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 "compat.h"
  33 #include "ctree.h"
  34 #include "disk-io.h"
  35 #include "transaction.h"
  36 #include "btrfs_inode.h"
  37 #include "volumes.h"
  38 #include "print-tree.h"
  39 #include "async-thread.h"
  40 #include "locking.h"
  41 #include "tree-log.h"
  42 #include "free-space-cache.h"
  43
  44 static struct extent_io_ops btree_extent_io_ops;
  45 static void end_workqueue_fn(struct btrfs_work *work);
  46 static void free_fs_root(struct btrfs_root *root);
  47 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  48                                     int read_only);
  49 static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
  50 static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
  51 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  52                                       struct btrfs_root *root);
  53 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
  54 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  55 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
  56                                         struct extent_io_tree *dirty_pages,
  57                                         int mark);
  58 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
  59                                        struct extent_io_tree *pinned_extents);
  60 static int btrfs_cleanup_transaction(struct btrfs_root *root);
  61
  62 /*
  63  * end_io_wq structs are used to do processing in task context when an IO is
  64  * complete.  This is used during reads to verify checksums, and it is used
  65  * by writes to insert metadata for new file extents after IO is complete.
  66  */
  67 struct end_io_wq {
  68         struct bio *bio;
  69         bio_end_io_t *end_io;
  70         void *private;
  71         struct btrfs_fs_info *info;
  72         int error;
  73         int metadata;
  74         struct list_head list;
  75         struct btrfs_work work;
  76 };
  77
  78 /*
  79  * async submit bios are used to offload expensive checksumming
  80  * onto the worker threads.  They checksum file and metadata bios
  81  * just before they are sent down the IO stack.
  82  */
  83 struct async_submit_bio {
  84         struct inode *inode;
  85         struct bio *bio;
  86         struct list_head list;
  87         extent_submit_bio_hook_t *submit_bio_start;
  88         extent_submit_bio_hook_t *submit_bio_done;
  89         int rw;
  90         int mirror_num;
  91         unsigned long bio_flags;
  92         /*
  93          * bio_offset is optional, can be used if the pages in the bio
  94          * can't tell us where in the file the bio should go
  95          */
  96         u64 bio_offset;
  97         struct btrfs_work work;
  98 };
  99
 100 /* These are used to set the lockdep class on the extent buffer locks.
 101  * The class is set by the readpage_end_io_hook after the buffer has
 102  * passed csum validation but before the pages are unlocked.
 103  *
 104  * The lockdep class is also set by btrfs_init_new_buffer on freshly
 105  * allocated blocks.
 106  *
 107  * The class is based on the level in the tree block, which allows lockdep
 108  * to know that lower nodes nest inside the locks of higher nodes.
 109  *
 110  * We also add a check to make sure the highest level of the tree is
 111  * the same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this
 112  * code needs update as well.
 113  */
 114 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 115 # if BTRFS_MAX_LEVEL != 8
 116 #  error
 117 # endif
 118 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
 119 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
 120         /* leaf */
 121         "btrfs-extent-00",
 122         "btrfs-extent-01",
 123         "btrfs-extent-02",
 124         "btrfs-extent-03",
 125         "btrfs-extent-04",
 126         "btrfs-extent-05",
 127         "btrfs-extent-06",
 128         "btrfs-extent-07",
 129         /* highest possible level */
 130         "btrfs-extent-08",
 131 };
 132 #endif
 133
 134 /*
 135  * extents on the btree inode are pretty simple, there's one extent
 136  * that covers the entire device
 137  */
 138 static struct extent_map *btree_get_extent(struct inode *inode,
 139                 struct page *page, size_t page_offset, u64 start, u64 len,
 140                 int create)
 141 {
 142         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 143         struct extent_map *em;
 144         int ret;
 145
 146         read_lock(&em_tree->lock);
 147         em = lookup_extent_mapping(em_tree, start, len);
 148         if (em) {
 149                 em->bdev =
 150                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 151                 read_unlock(&em_tree->lock);
 152                 goto out;
 153         }
 154         read_unlock(&em_tree->lock);
 155
 156         em = alloc_extent_map(GFP_NOFS);
 157         if (!em) {
 158                 em = ERR_PTR(-ENOMEM);
 159                 goto out;
 160         }
 161         em->start = 0;
 162         em->len = (u64)-1;
 163         em->block_len = (u64)-1;
 164         em->block_start = 0;
 165         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 166
 167         write_lock(&em_tree->lock);
 168         ret = add_extent_mapping(em_tree, em);
 169         if (ret == -EEXIST) {
 170                 u64 failed_start = em->start;
 171                 u64 failed_len = em->len;
 172
 173                 free_extent_map(em);
 174                 em = lookup_extent_mapping(em_tree, start, len);
 175                 if (em) {
 176                         ret = 0;
 177                 } else {
 178                         em = lookup_extent_mapping(em_tree, failed_start,
 179                                                    failed_len);
 180                         ret = -EIO;
 181                 }
 182         } else if (ret) {
 183                 free_extent_map(em);
 184                 em = NULL;
 185         }
 186         write_unlock(&em_tree->lock);
 187
 188         if (ret)
 189                 em = ERR_PTR(ret);
 190 out:
 191         return em;
 192 }
 193
 194 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 195 {
 196         return crc32c(seed, data, len);
 197 }
 198
 199 void btrfs_csum_final(u32 crc, char *result)
 200 {
 201         *(__le32 *)result = ~cpu_to_le32(crc);
 202 }
 203
 204 /*
 205  * compute the csum for a btree block, and either verify it or write it
 206  * into the csum field of the block.
 207  */
 208 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 209                            int verify)
 210 {
 211         u16 csum_size =
 212                 btrfs_super_csum_size(&root->fs_info->super_copy);
 213         char *result = NULL;
 214         unsigned long len;
 215         unsigned long cur_len;
 216         unsigned long offset = BTRFS_CSUM_SIZE;
 217         char *map_token = NULL;
 218         char *kaddr;
 219         unsigned long map_start;
 220         unsigned long map_len;
 221         int err;
 222         u32 crc = ~(u32)0;
 223         unsigned long inline_result;
 224
 225         len = buf->len - offset;
 226         while (len > 0) {
 227                 err = map_private_extent_buffer(buf, offset, 32,
 228                                         &map_token, &kaddr,
 229                                         &map_start, &map_len, KM_USER0);
 230                 if (err)
 231                         return 1;
 232                 cur_len = min(len, map_len - (offset - map_start));
 233                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 234                                       crc, cur_len);
 235                 len -= cur_len;
 236                 offset += cur_len;
 237                 unmap_extent_buffer(buf, map_token, KM_USER0);
 238         }
 239         if (csum_size > sizeof(inline_result)) {
 240                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
 241                 if (!result)
 242                         return 1;
 243         } else {
 244                 result = (char *)&inline_result;
 245         }
 246
 247         btrfs_csum_final(crc, result);
 248
 249         if (verify) {
 250                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 251                         u32 val;
 252                         u32 found = 0;
 253                         memcpy(&found, result, csum_size);
 254
 255                         read_extent_buffer(buf, &val, 0, csum_size);
 256                         if (printk_ratelimit()) {
 257                                 printk(KERN_INFO "btrfs: %s checksum verify "
 258                                        "failed on %llu wanted %X found %X "
 259                                        "level %d\n",
 260                                        root->fs_info->sb->s_id,
 261                                        (unsigned long long)buf->start, val, found,
 262                                        btrfs_header_level(buf));
 263                         }
 264                         if (result != (char *)&inline_result)
 265                                 kfree(result);
 266                         return 1;
 267                 }
 268         } else {
 269                 write_extent_buffer(buf, result, 0, csum_size);
 270         }
 271         if (result != (char *)&inline_result)
 272                 kfree(result);
 273         return 0;
 274 }
 275
 276 /*
 277  * we can't consider a given block up to date unless the transid of the
 278  * block matches the transid in the parent node's pointer.  This is how we
 279  * detect blocks that either didn't get written at all or got written
 280  * in the wrong place.
 281  */
 282 static int verify_parent_transid(struct extent_io_tree *io_tree,
 283                                  struct extent_buffer *eb, u64 parent_transid)
 284 {
 285         struct extent_state *cached_state = NULL;
 286         int ret;
 287
 288         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 289                 return 0;
 290
 291         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 292                          0, &cached_state, GFP_NOFS);
 293         if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
 294             btrfs_header_generation(eb) == parent_transid) {
 295                 ret = 0;
 296                 goto out;
 297         }
 298         if (printk_ratelimit()) {
 299                 printk("parent transid verify failed on %llu wanted %llu "
 300                        "found %llu\n",
 301                        (unsigned long long)eb->start,
 302                        (unsigned long long)parent_transid,
 303                        (unsigned long long)btrfs_header_generation(eb));
 304         }
 305         ret = 1;
 306         clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
 307 out:
 308         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 309                              &cached_state, GFP_NOFS);
 310         return ret;
 311 }
 312
 313 /*
 314  * helper to read a given tree block, doing retries as required when
 315  * the checksums don't match and we have alternate mirrors to try.
 316  */
 317 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 318                                           struct extent_buffer *eb,
 319                                           u64 start, u64 parent_transid)
 320 {
 321         struct extent_io_tree *io_tree;
 322         int ret;
 323         int num_copies = 0;
 324         int mirror_num = 0;
 325
 326         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 327         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 328         while (1) {
 329                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 330                                                btree_get_extent, mirror_num);
 331                 if (!ret &&
 332                     !verify_parent_transid(io_tree, eb, parent_transid))
 333                         return ret;
 334
 335                 /*
 336                  * This buffer's crc is fine, but its contents are corrupted, so
 337                  * there is no reason to read the other copies, they won't be
 338                  * any less wrong.
 339                  */
 340                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 341                         return ret;
 342
 343                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 344                                               eb->start, eb->len);
 345                 if (num_copies == 1)
 346                         return ret;
 347
 348                 mirror_num++;
 349                 if (mirror_num > num_copies)
 350                         return ret;
 351         }
 352         return -EIO;
 353 }
 354
 355 /*
 356  * checksum a dirty tree block before IO.  This has extra checks to make sure
 357  * we only fill in the checksum field in the first page of a multi-page block
 358  */
 359
 360 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 361 {
 362         struct extent_io_tree *tree;
 363         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 364         u64 found_start;
 365         unsigned long len;
 366         struct extent_buffer *eb;
 367         int ret;
 368
 369         tree = &BTRFS_I(page->mapping->host)->io_tree;
 370
 371         if (page->private == EXTENT_PAGE_PRIVATE) {
 372                 WARN_ON(1);
 373                 goto out;
 374         }
 375         if (!page->private) {
 376                 WARN_ON(1);
 377                 goto out;
 378         }
 379         len = page->private >> 2;
 380         WARN_ON(len == 0);
 381
 382         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 383         if (eb == NULL) {
 384                 WARN_ON(1);
 385                 goto out;
 386         }
 387         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 388                                              btrfs_header_generation(eb));
 389         BUG_ON(ret);
 390         WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
 391
 392         found_start = btrfs_header_bytenr(eb);
 393         if (found_start != start) {
 394                 WARN_ON(1);
 395                 goto err;
 396         }
 397         if (eb->first_page != page) {
 398                 WARN_ON(1);
 399                 goto err;
 400         }
 401         if (!PageUptodate(page)) {
 402                 WARN_ON(1);
 403                 goto err;
 404         }
 405         csum_tree_block(root, eb, 0);
 406 err:
 407         free_extent_buffer(eb);
 408 out:
 409         return 0;
 410 }
 411
 412 static int check_tree_block_fsid(struct btrfs_root *root,
 413                                  struct extent_buffer *eb)
 414 {
 415         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 416         u8 fsid[BTRFS_UUID_SIZE];
 417         int ret = 1;
 418
 419         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
 420                            BTRFS_FSID_SIZE);
 421         while (fs_devices) {
 422                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 423                         ret = 0;
 424                         break;
 425                 }
 426                 fs_devices = fs_devices->seed;
 427         }
 428         return ret;
 429 }
 430
 431 #define CORRUPT(reason, eb, root, slot)                         \
 432         printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
 433                "root=%llu, slot=%d\n", reason,                  \
 434                (unsigned long long)btrfs_header_bytenr(eb),     \
 435                (unsigned long long)root->objectid, slot)
 436
 437 static noinline int check_leaf(struct btrfs_root *root,
 438                                struct extent_buffer *leaf)
 439 {
 440         struct btrfs_key key;
 441         struct btrfs_key leaf_key;
 442         u32 nritems = btrfs_header_nritems(leaf);
 443         int slot;
 444
 445         if (nritems == 0)
 446                 return 0;
 447
 448         /* Check the 0 item */
 449         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 450             BTRFS_LEAF_DATA_SIZE(root)) {
 451                 CORRUPT("invalid item offset size pair", leaf, root, 0);
 452                 return -EIO;
 453         }
 454
 455         /*
 456          * Check to make sure each items keys are in the correct order and their
 457          * offsets make sense.  We only have to loop through nritems-1 because
 458          * we check the current slot against the next slot, which verifies the
 459          * next slot's offset+size makes sense and that the current's slot
 460          * offset is correct.
