fs/btrfs/disk-io.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/fs.h>
  20 #include <linux/blkdev.h>
  21 #include <linux/scatterlist.h>
  22 #include <linux/swap.h>
  23 #include <linux/radix-tree.h>
  24 #include <linux/writeback.h>
  25 #include <linux/buffer_head.h>
  26 #include <linux/workqueue.h>
  27 #include <linux/kthread.h>
  28 #include <linux/freezer.h>
  29 #include <linux/crc32c.h>
  30 #include <linux/slab.h>
  31 #include <linux/migrate.h>
  32 #include <linux/ratelimit.h>
  33 #include <asm/unaligned.h>
  34 #include "compat.h"
  35 #include "ctree.h"
  36 #include "disk-io.h"
  37 #include "transaction.h"
  38 #include "btrfs_inode.h"
  39 #include "volumes.h"
  40 #include "print-tree.h"
  41 #include "async-thread.h"
  42 #include "locking.h"
  43 #include "tree-log.h"
  44 #include "free-space-cache.h"
  45 #include "inode-map.h"
  46
  47 static struct extent_io_ops btree_extent_io_ops;
  48 static void end_workqueue_fn(struct btrfs_work *work);
  49 static void free_fs_root(struct btrfs_root *root);
  50 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  51                                     int read_only);
  52 static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
  53 static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
  54 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  55                                       struct btrfs_root *root);
  56 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
  57 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  58 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
  59                                         struct extent_io_tree *dirty_pages,
  60                                         int mark);
  61 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
  62                                        struct extent_io_tree *pinned_extents);
  63 static int btrfs_cleanup_transaction(struct btrfs_root *root);
  64
  65 /*
  66  * end_io_wq structs are used to do processing in task context when an IO is
  67  * complete.  This is used during reads to verify checksums, and it is used
  68  * by writes to insert metadata for new file extents after IO is complete.
  69  */
  70 struct end_io_wq {
  71         struct bio *bio;
  72         bio_end_io_t *end_io;
  73         void *private;
  74         struct btrfs_fs_info *info;
  75         int error;
  76         int metadata;
  77         struct list_head list;
  78         struct btrfs_work work;
  79 };
  80
  81 /*
  82  * async submit bios are used to offload expensive checksumming
  83  * onto the worker threads.  They checksum file and metadata bios
  84  * just before they are sent down the IO stack.
  85  */
  86 struct async_submit_bio {
  87         struct inode *inode;
  88         struct bio *bio;
  89         struct list_head list;
  90         extent_submit_bio_hook_t *submit_bio_start;
  91         extent_submit_bio_hook_t *submit_bio_done;
  92         int rw;
  93         int mirror_num;
  94         unsigned long bio_flags;
  95         /*
  96          * bio_offset is optional, can be used if the pages in the bio
  97          * can't tell us where in the file the bio should go
  98          */
  99         u64 bio_offset;
 100         struct btrfs_work work;
 101 };
 102
 103 /* These are used to set the lockdep class on the extent buffer locks.
 104  * The class is set by the readpage_end_io_hook after the buffer has
 105  * passed csum validation but before the pages are unlocked.
 106  *
 107  * The lockdep class is also set by btrfs_init_new_buffer on freshly
 108  * allocated blocks.
 109  *
 110  * The class is based on the level in the tree block, which allows lockdep
 111  * to know that lower nodes nest inside the locks of higher nodes.
 112  *
 113  * We also add a check to make sure the highest level of the tree is
 114  * the same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this
 115  * code needs update as well.
 116  */
 117 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 118 # if BTRFS_MAX_LEVEL != 8
 119 #  error
 120 # endif
 121 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
 122 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
 123         /* leaf */
 124         "btrfs-extent-00",
 125         "btrfs-extent-01",
 126         "btrfs-extent-02",
 127         "btrfs-extent-03",
 128         "btrfs-extent-04",
 129         "btrfs-extent-05",
 130         "btrfs-extent-06",
 131         "btrfs-extent-07",
 132         /* highest possible level */
 133         "btrfs-extent-08",
 134 };
 135 #endif
 136
 137 /*
 138  * extents on the btree inode are pretty simple, there's one extent
 139  * that covers the entire device
 140  */
 141 static struct extent_map *btree_get_extent(struct inode *inode,
 142                 struct page *page, size_t pg_offset, u64 start, u64 len,
 143                 int create)
 144 {
 145         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 146         struct extent_map *em;
 147         int ret;
 148
 149         read_lock(&em_tree->lock);
 150         em = lookup_extent_mapping(em_tree, start, len);
 151         if (em) {
 152                 em->bdev =
 153                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 154                 read_unlock(&em_tree->lock);
 155                 goto out;
 156         }
 157         read_unlock(&em_tree->lock);
 158
 159         em = alloc_extent_map();
 160         if (!em) {
 161                 em = ERR_PTR(-ENOMEM);
 162                 goto out;
 163         }
 164         em->start = 0;
 165         em->len = (u64)-1;
 166         em->block_len = (u64)-1;
 167         em->block_start = 0;
 168         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 169
 170         write_lock(&em_tree->lock);
 171         ret = add_extent_mapping(em_tree, em);
 172         if (ret == -EEXIST) {
 173                 u64 failed_start = em->start;
 174                 u64 failed_len = em->len;
 175
 176                 free_extent_map(em);
 177                 em = lookup_extent_mapping(em_tree, start, len);
 178                 if (em) {
 179                         ret = 0;
 180                 } else {
 181                         em = lookup_extent_mapping(em_tree, failed_start,
 182                                                    failed_len);
 183                         ret = -EIO;
 184                 }
 185         } else if (ret) {
 186                 free_extent_map(em);
 187                 em = NULL;
 188         }
 189         write_unlock(&em_tree->lock);
 190
 191         if (ret)
 192                 em = ERR_PTR(ret);
 193 out:
 194         return em;
 195 }
 196
 197 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 198 {
 199         return crc32c(seed, data, len);
 200 }
 201
 202 void btrfs_csum_final(u32 crc, char *result)
 203 {
 204         put_unaligned_le32(~crc, result);
 205 }
 206
 207 /*
 208  * compute the csum for a btree block, and either verify it or write it
 209  * into the csum field of the block.
 210  */
 211 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 212                            int verify)
 213 {
 214         u16 csum_size =
 215                 btrfs_super_csum_size(&root->fs_info->super_copy);
 216         char *result = NULL;
 217         unsigned long len;
 218         unsigned long cur_len;
 219         unsigned long offset = BTRFS_CSUM_SIZE;
 220         char *map_token = NULL;
 221         char *kaddr;
 222         unsigned long map_start;
 223         unsigned long map_len;
 224         int err;
 225         u32 crc = ~(u32)0;
 226         unsigned long inline_result;
 227
 228         len = buf->len - offset;
 229         while (len > 0) {
 230                 err = map_private_extent_buffer(buf, offset, 32,
 231                                         &map_token, &kaddr,
 232                                         &map_start, &map_len, KM_USER0);
 233                 if (err)
 234                         return 1;
 235                 cur_len = min(len, map_len - (offset - map_start));
 236                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 237                                       crc, cur_len);
 238                 len -= cur_len;
 239                 offset += cur_len;
 240                 unmap_extent_buffer(buf, map_token, KM_USER0);
 241         }
 242         if (csum_size > sizeof(inline_result)) {
 243                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
 244                 if (!result)
 245                         return 1;
 246         } else {
 247                 result = (char *)&inline_result;
 248         }
 249
 250         btrfs_csum_final(crc, result);
 251
 252         if (verify) {
 253                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 254                         u32 val;
 255                         u32 found = 0;
 256                         memcpy(&found, result, csum_size);
 257
 258                         read_extent_buffer(buf, &val, 0, csum_size);
 259                         printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
 260                                        "failed on %llu wanted %X found %X "
 261                                        "level %d\n",
 262                                        root->fs_info->sb->s_id,
 263                                        (unsigned long long)buf->start, val, found,
 264                                        btrfs_header_level(buf));
 265                         if (result != (char *)&inline_result)
 266                                 kfree(result);
 267                         return 1;
 268                 }
 269         } else {
 270                 write_extent_buffer(buf, result, 0, csum_size);
 271         }
 272         if (result != (char *)&inline_result)
 273                 kfree(result);
 274         return 0;
 275 }
 276
 277 /*
 278  * we can't consider a given block up to date unless the transid of the
 279  * block matches the transid in the parent node's pointer.  This is how we
 280  * detect blocks that either didn't get written at all or got written
 281  * in the wrong place.
 282  */
 283 static int verify_parent_transid(struct extent_io_tree *io_tree,
 284                                  struct extent_buffer *eb, u64 parent_transid)
 285 {
 286         struct extent_state *cached_state = NULL;
 287         int ret;
 288
 289         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 290                 return 0;
 291
 292         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 293                          0, &cached_state, GFP_NOFS);
 294         if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
 295             btrfs_header_generation(eb) == parent_transid) {
 296                 ret = 0;
 297                 goto out;
 298         }
 299         printk_ratelimited("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         ret = 1;
 305         clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
 306 out:
 307         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 308                              &cached_state, GFP_NOFS);
 309         return ret;
 310 }
 311
 312 /*
 313  * helper to read a given tree block, doing retries as required when
 314  * the checksums don't match and we have alternate mirrors to try.
 315  */
 316 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 317                                           struct extent_buffer *eb,
 318                                           u64 start, u64 parent_transid)
 319 {
 320         struct extent_io_tree *io_tree;
 321         int ret;
 322         int num_copies = 0;
 323         int mirror_num = 0;
 324
 325         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 326         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 327         while (1) {
 328                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 329                                                btree_get_extent, mirror_num);
 330                 if (!ret &&
 331                     !verify_parent_transid(io_tree, eb, parent_transid))
 332                         return ret;
 333
 334                 /*
 335                  * This buffer's crc is fine, but its contents are corrupted, so
 336                  * there is no reason to read the other copies, they won't be
 337                  * any less wrong.
 338                  */
 339                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 340                         return ret;
 341
 342                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 343                                               eb->start, eb->len);
 344                 if (num_copies == 1)
 345                         return ret;
 346
 347                 mirror_num++;
 348                 if (mirror_num > num_copies)
 349                         return ret;
 350         }
 351         return -EIO;
 352 }
 353
 354 /*
 355  * checksum a dirty tree block before IO.  This has extra checks to make sure
 356  * we only fill in the checksum field in the first page of a multi-page block
 357  */
 358
 359 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 360 {
 361         struct extent_io_tree *tree;
 362         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 363         u64 found_start;
 364         unsigned long len;
 365         struct extent_buffer *eb;
 366         int ret;
 367
 368         tree = &BTRFS_I(page->mapping->host)->io_tree;
 369
 370         if (page->private == EXTENT_PAGE_PRIVATE) {
 371                 WARN_ON(1);
 372                 goto out;
 373         }
 374         if (!page->private) {
 375                 WARN_ON(1);
 376                 goto out;
 377         }
 378         len = page->private >> 2;
 379         WARN_ON(len == 0);
 380
 381         eb = alloc_extent_buffer(tree, start, len, page);
 382         if (eb == NULL) {
 383                 WARN_ON(1);
 384                 goto out;
 385         }
 386         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 387                                              btrfs_header_generation(eb));
 388         BUG_ON(ret);
 389         WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
 390
 391         found_start = btrfs_header_bytenr(eb);
 392         if (found_start != start) {
 393                 WARN_ON(1);
 394                 goto err;
 395         }
 396         if (eb->first_page != page) {
 397                 WARN_ON(1);
 398                 goto err;
 399         }
 400         if (!PageUptodate(page)) {
 401                 WARN_ON(1);
 402                 goto err;
 403         }
 404         csum_tree_block(root, eb, 0);
 405 err:
 406         free_extent_buffer(eb);
 407 out:
 408         return 0;
 409 }
 410
 411 static int check_tree_block_fsid(struct btrfs_root *root,
 412                                  struct extent_buffer *eb)
 413 {
 414         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 415         u8 fsid[BTRFS_UUID_SIZE];
 416         int ret = 1;
 417
 418         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
 419                            BTRFS_FSID_SIZE);
 420         while (fs_devices) {
 421                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 422                         ret = 0;
 423                         break;
 424                 }
 425                 fs_devices = fs_devices->seed;
 426         }
 427         return ret;
 428 }
 429
 430 #define CORRUPT(reason, eb, root, slot)                         \
 431         printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
 432                "root=%llu, slot=%d\n", reason,                  \
 433                (unsigned long long)btrfs_header_bytenr(eb),     \
 434                (unsigned long long)root->objectid, slot)
 435
 436 static noinline int check_leaf(struct btrfs_root *root,
 437                                struct extent_buffer *leaf)
 438 {
 439         struct btrfs_key key;
 440         struct btrfs_key leaf_key;
 441         u32 nritems = btrfs_header_nritems(leaf);
 442         int slot;
 443
 444         if (nritems == 0)
 445                 return 0;
 446
 447         /* Check the 0 item */
 448         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 449             BTRFS_LEAF_DATA_SIZE(root)) {
 450                 CORRUPT("invalid item offset size pair", leaf, root, 0);
 451                 return -EIO;
 452         }
 453
 454         /*
 455          * Check to make sure each items keys are in the correct order and their
 456          * offsets make sense.  We only have to loop through nritems-1 because
 457          * we check the current slot against the next slot, which verifies the
 458          * next slot's offset+size makes sense and that the current's slot
 459          * offset is correct.
