 191 out:
 192         return em;
 193 }
 194
 195 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 196 {
 197         return crc32c(seed, data, len);
 198 }
 199
 200 void btrfs_csum_final(u32 crc, char *result)
 201 {
 202         put_unaligned_le32(~crc, result);
 203 }
 204
 205 /*
 206  * compute the csum for a btree block, and either verify it or write it
 207  * into the csum field of the block.
 208  */
 209 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 210                            int verify)
 211 {
 212         u16 csum_size =
 213                 btrfs_super_csum_size(&root->fs_info->super_copy);
 214         char *result = NULL;
 215         unsigned long len;
 216         unsigned long cur_len;
 217         unsigned long offset = BTRFS_CSUM_SIZE;
 218         char *map_token = NULL;
 219         char *kaddr;
 220         unsigned long map_start;
 221         unsigned long map_len;
 222         int err;
 223         u32 crc = ~(u32)0;
 224         unsigned long inline_result;
 225
 226         len = buf->len - offset;
 227         while (len > 0) {
 228                 err = map_private_extent_buffer(buf, offset, 32,
 229                                         &map_token, &kaddr,
 230                                         &map_start, &map_len, KM_USER0);
 231                 if (err)
 232                         return 1;
 233                 cur_len = min(len, map_len - (offset - map_start));
 234                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 235                                       crc, cur_len);
 236                 len -= cur_len;
 237                 offset += cur_len;
 238                 unmap_extent_buffer(buf, map_token, KM_USER0);
 239         }
 240         if (csum_size > sizeof(inline_result)) {
 241                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
 242                 if (!result)
 243                         return 1;
 244         } else {
 245                 result = (char *)&inline_result;
 246         }
 247
 248         btrfs_csum_final(crc, result);
 249
 250         if (verify) {
 251                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 252                         u32 val;
 253                         u32 found = 0;
 254                         memcpy(&found, result, csum_size);
 255
 256                         read_extent_buffer(buf, &val, 0, csum_size);
 257                         if (printk_ratelimit()) {
 258                                 printk(KERN_INFO "btrfs: %s checksum verify "
 259                                        "failed on %llu wanted %X found %X "
 260                                        "level %d\n",
 261                                        root->fs_info->sb->s_id,
 262                                        (unsigned long long)buf->start, val, found,
 263                                        btrfs_header_level(buf));
 264                         }
 265                         if (result != (char *)&inline_result)
 266                                 kfree(result);
 267                         return 1;
 268                 }
 269         } else {
 270                 write_extent_buffer(buf, result, 0, csum_size);
 271         }
 272         if (result != (char *)&inline_result)
 273                 kfree(result);
 274         return 0;
 275 }
 276
 277 /*
 278  * we can't consider a given block up to date unless the transid of the
 279  * block matches the transid in the parent node's pointer.  This is how we
 280  * detect blocks that either didn't get written at all or got written
 281  * in the wrong place.
 282  */
 283 static int verify_parent_transid(struct extent_io_tree *io_tree,
 284                                  struct extent_buffer *eb, u64 parent_transid)
 285 {
 286         struct extent_state *cached_state = NULL;
 287         int ret;
 288
 289         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 290                 return 0;
 291
 292         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 293                          0, &cached_state, GFP_NOFS);
 294         if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
 295             btrfs_header_generation(eb) == parent_transid) {
 296                 ret = 0;
 297                 goto out;
 298         }
 299         if (printk_ratelimit()) {
 300                 printk("parent transid verify failed on %llu wanted %llu "
 301                        "found %llu\n",
 302                        (unsigned long long)eb->start,
 303                        (unsigned long long)parent_transid,
 304                        (unsigned long long)btrfs_header_generation(eb));
 305         }
 306         ret = 1;
 307         clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
 308 out:
 309         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 310                              &cached_state, GFP_NOFS);
 311         return ret;
 312 }
 313
 314 /*
 315  * helper to read a given tree block, doing retries as required when
 316  * the checksums don't match and we have alternate mirrors to try.
 317  */
 318 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 319                                           struct extent_buffer *eb,
 320                                           u64 start, u64 parent_transid)
 321 {
 322         struct extent_io_tree *io_tree;
 323         int ret;
 324         int num_copies = 0;
 325         int mirror_num = 0;
 326
 327         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 328         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 329         while (1) {
 330                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 331                                                btree_get_extent, mirror_num);
 332                 if (!ret &&
 333                     !verify_parent_transid(io_tree, eb, parent_transid))
 334                         return ret;
 335
 336                 /*
 337                  * This buffer's crc is fine, but its contents are corrupted, so
 338                  * there is no reason to read the other copies, they won't be
 339                  * any less wrong.
 340                  */
 341                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 342                         return ret;
 343
 344                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 345                                               eb->start, eb->len);
 346                 if (num_copies == 1)
 347                         return ret;
 348
 349                 mirror_num++;
 350                 if (mirror_num > num_copies)
 351                         return ret;
 352         }
 353         return -EIO;
 354 }
 355
 356 /*
 357  * checksum a dirty tree block before IO.  This has extra checks to make sure
 358  * we only fill in the checksum field in the first page of a multi-page block
 359  */
 360
 361 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 362 {
 363         struct extent_io_tree *tree;
 364         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 365         u64 found_start;
 366         unsigned long len;
 367         struct extent_buffer *eb;
 368         int ret;
 369
 370         tree = &BTRFS_I(page->mapping->host)->io_tree;
 371
 372         if (page->private == EXTENT_PAGE_PRIVATE) {
 373                 WARN_ON(1);
 374                 goto out;
 375         }
 376         if (!page->private) {
 377                 WARN_ON(1);
 378                 goto out;
 379         }
 380         len = page->private >> 2;
 381         WARN_ON(len == 0);
 382
 383         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 384         if (eb == NULL) {
 385                 WARN_ON(1);
 386                 goto out;
 387         }
 388         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 389                                              btrfs_header_generation(eb));
 390         BUG_ON(ret);
 391         WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
 392
 393         found_start = btrfs_header_bytenr(eb);
 394         if (found_start != start) {
 395                 WARN_ON(1);
 396                 goto err;
 397         }
 398         if (eb->first_page != page) {
 399                 WARN_ON(1);
 400                 goto err;
 401         }
 402         if (!PageUptodate(page)) {
 403                 WARN_ON(1);
 404                 goto err;
 405         }
 406         csum_tree_block(root, eb, 0);
 407 err:
 408         free_extent_buffer(eb);
 409 out:
 410         return 0;
 411 }
 412
 413 static int check_tree_block_fsid(struct btrfs_root *root,
 414                                  struct extent_buffer *eb)
 415 {
 416         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 417         u8 fsid[BTRFS_UUID_SIZE];
 418         int ret = 1;
 419
 420         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
 421                            BTRFS_FSID_SIZE);
 422         while (fs_devices) {
 423                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 424                         ret = 0;
 425                         break;
 426                 }
 427                 fs_devices = fs_devices->seed;
 428         }
 429         return ret;
 430 }
 431
 432 #define CORRUPT(reason, eb, root, slot)                         \
 433         printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
 434                "root=%llu, slot=%d\n", reason,                  \
 435                (unsigned long long)btrfs_header_bytenr(eb),     \
 436                (unsigned long long)root->objectid, slot)
 437
 438 static noinline int check_leaf(struct btrfs_root *root,
 439                                struct extent_buffer *leaf)
 440 {
 441         struct btrfs_key key;
 442         struct btrfs_key leaf_key;
 443         u32 nritems = btrfs_header_nritems(leaf);
 444         int slot;
 445
 446         if (nritems == 0)
 447                 return 0;
 448
 449         /* Check the 0 item */
 450         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 451             BTRFS_LEAF_DATA_SIZE(root)) {
 452                 CORRUPT("invalid item offset size pair", leaf, root, 0);
 453                 return -EIO;
 454         }
 455
 456         /*
 457          * Check to make sure each items keys are in the correct order and their
 458          * offsets make sense.  We only have to loop through nritems-1 because
 459          * we check the current slot against the next slot, which verifies the
 460          * next slot's offset+size makes sense and that the current's slot
 461          * offset is correct.
 462          */
 463         for (slot = 0; slot < nritems - 1; slot++) {
 464                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 465                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 466
 467                 /* Make sure the keys are in the right order */
 468                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 469                         CORRUPT("bad key order", leaf, root, slot);
 470                         return -EIO;
 471                 }
 472
 473                 /*
 474                  * Make sure the offset and ends are right, remember that the
 475                  * item data starts at the end of the leaf and grows towards the
 476                  * front.
 477                  */
 478                 if (btrfs_item_offset_nr(leaf, slot) !=
 479                         btrfs_item_end_nr(leaf, slot + 1)) {
 480                         CORRUPT("slot offset bad", leaf, root, slot);
 481                         return -EIO;
 482                 }
 483
 484                 /*
 485                  * Check to make sure that we don't point outside of the leaf,
 486                  * just incase all the items are consistent to eachother, but
 487                  * all point outside of the leaf.
 488                  */
 489                 if (btrfs_item_end_nr(leaf, slot) >
 490                     BTRFS_LEAF_DATA_SIZE(root)) {
 491                         CORRUPT("slot end outside of leaf", leaf, root, slot);
 492                         return -EIO;
 493                 }
 494         }
 495
 496         return 0;
 497 }
 498
 499 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 500 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
 501 {
 502         lockdep_set_class_and_name(&eb->lock,
 503                            &btrfs_eb_class[level],
 504                            btrfs_eb_name[level]);
 505 }
 506 #endif
 507
 508 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 509                                struct extent_state *state)
 510 {
 511         struct extent_io_tree *tree;
 512         u64 found_start;
 513         int found_level;
 514         unsigned long len;
 515         struct extent_buffer *eb;
 516         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 517         int ret = 0;
 518
 519         tree = &BTRFS_I(page->mapping->host)->io_tree;
 520         if (page->private == EXTENT_PAGE_PRIVATE)
 521                 goto out;
 522         if (!page->private)
 523                 goto out;
 524
 525         len = page->private >> 2;
 526         WARN_ON(len == 0);
 527
 528         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 529         if (eb == NULL) {
 530                 ret = -EIO;
 531                 goto out;
 532         }
 533
 534         found_start = btrfs_header_bytenr(eb);
 535         if (found_start != start) {
 536                 if (printk_ratelimit()) {
 537                         printk(KERN_INFO "btrfs bad tree block start "
 538                                "%llu %llu\n",
 539                                (unsigned long long)found_start,
 540                                (unsigned long long)eb->start);
 541                 }
 542                 ret = -EIO;
 543                 goto err;
 544         }
 545         if (eb->first_page != page) {
 546                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
 547                        eb->first_page->index, page->index);
 548                 WARN_ON(1);
 549                 ret = -EIO;
 550                 goto err;
 551         }
 552         if (check_tree_block_fsid(root, eb)) {
 553                 if (printk_ratelimit()) {
 554                         printk(KERN_INFO "btrfs bad fsid on block %llu\n",
 555                                (unsigned long long)eb->start);
 556                 }
 557                 ret = -EIO;
 558                 goto err;
 559         }
 560         found_level = btrfs_header_level(eb);
 561
 562         btrfs_set_buffer_lockdep_class(eb, found_level);
 563
 564         ret = csum_tree_block(root, eb, 1);
 565         if (ret) {
 566                 ret = -EIO;
 567                 goto err;
 568         }
 569
 570         /*
 571          * If this is a leaf block and it is corrupt, set the corrupt bit so
 572          * that we don't try and read the other copies of this block, just
 573          * return -EIO.
