fs/btrfs/disk-io.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/fs.h>
  20 #include <linux/blkdev.h>
  21 #include <linux/scatterlist.h>
  22 #include <linux/swap.h>
  23 #include <linux/radix-tree.h>
  24 #include <linux/writeback.h>
  25 #include <linux/buffer_head.h>
  26 #include <linux/workqueue.h>
  27 #include <linux/kthread.h>
  28 #include <linux/freezer.h>
  29 #include <linux/crc32c.h>
  30 #include <linux/slab.h>
  31 #include <linux/migrate.h>
  32 #include <asm/unaligned.h>
  33 #include "compat.h"
  34 #include "ctree.h"
  35 #include "disk-io.h"
  36 #include "transaction.h"
  37 #include "btrfs_inode.h"
  38 #include "volumes.h"
  39 #include "print-tree.h"
  40 #include "async-thread.h"
  41 #include "locking.h"
  42 #include "tree-log.h"
  43 #include "free-space-cache.h"
  44 #include "inode-map.h"
  45
  46 static struct extent_io_ops btree_extent_io_ops;
  47 static void end_workqueue_fn(struct btrfs_work *work);
  48 static void free_fs_root(struct btrfs_root *root);
  49 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  50                                     int read_only);
  51 static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
  52 static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
  53 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  54                                       struct btrfs_root *root);
  55 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
  56 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  57 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
  58                                         struct extent_io_tree *dirty_pages,
  59                                         int mark);
  60 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
  61                                        struct extent_io_tree *pinned_extents);
  62 static int btrfs_cleanup_transaction(struct btrfs_root *root);
  63
  64 /*
  65  * end_io_wq structs are used to do processing in task context when an IO is
  66  * complete.  This is used during reads to verify checksums, and it is used
  67  * by writes to insert metadata for new file extents after IO is complete.
  68  */
  69 struct end_io_wq {
  70         struct bio *bio;
  71         bio_end_io_t *end_io;
  72         void *private;
  73         struct btrfs_fs_info *info;
  74         int error;
  75         int metadata;
  76         struct list_head list;
  77         struct btrfs_work work;
  78 };
  79
  80 /*
  81  * async submit bios are used to offload expensive checksumming
  82  * onto the worker threads.  They checksum file and metadata bios
  83  * just before they are sent down the IO stack.
  84  */
  85 struct async_submit_bio {
  86         struct inode *inode;
  87         struct bio *bio;
  88         struct list_head list;
  89         extent_submit_bio_hook_t *submit_bio_start;
  90         extent_submit_bio_hook_t *submit_bio_done;
  91         int rw;
  92         int mirror_num;
  93         unsigned long bio_flags;
  94         /*
  95          * bio_offset is optional, can be used if the pages in the bio
  96          * can't tell us where in the file the bio should go
  97          */
  98         u64 bio_offset;
  99         struct btrfs_work work;
 100 };
 101
 102 /* These are used to set the lockdep class on the extent buffer locks.
 103  * The class is set by the readpage_end_io_hook after the buffer has
 104  * passed csum validation but before the pages are unlocked.
 105  *
 106  * The lockdep class is also set by btrfs_init_new_buffer on freshly
 107  * allocated blocks.
 108  *
 109  * The class is based on the level in the tree block, which allows lockdep
 110  * to know that lower nodes nest inside the locks of higher nodes.
 111  *
 112  * We also add a check to make sure the highest level of the tree is
 113  * the same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this
 114  * code needs update as well.
 115  */
 116 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 117 # if BTRFS_MAX_LEVEL != 8
 118 #  error
 119 # endif
 120 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
 121 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
 122         /* leaf */
 123         "btrfs-extent-00",
 124         "btrfs-extent-01",
 125         "btrfs-extent-02",
 126         "btrfs-extent-03",
 127         "btrfs-extent-04",
 128         "btrfs-extent-05",
 129         "btrfs-extent-06",
 130         "btrfs-extent-07",
 131         /* highest possible level */
 132         "btrfs-extent-08",
 133 };
 134 #endif
 135
 136 /*
 137  * extents on the btree inode are pretty simple, there's one extent
 138  * that covers the entire device
 139  */
 140 static struct extent_map *btree_get_extent(struct inode *inode,
 141                 struct page *page, size_t page_offset, u64 start, u64 len,
 142                 int create)
 143 {
 144         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 145         struct extent_map *em;
 146         int ret;
 147
 148         read_lock(&em_tree->lock);
 149         em = lookup_extent_mapping(em_tree, start, len);
 150         if (em) {
 151                 em->bdev =
 152                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 153                 read_unlock(&em_tree->lock);
 154                 goto out;
 155         }
 156         read_unlock(&em_tree->lock);
 157
 158         em = alloc_extent_map(GFP_NOFS);
 159         if (!em) {
 160                 em = ERR_PTR(-ENOMEM);
 161                 goto out;
 162         }
 163         em->start = 0;
 164         em->len = (u64)-1;
 165         em->block_len = (u64)-1;
 166         em->block_start = 0;
 167         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 168
 169         write_lock(&em_tree->lock);
 170         ret = add_extent_mapping(em_tree, em);
 171         if (ret == -EEXIST) {
 172                 u64 failed_start = em->start;
 173                 u64 failed_len = em->len;
 174
 175                 free_extent_map(em);
 176                 em = lookup_extent_mapping(em_tree, start, len);
 177                 if (em) {
 178                         ret = 0;
 179                 } else {
 180                         em = lookup_extent_mapping(em_tree, failed_start,
 181                                                    failed_len);
 182                         ret = -EIO;
 183                 }
 184         } else if (ret) {
 185                 free_extent_map(em);
 186                 em = NULL;
 187         }
 188         write_unlock(&em_tree->lock);
 189
 190         if (ret)
 191                 em = ERR_PTR(ret);
 192 out:
 193         return em;
 194 }
 195
 196 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 197 {
 198         return crc32c(seed, data, len);
 199 }
 200
 201 void btrfs_csum_final(u32 crc, char *result)
 202 {
 203         put_unaligned_le32(~crc, result);
 204 }
 205
 206 /*
 207  * compute the csum for a btree block, and either verify it or write it
 208  * into the csum field of the block.
 209  */
 210 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 211                            int verify)
 212 {
 213         u16 csum_size =
 214                 btrfs_super_csum_size(&root->fs_info->super_copy);
 215         char *result = NULL;
 216         unsigned long len;
 217         unsigned long cur_len;
 218         unsigned long offset = BTRFS_CSUM_SIZE;
 219         char *map_token = NULL;
 220         char *kaddr;
 221         unsigned long map_start;
 222         unsigned long map_len;
 223         int err;
 224         u32 crc = ~(u32)0;
 225         unsigned long inline_result;
 226
 227         len = buf->len - offset;
 228         while (len > 0) {
 229                 err = map_private_extent_buffer(buf, offset, 32,
 230                                         &map_token, &kaddr,
 231                                         &map_start, &map_len, KM_USER0);
 232                 if (err)
 233                         return 1;
 234                 cur_len = min(len, map_len - (offset - map_start));
 235                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 236                                       crc, cur_len);
 237                 len -= cur_len;
 238                 offset += cur_len;
 239                 unmap_extent_buffer(buf, map_token, KM_USER0);
 240         }
 241         if (csum_size > sizeof(inline_result)) {
 242                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
 243                 if (!result)
 244                         return 1;
 245         } else {
 246                 result = (char *)&inline_result;
 247         }
 248
 249         btrfs_csum_final(crc, result);
 250
 251         if (verify) {
 252                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 253                         u32 val;
 254                         u32 found = 0;
 255                         memcpy(&found, result, csum_size);
 256
 257                         read_extent_buffer(buf, &val, 0, csum_size);
 258                         if (printk_ratelimit()) {
 259                                 printk(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                         }
 266                         if (result != (char *)&inline_result)
 267                                 kfree(result);
 268                         return 1;
 269                 }
 270         } else {
 271                 write_extent_buffer(buf, result, 0, csum_size);
 272         }
 273         if (result != (char *)&inline_result)
 274                 kfree(result);
 275         return 0;
 276 }
 277
 278 /*
 279  * we can't consider a given block up to date unless the transid of the
 280  * block matches the transid in the parent node's pointer.  This is how we
 281  * detect blocks that either didn't get written at all or got written
 282  * in the wrong place.
 283  */
 284 static int verify_parent_transid(struct extent_io_tree *io_tree,
 285                                  struct extent_buffer *eb, u64 parent_transid)
 286 {
 287         struct extent_state *cached_state = NULL;
 288         int ret;
 289
 290         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 291                 return 0;
 292
 293         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 294                          0, &cached_state, GFP_NOFS);
 295         if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
 296             btrfs_header_generation(eb) == parent_transid) {
 297                 ret = 0;
 298                 goto out;
 299         }
 300         if (printk_ratelimit()) {
 301                 printk("parent transid verify failed on %llu wanted %llu "
 302                        "found %llu\n",
 303                        (unsigned long long)eb->start,
 304                        (unsigned long long)parent_transid,
 305                        (unsigned long long)btrfs_header_generation(eb));
 306         }
 307         ret = 1;
 308         clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
 309 out:
 310         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 311                              &cached_state, GFP_NOFS);
 312         return ret;
 313 }
 314
 315 /*
 316  * helper to read a given tree block, doing retries as required when
 317  * the checksums don't match and we have alternate mirrors to try.
 318  */
 319 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 320                                           struct extent_buffer *eb,
 321                                           u64 start, u64 parent_transid)
 322 {
 323         struct extent_io_tree *io_tree;
 324         int ret;
 325         int num_copies = 0;
 326         int mirror_num = 0;
 327
 328         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 329         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 330         while (1) {
 331                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 332                                                btree_get_extent, mirror_num);
 333                 if (!ret &&
 334                     !verify_parent_transid(io_tree, eb, parent_transid))
 335                         return ret;
 336
 337                 /*
 338                  * This buffer's crc is fine, but its contents are corrupted, so
 339                  * there is no reason to read the other copies, they won't be
 340                  * any less wrong.
 341                  */
 342                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 343                         return ret;
 344
 345                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 346                                               eb->start, eb->len);
 347                 if (num_copies == 1)
 348                         return ret;
 349
 350                 mirror_num++;
 351                 if (mirror_num > num_copies)
 352                         return ret;
 353         }
 354         return -EIO;
 355 }
 356
 357 /*
 358  * checksum a dirty tree block before IO.  This has extra checks to make sure
 359  * we only fill in the checksum field in the first page of a multi-page block
 360  */
 361
 362 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 363 {
 364         struct extent_io_tree *tree;
 365         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 366         u64 found_start;
 367         unsigned long len;
 368         struct extent_buffer *eb;
 369         int ret;
 370
 371         tree = &BTRFS_I(page->mapping->host)->io_tree;
 372
 373         if (page->private == EXTENT_PAGE_PRIVATE) {
 374                 WARN_ON(1);
 375                 goto out;
 376         }
 377         if (!page->private) {
 378                 WARN_ON(1);
 379                 goto out;
 380         }
 381         len = page->private >> 2;
 382         WARN_ON(len == 0);
 383
 384         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 385         if (eb == NULL) {
 386                 WARN_ON(1);
 387                 goto out;
 388         }
 389         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 390                                              btrfs_header_generation(eb));
 391         BUG_ON(ret);
 392         WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
 393
 394         found_start = btrfs_header_bytenr(eb);
 395         if (found_start != start) {
 396                 WARN_ON(1);
 397                 goto err;
 398         }
 399         if (eb->first_page != page) {
 400                 WARN_ON(1);
 401                 goto err;
 402         }
 403         if (!PageUptodate(page)) {
 404                 WARN_ON(1);
 405                 goto err;
 406         }
 407         csum_tree_block(root, eb, 0);
 408 err:
 409         free_extent_buffer(eb);
 410 out:
 411         return 0;
 412 }
 413
 414 static int check_tree_block_fsid(struct btrfs_root *root,
 415                                  struct extent_buffer *eb)
 416 {
 417         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 418         u8 fsid[BTRFS_UUID_SIZE];
 419         int ret = 1;
 420
 421         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
 422                            BTRFS_FSID_SIZE);
 423         while (fs_devices) {
 424                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 425                         ret = 0;
 426                         break;
 427                 }
 428                 fs_devices = fs_devices->seed;
 429         }
 430         return ret;
 431 }
 432
 433 #define CORRUPT(reason, eb, root, slot)                         \
 434         printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
 435                "root=%llu, slot=%d\n", reason,                  \
 436                (unsigned long long)btrfs_header_bytenr(eb),     \
 437                (unsigned long long)root->objectid, slot)
 438
 439 static noinline int check_leaf(struct btrfs_root *root,
 440                                struct extent_buffer *leaf)
 441 {
 442         struct btrfs_key key;
 443         struct btrfs_key leaf_key;
 444         u32 nritems = btrfs_header_nritems(leaf);
 445         int slot;
 446
 447         if (nritems == 0)
 448                 return 0;
 449
 450         /* Check the 0 item */
 451         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 452             BTRFS_LEAF_DATA_SIZE(root)) {
 453                 CORRUPT("invalid item offset size pair", leaf, root, 0);
 454                 return -EIO;
 455         }
 456
 457         /*
 458          * Check to make sure each items keys are in the correct order and their
 459          * offsets make sense.  We only have to loop through nritems-1 because
 460          * we check the current slot against the next slot, which verifies the
 461          * next slot's offset+size makes sense and that the current's slot
 462          * offset is correct.
