fs/btrfs/disk-io.c

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