 460          */
 461         for (slot = 0; slot < nritems - 1; slot++) {
 462                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 463                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 464
 465                 /* Make sure the keys are in the right order */
 466                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 467                         CORRUPT("bad key order", leaf, root, slot);
 468                         return -EIO;
 469                 }
 470
 471                 /*
 472                  * Make sure the offset and ends are right, remember that the
 473                  * item data starts at the end of the leaf and grows towards the
 474                  * front.
 475                  */
 476                 if (btrfs_item_offset_nr(leaf, slot) !=
 477                         btrfs_item_end_nr(leaf, slot + 1)) {
 478                         CORRUPT("slot offset bad", leaf, root, slot);
 479                         return -EIO;
 480                 }
 481
 482                 /*
 483                  * Check to make sure that we don't point outside of the leaf,
 484                  * just incase all the items are consistent to eachother, but
 485                  * all point outside of the leaf.
 486                  */
 487                 if (btrfs_item_end_nr(leaf, slot) >
 488                     BTRFS_LEAF_DATA_SIZE(root)) {
 489                         CORRUPT("slot end outside of leaf", leaf, root, slot);
 490                         return -EIO;
 491                 }
 492         }
 493
 494         return 0;
 495 }
 496
 497 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 498 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
 499 {
 500         lockdep_set_class_and_name(&eb->lock,
 501                            &btrfs_eb_class[level],
 502                            btrfs_eb_name[level]);
 503 }
 504 #endif
 505
 506 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 507                                struct extent_state *state)
 508 {
 509         struct extent_io_tree *tree;
 510         u64 found_start;
 511         int found_level;
 512         unsigned long len;
 513         struct extent_buffer *eb;
 514         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 515         int ret = 0;
 516
 517         tree = &BTRFS_I(page->mapping->host)->io_tree;
 518         if (page->private == EXTENT_PAGE_PRIVATE)
 519                 goto out;
 520         if (!page->private)
 521                 goto out;
 522
 523         len = page->private >> 2;
 524         WARN_ON(len == 0);
 525
 526         eb = alloc_extent_buffer(tree, start, len, page);
 527         if (eb == NULL) {
 528                 ret = -EIO;
 529                 goto out;
 530         }
 531
 532         found_start = btrfs_header_bytenr(eb);
 533         if (found_start != start) {
 534                 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
 535                                "%llu %llu\n",
 536                                (unsigned long long)found_start,
 537                                (unsigned long long)eb->start);
 538                 ret = -EIO;
 539                 goto err;
 540         }
 541         if (eb->first_page != page) {
 542                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
 543                        eb->first_page->index, page->index);
 544                 WARN_ON(1);
 545                 ret = -EIO;
 546                 goto err;
 547         }
 548         if (check_tree_block_fsid(root, eb)) {
 549                 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
 550                                (unsigned long long)eb->start);
 551                 ret = -EIO;
 552                 goto err;
 553         }
 554         found_level = btrfs_header_level(eb);
 555
 556         btrfs_set_buffer_lockdep_class(eb, found_level);
 557
 558         ret = csum_tree_block(root, eb, 1);
 559         if (ret) {
 560                 ret = -EIO;
 561                 goto err;
 562         }
 563
 564         /*
 565          * If this is a leaf block and it is corrupt, set the corrupt bit so
 566          * that we don't try and read the other copies of this block, just
 567          * return -EIO.
 568          */
 569         if (found_level == 0 && check_leaf(root, eb)) {
 570                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 571                 ret = -EIO;
 572         }
 573
 574         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 575         end = eb->start + end - 1;
 576 err:
 577         free_extent_buffer(eb);
 578 out:
 579         return ret;
 580 }
 581
 582 static void end_workqueue_bio(struct bio *bio, int err)
 583 {
 584         struct end_io_wq *end_io_wq = bio->bi_private;
 585         struct btrfs_fs_info *fs_info;
 586
 587         fs_info = end_io_wq->info;
 588         end_io_wq->error = err;
 589         end_io_wq->work.func = end_workqueue_fn;
 590         end_io_wq->work.flags = 0;
 591
 592         if (bio->bi_rw & REQ_WRITE) {
 593                 if (end_io_wq->metadata == 1)
 594                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
 595                                            &end_io_wq->work);
 596                 else if (end_io_wq->metadata == 2)
 597                         btrfs_queue_worker(&fs_info->endio_freespace_worker,
 598                                            &end_io_wq->work);
 599                 else
 600                         btrfs_queue_worker(&fs_info->endio_write_workers,
 601                                            &end_io_wq->work);
 602         } else {
 603                 if (end_io_wq->metadata)
 604                         btrfs_queue_worker(&fs_info->endio_meta_workers,
 605                                            &end_io_wq->work);
 606                 else
 607                         btrfs_queue_worker(&fs_info->endio_workers,
 608                                            &end_io_wq->work);
 609         }
 610 }
 611
 612 /*
 613  * For the metadata arg you want
 614  *
 615  * 0 - if data
 616  * 1 - if normal metadta
 617  * 2 - if writing to the free space cache area
 618  */
 619 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 620                         int metadata)
 621 {
 622         struct end_io_wq *end_io_wq;
 623         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 624         if (!end_io_wq)
 625                 return -ENOMEM;
 626
 627         end_io_wq->private = bio->bi_private;
 628         end_io_wq->end_io = bio->bi_end_io;
 629         end_io_wq->info = info;
 630         end_io_wq->error = 0;
 631         end_io_wq->bio = bio;
 632         end_io_wq->metadata = metadata;
 633
 634         bio->bi_private = end_io_wq;
 635         bio->bi_end_io = end_workqueue_bio;
 636         return 0;
 637 }
 638
 639 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 640 {
 641         unsigned long limit = min_t(unsigned long,
 642                                     info->workers.max_workers,
 643                                     info->fs_devices->open_devices);
 644         return 256 * limit;
 645 }
 646
 647 static void run_one_async_start(struct btrfs_work *work)
 648 {
 649         struct async_submit_bio *async;
 650
 651         async = container_of(work, struct  async_submit_bio, work);
 652         async->submit_bio_start(async->inode, async->rw, async->bio,
 653                                async->mirror_num, async->bio_flags,
 654                                async->bio_offset);
 655 }
 656
 657 static void run_one_async_done(struct btrfs_work *work)
 658 {
 659         struct btrfs_fs_info *fs_info;
 660         struct async_submit_bio *async;
 661         int limit;
 662
 663         async = container_of(work, struct  async_submit_bio, work);
 664         fs_info = BTRFS_I(async->inode)->root->fs_info;
 665
 666         limit = btrfs_async_submit_limit(fs_info);
 667         limit = limit * 2 / 3;
 668
 669         atomic_dec(&fs_info->nr_async_submits);
 670
 671         if (atomic_read(&fs_info->nr_async_submits) < limit &&
 672             waitqueue_active(&fs_info->async_submit_wait))
 673                 wake_up(&fs_info->async_submit_wait);
 674
 675         async->submit_bio_done(async->inode, async->rw, async->bio,
 676                                async->mirror_num, async->bio_flags,
 677                                async->bio_offset);
 678 }
 679
 680 static void run_one_async_free(struct btrfs_work *work)
 681 {
 682         struct async_submit_bio *async;
 683
 684         async = container_of(work, struct  async_submit_bio, work);
 685         kfree(async);
 686 }
 687
 688 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 689                         int rw, struct bio *bio, int mirror_num,
 690                         unsigned long bio_flags,
 691                         u64 bio_offset,
 692                         extent_submit_bio_hook_t *submit_bio_start,
 693                         extent_submit_bio_hook_t *submit_bio_done)
 694 {
 695         struct async_submit_bio *async;
 696
 697         async = kmalloc(sizeof(*async), GFP_NOFS);
 698         if (!async)
 699                 return -ENOMEM;
 700
 701         async->inode = inode;
 702         async->rw = rw;
 703         async->bio = bio;
 704         async->mirror_num = mirror_num;
 705         async->submit_bio_start = submit_bio_start;
 706         async->submit_bio_done = submit_bio_done;
 707
 708         async->work.func = run_one_async_start;
 709         async->work.ordered_func = run_one_async_done;
 710         async->work.ordered_free = run_one_async_free;
 711
 712         async->work.flags = 0;
 713         async->bio_flags = bio_flags;
 714         async->bio_offset = bio_offset;
 715
 716         atomic_inc(&fs_info->nr_async_submits);
 717
 718         if (rw & REQ_SYNC)
 719                 btrfs_set_work_high_prio(&async->work);
 720
 721         btrfs_queue_worker(&fs_info->workers, &async->work);
 722
 723         while (atomic_read(&fs_info->async_submit_draining) &&
 724               atomic_read(&fs_info->nr_async_submits)) {
 725                 wait_event(fs_info->async_submit_wait,
 726                            (atomic_read(&fs_info->nr_async_submits) == 0));
 727         }
 728
 729         return 0;
 730 }
 731
 732 static int btree_csum_one_bio(struct bio *bio)
 733 {
 734         struct bio_vec *bvec = bio->bi_io_vec;
 735         int bio_index = 0;
 736         struct btrfs_root *root;
 737
 738         WARN_ON(bio->bi_vcnt <= 0);
 739         while (bio_index < bio->bi_vcnt) {
 740                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 741                 csum_dirty_buffer(root, bvec->bv_page);
 742                 bio_index++;
 743                 bvec++;
 744         }
 745         return 0;
 746 }
 747
 748 static int __btree_submit_bio_start(struct inode *inode, int rw,
 749                                     struct bio *bio, int mirror_num,
 750                                     unsigned long bio_flags,
 751                                     u64 bio_offset)
 752 {
 753         /*
 754          * when we're called for a write, we're already in the async
 755          * submission context.  Just jump into btrfs_map_bio
 756          */
 757         btree_csum_one_bio(bio);
 758         return 0;
 759 }
 760
 761 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 762                                  int mirror_num, unsigned long bio_flags,
 763                                  u64 bio_offset)
 764 {
 765         /*
 766          * when we're called for a write, we're already in the async
 767          * submission context.  Just jump into btrfs_map_bio
 768          */
 769         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 770 }
 771
 772 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 773                                  int mirror_num, unsigned long bio_flags,
 774                                  u64 bio_offset)
 775 {
 776         int ret;
 777
 778         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 779                                           bio, 1);
 780         BUG_ON(ret);
 781
 782         if (!(rw & REQ_WRITE)) {
 783                 /*
 784                  * called for a read, do the setup so that checksum validation
 785                  * can happen in the async kernel threads
 786                  */
 787                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 788                                      mirror_num, 0);
 789         }
 790
 791         /*
 792          * kthread helpers are used to submit writes so that checksumming
 793          * can happen in parallel across all CPUs
 794          */
 795         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 796                                    inode, rw, bio, mirror_num, 0,
 797                                    bio_offset,
 798                                    __btree_submit_bio_start,
 799                                    __btree_submit_bio_done);
 800 }
 801
 802 #ifdef CONFIG_MIGRATION
 803 static int btree_migratepage(struct address_space *mapping,
 804                         struct page *newpage, struct page *page)
 805 {
 806         /*
 807          * we can't safely write a btree page from here,
 808          * we haven't done the locking hook
 809          */
 810         if (PageDirty(page))
 811                 return -EAGAIN;
 812         /*
 813          * Buffers may be managed in a filesystem specific way.
 814          * We must have no buffers or drop them.