 463          */
 464         for (slot = 0; slot < nritems - 1; slot++) {
 465                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 466                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 467
 468                 /* Make sure the keys are in the right order */
 469                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 470                         CORRUPT("bad key order", leaf, root, slot);
 471                         return -EIO;
 472                 }
 473
 474                 /*
 475                  * Make sure the offset and ends are right, remember that the
 476                  * item data starts at the end of the leaf and grows towards the
 477                  * front.
 478                  */
 479                 if (btrfs_item_offset_nr(leaf, slot) !=
 480                         btrfs_item_end_nr(leaf, slot + 1)) {
 481                         CORRUPT("slot offset bad", leaf, root, slot);
 482                         return -EIO;
 483                 }
 484
 485                 /*
 486                  * Check to make sure that we don't point outside of the leaf,
 487                  * just incase all the items are consistent to eachother, but
 488                  * all point outside of the leaf.
 489                  */
 490                 if (btrfs_item_end_nr(leaf, slot) >
 491                     BTRFS_LEAF_DATA_SIZE(root)) {
 492                         CORRUPT("slot end outside of leaf", leaf, root, slot);
 493                         return -EIO;
 494                 }
 495         }
 496
 497         return 0;
 498 }
 499
 500 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 501 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
 502 {
 503         lockdep_set_class_and_name(&eb->lock,
 504                            &btrfs_eb_class[level],
 505                            btrfs_eb_name[level]);
 506 }
 507 #endif
 508
 509 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 510                                struct extent_state *state)
 511 {
 512         struct extent_io_tree *tree;
 513         u64 found_start;
 514         int found_level;
 515         unsigned long len;
 516         struct extent_buffer *eb;
 517         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 518         int ret = 0;
 519
 520         tree = &BTRFS_I(page->mapping->host)->io_tree;
 521         if (page->private == EXTENT_PAGE_PRIVATE)
 522                 goto out;
 523         if (!page->private)
 524                 goto out;
 525
 526         len = page->private >> 2;
 527         WARN_ON(len == 0);
 528
 529         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 530         if (eb == NULL) {
 531                 ret = -EIO;
 532                 goto out;
 533         }
 534
 535         found_start = btrfs_header_bytenr(eb);
 536         if (found_start != start) {
 537                 if (printk_ratelimit()) {
 538                         printk(KERN_INFO "btrfs bad tree block start "
 539                                "%llu %llu\n",
 540                                (unsigned long long)found_start,
 541                                (unsigned long long)eb->start);
 542                 }
 543                 ret = -EIO;
 544                 goto err;
 545         }
 546         if (eb->first_page != page) {
 547                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
 548                        eb->first_page->index, page->index);
 549                 WARN_ON(1);
 550                 ret = -EIO;
 551                 goto err;
 552         }
 553         if (check_tree_block_fsid(root, eb)) {
 554                 if (printk_ratelimit()) {
 555                         printk(KERN_INFO "btrfs bad fsid on block %llu\n",
 556                                (unsigned long long)eb->start);
 557                 }
 558                 ret = -EIO;
 559                 goto err;
 560         }
 561         found_level = btrfs_header_level(eb);
 562
 563         btrfs_set_buffer_lockdep_class(eb, found_level);
 564
 565         ret = csum_tree_block(root, eb, 1);
 566         if (ret) {
 567                 ret = -EIO;
 568                 goto err;
 569         }
 570
 571         /*
 572          * If this is a leaf block and it is corrupt, set the corrupt bit so
 573          * that we don't try and read the other copies of this block, just
 574          * return -EIO.
 575          */
 576         if (found_level == 0 && check_leaf(root, eb)) {
 577                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 578                 ret = -EIO;
 579         }
 580
 581         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 582         end = eb->start + end - 1;
 583 err:
 584         free_extent_buffer(eb);
 585 out:
 586         return ret;
 587 }
 588
 589 static void end_workqueue_bio(struct bio *bio, int err)
 590 {
 591         struct end_io_wq *end_io_wq = bio->bi_private;
 592         struct btrfs_fs_info *fs_info;
 593
 594         fs_info = end_io_wq->info;
 595         end_io_wq->error = err;
 596         end_io_wq->work.func = end_workqueue_fn;
 597         end_io_wq->work.flags = 0;
 598
 599         if (bio->bi_rw & REQ_WRITE) {
 600                 if (end_io_wq->metadata == 1)
 601                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
 602                                            &end_io_wq->work);
 603                 else if (end_io_wq->metadata == 2)
 604                         btrfs_queue_worker(&fs_info->endio_freespace_worker,
 605                                            &end_io_wq->work);
 606                 else
 607                         btrfs_queue_worker(&fs_info->endio_write_workers,
 608                                            &end_io_wq->work);
 609         } else {
 610                 if (end_io_wq->metadata)
 611                         btrfs_queue_worker(&fs_info->endio_meta_workers,
 612                                            &end_io_wq->work);
 613                 else
 614                         btrfs_queue_worker(&fs_info->endio_workers,
 615                                            &end_io_wq->work);
 616         }
 617 }
 618
 619 /*
 620  * For the metadata arg you want
 621  *
 622  * 0 - if data
 623  * 1 - if normal metadta
 624  * 2 - if writing to the free space cache area
 625  */
 626 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 627                         int metadata)
 628 {
 629         struct end_io_wq *end_io_wq;
 630         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 631         if (!end_io_wq)
 632                 return -ENOMEM;
 633
 634         end_io_wq->private = bio->bi_private;
 635         end_io_wq->end_io = bio->bi_end_io;
 636         end_io_wq->info = info;
 637         end_io_wq->error = 0;
 638         end_io_wq->bio = bio;
 639         end_io_wq->metadata = metadata;
 640
 641         bio->bi_private = end_io_wq;
 642         bio->bi_end_io = end_workqueue_bio;
 643         return 0;
 644 }
 645
 646 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 647 {
 648         unsigned long limit = min_t(unsigned long,
 649                                     info->workers.max_workers,
 650                                     info->fs_devices->open_devices);
 651         return 256 * limit;
 652 }
 653
 654 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
 655 {
 656         return atomic_read(&info->nr_async_bios) >
 657                 btrfs_async_submit_limit(info);
 658 }
 659
 660 static void run_one_async_start(struct btrfs_work *work)
 661 {
 662         struct async_submit_bio *async;
 663
 664         async = container_of(work, struct  async_submit_bio, work);
 665         async->submit_bio_start(async->inode, async->rw, async->bio,
 666                                async->mirror_num, async->bio_flags,
 667                                async->bio_offset);
 668 }
 669
 670 static void run_one_async_done(struct btrfs_work *work)
 671 {
 672         struct btrfs_fs_info *fs_info;
 673         struct async_submit_bio *async;
 674         int limit;
 675
 676         async = container_of(work, struct  async_submit_bio, work);
 677         fs_info = BTRFS_I(async->inode)->root->fs_info;
 678
 679         limit = btrfs_async_submit_limit(fs_info);
 680         limit = limit * 2 / 3;
 681
 682         atomic_dec(&fs_info->nr_async_submits);
 683
 684         if (atomic_read(&fs_info->nr_async_submits) < limit &&
 685             waitqueue_active(&fs_info->async_submit_wait))
 686                 wake_up(&fs_info->async_submit_wait);
 687
 688         async->submit_bio_done(async->inode, async->rw, async->bio,
 689                                async->mirror_num, async->bio_flags,
 690                                async->bio_offset);
 691 }
 692
 693 static void run_one_async_free(struct btrfs_work *work)
 694 {
 695         struct async_submit_bio *async;
 696
 697         async = container_of(work, struct  async_submit_bio, work);
 698         kfree(async);
 699 }
 700
 701 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 702                         int rw, struct bio *bio, int mirror_num,
 703                         unsigned long bio_flags,
 704                         u64 bio_offset,
 705                         extent_submit_bio_hook_t *submit_bio_start,
 706                         extent_submit_bio_hook_t *submit_bio_done)
 707 {
 708         struct async_submit_bio *async;
 709
 710         async = kmalloc(sizeof(*async), GFP_NOFS);
 711         if (!async)
 712                 return -ENOMEM;
 713
 714         async->inode = inode;
 715         async->rw = rw;
 716         async->bio = bio;
 717         async->mirror_num = mirror_num;
 718         async->submit_bio_start = submit_bio_start;
 719         async->submit_bio_done = submit_bio_done;
 720
 721         async->work.func = run_one_async_start;
 722         async->work.ordered_func = run_one_async_done;
 723         async->work.ordered_free = run_one_async_free;
 724
 725         async->work.flags = 0;
 726         async->bio_flags = bio_flags;
 727         async->bio_offset = bio_offset;
 728
 729         atomic_inc(&fs_info->nr_async_submits);
 730
 731         if (rw & REQ_SYNC)
 732                 btrfs_set_work_high_prio(&async->work);
 733
 734         btrfs_queue_worker(&fs_info->workers, &async->work);
 735
 736         while (atomic_read(&fs_info->async_submit_draining) &&
 737               atomic_read(&fs_info->nr_async_submits)) {
 738                 wait_event(fs_info->async_submit_wait,
 739                            (atomic_read(&fs_info->nr_async_submits) == 0));
 740         }
 741
 742         return 0;
 743 }
 744
 745 static int btree_csum_one_bio(struct bio *bio)
 746 {
 747         struct bio_vec *bvec = bio->bi_io_vec;
 748         int bio_index = 0;
 749         struct btrfs_root *root;
 750
 751         WARN_ON(bio->bi_vcnt <= 0);
 752         while (bio_index < bio->bi_vcnt) {
 753                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 754                 csum_dirty_buffer(root, bvec->bv_page);
 755                 bio_index++;
 756                 bvec++;
 757         }
 758         return 0;
 759 }
 760
 761 static int __btree_submit_bio_start(struct inode *inode, int rw,
 762                                     struct bio *bio, int mirror_num,
 763                                     unsigned long bio_flags,
 764                                     u64 bio_offset)
 765 {
 766         /*
 767          * when we're called for a write, we're already in the async
 768          * submission context.  Just jump into btrfs_map_bio
 769          */
 770         btree_csum_one_bio(bio);
 771         return 0;
 772 }
 773
 774 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 775                                  int mirror_num, unsigned long bio_flags,
 776                                  u64 bio_offset)
 777 {
 778         /*
 779          * when we're called for a write, we're already in the async
 780          * submission context.  Just jump into btrfs_map_bio
 781          */
 782         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 783 }
 784
 785 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 786                                  int mirror_num, unsigned long bio_flags,
 787                                  u64 bio_offset)
 788 {
 789         int ret;
 790
 791         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 792                                           bio, 1);
 793         BUG_ON(ret);
 794
 795         if (!(rw & REQ_WRITE)) {
 796                 /*
 797                  * called for a read, do the setup so that checksum validation
 798                  * can happen in the async kernel threads
 799                  */
 800                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 801                                      mirror_num, 0);
 802         }
 803
 804         /*
 805          * kthread helpers are used to submit writes so that checksumming
 806          * can happen in parallel across all CPUs
 807          */
 808         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 809                                    inode, rw, bio, mirror_num, 0,
 810                                    bio_offset,
 811                                    __btree_submit_bio_start,
 812                                    __btree_submit_bio_done);
 813 }
 814
 815 #ifdef CONFIG_MIGRATION
 816 static int btree_migratepage(struct address_space *mapping,
 817                         struct page *newpage, struct page *page)
 818 {
 819         /*
 820          * we can't safely write a btree page from here,
 821          * we haven't done the locking hook
 822          */
 823         if (PageDirty(page))
 824                 return -EAGAIN;
 825         /*
 826          * Buffers may be managed in a filesystem specific way.