 574          */
 575         if (found_level == 0 && check_leaf(root, eb)) {
 576                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 577                 ret = -EIO;
 578         }
 579
 580         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 581         end = eb->start + end - 1;
 582 err:
 583         free_extent_buffer(eb);
 584 out:
 585         return ret;
 586 }
 587
 588 static void end_workqueue_bio(struct bio *bio, int err)
 589 {
 590         struct end_io_wq *end_io_wq = bio->bi_private;
 591         struct btrfs_fs_info *fs_info;
 592
 593         fs_info = end_io_wq->info;
 594         end_io_wq->error = err;
 595         end_io_wq->work.func = end_workqueue_fn;
 596         end_io_wq->work.flags = 0;
 597
 598         if (bio->bi_rw & REQ_WRITE) {
 599                 if (end_io_wq->metadata == 1)
 600                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
 601                                            &end_io_wq->work);
 602                 else if (end_io_wq->metadata == 2)
 603                         btrfs_queue_worker(&fs_info->endio_freespace_worker,
 604                                            &end_io_wq->work);
 605                 else
 606                         btrfs_queue_worker(&fs_info->endio_write_workers,
 607                                            &end_io_wq->work);
 608         } else {
 609                 if (end_io_wq->metadata)
 610                         btrfs_queue_worker(&fs_info->endio_meta_workers,
 611                                            &end_io_wq->work);
 612                 else
 613                         btrfs_queue_worker(&fs_info->endio_workers,
 614                                            &end_io_wq->work);
 615         }
 616 }
 617
 618 /*
 619  * For the metadata arg you want
 620  *
 621  * 0 - if data
 622  * 1 - if normal metadta
 623  * 2 - if writing to the free space cache area
 624  */
 625 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 626                         int metadata)
 627 {
 628         struct end_io_wq *end_io_wq;
 629         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 630         if (!end_io_wq)
 631                 return -ENOMEM;
 632
 633         end_io_wq->private = bio->bi_private;
 634         end_io_wq->end_io = bio->bi_end_io;
 635         end_io_wq->info = info;
 636         end_io_wq->error = 0;
 637         end_io_wq->bio = bio;
 638         end_io_wq->metadata = metadata;
 639
 640         bio->bi_private = end_io_wq;
 641         bio->bi_end_io = end_workqueue_bio;
 642         return 0;
 643 }
 644
 645 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 646 {
 647         unsigned long limit = min_t(unsigned long,
 648                                     info->workers.max_workers,
 649                                     info->fs_devices->open_devices);
 650         return 256 * limit;
 651 }
 652
 653 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
 654 {
 655         return atomic_read(&info->nr_async_bios) >
 656                 btrfs_async_submit_limit(info);
 657 }
 658
 659 static void run_one_async_start(struct btrfs_work *work)
 660 {
 661         struct async_submit_bio *async;
 662
 663         async = container_of(work, struct  async_submit_bio, work);
 664         async->submit_bio_start(async->inode, async->rw, async->bio,
 665                                async->mirror_num, async->bio_flags,
 666                                async->bio_offset);
 667 }
 668
 669 static void run_one_async_done(struct btrfs_work *work)
 670 {
 671         struct btrfs_fs_info *fs_info;
 672         struct async_submit_bio *async;
 673         int limit;
 674
 675         async = container_of(work, struct  async_submit_bio, work);
 676         fs_info = BTRFS_I(async->inode)->root->fs_info;
 677
 678         limit = btrfs_async_submit_limit(fs_info);
 679         limit = limit * 2 / 3;
 680
 681         atomic_dec(&fs_info->nr_async_submits);
 682
 683         if (atomic_read(&fs_info->nr_async_submits) < limit &&
 684             waitqueue_active(&fs_info->async_submit_wait))
 685                 wake_up(&fs_info->async_submit_wait);
 686
 687         async->submit_bio_done(async->inode, async->rw, async->bio,
 688                                async->mirror_num, async->bio_flags,
 689                                async->bio_offset);
 690 }
 691
 692 static void run_one_async_free(struct btrfs_work *work)
 693 {
 694         struct async_submit_bio *async;
 695
 696         async = container_of(work, struct  async_submit_bio, work);
 697         kfree(async);
 698 }
 699
 700 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 701                         int rw, struct bio *bio, int mirror_num,
 702                         unsigned long bio_flags,
 703                         u64 bio_offset,
 704                         extent_submit_bio_hook_t *submit_bio_start,
 705                         extent_submit_bio_hook_t *submit_bio_done)
 706 {
 707         struct async_submit_bio *async;
 708
 709         async = kmalloc(sizeof(*async), GFP_NOFS);
 710         if (!async)
 711                 return -ENOMEM;
 712
 713         async->inode = inode;
 714         async->rw = rw;
 715         async->bio = bio;
 716         async->mirror_num = mirror_num;
 717         async->submit_bio_start = submit_bio_start;
 718         async->submit_bio_done = submit_bio_done;
 719
 720         async->work.func = run_one_async_start;
 721         async->work.ordered_func = run_one_async_done;
 722         async->work.ordered_free = run_one_async_free;
 723
 724         async->work.flags = 0;
 725         async->bio_flags = bio_flags;
 726         async->bio_offset = bio_offset;
 727
 728         atomic_inc(&fs_info->nr_async_submits);
 729
 730         if (rw & REQ_SYNC)
 731                 btrfs_set_work_high_prio(&async->work);
 732
 733         btrfs_queue_worker(&fs_info->workers, &async->work);
 734
 735         while (atomic_read(&fs_info->async_submit_draining) &&
 736               atomic_read(&fs_info->nr_async_submits)) {
 737                 wait_event(fs_info->async_submit_wait,
 738                            (atomic_read(&fs_info->nr_async_submits) == 0));
 739         }
 740
 741         return 0;
 742 }
 743
 744 static int btree_csum_one_bio(struct bio *bio)
 745 {
 746         struct bio_vec *bvec = bio->bi_io_vec;
 747         int bio_index = 0;
 748         struct btrfs_root *root;
 749
 750         WARN_ON(bio->bi_vcnt <= 0);
 751         while (bio_index < bio->bi_vcnt) {
 752                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 753                 csum_dirty_buffer(root, bvec->bv_page);
 754                 bio_index++;
 755                 bvec++;
 756         }
 757         return 0;
 758 }
 759
 760 static int __btree_submit_bio_start(struct inode *inode, int rw,
 761                                     struct bio *bio, int mirror_num,
 762                                     unsigned long bio_flags,
 763                                     u64 bio_offset)
 764 {
 765         /*
 766          * when we're called for a write, we're already in the async
 767          * submission context.  Just jump into btrfs_map_bio
 768          */
 769         btree_csum_one_bio(bio);
 770         return 0;
 771 }
 772
 773 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 774                                  int mirror_num, unsigned long bio_flags,
 775                                  u64 bio_offset)
 776 {
 777         /*
 778          * when we're called for a write, we're already in the async
 779          * submission context.  Just jump into btrfs_map_bio
 780          */
 781         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 782 }
 783
 784 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 785                                  int mirror_num, unsigned long bio_flags,
 786                                  u64 bio_offset)
 787 {
 788         int ret;
 789
 790         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 791                                           bio, 1);
 792         BUG_ON(ret);
 793
 794         if (!(rw & REQ_WRITE)) {
 795                 /*
 796                  * called for a read, do the setup so that checksum validation
 797                  * can happen in the async kernel threads
 798                  */
 799                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 800                                      mirror_num, 0);
 801         }
 802
 803         /*
 804          * kthread helpers are used to submit writes so that checksumming
 805          * can happen in parallel across all CPUs
 806          */
 807         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 808                                    inode, rw, bio, mirror_num, 0,
 809                                    bio_offset,
 810                                    __btree_submit_bio_start,
 811                                    __btree_submit_bio_done);
 812 }
 813
 814 #ifdef CONFIG_MIGRATION
 815 static int btree_migratepage(struct address_space *mapping,
 816                         struct page *newpage, struct page *page)
 817 {
 818         /*
 819          * we can't safely write a btree page from here,
 820          * we haven't done the locking hook
 821          */
 822         if (PageDirty(page))
 823                 return -EAGAIN;
 824         /*
 825          * Buffers may be managed in a filesystem specific way.
 826          * We must have no buffers or drop them.
 827          */
 828         if (page_has_private(page) &&
 829             !try_to_release_page(page, GFP_KERNEL))
 830                 return -EAGAIN;
 831         return migrate_page(mapping, newpage, page);
 832 }
 833 #endif
 834
 835 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 836 {
 837         struct extent_io_tree *tree;
 838         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 839         struct extent_buffer *eb;
 840         int was_dirty;
 841
 842         tree = &BTRFS_I(page->mapping->host)->io_tree;
 843         if (!(current->flags & PF_MEMALLOC)) {
 844                 return extent_write_full_page(tree, page,
 845                                               btree_get_extent, wbc);
 846         }
 847
 848         redirty_page_for_writepage(wbc, page);
 849         eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
 850         WARN_ON(!eb);
 851
 852         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
 853         if (!was_dirty) {
 854                 spin_lock(&root->fs_info->delalloc_lock);
 855                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
 856                 spin_unlock(&root->fs_info->delalloc_lock);
 857         }
 858         free_extent_buffer(eb);
 859
 860         unlock_page(page);
 861         return 0;
 862 }
 863
 864 static int btree_writepages(struct address_space *mapping,
 865                             struct writeback_control *wbc)
 866 {
 867         struct extent_io_tree *tree;
 868         tree = &BTRFS_I(mapping->host)->io_tree;
 869         if (wbc->sync_mode == WB_SYNC_NONE) {
 870                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
 871                 u64 num_dirty;
 872                 unsigned long thresh = 32 * 1024 * 1024;
 873
 874                 if (wbc->for_kupdate)
 875                         return 0;
 876
 877                 /* this is a bit racy, but that's ok */
 878                 num_dirty = root->fs_info->dirty_metadata_bytes;
 879                 if (num_dirty < thresh)
 880                         return 0;
 881         }
 882         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 883 }
 884
 885 static int btree_readpage(struct file *file, struct page *page)
 886 {
 887         struct extent_io_tree *tree;
 888         tree = &BTRFS_I(page->mapping->host)->io_tree;
 889         return extent_read_full_page(tree, page, btree_get_extent);
 890 }
 891
 892 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 893 {
 894         struct extent_io_tree *tree;
 895         struct extent_map_tree *map;
 896         int ret;
 897
 898         if (PageWriteback(page) || PageDirty(page))
 899                 return 0;
 900
 901         tree = &BTRFS_I(page->mapping->host)->io_tree;
 902         map = &BTRFS_I(page->mapping->host)->extent_tree;
 903
 904         ret = try_release_extent_state(map, tree, page, gfp_flags);
 905         if (!ret)
 906                 return 0;
 907
 908         ret = try_release_extent_buffer(tree, page);
 909         if (ret == 1) {
 910                 ClearPagePrivate(page);
 911                 set_page_private(page, 0);
 912                 page_cache_release(page);
 913         }
 914
 915         return ret;
 916 }
 917
 918 static void btree_invalidatepage(struct page *page, unsigned long offset)
 919 {
 920         struct extent_io_tree *tree;
 921         tree = &BTRFS_I(page->mapping->host)->io_tree;
 922         extent_invalidatepage(tree, page, offset);
 923         btree_releasepage(page, GFP_NOFS);
 924         if (PagePrivate(page)) {
 925                 printk(KERN_WARNING "btrfs warning page private not zero "
 926                        "on page %llu\n", (unsigned long long)page_offset(page));
 927                 ClearPagePrivate(page);
 928                 set_page_private(page, 0);
 929                 page_cache_release(page);
 930         }
 931 }
 932
 933 static const struct address_space_operations btree_aops = {
 934         .readpage       = btree_readpage,
 935         .writepage      = btree_writepage,
 936         .writepages     = btree_writepages,
 937         .releasepage    = btree_releasepage,
 938         .invalidatepage = btree_invalidatepage,
 939         .sync_page      = block_sync_page,
 940 #ifdef CONFIG_MIGRATION
 941         .migratepage    = btree_migratepage,
 942 #endif
 943 };
 944
 945 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 946                          u64 parent_transid)
 947 {
 948         struct extent_buffer *buf = NULL;
 949         struct inode *btree_inode = root->fs_info->btree_inode;
 950         int ret = 0;
 951
 952         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 953         if (!buf)
 954                 return 0;
 955         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 956                                  buf, 0, 0, btree_get_extent, 0);
 957         free_extent_buffer(buf);
 958         return ret;
 959 }
 960
 961 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 962                                             u64 bytenr, u32 blocksize)
 963 {
 964         struct inode *btree_inode = root->fs_info->btree_inode;
 965         struct extent_buffer *eb;
 966         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 967                                 bytenr, blocksize, GFP_NOFS);
 968         return eb;
 969 }
 970
 971 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 972                                                  u64 bytenr, u32 blocksize)
 973 {
 974         struct inode *btree_inode = root->fs_info->btree_inode;
 975         struct extent_buffer *eb;
 976
 977         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 978                                  bytenr, blocksize, NULL, GFP_NOFS);
 979         return eb;
 980 }
 981
 982
 983 int btrfs_write_tree_block(struct extent_buffer *buf)
 984 {
 985         return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
 986                                         buf->start + buf->len - 1);
 987 }
 988
 989 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 990 {
 991         return filemap_fdatawait_range(buf->first_page->mapping,
 992                                        buf->start, buf->start + buf->len - 1);
 993 }