 815          */
 816         if (page_has_private(page) &&
 817             !try_to_release_page(page, GFP_KERNEL))
 818                 return -EAGAIN;
 819         return migrate_page(mapping, newpage, page);
 820 }
 821 #endif
 822
 823 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 824 {
 825         struct extent_io_tree *tree;
 826         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 827         struct extent_buffer *eb;
 828         int was_dirty;
 829
 830         tree = &BTRFS_I(page->mapping->host)->io_tree;
 831         if (!(current->flags & PF_MEMALLOC)) {
 832                 return extent_write_full_page(tree, page,
 833                                               btree_get_extent, wbc);
 834         }
 835
 836         redirty_page_for_writepage(wbc, page);
 837         eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
 838         WARN_ON(!eb);
 839
 840         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
 841         if (!was_dirty) {
 842                 spin_lock(&root->fs_info->delalloc_lock);
 843                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
 844                 spin_unlock(&root->fs_info->delalloc_lock);
 845         }
 846         free_extent_buffer(eb);
 847
 848         unlock_page(page);
 849         return 0;
 850 }
 851
 852 static int btree_writepages(struct address_space *mapping,
 853                             struct writeback_control *wbc)
 854 {
 855         struct extent_io_tree *tree;
 856         tree = &BTRFS_I(mapping->host)->io_tree;
 857         if (wbc->sync_mode == WB_SYNC_NONE) {
 858                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
 859                 u64 num_dirty;
 860                 unsigned long thresh = 32 * 1024 * 1024;
 861
 862                 if (wbc->for_kupdate)
 863                         return 0;
 864
 865                 /* this is a bit racy, but that's ok */
 866                 num_dirty = root->fs_info->dirty_metadata_bytes;
 867                 if (num_dirty < thresh)
 868                         return 0;
 869         }
 870         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 871 }
 872
 873 static int btree_readpage(struct file *file, struct page *page)
 874 {
 875         struct extent_io_tree *tree;
 876         tree = &BTRFS_I(page->mapping->host)->io_tree;
 877         return extent_read_full_page(tree, page, btree_get_extent);
 878 }
 879
 880 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 881 {
 882         struct extent_io_tree *tree;
 883         struct extent_map_tree *map;
 884         int ret;
 885
 886         if (PageWriteback(page) || PageDirty(page))
 887                 return 0;
 888
 889         tree = &BTRFS_I(page->mapping->host)->io_tree;
 890         map = &BTRFS_I(page->mapping->host)->extent_tree;
 891
 892         ret = try_release_extent_state(map, tree, page, gfp_flags);
 893         if (!ret)
 894                 return 0;
 895
 896         ret = try_release_extent_buffer(tree, page);
 897         if (ret == 1) {
 898                 ClearPagePrivate(page);
 899                 set_page_private(page, 0);
 900                 page_cache_release(page);
 901         }
 902
 903         return ret;
 904 }
 905
 906 static void btree_invalidatepage(struct page *page, unsigned long offset)
 907 {
 908         struct extent_io_tree *tree;
 909         tree = &BTRFS_I(page->mapping->host)->io_tree;
 910         extent_invalidatepage(tree, page, offset);
 911         btree_releasepage(page, GFP_NOFS);
 912         if (PagePrivate(page)) {
 913                 printk(KERN_WARNING "btrfs warning page private not zero "
 914                        "on page %llu\n", (unsigned long long)page_offset(page));
 915                 ClearPagePrivate(page);
 916                 set_page_private(page, 0);
 917                 page_cache_release(page);
 918         }
 919 }
 920
 921 static const struct address_space_operations btree_aops = {
 922         .readpage       = btree_readpage,
 923         .writepage      = btree_writepage,
 924         .writepages     = btree_writepages,
 925         .releasepage    = btree_releasepage,
 926         .invalidatepage = btree_invalidatepage,
 927 #ifdef CONFIG_MIGRATION
 928         .migratepage    = btree_migratepage,
 929 #endif
 930 };
 931
 932 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 933                          u64 parent_transid)
 934 {
 935         struct extent_buffer *buf = NULL;
 936         struct inode *btree_inode = root->fs_info->btree_inode;
 937         int ret = 0;
 938
 939         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 940         if (!buf)
 941                 return 0;
 942         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 943                                  buf, 0, 0, btree_get_extent, 0);
 944         free_extent_buffer(buf);
 945         return ret;
 946 }
 947
 948 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 949                                             u64 bytenr, u32 blocksize)
 950 {
 951         struct inode *btree_inode = root->fs_info->btree_inode;
 952         struct extent_buffer *eb;
 953         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 954                                 bytenr, blocksize);
 955         return eb;
 956 }
 957
 958 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 959                                                  u64 bytenr, u32 blocksize)
 960 {
 961         struct inode *btree_inode = root->fs_info->btree_inode;
 962         struct extent_buffer *eb;
 963
 964         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 965                                  bytenr, blocksize, NULL);
 966         return eb;
 967 }
 968
 969
 970 int btrfs_write_tree_block(struct extent_buffer *buf)
 971 {
 972         return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
 973                                         buf->start + buf->len - 1);
 974 }
 975
 976 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 977 {
 978         return filemap_fdatawait_range(buf->first_page->mapping,
 979                                        buf->start, buf->start + buf->len - 1);
 980 }
 981
 982 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 983                                       u32 blocksize, u64 parent_transid)
 984 {
 985         struct extent_buffer *buf = NULL;
 986         int ret;
 987
 988         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 989         if (!buf)
 990                 return NULL;
 991
 992         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 993
 994         if (ret == 0)
 995                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
 996         return buf;
 997
 998 }
 999
1000 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1001                      struct extent_buffer *buf)
1002 {
1003         struct inode *btree_inode = root->fs_info->btree_inode;
1004         if (btrfs_header_generation(buf) ==
1005             root->fs_info->running_transaction->transid) {
1006                 btrfs_assert_tree_locked(buf);
1007
1008                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1009                         spin_lock(&root->fs_info->delalloc_lock);
1010                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
1011                                 root->fs_info->dirty_metadata_bytes -= buf->len;
1012                         else
1013                                 WARN_ON(1);
1014                         spin_unlock(&root->fs_info->delalloc_lock);
1015                 }
1016
1017                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1018                 btrfs_set_lock_blocking(buf);
1019                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1020                                           buf);
1021         }
1022         return 0;
1023 }
1024
1025 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1026                         u32 stripesize, struct btrfs_root *root,
1027                         struct btrfs_fs_info *fs_info,
1028                         u64 objectid)
1029 {
1030         root->node = NULL;
1031         root->commit_root = NULL;
1032         root->sectorsize = sectorsize;
1033         root->nodesize = nodesize;
1034         root->leafsize = leafsize;
1035         root->stripesize = stripesize;
1036         root->ref_cows = 0;
1037         root->track_dirty = 0;
1038         root->in_radix = 0;
1039         root->orphan_item_inserted = 0;
1040         root->orphan_cleanup_state = 0;
1041
1042         root->fs_info = fs_info;
1043         root->objectid = objectid;
1044         root->last_trans = 0;
1045         root->highest_objectid = 0;
1046         root->name = NULL;
1047         root->in_sysfs = 0;
1048         root->inode_tree = RB_ROOT;
1049         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1050         root->block_rsv = NULL;
1051         root->orphan_block_rsv = NULL;
1052
1053         INIT_LIST_HEAD(&root->dirty_list);
1054         INIT_LIST_HEAD(&root->orphan_list);
1055         INIT_LIST_HEAD(&root->root_list);
1056         spin_lock_init(&root->orphan_lock);
1057         spin_lock_init(&root->inode_lock);
1058         spin_lock_init(&root->accounting_lock);
1059         mutex_init(&root->objectid_mutex);
1060         mutex_init(&root->log_mutex);
1061         init_waitqueue_head(&root->log_writer_wait);
1062         init_waitqueue_head(&root->log_commit_wait[0]);
1063         init_waitqueue_head(&root->log_commit_wait[1]);
1064         atomic_set(&root->log_commit[0], 0);
1065         atomic_set(&root->log_commit[1], 0);
1066         atomic_set(&root->log_writers, 0);
1067         root->log_batch = 0;
1068         root->log_transid = 0;
1069         root->last_log_commit = 0;
1070         extent_io_tree_init(&root->dirty_log_pages,
1071                              fs_info->btree_inode->i_mapping);
1072
1073         memset(&root->root_key, 0, sizeof(root->root_key));
1074         memset(&root->root_item, 0, sizeof(root->root_item));
1075         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1076         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1077         root->defrag_trans_start = fs_info->generation;
1078         init_completion(&root->kobj_unregister);
1079         root->defrag_running = 0;
1080         root->root_key.objectid = objectid;
1081         root->anon_super.s_root = NULL;
1082         root->anon_super.s_dev = 0;
1083         INIT_LIST_HEAD(&root->anon_super.s_list);
1084         INIT_LIST_HEAD(&root->anon_super.s_instances);
1085         init_rwsem(&root->anon_super.s_umount);
1086
1087         return 0;
1088 }
1089
1090 static int find_and_setup_root(struct btrfs_root *tree_root,
1091                                struct btrfs_fs_info *fs_info,
1092                                u64 objectid,
1093                                struct btrfs_root *root)
1094 {
1095         int ret;
1096         u32 blocksize;
1097         u64 generation;
1098
1099         __setup_root(tree_root->nodesize, tree_root->leafsize,
1100                      tree_root->sectorsize, tree_root->stripesize,
1101                      root, fs_info, objectid);
1102         ret = btrfs_find_last_root(tree_root, objectid,
1103                                    &root->root_item, &root->root_key);
1104         if (ret > 0)
1105                 return -ENOENT;
1106         BUG_ON(ret);
1107
1108         generation = btrfs_root_generation(&root->root_item);
1109         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1110         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1111                                      blocksize, generation);
1112         if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1113                 free_extent_buffer(root->node);
1114                 return -EIO;
1115         }
1116         root->commit_root = btrfs_root_node(root);
1117         return 0;
1118 }
1119
1120 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1121                                          struct btrfs_fs_info *fs_info)
1122 {
1123         struct btrfs_root *root;
1124         struct btrfs_root *tree_root = fs_info->tree_root;
1125         struct extent_buffer *leaf;
1126
1127         root = kzalloc(sizeof(*root), GFP_NOFS);
1128         if (!root)
1129                 return ERR_PTR(-ENOMEM);
1130
1131         __setup_root(tree_root->nodesize, tree_root->leafsize,
1132                      tree_root->sectorsize, tree_root->stripesize,
1133                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1134
1135         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1136         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1137         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1138         /*
1139          * log trees do not get reference counted because they go away
1140          * before a real commit is actually done.  They do store pointers
1141          * to file data extents, and those reference counts still get
1142          * updated (along with back refs to the log tree).