 827          * We must have no buffers or drop them.
 828          */
 829         if (page_has_private(page) &&
 830             !try_to_release_page(page, GFP_KERNEL))
 831                 return -EAGAIN;
 832         return migrate_page(mapping, newpage, page);
 833 }
 834 #endif
 835
 836 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 837 {
 838         struct extent_io_tree *tree;
 839         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 840         struct extent_buffer *eb;
 841         int was_dirty;
 842
 843         tree = &BTRFS_I(page->mapping->host)->io_tree;
 844         if (!(current->flags & PF_MEMALLOC)) {
 845                 return extent_write_full_page(tree, page,
 846                                               btree_get_extent, wbc);
 847         }
 848
 849         redirty_page_for_writepage(wbc, page);
 850         eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
 851         WARN_ON(!eb);
 852
 853         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
 854         if (!was_dirty) {
 855                 spin_lock(&root->fs_info->delalloc_lock);
 856                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
 857                 spin_unlock(&root->fs_info->delalloc_lock);
 858         }
 859         free_extent_buffer(eb);
 860
 861         unlock_page(page);
 862         return 0;
 863 }
 864
 865 static int btree_writepages(struct address_space *mapping,
 866                             struct writeback_control *wbc)
 867 {
 868         struct extent_io_tree *tree;
 869         tree = &BTRFS_I(mapping->host)->io_tree;
 870         if (wbc->sync_mode == WB_SYNC_NONE) {
 871                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
 872                 u64 num_dirty;
 873                 unsigned long thresh = 32 * 1024 * 1024;
 874
 875                 if (wbc->for_kupdate)
 876                         return 0;
 877
 878                 /* this is a bit racy, but that's ok */
 879                 num_dirty = root->fs_info->dirty_metadata_bytes;
 880                 if (num_dirty < thresh)
 881                         return 0;
 882         }
 883         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 884 }
 885
 886 static int btree_readpage(struct file *file, struct page *page)
 887 {
 888         struct extent_io_tree *tree;
 889         tree = &BTRFS_I(page->mapping->host)->io_tree;
 890         return extent_read_full_page(tree, page, btree_get_extent);
 891 }
 892
 893 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 894 {
 895         struct extent_io_tree *tree;
 896         struct extent_map_tree *map;
 897         int ret;
 898
 899         if (PageWriteback(page) || PageDirty(page))
 900                 return 0;
 901
 902         tree = &BTRFS_I(page->mapping->host)->io_tree;
 903         map = &BTRFS_I(page->mapping->host)->extent_tree;
 904
 905         ret = try_release_extent_state(map, tree, page, gfp_flags);
 906         if (!ret)
 907                 return 0;
 908
 909         ret = try_release_extent_buffer(tree, page);
 910         if (ret == 1) {
 911                 ClearPagePrivate(page);
 912                 set_page_private(page, 0);
 913                 page_cache_release(page);
 914         }
 915
 916         return ret;
 917 }
 918
 919 static void btree_invalidatepage(struct page *page, unsigned long offset)
 920 {
 921         struct extent_io_tree *tree;
 922         tree = &BTRFS_I(page->mapping->host)->io_tree;
 923         extent_invalidatepage(tree, page, offset);
 924         btree_releasepage(page, GFP_NOFS);
 925         if (PagePrivate(page)) {
 926                 printk(KERN_WARNING "btrfs warning page private not zero "
 927                        "on page %llu\n", (unsigned long long)page_offset(page));
 928                 ClearPagePrivate(page);
 929                 set_page_private(page, 0);
 930                 page_cache_release(page);
 931         }
 932 }
 933
 934 static const struct address_space_operations btree_aops = {
 935         .readpage       = btree_readpage,
 936         .writepage      = btree_writepage,
 937         .writepages     = btree_writepages,
 938         .releasepage    = btree_releasepage,
 939         .invalidatepage = btree_invalidatepage,
 940 #ifdef CONFIG_MIGRATION
 941         .migratepage    = btree_migratepage,
 942 #endif
 943 };
 944
 945 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 946                          u64 parent_transid)
 947 {
 948         struct extent_buffer *buf = NULL;
 949         struct inode *btree_inode = root->fs_info->btree_inode;
 950         int ret = 0;
 951
 952         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 953         if (!buf)
 954                 return 0;
 955         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 956                                  buf, 0, 0, btree_get_extent, 0);
 957         free_extent_buffer(buf);
 958         return ret;
 959 }
 960
 961 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 962                                             u64 bytenr, u32 blocksize)
 963 {
 964         struct inode *btree_inode = root->fs_info->btree_inode;
 965         struct extent_buffer *eb;
 966         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 967                                 bytenr, blocksize, GFP_NOFS);
 968         return eb;
 969 }
 970
 971 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 972                                                  u64 bytenr, u32 blocksize)
 973 {
 974         struct inode *btree_inode = root->fs_info->btree_inode;
 975         struct extent_buffer *eb;
 976
 977         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 978                                  bytenr, blocksize, NULL, GFP_NOFS);
 979         return eb;
 980 }
 981
 982
 983 int btrfs_write_tree_block(struct extent_buffer *buf)
 984 {
 985         return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
 986                                         buf->start + buf->len - 1);
 987 }
 988
 989 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 990 {
 991         return filemap_fdatawait_range(buf->first_page->mapping,
 992                                        buf->start, buf->start + buf->len - 1);
 993 }
 994
 995 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 996                                       u32 blocksize, u64 parent_transid)
 997 {
 998         struct extent_buffer *buf = NULL;
 999         int ret;
1000
1001         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1002         if (!buf)
1003                 return NULL;
1004
1005         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1006
1007         if (ret == 0)
1008                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1009         return buf;
1010
1011 }
1012
1013 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1014                      struct extent_buffer *buf)
1015 {
1016         struct inode *btree_inode = root->fs_info->btree_inode;
1017         if (btrfs_header_generation(buf) ==
1018             root->fs_info->running_transaction->transid) {
1019                 btrfs_assert_tree_locked(buf);
1020
1021                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1022                         spin_lock(&root->fs_info->delalloc_lock);
1023                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
1024                                 root->fs_info->dirty_metadata_bytes -= buf->len;
1025                         else
1026                                 WARN_ON(1);
1027                         spin_unlock(&root->fs_info->delalloc_lock);
1028                 }
1029
1030                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1031                 btrfs_set_lock_blocking(buf);
1032                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1033                                           buf);
1034         }
1035         return 0;
1036 }
1037
1038 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1039                         u32 stripesize, struct btrfs_root *root,
1040                         struct btrfs_fs_info *fs_info,
1041                         u64 objectid)
1042 {
1043         root->node = NULL;
1044         root->commit_root = NULL;
1045         root->sectorsize = sectorsize;
1046         root->nodesize = nodesize;
1047         root->leafsize = leafsize;
1048         root->stripesize = stripesize;
1049         root->ref_cows = 0;
1050         root->track_dirty = 0;
1051         root->in_radix = 0;
1052         root->orphan_item_inserted = 0;
1053         root->orphan_cleanup_state = 0;
1054
1055         root->fs_info = fs_info;
1056         root->objectid = objectid;
1057         root->last_trans = 0;
1058         root->highest_objectid = 0;
1059         root->name = NULL;
1060         root->in_sysfs = 0;
1061         root->inode_tree = RB_ROOT;
1062         root->block_rsv = NULL;
1063         root->orphan_block_rsv = NULL;
1064
1065         INIT_LIST_HEAD(&root->dirty_list);
1066         INIT_LIST_HEAD(&root->orphan_list);
1067         INIT_LIST_HEAD(&root->root_list);
1068         spin_lock_init(&root->node_lock);
1069         spin_lock_init(&root->orphan_lock);
1070         spin_lock_init(&root->inode_lock);
1071         spin_lock_init(&root->accounting_lock);
1072         mutex_init(&root->objectid_mutex);
1073         mutex_init(&root->log_mutex);
1074         init_waitqueue_head(&root->log_writer_wait);
1075         init_waitqueue_head(&root->log_commit_wait[0]);
1076         init_waitqueue_head(&root->log_commit_wait[1]);
1077         atomic_set(&root->log_commit[0], 0);
1078         atomic_set(&root->log_commit[1], 0);
1079         atomic_set(&root->log_writers, 0);
1080         root->log_batch = 0;
1081         root->log_transid = 0;
1082         root->last_log_commit = 0;
1083         extent_io_tree_init(&root->dirty_log_pages,
1084                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1085
1086         memset(&root->root_key, 0, sizeof(root->root_key));
1087         memset(&root->root_item, 0, sizeof(root->root_item));
1088         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1089         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1090         root->defrag_trans_start = fs_info->generation;
1091         init_completion(&root->kobj_unregister);
1092         root->defrag_running = 0;
1093         root->root_key.objectid = objectid;
1094         root->anon_super.s_root = NULL;
1095         root->anon_super.s_dev = 0;
1096         INIT_LIST_HEAD(&root->anon_super.s_list);
1097         INIT_LIST_HEAD(&root->anon_super.s_instances);
1098         init_rwsem(&root->anon_super.s_umount);
1099
1100         return 0;
1101 }
1102
1103 static int find_and_setup_root(struct btrfs_root *tree_root,
1104                                struct btrfs_fs_info *fs_info,
1105                                u64 objectid,
1106                                struct btrfs_root *root)
1107 {
1108         int ret;
1109         u32 blocksize;
1110         u64 generation;
1111
1112         __setup_root(tree_root->nodesize, tree_root->leafsize,
1113                      tree_root->sectorsize, tree_root->stripesize,
1114                      root, fs_info, objectid);
1115         ret = btrfs_find_last_root(tree_root, objectid,
1116                                    &root->root_item, &root->root_key);
1117         if (ret > 0)
1118                 return -ENOENT;
1119         BUG_ON(ret);
1120
1121         generation = btrfs_root_generation(&root->root_item);
1122         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1123         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1124                                      blocksize, generation);
1125         if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1126                 free_extent_buffer(root->node);
1127                 return -EIO;
1128         }
1129         root->commit_root = btrfs_root_node(root);
1130         return 0;
1131 }
1132
1133 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1134                                          struct btrfs_fs_info *fs_info)
1135 {
1136         struct btrfs_root *root;
1137         struct btrfs_root *tree_root = fs_info->tree_root;
1138         struct extent_buffer *leaf;
1139
1140         root = kzalloc(sizeof(*root), GFP_NOFS);
1141         if (!root)
1142                 return ERR_PTR(-ENOMEM);
1143
1144         __setup_root(tree_root->nodesize, tree_root->leafsize,
1145                      tree_root->sectorsize, tree_root->stripesize,
1146                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1147
1148         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1149         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1150         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1151         /*
1152          * log trees do not get reference counted because they go away
1153          * before a real commit is actually done.  They do store pointers
1154          * to file data extents, and those reference counts still get
1155          * updated (along with back refs to the log tree).