 994
 995 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 996                                       u32 blocksize, u64 parent_transid)
 997 {
 998         struct extent_buffer *buf = NULL;
 999         int ret;
1000
1001         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1002         if (!buf)
1003                 return NULL;
1004
1005         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1006
1007         if (ret == 0)
1008                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1009         return buf;
1010
1011 }
1012
1013 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1014                      struct extent_buffer *buf)
1015 {
1016         struct inode *btree_inode = root->fs_info->btree_inode;
1017         if (btrfs_header_generation(buf) ==
1018             root->fs_info->running_transaction->transid) {
1019                 btrfs_assert_tree_locked(buf);
1020
1021                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1022                         spin_lock(&root->fs_info->delalloc_lock);
1023                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
1024                                 root->fs_info->dirty_metadata_bytes -= buf->len;
1025                         else
1026                                 WARN_ON(1);
1027                         spin_unlock(&root->fs_info->delalloc_lock);
1028                 }
1029
1030                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1031                 btrfs_set_lock_blocking(buf);
1032                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1033                                           buf);
1034         }
1035         return 0;
1036 }
1037
1038 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1039                         u32 stripesize, struct btrfs_root *root,
1040                         struct btrfs_fs_info *fs_info,
1041                         u64 objectid)
1042 {
1043         root->node = NULL;
1044         root->commit_root = NULL;
1045         root->sectorsize = sectorsize;
1046         root->nodesize = nodesize;
1047         root->leafsize = leafsize;
1048         root->stripesize = stripesize;
1049         root->ref_cows = 0;
1050         root->track_dirty = 0;
1051         root->in_radix = 0;
1052         root->orphan_item_inserted = 0;
1053         root->orphan_cleanup_state = 0;
1054
1055         root->fs_info = fs_info;
1056         root->objectid = objectid;
1057         root->last_trans = 0;
1058         root->highest_objectid = 0;
1059         root->name = NULL;
1060         root->in_sysfs = 0;
1061         root->inode_tree = RB_ROOT;
1062         root->block_rsv = NULL;
1063         root->orphan_block_rsv = NULL;
1064
1065         INIT_LIST_HEAD(&root->dirty_list);
1066         INIT_LIST_HEAD(&root->orphan_list);
1067         INIT_LIST_HEAD(&root->root_list);
1068         spin_lock_init(&root->node_lock);
1069         spin_lock_init(&root->orphan_lock);
1070         spin_lock_init(&root->inode_lock);
1071         spin_lock_init(&root->accounting_lock);
1072         mutex_init(&root->objectid_mutex);
1073         mutex_init(&root->log_mutex);
1074         init_waitqueue_head(&root->log_writer_wait);
1075         init_waitqueue_head(&root->log_commit_wait[0]);
1076         init_waitqueue_head(&root->log_commit_wait[1]);
1077         atomic_set(&root->log_commit[0], 0);
1078         atomic_set(&root->log_commit[1], 0);
1079         atomic_set(&root->log_writers, 0);
1080         root->log_batch = 0;
1081         root->log_transid = 0;
1082         root->last_log_commit = 0;
1083         extent_io_tree_init(&root->dirty_log_pages,
1084                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1085
1086         memset(&root->root_key, 0, sizeof(root->root_key));
1087         memset(&root->root_item, 0, sizeof(root->root_item));
1088         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1089         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1090         root->defrag_trans_start = fs_info->generation;
1091         init_completion(&root->kobj_unregister);
1092         root->defrag_running = 0;
1093         root->root_key.objectid = objectid;
1094         root->anon_super.s_root = NULL;
1095         root->anon_super.s_dev = 0;
1096         INIT_LIST_HEAD(&root->anon_super.s_list);
1097         INIT_LIST_HEAD(&root->anon_super.s_instances);
1098         init_rwsem(&root->anon_super.s_umount);
1099
1100         return 0;
1101 }
1102
1103 static int find_and_setup_root(struct btrfs_root *tree_root,
1104                                struct btrfs_fs_info *fs_info,
1105                                u64 objectid,
1106                                struct btrfs_root *root)
1107 {
1108         int ret;
1109         u32 blocksize;
1110         u64 generation;
1111
1112         __setup_root(tree_root->nodesize, tree_root->leafsize,
1113                      tree_root->sectorsize, tree_root->stripesize,
1114                      root, fs_info, objectid);
1115         ret = btrfs_find_last_root(tree_root, objectid,
1116                                    &root->root_item, &root->root_key);
1117         if (ret > 0)
1118                 return -ENOENT;
1119         BUG_ON(ret);
1120
1121         generation = btrfs_root_generation(&root->root_item);
1122         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1123         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1124                                      blocksize, generation);
1125         if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1126                 free_extent_buffer(root->node);
1127                 return -EIO;
1128         }
1129         root->commit_root = btrfs_root_node(root);
1130         return 0;
1131 }
1132
1133 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1134                                          struct btrfs_fs_info *fs_info)
1135 {
1136         struct btrfs_root *root;
1137         struct btrfs_root *tree_root = fs_info->tree_root;
1138         struct extent_buffer *leaf;
1139
1140         root = kzalloc(sizeof(*root), GFP_NOFS);
1141         if (!root)
1142                 return ERR_PTR(-ENOMEM);
1143
1144         __setup_root(tree_root->nodesize, tree_root->leafsize,
1145                      tree_root->sectorsize, tree_root->stripesize,
1146                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1147
1148         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1149         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1150         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1151         /*
1152          * log trees do not get reference counted because they go away
1153          * before a real commit is actually done.  They do store pointers
1154          * to file data extents, and those reference counts still get
1155          * updated (along with back refs to the log tree).
1156          */
1157         root->ref_cows = 0;
1158
1159         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1160                                       BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1161         if (IS_ERR(leaf)) {
1162                 kfree(root);
1163                 return ERR_CAST(leaf);
1164         }
1165
1166         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1167         btrfs_set_header_bytenr(leaf, leaf->start);
1168         btrfs_set_header_generation(leaf, trans->transid);
1169         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1170         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1171         root->node = leaf;
1172
1173         write_extent_buffer(root->node, root->fs_info->fsid,
1174                             (unsigned long)btrfs_header_fsid(root->node),
1175                             BTRFS_FSID_SIZE);
1176         btrfs_mark_buffer_dirty(root->node);
1177         btrfs_tree_unlock(root->node);
1178         return root;
1179 }
1180
1181 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1182                              struct btrfs_fs_info *fs_info)
1183 {
1184         struct btrfs_root *log_root;
1185
1186         log_root = alloc_log_tree(trans, fs_info);
1187         if (IS_ERR(log_root))
1188                 return PTR_ERR(log_root);
1189         WARN_ON(fs_info->log_root_tree);
1190         fs_info->log_root_tree = log_root;
1191         return 0;
1192 }
1193
1194 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1195                        struct btrfs_root *root)
1196 {
1197         struct btrfs_root *log_root;
1198         struct btrfs_inode_item *inode_item;
1199
1200         log_root = alloc_log_tree(trans, root->fs_info);
1201         if (IS_ERR(log_root))
1202                 return PTR_ERR(log_root);
1203
1204         log_root->last_trans = trans->transid;
1205         log_root->root_key.offset = root->root_key.objectid;
1206
1207         inode_item = &log_root->root_item.inode;
1208         inode_item->generation = cpu_to_le64(1);
1209         inode_item->size = cpu_to_le64(3);
1210         inode_item->nlink = cpu_to_le32(1);
1211         inode_item->nbytes = cpu_to_le64(root->leafsize);
1212         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1213
1214         btrfs_set_root_node(&log_root->root_item, log_root->node);
1215
1216         WARN_ON(root->log_root);
1217         root->log_root = log_root;
1218         root->log_transid = 0;
1219         root->last_log_commit = 0;
1220         return 0;
1221 }
1222
1223 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1224                                                struct btrfs_key *location)
1225 {
1226         struct btrfs_root *root;
1227         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1228         struct btrfs_path *path;
1229         struct extent_buffer *l;
1230         u64 generation;
1231         u32 blocksize;
1232         int ret = 0;
1233
1234         root = kzalloc(sizeof(*root), GFP_NOFS);
1235         if (!root)
1236                 return ERR_PTR(-ENOMEM);
1237         if (location->offset == (u64)-1) {
1238                 ret = find_and_setup_root(tree_root, fs_info,
1239                                           location->objectid, root);
1240                 if (ret) {
1241                         kfree(root);
1242                         return ERR_PTR(ret);
1243                 }
1244                 goto out;
1245         }
1246
1247         __setup_root(tree_root->nodesize, tree_root->leafsize,
1248                      tree_root->sectorsize, tree_root->stripesize,
1249                      root, fs_info, location->objectid);
1250
1251         path = btrfs_alloc_path();
1252         if (!path) {
1253                 kfree(root);
1254                 return ERR_PTR(-ENOMEM);
1255         }
1256         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1257         if (ret == 0) {
1258                 l = path->nodes[0];
1259                 read_extent_buffer(l, &root->root_item,
1260                                 btrfs_item_ptr_offset(l, path->slots[0]),
1261                                 sizeof(root->root_item));
1262                 memcpy(&root->root_key, location, sizeof(*location));
1263         }
1264         btrfs_free_path(path);
1265         if (ret) {
1266                 kfree(root);
1267                 if (ret > 0)
1268                         ret = -ENOENT;
1269                 return ERR_PTR(ret);
1270         }
1271
1272         generation = btrfs_root_generation(&root->root_item);
1273         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1274         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1275                                      blocksize, generation);
1276         root->commit_root = btrfs_root_node(root);
1277         BUG_ON(!root->node);
1278 out:
1279         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1280                 root->ref_cows = 1;
1281                 btrfs_check_and_init_root_item(&root->root_item);
1282         }
1283
1284         return root;
1285 }
1286
1287 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1288                                         u64 root_objectid)
1289 {
1290         struct btrfs_root *root;
1291
1292         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1293                 return fs_info->tree_root;
1294         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1295                 return fs_info->extent_root;
1296
1297         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1298                                  (unsigned long)root_objectid);
1299         return root;
1300 }
1301
1302 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1303                                               struct btrfs_key *location)
1304 {
1305         struct btrfs_root *root;
1306         int ret;
1307
1308         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1309                 return fs_info->tree_root;
1310         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1311                 return fs_info->extent_root;
1312         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1313                 return fs_info->chunk_root;
1314         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1315                 return fs_info->dev_root;
1316         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1317                 return fs_info->csum_root;
1318 again:
1319         spin_lock(&fs_info->fs_roots_radix_lock);
1320         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1321                                  (unsigned long)location->objectid);
1322         spin_unlock(&fs_info->fs_roots_radix_lock);
1323         if (root)
1324                 return root;
1325
1326         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1327         if (IS_ERR(root))
1328                 return root;
1329
1330         set_anon_super(&root->anon_super, NULL);
1331
1332         if (btrfs_root_refs(&root->root_item) == 0) {
1333                 ret = -ENOENT;
1334                 goto fail;
1335         }
1336
1337         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1338         if (ret < 0)
1339                 goto fail;
1340         if (ret == 0)
1341                 root->orphan_item_inserted = 1;
1342
1343         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1344         if (ret)
1345                 goto fail;
1346
1347         spin_lock(&fs_info->fs_roots_radix_lock);
1348         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1349                                 (unsigned long)root->root_key.objectid,
1350                                 root);
1351         if (ret == 0)
1352                 root->in_radix = 1;
1353
1354         spin_unlock(&fs_info->fs_roots_radix_lock);
1355         radix_tree_preload_end();
1356         if (ret) {
1357                 if (ret == -EEXIST) {
1358                         free_fs_root(root);
1359                         goto again;
1360                 }
1361                 goto fail;
1362         }
1363
1364         ret = btrfs_find_dead_roots(fs_info->tree_root,
1365                                     root->root_key.objectid);
1366         WARN_ON(ret);
1367         return root;
1368 fail:
1369         free_fs_root(root);
1370         return ERR_PTR(ret);
1371 }
1372
1373 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1374                                       struct btrfs_key *location,
1375                                       const char *name, int namelen)
1376 {
1377         return btrfs_read_fs_root_no_name(fs_info, location);
1378 #if 0
1379         struct btrfs_root *root;
1380         int ret;
1381
1382         root = btrfs_read_fs_root_no_name(fs_info, location);
1383         if (!root)
1384                 return NULL;
1385
1386         if (root->in_sysfs)
1387                 return root;