 461          */
 462         for (slot = 0; slot < nritems - 1; slot++) {
 463                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 464                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 465
 466                 /* Make sure the keys are in the right order */
 467                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 468                         CORRUPT("bad key order", leaf, root, slot);
 469                         return -EIO;
 470                 }
 471
 472                 /*
 473                  * Make sure the offset and ends are right, remember that the
 474                  * item data starts at the end of the leaf and grows towards the
 475                  * front.
 476                  */
 477                 if (btrfs_item_offset_nr(leaf, slot) !=
 478                         btrfs_item_end_nr(leaf, slot + 1)) {
 479                         CORRUPT("slot offset bad", leaf, root, slot);
 480                         return -EIO;
 481                 }
 482
 483                 /*
 484                  * Check to make sure that we don't point outside of the leaf,
 485                  * just incase all the items are consistent to eachother, but
 486                  * all point outside of the leaf.
 487                  */
 488                 if (btrfs_item_end_nr(leaf, slot) >
 489                     BTRFS_LEAF_DATA_SIZE(root)) {
 490                         CORRUPT("slot end outside of leaf", leaf, root, slot);
 491                         return -EIO;
 492                 }
 493         }
 494
 495         return 0;
 496 }
 497
 498 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 499 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
 500 {
 501         lockdep_set_class_and_name(&eb->lock,
 502                            &btrfs_eb_class[level],
 503                            btrfs_eb_name[level]);
 504 }
 505 #endif
 506
 507 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 508                                struct extent_state *state)
 509 {
 510         struct extent_io_tree *tree;
 511         u64 found_start;
 512         int found_level;
 513         unsigned long len;
 514         struct extent_buffer *eb;
 515         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 516         int ret = 0;
 517
 518         tree = &BTRFS_I(page->mapping->host)->io_tree;
 519         if (page->private == EXTENT_PAGE_PRIVATE)
 520                 goto out;
 521         if (!page->private)
 522                 goto out;
 523
 524         len = page->private >> 2;
 525         WARN_ON(len == 0);
 526
 527         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 528         if (eb == NULL) {
 529                 ret = -EIO;
 530                 goto out;
 531         }
 532
 533         found_start = btrfs_header_bytenr(eb);
 534         if (found_start != start) {
 535                 if (printk_ratelimit()) {
 536                         printk(KERN_INFO "btrfs bad tree block start "
 537                                "%llu %llu\n",
 538                                (unsigned long long)found_start,
 539                                (unsigned long long)eb->start);
 540                 }
 541                 ret = -EIO;
 542                 goto err;
 543         }
 544         if (eb->first_page != page) {
 545                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
 546                        eb->first_page->index, page->index);
 547                 WARN_ON(1);
 548                 ret = -EIO;
 549                 goto err;
 550         }
 551         if (check_tree_block_fsid(root, eb)) {
 552                 if (printk_ratelimit()) {
 553                         printk(KERN_INFO "btrfs bad fsid on block %llu\n",
 554                                (unsigned long long)eb->start);
 555                 }
 556                 ret = -EIO;
 557                 goto err;
 558         }
 559         found_level = btrfs_header_level(eb);
 560
 561         btrfs_set_buffer_lockdep_class(eb, found_level);
 562
 563         ret = csum_tree_block(root, eb, 1);
 564         if (ret) {
 565                 ret = -EIO;
 566                 goto err;
 567         }
 568
 569         /*
 570          * If this is a leaf block and it is corrupt, set the corrupt bit so
 571          * that we don't try and read the other copies of this block, just
 572          * return -EIO.
 573          */
 574         if (found_level == 0 && check_leaf(root, eb)) {
 575                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 576                 ret = -EIO;
 577         }
 578
 579         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 580         end = eb->start + end - 1;
 581 err:
 582         free_extent_buffer(eb);
 583 out:
 584         return ret;
 585 }
 586
 587 static void end_workqueue_bio(struct bio *bio, int err)
 588 {
 589         struct end_io_wq *end_io_wq = bio->bi_private;
 590         struct btrfs_fs_info *fs_info;
 591
 592         fs_info = end_io_wq->info;
 593         end_io_wq->error = err;
 594         end_io_wq->work.func = end_workqueue_fn;
 595         end_io_wq->work.flags = 0;
 596
 597         if (bio->bi_rw & REQ_WRITE) {
 598                 if (end_io_wq->metadata == 1)
 599                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
 600                                            &end_io_wq->work);
 601                 else if (end_io_wq->metadata == 2)
 602                         btrfs_queue_worker(&fs_info->endio_freespace_worker,
 603                                            &end_io_wq->work);
 604                 else
 605                         btrfs_queue_worker(&fs_info->endio_write_workers,
 606                                            &end_io_wq->work);
 607         } else {
 608                 if (end_io_wq->metadata)
 609                         btrfs_queue_worker(&fs_info->endio_meta_workers,
 610                                            &end_io_wq->work);
 611                 else
 612                         btrfs_queue_worker(&fs_info->endio_workers,
 613                                            &end_io_wq->work);
 614         }
 615 }
 616
 617 /*
 618  * For the metadata arg you want
 619  *
 620  * 0 - if data
 621  * 1 - if normal metadta
 622  * 2 - if writing to the free space cache area
 623  */
 624 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 625                         int metadata)
 626 {
 627         struct end_io_wq *end_io_wq;
 628         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 629         if (!end_io_wq)
 630                 return -ENOMEM;
 631
 632         end_io_wq->private = bio->bi_private;
 633         end_io_wq->end_io = bio->bi_end_io;
 634         end_io_wq->info = info;
 635         end_io_wq->error = 0;
 636         end_io_wq->bio = bio;
 637         end_io_wq->metadata = metadata;
 638
 639         bio->bi_private = end_io_wq;
 640         bio->bi_end_io = end_workqueue_bio;
 641         return 0;
 642 }
 643
 644 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 645 {
 646         unsigned long limit = min_t(unsigned long,
 647                                     info->workers.max_workers,
 648                                     info->fs_devices->open_devices);
 649         return 256 * limit;
 650 }
 651
 652 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
 653 {
 654         return atomic_read(&info->nr_async_bios) >
 655                 btrfs_async_submit_limit(info);
 656 }
 657
 658 static void run_one_async_start(struct btrfs_work *work)
 659 {
 660         struct async_submit_bio *async;
 661
 662         async = container_of(work, struct  async_submit_bio, work);
 663         async->submit_bio_start(async->inode, async->rw, async->bio,
 664                                async->mirror_num, async->bio_flags,
 665                                async->bio_offset);
 666 }
 667
 668 static void run_one_async_done(struct btrfs_work *work)
 669 {
 670         struct btrfs_fs_info *fs_info;
 671         struct async_submit_bio *async;
 672         int limit;
 673
 674         async = container_of(work, struct  async_submit_bio, work);
 675         fs_info = BTRFS_I(async->inode)->root->fs_info;
 676
 677         limit = btrfs_async_submit_limit(fs_info);
 678         limit = limit * 2 / 3;
 679
 680         atomic_dec(&fs_info->nr_async_submits);
 681
 682         if (atomic_read(&fs_info->nr_async_submits) < limit &&
 683             waitqueue_active(&fs_info->async_submit_wait))
 684                 wake_up(&fs_info->async_submit_wait);
 685
 686         async->submit_bio_done(async->inode, async->rw, async->bio,
 687                                async->mirror_num, async->bio_flags,
 688                                async->bio_offset);
 689 }
 690
 691 static void run_one_async_free(struct btrfs_work *work)
 692 {
 693         struct async_submit_bio *async;
 694
 695         async = container_of(work, struct  async_submit_bio, work);
 696         kfree(async);
 697 }
 698
 699 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 700                         int rw, struct bio *bio, int mirror_num,
 701                         unsigned long bio_flags,
 702                         u64 bio_offset,
 703                         extent_submit_bio_hook_t *submit_bio_start,
 704                         extent_submit_bio_hook_t *submit_bio_done)
 705 {
 706         struct async_submit_bio *async;
 707
 708         async = kmalloc(sizeof(*async), GFP_NOFS);
 709         if (!async)
 710                 return -ENOMEM;
 711
 712         async->inode = inode;
 713         async->rw = rw;
 714         async->bio = bio;
 715         async->mirror_num = mirror_num;
 716         async->submit_bio_start = submit_bio_start;
 717         async->submit_bio_done = submit_bio_done;
 718
 719         async->work.func = run_one_async_start;
 720         async->work.ordered_func = run_one_async_done;
 721         async->work.ordered_free = run_one_async_free;
 722
 723         async->work.flags = 0;
 724         async->bio_flags = bio_flags;
 725         async->bio_offset = bio_offset;
 726
 727         atomic_inc(&fs_info->nr_async_submits);
 728
 729         if (rw & REQ_SYNC)
 730                 btrfs_set_work_high_prio(&async->work);
 731
 732         btrfs_queue_worker(&fs_info->workers, &async->work);
 733
 734         while (atomic_read(&fs_info->async_submit_draining) &&
 735               atomic_read(&fs_info->nr_async_submits)) {
 736                 wait_event(fs_info->async_submit_wait,
 737                            (atomic_read(&fs_info->nr_async_submits) == 0));
 738         }
 739
 740         return 0;
 741 }
 742
 743 static int btree_csum_one_bio(struct bio *bio)
 744 {
 745         struct bio_vec *bvec = bio->bi_io_vec;
 746         int bio_index = 0;
 747         struct btrfs_root *root;
 748
 749         WARN_ON(bio->bi_vcnt <= 0);
 750         while (bio_index < bio->bi_vcnt) {
 751                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 752                 csum_dirty_buffer(root, bvec->bv_page);
 753                 bio_index++;
 754                 bvec++;
 755         }
 756         return 0;
 757 }
 758
 759 static int __btree_submit_bio_start(struct inode *inode, int rw,
 760                                     struct bio *bio, int mirror_num,
 761                                     unsigned long bio_flags,
 762                                     u64 bio_offset)
 763 {
 764         /*
 765          * when we're called for a write, we're already in the async
 766          * submission context.  Just jump into btrfs_map_bio
 767          */
 768         btree_csum_one_bio(bio);
 769         return 0;
 770 }
 771
 772 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 773                                  int mirror_num, unsigned long bio_flags,
 774                                  u64 bio_offset)
 775 {
 776         /*
 777          * when we're called for a write, we're already in the async
 778          * submission context.  Just jump into btrfs_map_bio
 779          */
 780         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 781 }
 782
 783 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 784                                  int mirror_num, unsigned long bio_flags,
 785                                  u64 bio_offset)
 786 {
 787         int ret;
 788
 789         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 790                                           bio, 1);
 791         BUG_ON(ret);
 792
 793         if (!(rw & REQ_WRITE)) {
 794                 /*
 795                  * called for a read, do the setup so that checksum validation
 796                  * can happen in the async kernel threads
 797                  */
 798                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 799                                      mirror_num, 0);
 800         }
 801
 802         /*
 803          * kthread helpers are used to submit writes so that checksumming
 804          * can happen in parallel across all CPUs
 805          */
 806         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 807                                    inode, rw, bio, mirror_num, 0,
 808                                    bio_offset,
 809                                    __btree_submit_bio_start,
 810                                    __btree_submit_bio_done);
 811 }
 812
 813 #ifdef CONFIG_MIGRATION
 814 static int btree_migratepage(struct address_space *mapping,
 815                         struct page *newpage, struct page *page)
 816 {
 817         /*
 818          * we can't safely write a btree page from here,
 819          * we haven't done the locking hook
 820          */
 821         if (PageDirty(page))
 822                 return -EAGAIN;
 823         /*
 824          * Buffers may be managed in a filesystem specific way.