1156          */
1157         root->ref_cows = 0;
1158
1159         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1160                                       BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1161         if (IS_ERR(leaf)) {
1162                 kfree(root);
1163                 return ERR_CAST(leaf);
1164         }
1165
1166         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1167         btrfs_set_header_bytenr(leaf, leaf->start);
1168         btrfs_set_header_generation(leaf, trans->transid);
1169         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1170         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1171         root->node = leaf;
1172
1173         write_extent_buffer(root->node, root->fs_info->fsid,
1174                             (unsigned long)btrfs_header_fsid(root->node),
1175                             BTRFS_FSID_SIZE);
1176         btrfs_mark_buffer_dirty(root->node);
1177         btrfs_tree_unlock(root->node);
1178         return root;
1179 }
1180
1181 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1182                              struct btrfs_fs_info *fs_info)
1183 {
1184         struct btrfs_root *log_root;
1185
1186         log_root = alloc_log_tree(trans, fs_info);
1187         if (IS_ERR(log_root))
1188                 return PTR_ERR(log_root);
1189         WARN_ON(fs_info->log_root_tree);
1190         fs_info->log_root_tree = log_root;
1191         return 0;
1192 }
1193
1194 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1195                        struct btrfs_root *root)
1196 {
1197         struct btrfs_root *log_root;
1198         struct btrfs_inode_item *inode_item;
1199
1200         log_root = alloc_log_tree(trans, root->fs_info);
1201         if (IS_ERR(log_root))
1202                 return PTR_ERR(log_root);
1203
1204         log_root->last_trans = trans->transid;
1205         log_root->root_key.offset = root->root_key.objectid;
1206
1207         inode_item = &log_root->root_item.inode;
1208         inode_item->generation = cpu_to_le64(1);
1209         inode_item->size = cpu_to_le64(3);
1210         inode_item->nlink = cpu_to_le32(1);
1211         inode_item->nbytes = cpu_to_le64(root->leafsize);
1212         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1213
1214         btrfs_set_root_node(&log_root->root_item, log_root->node);
1215
1216         WARN_ON(root->log_root);
1217         root->log_root = log_root;
1218         root->log_transid = 0;
1219         root->last_log_commit = 0;
1220         return 0;
1221 }
1222
1223 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1224                                                struct btrfs_key *location)
1225 {
1226         struct btrfs_root *root;
1227         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1228         struct btrfs_path *path;
1229         struct extent_buffer *l;
1230         u64 generation;
1231         u32 blocksize;
1232         int ret = 0;
1233
1234         root = kzalloc(sizeof(*root), GFP_NOFS);
1235         if (!root)
1236                 return ERR_PTR(-ENOMEM);
1237         if (location->offset == (u64)-1) {
1238                 ret = find_and_setup_root(tree_root, fs_info,
1239                                           location->objectid, root);
1240                 if (ret) {
1241                         kfree(root);
1242                         return ERR_PTR(ret);
1243                 }
1244                 goto out;
1245         }
1246
1247         __setup_root(tree_root->nodesize, tree_root->leafsize,
1248                      tree_root->sectorsize, tree_root->stripesize,
1249                      root, fs_info, location->objectid);
1250
1251         path = btrfs_alloc_path();
1252         if (!path) {
1253                 kfree(root);
1254                 return ERR_PTR(-ENOMEM);
1255         }
1256         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1257         if (ret == 0) {
1258                 l = path->nodes[0];
1259                 read_extent_buffer(l, &root->root_item,
1260                                 btrfs_item_ptr_offset(l, path->slots[0]),
1261                                 sizeof(root->root_item));
1262                 memcpy(&root->root_key, location, sizeof(*location));
1263         }
1264         btrfs_free_path(path);
1265         if (ret) {
1266                 kfree(root);
1267                 if (ret > 0)
1268                         ret = -ENOENT;
1269                 return ERR_PTR(ret);
1270         }
1271
1272         generation = btrfs_root_generation(&root->root_item);
1273         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1274         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1275                                      blocksize, generation);
1276         root->commit_root = btrfs_root_node(root);
1277         BUG_ON(!root->node);
1278 out:
1279         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1280                 root->ref_cows = 1;
1281                 btrfs_check_and_init_root_item(&root->root_item);
1282         }
1283
1284         return root;
1285 }
1286
1287 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1288                                         u64 root_objectid)
1289 {
1290         struct btrfs_root *root;
1291
1292         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1293                 return fs_info->tree_root;
1294         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1295                 return fs_info->extent_root;
1296
1297         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1298                                  (unsigned long)root_objectid);
1299         return root;
1300 }
1301
1302 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1303                                               struct btrfs_key *location)
1304 {
1305         struct btrfs_root *root;
1306         int ret;
1307
1308         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1309                 return fs_info->tree_root;
1310         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1311                 return fs_info->extent_root;
1312         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1313                 return fs_info->chunk_root;
1314         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1315                 return fs_info->dev_root;
1316         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1317                 return fs_info->csum_root;
1318 again:
1319         spin_lock(&fs_info->fs_roots_radix_lock);
1320         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1321                                  (unsigned long)location->objectid);
1322         spin_unlock(&fs_info->fs_roots_radix_lock);
1323         if (root)
1324                 return root;
1325
1326         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1327         if (IS_ERR(root))
1328                 return root;
1329
1330         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1331         if (!root->free_ino_ctl)
1332                 goto fail;
1333         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1334                                         GFP_NOFS);
1335         if (!root->free_ino_pinned)
1336                 goto fail;
1337
1338         btrfs_init_free_ino_ctl(root);
1339         mutex_init(&root->fs_commit_mutex);
1340         spin_lock_init(&root->cache_lock);
1341         init_waitqueue_head(&root->cache_wait);
1342
1343         set_anon_super(&root->anon_super, NULL);
1344
1345         if (btrfs_root_refs(&root->root_item) == 0) {
1346                 ret = -ENOENT;
1347                 goto fail;
1348         }
1349
1350         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1351         if (ret < 0)
1352                 goto fail;
1353         if (ret == 0)
1354                 root->orphan_item_inserted = 1;
1355
1356         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1357         if (ret)
1358                 goto fail;
1359
1360         spin_lock(&fs_info->fs_roots_radix_lock);
1361         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1362                                 (unsigned long)root->root_key.objectid,
1363                                 root);
1364         if (ret == 0)
1365                 root->in_radix = 1;
1366
1367         spin_unlock(&fs_info->fs_roots_radix_lock);
1368         radix_tree_preload_end();
1369         if (ret) {
1370                 if (ret == -EEXIST) {
1371                         free_fs_root(root);
1372                         goto again;
1373                 }
1374                 goto fail;
1375         }
1376
1377         ret = btrfs_find_dead_roots(fs_info->tree_root,
1378                                     root->root_key.objectid);
1379         WARN_ON(ret);
1380         return root;
1381 fail:
1382         free_fs_root(root);
1383         return ERR_PTR(ret);
1384 }
1385
1386 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1387                                       struct btrfs_key *location,
1388                                       const char *name, int namelen)
1389 {
1390         return btrfs_read_fs_root_no_name(fs_info, location);
1391 #if 0
1392         struct btrfs_root *root;
1393         int ret;
1394
1395         root = btrfs_read_fs_root_no_name(fs_info, location);
1396         if (!root)
1397                 return NULL;
1398
1399         if (root->in_sysfs)
1400                 return root;
1401
1402         ret = btrfs_set_root_name(root, name, namelen);
1403         if (ret) {
1404                 free_extent_buffer(root->node);
1405                 kfree(root);
1406                 return ERR_PTR(ret);
1407         }
1408
1409         ret = btrfs_sysfs_add_root(root);
1410         if (ret) {
1411                 free_extent_buffer(root->node);
1412                 kfree(root->name);
1413                 kfree(root);
1414                 return ERR_PTR(ret);
1415         }
1416         root->in_sysfs = 1;
1417         return root;
1418 #endif
1419 }
1420
1421 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1422 {
1423         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1424         int ret = 0;
1425         struct btrfs_device *device;
1426         struct backing_dev_info *bdi;
1427
1428         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1429                 if (!device->bdev)
1430                         continue;
1431                 bdi = blk_get_backing_dev_info(device->bdev);
1432                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1433                         ret = 1;
1434                         break;
1435                 }
1436         }
1437         return ret;
1438 }
1439
1440 /*
1441  * If this fails, caller must call bdi_destroy() to get rid of the
1442  * bdi again.
1443  */
1444 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1445 {
1446         int err;
1447
1448         bdi->capabilities = BDI_CAP_MAP_COPY;
1449         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1450         if (err)
1451                 return err;
1452
1453         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1454         bdi->congested_fn       = btrfs_congested_fn;
1455         bdi->congested_data     = info;
1456         return 0;
1457 }
1458
1459 static int bio_ready_for_csum(struct bio *bio)
1460 {
1461         u64 length = 0;
1462         u64 buf_len = 0;
1463         u64 start = 0;
1464         struct page *page;
1465         struct extent_io_tree *io_tree = NULL;
1466         struct bio_vec *bvec;
1467         int i;
1468         int ret;
1469
1470         bio_for_each_segment(bvec, bio, i) {
1471                 page = bvec->bv_page;
1472                 if (page->private == EXTENT_PAGE_PRIVATE) {
1473                         length += bvec->bv_len;
1474                         continue;
1475                 }
1476                 if (!page->private) {
1477                         length += bvec->bv_len;
1478                         continue;
1479                 }
1480                 length = bvec->bv_len;
1481                 buf_len = page->private >> 2;
1482                 start = page_offset(page) + bvec->bv_offset;
1483                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1484         }
1485         /* are we fully contained in this bio? */
1486         if (buf_len <= length)
1487                 return 1;
1488
1489         ret = extent_range_uptodate(io_tree, start + length,
1490                                     start + buf_len - 1);
1491         return ret;
1492 }
1493
1494 /*
1495  * called by the kthread helper functions to finally call the bio end_io
1496  * functions.  This is where read checksum verification actually happens
1497  */
1498 static void end_workqueue_fn(struct btrfs_work *work)
1499 {
1500         struct bio *bio;
1501         struct end_io_wq *end_io_wq;
1502         struct btrfs_fs_info *fs_info;
1503         int error;
1504
1505         end_io_wq = container_of(work, struct end_io_wq, work);
1506         bio = end_io_wq->bio;
1507         fs_info = end_io_wq->info;
1508
1509         /* metadata bio reads are special because the whole tree block must
1510          * be checksummed at once.  This makes sure the entire block is in
1511          * ram and up to date before trying to verify things.  For
1512          * blocksize <= pagesize, it is basically a noop
1513          */
1514         if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1515             !bio_ready_for_csum(bio)) {
1516                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1517                                    &end_io_wq->work);
1518                 return;
1519         }
1520         error = end_io_wq->error;
1521         bio->bi_private = end_io_wq->private;
1522         bio->bi_end_io = end_io_wq->end_io;
1523         kfree(end_io_wq);
1524         bio_endio(bio, error);
1525 }
1526
1527 static int cleaner_kthread(void *arg)
1528 {
1529         struct btrfs_root *root = arg;
1530
1531         do {
1532                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1533
1534                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1535                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1536                         btrfs_run_delayed_iputs(root);
1537                         btrfs_clean_old_snapshots(root);
1538                         mutex_unlock(&root->fs_info->cleaner_mutex);
1539                 }
1540
1541                 if (freezing(current)) {
1542                         refrigerator();
1543                 } else {
1544                         set_current_state(TASK_INTERRUPTIBLE);
1545                         if (!kthread_should_stop())
1546                                 schedule();
1547                         __set_current_state(TASK_RUNNING);
1548                 }
1549         } while (!kthread_should_stop());
1550         return 0;
1551 }
1552
1553 static int transaction_kthread(void *arg)
1554 {
1555         struct btrfs_root *root = arg;