1143          */
1144         root->ref_cows = 0;
1145
1146         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1147                                       BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1148         if (IS_ERR(leaf)) {
1149                 kfree(root);
1150                 return ERR_CAST(leaf);
1151         }
1152
1153         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1154         btrfs_set_header_bytenr(leaf, leaf->start);
1155         btrfs_set_header_generation(leaf, trans->transid);
1156         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1157         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1158         root->node = leaf;
1159
1160         write_extent_buffer(root->node, root->fs_info->fsid,
1161                             (unsigned long)btrfs_header_fsid(root->node),
1162                             BTRFS_FSID_SIZE);
1163         btrfs_mark_buffer_dirty(root->node);
1164         btrfs_tree_unlock(root->node);
1165         return root;
1166 }
1167
1168 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1169                              struct btrfs_fs_info *fs_info)
1170 {
1171         struct btrfs_root *log_root;
1172
1173         log_root = alloc_log_tree(trans, fs_info);
1174         if (IS_ERR(log_root))
1175                 return PTR_ERR(log_root);
1176         WARN_ON(fs_info->log_root_tree);
1177         fs_info->log_root_tree = log_root;
1178         return 0;
1179 }
1180
1181 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1182                        struct btrfs_root *root)
1183 {
1184         struct btrfs_root *log_root;
1185         struct btrfs_inode_item *inode_item;
1186
1187         log_root = alloc_log_tree(trans, root->fs_info);
1188         if (IS_ERR(log_root))
1189                 return PTR_ERR(log_root);
1190
1191         log_root->last_trans = trans->transid;
1192         log_root->root_key.offset = root->root_key.objectid;
1193
1194         inode_item = &log_root->root_item.inode;
1195         inode_item->generation = cpu_to_le64(1);
1196         inode_item->size = cpu_to_le64(3);
1197         inode_item->nlink = cpu_to_le32(1);
1198         inode_item->nbytes = cpu_to_le64(root->leafsize);
1199         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1200
1201         btrfs_set_root_node(&log_root->root_item, log_root->node);
1202
1203         WARN_ON(root->log_root);
1204         root->log_root = log_root;
1205         root->log_transid = 0;
1206         root->last_log_commit = 0;
1207         return 0;
1208 }
1209
1210 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1211                                                struct btrfs_key *location)
1212 {
1213         struct btrfs_root *root;
1214         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1215         struct btrfs_path *path;
1216         struct extent_buffer *l;
1217         u64 generation;
1218         u32 blocksize;
1219         int ret = 0;
1220
1221         root = kzalloc(sizeof(*root), GFP_NOFS);
1222         if (!root)
1223                 return ERR_PTR(-ENOMEM);
1224         if (location->offset == (u64)-1) {
1225                 ret = find_and_setup_root(tree_root, fs_info,
1226                                           location->objectid, root);
1227                 if (ret) {
1228                         kfree(root);
1229                         return ERR_PTR(ret);
1230                 }
1231                 goto out;
1232         }
1233
1234         __setup_root(tree_root->nodesize, tree_root->leafsize,
1235                      tree_root->sectorsize, tree_root->stripesize,
1236                      root, fs_info, location->objectid);
1237
1238         path = btrfs_alloc_path();
1239         if (!path) {
1240                 kfree(root);
1241                 return ERR_PTR(-ENOMEM);
1242         }
1243         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1244         if (ret == 0) {
1245                 l = path->nodes[0];
1246                 read_extent_buffer(l, &root->root_item,
1247                                 btrfs_item_ptr_offset(l, path->slots[0]),
1248                                 sizeof(root->root_item));
1249                 memcpy(&root->root_key, location, sizeof(*location));
1250         }
1251         btrfs_free_path(path);
1252         if (ret) {
1253                 kfree(root);
1254                 if (ret > 0)
1255                         ret = -ENOENT;
1256                 return ERR_PTR(ret);
1257         }
1258
1259         generation = btrfs_root_generation(&root->root_item);
1260         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1261         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1262                                      blocksize, generation);
1263         root->commit_root = btrfs_root_node(root);
1264         BUG_ON(!root->node);
1265 out:
1266         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1267                 root->ref_cows = 1;
1268                 btrfs_check_and_init_root_item(&root->root_item);
1269         }
1270
1271         return root;
1272 }
1273
1274 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1275                                               struct btrfs_key *location)
1276 {
1277         struct btrfs_root *root;
1278         int ret;
1279
1280         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1281                 return fs_info->tree_root;
1282         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1283                 return fs_info->extent_root;
1284         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1285                 return fs_info->chunk_root;
1286         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1287                 return fs_info->dev_root;
1288         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1289                 return fs_info->csum_root;
1290 again:
1291         spin_lock(&fs_info->fs_roots_radix_lock);
1292         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1293                                  (unsigned long)location->objectid);
1294         spin_unlock(&fs_info->fs_roots_radix_lock);
1295         if (root)
1296                 return root;
1297
1298         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1299         if (IS_ERR(root))
1300                 return root;
1301
1302         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1303         if (!root->free_ino_ctl)
1304                 goto fail;
1305         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1306                                         GFP_NOFS);
1307         if (!root->free_ino_pinned)
1308                 goto fail;
1309
1310         btrfs_init_free_ino_ctl(root);
1311         mutex_init(&root->fs_commit_mutex);
1312         spin_lock_init(&root->cache_lock);
1313         init_waitqueue_head(&root->cache_wait);
1314
1315         set_anon_super(&root->anon_super, NULL);
1316
1317         if (btrfs_root_refs(&root->root_item) == 0) {
1318                 ret = -ENOENT;
1319                 goto fail;
1320         }
1321
1322         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1323         if (ret < 0)
1324                 goto fail;
1325         if (ret == 0)
1326                 root->orphan_item_inserted = 1;
1327
1328         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1329         if (ret)
1330                 goto fail;
1331
1332         spin_lock(&fs_info->fs_roots_radix_lock);
1333         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1334                                 (unsigned long)root->root_key.objectid,
1335                                 root);
1336         if (ret == 0)
1337                 root->in_radix = 1;
1338
1339         spin_unlock(&fs_info->fs_roots_radix_lock);
1340         radix_tree_preload_end();
1341         if (ret) {
1342                 if (ret == -EEXIST) {
1343                         free_fs_root(root);
1344                         goto again;
1345                 }
1346                 goto fail;
1347         }
1348
1349         ret = btrfs_find_dead_roots(fs_info->tree_root,
1350                                     root->root_key.objectid);
1351         WARN_ON(ret);
1352         return root;
1353 fail:
1354         free_fs_root(root);
1355         return ERR_PTR(ret);
1356 }
1357
1358 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1359 {
1360         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1361         int ret = 0;
1362         struct btrfs_device *device;
1363         struct backing_dev_info *bdi;
1364
1365         rcu_read_lock();
1366         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1367                 if (!device->bdev)
1368                         continue;
1369                 bdi = blk_get_backing_dev_info(device->bdev);
1370                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1371                         ret = 1;
1372                         break;
1373                 }
1374         }
1375         rcu_read_unlock();
1376         return ret;
1377 }
1378
1379 /*
1380  * If this fails, caller must call bdi_destroy() to get rid of the
1381  * bdi again.
1382  */
1383 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1384 {
1385         int err;
1386
1387         bdi->capabilities = BDI_CAP_MAP_COPY;
1388         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1389         if (err)
1390                 return err;
1391
1392         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1393         bdi->congested_fn       = btrfs_congested_fn;
1394         bdi->congested_data     = info;
1395         return 0;
1396 }
1397
1398 static int bio_ready_for_csum(struct bio *bio)
1399 {
1400         u64 length = 0;
1401         u64 buf_len = 0;
1402         u64 start = 0;
1403         struct page *page;
1404         struct extent_io_tree *io_tree = NULL;
1405         struct bio_vec *bvec;
1406         int i;
1407         int ret;
1408
1409         bio_for_each_segment(bvec, bio, i) {
1410                 page = bvec->bv_page;
1411                 if (page->private == EXTENT_PAGE_PRIVATE) {
1412                         length += bvec->bv_len;
1413                         continue;
1414                 }
1415                 if (!page->private) {
1416                         length += bvec->bv_len;
1417                         continue;
1418                 }
1419                 length = bvec->bv_len;
1420                 buf_len = page->private >> 2;
1421                 start = page_offset(page) + bvec->bv_offset;
1422                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1423         }
1424         /* are we fully contained in this bio? */
1425         if (buf_len <= length)
1426                 return 1;
1427
1428         ret = extent_range_uptodate(io_tree, start + length,
1429                                     start + buf_len - 1);
1430         return ret;
1431 }
1432
1433 /*
1434  * called by the kthread helper functions to finally call the bio end_io
1435  * functions.  This is where read checksum verification actually happens
1436  */
1437 static void end_workqueue_fn(struct btrfs_work *work)
1438 {
1439         struct bio *bio;
1440         struct end_io_wq *end_io_wq;
1441         struct btrfs_fs_info *fs_info;
1442         int error;
1443
1444         end_io_wq = container_of(work, struct end_io_wq, work);
1445         bio = end_io_wq->bio;
1446         fs_info = end_io_wq->info;
1447
1448         /* metadata bio reads are special because the whole tree block must
1449          * be checksummed at once.  This makes sure the entire block is in
1450          * ram and up to date before trying to verify things.  For
1451          * blocksize <= pagesize, it is basically a noop
1452          */
1453         if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1454             !bio_ready_for_csum(bio)) {
1455                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1456                                    &end_io_wq->work);
1457                 return;
1458         }
1459         error = end_io_wq->error;
1460         bio->bi_private = end_io_wq->private;
1461         bio->bi_end_io = end_io_wq->end_io;
1462         kfree(end_io_wq);
1463         bio_endio(bio, error);
1464 }
1465
1466 static int cleaner_kthread(void *arg)
1467 {
1468         struct btrfs_root *root = arg;
1469
1470         do {
1471                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1472
1473                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1474                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1475                         btrfs_run_delayed_iputs(root);
1476                         btrfs_clean_old_snapshots(root);
1477                         mutex_unlock(&root->fs_info->cleaner_mutex);
1478                         btrfs_run_defrag_inodes(root->fs_info);
1479                 }
1480
1481                 if (freezing(current)) {
1482                         refrigerator();
1483                 } else {
1484                         set_current_state(TASK_INTERRUPTIBLE);
1485                         if (!kthread_should_stop())
1486                                 schedule();
1487                         __set_current_state(TASK_RUNNING);
1488                 }
1489         } while (!kthread_should_stop());
1490         return 0;
1491 }
1492
1493 static int transaction_kthread(void *arg)
1494 {
1495         struct btrfs_root *root = arg;
1496         struct btrfs_trans_handle *trans;
1497         struct btrfs_transaction *cur;
1498         u64 transid;
1499         unsigned long now;
1500         unsigned long delay;
1501         int ret;
1502
1503         do {
1504                 delay = HZ * 30;
1505                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1506                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1507
1508                 spin_lock(&root->fs_info->trans_lock);
1509                 cur = root->fs_info->running_transaction;
1510                 if (!cur) {
1511                         spin_unlock(&root->fs_info->trans_lock);
1512                         goto sleep;
1513                 }
1514
1515                 now = get_seconds();
1516                 if (!cur->blocked &&
1517                     (now < cur->start_time || now - cur->start_time < 30)) {
1518                         spin_unlock(&root->fs_info->trans_lock);
1519                         delay = HZ * 5;
1520                         goto sleep;
1521                 }
1522                 transid = cur->transid;
1523                 spin_unlock(&root->fs_info->trans_lock);
1524
1525                 trans = btrfs_join_transaction(root);
1526                 BUG_ON(IS_ERR(trans));
1527                 if (transid == trans->transid) {
1528                         ret = btrfs_commit_transaction(trans, root);
1529                         BUG_ON(ret);
1530                 } else {
1531                         btrfs_end_transaction(trans, root);
1532                 }
1533 sleep:
1534                 wake_up_process(root->fs_info->cleaner_kthread);
1535                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1536
1537                 if (freezing(current)) {
1538                         refrigerator();
1539                 } else {
1540                         set_current_state(TASK_INTERRUPTIBLE);
1541                         if (!kthread_should_stop() &&
1542                             !btrfs_transaction_blocked(root->fs_info))
1543                                 schedule_timeout(delay);
1544                         __set_current_state(TASK_RUNNING);
1545                 }
1546         } while (!kthread_should_stop());
1547         return 0;
1548 }
1549
1550 struct btrfs_root *open_ctree(struct super_block *sb,
1551                               struct btrfs_fs_devices *fs_devices,
1552                               char *options)
1553 {
1554         u32 sectorsize;
1555         u32 nodesize;
1556         u32 leafsize;
1557         u32 blocksize;
1558         u32 stripesize;
1559         u64 generation;
1560         u64 features;
1561         struct btrfs_key location;
1562         struct buffer_head *bh;
1563         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1564                                                  GFP_NOFS);
1565         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1566                                                  GFP_NOFS);
1567         struct btrfs_root *tree_root = btrfs_sb(sb);
1568         struct btrfs_fs_info *fs_info = NULL;
1569         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1570                                                 GFP_NOFS);
1571         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1572                                               GFP_NOFS);
1573         struct btrfs_root *log_tree_root;
1574
1575         int ret;
1576         int err = -EINVAL;
1577
1578         struct btrfs_super_block *disk_super;
1579
1580         if (!extent_root || !tree_root || !tree_root->fs_info ||
1581             !chunk_root || !dev_root || !csum_root) {
1582                 err = -ENOMEM;
1583                 goto fail;
1584         }
1585         fs_info = tree_root->fs_info;
1586
1587         ret = init_srcu_struct(&fs_info->subvol_srcu);
1588         if (ret) {
1589                 err = ret;
1590                 goto fail;
1591         }
1592
1593         ret = setup_bdi(fs_info, &fs_info->bdi);
1594         if (ret) {
1595                 err = ret;
1596                 goto fail_srcu;
1597         }
1598
1599         fs_info->btree_inode = new_inode(sb);
1600         if (!fs_info->btree_inode) {
1601                 err = -ENOMEM;
1602                 goto fail_bdi;
1603         }
1604
1605         fs_info->btree_inode->i_mapping->flags &= ~__GFP_FS;
1606
1607         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1608         INIT_LIST_HEAD(&fs_info->trans_list);
1609         INIT_LIST_HEAD(&fs_info->dead_roots);
1610         INIT_LIST_HEAD(&fs_info->delayed_iputs);
1611         INIT_LIST_HEAD(&fs_info->hashers);
1612         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1613         INIT_LIST_HEAD(&fs_info->ordered_operations);
1614         INIT_LIST_HEAD(&fs_info->caching_block_groups);
1615         spin_lock_init(&fs_info->delalloc_lock);
1616         spin_lock_init(&fs_info->trans_lock);
1617         spin_lock_init(&fs_info->ref_cache_lock);
1618         spin_lock_init(&fs_info->fs_roots_radix_lock);
1619         spin_lock_init(&fs_info->delayed_iput_lock);
1620         spin_lock_init(&fs_info->defrag_inodes_lock);
1621
1622         init_completion(&fs_info->kobj_unregister);
1623         fs_info->tree_root = tree_root;
1624         fs_info->extent_root = extent_root;
1625         fs_info->csum_root = csum_root;
1626         fs_info->chunk_root = chunk_root;
1627         fs_info->dev_root = dev_root;
1628         fs_info->fs_devices = fs_devices;
1629         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1630         INIT_LIST_HEAD(&fs_info->space_info);
1631         btrfs_mapping_init(&fs_info->mapping_tree);
1632         btrfs_init_block_rsv(&fs_info->global_block_rsv);
1633         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1634         btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1635         btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1636         btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1637         INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1638         mutex_init(&fs_info->durable_block_rsv_mutex);
1639         atomic_set(&fs_info->nr_async_submits, 0);
1640         atomic_set(&fs_info->async_delalloc_pages, 0);
1641         atomic_set(&fs_info->async_submit_draining, 0);
1642         atomic_set(&fs_info->nr_async_bios, 0);
1643         atomic_set(&fs_info->defrag_running, 0);
1644         fs_info->sb = sb;
1645         fs_info->max_inline = 8192 * 1024;
1646         fs_info->metadata_ratio = 0;
1647         fs_info->defrag_inodes = RB_ROOT;
1648         fs_info->trans_no_join = 0;
1649
1650         fs_info->thread_pool_size = min_t(unsigned long,
1651                                           num_online_cpus() + 2, 8);
1652
1653         INIT_LIST_HEAD(&fs_info->ordered_extents);
1654         spin_lock_init(&fs_info->ordered_extent_lock);
1655         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1656                                         GFP_NOFS);
1657         if (!fs_info->delayed_root) {
1658                 err = -ENOMEM;
1659                 goto fail_iput;
1660         }
1661         btrfs_init_delayed_root(fs_info->delayed_root);
1662
1663         mutex_init(&fs_info->scrub_lock);
1664         atomic_set(&fs_info->scrubs_running, 0);
1665         atomic_set(&fs_info->scrub_pause_req, 0);
1666         atomic_set(&fs_info->scrubs_paused, 0);
1667         atomic_set(&fs_info->scrub_cancel_req, 0);
1668         init_waitqueue_head(&fs_info->scrub_pause_wait);
1669         init_rwsem(&fs_info->scrub_super_lock);
1670         fs_info->scrub_workers_refcnt = 0;
1671
1672         sb->s_blocksize = 4096;
1673         sb->s_blocksize_bits = blksize_bits(4096);
1674         sb->s_bdi = &fs_info->bdi;
1675
1676         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1677         fs_info->btree_inode->i_nlink = 1;
1678         /*
1679          * we set the i_size on the btree inode to the max possible int.