1388
1389         ret = btrfs_set_root_name(root, name, namelen);
1390         if (ret) {
1391                 free_extent_buffer(root->node);
1392                 kfree(root);
1393                 return ERR_PTR(ret);
1394         }
1395
1396         ret = btrfs_sysfs_add_root(root);
1397         if (ret) {
1398                 free_extent_buffer(root->node);
1399                 kfree(root->name);
1400                 kfree(root);
1401                 return ERR_PTR(ret);
1402         }
1403         root->in_sysfs = 1;
1404         return root;
1405 #endif
1406 }
1407
1408 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1409 {
1410         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1411         int ret = 0;
1412         struct btrfs_device *device;
1413         struct backing_dev_info *bdi;
1414
1415         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1416                 if (!device->bdev)
1417                         continue;
1418                 bdi = blk_get_backing_dev_info(device->bdev);
1419                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1420                         ret = 1;
1421                         break;
1422                 }
1423         }
1424         return ret;
1425 }
1426
1427 /*
1428  * this unplugs every device on the box, and it is only used when page
1429  * is null
1430  */
1431 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1432 {
1433         struct btrfs_device *device;
1434         struct btrfs_fs_info *info;
1435
1436         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1437         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1438                 if (!device->bdev)
1439                         continue;
1440
1441                 bdi = blk_get_backing_dev_info(device->bdev);
1442                 if (bdi->unplug_io_fn)
1443                         bdi->unplug_io_fn(bdi, page);
1444         }
1445 }
1446
1447 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1448 {
1449         struct inode *inode;
1450         struct extent_map_tree *em_tree;
1451         struct extent_map *em;
1452         struct address_space *mapping;
1453         u64 offset;
1454
1455         /* the generic O_DIRECT read code does this */
1456         if (1 || !page) {
1457                 __unplug_io_fn(bdi, page);
1458                 return;
1459         }
1460
1461         /*
1462          * page->mapping may change at any time.  Get a consistent copy
1463          * and use that for everything below
1464          */
1465         smp_mb();
1466         mapping = page->mapping;
1467         if (!mapping)
1468                 return;
1469
1470         inode = mapping->host;
1471
1472         /*
1473          * don't do the expensive searching for a small number of
1474          * devices
1475          */
1476         if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1477                 __unplug_io_fn(bdi, page);
1478                 return;
1479         }
1480
1481         offset = page_offset(page);
1482
1483         em_tree = &BTRFS_I(inode)->extent_tree;
1484         read_lock(&em_tree->lock);
1485         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1486         read_unlock(&em_tree->lock);
1487         if (!em) {
1488                 __unplug_io_fn(bdi, page);
1489                 return;
1490         }
1491
1492         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1493                 free_extent_map(em);
1494                 __unplug_io_fn(bdi, page);
1495                 return;
1496         }
1497         offset = offset - em->start;
1498         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1499                           em->block_start + offset, page);
1500         free_extent_map(em);
1501 }
1502
1503 /*
1504  * If this fails, caller must call bdi_destroy() to get rid of the
1505  * bdi again.
1506  */
1507 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1508 {
1509         int err;
1510
1511         bdi->capabilities = BDI_CAP_MAP_COPY;
1512         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1513         if (err)
1514                 return err;
1515
1516         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1517         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1518         bdi->unplug_io_data     = info;
1519         bdi->congested_fn       = btrfs_congested_fn;
1520         bdi->congested_data     = info;
1521         return 0;
1522 }
1523
1524 static int bio_ready_for_csum(struct bio *bio)
1525 {
1526         u64 length = 0;
1527         u64 buf_len = 0;
1528         u64 start = 0;
1529         struct page *page;
1530         struct extent_io_tree *io_tree = NULL;
1531         struct bio_vec *bvec;
1532         int i;
1533         int ret;
1534
1535         bio_for_each_segment(bvec, bio, i) {
1536                 page = bvec->bv_page;
1537                 if (page->private == EXTENT_PAGE_PRIVATE) {
1538                         length += bvec->bv_len;
1539                         continue;
1540                 }
1541                 if (!page->private) {
1542                         length += bvec->bv_len;
1543                         continue;
1544                 }
1545                 length = bvec->bv_len;
1546                 buf_len = page->private >> 2;
1547                 start = page_offset(page) + bvec->bv_offset;
1548                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1549         }
1550         /* are we fully contained in this bio? */
1551         if (buf_len <= length)
1552                 return 1;
1553
1554         ret = extent_range_uptodate(io_tree, start + length,
1555                                     start + buf_len - 1);
1556         return ret;
1557 }
1558
1559 /*
1560  * called by the kthread helper functions to finally call the bio end_io
1561  * functions.  This is where read checksum verification actually happens
1562  */
1563 static void end_workqueue_fn(struct btrfs_work *work)
1564 {
1565         struct bio *bio;
1566         struct end_io_wq *end_io_wq;
1567         struct btrfs_fs_info *fs_info;
1568         int error;
1569
1570         end_io_wq = container_of(work, struct end_io_wq, work);
1571         bio = end_io_wq->bio;
1572         fs_info = end_io_wq->info;
1573
1574         /* metadata bio reads are special because the whole tree block must
1575          * be checksummed at once.  This makes sure the entire block is in
1576          * ram and up to date before trying to verify things.  For
1577          * blocksize <= pagesize, it is basically a noop
1578          */
1579         if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1580             !bio_ready_for_csum(bio)) {
1581                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1582                                    &end_io_wq->work);
1583                 return;
1584         }
1585         error = end_io_wq->error;
1586         bio->bi_private = end_io_wq->private;
1587         bio->bi_end_io = end_io_wq->end_io;
1588         kfree(end_io_wq);
1589         bio_endio(bio, error);
1590 }
1591
1592 static int cleaner_kthread(void *arg)
1593 {
1594         struct btrfs_root *root = arg;
1595
1596         do {
1597                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1598
1599                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1600                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1601                         btrfs_run_delayed_iputs(root);
1602                         btrfs_clean_old_snapshots(root);
1603                         mutex_unlock(&root->fs_info->cleaner_mutex);
1604                 }
1605
1606                 if (freezing(current)) {
1607                         refrigerator();
1608                 } else {
1609                         set_current_state(TASK_INTERRUPTIBLE);
1610                         if (!kthread_should_stop())
1611                                 schedule();
1612                         __set_current_state(TASK_RUNNING);
1613                 }
1614         } while (!kthread_should_stop());
1615         return 0;
1616 }
1617
1618 static int transaction_kthread(void *arg)
1619 {
1620         struct btrfs_root *root = arg;
1621         struct btrfs_trans_handle *trans;
1622         struct btrfs_transaction *cur;
1623         u64 transid;
1624         unsigned long now;
1625         unsigned long delay;
1626         int ret;
1627
1628         do {
1629                 delay = HZ * 30;
1630                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1631                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1632
1633                 spin_lock(&root->fs_info->new_trans_lock);
1634                 cur = root->fs_info->running_transaction;
1635                 if (!cur) {
1636                         spin_unlock(&root->fs_info->new_trans_lock);
1637                         goto sleep;
1638                 }
1639
1640                 now = get_seconds();
1641                 if (!cur->blocked &&
1642                     (now < cur->start_time || now - cur->start_time < 30)) {
1643                         spin_unlock(&root->fs_info->new_trans_lock);
1644                         delay = HZ * 5;
1645                         goto sleep;
1646                 }
1647                 transid = cur->transid;
1648                 spin_unlock(&root->fs_info->new_trans_lock);
1649
1650                 trans = btrfs_join_transaction(root, 1);
1651                 BUG_ON(IS_ERR(trans));
1652                 if (transid == trans->transid) {
1653                         ret = btrfs_commit_transaction(trans, root);
1654                         BUG_ON(ret);
1655                 } else {
1656                         btrfs_end_transaction(trans, root);
1657                 }
1658 sleep:
1659                 wake_up_process(root->fs_info->cleaner_kthread);
1660                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1661
1662                 if (freezing(current)) {
1663                         refrigerator();
1664                 } else {
1665                         set_current_state(TASK_INTERRUPTIBLE);
1666                         if (!kthread_should_stop() &&
1667                             !btrfs_transaction_blocked(root->fs_info))
1668                                 schedule_timeout(delay);
1669                         __set_current_state(TASK_RUNNING);
1670                 }
1671         } while (!kthread_should_stop());
1672         return 0;
1673 }
1674
1675 struct btrfs_root *open_ctree(struct super_block *sb,
1676                               struct btrfs_fs_devices *fs_devices,
1677                               char *options)
1678 {
1679         u32 sectorsize;
1680         u32 nodesize;
1681         u32 leafsize;
1682         u32 blocksize;
1683         u32 stripesize;
1684         u64 generation;
1685         u64 features;
1686         struct btrfs_key location;
1687         struct buffer_head *bh;
1688         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1689                                                  GFP_NOFS);
1690         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1691                                                  GFP_NOFS);
1692         struct btrfs_root *tree_root = btrfs_sb(sb);
1693         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1694         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1695                                                 GFP_NOFS);
1696         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1697                                               GFP_NOFS);
1698         struct btrfs_root *log_tree_root;
1699
1700         int ret;
1701         int err = -EINVAL;
1702
1703         struct btrfs_super_block *disk_super;
1704
1705         if (!extent_root || !tree_root || !fs_info ||
1706             !chunk_root || !dev_root || !csum_root) {
1707                 err = -ENOMEM;
1708                 goto fail;
1709         }
1710
1711         ret = init_srcu_struct(&fs_info->subvol_srcu);
1712         if (ret) {
1713                 err = ret;
1714                 goto fail;
1715         }
1716
1717         ret = setup_bdi(fs_info, &fs_info->bdi);
1718         if (ret) {
1719                 err = ret;
1720                 goto fail_srcu;
1721         }
1722
1723         fs_info->btree_inode = new_inode(sb);
1724         if (!fs_info->btree_inode) {
1725                 err = -ENOMEM;
1726                 goto fail_bdi;
1727         }
1728
1729         fs_info->btree_inode->i_mapping->flags &= ~__GFP_FS;
1730
1731         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1732         INIT_LIST_HEAD(&fs_info->trans_list);
1733         INIT_LIST_HEAD(&fs_info->dead_roots);
1734         INIT_LIST_HEAD(&fs_info->delayed_iputs);
1735         INIT_LIST_HEAD(&fs_info->hashers);
1736         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1737         INIT_LIST_HEAD(&fs_info->ordered_operations);
1738         INIT_LIST_HEAD(&fs_info->caching_block_groups);
1739         spin_lock_init(&fs_info->delalloc_lock);
1740         spin_lock_init(&fs_info->new_trans_lock);
1741         spin_lock_init(&fs_info->ref_cache_lock);
1742         spin_lock_init(&fs_info->fs_roots_radix_lock);
1743         spin_lock_init(&fs_info->delayed_iput_lock);
1744
1745         init_completion(&fs_info->kobj_unregister);
1746         fs_info->tree_root = tree_root;
1747         fs_info->extent_root = extent_root;
1748         fs_info->csum_root = csum_root;
1749         fs_info->chunk_root = chunk_root;
1750         fs_info->dev_root = dev_root;
1751         fs_info->fs_devices = fs_devices;
1752         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1753         INIT_LIST_HEAD(&fs_info->space_info);
1754         btrfs_mapping_init(&fs_info->mapping_tree);
1755         btrfs_init_block_rsv(&fs_info->global_block_rsv);
1756         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1757         btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1758         btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1759         btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1760         INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1761         mutex_init(&fs_info->durable_block_rsv_mutex);
1762         atomic_set(&fs_info->nr_async_submits, 0);
1763         atomic_set(&fs_info->async_delalloc_pages, 0);
1764         atomic_set(&fs_info->async_submit_draining, 0);
1765         atomic_set(&fs_info->nr_async_bios, 0);
1766         fs_info->sb = sb;
1767         fs_info->max_inline = 8192 * 1024;
1768         fs_info->metadata_ratio = 0;
1769
1770         fs_info->thread_pool_size = min_t(unsigned long,
1771                                           num_online_cpus() + 2, 8);
1772
1773         INIT_LIST_HEAD(&fs_info->ordered_extents);
1774         spin_lock_init(&fs_info->ordered_extent_lock);
1775
1776         mutex_init(&fs_info->scrub_lock);
1777         atomic_set(&fs_info->scrubs_running, 0);
1778         atomic_set(&fs_info->scrub_pause_req, 0);
1779         atomic_set(&fs_info->scrubs_paused, 0);
1780         atomic_set(&fs_info->scrub_cancel_req, 0);
1781         init_waitqueue_head(&fs_info->scrub_pause_wait);
1782         init_rwsem(&fs_info->scrub_super_lock);
1783         fs_info->scrub_workers_refcnt = 0;
1784         btrfs_init_workers(&fs_info->scrub_workers, "scrub",
1785                            fs_info->thread_pool_size, &fs_info->generic_worker);
1786
1787         sb->s_blocksize = 4096;
1788         sb->s_blocksize_bits = blksize_bits(4096);
1789         sb->s_bdi = &fs_info->bdi;
1790
1791         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1792         fs_info->btree_inode->i_nlink = 1;
1793         /*
1794          * we set the i_size on the btree inode to the max possible int.