1556         struct btrfs_trans_handle *trans;
1557         struct btrfs_transaction *cur;
1558         u64 transid;
1559         unsigned long now;
1560         unsigned long delay;
1561         int ret;
1562
1563         do {
1564                 delay = HZ * 30;
1565                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1566                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1567
1568                 spin_lock(&root->fs_info->new_trans_lock);
1569                 cur = root->fs_info->running_transaction;
1570                 if (!cur) {
1571                         spin_unlock(&root->fs_info->new_trans_lock);
1572                         goto sleep;
1573                 }
1574
1575                 now = get_seconds();
1576                 if (!cur->blocked &&
1577                     (now < cur->start_time || now - cur->start_time < 30)) {
1578                         spin_unlock(&root->fs_info->new_trans_lock);
1579                         delay = HZ * 5;
1580                         goto sleep;
1581                 }
1582                 transid = cur->transid;
1583                 spin_unlock(&root->fs_info->new_trans_lock);
1584
1585                 trans = btrfs_join_transaction(root, 1);
1586                 BUG_ON(IS_ERR(trans));
1587                 if (transid == trans->transid) {
1588                         ret = btrfs_commit_transaction(trans, root);
1589                         BUG_ON(ret);
1590                 } else {
1591                         btrfs_end_transaction(trans, root);
1592                 }
1593 sleep:
1594                 wake_up_process(root->fs_info->cleaner_kthread);
1595                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1596
1597                 if (freezing(current)) {
1598                         refrigerator();
1599                 } else {
1600                         set_current_state(TASK_INTERRUPTIBLE);
1601                         if (!kthread_should_stop() &&
1602                             !btrfs_transaction_blocked(root->fs_info))
1603                                 schedule_timeout(delay);
1604                         __set_current_state(TASK_RUNNING);
1605                 }
1606         } while (!kthread_should_stop());
1607         return 0;
1608 }
1609
1610 struct btrfs_root *open_ctree(struct super_block *sb,
1611                               struct btrfs_fs_devices *fs_devices,
1612                               char *options)
1613 {
1614         u32 sectorsize;
1615         u32 nodesize;
1616         u32 leafsize;
1617         u32 blocksize;
1618         u32 stripesize;
1619         u64 generation;
1620         u64 features;
1621         struct btrfs_key location;
1622         struct buffer_head *bh;
1623         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1624                                                  GFP_NOFS);
1625         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1626                                                  GFP_NOFS);
1627         struct btrfs_root *tree_root = btrfs_sb(sb);
1628         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1629         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1630                                                 GFP_NOFS);
1631         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1632                                               GFP_NOFS);
1633         struct btrfs_root *log_tree_root;
1634
1635         int ret;
1636         int err = -EINVAL;
1637
1638         struct btrfs_super_block *disk_super;
1639
1640         if (!extent_root || !tree_root || !fs_info ||
1641             !chunk_root || !dev_root || !csum_root) {
1642                 err = -ENOMEM;
1643                 goto fail;
1644         }
1645
1646         ret = init_srcu_struct(&fs_info->subvol_srcu);
1647         if (ret) {
1648                 err = ret;
1649                 goto fail;
1650         }
1651
1652         ret = setup_bdi(fs_info, &fs_info->bdi);
1653         if (ret) {
1654                 err = ret;
1655                 goto fail_srcu;
1656         }
1657
1658         fs_info->btree_inode = new_inode(sb);
1659         if (!fs_info->btree_inode) {
1660                 err = -ENOMEM;
1661                 goto fail_bdi;
1662         }
1663
1664         fs_info->btree_inode->i_mapping->flags &= ~__GFP_FS;
1665
1666         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1667         INIT_LIST_HEAD(&fs_info->trans_list);
1668         INIT_LIST_HEAD(&fs_info->dead_roots);
1669         INIT_LIST_HEAD(&fs_info->delayed_iputs);
1670         INIT_LIST_HEAD(&fs_info->hashers);
1671         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1672         INIT_LIST_HEAD(&fs_info->ordered_operations);
1673         INIT_LIST_HEAD(&fs_info->caching_block_groups);
1674         spin_lock_init(&fs_info->delalloc_lock);
1675         spin_lock_init(&fs_info->new_trans_lock);
1676         spin_lock_init(&fs_info->ref_cache_lock);
1677         spin_lock_init(&fs_info->fs_roots_radix_lock);
1678         spin_lock_init(&fs_info->delayed_iput_lock);
1679
1680         init_completion(&fs_info->kobj_unregister);
1681         fs_info->tree_root = tree_root;
1682         fs_info->extent_root = extent_root;
1683         fs_info->csum_root = csum_root;
1684         fs_info->chunk_root = chunk_root;
1685         fs_info->dev_root = dev_root;
1686         fs_info->fs_devices = fs_devices;
1687         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1688         INIT_LIST_HEAD(&fs_info->space_info);
1689         btrfs_mapping_init(&fs_info->mapping_tree);
1690         btrfs_init_block_rsv(&fs_info->global_block_rsv);
1691         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1692         btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1693         btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1694         btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1695         INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1696         mutex_init(&fs_info->durable_block_rsv_mutex);
1697         atomic_set(&fs_info->nr_async_submits, 0);
1698         atomic_set(&fs_info->async_delalloc_pages, 0);
1699         atomic_set(&fs_info->async_submit_draining, 0);
1700         atomic_set(&fs_info->nr_async_bios, 0);
1701         fs_info->sb = sb;
1702         fs_info->max_inline = 8192 * 1024;
1703         fs_info->metadata_ratio = 0;
1704
1705         fs_info->thread_pool_size = min_t(unsigned long,
1706                                           num_online_cpus() + 2, 8);
1707
1708         INIT_LIST_HEAD(&fs_info->ordered_extents);
1709         spin_lock_init(&fs_info->ordered_extent_lock);
1710
1711         sb->s_blocksize = 4096;
1712         sb->s_blocksize_bits = blksize_bits(4096);
1713         sb->s_bdi = &fs_info->bdi;
1714
1715         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1716         fs_info->btree_inode->i_nlink = 1;
1717         /*
1718          * we set the i_size on the btree inode to the max possible int.
1719          * the real end of the address space is determined by all of
1720          * the devices in the system
1721          */
1722         fs_info->btree_inode->i_size = OFFSET_MAX;
1723         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1724         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1725
1726         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1727         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1728                              fs_info->btree_inode->i_mapping,
1729                              GFP_NOFS);
1730         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1731                              GFP_NOFS);
1732
1733         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1734
1735         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1736         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1737                sizeof(struct btrfs_key));
1738         BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1739         insert_inode_hash(fs_info->btree_inode);
1740
1741         spin_lock_init(&fs_info->block_group_cache_lock);
1742         fs_info->block_group_cache_tree = RB_ROOT;
1743
1744         extent_io_tree_init(&fs_info->freed_extents[0],
1745                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1746         extent_io_tree_init(&fs_info->freed_extents[1],
1747                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1748         fs_info->pinned_extents = &fs_info->freed_extents[0];
1749         fs_info->do_barriers = 1;
1750
1751
1752         mutex_init(&fs_info->trans_mutex);
1753         mutex_init(&fs_info->ordered_operations_mutex);
1754         mutex_init(&fs_info->tree_log_mutex);
1755         mutex_init(&fs_info->chunk_mutex);
1756         mutex_init(&fs_info->transaction_kthread_mutex);
1757         mutex_init(&fs_info->cleaner_mutex);
1758         mutex_init(&fs_info->volume_mutex);
1759         init_rwsem(&fs_info->extent_commit_sem);
1760         init_rwsem(&fs_info->cleanup_work_sem);
1761         init_rwsem(&fs_info->subvol_sem);
1762
1763         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1764         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1765
1766         init_waitqueue_head(&fs_info->transaction_throttle);
1767         init_waitqueue_head(&fs_info->transaction_wait);
1768         init_waitqueue_head(&fs_info->transaction_blocked_wait);
1769         init_waitqueue_head(&fs_info->async_submit_wait);
1770
1771         __setup_root(4096, 4096, 4096, 4096, tree_root,
1772                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1773
1774         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1775         if (!bh) {
1776                 err = -EINVAL;
1777                 goto fail_iput;
1778         }
1779
1780         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1781         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1782                sizeof(fs_info->super_for_commit));
1783         brelse(bh);
1784
1785         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1786
1787         disk_super = &fs_info->super_copy;
1788         if (!btrfs_super_root(disk_super))
1789                 goto fail_iput;
1790
1791         /* check FS state, whether FS is broken. */
1792         fs_info->fs_state |= btrfs_super_flags(disk_super);
1793
1794         btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
1795
1796         /*
1797          * In the long term, we'll store the compression type in the super
1798          * block, and it'll be used for per file compression control.
1799          */
1800         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
1801
1802         ret = btrfs_parse_options(tree_root, options);
1803         if (ret) {
1804                 err = ret;
1805                 goto fail_iput;
1806         }
1807
1808         features = btrfs_super_incompat_flags(disk_super) &
1809                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1810         if (features) {
1811                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1812                        "unsupported optional features (%Lx).\n",
1813                        (unsigned long long)features);
1814                 err = -EINVAL;
1815                 goto fail_iput;
1816         }
1817
1818         features = btrfs_super_incompat_flags(disk_super);
1819         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1820         if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1821                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1822         btrfs_set_super_incompat_flags(disk_super, features);
1823
1824         features = btrfs_super_compat_ro_flags(disk_super) &
1825                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1826         if (!(sb->s_flags & MS_RDONLY) && features) {
1827                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1828                        "unsupported option features (%Lx).\n",
1829                        (unsigned long long)features);
1830                 err = -EINVAL;
1831                 goto fail_iput;
1832         }
1833
1834         btrfs_init_workers(&fs_info->generic_worker,
1835                            "genwork", 1, NULL);
1836
1837         btrfs_init_workers(&fs_info->workers, "worker",
1838                            fs_info->thread_pool_size,
1839                            &fs_info->generic_worker);
1840
1841         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1842                            fs_info->thread_pool_size,
1843                            &fs_info->generic_worker);
1844
1845         btrfs_init_workers(&fs_info->submit_workers, "submit",
1846                            min_t(u64, fs_devices->num_devices,
1847                            fs_info->thread_pool_size),
1848                            &fs_info->generic_worker);
1849
1850         /* a higher idle thresh on the submit workers makes it much more
1851          * likely that bios will be send down in a sane order to the
1852          * devices
1853          */
1854         fs_info->submit_workers.idle_thresh = 64;
1855
1856         fs_info->workers.idle_thresh = 16;
1857         fs_info->workers.ordered = 1;
1858
1859         fs_info->delalloc_workers.idle_thresh = 2;
1860         fs_info->delalloc_workers.ordered = 1;
1861
1862         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1863                            &fs_info->generic_worker);
1864         btrfs_init_workers(&fs_info->endio_workers, "endio",
1865                            fs_info->thread_pool_size,
1866                            &fs_info->generic_worker);
1867         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1868                            fs_info->thread_pool_size,
1869                            &fs_info->generic_worker);
1870         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1871                            "endio-meta-write", fs_info->thread_pool_size,
1872                            &fs_info->generic_worker);
1873         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1874                            fs_info->thread_pool_size,
1875                            &fs_info->generic_worker);
1876         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1877                            1, &fs_info->generic_worker);
1878
1879         /*
1880          * endios are largely parallel and should have a very
1881          * low idle thresh
1882          */
1883         fs_info->endio_workers.idle_thresh = 4;
1884         fs_info->endio_meta_workers.idle_thresh = 4;
1885
1886         fs_info->endio_write_workers.idle_thresh = 2;
1887         fs_info->endio_meta_write_workers.idle_thresh = 2;
1888
1889         btrfs_start_workers(&fs_info->workers, 1);
1890         btrfs_start_workers(&fs_info->generic_worker, 1);
1891         btrfs_start_workers(&fs_info->submit_workers, 1);
1892         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1893         btrfs_start_workers(&fs_info->fixup_workers, 1);