1680          * the real end of the address space is determined by all of
1681          * the devices in the system
1682          */
1683         fs_info->btree_inode->i_size = OFFSET_MAX;
1684         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1685         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1686
1687         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1688         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1689                              fs_info->btree_inode->i_mapping);
1690         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
1691
1692         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1693
1694         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1695         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1696                sizeof(struct btrfs_key));
1697         BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1698         insert_inode_hash(fs_info->btree_inode);
1699
1700         spin_lock_init(&fs_info->block_group_cache_lock);
1701         fs_info->block_group_cache_tree = RB_ROOT;
1702
1703         extent_io_tree_init(&fs_info->freed_extents[0],
1704                              fs_info->btree_inode->i_mapping);
1705         extent_io_tree_init(&fs_info->freed_extents[1],
1706                              fs_info->btree_inode->i_mapping);
1707         fs_info->pinned_extents = &fs_info->freed_extents[0];
1708         fs_info->do_barriers = 1;
1709
1710
1711         mutex_init(&fs_info->ordered_operations_mutex);
1712         mutex_init(&fs_info->tree_log_mutex);
1713         mutex_init(&fs_info->chunk_mutex);
1714         mutex_init(&fs_info->transaction_kthread_mutex);
1715         mutex_init(&fs_info->cleaner_mutex);
1716         mutex_init(&fs_info->volume_mutex);
1717         init_rwsem(&fs_info->extent_commit_sem);
1718         init_rwsem(&fs_info->cleanup_work_sem);
1719         init_rwsem(&fs_info->subvol_sem);
1720
1721         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1722         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1723
1724         init_waitqueue_head(&fs_info->transaction_throttle);
1725         init_waitqueue_head(&fs_info->transaction_wait);
1726         init_waitqueue_head(&fs_info->transaction_blocked_wait);
1727         init_waitqueue_head(&fs_info->async_submit_wait);
1728
1729         __setup_root(4096, 4096, 4096, 4096, tree_root,
1730                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1731
1732         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1733         if (!bh) {
1734                 err = -EINVAL;
1735                 goto fail_alloc;
1736         }
1737
1738         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1739         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1740                sizeof(fs_info->super_for_commit));
1741         brelse(bh);
1742
1743         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1744
1745         disk_super = &fs_info->super_copy;
1746         if (!btrfs_super_root(disk_super))
1747                 goto fail_alloc;
1748
1749         /* check FS state, whether FS is broken. */
1750         fs_info->fs_state |= btrfs_super_flags(disk_super);
1751
1752         btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
1753
1754         /*
1755          * In the long term, we'll store the compression type in the super
1756          * block, and it'll be used for per file compression control.
1757          */
1758         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
1759
1760         ret = btrfs_parse_options(tree_root, options);
1761         if (ret) {
1762                 err = ret;
1763                 goto fail_alloc;
1764         }
1765
1766         features = btrfs_super_incompat_flags(disk_super) &
1767                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1768         if (features) {
1769                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1770                        "unsupported optional features (%Lx).\n",
1771                        (unsigned long long)features);
1772                 err = -EINVAL;
1773                 goto fail_alloc;
1774         }
1775
1776         features = btrfs_super_incompat_flags(disk_super);
1777         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1778         if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1779                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1780         btrfs_set_super_incompat_flags(disk_super, features);
1781
1782         features = btrfs_super_compat_ro_flags(disk_super) &
1783                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1784         if (!(sb->s_flags & MS_RDONLY) && features) {
1785                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1786                        "unsupported option features (%Lx).\n",
1787                        (unsigned long long)features);
1788                 err = -EINVAL;
1789                 goto fail_alloc;
1790         }
1791
1792         btrfs_init_workers(&fs_info->generic_worker,
1793                            "genwork", 1, NULL);
1794
1795         btrfs_init_workers(&fs_info->workers, "worker",
1796                            fs_info->thread_pool_size,
1797                            &fs_info->generic_worker);
1798
1799         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1800                            fs_info->thread_pool_size,
1801                            &fs_info->generic_worker);
1802
1803         btrfs_init_workers(&fs_info->submit_workers, "submit",
1804                            min_t(u64, fs_devices->num_devices,
1805                            fs_info->thread_pool_size),
1806                            &fs_info->generic_worker);
1807
1808         /* a higher idle thresh on the submit workers makes it much more
1809          * likely that bios will be send down in a sane order to the
1810          * devices
1811          */
1812         fs_info->submit_workers.idle_thresh = 64;
1813
1814         fs_info->workers.idle_thresh = 16;
1815         fs_info->workers.ordered = 1;
1816
1817         fs_info->delalloc_workers.idle_thresh = 2;
1818         fs_info->delalloc_workers.ordered = 1;
1819
1820         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1821                            &fs_info->generic_worker);
1822         btrfs_init_workers(&fs_info->endio_workers, "endio",
1823                            fs_info->thread_pool_size,
1824                            &fs_info->generic_worker);
1825         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1826                            fs_info->thread_pool_size,
1827                            &fs_info->generic_worker);
1828         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1829                            "endio-meta-write", fs_info->thread_pool_size,
1830                            &fs_info->generic_worker);
1831         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1832                            fs_info->thread_pool_size,
1833                            &fs_info->generic_worker);
1834         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1835                            1, &fs_info->generic_worker);
1836         btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
1837                            fs_info->thread_pool_size,
1838                            &fs_info->generic_worker);
1839
1840         /*
1841          * endios are largely parallel and should have a very
1842          * low idle thresh
1843          */
1844         fs_info->endio_workers.idle_thresh = 4;
1845         fs_info->endio_meta_workers.idle_thresh = 4;
1846
1847         fs_info->endio_write_workers.idle_thresh = 2;
1848         fs_info->endio_meta_write_workers.idle_thresh = 2;
1849
1850         btrfs_start_workers(&fs_info->workers, 1);
1851         btrfs_start_workers(&fs_info->generic_worker, 1);
1852         btrfs_start_workers(&fs_info->submit_workers, 1);
1853         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1854         btrfs_start_workers(&fs_info->fixup_workers, 1);
1855         btrfs_start_workers(&fs_info->endio_workers, 1);
1856         btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1857         btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1858         btrfs_start_workers(&fs_info->endio_write_workers, 1);
1859         btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1860         btrfs_start_workers(&fs_info->delayed_workers, 1);
1861
1862         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1863         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1864                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1865
1866         nodesize = btrfs_super_nodesize(disk_super);
1867         leafsize = btrfs_super_leafsize(disk_super);
1868         sectorsize = btrfs_super_sectorsize(disk_super);
1869         stripesize = btrfs_super_stripesize(disk_super);
1870         tree_root->nodesize = nodesize;
1871         tree_root->leafsize = leafsize;
1872         tree_root->sectorsize = sectorsize;
1873         tree_root->stripesize = stripesize;
1874
1875         sb->s_blocksize = sectorsize;
1876         sb->s_blocksize_bits = blksize_bits(sectorsize);
1877
1878         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1879                     sizeof(disk_super->magic))) {
1880                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1881                 goto fail_sb_buffer;
1882         }
1883
1884         mutex_lock(&fs_info->chunk_mutex);
1885         ret = btrfs_read_sys_array(tree_root);
1886         mutex_unlock(&fs_info->chunk_mutex);
1887         if (ret) {
1888                 printk(KERN_WARNING "btrfs: failed to read the system "
1889                        "array on %s\n", sb->s_id);
1890                 goto fail_sb_buffer;
1891         }
1892
1893         blocksize = btrfs_level_size(tree_root,
1894                                      btrfs_super_chunk_root_level(disk_super));
1895         generation = btrfs_super_chunk_root_generation(disk_super);
1896
1897         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1898                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1899
1900         chunk_root->node = read_tree_block(chunk_root,
1901                                            btrfs_super_chunk_root(disk_super),
1902                                            blocksize, generation);
1903         BUG_ON(!chunk_root->node);
1904         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1905                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1906                        sb->s_id);
1907                 goto fail_chunk_root;
1908         }
1909         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1910         chunk_root->commit_root = btrfs_root_node(chunk_root);
1911
1912         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1913            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1914            BTRFS_UUID_SIZE);
1915
1916         mutex_lock(&fs_info->chunk_mutex);
1917         ret = btrfs_read_chunk_tree(chunk_root);
1918         mutex_unlock(&fs_info->chunk_mutex);
1919         if (ret) {
1920                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1921                        sb->s_id);
1922                 goto fail_chunk_root;
1923         }
1924
1925         btrfs_close_extra_devices(fs_devices);
1926
1927         blocksize = btrfs_level_size(tree_root,
1928                                      btrfs_super_root_level(disk_super));