 825          * We must have no buffers or drop them.
 826          */
 827         if (page_has_private(page) &&
 828             !try_to_release_page(page, GFP_KERNEL))
 829                 return -EAGAIN;
 830         return migrate_page(mapping, newpage, page);
 831 }
 832 #endif
 833
 834 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 835 {
 836         struct extent_io_tree *tree;
 837         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 838         struct extent_buffer *eb;
 839         int was_dirty;
 840
 841         tree = &BTRFS_I(page->mapping->host)->io_tree;
 842         if (!(current->flags & PF_MEMALLOC)) {
 843                 return extent_write_full_page(tree, page,
 844                                               btree_get_extent, wbc);
 845         }
 846
 847         redirty_page_for_writepage(wbc, page);
 848         eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
 849         WARN_ON(!eb);
 850
 851         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
 852         if (!was_dirty) {
 853                 spin_lock(&root->fs_info->delalloc_lock);
 854                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
 855                 spin_unlock(&root->fs_info->delalloc_lock);
 856         }
 857         free_extent_buffer(eb);
 858
 859         unlock_page(page);
 860         return 0;
 861 }
 862
 863 static int btree_writepages(struct address_space *mapping,
 864                             struct writeback_control *wbc)
 865 {
 866         struct extent_io_tree *tree;
 867         tree = &BTRFS_I(mapping->host)->io_tree;
 868         if (wbc->sync_mode == WB_SYNC_NONE) {
 869                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
 870                 u64 num_dirty;
 871                 unsigned long thresh = 32 * 1024 * 1024;
 872
 873                 if (wbc->for_kupdate)
 874                         return 0;
 875
 876                 /* this is a bit racy, but that's ok */
 877                 num_dirty = root->fs_info->dirty_metadata_bytes;
 878                 if (num_dirty < thresh)
 879                         return 0;
 880         }
 881         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 882 }
 883
 884 static int btree_readpage(struct file *file, struct page *page)
 885 {
 886         struct extent_io_tree *tree;
 887         tree = &BTRFS_I(page->mapping->host)->io_tree;
 888         return extent_read_full_page(tree, page, btree_get_extent);
 889 }
 890
 891 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 892 {
 893         struct extent_io_tree *tree;
 894         struct extent_map_tree *map;
 895         int ret;
 896
 897         if (PageWriteback(page) || PageDirty(page))
 898                 return 0;
 899
 900         tree = &BTRFS_I(page->mapping->host)->io_tree;
 901         map = &BTRFS_I(page->mapping->host)->extent_tree;
 902
 903         ret = try_release_extent_state(map, tree, page, gfp_flags);
 904         if (!ret)
 905                 return 0;
 906
 907         ret = try_release_extent_buffer(tree, page);
 908         if (ret == 1) {
 909                 ClearPagePrivate(page);
 910                 set_page_private(page, 0);
 911                 page_cache_release(page);
 912         }
 913
 914         return ret;
 915 }
 916
 917 static void btree_invalidatepage(struct page *page, unsigned long offset)
 918 {
 919         struct extent_io_tree *tree;
 920         tree = &BTRFS_I(page->mapping->host)->io_tree;
 921         extent_invalidatepage(tree, page, offset);
 922         btree_releasepage(page, GFP_NOFS);
 923         if (PagePrivate(page)) {
 924                 printk(KERN_WARNING "btrfs warning page private not zero "
 925                        "on page %llu\n", (unsigned long long)page_offset(page));
 926                 ClearPagePrivate(page);
 927                 set_page_private(page, 0);
 928                 page_cache_release(page);
 929         }
 930 }
 931
 932 static const struct address_space_operations btree_aops = {
 933         .readpage       = btree_readpage,
 934         .writepage      = btree_writepage,
 935         .writepages     = btree_writepages,
 936         .releasepage    = btree_releasepage,
 937         .invalidatepage = btree_invalidatepage,
 938         .sync_page      = block_sync_page,
 939 #ifdef CONFIG_MIGRATION
 940         .migratepage    = btree_migratepage,
 941 #endif
 942 };
 943
 944 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 945                          u64 parent_transid)
 946 {
 947         struct extent_buffer *buf = NULL;
 948         struct inode *btree_inode = root->fs_info->btree_inode;
 949         int ret = 0;
 950
 951         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 952         if (!buf)
 953                 return 0;
 954         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 955                                  buf, 0, 0, btree_get_extent, 0);
 956         free_extent_buffer(buf);
 957         return ret;
 958 }
 959
 960 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 961                                             u64 bytenr, u32 blocksize)
 962 {
 963         struct inode *btree_inode = root->fs_info->btree_inode;
 964         struct extent_buffer *eb;
 965         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 966                                 bytenr, blocksize, GFP_NOFS);
 967         return eb;
 968 }
 969
 970 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 971                                                  u64 bytenr, u32 blocksize)
 972 {
 973         struct inode *btree_inode = root->fs_info->btree_inode;
 974         struct extent_buffer *eb;
 975
 976         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 977                                  bytenr, blocksize, NULL, GFP_NOFS);
 978         return eb;
 979 }
 980
 981
 982 int btrfs_write_tree_block(struct extent_buffer *buf)
 983 {
 984         return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
 985                                         buf->start + buf->len - 1);
 986 }
 987
 988 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 989 {
 990         return filemap_fdatawait_range(buf->first_page->mapping,
 991                                        buf->start, buf->start + buf->len - 1);
 992 }
 993
 994 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 995                                       u32 blocksize, u64 parent_transid)
 996 {
 997         struct extent_buffer *buf = NULL;
 998         int ret;
 999
1000         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1001         if (!buf)
1002                 return NULL;
1003
1004         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1005
1006         if (ret == 0)
1007                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1008         return buf;
1009
1010 }
1011
1012 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1013                      struct extent_buffer *buf)
1014 {
1015         struct inode *btree_inode = root->fs_info->btree_inode;
1016         if (btrfs_header_generation(buf) ==
1017             root->fs_info->running_transaction->transid) {
1018                 btrfs_assert_tree_locked(buf);
1019
1020                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1021                         spin_lock(&root->fs_info->delalloc_lock);
1022                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
1023                                 root->fs_info->dirty_metadata_bytes -= buf->len;
1024                         else
1025                                 WARN_ON(1);
1026                         spin_unlock(&root->fs_info->delalloc_lock);
1027                 }
1028
1029                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1030                 btrfs_set_lock_blocking(buf);
1031                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1032                                           buf);
1033         }
1034         return 0;
1035 }
1036
1037 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1038                         u32 stripesize, struct btrfs_root *root,
1039                         struct btrfs_fs_info *fs_info,
1040                         u64 objectid)
1041 {
1042         root->node = NULL;
1043         root->commit_root = NULL;
1044         root->sectorsize = sectorsize;
1045         root->nodesize = nodesize;
1046         root->leafsize = leafsize;
1047         root->stripesize = stripesize;
1048         root->ref_cows = 0;
1049         root->track_dirty = 0;
1050         root->in_radix = 0;
1051         root->orphan_item_inserted = 0;
1052         root->orphan_cleanup_state = 0;
1053
1054         root->fs_info = fs_info;
1055         root->objectid = objectid;
1056         root->last_trans = 0;
1057         root->highest_objectid = 0;
1058         root->name = NULL;
1059         root->in_sysfs = 0;
1060         root->inode_tree = RB_ROOT;
1061         root->block_rsv = NULL;
1062         root->orphan_block_rsv = NULL;
1063
1064         INIT_LIST_HEAD(&root->dirty_list);
1065         INIT_LIST_HEAD(&root->orphan_list);
1066         INIT_LIST_HEAD(&root->root_list);
1067         spin_lock_init(&root->node_lock);
1068         spin_lock_init(&root->orphan_lock);
1069         spin_lock_init(&root->inode_lock);
1070         spin_lock_init(&root->accounting_lock);
1071         mutex_init(&root->objectid_mutex);
1072         mutex_init(&root->log_mutex);
1073         init_waitqueue_head(&root->log_writer_wait);
1074         init_waitqueue_head(&root->log_commit_wait[0]);
1075         init_waitqueue_head(&root->log_commit_wait[1]);
1076         atomic_set(&root->log_commit[0], 0);
1077         atomic_set(&root->log_commit[1], 0);
1078         atomic_set(&root->log_writers, 0);
1079         root->log_batch = 0;
1080         root->log_transid = 0;
1081         root->last_log_commit = 0;
1082         extent_io_tree_init(&root->dirty_log_pages,
1083                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1084
1085         memset(&root->root_key, 0, sizeof(root->root_key));
1086         memset(&root->root_item, 0, sizeof(root->root_item));
1087         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1088         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1089         root->defrag_trans_start = fs_info->generation;
1090         init_completion(&root->kobj_unregister);
1091         root->defrag_running = 0;
1092         root->root_key.objectid = objectid;
1093         root->anon_super.s_root = NULL;
1094         root->anon_super.s_dev = 0;
1095         INIT_LIST_HEAD(&root->anon_super.s_list);
1096         INIT_LIST_HEAD(&root->anon_super.s_instances);
1097         init_rwsem(&root->anon_super.s_umount);
1098
1099         return 0;
1100 }
1101
1102 static int find_and_setup_root(struct btrfs_root *tree_root,
1103                                struct btrfs_fs_info *fs_info,
1104                                u64 objectid,
1105                                struct btrfs_root *root)
1106 {
1107         int ret;
1108         u32 blocksize;
1109         u64 generation;
1110
1111         __setup_root(tree_root->nodesize, tree_root->leafsize,
1112                      tree_root->sectorsize, tree_root->stripesize,
1113                      root, fs_info, objectid);
1114         ret = btrfs_find_last_root(tree_root, objectid,
1115                                    &root->root_item, &root->root_key);
1116         if (ret > 0)
1117                 return -ENOENT;
1118         BUG_ON(ret);
1119
1120         generation = btrfs_root_generation(&root->root_item);
1121         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1122         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1123                                      blocksize, generation);
1124         if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1125                 free_extent_buffer(root->node);
1126                 return -EIO;
1127         }
1128         root->commit_root = btrfs_root_node(root);
1129         return 0;
1130 }
1131
1132 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1133                                          struct btrfs_fs_info *fs_info)
1134 {
1135         struct btrfs_root *root;
1136         struct btrfs_root *tree_root = fs_info->tree_root;
1137         struct extent_buffer *leaf;
1138
1139         root = kzalloc(sizeof(*root), GFP_NOFS);
1140         if (!root)
1141                 return ERR_PTR(-ENOMEM);
1142
1143         __setup_root(tree_root->nodesize, tree_root->leafsize,
1144                      tree_root->sectorsize, tree_root->stripesize,
1145                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1146
1147         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1148         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1149         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1150         /*
1151          * log trees do not get reference counted because they go away
1152          * before a real commit is actually done.  They do store pointers
1153          * to file data extents, and those reference counts still get
1154          * updated (along with back refs to the log tree).