1894         btrfs_start_workers(&fs_info->endio_workers, 1);
1895         btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1896         btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1897         btrfs_start_workers(&fs_info->endio_write_workers, 1);
1898         btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1899
1900         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1901         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1902                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1903
1904         nodesize = btrfs_super_nodesize(disk_super);
1905         leafsize = btrfs_super_leafsize(disk_super);
1906         sectorsize = btrfs_super_sectorsize(disk_super);
1907         stripesize = btrfs_super_stripesize(disk_super);
1908         tree_root->nodesize = nodesize;
1909         tree_root->leafsize = leafsize;
1910         tree_root->sectorsize = sectorsize;
1911         tree_root->stripesize = stripesize;
1912
1913         sb->s_blocksize = sectorsize;
1914         sb->s_blocksize_bits = blksize_bits(sectorsize);
1915
1916         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1917                     sizeof(disk_super->magic))) {
1918                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1919                 goto fail_sb_buffer;
1920         }
1921
1922         mutex_lock(&fs_info->chunk_mutex);
1923         ret = btrfs_read_sys_array(tree_root);
1924         mutex_unlock(&fs_info->chunk_mutex);
1925         if (ret) {
1926                 printk(KERN_WARNING "btrfs: failed to read the system "
1927                        "array on %s\n", sb->s_id);
1928                 goto fail_sb_buffer;
1929         }
1930
1931         blocksize = btrfs_level_size(tree_root,
1932                                      btrfs_super_chunk_root_level(disk_super));
1933         generation = btrfs_super_chunk_root_generation(disk_super);
1934
1935         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1936                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1937
1938         chunk_root->node = read_tree_block(chunk_root,
1939                                            btrfs_super_chunk_root(disk_super),
1940                                            blocksize, generation);
1941         BUG_ON(!chunk_root->node);
1942         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1943                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1944                        sb->s_id);
1945                 goto fail_chunk_root;
1946         }
1947         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1948         chunk_root->commit_root = btrfs_root_node(chunk_root);
1949
1950         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1951            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1952            BTRFS_UUID_SIZE);
1953
1954         mutex_lock(&fs_info->chunk_mutex);
1955         ret = btrfs_read_chunk_tree(chunk_root);
1956         mutex_unlock(&fs_info->chunk_mutex);
1957         if (ret) {
1958                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1959                        sb->s_id);
1960                 goto fail_chunk_root;
1961         }
1962
1963         btrfs_close_extra_devices(fs_devices);
1964
1965         blocksize = btrfs_level_size(tree_root,
1966                                      btrfs_super_root_level(disk_super));
1967         generation = btrfs_super_generation(disk_super);
1968
1969         tree_root->node = read_tree_block(tree_root,
1970                                           btrfs_super_root(disk_super),
1971                                           blocksize, generation);
1972         if (!tree_root->node)
1973                 goto fail_chunk_root;
1974         if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1975                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1976                        sb->s_id);
1977                 goto fail_tree_root;
1978         }
1979         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1980         tree_root->commit_root = btrfs_root_node(tree_root);
1981
1982         ret = find_and_setup_root(tree_root, fs_info,
1983                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1984         if (ret)
1985                 goto fail_tree_root;
1986         extent_root->track_dirty = 1;
1987
1988         ret = find_and_setup_root(tree_root, fs_info,
1989                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1990         if (ret)
1991                 goto fail_extent_root;
1992         dev_root->track_dirty = 1;
1993
1994         ret = find_and_setup_root(tree_root, fs_info,
1995                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
1996         if (ret)
1997                 goto fail_dev_root;
1998
1999         csum_root->track_dirty = 1;
2000
2001         fs_info->generation = generation;
2002         fs_info->last_trans_committed = generation;
2003         fs_info->data_alloc_profile = (u64)-1;
2004         fs_info->metadata_alloc_profile = (u64)-1;
2005         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
2006
2007         ret = btrfs_init_space_info(fs_info);
2008         if (ret) {
2009                 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2010                 goto fail_block_groups;
2011         }
2012
2013         ret = btrfs_read_block_groups(extent_root);
2014         if (ret) {
2015                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2016                 goto fail_block_groups;
2017         }
2018
2019         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2020                                                "btrfs-cleaner");
2021         if (IS_ERR(fs_info->cleaner_kthread))
2022                 goto fail_block_groups;
2023
2024         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2025                                                    tree_root,
2026                                                    "btrfs-transaction");
2027         if (IS_ERR(fs_info->transaction_kthread))
2028                 goto fail_cleaner;
2029
2030         if (!btrfs_test_opt(tree_root, SSD) &&
2031             !btrfs_test_opt(tree_root, NOSSD) &&
2032             !fs_info->fs_devices->rotating) {
2033                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2034                        "mode\n");
2035                 btrfs_set_opt(fs_info->mount_opt, SSD);
2036         }
2037
2038         /* do not make disk changes in broken FS */
2039         if (btrfs_super_log_root(disk_super) != 0 &&
2040             !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2041                 u64 bytenr = btrfs_super_log_root(disk_super);
2042
2043                 if (fs_devices->rw_devices == 0) {
2044                         printk(KERN_WARNING "Btrfs log replay required "
2045                                "on RO media\n");
2046                         err = -EIO;
2047                         goto fail_trans_kthread;
2048                 }
2049                 blocksize =
2050                      btrfs_level_size(tree_root,
2051                                       btrfs_super_log_root_level(disk_super));
2052
2053                 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2054                 if (!log_tree_root) {
2055                         err = -ENOMEM;
2056                         goto fail_trans_kthread;
2057                 }
2058
2059                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2060                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2061
2062                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2063                                                       blocksize,
2064                                                       generation + 1);
2065                 ret = btrfs_recover_log_trees(log_tree_root);
2066                 BUG_ON(ret);
2067
2068                 if (sb->s_flags & MS_RDONLY) {
2069                         ret =  btrfs_commit_super(tree_root);
2070                         BUG_ON(ret);
2071                 }
2072         }
2073
2074         ret = btrfs_find_orphan_roots(tree_root);
2075         BUG_ON(ret);
2076
2077         if (!(sb->s_flags & MS_RDONLY)) {
2078                 ret = btrfs_cleanup_fs_roots(fs_info);
2079                 BUG_ON(ret);
2080
2081                 ret = btrfs_recover_relocation(tree_root);
2082                 if (ret < 0) {
2083                         printk(KERN_WARNING
2084                                "btrfs: failed to recover relocation\n");
2085                         err = -EINVAL;
2086                         goto fail_trans_kthread;
2087                 }
2088         }
2089
2090         location.objectid = BTRFS_FS_TREE_OBJECTID;
2091         location.type = BTRFS_ROOT_ITEM_KEY;
2092         location.offset = (u64)-1;
2093
2094         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2095         if (!fs_info->fs_root)
2096                 goto fail_trans_kthread;
2097         if (IS_ERR(fs_info->fs_root)) {
2098                 err = PTR_ERR(fs_info->fs_root);
2099                 goto fail_trans_kthread;
2100         }
2101
2102         if (!(sb->s_flags & MS_RDONLY)) {
2103                 down_read(&fs_info->cleanup_work_sem);
2104                 err = btrfs_orphan_cleanup(fs_info->fs_root);
2105                 if (!err)
2106                         err = btrfs_orphan_cleanup(fs_info->tree_root);
2107                 up_read(&fs_info->cleanup_work_sem);
2108                 if (err) {
2109                         close_ctree(tree_root);
2110                         return ERR_PTR(err);
2111                 }
2112         }
2113
2114         return tree_root;
2115
2116 fail_trans_kthread:
2117         kthread_stop(fs_info->transaction_kthread);
2118 fail_cleaner:
2119         kthread_stop(fs_info->cleaner_kthread);
2120
2121         /*
2122          * make sure we're done with the btree inode before we stop our
2123          * kthreads
2124          */
2125         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2126         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2127
2128 fail_block_groups:
2129         btrfs_free_block_groups(fs_info);
2130         free_extent_buffer(csum_root->node);
2131         free_extent_buffer(csum_root->commit_root);
2132 fail_dev_root:
2133         free_extent_buffer(dev_root->node);
2134         free_extent_buffer(dev_root->commit_root);
2135 fail_extent_root:
2136         free_extent_buffer(extent_root->node);
2137         free_extent_buffer(extent_root->commit_root);
2138 fail_tree_root:
2139         free_extent_buffer(tree_root->node);
2140         free_extent_buffer(tree_root->commit_root);
2141 fail_chunk_root:
2142         free_extent_buffer(chunk_root->node);
2143         free_extent_buffer(chunk_root->commit_root);
2144 fail_sb_buffer:
2145         btrfs_stop_workers(&fs_info->generic_worker);
2146         btrfs_stop_workers(&fs_info->fixup_workers);
2147         btrfs_stop_workers(&fs_info->delalloc_workers);
2148         btrfs_stop_workers(&fs_info->workers);
2149         btrfs_stop_workers(&fs_info->endio_workers);
2150         btrfs_stop_workers(&fs_info->endio_meta_workers);
2151         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2152         btrfs_stop_workers(&fs_info->endio_write_workers);
2153         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2154         btrfs_stop_workers(&fs_info->submit_workers);
2155 fail_iput:
2156         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2157         iput(fs_info->btree_inode);
2158
2159         btrfs_close_devices(fs_info->fs_devices);
2160         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2161 fail_bdi:
2162         bdi_destroy(&fs_info->bdi);
2163 fail_srcu:
2164         cleanup_srcu_struct(&fs_info->subvol_srcu);
2165 fail:
2166         kfree(extent_root);
2167         kfree(tree_root);
2168         kfree(fs_info);
2169         kfree(chunk_root);
2170         kfree(dev_root);
2171         kfree(csum_root);
2172         return ERR_PTR(err);
2173 }
2174
2175 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2176 {
2177         char b[BDEVNAME_SIZE];
2178
2179         if (uptodate) {
2180                 set_buffer_uptodate(bh);
2181         } else {
2182                 if (printk_ratelimit()) {
2183                         printk(KERN_WARNING "lost page write due to "
2184                                         "I/O error on %s\n",
2185                                        bdevname(bh->b_bdev, b));
2186                 }
2187                 /* note, we dont' set_buffer_write_io_error because we have
2188                  * our own ways of dealing with the IO errors
2189                  */
2190                 clear_buffer_uptodate(bh);
2191         }
2192         unlock_buffer(bh);
2193         put_bh(bh);
2194 }
2195
2196 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2197 {
2198         struct buffer_head *bh;
2199         struct buffer_head *latest = NULL;
2200         struct btrfs_super_block *super;
2201         int i;
2202         u64 transid = 0;
2203         u64 bytenr;
2204
2205         /* we would like to check all the supers, but that would make
2206          * a btrfs mount succeed after a mkfs from a different FS.
2207          * So, we need to add a special mount option to scan for
2208          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2209          */
2210         for (i = 0; i < 1; i++) {
2211                 bytenr = btrfs_sb_offset(i);
2212                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2213                         break;
2214                 bh = __bread(bdev, bytenr / 4096, 4096);
2215                 if (!bh)
2216                         continue;
2217
2218                 super = (struct btrfs_super_block *)bh->b_data;
2219                 if (btrfs_super_bytenr(super) != bytenr ||
2220                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2221                             sizeof(super->magic))) {
2222                         brelse(bh);
2223                         continue;
2224                 }
2225
2226                 if (!latest || btrfs_super_generation(super) > transid) {
2227                         brelse(latest);
2228                         latest = bh;
2229                         transid = btrfs_super_generation(super);
2230                 } else {
2231                         brelse(bh);
2232                 }
2233         }
2234         return latest;
2235 }
2236
2237 /*
2238  * this should be called twice, once with wait == 0 and
2239  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2240  * we write are pinned.
2241  *
2242  * They are released when wait == 1 is done.
2243  * max_mirrors must be the same for both runs, and it indicates how
2244  * many supers on this one device should be written.
2245  *
2246  * max_mirrors == 0 means to write them all.