1795          * the real end of the address space is determined by all of
1796          * the devices in the system
1797          */
1798         fs_info->btree_inode->i_size = OFFSET_MAX;
1799         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1800         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1801
1802         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1803         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1804                              fs_info->btree_inode->i_mapping,
1805                              GFP_NOFS);
1806         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1807                              GFP_NOFS);
1808
1809         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1810
1811         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1812         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1813                sizeof(struct btrfs_key));
1814         BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1815         insert_inode_hash(fs_info->btree_inode);
1816
1817         spin_lock_init(&fs_info->block_group_cache_lock);
1818         fs_info->block_group_cache_tree = RB_ROOT;
1819
1820         extent_io_tree_init(&fs_info->freed_extents[0],
1821                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1822         extent_io_tree_init(&fs_info->freed_extents[1],
1823                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1824         fs_info->pinned_extents = &fs_info->freed_extents[0];
1825         fs_info->do_barriers = 1;
1826
1827
1828         mutex_init(&fs_info->trans_mutex);
1829         mutex_init(&fs_info->ordered_operations_mutex);
1830         mutex_init(&fs_info->tree_log_mutex);
1831         mutex_init(&fs_info->chunk_mutex);
1832         mutex_init(&fs_info->transaction_kthread_mutex);
1833         mutex_init(&fs_info->cleaner_mutex);
1834         mutex_init(&fs_info->volume_mutex);
1835         init_rwsem(&fs_info->extent_commit_sem);
1836         init_rwsem(&fs_info->cleanup_work_sem);
1837         init_rwsem(&fs_info->subvol_sem);
1838
1839         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1840         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1841
1842         init_waitqueue_head(&fs_info->transaction_throttle);
1843         init_waitqueue_head(&fs_info->transaction_wait);
1844         init_waitqueue_head(&fs_info->transaction_blocked_wait);
1845         init_waitqueue_head(&fs_info->async_submit_wait);
1846
1847         __setup_root(4096, 4096, 4096, 4096, tree_root,
1848                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1849
1850         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1851         if (!bh) {
1852                 err = -EINVAL;
1853                 goto fail_iput;
1854         }
1855
1856         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1857         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1858                sizeof(fs_info->super_for_commit));
1859         brelse(bh);
1860
1861         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1862
1863         disk_super = &fs_info->super_copy;
1864         if (!btrfs_super_root(disk_super))
1865                 goto fail_iput;
1866
1867         /* check FS state, whether FS is broken. */
1868         fs_info->fs_state |= btrfs_super_flags(disk_super);
1869
1870         btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
1871
1872         /*
1873          * In the long term, we'll store the compression type in the super
1874          * block, and it'll be used for per file compression control.
1875          */
1876         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
1877
1878         ret = btrfs_parse_options(tree_root, options);
1879         if (ret) {
1880                 err = ret;
1881                 goto fail_iput;
1882         }
1883
1884         features = btrfs_super_incompat_flags(disk_super) &
1885                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1886         if (features) {
1887                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1888                        "unsupported optional features (%Lx).\n",
1889                        (unsigned long long)features);
1890                 err = -EINVAL;
1891                 goto fail_iput;
1892         }
1893
1894         features = btrfs_super_incompat_flags(disk_super);
1895         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1896         if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1897                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1898         btrfs_set_super_incompat_flags(disk_super, features);
1899
1900         features = btrfs_super_compat_ro_flags(disk_super) &
1901                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1902         if (!(sb->s_flags & MS_RDONLY) && features) {
1903                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1904                        "unsupported option features (%Lx).\n",
1905                        (unsigned long long)features);
1906                 err = -EINVAL;
1907                 goto fail_iput;
1908         }
1909
1910         btrfs_init_workers(&fs_info->generic_worker,
1911                            "genwork", 1, NULL);
1912
1913         btrfs_init_workers(&fs_info->workers, "worker",
1914                            fs_info->thread_pool_size,
1915                            &fs_info->generic_worker);
1916
1917         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1918                            fs_info->thread_pool_size,
1919                            &fs_info->generic_worker);
1920
1921         btrfs_init_workers(&fs_info->submit_workers, "submit",
1922                            min_t(u64, fs_devices->num_devices,
1923                            fs_info->thread_pool_size),
1924                            &fs_info->generic_worker);
1925
1926         /* a higher idle thresh on the submit workers makes it much more
1927          * likely that bios will be send down in a sane order to the
1928          * devices
1929          */
1930         fs_info->submit_workers.idle_thresh = 64;
1931
1932         fs_info->workers.idle_thresh = 16;
1933         fs_info->workers.ordered = 1;
1934
1935         fs_info->delalloc_workers.idle_thresh = 2;
1936         fs_info->delalloc_workers.ordered = 1;
1937
1938         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1939                            &fs_info->generic_worker);
1940         btrfs_init_workers(&fs_info->endio_workers, "endio",
1941                            fs_info->thread_pool_size,
1942                            &fs_info->generic_worker);
1943         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1944                            fs_info->thread_pool_size,
1945                            &fs_info->generic_worker);
1946         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1947                            "endio-meta-write", fs_info->thread_pool_size,
1948                            &fs_info->generic_worker);
1949         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1950                            fs_info->thread_pool_size,
1951                            &fs_info->generic_worker);
1952         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1953                            1, &fs_info->generic_worker);
1954
1955         /*
1956          * endios are largely parallel and should have a very
1957          * low idle thresh
1958          */
1959         fs_info->endio_workers.idle_thresh = 4;
1960         fs_info->endio_meta_workers.idle_thresh = 4;
1961
1962         fs_info->endio_write_workers.idle_thresh = 2;
1963         fs_info->endio_meta_write_workers.idle_thresh = 2;
1964
1965         btrfs_start_workers(&fs_info->workers, 1);
1966         btrfs_start_workers(&fs_info->generic_worker, 1);
1967         btrfs_start_workers(&fs_info->submit_workers, 1);
1968         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1969         btrfs_start_workers(&fs_info->fixup_workers, 1);
1970         btrfs_start_workers(&fs_info->endio_workers, 1);
1971         btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1972         btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1973         btrfs_start_workers(&fs_info->endio_write_workers, 1);
1974         btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1975
1976         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1977         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1978                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1979
1980         nodesize = btrfs_super_nodesize(disk_super);
1981         leafsize = btrfs_super_leafsize(disk_super);
1982         sectorsize = btrfs_super_sectorsize(disk_super);
1983         stripesize = btrfs_super_stripesize(disk_super);
1984         tree_root->nodesize = nodesize;
1985         tree_root->leafsize = leafsize;
1986         tree_root->sectorsize = sectorsize;
1987         tree_root->stripesize = stripesize;
1988
1989         sb->s_blocksize = sectorsize;
1990         sb->s_blocksize_bits = blksize_bits(sectorsize);
1991
1992         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1993                     sizeof(disk_super->magic))) {
1994                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1995                 goto fail_sb_buffer;
1996         }
1997
1998         mutex_lock(&fs_info->chunk_mutex);
1999         ret = btrfs_read_sys_array(tree_root);
2000         mutex_unlock(&fs_info->chunk_mutex);
2001         if (ret) {
2002                 printk(KERN_WARNING "btrfs: failed to read the system "
2003                        "array on %s\n", sb->s_id);
2004                 goto fail_sb_buffer;
2005         }
2006
2007         blocksize = btrfs_level_size(tree_root,
2008                                      btrfs_super_chunk_root_level(disk_super));
2009         generation = btrfs_super_chunk_root_generation(disk_super);
2010
2011         __setup_root(nodesize, leafsize, sectorsize, stripesize,
2012                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2013
2014         chunk_root->node = read_tree_block(chunk_root,
2015                                            btrfs_super_chunk_root(disk_super),
2016                                            blocksize, generation);
2017         BUG_ON(!chunk_root->node);
2018         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2019                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2020                        sb->s_id);
2021                 goto fail_chunk_root;
2022         }
2023         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2024         chunk_root->commit_root = btrfs_root_node(chunk_root);
2025
2026         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2027            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2028            BTRFS_UUID_SIZE);
2029
2030         mutex_lock(&fs_info->chunk_mutex);
2031         ret = btrfs_read_chunk_tree(chunk_root);
2032         mutex_unlock(&fs_info->chunk_mutex);
2033         if (ret) {
2034                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2035                        sb->s_id);
2036                 goto fail_chunk_root;
2037         }
2038
2039         btrfs_close_extra_devices(fs_devices);
2040
2041         blocksize = btrfs_level_size(tree_root,
2042                                      btrfs_super_root_level(disk_super));
2043         generation = btrfs_super_generation(disk_super);