1929         generation = btrfs_super_generation(disk_super);
1930
1931         tree_root->node = read_tree_block(tree_root,
1932                                           btrfs_super_root(disk_super),
1933                                           blocksize, generation);
1934         if (!tree_root->node)
1935                 goto fail_chunk_root;
1936         if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1937                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1938                        sb->s_id);
1939                 goto fail_tree_root;
1940         }
1941         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1942         tree_root->commit_root = btrfs_root_node(tree_root);
1943
1944         ret = find_and_setup_root(tree_root, fs_info,
1945                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1946         if (ret)
1947                 goto fail_tree_root;
1948         extent_root->track_dirty = 1;
1949
1950         ret = find_and_setup_root(tree_root, fs_info,
1951                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1952         if (ret)
1953                 goto fail_extent_root;
1954         dev_root->track_dirty = 1;
1955
1956         ret = find_and_setup_root(tree_root, fs_info,
1957                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
1958         if (ret)
1959                 goto fail_dev_root;
1960
1961         csum_root->track_dirty = 1;
1962
1963         fs_info->generation = generation;
1964         fs_info->last_trans_committed = generation;
1965         fs_info->data_alloc_profile = (u64)-1;
1966         fs_info->metadata_alloc_profile = (u64)-1;
1967         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1968
1969         ret = btrfs_init_space_info(fs_info);
1970         if (ret) {
1971                 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
1972                 goto fail_block_groups;
1973         }
1974
1975         ret = btrfs_read_block_groups(extent_root);
1976         if (ret) {
1977                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
1978                 goto fail_block_groups;
1979         }
1980
1981         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1982                                                "btrfs-cleaner");
1983         if (IS_ERR(fs_info->cleaner_kthread))
1984                 goto fail_block_groups;
1985
1986         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1987                                                    tree_root,
1988                                                    "btrfs-transaction");
1989         if (IS_ERR(fs_info->transaction_kthread))
1990                 goto fail_cleaner;
1991
1992         if (!btrfs_test_opt(tree_root, SSD) &&
1993             !btrfs_test_opt(tree_root, NOSSD) &&
1994             !fs_info->fs_devices->rotating) {
1995                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
1996                        "mode\n");
1997                 btrfs_set_opt(fs_info->mount_opt, SSD);
1998         }
1999
2000         /* do not make disk changes in broken FS */
2001         if (btrfs_super_log_root(disk_super) != 0 &&
2002             !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2003                 u64 bytenr = btrfs_super_log_root(disk_super);
2004
2005                 if (fs_devices->rw_devices == 0) {
2006                         printk(KERN_WARNING "Btrfs log replay required "
2007                                "on RO media\n");
2008                         err = -EIO;
2009                         goto fail_trans_kthread;
2010                 }
2011                 blocksize =
2012                      btrfs_level_size(tree_root,
2013                                       btrfs_super_log_root_level(disk_super));
2014
2015                 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2016                 if (!log_tree_root) {
2017                         err = -ENOMEM;
2018                         goto fail_trans_kthread;
2019                 }
2020
2021                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2022                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2023
2024                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2025                                                       blocksize,
2026                                                       generation + 1);
2027                 ret = btrfs_recover_log_trees(log_tree_root);
2028                 BUG_ON(ret);
2029
2030                 if (sb->s_flags & MS_RDONLY) {
2031                         ret =  btrfs_commit_super(tree_root);
2032                         BUG_ON(ret);
2033                 }
2034         }
2035
2036         ret = btrfs_find_orphan_roots(tree_root);
2037         BUG_ON(ret);
2038
2039         if (!(sb->s_flags & MS_RDONLY)) {
2040                 ret = btrfs_cleanup_fs_roots(fs_info);
2041                 BUG_ON(ret);
2042
2043                 ret = btrfs_recover_relocation(tree_root);
2044                 if (ret < 0) {
2045                         printk(KERN_WARNING
2046                                "btrfs: failed to recover relocation\n");
2047                         err = -EINVAL;
2048                         goto fail_trans_kthread;
2049                 }
2050         }
2051
2052         location.objectid = BTRFS_FS_TREE_OBJECTID;
2053         location.type = BTRFS_ROOT_ITEM_KEY;
2054         location.offset = (u64)-1;
2055
2056         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2057         if (!fs_info->fs_root)
2058                 goto fail_trans_kthread;
2059         if (IS_ERR(fs_info->fs_root)) {
2060                 err = PTR_ERR(fs_info->fs_root);
2061                 goto fail_trans_kthread;
2062         }
2063
2064         if (!(sb->s_flags & MS_RDONLY)) {
2065                 down_read(&fs_info->cleanup_work_sem);
2066                 err = btrfs_orphan_cleanup(fs_info->fs_root);
2067                 if (!err)
2068                         err = btrfs_orphan_cleanup(fs_info->tree_root);
2069                 up_read(&fs_info->cleanup_work_sem);
2070                 if (err) {
2071                         close_ctree(tree_root);
2072                         return ERR_PTR(err);
2073                 }
2074         }
2075
2076         return tree_root;
2077
2078 fail_trans_kthread:
2079         kthread_stop(fs_info->transaction_kthread);
2080 fail_cleaner:
2081         kthread_stop(fs_info->cleaner_kthread);
2082
2083         /*
2084          * make sure we're done with the btree inode before we stop our
2085          * kthreads
2086          */
2087         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2088         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2089
2090 fail_block_groups:
2091         btrfs_free_block_groups(fs_info);
2092         free_extent_buffer(csum_root->node);
2093         free_extent_buffer(csum_root->commit_root);
2094 fail_dev_root:
2095         free_extent_buffer(dev_root->node);
2096         free_extent_buffer(dev_root->commit_root);
2097 fail_extent_root:
2098         free_extent_buffer(extent_root->node);
2099         free_extent_buffer(extent_root->commit_root);
2100 fail_tree_root:
2101         free_extent_buffer(tree_root->node);
2102         free_extent_buffer(tree_root->commit_root);
2103 fail_chunk_root:
2104         free_extent_buffer(chunk_root->node);
2105         free_extent_buffer(chunk_root->commit_root);
2106 fail_sb_buffer:
2107         btrfs_stop_workers(&fs_info->generic_worker);
2108         btrfs_stop_workers(&fs_info->fixup_workers);
2109         btrfs_stop_workers(&fs_info->delalloc_workers);
2110         btrfs_stop_workers(&fs_info->workers);
2111         btrfs_stop_workers(&fs_info->endio_workers);
2112         btrfs_stop_workers(&fs_info->endio_meta_workers);
2113         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2114         btrfs_stop_workers(&fs_info->endio_write_workers);
2115         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2116         btrfs_stop_workers(&fs_info->submit_workers);
2117         btrfs_stop_workers(&fs_info->delayed_workers);
2118 fail_alloc:
2119         kfree(fs_info->delayed_root);
2120 fail_iput:
2121         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2122         iput(fs_info->btree_inode);
2123
2124         btrfs_close_devices(fs_info->fs_devices);
2125         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2126 fail_bdi:
2127         bdi_destroy(&fs_info->bdi);
2128 fail_srcu:
2129         cleanup_srcu_struct(&fs_info->subvol_srcu);
2130 fail:
2131         kfree(extent_root);
2132         kfree(tree_root);
2133         kfree(fs_info);
2134         kfree(chunk_root);
2135         kfree(dev_root);
2136         kfree(csum_root);
2137         return ERR_PTR(err);
2138 }
2139
2140 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2141 {
2142         char b[BDEVNAME_SIZE];
2143
2144         if (uptodate) {
2145                 set_buffer_uptodate(bh);
2146         } else {
2147                 printk_ratelimited(KERN_WARNING "lost page write due to "
2148                                         "I/O error on %s\n",
2149                                        bdevname(bh->b_bdev, b));
2150                 /* note, we dont' set_buffer_write_io_error because we have
2151                  * our own ways of dealing with the IO errors
2152                  */
2153                 clear_buffer_uptodate(bh);
2154         }
2155         unlock_buffer(bh);
2156         put_bh(bh);
2157 }
2158
2159 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2160 {
2161         struct buffer_head *bh;
2162         struct buffer_head *latest = NULL;
2163         struct btrfs_super_block *super;
2164         int i;
2165         u64 transid = 0;
2166         u64 bytenr;
2167
2168         /* we would like to check all the supers, but that would make
2169          * a btrfs mount succeed after a mkfs from a different FS.
2170          * So, we need to add a special mount option to scan for
2171          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2172          */
2173         for (i = 0; i < 1; i++) {
2174                 bytenr = btrfs_sb_offset(i);
2175                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2176                         break;
2177                 bh = __bread(bdev, bytenr / 4096, 4096);
2178                 if (!bh)
2179                         continue;
2180
2181                 super = (struct btrfs_super_block *)bh->b_data;
2182                 if (btrfs_super_bytenr(super) != bytenr ||
2183                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2184                             sizeof(super->magic))) {
2185                         brelse(bh);
2186                         continue;
2187                 }
2188
2189                 if (!latest || btrfs_super_generation(super) > transid) {
2190                         brelse(latest);
2191                         latest = bh;
2192                         transid = btrfs_super_generation(super);
2193                 } else {
2194                         brelse(bh);
2195                 }
2196         }
2197         return latest;
2198 }
2199
2200 /*
2201  * this should be called twice, once with wait == 0 and
2202  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2203  * we write are pinned.
2204  *
2205  * They are released when wait == 1 is done.
2206  * max_mirrors must be the same for both runs, and it indicates how
2207  * many supers on this one device should be written.
2208  *
2209  * max_mirrors == 0 means to write them all.