2247  */
2248 static int write_dev_supers(struct btrfs_device *device,
2249                             struct btrfs_super_block *sb,
2250                             int do_barriers, int wait, int max_mirrors)
2251 {
2252         struct buffer_head *bh;
2253         int i;
2254         int ret;
2255         int errors = 0;
2256         u32 crc;
2257         u64 bytenr;
2258         int last_barrier = 0;
2259
2260         if (max_mirrors == 0)
2261                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2262
2263         /* make sure only the last submit_bh does a barrier */
2264         if (do_barriers) {
2265                 for (i = 0; i < max_mirrors; i++) {
2266                         bytenr = btrfs_sb_offset(i);
2267                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2268                             device->total_bytes)
2269                                 break;
2270                         last_barrier = i;
2271                 }
2272         }
2273
2274         for (i = 0; i < max_mirrors; i++) {
2275                 bytenr = btrfs_sb_offset(i);
2276                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2277                         break;
2278
2279                 if (wait) {
2280                         bh = __find_get_block(device->bdev, bytenr / 4096,
2281                                               BTRFS_SUPER_INFO_SIZE);
2282                         BUG_ON(!bh);
2283                         wait_on_buffer(bh);
2284                         if (!buffer_uptodate(bh))
2285                                 errors++;
2286
2287                         /* drop our reference */
2288                         brelse(bh);
2289
2290                         /* drop the reference from the wait == 0 run */
2291                         brelse(bh);
2292                         continue;
2293                 } else {
2294                         btrfs_set_super_bytenr(sb, bytenr);
2295
2296                         crc = ~(u32)0;
2297                         crc = btrfs_csum_data(NULL, (char *)sb +
2298                                               BTRFS_CSUM_SIZE, crc,
2299                                               BTRFS_SUPER_INFO_SIZE -
2300                                               BTRFS_CSUM_SIZE);
2301                         btrfs_csum_final(crc, sb->csum);
2302
2303                         /*
2304                          * one reference for us, and we leave it for the
2305                          * caller
2306                          */
2307                         bh = __getblk(device->bdev, bytenr / 4096,
2308                                       BTRFS_SUPER_INFO_SIZE);
2309                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2310
2311                         /* one reference for submit_bh */
2312                         get_bh(bh);
2313
2314                         set_buffer_uptodate(bh);
2315                         lock_buffer(bh);
2316                         bh->b_end_io = btrfs_end_buffer_write_sync;
2317                 }
2318
2319                 if (i == last_barrier && do_barriers)
2320                         ret = submit_bh(WRITE_FLUSH_FUA, bh);
2321                 else
2322                         ret = submit_bh(WRITE_SYNC, bh);
2323
2324                 if (ret)
2325                         errors++;
2326         }
2327         return errors < i ? 0 : -1;
2328 }
2329
2330 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2331 {
2332         struct list_head *head;
2333         struct btrfs_device *dev;
2334         struct btrfs_super_block *sb;
2335         struct btrfs_dev_item *dev_item;
2336         int ret;
2337         int do_barriers;
2338         int max_errors;
2339         int total_errors = 0;
2340         u64 flags;
2341
2342         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2343         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2344
2345         sb = &root->fs_info->super_for_commit;
2346         dev_item = &sb->dev_item;
2347
2348         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2349         head = &root->fs_info->fs_devices->devices;
2350         list_for_each_entry(dev, head, dev_list) {
2351                 if (!dev->bdev) {
2352                         total_errors++;
2353                         continue;
2354                 }
2355                 if (!dev->in_fs_metadata || !dev->writeable)
2356                         continue;
2357
2358                 btrfs_set_stack_device_generation(dev_item, 0);
2359                 btrfs_set_stack_device_type(dev_item, dev->type);
2360                 btrfs_set_stack_device_id(dev_item, dev->devid);
2361                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2362                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2363                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2364                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2365                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2366                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2367                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2368
2369                 flags = btrfs_super_flags(sb);
2370                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2371
2372                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2373                 if (ret)
2374                         total_errors++;
2375         }
2376         if (total_errors > max_errors) {
2377                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2378                        total_errors);
2379                 BUG();
2380         }
2381
2382         total_errors = 0;
2383         list_for_each_entry(dev, head, dev_list) {
2384                 if (!dev->bdev)
2385                         continue;
2386                 if (!dev->in_fs_metadata || !dev->writeable)
2387                         continue;
2388
2389                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2390                 if (ret)
2391                         total_errors++;
2392         }
2393         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2394         if (total_errors > max_errors) {
2395                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2396                        total_errors);
2397                 BUG();
2398         }
2399         return 0;
2400 }
2401
2402 int write_ctree_super(struct btrfs_trans_handle *trans,
2403                       struct btrfs_root *root, int max_mirrors)
2404 {
2405         int ret;
2406
2407         ret = write_all_supers(root, max_mirrors);
2408         return ret;
2409 }
2410
2411 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2412 {
2413         spin_lock(&fs_info->fs_roots_radix_lock);
2414         radix_tree_delete(&fs_info->fs_roots_radix,
2415                           (unsigned long)root->root_key.objectid);
2416         spin_unlock(&fs_info->fs_roots_radix_lock);
2417
2418         if (btrfs_root_refs(&root->root_item) == 0)
2419                 synchronize_srcu(&fs_info->subvol_srcu);
2420
2421         __btrfs_remove_free_space_cache(root->free_ino_pinned);
2422         __btrfs_remove_free_space_cache(root->free_ino_ctl);
2423         free_fs_root(root);
2424         return 0;
2425 }
2426
2427 static void free_fs_root(struct btrfs_root *root)
2428 {
2429         iput(root->cache_inode);
2430         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2431         if (root->anon_super.s_dev) {
2432                 down_write(&root->anon_super.s_umount);
2433                 kill_anon_super(&root->anon_super);
2434         }
2435         free_extent_buffer(root->node);
2436         free_extent_buffer(root->commit_root);
2437         kfree(root->free_ino_ctl);
2438         kfree(root->free_ino_pinned);
2439         kfree(root->name);
2440         kfree(root);
2441 }
2442
2443 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2444 {
2445         int ret;
2446         struct btrfs_root *gang[8];
2447         int i;
2448
2449         while (!list_empty(&fs_info->dead_roots)) {
2450                 gang[0] = list_entry(fs_info->dead_roots.next,
2451                                      struct btrfs_root, root_list);
2452                 list_del(&gang[0]->root_list);
2453
2454                 if (gang[0]->in_radix) {
2455                         btrfs_free_fs_root(fs_info, gang[0]);
2456                 } else {
2457                         free_extent_buffer(gang[0]->node);
2458                         free_extent_buffer(gang[0]->commit_root);
2459                         kfree(gang[0]);
2460                 }
2461         }
2462
2463         while (1) {
2464                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2465                                              (void **)gang, 0,
2466                                              ARRAY_SIZE(gang));
2467                 if (!ret)
2468                         break;
2469                 for (i = 0; i < ret; i++)
2470                         btrfs_free_fs_root(fs_info, gang[i]);
2471         }
2472         return 0;
2473 }
2474
2475 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2476 {
2477         u64 root_objectid = 0;
2478         struct btrfs_root *gang[8];
2479         int i;
2480         int ret;
2481
2482         while (1) {
2483                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2484                                              (void **)gang, root_objectid,
2485                                              ARRAY_SIZE(gang));
2486                 if (!ret)
2487                         break;
2488
2489                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2490                 for (i = 0; i < ret; i++) {
2491                         int err;
2492
2493                         root_objectid = gang[i]->root_key.objectid;
2494                         err = btrfs_orphan_cleanup(gang[i]);
2495                         if (err)
2496                                 return err;
2497                 }
2498                 root_objectid++;
2499         }
2500         return 0;
2501 }
2502
2503 int btrfs_commit_super(struct btrfs_root *root)
2504 {
2505         struct btrfs_trans_handle *trans;
2506         int ret;
2507
2508         mutex_lock(&root->fs_info->cleaner_mutex);
2509         btrfs_run_delayed_iputs(root);
2510         btrfs_clean_old_snapshots(root);
2511         mutex_unlock(&root->fs_info->cleaner_mutex);
2512
2513         /* wait until ongoing cleanup work done */
2514         down_write(&root->fs_info->cleanup_work_sem);
2515         up_write(&root->fs_info->cleanup_work_sem);
2516
2517         trans = btrfs_join_transaction(root, 1);
2518         if (IS_ERR(trans))
2519                 return PTR_ERR(trans);
2520         ret = btrfs_commit_transaction(trans, root);
2521         BUG_ON(ret);
2522         /* run commit again to drop the original snapshot */
2523         trans = btrfs_join_transaction(root, 1);
2524         if (IS_ERR(trans))
2525                 return PTR_ERR(trans);
2526         btrfs_commit_transaction(trans, root);
2527         ret = btrfs_write_and_wait_transaction(NULL, root);
2528         BUG_ON(ret);
2529
2530         ret = write_ctree_super(NULL, root, 0);
2531         return ret;
2532 }
2533
2534 int close_ctree(struct btrfs_root *root)
2535 {
2536         struct btrfs_fs_info *fs_info = root->fs_info;
2537         int ret;
2538
2539         fs_info->closing = 1;
2540         smp_mb();
2541
2542         btrfs_put_block_group_cache(fs_info);
2543
2544         /*
2545          * Here come 2 situations when btrfs is broken to flip readonly:
2546          *
2547          * 1. when btrfs flips readonly somewhere else before
2548          * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2549          * and btrfs will skip to write sb directly to keep
2550          * ERROR state on disk.
2551          *
2552          * 2. when btrfs flips readonly just in btrfs_commit_super,
2553          * and in such case, btrfs cannot write sb via btrfs_commit_super,
2554          * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2555          * btrfs will cleanup all FS resources first and write sb then.
2556          */
2557         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2558                 ret = btrfs_commit_super(root);
2559                 if (ret)
2560                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2561         }
2562
2563         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2564                 ret = btrfs_error_commit_super(root);
2565                 if (ret)
2566                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2567         }
2568
2569         kthread_stop(root->fs_info->transaction_kthread);
2570         kthread_stop(root->fs_info->cleaner_kthread);
2571
2572         fs_info->closing = 2;
2573         smp_mb();
2574
2575         if (fs_info->delalloc_bytes) {
2576                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2577                        (unsigned long long)fs_info->delalloc_bytes);
2578         }
2579         if (fs_info->total_ref_cache_size) {
2580                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2581                        (unsigned long long)fs_info->total_ref_cache_size);
2582         }
2583
2584         free_extent_buffer(fs_info->extent_root->node);
2585         free_extent_buffer(fs_info->extent_root->commit_root);
2586         free_extent_buffer(fs_info->tree_root->node);
2587         free_extent_buffer(fs_info->tree_root->commit_root);
2588         free_extent_buffer(root->fs_info->chunk_root->node);
2589         free_extent_buffer(root->fs_info->chunk_root->commit_root);
2590         free_extent_buffer(root->fs_info->dev_root->node);
2591         free_extent_buffer(root->fs_info->dev_root->commit_root);
2592         free_extent_buffer(root->fs_info->csum_root->node);
2593         free_extent_buffer(root->fs_info->csum_root->commit_root);
2594
2595         btrfs_free_block_groups(root->fs_info);
2596
2597         del_fs_roots(fs_info);
2598
2599         iput(fs_info->btree_inode);
2600
2601         btrfs_stop_workers(&fs_info->generic_worker);
2602         btrfs_stop_workers(&fs_info->fixup_workers);
2603         btrfs_stop_workers(&fs_info->delalloc_workers);
2604         btrfs_stop_workers(&fs_info->workers);
2605         btrfs_stop_workers(&fs_info->endio_workers);
2606         btrfs_stop_workers(&fs_info->endio_meta_workers);
2607         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2608         btrfs_stop_workers(&fs_info->endio_write_workers);
2609         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2610         btrfs_stop_workers(&fs_info->submit_workers);
2611
2612         btrfs_close_devices(fs_info->fs_devices);
2613         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2614
2615         bdi_destroy(&fs_info->bdi);
2616         cleanup_srcu_struct(&fs_info->subvol_srcu);
2617
2618         kfree(fs_info->extent_root);
2619         kfree(fs_info->tree_root);
2620         kfree(fs_info->chunk_root);
2621         kfree(fs_info->dev_root);
2622         kfree(fs_info->csum_root);
2623         kfree(fs_info);
2624
2625         return 0;
2626 }
2627
2628 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2629 {
2630         int ret;
2631         struct inode *btree_inode = buf->first_page->mapping->host;
2632
2633         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2634                                      NULL);
2635         if (!ret)
2636                 return ret;
2637
2638         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2639                                     parent_transid);
2640         return !ret;
2641 }
2642
2643 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2644 {
2645         struct inode *btree_inode = buf->first_page->mapping->host;
2646         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2647                                           buf);
2648 }
2649
2650 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2651 {
2652         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2653         u64 transid = btrfs_header_generation(buf);
2654         struct inode *btree_inode = root->fs_info->btree_inode;
2655         int was_dirty;
2656
2657         btrfs_assert_tree_locked(buf);
2658         if (transid != root->fs_info->generation) {
2659                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2660                        "found %llu running %llu\n",
2661                         (unsigned long long)buf->start,
