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