2044
2045         tree_root->node = read_tree_block(tree_root,
2046                                           btrfs_super_root(disk_super),
2047                                           blocksize, generation);
2048         if (!tree_root->node)
2049                 goto fail_chunk_root;
2050         if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2051                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2052                        sb->s_id);
2053                 goto fail_tree_root;
2054         }
2055         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2056         tree_root->commit_root = btrfs_root_node(tree_root);
2057
2058         ret = find_and_setup_root(tree_root, fs_info,
2059                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2060         if (ret)
2061                 goto fail_tree_root;
2062         extent_root->track_dirty = 1;
2063
2064         ret = find_and_setup_root(tree_root, fs_info,
2065                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
2066         if (ret)
2067                 goto fail_extent_root;
2068         dev_root->track_dirty = 1;
2069
2070         ret = find_and_setup_root(tree_root, fs_info,
2071                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
2072         if (ret)
2073                 goto fail_dev_root;
2074
2075         csum_root->track_dirty = 1;
2076
2077         fs_info->generation = generation;
2078         fs_info->last_trans_committed = generation;
2079         fs_info->data_alloc_profile = (u64)-1;
2080         fs_info->metadata_alloc_profile = (u64)-1;
2081         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
2082
2083         ret = btrfs_init_space_info(fs_info);
2084         if (ret) {
2085                 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2086                 goto fail_block_groups;
2087         }
2088
2089         ret = btrfs_read_block_groups(extent_root);
2090         if (ret) {
2091                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2092                 goto fail_block_groups;
2093         }
2094
2095         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2096                                                "btrfs-cleaner");
2097         if (IS_ERR(fs_info->cleaner_kthread))
2098                 goto fail_block_groups;
2099
2100         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2101                                                    tree_root,
2102                                                    "btrfs-transaction");
2103         if (IS_ERR(fs_info->transaction_kthread))
2104                 goto fail_cleaner;
2105
2106         if (!btrfs_test_opt(tree_root, SSD) &&
2107             !btrfs_test_opt(tree_root, NOSSD) &&
2108             !fs_info->fs_devices->rotating) {
2109                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2110                        "mode\n");
2111                 btrfs_set_opt(fs_info->mount_opt, SSD);
2112         }
2113
2114         /* do not make disk changes in broken FS */
2115         if (btrfs_super_log_root(disk_super) != 0 &&
2116             !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2117                 u64 bytenr = btrfs_super_log_root(disk_super);
2118
2119                 if (fs_devices->rw_devices == 0) {
2120                         printk(KERN_WARNING "Btrfs log replay required "
2121                                "on RO media\n");
2122                         err = -EIO;
2123                         goto fail_trans_kthread;
2124                 }
2125                 blocksize =
2126                      btrfs_level_size(tree_root,
2127                                       btrfs_super_log_root_level(disk_super));
2128
2129                 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2130                 if (!log_tree_root) {
2131                         err = -ENOMEM;
2132                         goto fail_trans_kthread;
2133                 }
2134
2135                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2136                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2137
2138                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2139                                                       blocksize,
2140                                                       generation + 1);
2141                 ret = btrfs_recover_log_trees(log_tree_root);
2142                 BUG_ON(ret);
2143
2144                 if (sb->s_flags & MS_RDONLY) {
2145                         ret =  btrfs_commit_super(tree_root);
2146                         BUG_ON(ret);
2147                 }
2148         }
2149
2150         ret = btrfs_find_orphan_roots(tree_root);
2151         BUG_ON(ret);
2152
2153         if (!(sb->s_flags & MS_RDONLY)) {
2154                 ret = btrfs_cleanup_fs_roots(fs_info);
2155                 BUG_ON(ret);
2156
2157                 ret = btrfs_recover_relocation(tree_root);
2158                 if (ret < 0) {
2159                         printk(KERN_WARNING
2160                                "btrfs: failed to recover relocation\n");
2161                         err = -EINVAL;
2162                         goto fail_trans_kthread;
2163                 }
2164         }
2165
2166         location.objectid = BTRFS_FS_TREE_OBJECTID;
2167         location.type = BTRFS_ROOT_ITEM_KEY;
2168         location.offset = (u64)-1;
2169
2170         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2171         if (!fs_info->fs_root)
2172                 goto fail_trans_kthread;
2173         if (IS_ERR(fs_info->fs_root)) {
2174                 err = PTR_ERR(fs_info->fs_root);
2175                 goto fail_trans_kthread;
2176         }
2177
2178         if (!(sb->s_flags & MS_RDONLY)) {
2179                 down_read(&fs_info->cleanup_work_sem);
2180                 err = btrfs_orphan_cleanup(fs_info->fs_root);
2181                 if (!err)
2182                         err = btrfs_orphan_cleanup(fs_info->tree_root);
2183                 up_read(&fs_info->cleanup_work_sem);
2184                 if (err) {
2185                         close_ctree(tree_root);
2186                         return ERR_PTR(err);
2187                 }
2188         }
2189
2190         return tree_root;
2191
2192 fail_trans_kthread:
2193         kthread_stop(fs_info->transaction_kthread);
2194 fail_cleaner:
2195         kthread_stop(fs_info->cleaner_kthread);
2196
2197         /*
2198          * make sure we're done with the btree inode before we stop our
2199          * kthreads
2200          */
2201         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2202         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2203
2204 fail_block_groups:
2205         btrfs_free_block_groups(fs_info);
2206         free_extent_buffer(csum_root->node);
2207         free_extent_buffer(csum_root->commit_root);
2208 fail_dev_root:
2209         free_extent_buffer(dev_root->node);
2210         free_extent_buffer(dev_root->commit_root);
2211 fail_extent_root:
2212         free_extent_buffer(extent_root->node);
2213         free_extent_buffer(extent_root->commit_root);
2214 fail_tree_root:
2215         free_extent_buffer(tree_root->node);
2216         free_extent_buffer(tree_root->commit_root);
2217 fail_chunk_root:
2218         free_extent_buffer(chunk_root->node);
2219         free_extent_buffer(chunk_root->commit_root);
2220 fail_sb_buffer:
2221         btrfs_stop_workers(&fs_info->generic_worker);
2222         btrfs_stop_workers(&fs_info->fixup_workers);
2223         btrfs_stop_workers(&fs_info->delalloc_workers);
2224         btrfs_stop_workers(&fs_info->workers);
2225         btrfs_stop_workers(&fs_info->endio_workers);
2226         btrfs_stop_workers(&fs_info->endio_meta_workers);
2227         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2228         btrfs_stop_workers(&fs_info->endio_write_workers);
2229         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2230         btrfs_stop_workers(&fs_info->submit_workers);
2231 fail_iput:
2232         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2233         iput(fs_info->btree_inode);
2234
2235         btrfs_close_devices(fs_info->fs_devices);
2236         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2237 fail_bdi:
2238         bdi_destroy(&fs_info->bdi);
2239 fail_srcu:
2240         cleanup_srcu_struct(&fs_info->subvol_srcu);
2241 fail:
2242         kfree(extent_root);
2243         kfree(tree_root);
2244         kfree(fs_info);
2245         kfree(chunk_root);
2246         kfree(dev_root);
2247         kfree(csum_root);
2248         return ERR_PTR(err);
2249 }
2250
2251 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2252 {
2253         char b[BDEVNAME_SIZE];
2254
2255         if (uptodate) {
2256                 set_buffer_uptodate(bh);
2257         } else {
2258                 if (printk_ratelimit()) {
2259                         printk(KERN_WARNING "lost page write due to "
2260                                         "I/O error on %s\n",
2261                                        bdevname(bh->b_bdev, b));
2262                 }
2263                 /* note, we dont' set_buffer_write_io_error because we have
2264                  * our own ways of dealing with the IO errors
2265                  */
2266                 clear_buffer_uptodate(bh);
2267         }
2268         unlock_buffer(bh);
2269         put_bh(bh);
2270 }
2271
2272 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2273 {
2274         struct buffer_head *bh;
2275         struct buffer_head *latest = NULL;
2276         struct btrfs_super_block *super;
2277         int i;
2278         u64 transid = 0;
2279         u64 bytenr;
2280
2281         /* we would like to check all the supers, but that would make
2282          * a btrfs mount succeed after a mkfs from a different FS.
2283          * So, we need to add a special mount option to scan for
2284          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2285          */
2286         for (i = 0; i < 1; i++) {
2287                 bytenr = btrfs_sb_offset(i);
2288                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2289                         break;
2290                 bh = __bread(bdev, bytenr / 4096, 4096);
2291                 if (!bh)
2292                         continue;
2293
2294                 super = (struct btrfs_super_block *)bh->b_data;
2295                 if (btrfs_super_bytenr(super) != bytenr ||
2296                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2297                             sizeof(super->magic))) {
2298                         brelse(bh);
2299                         continue;
2300                 }
2301
2302                 if (!latest || btrfs_super_generation(super) > transid) {
2303                         brelse(latest);
2304                         latest = bh;
2305                         transid = btrfs_super_generation(super);
2306                 } else {
2307                         brelse(bh);
2308                 }
2309         }
2310         return latest;
2311 }
2312
2313 /*
2314  * this should be called twice, once with wait == 0 and
2315  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2316  * we write are pinned.
2317  *
2318  * They are released when wait == 1 is done.
2319  * max_mirrors must be the same for both runs, and it indicates how
2320  * many supers on this one device should be written.
2321  *
2322  * max_mirrors == 0 means to write them all.
