fs/btrfs/disk-io.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/fs.h>
  20 #include <linux/blkdev.h>
  21 #include <linux/scatterlist.h>
  22 #include <linux/swap.h>
  23 #include <linux/radix-tree.h>
  24 #include <linux/writeback.h>
  25 #include <linux/buffer_head.h>
  26 #include <linux/workqueue.h>
  27 #include <linux/kthread.h>
  28 #include <linux/freezer.h>
  29 #include <linux/crc32c.h>
  30 #include <linux/slab.h>
  31 #include <linux/migrate.h>
  32 #include "compat.h"
  33 #include "ctree.h"
  34 #include "disk-io.h"
  35 #include "transaction.h"
  36 #include "btrfs_inode.h"
  37 #include "volumes.h"
  38 #include "print-tree.h"
  39 #include "async-thread.h"
  40 #include "locking.h"
  41 #include "tree-log.h"
  42 #include "free-space-cache.h"
  43
  44 static struct extent_io_ops btree_extent_io_ops;
  45 static void end_workqueue_fn(struct btrfs_work *work);
  46 static void free_fs_root(struct btrfs_root *root);
  47
  48 /*
  49  * end_io_wq structs are used to do processing in task context when an IO is
  50  * complete.  This is used during reads to verify checksums, and it is used
  51  * by writes to insert metadata for new file extents after IO is complete.
  52  */
  53 struct end_io_wq {
  54         struct bio *bio;
  55         bio_end_io_t *end_io;
  56         void *private;
  57         struct btrfs_fs_info *info;
  58         int error;
  59         int metadata;
  60         struct list_head list;
  61         struct btrfs_work work;
  62 };
  63
  64 /*
  65  * async submit bios are used to offload expensive checksumming
  66  * onto the worker threads.  They checksum file and metadata bios
  67  * just before they are sent down the IO stack.
  68  */
  69 struct async_submit_bio {
  70         struct inode *inode;
  71         struct bio *bio;
  72         struct list_head list;
  73         extent_submit_bio_hook_t *submit_bio_start;
  74         extent_submit_bio_hook_t *submit_bio_done;
  75         int rw;
  76         int mirror_num;
  77         unsigned long bio_flags;
  78         /*
  79          * bio_offset is optional, can be used if the pages in the bio
  80          * can't tell us where in the file the bio should go
  81          */
  82         u64 bio_offset;
  83         struct btrfs_work work;
  84 };
  85
  86 /* These are used to set the lockdep class on the extent buffer locks.
  87  * The class is set by the readpage_end_io_hook after the buffer has
  88  * passed csum validation but before the pages are unlocked.
  89  *
  90  * The lockdep class is also set by btrfs_init_new_buffer on freshly
  91  * allocated blocks.
  92  *
  93  * The class is based on the level in the tree block, which allows lockdep
  94  * to know that lower nodes nest inside the locks of higher nodes.
  95  *
  96  * We also add a check to make sure the highest level of the tree is
  97  * the same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this
  98  * code needs update as well.
  99  */
 100 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 101 # if BTRFS_MAX_LEVEL != 8
 102 #  error
 103 # endif
 104 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
 105 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
 106         /* leaf */
 107         "btrfs-extent-00",
 108         "btrfs-extent-01",
 109         "btrfs-extent-02",
 110         "btrfs-extent-03",
 111         "btrfs-extent-04",
 112         "btrfs-extent-05",
 113         "btrfs-extent-06",
 114         "btrfs-extent-07",
 115         /* highest possible level */
 116         "btrfs-extent-08",
 117 };
 118 #endif
 119
 120 /*
 121  * extents on the btree inode are pretty simple, there's one extent
 122  * that covers the entire device
 123  */
 124 static struct extent_map *btree_get_extent(struct inode *inode,
 125                 struct page *page, size_t page_offset, u64 start, u64 len,
 126                 int create)
 127 {
 128         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 129         struct extent_map *em;
 130         int ret;
 131
 132         read_lock(&em_tree->lock);
 133         em = lookup_extent_mapping(em_tree, start, len);
 134         if (em) {
 135                 em->bdev =
 136                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 137                 read_unlock(&em_tree->lock);
 138                 goto out;
 139         }
 140         read_unlock(&em_tree->lock);
 141
 142         em = alloc_extent_map(GFP_NOFS);
 143         if (!em) {
 144                 em = ERR_PTR(-ENOMEM);
 145                 goto out;
 146         }
 147         em->start = 0;
 148         em->len = (u64)-1;
 149         em->block_len = (u64)-1;
 150         em->block_start = 0;
 151         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 152
 153         write_lock(&em_tree->lock);
 154         ret = add_extent_mapping(em_tree, em);
 155         if (ret == -EEXIST) {
 156                 u64 failed_start = em->start;
 157                 u64 failed_len = em->len;
 158
 159                 free_extent_map(em);
 160                 em = lookup_extent_mapping(em_tree, start, len);
 161                 if (em) {
 162                         ret = 0;
 163                 } else {
 164                         em = lookup_extent_mapping(em_tree, failed_start,
 165                                                    failed_len);
 166                         ret = -EIO;
 167                 }
 168         } else if (ret) {
 169                 free_extent_map(em);
 170                 em = NULL;
 171         }
 172         write_unlock(&em_tree->lock);
 173
 174         if (ret)
 175                 em = ERR_PTR(ret);
 176 out:
 177         return em;
 178 }
 179
 180 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 181 {
 182         return crc32c(seed, data, len);
 183 }
 184
 185 void btrfs_csum_final(u32 crc, char *result)
 186 {
 187         *(__le32 *)result = ~cpu_to_le32(crc);
 188 }
 189
 190 /*
 191  * compute the csum for a btree block, and either verify it or write it
 192  * into the csum field of the block.