1155          */
1156         root->ref_cows = 0;
1157
1158         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1159                                       BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1160         if (IS_ERR(leaf)) {
1161                 kfree(root);
1162                 return ERR_CAST(leaf);
1163         }
1164
1165         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1166         btrfs_set_header_bytenr(leaf, leaf->start);
1167         btrfs_set_header_generation(leaf, trans->transid);
1168         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1169         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1170         root->node = leaf;
1171
1172         write_extent_buffer(root->node, root->fs_info->fsid,
1173                             (unsigned long)btrfs_header_fsid(root->node),
1174                             BTRFS_FSID_SIZE);
1175         btrfs_mark_buffer_dirty(root->node);
1176         btrfs_tree_unlock(root->node);
1177         return root;
1178 }
1179
1180 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1181                              struct btrfs_fs_info *fs_info)
1182 {
1183         struct btrfs_root *log_root;
1184
1185         log_root = alloc_log_tree(trans, fs_info);
1186         if (IS_ERR(log_root))
1187                 return PTR_ERR(log_root);
1188         WARN_ON(fs_info->log_root_tree);
1189         fs_info->log_root_tree = log_root;
1190         return 0;
1191 }
1192
1193 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1194                        struct btrfs_root *root)
1195 {
1196         struct btrfs_root *log_root;
1197         struct btrfs_inode_item *inode_item;
1198
1199         log_root = alloc_log_tree(trans, root->fs_info);
1200         if (IS_ERR(log_root))
1201                 return PTR_ERR(log_root);
1202
1203         log_root->last_trans = trans->transid;
1204         log_root->root_key.offset = root->root_key.objectid;
1205
1206         inode_item = &log_root->root_item.inode;
1207         inode_item->generation = cpu_to_le64(1);
1208         inode_item->size = cpu_to_le64(3);
1209         inode_item->nlink = cpu_to_le32(1);
1210         inode_item->nbytes = cpu_to_le64(root->leafsize);
1211         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1212
1213         btrfs_set_root_node(&log_root->root_item, log_root->node);
1214
1215         WARN_ON(root->log_root);
1216         root->log_root = log_root;
1217         root->log_transid = 0;
1218         root->last_log_commit = 0;
1219         return 0;
1220 }
1221
1222 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1223                                                struct btrfs_key *location)
1224 {
1225         struct btrfs_root *root;
1226         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1227         struct btrfs_path *path;
1228         struct extent_buffer *l;
1229         u64 generation;
1230         u32 blocksize;
1231         int ret = 0;
1232
1233         root = kzalloc(sizeof(*root), GFP_NOFS);
1234         if (!root)
1235                 return ERR_PTR(-ENOMEM);
1236         if (location->offset == (u64)-1) {
1237                 ret = find_and_setup_root(tree_root, fs_info,
1238                                           location->objectid, root);
1239                 if (ret) {
1240                         kfree(root);
1241                         return ERR_PTR(ret);
1242                 }
1243                 goto out;
1244         }
1245
1246         __setup_root(tree_root->nodesize, tree_root->leafsize,
1247                      tree_root->sectorsize, tree_root->stripesize,
1248                      root, fs_info, location->objectid);
1249
1250         path = btrfs_alloc_path();
1251         BUG_ON(!path);
1252         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1253         if (ret == 0) {
1254                 l = path->nodes[0];
1255                 read_extent_buffer(l, &root->root_item,
1256                                 btrfs_item_ptr_offset(l, path->slots[0]),
1257                                 sizeof(root->root_item));
1258                 memcpy(&root->root_key, location, sizeof(*location));
1259         }
1260         btrfs_free_path(path);
1261         if (ret) {
1262                 kfree(root);
1263                 if (ret > 0)
1264                         ret = -ENOENT;
1265                 return ERR_PTR(ret);
1266         }
1267
1268         generation = btrfs_root_generation(&root->root_item);
1269         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1270         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1271                                      blocksize, generation);
1272         root->commit_root = btrfs_root_node(root);
1273         BUG_ON(!root->node);
1274 out:
1275         if (location->objectid != BTRFS_TREE_LOG_OBJECTID)
1276                 root->ref_cows = 1;
1277
1278         return root;
1279 }
1280
1281 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1282                                         u64 root_objectid)
1283 {
1284         struct btrfs_root *root;
1285
1286         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1287                 return fs_info->tree_root;
1288         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1289                 return fs_info->extent_root;
1290
1291         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1292                                  (unsigned long)root_objectid);
1293         return root;
1294 }
1295
1296 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1297                                               struct btrfs_key *location)
1298 {
1299         struct btrfs_root *root;
1300         int ret;
1301
1302         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1303                 return fs_info->tree_root;
1304         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1305                 return fs_info->extent_root;
1306         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1307                 return fs_info->chunk_root;
1308         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1309                 return fs_info->dev_root;
1310         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1311                 return fs_info->csum_root;
1312 again:
1313         spin_lock(&fs_info->fs_roots_radix_lock);
1314         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1315                                  (unsigned long)location->objectid);
1316         spin_unlock(&fs_info->fs_roots_radix_lock);
1317         if (root)
1318                 return root;
1319
1320         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1321         if (IS_ERR(root))
1322                 return root;
1323
1324         set_anon_super(&root->anon_super, NULL);
1325
1326         if (btrfs_root_refs(&root->root_item) == 0) {
1327                 ret = -ENOENT;
1328                 goto fail;
1329         }
1330
1331         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1332         if (ret < 0)
1333                 goto fail;
1334         if (ret == 0)
1335                 root->orphan_item_inserted = 1;
1336
1337         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1338         if (ret)
1339                 goto fail;
1340
1341         spin_lock(&fs_info->fs_roots_radix_lock);
1342         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1343                                 (unsigned long)root->root_key.objectid,
1344                                 root);
1345         if (ret == 0)
1346                 root->in_radix = 1;
1347
1348         spin_unlock(&fs_info->fs_roots_radix_lock);
1349         radix_tree_preload_end();
1350         if (ret) {
1351                 if (ret == -EEXIST) {
1352                         free_fs_root(root);
1353                         goto again;
1354                 }
1355                 goto fail;
1356         }
1357
1358         ret = btrfs_find_dead_roots(fs_info->tree_root,
1359                                     root->root_key.objectid);
1360         WARN_ON(ret);
1361         return root;
1362 fail:
1363         free_fs_root(root);
1364         return ERR_PTR(ret);
1365 }
1366
1367 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1368                                       struct btrfs_key *location,
1369                                       const char *name, int namelen)
1370 {
1371         return btrfs_read_fs_root_no_name(fs_info, location);
1372 #if 0
1373         struct btrfs_root *root;
1374         int ret;
1375
1376         root = btrfs_read_fs_root_no_name(fs_info, location);
1377         if (!root)
1378                 return NULL;
1379
1380         if (root->in_sysfs)
1381                 return root;
1382
1383         ret = btrfs_set_root_name(root, name, namelen);
1384         if (ret) {
1385                 free_extent_buffer(root->node);
1386                 kfree(root);
1387                 return ERR_PTR(ret);
1388         }
1389
1390         ret = btrfs_sysfs_add_root(root);
1391         if (ret) {
1392                 free_extent_buffer(root->node);
1393                 kfree(root->name);
1394                 kfree(root);
1395                 return ERR_PTR(ret);
1396         }
1397         root->in_sysfs = 1;
1398         return root;
1399 #endif
1400 }
1401
1402 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1403 {
1404         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1405         int ret = 0;
1406         struct btrfs_device *device;
1407         struct backing_dev_info *bdi;
1408
1409         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1410                 if (!device->bdev)
1411                         continue;
1412                 bdi = blk_get_backing_dev_info(device->bdev);
1413                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1414                         ret = 1;
1415                         break;
1416                 }
1417         }
1418         return ret;
1419 }
1420
1421 /*
1422  * this unplugs every device on the box, and it is only used when page
1423  * is null
1424  */
1425 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1426 {
1427         struct btrfs_device *device;
1428         struct btrfs_fs_info *info;
1429
1430         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1431         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1432                 if (!device->bdev)
1433                         continue;
1434
1435                 bdi = blk_get_backing_dev_info(device->bdev);
1436                 if (bdi->unplug_io_fn)
1437                         bdi->unplug_io_fn(bdi, page);
1438         }
1439 }
1440
1441 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1442 {
1443         struct inode *inode;
1444         struct extent_map_tree *em_tree;
1445         struct extent_map *em;
1446         struct address_space *mapping;
1447         u64 offset;
1448
1449         /* the generic O_DIRECT read code does this */
1450         if (1 || !page) {
1451                 __unplug_io_fn(bdi, page);
1452                 return;
1453         }
1454
1455         /*
1456          * page->mapping may change at any time.  Get a consistent copy
1457          * and use that for everything below
1458          */
1459         smp_mb();
1460         mapping = page->mapping;
1461         if (!mapping)
1462                 return;
1463
1464         inode = mapping->host;
1465
1466         /*
1467          * don't do the expensive searching for a small number of
1468          * devices
1469          */
1470         if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1471                 __unplug_io_fn(bdi, page);
1472                 return;
1473         }
1474
1475         offset = page_offset(page);
1476
1477         em_tree = &BTRFS_I(inode)->extent_tree;
1478         read_lock(&em_tree->lock);
1479         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1480         read_unlock(&em_tree->lock);
1481         if (!em) {
1482                 __unplug_io_fn(bdi, page);
1483                 return;
1484         }
1485
1486         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1487                 free_extent_map(em);
1488                 __unplug_io_fn(bdi, page);
1489                 return;
1490         }
1491         offset = offset - em->start;
1492         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1493                           em->block_start + offset, page);
1494         free_extent_map(em);
1495 }
1496
1497 /*
1498  * If this fails, caller must call bdi_destroy() to get rid of the
1499  * bdi again.
1500  */
1501 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1502 {
1503         int err;
1504
1505         bdi->capabilities = BDI_CAP_MAP_COPY;
1506         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1507         if (err)
1508                 return err;
1509
1510         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1511         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1512         bdi->unplug_io_data     = info;
1513         bdi->congested_fn       = btrfs_congested_fn;
1514         bdi->congested_data     = info;
1515         return 0;
1516 }
1517
1518 static int bio_ready_for_csum(struct bio *bio)
1519 {
1520         u64 length = 0;
1521         u64 buf_len = 0;
1522         u64 start = 0;
1523         struct page *page;
1524         struct extent_io_tree *io_tree = NULL;
1525         struct bio_vec *bvec;
1526         int i;
1527         int ret;
1528
1529         bio_for_each_segment(bvec, bio, i) {
1530                 page = bvec->bv_page;
1531                 if (page->private == EXTENT_PAGE_PRIVATE) {
1532                         length += bvec->bv_len;
1533                         continue;
1534                 }
1535                 if (!page->private) {
1536                         length += bvec->bv_len;
1537                         continue;
1538                 }
1539                 length = bvec->bv_len;
1540                 buf_len = page->private >> 2;
1541                 start = page_offset(page) + bvec->bv_offset;
1542                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1543         }
1544         /* are we fully contained in this bio? */
1545         if (buf_len <= length)
1546                 return 1;
1547
1548         ret = extent_range_uptodate(io_tree, start + length,
1549                                     start + buf_len - 1);
1550         return ret;
1551 }
1552
1553 /*
1554  * called by the kthread helper functions to finally call the bio end_io
1555  * functions.  This is where read checksum verification actually happens
1556  */
1557 static void end_workqueue_fn(struct btrfs_work *work)
1558 {
1559         struct bio *bio;
1560         struct end_io_wq *end_io_wq;
1561         struct btrfs_fs_info *fs_info;
1562         int error;
1563
1564         end_io_wq = container_of(work, struct end_io_wq, work);
1565         bio = end_io_wq->bio;
1566         fs_info = end_io_wq->info;
1567
1568         /* metadata bio reads are special because the whole tree block must
1569          * be checksummed at once.  This makes sure the entire block is in
1570          * ram and up to date before trying to verify things.  For
1571          * blocksize <= pagesize, it is basically a noop
1572          */
1573         if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1574             !bio_ready_for_csum(bio)) {