2323  */
2324 static int write_dev_supers(struct btrfs_device *device,
2325                             struct btrfs_super_block *sb,
2326                             int do_barriers, int wait, int max_mirrors)
2327 {
2328         struct buffer_head *bh;
2329         int i;
2330         int ret;
2331         int errors = 0;
2332         u32 crc;
2333         u64 bytenr;
2334         int last_barrier = 0;
2335
2336         if (max_mirrors == 0)
2337                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2338
2339         /* make sure only the last submit_bh does a barrier */
2340         if (do_barriers) {
2341                 for (i = 0; i < max_mirrors; i++) {
2342                         bytenr = btrfs_sb_offset(i);
2343                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2344                             device->total_bytes)
2345                                 break;
2346                         last_barrier = i;
2347                 }
2348         }
2349
2350         for (i = 0; i < max_mirrors; i++) {
2351                 bytenr = btrfs_sb_offset(i);
2352                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2353                         break;
2354
2355                 if (wait) {
2356                         bh = __find_get_block(device->bdev, bytenr / 4096,
2357                                               BTRFS_SUPER_INFO_SIZE);
2358                         BUG_ON(!bh);
2359                         wait_on_buffer(bh);
2360                         if (!buffer_uptodate(bh))
2361                                 errors++;
2362
2363                         /* drop our reference */
2364                         brelse(bh);
2365
2366                         /* drop the reference from the wait == 0 run */
2367                         brelse(bh);
2368                         continue;
2369                 } else {
2370                         btrfs_set_super_bytenr(sb, bytenr);
2371
2372                         crc = ~(u32)0;
2373                         crc = btrfs_csum_data(NULL, (char *)sb +
2374                                               BTRFS_CSUM_SIZE, crc,
2375                                               BTRFS_SUPER_INFO_SIZE -
2376                                               BTRFS_CSUM_SIZE);
2377                         btrfs_csum_final(crc, sb->csum);
2378
2379                         /*
2380                          * one reference for us, and we leave it for the
2381                          * caller
2382                          */
2383                         bh = __getblk(device->bdev, bytenr / 4096,
2384                                       BTRFS_SUPER_INFO_SIZE);
2385                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2386
2387                         /* one reference for submit_bh */
2388                         get_bh(bh);
2389
2390                         set_buffer_uptodate(bh);
2391                         lock_buffer(bh);
2392                         bh->b_end_io = btrfs_end_buffer_write_sync;
2393                 }
2394
2395                 if (i == last_barrier && do_barriers)
2396                         ret = submit_bh(WRITE_FLUSH_FUA, bh);
2397                 else
2398                         ret = submit_bh(WRITE_SYNC, bh);
2399
2400                 if (ret)
2401                         errors++;
2402         }
2403         return errors < i ? 0 : -1;
2404 }
2405
2406 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2407 {
2408         struct list_head *head;
2409         struct btrfs_device *dev;
2410         struct btrfs_super_block *sb;
2411         struct btrfs_dev_item *dev_item;
2412         int ret;
2413         int do_barriers;
2414         int max_errors;
2415         int total_errors = 0;
2416         u64 flags;
2417
2418         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2419         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2420
2421         sb = &root->fs_info->super_for_commit;
2422         dev_item = &sb->dev_item;
2423
2424         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2425         head = &root->fs_info->fs_devices->devices;
2426         list_for_each_entry(dev, head, dev_list) {
2427                 if (!dev->bdev) {
2428                         total_errors++;
2429                         continue;
2430                 }
2431                 if (!dev->in_fs_metadata || !dev->writeable)
2432                         continue;
2433
2434                 btrfs_set_stack_device_generation(dev_item, 0);
2435                 btrfs_set_stack_device_type(dev_item, dev->type);
2436                 btrfs_set_stack_device_id(dev_item, dev->devid);
2437                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2438                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2439                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2440                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2441                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2442                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2443                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2444
2445                 flags = btrfs_super_flags(sb);
2446                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2447
2448                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2449                 if (ret)
2450                         total_errors++;
2451         }
2452         if (total_errors > max_errors) {
2453                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2454                        total_errors);
2455                 BUG();
2456         }
2457
2458         total_errors = 0;
2459         list_for_each_entry(dev, head, dev_list) {
2460                 if (!dev->bdev)
2461                         continue;
2462                 if (!dev->in_fs_metadata || !dev->writeable)
2463                         continue;
2464
2465                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2466                 if (ret)
2467                         total_errors++;
2468         }
2469         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2470         if (total_errors > max_errors) {
2471                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2472                        total_errors);
2473                 BUG();
2474         }
2475         return 0;
2476 }
2477
2478 int write_ctree_super(struct btrfs_trans_handle *trans,
2479                       struct btrfs_root *root, int max_mirrors)
2480 {
2481         int ret;
2482
2483         ret = write_all_supers(root, max_mirrors);
2484         return ret;
2485 }
2486
2487 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2488 {
2489         spin_lock(&fs_info->fs_roots_radix_lock);
2490         radix_tree_delete(&fs_info->fs_roots_radix,
2491                           (unsigned long)root->root_key.objectid);
2492         spin_unlock(&fs_info->fs_roots_radix_lock);
2493
2494         if (btrfs_root_refs(&root->root_item) == 0)
2495                 synchronize_srcu(&fs_info->subvol_srcu);
2496
2497         free_fs_root(root);
2498         return 0;
2499 }
2500
2501 static void free_fs_root(struct btrfs_root *root)
2502 {
2503         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2504         if (root->anon_super.s_dev) {
2505                 down_write(&root->anon_super.s_umount);
2506                 kill_anon_super(&root->anon_super);
2507         }
2508         free_extent_buffer(root->node);
2509         free_extent_buffer(root->commit_root);
2510         kfree(root->name);
2511         kfree(root);
2512 }
2513
2514 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2515 {
2516         int ret;
2517         struct btrfs_root *gang[8];
2518         int i;
2519
2520         while (!list_empty(&fs_info->dead_roots)) {
2521                 gang[0] = list_entry(fs_info->dead_roots.next,
2522                                      struct btrfs_root, root_list);
2523                 list_del(&gang[0]->root_list);
2524
2525                 if (gang[0]->in_radix) {
2526                         btrfs_free_fs_root(fs_info, gang[0]);
2527                 } else {
2528                         free_extent_buffer(gang[0]->node);
2529                         free_extent_buffer(gang[0]->commit_root);
2530                         kfree(gang[0]);
2531                 }
2532         }
2533
2534         while (1) {
2535                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2536                                              (void **)gang, 0,
2537                                              ARRAY_SIZE(gang));
2538                 if (!ret)
2539                         break;
2540                 for (i = 0; i < ret; i++)
2541                         btrfs_free_fs_root(fs_info, gang[i]);
2542         }
2543         return 0;
2544 }
2545
2546 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2547 {
2548         u64 root_objectid = 0;
2549         struct btrfs_root *gang[8];
2550         int i;
2551         int ret;
2552
2553         while (1) {
2554                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2555                                              (void **)gang, root_objectid,
2556                                              ARRAY_SIZE(gang));
2557                 if (!ret)
2558                         break;
2559
2560                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2561                 for (i = 0; i < ret; i++) {
2562                         int err;
2563
2564                         root_objectid = gang[i]->root_key.objectid;
2565                         err = btrfs_orphan_cleanup(gang[i]);
2566                         if (err)
2567                                 return err;
2568                 }
2569                 root_objectid++;
2570         }
2571         return 0;
2572 }
2573
2574 int btrfs_commit_super(struct btrfs_root *root)
2575 {
2576         struct btrfs_trans_handle *trans;
2577         int ret;
2578
2579         mutex_lock(&root->fs_info->cleaner_mutex);
2580         btrfs_run_delayed_iputs(root);
2581         btrfs_clean_old_snapshots(root);
2582         mutex_unlock(&root->fs_info->cleaner_mutex);
2583
2584         /* wait until ongoing cleanup work done */
2585         down_write(&root->fs_info->cleanup_work_sem);
2586         up_write(&root->fs_info->cleanup_work_sem);
2587
2588         trans = btrfs_join_transaction(root, 1);
2589         if (IS_ERR(trans))
2590                 return PTR_ERR(trans);
2591         ret = btrfs_commit_transaction(trans, root);
2592         BUG_ON(ret);
2593         /* run commit again to drop the original snapshot */
2594         trans = btrfs_join_transaction(root, 1);
2595         if (IS_ERR(trans))
2596                 return PTR_ERR(trans);
2597         btrfs_commit_transaction(trans, root);
2598         ret = btrfs_write_and_wait_transaction(NULL, root);
2599         BUG_ON(ret);
2600
2601         ret = write_ctree_super(NULL, root, 0);
2602         return ret;
2603 }
2604
2605 int close_ctree(struct btrfs_root *root)
2606 {
2607         struct btrfs_fs_info *fs_info = root->fs_info;
2608         int ret;
2609
2610         fs_info->closing = 1;
2611         smp_mb();
2612
2613         btrfs_scrub_cancel(root);
2614         btrfs_put_block_group_cache(fs_info);
2615
2616         /*
2617          * Here come 2 situations when btrfs is broken to flip readonly:
2618          *
2619          * 1. when btrfs flips readonly somewhere else before
2620          * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2621          * and btrfs will skip to write sb directly to keep
2622          * ERROR state on disk.
2623          *
2624          * 2. when btrfs flips readonly just in btrfs_commit_super,
2625          * and in such case, btrfs cannnot write sb via btrfs_commit_super,
2626          * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2627          * btrfs will cleanup all FS resources first and write sb then.
2628          */
2629         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2630                 ret = btrfs_commit_super(root);
2631                 if (ret)
2632                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2633         }
2634
2635         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2636                 ret = btrfs_error_commit_super(root);
2637                 if (ret)
2638                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2639         }
2640
2641         kthread_stop(root->fs_info->transaction_kthread);
2642         kthread_stop(root->fs_info->cleaner_kthread);
2643
2644         fs_info->closing = 2;
2645         smp_mb();
2646
2647         if (fs_info->delalloc_bytes) {
2648                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2649                        (unsigned long long)fs_info->delalloc_bytes);
2650         }
2651         if (fs_info->total_ref_cache_size) {
2652                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2653                        (unsigned long long)fs_info->total_ref_cache_size);
2654         }
2655
2656         free_extent_buffer(fs_info->extent_root->node);
2657         free_extent_buffer(fs_info->extent_root->commit_root);
2658         free_extent_buffer(fs_info->tree_root->node);
2659         free_extent_buffer(fs_info->tree_root->commit_root);
2660         free_extent_buffer(root->fs_info->chunk_root->node);
2661         free_extent_buffer(root->fs_info->chunk_root->commit_root);
2662         free_extent_buffer(root->fs_info->dev_root->node);
2663         free_extent_buffer(root->fs_info->dev_root->commit_root);
2664         free_extent_buffer(root->fs_info->csum_root->node);
2665         free_extent_buffer(root->fs_info->csum_root->commit_root);
2666
2667         btrfs_free_block_groups(root->fs_info);
2668
2669         del_fs_roots(fs_info);
2670
2671         iput(fs_info->btree_inode);
2672
2673         btrfs_stop_workers(&fs_info->generic_worker);
2674         btrfs_stop_workers(&fs_info->fixup_workers);
2675         btrfs_stop_workers(&fs_info->delalloc_workers);
2676         btrfs_stop_workers(&fs_info->workers);
2677         btrfs_stop_workers(&fs_info->endio_workers);
2678         btrfs_stop_workers(&fs_info->endio_meta_workers);
2679         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2680         btrfs_stop_workers(&fs_info->endio_write_workers);
2681         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2682         btrfs_stop_workers(&fs_info->submit_workers);
2683
2684         btrfs_close_devices(fs_info->fs_devices);
2685         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2686
2687         bdi_destroy(&fs_info->bdi);
2688         cleanup_srcu_struct(&fs_info->subvol_srcu);
2689
2690         kfree(fs_info->extent_root);
2691         kfree(fs_info->tree_root);
2692         kfree(fs_info->chunk_root);
2693         kfree(fs_info->dev_root);
2694         kfree(fs_info->csum_root);
2695         kfree(fs_info);
2696
2697         return 0;
2698 }
2699
2700 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2701 {
2702         int ret;
2703         struct inode *btree_inode = buf->first_page->mapping->host;
2704
2705         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2706                                      NULL);
2707         if (!ret)
2708                 return ret;
2709
2710         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2711                                     parent_transid);
2712         return !ret;
2713 }
2714
2715 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2716 {
2717         struct inode *btree_inode = buf->first_page->mapping->host;
2718         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2719                                           buf);
2720 }
2721
2722 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2723 {
2724         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2725         u64 transid = btrfs_header_generation(buf);
2726         struct inode *btree_inode = root->fs_info->btree_inode;
2727         int was_dirty;
2728
2729         btrfs_assert_tree_locked(buf);
2730         if (transid != root->fs_info->generation) {
2731                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2732                        "found %llu running %llu\n",
2733                         (unsigned long long)buf->start,
2734                         (unsigned long long)transid,