 193  */
 194 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 195                            int verify)
 196 {
 197         u16 csum_size =
 198                 btrfs_super_csum_size(&root->fs_info->super_copy);
 199         char *result = NULL;
 200         unsigned long len;
 201         unsigned long cur_len;
 202         unsigned long offset = BTRFS_CSUM_SIZE;
 203         char *map_token = NULL;
 204         char *kaddr;
 205         unsigned long map_start;
 206         unsigned long map_len;
 207         int err;
 208         u32 crc = ~(u32)0;
 209         unsigned long inline_result;
 210
 211         len = buf->len - offset;
 212         while (len > 0) {
 213                 err = map_private_extent_buffer(buf, offset, 32,
 214                                         &map_token, &kaddr,
 215                                         &map_start, &map_len, KM_USER0);
 216                 if (err)
 217                         return 1;
 218                 cur_len = min(len, map_len - (offset - map_start));
 219                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 220                                       crc, cur_len);
 221                 len -= cur_len;
 222                 offset += cur_len;
 223                 unmap_extent_buffer(buf, map_token, KM_USER0);
 224         }
 225         if (csum_size > sizeof(inline_result)) {
 226                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
 227                 if (!result)
 228                         return 1;
 229         } else {
 230                 result = (char *)&inline_result;
 231         }
 232
 233         btrfs_csum_final(crc, result);
 234
 235         if (verify) {
 236                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 237                         u32 val;
 238                         u32 found = 0;
 239                         memcpy(&found, result, csum_size);
 240
 241                         read_extent_buffer(buf, &val, 0, csum_size);
 242                         if (printk_ratelimit()) {
 243                                 printk(KERN_INFO "btrfs: %s checksum verify "
 244                                        "failed on %llu wanted %X found %X "
 245                                        "level %d\n",
 246                                        root->fs_info->sb->s_id,
 247                                        (unsigned long long)buf->start, val, found,
 248                                        btrfs_header_level(buf));
 249                         }
 250                         if (result != (char *)&inline_result)
 251                                 kfree(result);
 252                         return 1;
 253                 }
 254         } else {
 255                 write_extent_buffer(buf, result, 0, csum_size);
 256         }
 257         if (result != (char *)&inline_result)
 258                 kfree(result);
 259         return 0;
 260 }
 261
 262 /*
 263  * we can't consider a given block up to date unless the transid of the
 264  * block matches the transid in the parent node's pointer.  This is how we
 265  * detect blocks that either didn't get written at all or got written
 266  * in the wrong place.
 267  */
 268 static int verify_parent_transid(struct extent_io_tree *io_tree,
 269                                  struct extent_buffer *eb, u64 parent_transid)
 270 {
 271         struct extent_state *cached_state = NULL;
 272         int ret;
 273
 274         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 275                 return 0;
 276
 277         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 278                          0, &cached_state, GFP_NOFS);
 279         if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
 280             btrfs_header_generation(eb) == parent_transid) {
 281                 ret = 0;
 282                 goto out;
 283         }
 284         if (printk_ratelimit()) {
 285                 printk("parent transid verify failed on %llu wanted %llu "
 286                        "found %llu\n",
 287                        (unsigned long long)eb->start,
 288                        (unsigned long long)parent_transid,
 289                        (unsigned long long)btrfs_header_generation(eb));
 290         }
 291         ret = 1;
 292         clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
 293 out:
 294         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 295                              &cached_state, GFP_NOFS);
 296         return ret;
 297 }
 298
 299 /*
 300  * helper to read a given tree block, doing retries as required when
 301  * the checksums don't match and we have alternate mirrors to try.