1575                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1576                                    &end_io_wq->work);
1577                 return;
1578         }
1579         error = end_io_wq->error;
1580         bio->bi_private = end_io_wq->private;
1581         bio->bi_end_io = end_io_wq->end_io;
1582         kfree(end_io_wq);
1583         bio_endio(bio, error);
1584 }
1585
1586 static int cleaner_kthread(void *arg)
1587 {
1588         struct btrfs_root *root = arg;
1589
1590         do {
1591                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1592
1593                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1594                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1595                         btrfs_run_delayed_iputs(root);
1596                         btrfs_clean_old_snapshots(root);
1597                         mutex_unlock(&root->fs_info->cleaner_mutex);
1598                 }
1599
1600                 if (freezing(current)) {
1601                         refrigerator();
1602                 } else {
1603                         set_current_state(TASK_INTERRUPTIBLE);
1604                         if (!kthread_should_stop())
1605                                 schedule();
1606                         __set_current_state(TASK_RUNNING);
1607                 }
1608         } while (!kthread_should_stop());
1609         return 0;
1610 }
1611
1612 static int transaction_kthread(void *arg)
1613 {
1614         struct btrfs_root *root = arg;
1615         struct btrfs_trans_handle *trans;
1616         struct btrfs_transaction *cur;
1617         u64 transid;
1618         unsigned long now;
1619         unsigned long delay;
1620         int ret;
1621
1622         do {
1623                 delay = HZ * 30;
1624                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1625                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1626
1627                 spin_lock(&root->fs_info->new_trans_lock);
1628                 cur = root->fs_info->running_transaction;
1629                 if (!cur) {
1630                         spin_unlock(&root->fs_info->new_trans_lock);
1631                         goto sleep;
1632                 }
1633
1634                 now = get_seconds();
1635                 if (!cur->blocked &&
1636                     (now < cur->start_time || now - cur->start_time < 30)) {
1637                         spin_unlock(&root->fs_info->new_trans_lock);
1638                         delay = HZ * 5;
1639                         goto sleep;
1640                 }
1641                 transid = cur->transid;
1642                 spin_unlock(&root->fs_info->new_trans_lock);
1643
1644                 trans = btrfs_join_transaction(root, 1);
1645                 BUG_ON(IS_ERR(trans));
1646                 if (transid == trans->transid) {
1647                         ret = btrfs_commit_transaction(trans, root);
1648                         BUG_ON(ret);
1649                 } else {
1650                         btrfs_end_transaction(trans, root);
1651                 }
1652 sleep:
1653                 wake_up_process(root->fs_info->cleaner_kthread);
1654                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1655
1656                 if (freezing(current)) {
1657                         refrigerator();
1658                 } else {
1659                         set_current_state(TASK_INTERRUPTIBLE);
1660                         if (!kthread_should_stop() &&
1661                             !btrfs_transaction_blocked(root->fs_info))
1662                                 schedule_timeout(delay);
1663                         __set_current_state(TASK_RUNNING);
1664                 }
1665         } while (!kthread_should_stop());
1666         return 0;
1667 }
1668
1669 struct btrfs_root *open_ctree(struct super_block *sb,
1670                               struct btrfs_fs_devices *fs_devices,
1671                               char *options)
1672 {
1673         u32 sectorsize;
1674         u32 nodesize;
1675         u32 leafsize;
1676         u32 blocksize;
1677         u32 stripesize;
1678         u64 generation;
1679         u64 features;
1680         struct btrfs_key location;
1681         struct buffer_head *bh;
1682         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1683                                                  GFP_NOFS);
1684         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1685                                                  GFP_NOFS);
1686         struct btrfs_root *tree_root = btrfs_sb(sb);
1687         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1688         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1689                                                 GFP_NOFS);
1690         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1691                                               GFP_NOFS);
1692         struct btrfs_root *log_tree_root;
1693
1694         int ret;
1695         int err = -EINVAL;
1696
1697         struct btrfs_super_block *disk_super;
1698
1699         if (!extent_root || !tree_root || !fs_info ||
1700             !chunk_root || !dev_root || !csum_root) {
1701                 err = -ENOMEM;
1702                 goto fail;
1703         }
1704
1705         ret = init_srcu_struct(&fs_info->subvol_srcu);
1706         if (ret) {
1707                 err = ret;
1708                 goto fail;
1709         }
1710
1711         ret = setup_bdi(fs_info, &fs_info->bdi);
1712         if (ret) {
1713                 err = ret;
1714                 goto fail_srcu;
1715         }
1716
1717         fs_info->btree_inode = new_inode(sb);
1718         if (!fs_info->btree_inode) {
1719                 err = -ENOMEM;
1720                 goto fail_bdi;
1721         }
1722
1723         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1724         INIT_LIST_HEAD(&fs_info->trans_list);
1725         INIT_LIST_HEAD(&fs_info->dead_roots);
1726         INIT_LIST_HEAD(&fs_info->delayed_iputs);
1727         INIT_LIST_HEAD(&fs_info->hashers);
1728         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1729         INIT_LIST_HEAD(&fs_info->ordered_operations);
1730         INIT_LIST_HEAD(&fs_info->caching_block_groups);
1731         spin_lock_init(&fs_info->delalloc_lock);
1732         spin_lock_init(&fs_info->new_trans_lock);
1733         spin_lock_init(&fs_info->ref_cache_lock);
1734         spin_lock_init(&fs_info->fs_roots_radix_lock);
1735         spin_lock_init(&fs_info->delayed_iput_lock);
1736
1737         init_completion(&fs_info->kobj_unregister);
1738         fs_info->tree_root = tree_root;
1739         fs_info->extent_root = extent_root;
1740         fs_info->csum_root = csum_root;
1741         fs_info->chunk_root = chunk_root;
1742         fs_info->dev_root = dev_root;
1743         fs_info->fs_devices = fs_devices;
1744         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1745         INIT_LIST_HEAD(&fs_info->space_info);
1746         btrfs_mapping_init(&fs_info->mapping_tree);
1747         btrfs_init_block_rsv(&fs_info->global_block_rsv);
1748         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1749         btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1750         btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1751         btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1752         INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1753         mutex_init(&fs_info->durable_block_rsv_mutex);
1754         atomic_set(&fs_info->nr_async_submits, 0);
1755         atomic_set(&fs_info->async_delalloc_pages, 0);
1756         atomic_set(&fs_info->async_submit_draining, 0);
1757         atomic_set(&fs_info->nr_async_bios, 0);
1758         fs_info->sb = sb;
1759         fs_info->max_inline = 8192 * 1024;
1760         fs_info->metadata_ratio = 0;
1761
1762         fs_info->thread_pool_size = min_t(unsigned long,
1763                                           num_online_cpus() + 2, 8);
1764
1765         INIT_LIST_HEAD(&fs_info->ordered_extents);
1766         spin_lock_init(&fs_info->ordered_extent_lock);
1767
1768         sb->s_blocksize = 4096;
1769         sb->s_blocksize_bits = blksize_bits(4096);
1770         sb->s_bdi = &fs_info->bdi;
1771
1772         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1773         fs_info->btree_inode->i_nlink = 1;
1774         /*
1775          * we set the i_size on the btree inode to the max possible int.
1776          * the real end of the address space is determined by all of
1777          * the devices in the system
1778          */
1779         fs_info->btree_inode->i_size = OFFSET_MAX;
1780         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1781         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1782
1783         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1784         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1785                              fs_info->btree_inode->i_mapping,
1786                              GFP_NOFS);
1787         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1788                              GFP_NOFS);
1789
1790         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1791
1792         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1793         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1794                sizeof(struct btrfs_key));
1795         BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1796         insert_inode_hash(fs_info->btree_inode);
1797
1798         spin_lock_init(&fs_info->block_group_cache_lock);
1799         fs_info->block_group_cache_tree = RB_ROOT;
1800
1801         extent_io_tree_init(&fs_info->freed_extents[0],
1802                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1803         extent_io_tree_init(&fs_info->freed_extents[1],
1804                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1805         fs_info->pinned_extents = &fs_info->freed_extents[0];
1806         fs_info->do_barriers = 1;
1807
1808
1809         mutex_init(&fs_info->trans_mutex);
1810         mutex_init(&fs_info->ordered_operations_mutex);
1811         mutex_init(&fs_info->tree_log_mutex);
1812         mutex_init(&fs_info->chunk_mutex);
1813         mutex_init(&fs_info->transaction_kthread_mutex);
1814         mutex_init(&fs_info->cleaner_mutex);
1815         mutex_init(&fs_info->volume_mutex);
1816         init_rwsem(&fs_info->extent_commit_sem);
1817         init_rwsem(&fs_info->cleanup_work_sem);
1818         init_rwsem(&fs_info->subvol_sem);
1819
1820         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1821         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1822
1823         init_waitqueue_head(&fs_info->transaction_throttle);
1824         init_waitqueue_head(&fs_info->transaction_wait);
1825         init_waitqueue_head(&fs_info->transaction_blocked_wait);
1826         init_waitqueue_head(&fs_info->async_submit_wait);
1827
1828         __setup_root(4096, 4096, 4096, 4096, tree_root,
1829                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1830
1831         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1832         if (!bh) {
1833                 err = -EINVAL;
1834                 goto fail_iput;
1835         }
1836
1837         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1838         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1839                sizeof(fs_info->super_for_commit));
1840         brelse(bh);
1841
1842         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1843
1844         disk_super = &fs_info->super_copy;
1845         if (!btrfs_super_root(disk_super))
1846                 goto fail_iput;
1847
1848         /* check FS state, whether FS is broken. */
1849         fs_info->fs_state |= btrfs_super_flags(disk_super);
1850
1851         btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
1852
1853         ret = btrfs_parse_options(tree_root, options);
1854         if (ret) {
1855                 err = ret;
1856                 goto fail_iput;
1857         }
1858
1859         features = btrfs_super_incompat_flags(disk_super) &
1860                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1861         if (features) {
1862                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1863                        "unsupported optional features (%Lx).\n",
1864                        (unsigned long long)features);
1865                 err = -EINVAL;
1866                 goto fail_iput;
1867         }
1868
1869         features = btrfs_super_incompat_flags(disk_super);
1870         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1871         if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1872                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1873         btrfs_set_super_incompat_flags(disk_super, features);
1874
1875         features = btrfs_super_compat_ro_flags(disk_super) &
1876                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1877         if (!(sb->s_flags & MS_RDONLY) && features) {
1878                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1879                        "unsupported option features (%Lx).\n",
1880                        (unsigned long long)features);
1881                 err = -EINVAL;
1882                 goto fail_iput;
1883         }
1884
1885         btrfs_init_workers(&fs_info->generic_worker,
1886                            "genwork", 1, NULL);
1887
1888         btrfs_init_workers(&fs_info->workers, "worker",
1889                            fs_info->thread_pool_size,
1890                            &fs_info->generic_worker);
1891
1892         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1893                            fs_info->thread_pool_size,
1894                            &fs_info->generic_worker);
1895
1896         btrfs_init_workers(&fs_info->submit_workers, "submit",
1897                            min_t(u64, fs_devices->num_devices,
1898                            fs_info->thread_pool_size),
1899                            &fs_info->generic_worker);
1900
1901         /* a higher idle thresh on the submit workers makes it much more
1902          * likely that bios will be send down in a sane order to the
1903          * devices
1904          */
1905         fs_info->submit_workers.idle_thresh = 64;
1906
1907         fs_info->workers.idle_thresh = 16;
1908         fs_info->workers.ordered = 1;
1909
1910         fs_info->delalloc_workers.idle_thresh = 2;
1911         fs_info->delalloc_workers.ordered = 1;
1912
1913         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1914                            &fs_info->generic_worker);
1915         btrfs_init_workers(&fs_info->endio_workers, "endio",
1916                            fs_info->thread_pool_size,
1917                            &fs_info->generic_worker);
1918         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1919                            fs_info->thread_pool_size,
1920                            &fs_info->generic_worker);
1921         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1922                            "endio-meta-write", fs_info->thread_pool_size,
1923                            &fs_info->generic_worker);
1924         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1925                            fs_info->thread_pool_size,
1926                            &fs_info->generic_worker);
1927         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1928                            1, &fs_info->generic_worker);