2735                         (unsigned long long)root->fs_info->generation);
2736                 WARN_ON(1);
2737         }
2738         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2739                                             buf);
2740         if (!was_dirty) {
2741                 spin_lock(&root->fs_info->delalloc_lock);
2742                 root->fs_info->dirty_metadata_bytes += buf->len;
2743                 spin_unlock(&root->fs_info->delalloc_lock);
2744         }
2745 }
2746
2747 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2748 {
2749         /*
2750          * looks as though older kernels can get into trouble with
2751          * this code, they end up stuck in balance_dirty_pages forever
2752          */
2753         u64 num_dirty;
2754         unsigned long thresh = 32 * 1024 * 1024;
2755
2756         if (current->flags & PF_MEMALLOC)
2757                 return;
2758
2759         num_dirty = root->fs_info->dirty_metadata_bytes;
2760
2761         if (num_dirty > thresh) {
2762                 balance_dirty_pages_ratelimited_nr(
2763                                    root->fs_info->btree_inode->i_mapping, 1);
2764         }
2765         return;
2766 }
2767
2768 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2769 {
2770         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2771         int ret;
2772         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2773         if (ret == 0)
2774                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2775         return ret;
2776 }
2777
2778 int btree_lock_page_hook(struct page *page)
2779 {
2780         struct inode *inode = page->mapping->host;
2781         struct btrfs_root *root = BTRFS_I(inode)->root;
2782         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2783         struct extent_buffer *eb;
2784         unsigned long len;
2785         u64 bytenr = page_offset(page);
2786
2787         if (page->private == EXTENT_PAGE_PRIVATE)
2788                 goto out;
2789
2790         len = page->private >> 2;
2791         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2792         if (!eb)
2793                 goto out;
2794
2795         btrfs_tree_lock(eb);
2796         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2797
2798         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2799                 spin_lock(&root->fs_info->delalloc_lock);
2800                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2801                         root->fs_info->dirty_metadata_bytes -= eb->len;
2802                 else
2803                         WARN_ON(1);
2804                 spin_unlock(&root->fs_info->delalloc_lock);
2805         }
2806
2807         btrfs_tree_unlock(eb);
2808         free_extent_buffer(eb);
2809 out:
2810         lock_page(page);
2811         return 0;
2812 }
2813
2814 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
2815                               int read_only)
2816 {
2817         if (read_only)
2818                 return;
2819
2820         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
2821                 printk(KERN_WARNING "warning: mount fs with errors, "
2822                        "running btrfsck is recommended\n");
2823 }
2824
2825 int btrfs_error_commit_super(struct btrfs_root *root)
2826 {
2827         int ret;
2828
2829         mutex_lock(&root->fs_info->cleaner_mutex);
2830         btrfs_run_delayed_iputs(root);
2831         mutex_unlock(&root->fs_info->cleaner_mutex);
2832
2833         down_write(&root->fs_info->cleanup_work_sem);
2834         up_write(&root->fs_info->cleanup_work_sem);
2835
2836         /* cleanup FS via transaction */
2837         btrfs_cleanup_transaction(root);
2838
2839         ret = write_ctree_super(NULL, root, 0);
2840
2841         return ret;
2842 }
2843
2844 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
2845 {
2846         struct btrfs_inode *btrfs_inode;
2847         struct list_head splice;
2848
2849         INIT_LIST_HEAD(&splice);
2850
2851         mutex_lock(&root->fs_info->ordered_operations_mutex);
2852         spin_lock(&root->fs_info->ordered_extent_lock);
2853
2854         list_splice_init(&root->fs_info->ordered_operations, &splice);
2855         while (!list_empty(&splice)) {
2856                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2857                                          ordered_operations);
2858
2859                 list_del_init(&btrfs_inode->ordered_operations);
2860
2861                 btrfs_invalidate_inodes(btrfs_inode->root);
2862         }
2863
2864         spin_unlock(&root->fs_info->ordered_extent_lock);
2865         mutex_unlock(&root->fs_info->ordered_operations_mutex);
2866
2867         return 0;
2868 }
2869
2870 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
2871 {
2872         struct list_head splice;
2873         struct btrfs_ordered_extent *ordered;
2874         struct inode *inode;
2875
2876         INIT_LIST_HEAD(&splice);
2877
2878         spin_lock(&root->fs_info->ordered_extent_lock);
2879
2880         list_splice_init(&root->fs_info->ordered_extents, &splice);
2881         while (!list_empty(&splice)) {
2882                 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
2883                                      root_extent_list);
2884
2885                 list_del_init(&ordered->root_extent_list);
2886                 atomic_inc(&ordered->refs);
2887
2888                 /* the inode may be getting freed (in sys_unlink path). */
2889                 inode = igrab(ordered->inode);
2890
2891                 spin_unlock(&root->fs_info->ordered_extent_lock);
2892                 if (inode)
2893                         iput(inode);
2894
2895                 atomic_set(&ordered->refs, 1);
2896                 btrfs_put_ordered_extent(ordered);
2897
2898                 spin_lock(&root->fs_info->ordered_extent_lock);
2899         }
2900
2901         spin_unlock(&root->fs_info->ordered_extent_lock);
2902
2903         return 0;
2904 }
2905
2906 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
2907                                       struct btrfs_root *root)
2908 {
2909         struct rb_node *node;
2910         struct btrfs_delayed_ref_root *delayed_refs;
2911         struct btrfs_delayed_ref_node *ref;
2912         int ret = 0;
2913
2914         delayed_refs = &trans->delayed_refs;
2915
2916         spin_lock(&delayed_refs->lock);
2917         if (delayed_refs->num_entries == 0) {
2918                 spin_unlock(&delayed_refs->lock);
2919                 printk(KERN_INFO "delayed_refs has NO entry\n");
2920                 return ret;
2921         }
2922
2923         node = rb_first(&delayed_refs->root);
2924         while (node) {
2925                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2926                 node = rb_next(node);
2927
2928                 ref->in_tree = 0;
2929                 rb_erase(&ref->rb_node, &delayed_refs->root);
2930                 delayed_refs->num_entries--;
2931
2932                 atomic_set(&ref->refs, 1);
2933                 if (btrfs_delayed_ref_is_head(ref)) {
2934                         struct btrfs_delayed_ref_head *head;
2935
2936                         head = btrfs_delayed_node_to_head(ref);
2937                         mutex_lock(&head->mutex);
2938                         kfree(head->extent_op);
2939                         delayed_refs->num_heads--;
2940                         if (list_empty(&head->cluster))
2941                                 delayed_refs->num_heads_ready--;
2942                         list_del_init(&head->cluster);
2943                         mutex_unlock(&head->mutex);
2944                 }
2945
2946                 spin_unlock(&delayed_refs->lock);
2947                 btrfs_put_delayed_ref(ref);
2948
2949                 cond_resched();
2950                 spin_lock(&delayed_refs->lock);
2951         }
2952
2953         spin_unlock(&delayed_refs->lock);
2954
2955         return ret;
2956 }
2957
2958 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
2959 {
2960         struct btrfs_pending_snapshot *snapshot;
2961         struct list_head splice;
2962
2963         INIT_LIST_HEAD(&splice);
2964
2965         list_splice_init(&t->pending_snapshots, &splice);
2966
2967         while (!list_empty(&splice)) {
2968                 snapshot = list_entry(splice.next,
2969                                       struct btrfs_pending_snapshot,
2970                                       list);
2971
2972                 list_del_init(&snapshot->list);
2973
2974                 kfree(snapshot);
2975         }
2976
2977         return 0;
2978 }
2979
2980 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
2981 {
2982         struct btrfs_inode *btrfs_inode;
2983         struct list_head splice;
2984
2985         INIT_LIST_HEAD(&splice);
2986
2987         list_splice_init(&root->fs_info->delalloc_inodes, &splice);
2988
2989         spin_lock(&root->fs_info->delalloc_lock);
2990
2991         while (!list_empty(&splice)) {
2992                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2993                                     delalloc_inodes);
2994
2995                 list_del_init(&btrfs_inode->delalloc_inodes);
2996
2997                 btrfs_invalidate_inodes(btrfs_inode->root);
2998         }
2999
3000         spin_unlock(&root->fs_info->delalloc_lock);
3001
3002         return 0;
3003 }
3004
3005 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3006                                         struct extent_io_tree *dirty_pages,
3007                                         int mark)
3008 {
3009         int ret;
3010         struct page *page;
3011         struct inode *btree_inode = root->fs_info->btree_inode;
3012         struct extent_buffer *eb;
3013         u64 start = 0;
3014         u64 end;
3015         u64 offset;
3016         unsigned long index;
3017
3018         while (1) {
3019                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3020                                             mark);
3021                 if (ret)
3022                         break;
3023
3024                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3025                 while (start <= end) {
3026                         index = start >> PAGE_CACHE_SHIFT;
3027                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3028                         page = find_get_page(btree_inode->i_mapping, index);
3029                         if (!page)
3030                                 continue;
3031                         offset = page_offset(page);
3032
3033                         spin_lock(&dirty_pages->buffer_lock);
3034                         eb = radix_tree_lookup(
3035                              &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3036                                                offset >> PAGE_CACHE_SHIFT);
3037                         spin_unlock(&dirty_pages->buffer_lock);
3038                         if (eb) {
3039                                 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3040                                                          &eb->bflags);
3041                                 atomic_set(&eb->refs, 1);
3042                         }
3043                         if (PageWriteback(page))
3044                                 end_page_writeback(page);
3045
3046                         lock_page(page);
3047                         if (PageDirty(page)) {
3048                                 clear_page_dirty_for_io(page);
3049                                 spin_lock_irq(&page->mapping->tree_lock);
3050                                 radix_tree_tag_clear(&page->mapping->page_tree,
3051                                                         page_index(page),
3052                                                         PAGECACHE_TAG_DIRTY);
3053                                 spin_unlock_irq(&page->mapping->tree_lock);
3054                         }
3055
3056                         page->mapping->a_ops->invalidatepage(page, 0);
3057                         unlock_page(page);
3058                 }
3059         }
3060
3061         return ret;
3062 }
3063
3064 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3065                                        struct extent_io_tree *pinned_extents)
3066 {
3067         struct extent_io_tree *unpin;
3068         u64 start;
3069         u64 end;
3070         int ret;
3071
3072         unpin = pinned_extents;
3073         while (1) {
3074                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3075                                             EXTENT_DIRTY);
3076                 if (ret)
3077                         break;
3078
3079                 /* opt_discard */
3080                 if (btrfs_test_opt(root, DISCARD))
3081                         ret = btrfs_error_discard_extent(root, start,
3082                                                          end + 1 - start,
3083                                                          NULL);
3084
3085                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3086                 btrfs_error_unpin_extent_range(root, start, end);
3087                 cond_resched();
3088         }
3089
3090         return 0;
3091 }
3092
3093 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3094 {
3095         struct btrfs_transaction *t;
3096         LIST_HEAD(list);
3097
3098         WARN_ON(1);
3099
3100         mutex_lock(&root->fs_info->trans_mutex);
3101         mutex_lock(&root->fs_info->transaction_kthread_mutex);
3102
3103         list_splice_init(&root->fs_info->trans_list, &list);
3104         while (!list_empty(&list)) {
3105                 t = list_entry(list.next, struct btrfs_transaction, list);
3106                 if (!t)
3107                         break;
3108
3109                 btrfs_destroy_ordered_operations(root);
3110
3111                 btrfs_destroy_ordered_extents(root);
3112
3113                 btrfs_destroy_delayed_refs(t, root);
3114
3115                 btrfs_block_rsv_release(root,
3116                                         &root->fs_info->trans_block_rsv,
3117                                         t->dirty_pages.dirty_bytes);
3118
3119                 /* FIXME: cleanup wait for commit */
3120                 t->in_commit = 1;
3121                 t->blocked = 1;
3122                 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3123                         wake_up(&root->fs_info->transaction_blocked_wait);
3124
3125                 t->blocked = 0;
3126                 if (waitqueue_active(&root->fs_info->transaction_wait))
3127                         wake_up(&root->fs_info->transaction_wait);
3128                 mutex_unlock(&root->fs_info->trans_mutex);
3129
3130                 mutex_lock(&root->fs_info->trans_mutex);
3131                 t->commit_done = 1;
3132                 if (waitqueue_active(&t->commit_wait))
3133                         wake_up(&t->commit_wait);
3134                 mutex_unlock(&root->fs_info->trans_mutex);
3135
3136                 mutex_lock(&root->fs_info->trans_mutex);
3137
3138                 btrfs_destroy_pending_snapshots(t);
3139
3140                 btrfs_destroy_delalloc_inodes(root);
3141
3142                 spin_lock(&root->fs_info->new_trans_lock);
3143                 root->fs_info->running_transaction = NULL;
3144                 spin_unlock(&root->fs_info->new_trans_lock);
3145
3146                 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3147                                              EXTENT_DIRTY);
3148
3149                 btrfs_destroy_pinned_extent(root,
3150                                             root->fs_info->pinned_extents);
3151
3152                 atomic_set(&t->use_count, 0);
3153                 list_del_init(&t->list);
3154                 memset(t, 0, sizeof(*t));
3155                 kmem_cache_free(btrfs_transaction_cachep, t);
3156         }
3157
3158         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3159         mutex_unlock(&root->fs_info->trans_mutex);
3160
3161         return 0;
3162 }
3163
3164 static struct extent_io_ops btree_extent_io_ops = {
3165         .write_cache_pages_lock_hook = btree_lock_page_hook,
3166         .readpage_end_io_hook = btree_readpage_end_io_hook,
3167         .submit_bio_hook = btree_submit_bio_hook,
3168         /* note we're sharing with inode.c for the merge bio hook */
3169         .merge_bio_hook = btrfs_merge_bio_hook,
3170 };