2210  */
2211 static int write_dev_supers(struct btrfs_device *device,
2212                             struct btrfs_super_block *sb,
2213                             int do_barriers, int wait, int max_mirrors)
2214 {
2215         struct buffer_head *bh;
2216         int i;
2217         int ret;
2218         int errors = 0;
2219         u32 crc;
2220         u64 bytenr;
2221         int last_barrier = 0;
2222
2223         if (max_mirrors == 0)
2224                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2225
2226         /* make sure only the last submit_bh does a barrier */
2227         if (do_barriers) {
2228                 for (i = 0; i < max_mirrors; i++) {
2229                         bytenr = btrfs_sb_offset(i);
2230                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2231                             device->total_bytes)
2232                                 break;
2233                         last_barrier = i;
2234                 }
2235         }
2236
2237         for (i = 0; i < max_mirrors; i++) {
2238                 bytenr = btrfs_sb_offset(i);
2239                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2240                         break;
2241
2242                 if (wait) {
2243                         bh = __find_get_block(device->bdev, bytenr / 4096,
2244                                               BTRFS_SUPER_INFO_SIZE);
2245                         BUG_ON(!bh);
2246                         wait_on_buffer(bh);
2247                         if (!buffer_uptodate(bh))
2248                                 errors++;
2249
2250                         /* drop our reference */
2251                         brelse(bh);
2252
2253                         /* drop the reference from the wait == 0 run */
2254                         brelse(bh);
2255                         continue;
2256                 } else {
2257                         btrfs_set_super_bytenr(sb, bytenr);
2258
2259                         crc = ~(u32)0;
2260                         crc = btrfs_csum_data(NULL, (char *)sb +
2261                                               BTRFS_CSUM_SIZE, crc,
2262                                               BTRFS_SUPER_INFO_SIZE -
2263                                               BTRFS_CSUM_SIZE);
2264                         btrfs_csum_final(crc, sb->csum);
2265
2266                         /*
2267                          * one reference for us, and we leave it for the
2268                          * caller
2269                          */
2270                         bh = __getblk(device->bdev, bytenr / 4096,
2271                                       BTRFS_SUPER_INFO_SIZE);
2272                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2273
2274                         /* one reference for submit_bh */
2275                         get_bh(bh);
2276
2277                         set_buffer_uptodate(bh);
2278                         lock_buffer(bh);
2279                         bh->b_end_io = btrfs_end_buffer_write_sync;
2280                 }
2281
2282                 if (i == last_barrier && do_barriers)
2283                         ret = submit_bh(WRITE_FLUSH_FUA, bh);
2284                 else
2285                         ret = submit_bh(WRITE_SYNC, bh);
2286
2287                 if (ret)
2288                         errors++;
2289         }
2290         return errors < i ? 0 : -1;
2291 }
2292
2293 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2294 {
2295         struct list_head *head;
2296         struct btrfs_device *dev;
2297         struct btrfs_super_block *sb;
2298         struct btrfs_dev_item *dev_item;
2299         int ret;
2300         int do_barriers;
2301         int max_errors;
2302         int total_errors = 0;
2303         u64 flags;
2304
2305         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2306         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2307
2308         sb = &root->fs_info->super_for_commit;
2309         dev_item = &sb->dev_item;
2310
2311         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2312         head = &root->fs_info->fs_devices->devices;
2313         list_for_each_entry_rcu(dev, head, dev_list) {
2314                 if (!dev->bdev) {
2315                         total_errors++;
2316                         continue;
2317                 }
2318                 if (!dev->in_fs_metadata || !dev->writeable)
2319                         continue;
2320
2321                 btrfs_set_stack_device_generation(dev_item, 0);
2322                 btrfs_set_stack_device_type(dev_item, dev->type);
2323                 btrfs_set_stack_device_id(dev_item, dev->devid);
2324                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2325                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2326                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2327                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2328                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2329                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2330                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2331
2332                 flags = btrfs_super_flags(sb);
2333                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2334
2335                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2336                 if (ret)
2337                         total_errors++;
2338         }
2339         if (total_errors > max_errors) {
2340                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2341                        total_errors);
2342                 BUG();
2343         }
2344
2345         total_errors = 0;
2346         list_for_each_entry_rcu(dev, head, dev_list) {
2347                 if (!dev->bdev)
2348                         continue;
2349                 if (!dev->in_fs_metadata || !dev->writeable)
2350                         continue;
2351
2352                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2353                 if (ret)
2354                         total_errors++;
2355         }
2356         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2357         if (total_errors > max_errors) {
2358                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2359                        total_errors);
2360                 BUG();
2361         }
2362         return 0;
2363 }
2364
2365 int write_ctree_super(struct btrfs_trans_handle *trans,
2366                       struct btrfs_root *root, int max_mirrors)
2367 {
2368         int ret;
2369
2370         ret = write_all_supers(root, max_mirrors);
2371         return ret;
2372 }
2373
2374 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2375 {
2376         spin_lock(&fs_info->fs_roots_radix_lock);
2377         radix_tree_delete(&fs_info->fs_roots_radix,
2378                           (unsigned long)root->root_key.objectid);
2379         spin_unlock(&fs_info->fs_roots_radix_lock);
2380
2381         if (btrfs_root_refs(&root->root_item) == 0)
2382                 synchronize_srcu(&fs_info->subvol_srcu);
2383
2384         __btrfs_remove_free_space_cache(root->free_ino_pinned);
2385         __btrfs_remove_free_space_cache(root->free_ino_ctl);
2386         free_fs_root(root);
2387         return 0;
2388 }
2389
2390 static void free_fs_root(struct btrfs_root *root)
2391 {
2392         iput(root->cache_inode);
2393         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2394         if (root->anon_super.s_dev) {
2395                 down_write(&root->anon_super.s_umount);
2396                 kill_anon_super(&root->anon_super);
2397         }
2398         free_extent_buffer(root->node);
2399         free_extent_buffer(root->commit_root);
2400         kfree(root->free_ino_ctl);
2401         kfree(root->free_ino_pinned);
2402         kfree(root->name);
2403         kfree(root);
2404 }
2405
2406 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2407 {
2408         int ret;
2409         struct btrfs_root *gang[8];
2410         int i;
2411
2412         while (!list_empty(&fs_info->dead_roots)) {
2413                 gang[0] = list_entry(fs_info->dead_roots.next,
2414                                      struct btrfs_root, root_list);
2415                 list_del(&gang[0]->root_list);
2416
2417                 if (gang[0]->in_radix) {
2418                         btrfs_free_fs_root(fs_info, gang[0]);
2419                 } else {
2420                         free_extent_buffer(gang[0]->node);
2421                         free_extent_buffer(gang[0]->commit_root);
2422                         kfree(gang[0]);
2423                 }
2424         }
2425
2426         while (1) {
2427                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2428                                              (void **)gang, 0,
2429                                              ARRAY_SIZE(gang));
2430                 if (!ret)
2431                         break;
2432                 for (i = 0; i < ret; i++)
2433                         btrfs_free_fs_root(fs_info, gang[i]);
2434         }
2435         return 0;
2436 }
2437
2438 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2439 {
2440         u64 root_objectid = 0;
2441         struct btrfs_root *gang[8];
2442         int i;
2443         int ret;
2444
2445         while (1) {
2446                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2447                                              (void **)gang, root_objectid,
2448                                              ARRAY_SIZE(gang));
2449                 if (!ret)
2450                         break;
2451
2452                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2453                 for (i = 0; i < ret; i++) {
2454                         int err;
2455
2456                         root_objectid = gang[i]->root_key.objectid;
2457                         err = btrfs_orphan_cleanup(gang[i]);
2458                         if (err)
2459                                 return err;
2460                 }
2461                 root_objectid++;
2462         }
2463         return 0;
2464 }
2465
2466 int btrfs_commit_super(struct btrfs_root *root)
2467 {
2468         struct btrfs_trans_handle *trans;
2469         int ret;
2470
2471         mutex_lock(&root->fs_info->cleaner_mutex);
2472         btrfs_run_delayed_iputs(root);
2473         btrfs_clean_old_snapshots(root);
2474         mutex_unlock(&root->fs_info->cleaner_mutex);
2475
2476         /* wait until ongoing cleanup work done */
2477         down_write(&root->fs_info->cleanup_work_sem);
2478         up_write(&root->fs_info->cleanup_work_sem);
2479
2480         trans = btrfs_join_transaction(root);
2481         if (IS_ERR(trans))
2482                 return PTR_ERR(trans);
2483         ret = btrfs_commit_transaction(trans, root);
2484         BUG_ON(ret);
2485         /* run commit again to drop the original snapshot */
2486         trans = btrfs_join_transaction(root);
2487         if (IS_ERR(trans))
2488                 return PTR_ERR(trans);
2489         btrfs_commit_transaction(trans, root);
2490         ret = btrfs_write_and_wait_transaction(NULL, root);
2491         BUG_ON(ret);
2492
2493         ret = write_ctree_super(NULL, root, 0);
2494         return ret;
2495 }
2496
2497 int close_ctree(struct btrfs_root *root)
2498 {
2499         struct btrfs_fs_info *fs_info = root->fs_info;
2500         int ret;
2501
2502         fs_info->closing = 1;
2503         smp_mb();
2504
2505         btrfs_scrub_cancel(root);
2506
2507         /* wait for any defraggers to finish */
2508         wait_event(fs_info->transaction_wait,
2509                    (atomic_read(&fs_info->defrag_running) == 0));
2510
2511         /* clear out the rbtree of defraggable inodes */
2512         btrfs_run_defrag_inodes(root->fs_info);
2513
2514         btrfs_put_block_group_cache(fs_info);
2515
2516         /*
2517          * Here come 2 situations when btrfs is broken to flip readonly:
2518          *
2519          * 1. when btrfs flips readonly somewhere else before
2520          * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2521          * and btrfs will skip to write sb directly to keep
2522          * ERROR state on disk.
2523          *
2524          * 2. when btrfs flips readonly just in btrfs_commit_super,
2525          * and in such case, btrfs cannot write sb via btrfs_commit_super,
2526          * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2527          * btrfs will cleanup all FS resources first and write sb then.
2528          */
2529         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2530                 ret = btrfs_commit_super(root);
2531                 if (ret)
2532                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2533         }
2534
2535         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2536                 ret = btrfs_error_commit_super(root);
2537                 if (ret)
2538                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2539         }
2540
2541         kthread_stop(root->fs_info->transaction_kthread);
2542         kthread_stop(root->fs_info->cleaner_kthread);
2543
2544         fs_info->closing = 2;
2545         smp_mb();
2546
2547         if (fs_info->delalloc_bytes) {
2548                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2549                        (unsigned long long)fs_info->delalloc_bytes);
2550         }
2551         if (fs_info->total_ref_cache_size) {
2552                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2553                        (unsigned long long)fs_info->total_ref_cache_size);
2554         }
2555
2556         free_extent_buffer(fs_info->extent_root->node);
2557         free_extent_buffer(fs_info->extent_root->commit_root);
2558         free_extent_buffer(fs_info->tree_root->node);
2559         free_extent_buffer(fs_info->tree_root->commit_root);
2560         free_extent_buffer(root->fs_info->chunk_root->node);
2561         free_extent_buffer(root->fs_info->chunk_root->commit_root);
2562         free_extent_buffer(root->fs_info->dev_root->node);
2563         free_extent_buffer(root->fs_info->dev_root->commit_root);
2564         free_extent_buffer(root->fs_info->csum_root->node);
2565         free_extent_buffer(root->fs_info->csum_root->commit_root);
2566
2567         btrfs_free_block_groups(root->fs_info);
2568
2569         del_fs_roots(fs_info);
2570
2571         iput(fs_info->btree_inode);
2572         kfree(fs_info->delayed_root);
2573
2574         btrfs_stop_workers(&fs_info->generic_worker);
2575         btrfs_stop_workers(&fs_info->fixup_workers);
2576         btrfs_stop_workers(&fs_info->delalloc_workers);
2577         btrfs_stop_workers(&fs_info->workers);
2578         btrfs_stop_workers(&fs_info->endio_workers);
2579         btrfs_stop_workers(&fs_info->endio_meta_workers);
2580         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2581         btrfs_stop_workers(&fs_info->endio_write_workers);
2582         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2583         btrfs_stop_workers(&fs_info->submit_workers);
2584         btrfs_stop_workers(&fs_info->delayed_workers);
2585
2586         btrfs_close_devices(fs_info->fs_devices);
2587         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2588
2589         bdi_destroy(&fs_info->bdi);
2590         cleanup_srcu_struct(&fs_info->subvol_srcu);
2591
2592         kfree(fs_info->extent_root);
2593         kfree(fs_info->tree_root);
2594         kfree(fs_info->chunk_root);
2595         kfree(fs_info->dev_root);
2596         kfree(fs_info->csum_root);
2597         kfree(fs_info);
2598
2599         return 0;
2600 }
2601
2602 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2603 {
2604         int ret;
2605         struct inode *btree_inode = buf->first_page->mapping->host;
2606
2607         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2608                                      NULL);
2609         if (!ret)
2610                 return ret;
2611
2612         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2613                                     parent_transid);
2614         return !ret;
2615 }
2616
2617 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2618 {
2619         struct inode *btree_inode = buf->first_page->mapping->host;
2620         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2621                                           buf);
2622 }
2623
2624 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2625 {
2626         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2627         u64 transid = btrfs_header_generation(buf);
2628         struct inode *btree_inode = root->fs_info->btree_inode;
2629         int was_dirty;
2630
2631         btrfs_assert_tree_locked(buf);
2632         if (transid != root->fs_info->generation) {
2633                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2634                        "found %llu running %llu\n",
2635                         (unsigned long long)buf->start,
2636                         (unsigned long long)transid,
2637                         (unsigned long long)root->fs_info->generation);