1929
1930         /*
1931          * endios are largely parallel and should have a very
1932          * low idle thresh
1933          */
1934         fs_info->endio_workers.idle_thresh = 4;
1935         fs_info->endio_meta_workers.idle_thresh = 4;
1936
1937         fs_info->endio_write_workers.idle_thresh = 2;
1938         fs_info->endio_meta_write_workers.idle_thresh = 2;
1939
1940         btrfs_start_workers(&fs_info->workers, 1);
1941         btrfs_start_workers(&fs_info->generic_worker, 1);
1942         btrfs_start_workers(&fs_info->submit_workers, 1);
1943         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1944         btrfs_start_workers(&fs_info->fixup_workers, 1);
1945         btrfs_start_workers(&fs_info->endio_workers, 1);
1946         btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1947         btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1948         btrfs_start_workers(&fs_info->endio_write_workers, 1);
1949         btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1950
1951         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1952         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1953                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1954
1955         nodesize = btrfs_super_nodesize(disk_super);
1956         leafsize = btrfs_super_leafsize(disk_super);
1957         sectorsize = btrfs_super_sectorsize(disk_super);
1958         stripesize = btrfs_super_stripesize(disk_super);
1959         tree_root->nodesize = nodesize;
1960         tree_root->leafsize = leafsize;
1961         tree_root->sectorsize = sectorsize;
1962         tree_root->stripesize = stripesize;
1963
1964         sb->s_blocksize = sectorsize;
1965         sb->s_blocksize_bits = blksize_bits(sectorsize);
1966
1967         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1968                     sizeof(disk_super->magic))) {
1969                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1970                 goto fail_sb_buffer;
1971         }
1972
1973         mutex_lock(&fs_info->chunk_mutex);
1974         ret = btrfs_read_sys_array(tree_root);
1975         mutex_unlock(&fs_info->chunk_mutex);
1976         if (ret) {
1977                 printk(KERN_WARNING "btrfs: failed to read the system "
1978                        "array on %s\n", sb->s_id);
1979                 goto fail_sb_buffer;
1980         }
1981
1982         blocksize = btrfs_level_size(tree_root,
1983                                      btrfs_super_chunk_root_level(disk_super));
1984         generation = btrfs_super_chunk_root_generation(disk_super);
1985
1986         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1987                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1988
1989         chunk_root->node = read_tree_block(chunk_root,
1990                                            btrfs_super_chunk_root(disk_super),
1991                                            blocksize, generation);
1992         BUG_ON(!chunk_root->node);
1993         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1994                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1995                        sb->s_id);
1996                 goto fail_chunk_root;
1997         }
1998         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1999         chunk_root->commit_root = btrfs_root_node(chunk_root);
2000
2001         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2002            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2003            BTRFS_UUID_SIZE);
2004
2005         mutex_lock(&fs_info->chunk_mutex);
2006         ret = btrfs_read_chunk_tree(chunk_root);
2007         mutex_unlock(&fs_info->chunk_mutex);
2008         if (ret) {
2009                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2010                        sb->s_id);
2011                 goto fail_chunk_root;
2012         }
2013
2014         btrfs_close_extra_devices(fs_devices);
2015
2016         blocksize = btrfs_level_size(tree_root,
2017                                      btrfs_super_root_level(disk_super));
2018         generation = btrfs_super_generation(disk_super);
2019
2020         tree_root->node = read_tree_block(tree_root,
2021                                           btrfs_super_root(disk_super),
2022                                           blocksize, generation);
2023         if (!tree_root->node)
2024                 goto fail_chunk_root;
2025         if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2026                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2027                        sb->s_id);
2028                 goto fail_tree_root;
2029         }
2030         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2031         tree_root->commit_root = btrfs_root_node(tree_root);
2032
2033         ret = find_and_setup_root(tree_root, fs_info,
2034                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2035         if (ret)
2036                 goto fail_tree_root;
2037         extent_root->track_dirty = 1;
2038
2039         ret = find_and_setup_root(tree_root, fs_info,
2040                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
2041         if (ret)
2042                 goto fail_extent_root;
2043         dev_root->track_dirty = 1;
2044
2045         ret = find_and_setup_root(tree_root, fs_info,
2046                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
2047         if (ret)
2048                 goto fail_dev_root;
2049
2050         csum_root->track_dirty = 1;
2051
2052         fs_info->generation = generation;
2053         fs_info->last_trans_committed = generation;
2054         fs_info->data_alloc_profile = (u64)-1;
2055         fs_info->metadata_alloc_profile = (u64)-1;
2056         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
2057
2058         ret = btrfs_read_block_groups(extent_root);
2059         if (ret) {
2060                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2061                 goto fail_block_groups;
2062         }
2063
2064         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2065                                                "btrfs-cleaner");
2066         if (IS_ERR(fs_info->cleaner_kthread))
2067                 goto fail_block_groups;
2068
2069         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2070                                                    tree_root,
2071                                                    "btrfs-transaction");
2072         if (IS_ERR(fs_info->transaction_kthread))
2073                 goto fail_cleaner;
2074
2075         if (!btrfs_test_opt(tree_root, SSD) &&
2076             !btrfs_test_opt(tree_root, NOSSD) &&
2077             !fs_info->fs_devices->rotating) {
2078                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2079                        "mode\n");
2080                 btrfs_set_opt(fs_info->mount_opt, SSD);
2081         }
2082
2083         /* do not make disk changes in broken FS */
2084         if (btrfs_super_log_root(disk_super) != 0 &&
2085             !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2086                 u64 bytenr = btrfs_super_log_root(disk_super);
2087
2088                 if (fs_devices->rw_devices == 0) {
2089                         printk(KERN_WARNING "Btrfs log replay required "
2090                                "on RO media\n");
2091                         err = -EIO;
2092                         goto fail_trans_kthread;
2093                 }
2094                 blocksize =
2095                      btrfs_level_size(tree_root,
2096                                       btrfs_super_log_root_level(disk_super));
2097
2098                 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2099                 if (!log_tree_root) {
2100                         err = -ENOMEM;
2101                         goto fail_trans_kthread;
2102                 }
2103
2104                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2105                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2106
2107                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2108                                                       blocksize,
2109                                                       generation + 1);
2110                 ret = btrfs_recover_log_trees(log_tree_root);
2111                 BUG_ON(ret);
2112
2113                 if (sb->s_flags & MS_RDONLY) {
2114                         ret =  btrfs_commit_super(tree_root);
2115                         BUG_ON(ret);
2116                 }
2117         }
2118
2119         ret = btrfs_find_orphan_roots(tree_root);
2120         BUG_ON(ret);
2121
2122         if (!(sb->s_flags & MS_RDONLY)) {
2123                 ret = btrfs_cleanup_fs_roots(fs_info);
2124                 BUG_ON(ret);
2125
2126                 ret = btrfs_recover_relocation(tree_root);
2127                 if (ret < 0) {
2128                         printk(KERN_WARNING
2129                                "btrfs: failed to recover relocation\n");
2130                         err = -EINVAL;
2131                         goto fail_trans_kthread;
2132                 }
2133         }
2134
2135         location.objectid = BTRFS_FS_TREE_OBJECTID;
2136         location.type = BTRFS_ROOT_ITEM_KEY;
2137         location.offset = (u64)-1;
2138
2139         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2140         if (!fs_info->fs_root)
2141                 goto fail_trans_kthread;
2142         if (IS_ERR(fs_info->fs_root)) {
2143                 err = PTR_ERR(fs_info->fs_root);
2144                 goto fail_trans_kthread;
2145         }
2146
2147         if (!(sb->s_flags & MS_RDONLY)) {
2148                 down_read(&fs_info->cleanup_work_sem);
2149                 err = btrfs_orphan_cleanup(fs_info->fs_root);
2150                 if (!err)
2151                         err = btrfs_orphan_cleanup(fs_info->tree_root);
2152                 up_read(&fs_info->cleanup_work_sem);
2153                 if (err) {
2154                         close_ctree(tree_root);
2155                         return ERR_PTR(err);
2156                 }
2157         }
2158
2159         return tree_root;
2160
2161 fail_trans_kthread:
2162         kthread_stop(fs_info->transaction_kthread);
2163 fail_cleaner:
2164         kthread_stop(fs_info->cleaner_kthread);
2165
2166         /*
2167          * make sure we're done with the btree inode before we stop our
2168          * kthreads
2169          */
2170         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2171         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2172
2173 fail_block_groups:
2174         btrfs_free_block_groups(fs_info);
2175         free_extent_buffer(csum_root->node);
2176         free_extent_buffer(csum_root->commit_root);
2177 fail_dev_root:
2178         free_extent_buffer(dev_root->node);
2179         free_extent_buffer(dev_root->commit_root);
2180 fail_extent_root:
2181         free_extent_buffer(extent_root->node);
2182         free_extent_buffer(extent_root->commit_root);
2183 fail_tree_root:
2184         free_extent_buffer(tree_root->node);
2185         free_extent_buffer(tree_root->commit_root);
2186 fail_chunk_root:
2187         free_extent_buffer(chunk_root->node);
2188         free_extent_buffer(chunk_root->commit_root);
2189 fail_sb_buffer:
2190         btrfs_stop_workers(&fs_info->generic_worker);
2191         btrfs_stop_workers(&fs_info->fixup_workers);
2192         btrfs_stop_workers(&fs_info->delalloc_workers);
2193         btrfs_stop_workers(&fs_info->workers);
2194         btrfs_stop_workers(&fs_info->endio_workers);
2195         btrfs_stop_workers(&fs_info->endio_meta_workers);
2196         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2197         btrfs_stop_workers(&fs_info->endio_write_workers);
2198         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2199         btrfs_stop_workers(&fs_info->submit_workers);
2200 fail_iput:
2201         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2202         iput(fs_info->btree_inode);
2203
2204         btrfs_close_devices(fs_info->fs_devices);
2205         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2206 fail_bdi:
2207         bdi_destroy(&fs_info->bdi);
2208 fail_srcu:
2209         cleanup_srcu_struct(&fs_info->subvol_srcu);
2210 fail:
2211         kfree(extent_root);
2212         kfree(tree_root);
2213         kfree(fs_info);
2214         kfree(chunk_root);
2215         kfree(dev_root);
2216         kfree(csum_root);
2217         return ERR_PTR(err);
2218 }
2219
2220 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2221 {
2222         char b[BDEVNAME_SIZE];
2223
2224         if (uptodate) {
2225                 set_buffer_uptodate(bh);
2226         } else {
2227                 if (printk_ratelimit()) {
2228                         printk(KERN_WARNING "lost page write due to "
2229                                         "I/O error on %s\n",
2230                                        bdevname(bh->b_bdev, b));
2231                 }
2232                 /* note, we dont' set_buffer_write_io_error because we have
2233                  * our own ways of dealing with the IO errors
2234                  */
2235                 clear_buffer_uptodate(bh);
2236         }
2237         unlock_buffer(bh);
2238         put_bh(bh);
2239 }
2240
2241 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2242 {
2243         struct buffer_head *bh;
2244         struct buffer_head *latest = NULL;
2245         struct btrfs_super_block *super;
2246         int i;
2247         u64 transid = 0;
2248         u64 bytenr;
2249
2250         /* we would like to check all the supers, but that would make
2251          * a btrfs mount succeed after a mkfs from a different FS.
2252          * So, we need to add a special mount option to scan for
2253          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2254          */
2255         for (i = 0; i < 1; i++) {
2256                 bytenr = btrfs_sb_offset(i);
2257                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2258                         break;
2259                 bh = __bread(bdev, bytenr / 4096, 4096);
2260                 if (!bh)
2261                         continue;
2262
2263                 super = (struct btrfs_super_block *)bh->b_data;
2264                 if (btrfs_super_bytenr(super) != bytenr ||
2265                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2266                             sizeof(super->magic))) {
2267                         brelse(bh);
2268                         continue;
2269                 }
2270
2271                 if (!latest || btrfs_super_generation(super) > transid) {
2272                         brelse(latest);
2273                         latest = bh;
2274                         transid = btrfs_super_generation(super);
2275                 } else {
2276                         brelse(bh);
2277                 }
2278         }
2279         return latest;
2280 }
2281
2282 /*
2283  * this should be called twice, once with wait == 0 and
2284  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2285  * we write are pinned.
2286  *
2287  * They are released when wait == 1 is done.
2288  * max_mirrors must be the same for both runs, and it indicates how
2289  * many supers on this one device should be written.
2290  *
2291  * max_mirrors == 0 means to write them all.
