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