2638                 WARN_ON(1);
2639         }
2640         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2641                                             buf);
2642         if (!was_dirty) {
2643                 spin_lock(&root->fs_info->delalloc_lock);
2644                 root->fs_info->dirty_metadata_bytes += buf->len;
2645                 spin_unlock(&root->fs_info->delalloc_lock);
2646         }
2647 }
2648
2649 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2650 {
2651         /*
2652          * looks as though older kernels can get into trouble with
2653          * this code, they end up stuck in balance_dirty_pages forever
2654          */
2655         u64 num_dirty;
2656         unsigned long thresh = 32 * 1024 * 1024;
2657
2658         if (current->flags & PF_MEMALLOC)
2659                 return;
2660
2661         btrfs_balance_delayed_items(root);
2662
2663         num_dirty = root->fs_info->dirty_metadata_bytes;
2664
2665         if (num_dirty > thresh) {
2666                 balance_dirty_pages_ratelimited_nr(
2667                                    root->fs_info->btree_inode->i_mapping, 1);
2668         }
2669         return;
2670 }
2671
2672 void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2673 {
2674         /*
2675          * looks as though older kernels can get into trouble with
2676          * this code, they end up stuck in balance_dirty_pages forever
2677          */
2678         u64 num_dirty;
2679         unsigned long thresh = 32 * 1024 * 1024;
2680
2681         if (current->flags & PF_MEMALLOC)
2682                 return;
2683
2684         num_dirty = root->fs_info->dirty_metadata_bytes;
2685
2686         if (num_dirty > thresh) {
2687                 balance_dirty_pages_ratelimited_nr(
2688                                    root->fs_info->btree_inode->i_mapping, 1);
2689         }
2690         return;
2691 }
2692
2693 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2694 {
2695         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2696         int ret;
2697         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2698         if (ret == 0)
2699                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2700         return ret;
2701 }
2702
2703 int btree_lock_page_hook(struct page *page)
2704 {
2705         struct inode *inode = page->mapping->host;
2706         struct btrfs_root *root = BTRFS_I(inode)->root;
2707         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2708         struct extent_buffer *eb;
2709         unsigned long len;
2710         u64 bytenr = page_offset(page);
2711
2712         if (page->private == EXTENT_PAGE_PRIVATE)
2713                 goto out;
2714
2715         len = page->private >> 2;
2716         eb = find_extent_buffer(io_tree, bytenr, len);
2717         if (!eb)
2718                 goto out;
2719
2720         btrfs_tree_lock(eb);
2721         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2722
2723         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2724                 spin_lock(&root->fs_info->delalloc_lock);
2725                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2726                         root->fs_info->dirty_metadata_bytes -= eb->len;
2727                 else
2728                         WARN_ON(1);
2729                 spin_unlock(&root->fs_info->delalloc_lock);
2730         }
2731
2732         btrfs_tree_unlock(eb);
2733         free_extent_buffer(eb);
2734 out:
2735         lock_page(page);
2736         return 0;
2737 }
2738
2739 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
2740                               int read_only)
2741 {
2742         if (read_only)
2743                 return;
2744
2745         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
2746                 printk(KERN_WARNING "warning: mount fs with errors, "
2747                        "running btrfsck is recommended\n");
2748 }
2749
2750 int btrfs_error_commit_super(struct btrfs_root *root)
2751 {
2752         int ret;
2753
2754         mutex_lock(&root->fs_info->cleaner_mutex);
2755         btrfs_run_delayed_iputs(root);
2756         mutex_unlock(&root->fs_info->cleaner_mutex);
2757
2758         down_write(&root->fs_info->cleanup_work_sem);
2759         up_write(&root->fs_info->cleanup_work_sem);
2760
2761         /* cleanup FS via transaction */
2762         btrfs_cleanup_transaction(root);
2763
2764         ret = write_ctree_super(NULL, root, 0);
2765
2766         return ret;
2767 }
2768
2769 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
2770 {
2771         struct btrfs_inode *btrfs_inode;
2772         struct list_head splice;
2773
2774         INIT_LIST_HEAD(&splice);
2775
2776         mutex_lock(&root->fs_info->ordered_operations_mutex);
2777         spin_lock(&root->fs_info->ordered_extent_lock);
2778
2779         list_splice_init(&root->fs_info->ordered_operations, &splice);
2780         while (!list_empty(&splice)) {
2781                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2782                                          ordered_operations);
2783
2784                 list_del_init(&btrfs_inode->ordered_operations);
2785
2786                 btrfs_invalidate_inodes(btrfs_inode->root);
2787         }
2788
2789         spin_unlock(&root->fs_info->ordered_extent_lock);
2790         mutex_unlock(&root->fs_info->ordered_operations_mutex);
2791
2792         return 0;
2793 }
2794
2795 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
2796 {
2797         struct list_head splice;
2798         struct btrfs_ordered_extent *ordered;
2799         struct inode *inode;
2800
2801         INIT_LIST_HEAD(&splice);
2802
2803         spin_lock(&root->fs_info->ordered_extent_lock);
2804
2805         list_splice_init(&root->fs_info->ordered_extents, &splice);
2806         while (!list_empty(&splice)) {
2807                 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
2808                                      root_extent_list);
2809
2810                 list_del_init(&ordered->root_extent_list);
2811                 atomic_inc(&ordered->refs);
2812
2813                 /* the inode may be getting freed (in sys_unlink path). */
2814                 inode = igrab(ordered->inode);
2815
2816                 spin_unlock(&root->fs_info->ordered_extent_lock);
2817                 if (inode)
2818                         iput(inode);
2819
2820                 atomic_set(&ordered->refs, 1);
2821                 btrfs_put_ordered_extent(ordered);
2822
2823                 spin_lock(&root->fs_info->ordered_extent_lock);
2824         }
2825
2826         spin_unlock(&root->fs_info->ordered_extent_lock);
2827
2828         return 0;
2829 }
2830
2831 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
2832                                       struct btrfs_root *root)
2833 {
2834         struct rb_node *node;
2835         struct btrfs_delayed_ref_root *delayed_refs;
2836         struct btrfs_delayed_ref_node *ref;
2837         int ret = 0;
2838
2839         delayed_refs = &trans->delayed_refs;
2840
2841         spin_lock(&delayed_refs->lock);
2842         if (delayed_refs->num_entries == 0) {
2843                 spin_unlock(&delayed_refs->lock);
2844                 printk(KERN_INFO "delayed_refs has NO entry\n");
2845                 return ret;
2846         }
2847
2848         node = rb_first(&delayed_refs->root);
2849         while (node) {
2850                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2851                 node = rb_next(node);
2852
2853                 ref->in_tree = 0;
2854                 rb_erase(&ref->rb_node, &delayed_refs->root);
2855                 delayed_refs->num_entries--;
2856
2857                 atomic_set(&ref->refs, 1);
2858                 if (btrfs_delayed_ref_is_head(ref)) {
2859                         struct btrfs_delayed_ref_head *head;
2860
2861                         head = btrfs_delayed_node_to_head(ref);
2862                         mutex_lock(&head->mutex);
2863                         kfree(head->extent_op);
2864                         delayed_refs->num_heads--;
2865                         if (list_empty(&head->cluster))
2866                                 delayed_refs->num_heads_ready--;
2867                         list_del_init(&head->cluster);
2868                         mutex_unlock(&head->mutex);
2869                 }
2870
2871                 spin_unlock(&delayed_refs->lock);
2872                 btrfs_put_delayed_ref(ref);
2873
2874                 cond_resched();
2875                 spin_lock(&delayed_refs->lock);
2876         }
2877
2878         spin_unlock(&delayed_refs->lock);
2879
2880         return ret;
2881 }
2882
2883 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
2884 {
2885         struct btrfs_pending_snapshot *snapshot;
2886         struct list_head splice;
2887
2888         INIT_LIST_HEAD(&splice);
2889
2890         list_splice_init(&t->pending_snapshots, &splice);
2891
2892         while (!list_empty(&splice)) {
2893                 snapshot = list_entry(splice.next,
2894                                       struct btrfs_pending_snapshot,
2895                                       list);
2896
2897                 list_del_init(&snapshot->list);
2898
2899                 kfree(snapshot);
2900         }
2901
2902         return 0;
2903 }
2904
2905 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
2906 {
2907         struct btrfs_inode *btrfs_inode;
2908         struct list_head splice;
2909
2910         INIT_LIST_HEAD(&splice);
2911
2912         spin_lock(&root->fs_info->delalloc_lock);
2913         list_splice_init(&root->fs_info->delalloc_inodes, &splice);
2914
2915         while (!list_empty(&splice)) {
2916                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2917                                     delalloc_inodes);
2918
2919                 list_del_init(&btrfs_inode->delalloc_inodes);
2920
2921                 btrfs_invalidate_inodes(btrfs_inode->root);
2922         }
2923
2924         spin_unlock(&root->fs_info->delalloc_lock);
2925
2926         return 0;
2927 }
2928
2929 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
2930                                         struct extent_io_tree *dirty_pages,
2931                                         int mark)
2932 {
2933         int ret;
2934         struct page *page;
2935         struct inode *btree_inode = root->fs_info->btree_inode;
2936         struct extent_buffer *eb;
2937         u64 start = 0;
2938         u64 end;
2939         u64 offset;
2940         unsigned long index;
2941
2942         while (1) {
2943                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
2944                                             mark);
2945                 if (ret)
2946                         break;
2947
2948                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
2949                 while (start <= end) {
2950                         index = start >> PAGE_CACHE_SHIFT;
2951                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
2952                         page = find_get_page(btree_inode->i_mapping, index);
2953                         if (!page)
2954                                 continue;
2955                         offset = page_offset(page);
2956
2957                         spin_lock(&dirty_pages->buffer_lock);
2958                         eb = radix_tree_lookup(
2959                              &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
2960                                                offset >> PAGE_CACHE_SHIFT);
2961                         spin_unlock(&dirty_pages->buffer_lock);
2962                         if (eb) {
2963                                 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
2964                                                          &eb->bflags);
2965                                 atomic_set(&eb->refs, 1);
2966                         }
2967                         if (PageWriteback(page))
2968                                 end_page_writeback(page);
2969
2970                         lock_page(page);
2971                         if (PageDirty(page)) {
2972                                 clear_page_dirty_for_io(page);
2973                                 spin_lock_irq(&page->mapping->tree_lock);
2974                                 radix_tree_tag_clear(&page->mapping->page_tree,
2975                                                         page_index(page),
2976                                                         PAGECACHE_TAG_DIRTY);
2977                                 spin_unlock_irq(&page->mapping->tree_lock);
2978                         }
2979
2980                         page->mapping->a_ops->invalidatepage(page, 0);
2981                         unlock_page(page);
2982                 }
2983         }
2984
2985         return ret;
2986 }
2987
2988 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
2989                                        struct extent_io_tree *pinned_extents)
2990 {
2991         struct extent_io_tree *unpin;
2992         u64 start;
2993         u64 end;
2994         int ret;
2995
2996         unpin = pinned_extents;
2997         while (1) {
2998                 ret = find_first_extent_bit(unpin, 0, &start, &end,
2999                                             EXTENT_DIRTY);
3000                 if (ret)
3001                         break;
3002
3003                 /* opt_discard */
3004                 if (btrfs_test_opt(root, DISCARD))
3005                         ret = btrfs_error_discard_extent(root, start,
3006                                                          end + 1 - start,
3007                                                          NULL);
3008
3009                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3010                 btrfs_error_unpin_extent_range(root, start, end);
3011                 cond_resched();
3012         }
3013
3014         return 0;
3015 }
3016
3017 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3018 {
3019         struct btrfs_transaction *t;
3020         LIST_HEAD(list);
3021
3022         WARN_ON(1);
3023
3024         mutex_lock(&root->fs_info->transaction_kthread_mutex);
3025
3026         spin_lock(&root->fs_info->trans_lock);
3027         list_splice_init(&root->fs_info->trans_list, &list);
3028         root->fs_info->trans_no_join = 1;
3029         spin_unlock(&root->fs_info->trans_lock);
3030
3031         while (!list_empty(&list)) {
3032                 t = list_entry(list.next, struct btrfs_transaction, list);
3033                 if (!t)
3034                         break;
3035
3036                 btrfs_destroy_ordered_operations(root);
3037
3038                 btrfs_destroy_ordered_extents(root);
3039
3040                 btrfs_destroy_delayed_refs(t, root);
3041
3042                 btrfs_block_rsv_release(root,
3043                                         &root->fs_info->trans_block_rsv,
3044                                         t->dirty_pages.dirty_bytes);
3045
3046                 /* FIXME: cleanup wait for commit */
3047                 t->in_commit = 1;
3048                 t->blocked = 1;
3049                 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3050                         wake_up(&root->fs_info->transaction_blocked_wait);
3051
3052                 t->blocked = 0;
3053                 if (waitqueue_active(&root->fs_info->transaction_wait))
3054                         wake_up(&root->fs_info->transaction_wait);
3055
3056                 t->commit_done = 1;
3057                 if (waitqueue_active(&t->commit_wait))
3058                         wake_up(&t->commit_wait);
3059
3060                 btrfs_destroy_pending_snapshots(t);
3061
3062                 btrfs_destroy_delalloc_inodes(root);
3063
3064                 spin_lock(&root->fs_info->trans_lock);
3065                 root->fs_info->running_transaction = NULL;
3066                 spin_unlock(&root->fs_info->trans_lock);
3067
3068                 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3069                                              EXTENT_DIRTY);
3070
3071                 btrfs_destroy_pinned_extent(root,
3072                                             root->fs_info->pinned_extents);
3073
3074                 atomic_set(&t->use_count, 0);
3075                 list_del_init(&t->list);
3076                 memset(t, 0, sizeof(*t));
3077                 kmem_cache_free(btrfs_transaction_cachep, t);
3078         }
3079
3080         spin_lock(&root->fs_info->trans_lock);
3081         root->fs_info->trans_no_join = 0;
3082         spin_unlock(&root->fs_info->trans_lock);
3083         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3084
3085         return 0;
3086 }
3087
3088 static struct extent_io_ops btree_extent_io_ops = {
3089         .write_cache_pages_lock_hook = btree_lock_page_hook,
3090         .readpage_end_io_hook = btree_readpage_end_io_hook,
3091         .submit_bio_hook = btree_submit_bio_hook,
3092         /* note we're sharing with inode.c for the merge bio hook */
3093         .merge_bio_hook = btrfs_merge_bio_hook,
3094 };