2292  */
2293 static int write_dev_supers(struct btrfs_device *device,
2294                             struct btrfs_super_block *sb,
2295                             int do_barriers, int wait, int max_mirrors)
2296 {
2297         struct buffer_head *bh;
2298         int i;
2299         int ret;
2300         int errors = 0;
2301         u32 crc;
2302         u64 bytenr;
2303         int last_barrier = 0;
2304
2305         if (max_mirrors == 0)
2306                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2307
2308         /* make sure only the last submit_bh does a barrier */
2309         if (do_barriers) {
2310                 for (i = 0; i < max_mirrors; i++) {
2311                         bytenr = btrfs_sb_offset(i);
2312                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2313                             device->total_bytes)
2314                                 break;
2315                         last_barrier = i;
2316                 }
2317         }
2318
2319         for (i = 0; i < max_mirrors; i++) {
2320                 bytenr = btrfs_sb_offset(i);
2321                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2322                         break;
2323
2324                 if (wait) {
2325                         bh = __find_get_block(device->bdev, bytenr / 4096,
2326                                               BTRFS_SUPER_INFO_SIZE);
2327                         BUG_ON(!bh);
2328                         wait_on_buffer(bh);
2329                         if (!buffer_uptodate(bh))
2330                                 errors++;
2331
2332                         /* drop our reference */
2333                         brelse(bh);
2334
2335                         /* drop the reference from the wait == 0 run */
2336                         brelse(bh);
2337                         continue;
2338                 } else {
2339                         btrfs_set_super_bytenr(sb, bytenr);
2340
2341                         crc = ~(u32)0;
2342                         crc = btrfs_csum_data(NULL, (char *)sb +
2343                                               BTRFS_CSUM_SIZE, crc,
2344                                               BTRFS_SUPER_INFO_SIZE -
2345                                               BTRFS_CSUM_SIZE);
2346                         btrfs_csum_final(crc, sb->csum);
2347
2348                         /*
2349                          * one reference for us, and we leave it for the
2350                          * caller
2351                          */
2352                         bh = __getblk(device->bdev, bytenr / 4096,
2353                                       BTRFS_SUPER_INFO_SIZE);
2354                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2355
2356                         /* one reference for submit_bh */
2357                         get_bh(bh);
2358
2359                         set_buffer_uptodate(bh);
2360                         lock_buffer(bh);
2361                         bh->b_end_io = btrfs_end_buffer_write_sync;
2362                 }
2363
2364                 if (i == last_barrier && do_barriers)
2365                         ret = submit_bh(WRITE_FLUSH_FUA, bh);
2366                 else
2367                         ret = submit_bh(WRITE_SYNC, bh);
2368
2369                 if (ret)
2370                         errors++;
2371         }
2372         return errors < i ? 0 : -1;
2373 }
2374
2375 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2376 {
2377         struct list_head *head;
2378         struct btrfs_device *dev;
2379         struct btrfs_super_block *sb;
2380         struct btrfs_dev_item *dev_item;
2381         int ret;
2382         int do_barriers;
2383         int max_errors;
2384         int total_errors = 0;
2385         u64 flags;
2386
2387         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2388         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2389
2390         sb = &root->fs_info->super_for_commit;
2391         dev_item = &sb->dev_item;
2392
2393         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2394         head = &root->fs_info->fs_devices->devices;
2395         list_for_each_entry(dev, head, dev_list) {
2396                 if (!dev->bdev) {
2397                         total_errors++;
2398                         continue;
2399                 }
2400                 if (!dev->in_fs_metadata || !dev->writeable)
2401                         continue;
2402
2403                 btrfs_set_stack_device_generation(dev_item, 0);
2404                 btrfs_set_stack_device_type(dev_item, dev->type);
2405                 btrfs_set_stack_device_id(dev_item, dev->devid);
2406                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2407                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2408                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2409                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2410                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2411                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2412                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2413
2414                 flags = btrfs_super_flags(sb);
2415                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2416
2417                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2418                 if (ret)
2419                         total_errors++;
2420         }
2421         if (total_errors > max_errors) {
2422                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2423                        total_errors);
2424                 BUG();
2425         }
2426
2427         total_errors = 0;
2428         list_for_each_entry(dev, head, dev_list) {
2429                 if (!dev->bdev)
2430                         continue;
2431                 if (!dev->in_fs_metadata || !dev->writeable)
2432                         continue;
2433
2434                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2435                 if (ret)
2436                         total_errors++;
2437         }
2438         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2439         if (total_errors > max_errors) {
2440                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2441                        total_errors);
2442                 BUG();
2443         }
2444         return 0;
2445 }
2446
2447 int write_ctree_super(struct btrfs_trans_handle *trans,
2448                       struct btrfs_root *root, int max_mirrors)
2449 {
2450         int ret;
2451
2452         ret = write_all_supers(root, max_mirrors);
2453         return ret;
2454 }
2455
2456 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2457 {
2458         spin_lock(&fs_info->fs_roots_radix_lock);
2459         radix_tree_delete(&fs_info->fs_roots_radix,
2460                           (unsigned long)root->root_key.objectid);
2461         spin_unlock(&fs_info->fs_roots_radix_lock);
2462
2463         if (btrfs_root_refs(&root->root_item) == 0)
2464                 synchronize_srcu(&fs_info->subvol_srcu);
2465
2466         free_fs_root(root);
2467         return 0;
2468 }
2469
2470 static void free_fs_root(struct btrfs_root *root)
2471 {
2472         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2473         if (root->anon_super.s_dev) {
2474                 down_write(&root->anon_super.s_umount);
2475                 kill_anon_super(&root->anon_super);
2476         }
2477         free_extent_buffer(root->node);
2478         free_extent_buffer(root->commit_root);
2479         kfree(root->name);
2480         kfree(root);
2481 }
2482
2483 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2484 {
2485         int ret;
2486         struct btrfs_root *gang[8];
2487         int i;
2488
2489         while (!list_empty(&fs_info->dead_roots)) {
2490                 gang[0] = list_entry(fs_info->dead_roots.next,
2491                                      struct btrfs_root, root_list);
2492                 list_del(&gang[0]->root_list);
2493
2494                 if (gang[0]->in_radix) {
2495                         btrfs_free_fs_root(fs_info, gang[0]);
2496                 } else {
2497                         free_extent_buffer(gang[0]->node);
2498                         free_extent_buffer(gang[0]->commit_root);
2499                         kfree(gang[0]);
2500                 }
2501         }
2502
2503         while (1) {
2504                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2505                                              (void **)gang, 0,
2506                                              ARRAY_SIZE(gang));
2507                 if (!ret)
2508                         break;
2509                 for (i = 0; i < ret; i++)
2510                         btrfs_free_fs_root(fs_info, gang[i]);
2511         }
2512         return 0;
2513 }
2514
2515 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2516 {
2517         u64 root_objectid = 0;
2518         struct btrfs_root *gang[8];
2519         int i;
2520         int ret;
2521
2522         while (1) {
2523                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2524                                              (void **)gang, root_objectid,
2525                                              ARRAY_SIZE(gang));
2526                 if (!ret)
2527                         break;
2528
2529                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2530                 for (i = 0; i < ret; i++) {
2531                         int err;
2532
2533                         root_objectid = gang[i]->root_key.objectid;
2534                         err = btrfs_orphan_cleanup(gang[i]);
2535                         if (err)
2536                                 return err;
2537                 }
2538                 root_objectid++;
2539         }
2540         return 0;
2541 }
2542
2543 int btrfs_commit_super(struct btrfs_root *root)
2544 {
2545         struct btrfs_trans_handle *trans;
2546         int ret;
2547
2548         mutex_lock(&root->fs_info->cleaner_mutex);
2549         btrfs_run_delayed_iputs(root);
2550         btrfs_clean_old_snapshots(root);
2551         mutex_unlock(&root->fs_info->cleaner_mutex);
2552
2553         /* wait until ongoing cleanup work done */
2554         down_write(&root->fs_info->cleanup_work_sem);
2555         up_write(&root->fs_info->cleanup_work_sem);
2556
2557         trans = btrfs_join_transaction(root, 1);
2558         if (IS_ERR(trans))
2559                 return PTR_ERR(trans);
2560         ret = btrfs_commit_transaction(trans, root);
2561         BUG_ON(ret);
2562         /* run commit again to drop the original snapshot */
2563         trans = btrfs_join_transaction(root, 1);
2564         if (IS_ERR(trans))
2565                 return PTR_ERR(trans);
2566         btrfs_commit_transaction(trans, root);
2567         ret = btrfs_write_and_wait_transaction(NULL, root);
2568         BUG_ON(ret);
2569
2570         ret = write_ctree_super(NULL, root, 0);
2571         return ret;
2572 }
2573
2574 int close_ctree(struct btrfs_root *root)
2575 {
2576         struct btrfs_fs_info *fs_info = root->fs_info;
2577         int ret;
2578
2579         fs_info->closing = 1;
2580         smp_mb();
2581
2582         btrfs_put_block_group_cache(fs_info);
2583
2584         /*
2585          * Here come 2 situations when btrfs is broken to flip readonly:
2586          *
2587          * 1. when btrfs flips readonly somewhere else before
2588          * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2589          * and btrfs will skip to write sb directly to keep
2590          * ERROR state on disk.
2591          *
2592          * 2. when btrfs flips readonly just in btrfs_commit_super,
2593          * and in such case, btrfs cannnot write sb via btrfs_commit_super,
2594          * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2595          * btrfs will cleanup all FS resources first and write sb then.
2596          */
2597         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2598                 ret = btrfs_commit_super(root);
2599                 if (ret)
2600                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2601         }
2602
2603         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2604                 ret = btrfs_error_commit_super(root);
2605                 if (ret)
2606                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2607         }
2608
2609         kthread_stop(root->fs_info->transaction_kthread);
2610         kthread_stop(root->fs_info->cleaner_kthread);
2611
2612         fs_info->closing = 2;
2613         smp_mb();
2614
2615         if (fs_info->delalloc_bytes) {
2616                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2617                        (unsigned long long)fs_info->delalloc_bytes);
2618         }
2619         if (fs_info->total_ref_cache_size) {
2620                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2621                        (unsigned long long)fs_info->total_ref_cache_size);
2622         }
2623
2624         free_extent_buffer(fs_info->extent_root->node);
2625         free_extent_buffer(fs_info->extent_root->commit_root);
2626         free_extent_buffer(fs_info->tree_root->node);
2627         free_extent_buffer(fs_info->tree_root->commit_root);
2628         free_extent_buffer(root->fs_info->chunk_root->node);
2629         free_extent_buffer(root->fs_info->chunk_root->commit_root);
2630         free_extent_buffer(root->fs_info->dev_root->node);
2631         free_extent_buffer(root->fs_info->dev_root->commit_root);
2632         free_extent_buffer(root->fs_info->csum_root->node);
2633         free_extent_buffer(root->fs_info->csum_root->commit_root);
2634
2635         btrfs_free_block_groups(root->fs_info);
2636
2637         del_fs_roots(fs_info);
2638
2639         iput(fs_info->btree_inode);
2640
2641         btrfs_stop_workers(&fs_info->generic_worker);
2642         btrfs_stop_workers(&fs_info->fixup_workers);
2643         btrfs_stop_workers(&fs_info->delalloc_workers);
2644         btrfs_stop_workers(&fs_info->workers);
2645         btrfs_stop_workers(&fs_info->endio_workers);
2646         btrfs_stop_workers(&fs_info->endio_meta_workers);
2647         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2648         btrfs_stop_workers(&fs_info->endio_write_workers);
2649         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2650         btrfs_stop_workers(&fs_info->submit_workers);
2651
2652         btrfs_close_devices(fs_info->fs_devices);
2653         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2654
2655         bdi_destroy(&fs_info->bdi);
2656         cleanup_srcu_struct(&fs_info->subvol_srcu);
2657
2658         kfree(fs_info->extent_root);
2659         kfree(fs_info->tree_root);
2660         kfree(fs_info->chunk_root);
2661         kfree(fs_info->dev_root);
2662         kfree(fs_info->csum_root);
2663         kfree(fs_info);
2664
2665         return 0;
2666 }
2667
2668 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2669 {
2670         int ret;
2671         struct inode *btree_inode = buf->first_page->mapping->host;
2672
2673         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2674                                      NULL);
2675         if (!ret)
2676                 return ret;
2677
2678         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2679                                     parent_transid);
2680         return !ret;
2681 }
2682
2683 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2684 {
2685         struct inode *btree_inode = buf->first_page->mapping->host;
2686         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2687                                           buf);
2688 }
2689
2690 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2691 {
2692         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2693         u64 transid = btrfs_header_generation(buf);
2694         struct inode *btree_inode = root->fs_info->btree_inode;
2695         int was_dirty;
2696
2697         btrfs_assert_tree_locked(buf);
2698         if (transid != root->fs_info->generation) {
2699                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2700                        "found %llu running %llu\n",
2701                         (unsigned long long)buf->start,