 302  */
 303 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 304                                           struct extent_buffer *eb,
 305                                           u64 start, u64 parent_transid)
 306 {
 307         struct extent_io_tree *io_tree;
 308         int ret;
 309         int num_copies = 0;
 310         int mirror_num = 0;
 311
 312         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 313         while (1) {
 314                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 315                                                btree_get_extent, mirror_num);
 316                 if (!ret &&
 317                     !verify_parent_transid(io_tree, eb, parent_transid))
 318                         return ret;
 319
 320                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 321                                               eb->start, eb->len);
 322                 if (num_copies == 1)
 323                         return ret;
 324
 325                 mirror_num++;
 326                 if (mirror_num > num_copies)
 327                         return ret;
 328         }
 329         return -EIO;
 330 }
 331
 332 /*
 333  * checksum a dirty tree block before IO.  This has extra checks to make sure
 334  * we only fill in the checksum field in the first page of a multi-page block
 335  */
 336
 337 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 338 {
 339         struct extent_io_tree *tree;
 340         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 341         u64 found_start;
 342         unsigned long len;
 343         struct extent_buffer *eb;
 344         int ret;
 345
 346         tree = &BTRFS_I(page->mapping->host)->io_tree;
 347
 348         if (page->private == EXTENT_PAGE_PRIVATE)
 349                 goto out;
 350         if (!page->private)
 351                 goto out;
 352         len = page->private >> 2;
 353         WARN_ON(len == 0);
 354
 355         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 356         if (eb == NULL) {
 357                 WARN_ON(1);
 358                 goto out;
 359         }
 360         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 361                                              btrfs_header_generation(eb));
 362         BUG_ON(ret);
 363         WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
 364
 365         found_start = btrfs_header_bytenr(eb);
 366         if (found_start != start) {
 367                 WARN_ON(1);
 368                 goto err;
 369         }
 370         if (eb->first_page != page) {
 371                 WARN_ON(1);
 372                 goto err;
 373         }
 374         if (!PageUptodate(page)) {
 375                 WARN_ON(1);
 376                 goto err;
 377         }
 378         csum_tree_block(root, eb, 0);
 379 err:
 380         free_extent_buffer(eb);
 381 out:
 382         return 0;
 383 }
 384
 385 static int check_tree_block_fsid(struct btrfs_root *root,
 386                                  struct extent_buffer *eb)
 387 {
 388         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 389         u8 fsid[BTRFS_UUID_SIZE];
 390         int ret = 1;
 391
 392         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
 393                            BTRFS_FSID_SIZE);
 394         while (fs_devices) {
 395                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 396                         ret = 0;
 397                         break;
 398                 }
 399                 fs_devices = fs_devices->seed;
 400         }
 401         return ret;
 402 }
 403
 404 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 405 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
 406 {
 407         lockdep_set_class_and_name(&eb->lock,
 408                            &btrfs_eb_class[level],
 409                            btrfs_eb_name[level]);
 410 }
 411 #endif
 412
 413 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 414                                struct extent_state *state)
 415 {
 416         struct extent_io_tree *tree;
 417         u64 found_start;
 418         int found_level;
 419         unsigned long len;
 420         struct extent_buffer *eb;
 421         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 422         int ret = 0;
 423
 424         tree = &BTRFS_I(page->mapping->host)->io_tree;
 425         if (page->private == EXTENT_PAGE_PRIVATE)
 426                 goto out;
 427         if (!page->private)
 428                 goto out;
 429
 430         len = page->private >> 2;
 431         WARN_ON(len == 0);
 432
 433         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 434         if (eb == NULL) {
 435                 ret = -EIO;
 436                 goto out;
 437         }
 438
 439         found_start = btrfs_header_bytenr(eb);
 440         if (found_start != start) {
 441                 if (printk_ratelimit()) {
 442                         printk(KERN_INFO "btrfs bad tree block start "
 443                                "%llu %llu\n",
 444                                (unsigned long long)found_start,
 445                                (unsigned long long)eb->start);
 446                 }
 447                 ret = -EIO;
 448                 goto err;