2662                         (unsigned long long)transid,
2663                         (unsigned long long)root->fs_info->generation);
2664                 WARN_ON(1);
2665         }
2666         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2667                                             buf);
2668         if (!was_dirty) {
2669                 spin_lock(&root->fs_info->delalloc_lock);
2670                 root->fs_info->dirty_metadata_bytes += buf->len;
2671                 spin_unlock(&root->fs_info->delalloc_lock);
2672         }
2673 }
2674
2675 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2676 {
2677         /*
2678          * looks as though older kernels can get into trouble with
2679          * this code, they end up stuck in balance_dirty_pages forever
2680          */
2681         u64 num_dirty;
2682         unsigned long thresh = 32 * 1024 * 1024;
2683
2684         if (current->flags & PF_MEMALLOC)
2685                 return;
2686
2687         num_dirty = root->fs_info->dirty_metadata_bytes;
2688
2689         if (num_dirty > thresh) {
2690                 balance_dirty_pages_ratelimited_nr(
2691                                    root->fs_info->btree_inode->i_mapping, 1);
2692         }
2693         return;
2694 }
2695
2696 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2697 {
2698         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2699         int ret;
2700         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2701         if (ret == 0)
2702                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2703         return ret;
2704 }
2705
2706 int btree_lock_page_hook(struct page *page)
2707 {
2708         struct inode *inode = page->mapping->host;
2709         struct btrfs_root *root = BTRFS_I(inode)->root;
2710         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2711         struct extent_buffer *eb;
2712         unsigned long len;
2713         u64 bytenr = page_offset(page);
2714
2715         if (page->private == EXTENT_PAGE_PRIVATE)
2716                 goto out;
2717
2718         len = page->private >> 2;
2719         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2720         if (!eb)
2721                 goto out;
2722
2723         btrfs_tree_lock(eb);
2724         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2725
2726         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2727                 spin_lock(&root->fs_info->delalloc_lock);
2728                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2729                         root->fs_info->dirty_metadata_bytes -= eb->len;
2730                 else
2731                         WARN_ON(1);
2732                 spin_unlock(&root->fs_info->delalloc_lock);
2733         }
2734
2735         btrfs_tree_unlock(eb);
2736         free_extent_buffer(eb);
2737 out:
2738         lock_page(page);
2739         return 0;
2740 }
2741
2742 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
2743                               int read_only)
2744 {
2745         if (read_only)
2746                 return;
2747
2748         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
2749                 printk(KERN_WARNING "warning: mount fs with errors, "
2750                        "running btrfsck is recommended\n");
2751 }
2752
2753 int btrfs_error_commit_super(struct btrfs_root *root)
2754 {
2755         int ret;
2756
2757         mutex_lock(&root->fs_info->cleaner_mutex);
2758         btrfs_run_delayed_iputs(root);
2759         mutex_unlock(&root->fs_info->cleaner_mutex);
2760
2761         down_write(&root->fs_info->cleanup_work_sem);
2762         up_write(&root->fs_info->cleanup_work_sem);
2763
2764         /* cleanup FS via transaction */
2765         btrfs_cleanup_transaction(root);
2766
2767         ret = write_ctree_super(NULL, root, 0);
2768
2769         return ret;
2770 }
2771
2772 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
2773 {
2774         struct btrfs_inode *btrfs_inode;
2775         struct list_head splice;
2776
2777         INIT_LIST_HEAD(&splice);
2778
2779         mutex_lock(&root->fs_info->ordered_operations_mutex);
2780         spin_lock(&root->fs_info->ordered_extent_lock);
2781
2782         list_splice_init(&root->fs_info->ordered_operations, &splice);
2783         while (!list_empty(&splice)) {
2784                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2785                                          ordered_operations);
2786
2787                 list_del_init(&btrfs_inode->ordered_operations);
2788
2789                 btrfs_invalidate_inodes(btrfs_inode->root);
2790         }
2791
2792         spin_unlock(&root->fs_info->ordered_extent_lock);
2793         mutex_unlock(&root->fs_info->ordered_operations_mutex);
2794
2795         return 0;
2796 }
2797
2798 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
2799 {
2800         struct list_head splice;
2801         struct btrfs_ordered_extent *ordered;
2802         struct inode *inode;
2803
2804         INIT_LIST_HEAD(&splice);
2805
2806         spin_lock(&root->fs_info->ordered_extent_lock);
2807
2808         list_splice_init(&root->fs_info->ordered_extents, &splice);
2809         while (!list_empty(&splice)) {
2810                 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
2811                                      root_extent_list);
2812
2813                 list_del_init(&ordered->root_extent_list);
2814                 atomic_inc(&ordered->refs);
2815
2816                 /* the inode may be getting freed (in sys_unlink path). */
2817                 inode = igrab(ordered->inode);
2818
2819                 spin_unlock(&root->fs_info->ordered_extent_lock);
2820                 if (inode)
2821                         iput(inode);
2822
2823                 atomic_set(&ordered->refs, 1);
2824                 btrfs_put_ordered_extent(ordered);
2825
2826                 spin_lock(&root->fs_info->ordered_extent_lock);
2827         }
2828
2829         spin_unlock(&root->fs_info->ordered_extent_lock);
2830
2831         return 0;
2832 }
2833
2834 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
2835                                       struct btrfs_root *root)
2836 {
2837         struct rb_node *node;
2838         struct btrfs_delayed_ref_root *delayed_refs;
2839         struct btrfs_delayed_ref_node *ref;
2840         int ret = 0;
2841
2842         delayed_refs = &trans->delayed_refs;
2843
2844         spin_lock(&delayed_refs->lock);
2845         if (delayed_refs->num_entries == 0) {
2846                 spin_unlock(&delayed_refs->lock);
2847                 printk(KERN_INFO "delayed_refs has NO entry\n");
2848                 return ret;
2849         }
2850
2851         node = rb_first(&delayed_refs->root);
2852         while (node) {
2853                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2854                 node = rb_next(node);
2855
2856                 ref->in_tree = 0;
2857                 rb_erase(&ref->rb_node, &delayed_refs->root);
2858                 delayed_refs->num_entries--;
2859
2860                 atomic_set(&ref->refs, 1);
2861                 if (btrfs_delayed_ref_is_head(ref)) {
2862                         struct btrfs_delayed_ref_head *head;
2863
2864                         head = btrfs_delayed_node_to_head(ref);
2865                         mutex_lock(&head->mutex);
2866                         kfree(head->extent_op);
2867                         delayed_refs->num_heads--;
2868                         if (list_empty(&head->cluster))
2869                                 delayed_refs->num_heads_ready--;
2870                         list_del_init(&head->cluster);
2871                         mutex_unlock(&head->mutex);
2872                 }
2873
2874                 spin_unlock(&delayed_refs->lock);
2875                 btrfs_put_delayed_ref(ref);
2876
2877                 cond_resched();
2878                 spin_lock(&delayed_refs->lock);
2879         }
2880
2881         spin_unlock(&delayed_refs->lock);
2882
2883         return ret;
2884 }
2885
2886 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
2887 {
2888         struct btrfs_pending_snapshot *snapshot;
2889         struct list_head splice;
2890
2891         INIT_LIST_HEAD(&splice);
2892
2893         list_splice_init(&t->pending_snapshots, &splice);
2894
2895         while (!list_empty(&splice)) {
2896                 snapshot = list_entry(splice.next,
2897                                       struct btrfs_pending_snapshot,
2898                                       list);
2899
2900                 list_del_init(&snapshot->list);
2901
2902                 kfree(snapshot);
2903         }
2904
2905         return 0;
2906 }
2907
2908 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
2909 {
2910         struct btrfs_inode *btrfs_inode;
2911         struct list_head splice;
2912
2913         INIT_LIST_HEAD(&splice);
2914
2915         list_splice_init(&root->fs_info->delalloc_inodes, &splice);
2916
2917         spin_lock(&root->fs_info->delalloc_lock);
2918
2919         while (!list_empty(&splice)) {
2920                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2921                                     delalloc_inodes);
2922
2923                 list_del_init(&btrfs_inode->delalloc_inodes);
2924
2925                 btrfs_invalidate_inodes(btrfs_inode->root);
2926         }
2927
2928         spin_unlock(&root->fs_info->delalloc_lock);
2929
2930         return 0;
2931 }
2932
2933 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
2934                                         struct extent_io_tree *dirty_pages,
2935                                         int mark)
2936 {
2937         int ret;
2938         struct page *page;
2939         struct inode *btree_inode = root->fs_info->btree_inode;
2940         struct extent_buffer *eb;
2941         u64 start = 0;
2942         u64 end;
2943         u64 offset;
2944         unsigned long index;
2945
2946         while (1) {
2947                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
2948                                             mark);
2949                 if (ret)
2950                         break;
2951
2952                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
2953                 while (start <= end) {
2954                         index = start >> PAGE_CACHE_SHIFT;
2955                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
2956                         page = find_get_page(btree_inode->i_mapping, index);
2957                         if (!page)
2958                                 continue;
2959                         offset = page_offset(page);
2960
2961                         spin_lock(&dirty_pages->buffer_lock);
2962                         eb = radix_tree_lookup(
2963                              &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
2964                                                offset >> PAGE_CACHE_SHIFT);
2965                         spin_unlock(&dirty_pages->buffer_lock);
2966                         if (eb) {
2967                                 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
2968                                                          &eb->bflags);
2969                                 atomic_set(&eb->refs, 1);
2970                         }
2971                         if (PageWriteback(page))
2972                                 end_page_writeback(page);
2973
2974                         lock_page(page);
2975                         if (PageDirty(page)) {
2976                                 clear_page_dirty_for_io(page);
2977                                 spin_lock_irq(&page->mapping->tree_lock);
2978                                 radix_tree_tag_clear(&page->mapping->page_tree,
2979                                                         page_index(page),
2980                                                         PAGECACHE_TAG_DIRTY);
2981                                 spin_unlock_irq(&page->mapping->tree_lock);
2982                         }
2983
2984                         page->mapping->a_ops->invalidatepage(page, 0);
2985                         unlock_page(page);
2986                 }
2987         }
2988
2989         return ret;
2990 }
2991
2992 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
2993                                        struct extent_io_tree *pinned_extents)
2994 {
2995         struct extent_io_tree *unpin;
2996         u64 start;
2997         u64 end;
2998         int ret;
2999
3000         unpin = pinned_extents;
3001         while (1) {
3002                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3003                                             EXTENT_DIRTY);
3004                 if (ret)
3005                         break;
3006
3007                 /* opt_discard */
3008                 if (btrfs_test_opt(root, DISCARD))
3009                         ret = btrfs_error_discard_extent(root, start,
3010                                                          end + 1 - start,
3011                                                          NULL);
3012
3013                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3014                 btrfs_error_unpin_extent_range(root, start, end);
3015                 cond_resched();
3016         }
3017
3018         return 0;
3019 }
3020
3021 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3022 {
3023         struct btrfs_transaction *t;
3024         LIST_HEAD(list);
3025
3026         WARN_ON(1);
3027
3028         mutex_lock(&root->fs_info->trans_mutex);
3029         mutex_lock(&root->fs_info->transaction_kthread_mutex);
3030
3031         list_splice_init(&root->fs_info->trans_list, &list);
3032         while (!list_empty(&list)) {
3033                 t = list_entry(list.next, struct btrfs_transaction, list);
3034                 if (!t)
3035                         break;
3036
3037                 btrfs_destroy_ordered_operations(root);
3038
3039                 btrfs_destroy_ordered_extents(root);
3040
3041                 btrfs_destroy_delayed_refs(t, root);
3042
3043                 btrfs_block_rsv_release(root,
3044                                         &root->fs_info->trans_block_rsv,
3045                                         t->dirty_pages.dirty_bytes);
3046
3047                 /* FIXME: cleanup wait for commit */
3048                 t->in_commit = 1;
3049                 t->blocked = 1;
3050                 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3051                         wake_up(&root->fs_info->transaction_blocked_wait);
3052
3053                 t->blocked = 0;
3054                 if (waitqueue_active(&root->fs_info->transaction_wait))
3055                         wake_up(&root->fs_info->transaction_wait);
3056                 mutex_unlock(&root->fs_info->trans_mutex);
3057
3058                 mutex_lock(&root->fs_info->trans_mutex);
3059                 t->commit_done = 1;
3060                 if (waitqueue_active(&t->commit_wait))
3061                         wake_up(&t->commit_wait);
3062                 mutex_unlock(&root->fs_info->trans_mutex);
3063
3064                 mutex_lock(&root->fs_info->trans_mutex);
3065
3066                 btrfs_destroy_pending_snapshots(t);
3067
3068                 btrfs_destroy_delalloc_inodes(root);
3069
3070                 spin_lock(&root->fs_info->new_trans_lock);
3071                 root->fs_info->running_transaction = NULL;
3072                 spin_unlock(&root->fs_info->new_trans_lock);
3073
3074                 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3075                                              EXTENT_DIRTY);
3076
3077                 btrfs_destroy_pinned_extent(root,
3078                                             root->fs_info->pinned_extents);
3079
3080                 atomic_set(&t->use_count, 0);
3081                 list_del_init(&t->list);
3082                 memset(t, 0, sizeof(*t));
3083                 kmem_cache_free(btrfs_transaction_cachep, t);
3084         }
3085
3086         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3087         mutex_unlock(&root->fs_info->trans_mutex);
3088
3089         return 0;
3090 }
3091
3092 static struct extent_io_ops btree_extent_io_ops = {
3093         .write_cache_pages_lock_hook = btree_lock_page_hook,
3094         .readpage_end_io_hook = btree_readpage_end_io_hook,
3095         .submit_bio_hook = btree_submit_bio_hook,
3096         /* note we're sharing with inode.c for the merge bio hook */
3097         .merge_bio_hook = btrfs_merge_bio_hook,
3098 };