2702                         (unsigned long long)transid,
2703                         (unsigned long long)root->fs_info->generation);
2704                 WARN_ON(1);
2705         }
2706         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2707                                             buf);
2708         if (!was_dirty) {
2709                 spin_lock(&root->fs_info->delalloc_lock);
2710                 root->fs_info->dirty_metadata_bytes += buf->len;
2711                 spin_unlock(&root->fs_info->delalloc_lock);
2712         }
2713 }
2714
2715 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2716 {
2717         /*
2718          * looks as though older kernels can get into trouble with
2719          * this code, they end up stuck in balance_dirty_pages forever
2720          */
2721         u64 num_dirty;
2722         unsigned long thresh = 32 * 1024 * 1024;
2723
2724         if (current->flags & PF_MEMALLOC)
2725                 return;
2726
2727         num_dirty = root->fs_info->dirty_metadata_bytes;
2728
2729         if (num_dirty > thresh) {
2730                 balance_dirty_pages_ratelimited_nr(
2731                                    root->fs_info->btree_inode->i_mapping, 1);
2732         }
2733         return;
2734 }
2735
2736 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2737 {
2738         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2739         int ret;
2740         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2741         if (ret == 0)
2742                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2743         return ret;
2744 }
2745
2746 int btree_lock_page_hook(struct page *page)
2747 {
2748         struct inode *inode = page->mapping->host;
2749         struct btrfs_root *root = BTRFS_I(inode)->root;
2750         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2751         struct extent_buffer *eb;
2752         unsigned long len;
2753         u64 bytenr = page_offset(page);
2754
2755         if (page->private == EXTENT_PAGE_PRIVATE)
2756                 goto out;
2757
2758         len = page->private >> 2;
2759         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2760         if (!eb)
2761                 goto out;
2762
2763         btrfs_tree_lock(eb);
2764         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2765
2766         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2767                 spin_lock(&root->fs_info->delalloc_lock);
2768                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2769                         root->fs_info->dirty_metadata_bytes -= eb->len;
2770                 else
2771                         WARN_ON(1);
2772                 spin_unlock(&root->fs_info->delalloc_lock);
2773         }
2774
2775         btrfs_tree_unlock(eb);
2776         free_extent_buffer(eb);
2777 out:
2778         lock_page(page);
2779         return 0;
2780 }
2781
2782 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
2783                               int read_only)
2784 {
2785         if (read_only)
2786                 return;
2787
2788         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
2789                 printk(KERN_WARNING "warning: mount fs with errors, "
2790                        "running btrfsck is recommended\n");
2791 }
2792
2793 int btrfs_error_commit_super(struct btrfs_root *root)
2794 {
2795         int ret;
2796
2797         mutex_lock(&root->fs_info->cleaner_mutex);
2798         btrfs_run_delayed_iputs(root);
2799         mutex_unlock(&root->fs_info->cleaner_mutex);
2800
2801         down_write(&root->fs_info->cleanup_work_sem);
2802         up_write(&root->fs_info->cleanup_work_sem);
2803
2804         /* cleanup FS via transaction */
2805         btrfs_cleanup_transaction(root);
2806
2807         ret = write_ctree_super(NULL, root, 0);
2808
2809         return ret;
2810 }
2811
2812 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
2813 {
2814         struct btrfs_inode *btrfs_inode;
2815         struct list_head splice;
2816
2817         INIT_LIST_HEAD(&splice);
2818
2819         mutex_lock(&root->fs_info->ordered_operations_mutex);
2820         spin_lock(&root->fs_info->ordered_extent_lock);
2821
2822         list_splice_init(&root->fs_info->ordered_operations, &splice);
2823         while (!list_empty(&splice)) {
2824                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2825                                          ordered_operations);
2826
2827                 list_del_init(&btrfs_inode->ordered_operations);
2828
2829                 btrfs_invalidate_inodes(btrfs_inode->root);
2830         }
2831
2832         spin_unlock(&root->fs_info->ordered_extent_lock);
2833         mutex_unlock(&root->fs_info->ordered_operations_mutex);
2834
2835         return 0;
2836 }
2837
2838 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
2839 {
2840         struct list_head splice;
2841         struct btrfs_ordered_extent *ordered;
2842         struct inode *inode;
2843
2844         INIT_LIST_HEAD(&splice);
2845
2846         spin_lock(&root->fs_info->ordered_extent_lock);
2847
2848         list_splice_init(&root->fs_info->ordered_extents, &splice);
2849         while (!list_empty(&splice)) {
2850                 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
2851                                      root_extent_list);
2852
2853                 list_del_init(&ordered->root_extent_list);
2854                 atomic_inc(&ordered->refs);
2855
2856                 /* the inode may be getting freed (in sys_unlink path). */
2857                 inode = igrab(ordered->inode);
2858
2859                 spin_unlock(&root->fs_info->ordered_extent_lock);
2860                 if (inode)
2861                         iput(inode);
2862
2863                 atomic_set(&ordered->refs, 1);
2864                 btrfs_put_ordered_extent(ordered);
2865
2866                 spin_lock(&root->fs_info->ordered_extent_lock);
2867         }
2868
2869         spin_unlock(&root->fs_info->ordered_extent_lock);
2870
2871         return 0;
2872 }
2873
2874 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
2875                                       struct btrfs_root *root)
2876 {
2877         struct rb_node *node;
2878         struct btrfs_delayed_ref_root *delayed_refs;
2879         struct btrfs_delayed_ref_node *ref;
2880         int ret = 0;
2881
2882         delayed_refs = &trans->delayed_refs;
2883
2884         spin_lock(&delayed_refs->lock);
2885         if (delayed_refs->num_entries == 0) {
2886                 printk(KERN_INFO "delayed_refs has NO entry\n");
2887                 return ret;
2888         }
2889
2890         node = rb_first(&delayed_refs->root);
2891         while (node) {
2892                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2893                 node = rb_next(node);
2894
2895                 ref->in_tree = 0;
2896                 rb_erase(&ref->rb_node, &delayed_refs->root);
2897                 delayed_refs->num_entries--;
2898
2899                 atomic_set(&ref->refs, 1);
2900                 if (btrfs_delayed_ref_is_head(ref)) {
2901                         struct btrfs_delayed_ref_head *head;
2902
2903                         head = btrfs_delayed_node_to_head(ref);
2904                         mutex_lock(&head->mutex);
2905                         kfree(head->extent_op);
2906                         delayed_refs->num_heads--;
2907                         if (list_empty(&head->cluster))
2908                                 delayed_refs->num_heads_ready--;
2909                         list_del_init(&head->cluster);
2910                         mutex_unlock(&head->mutex);
2911                 }
2912
2913                 spin_unlock(&delayed_refs->lock);
2914                 btrfs_put_delayed_ref(ref);
2915
2916                 cond_resched();
2917                 spin_lock(&delayed_refs->lock);
2918         }
2919
2920         spin_unlock(&delayed_refs->lock);
2921
2922         return ret;
2923 }
2924
2925 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
2926 {
2927         struct btrfs_pending_snapshot *snapshot;
2928         struct list_head splice;
2929
2930         INIT_LIST_HEAD(&splice);
2931
2932         list_splice_init(&t->pending_snapshots, &splice);
2933
2934         while (!list_empty(&splice)) {
2935                 snapshot = list_entry(splice.next,
2936                                       struct btrfs_pending_snapshot,
2937                                       list);
2938
2939                 list_del_init(&snapshot->list);
2940
2941                 kfree(snapshot);
2942         }
2943
2944         return 0;
2945 }
2946
2947 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
2948 {
2949         struct btrfs_inode *btrfs_inode;
2950         struct list_head splice;
2951
2952         INIT_LIST_HEAD(&splice);
2953
2954         list_splice_init(&root->fs_info->delalloc_inodes, &splice);
2955
2956         spin_lock(&root->fs_info->delalloc_lock);
2957
2958         while (!list_empty(&splice)) {
2959                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2960                                     delalloc_inodes);
2961
2962                 list_del_init(&btrfs_inode->delalloc_inodes);
2963
2964                 btrfs_invalidate_inodes(btrfs_inode->root);
2965         }
2966
2967         spin_unlock(&root->fs_info->delalloc_lock);
2968
2969         return 0;
2970 }
2971
2972 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
2973                                         struct extent_io_tree *dirty_pages,
2974                                         int mark)
2975 {
2976         int ret;
2977         struct page *page;
2978         struct inode *btree_inode = root->fs_info->btree_inode;
2979         struct extent_buffer *eb;
2980         u64 start = 0;
2981         u64 end;
2982         u64 offset;
2983         unsigned long index;
2984
2985         while (1) {
2986                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
2987                                             mark);
2988                 if (ret)
2989                         break;
2990
2991                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
2992                 while (start <= end) {
2993                         index = start >> PAGE_CACHE_SHIFT;
2994                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
2995                         page = find_get_page(btree_inode->i_mapping, index);
2996                         if (!page)
2997                                 continue;
2998                         offset = page_offset(page);
2999
3000                         spin_lock(&dirty_pages->buffer_lock);
3001                         eb = radix_tree_lookup(
3002                              &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3003                                                offset >> PAGE_CACHE_SHIFT);
3004                         spin_unlock(&dirty_pages->buffer_lock);
3005                         if (eb) {
3006                                 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3007                                                          &eb->bflags);
3008                                 atomic_set(&eb->refs, 1);
3009                         }
3010                         if (PageWriteback(page))
3011                                 end_page_writeback(page);
3012
3013                         lock_page(page);
3014                         if (PageDirty(page)) {
3015                                 clear_page_dirty_for_io(page);
3016                                 spin_lock_irq(&page->mapping->tree_lock);
3017                                 radix_tree_tag_clear(&page->mapping->page_tree,
3018                                                         page_index(page),
3019                                                         PAGECACHE_TAG_DIRTY);
3020                                 spin_unlock_irq(&page->mapping->tree_lock);
3021                         }
3022
3023                         page->mapping->a_ops->invalidatepage(page, 0);
3024                         unlock_page(page);
3025                 }
3026         }
3027
3028         return ret;
3029 }
3030
3031 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3032                                        struct extent_io_tree *pinned_extents)
3033 {
3034         struct extent_io_tree *unpin;
3035         u64 start;
3036         u64 end;
3037         int ret;
3038
3039         unpin = pinned_extents;
3040         while (1) {
3041                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3042                                             EXTENT_DIRTY);
3043                 if (ret)
3044                         break;
3045
3046                 /* opt_discard */
3047                 ret = btrfs_error_discard_extent(root, start, end + 1 - start);
3048
3049                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3050                 btrfs_error_unpin_extent_range(root, start, end);
3051                 cond_resched();
3052         }
3053
3054         return 0;
3055 }
3056
3057 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3058 {
3059         struct btrfs_transaction *t;
3060         LIST_HEAD(list);
3061
3062         WARN_ON(1);
3063
3064         mutex_lock(&root->fs_info->trans_mutex);
3065         mutex_lock(&root->fs_info->transaction_kthread_mutex);
3066
3067         list_splice_init(&root->fs_info->trans_list, &list);
3068         while (!list_empty(&list)) {
3069                 t = list_entry(list.next, struct btrfs_transaction, list);
3070                 if (!t)
3071                         break;
3072
3073                 btrfs_destroy_ordered_operations(root);
3074
3075                 btrfs_destroy_ordered_extents(root);
3076
3077                 btrfs_destroy_delayed_refs(t, root);
3078
3079                 btrfs_block_rsv_release(root,
3080                                         &root->fs_info->trans_block_rsv,
3081                                         t->dirty_pages.dirty_bytes);
3082
3083                 /* FIXME: cleanup wait for commit */
3084                 t->in_commit = 1;
3085                 t->blocked = 1;
3086                 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3087                         wake_up(&root->fs_info->transaction_blocked_wait);
3088
3089                 t->blocked = 0;
3090                 if (waitqueue_active(&root->fs_info->transaction_wait))
3091                         wake_up(&root->fs_info->transaction_wait);
3092                 mutex_unlock(&root->fs_info->trans_mutex);
3093
3094                 mutex_lock(&root->fs_info->trans_mutex);
3095                 t->commit_done = 1;
3096                 if (waitqueue_active(&t->commit_wait))
3097                         wake_up(&t->commit_wait);
3098                 mutex_unlock(&root->fs_info->trans_mutex);
3099
3100                 mutex_lock(&root->fs_info->trans_mutex);
3101
3102                 btrfs_destroy_pending_snapshots(t);
3103
3104                 btrfs_destroy_delalloc_inodes(root);
3105
3106                 spin_lock(&root->fs_info->new_trans_lock);
3107                 root->fs_info->running_transaction = NULL;
3108                 spin_unlock(&root->fs_info->new_trans_lock);
3109
3110                 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3111                                              EXTENT_DIRTY);
3112
3113                 btrfs_destroy_pinned_extent(root,
3114                                             root->fs_info->pinned_extents);
3115
3116                 t->use_count = 0;
3117                 list_del_init(&t->list);
3118                 memset(t, 0, sizeof(*t));
3119                 kmem_cache_free(btrfs_transaction_cachep, t);
3120         }
3121
3122         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3123         mutex_unlock(&root->fs_info->trans_mutex);
3124
3125         return 0;
3126 }
3127
3128 static struct extent_io_ops btree_extent_io_ops = {
3129         .write_cache_pages_lock_hook = btree_lock_page_hook,
3130         .readpage_end_io_hook = btree_readpage_end_io_hook,
3131         .submit_bio_hook = btree_submit_bio_hook,
3132         /* note we're sharing with inode.c for the merge bio hook */
3133         .merge_bio_hook = btrfs_merge_bio_hook,
3134 };