 449         }
 450         if (eb->first_page != page) {
 451                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
 452                        eb->first_page->index, page->index);
 453                 WARN_ON(1);
 454                 ret = -EIO;
 455                 goto err;
 456         }
 457         if (check_tree_block_fsid(root, eb)) {
 458                 if (printk_ratelimit()) {
 459                         printk(KERN_INFO "btrfs bad fsid on block %llu\n",
 460                                (unsigned long long)eb->start);
 461                 }
 462                 ret = -EIO;
 463                 goto err;
 464         }
 465         found_level = btrfs_header_level(eb);
 466
 467         btrfs_set_buffer_lockdep_class(eb, found_level);
 468
 469         ret = csum_tree_block(root, eb, 1);
 470         if (ret)
 471                 ret = -EIO;
 472
 473         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 474         end = eb->start + end - 1;
 475 err:
 476         free_extent_buffer(eb);
 477 out:
 478         return ret;
 479 }
 480
 481 static void end_workqueue_bio(struct bio *bio, int err)
 482 {
 483         struct end_io_wq *end_io_wq = bio->bi_private;
 484         struct btrfs_fs_info *fs_info;
 485
 486         fs_info = end_io_wq->info;
 487         end_io_wq->error = err;
 488         end_io_wq->work.func = end_workqueue_fn;
 489         end_io_wq->work.flags = 0;
 490
 491         if (bio->bi_rw & REQ_WRITE) {
 492                 if (end_io_wq->metadata == 1)
 493                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
 494                                            &end_io_wq->work);
 495                 else if (end_io_wq->metadata == 2)
 496                         btrfs_queue_worker(&fs_info->endio_freespace_worker,
 497                                            &end_io_wq->work);
 498                 else
 499                         btrfs_queue_worker(&fs_info->endio_write_workers,
 500                                            &end_io_wq->work);
 501         } else {
 502                 if (end_io_wq->metadata)
 503                         btrfs_queue_worker(&fs_info->endio_meta_workers,
 504                                            &end_io_wq->work);
 505                 else
 506                         btrfs_queue_worker(&fs_info->endio_workers,
 507                                            &end_io_wq->work);
 508         }
 509 }
 510
 511 /*
 512  * For the metadata arg you want
 513  *
 514  * 0 - if data
 515  * 1 - if normal metadta
 516  * 2 - if writing to the free space cache area
 517  */
 518 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 519                         int metadata)
 520 {
 521         struct end_io_wq *end_io_wq;
 522         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 523         if (!end_io_wq)
 524                 return -ENOMEM;
 525
 526         end_io_wq->private = bio->bi_private;
 527         end_io_wq->end_io = bio->bi_end_io;
 528         end_io_wq->info = info;
 529         end_io_wq->error = 0;
 530         end_io_wq->bio = bio;
 531         end_io_wq->metadata = metadata;
 532
 533         bio->bi_private = end_io_wq;
 534         bio->bi_end_io = end_workqueue_bio;
 535         return 0;
 536 }
 537
 538 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 539 {
 540         unsigned long limit = min_t(unsigned long,
 541                                     info->workers.max_workers,
 542                                     info->fs_devices->open_devices);
 543         return 256 * limit;
 544 }
 545
 546 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
 547 {
 548         return atomic_read(&info->nr_async_bios) >
 549                 btrfs_async_submit_limit(info);
 550 }
 551
 552 static void run_one_async_start(struct btrfs_work *work)
 553 {
 554         struct async_submit_bio *async;
 555
 556         async = container_of(work, struct  async_submit_bio, work);
 557         async->submit_bio_start(async->inode, async->rw, async->bio,
 558                                async->mirror_num, async->bio_flags,
 559                                async->bio_offset);
 560 }
 561
 562 static void run_one_async_done(struct btrfs_work *work)
 563 {
 564         struct btrfs_fs_info *fs_info;
 565         struct async_submit_bio *async;
 566         int limit;
 567
 568         async = container_of(work, struct  async_submit_bio, work);
 569         fs_info = BTRFS_I(async->inode)->root->fs_info;
 570
 571         limit = btrfs_async_submit_limit(fs_info);
 572         limit = limit * 2 / 3;
 573
 574         atomic_dec(&fs_info->nr_async_submits);
 575
 576         if (atomic_read(&fs_info->nr_async_submits) < limit &&
 577             waitqueue_active(&fs_info->async_submit_wait))
 578                 wake_up(&fs_info->async_submit_wait);
 579
 580         async->submit_bio_done(async->inode, async->rw, async->bio,
 581                                async->mirror_num, async->bio_flags,
 582                                async->bio_offset);
 583 }
 584
 585 static void run_one_async_free(struct btrfs_work *work)
 586 {
 587         struct async_submit_bio *async;
 588
 589         async = container_of(work, struct  async_submit_bio, work);
 590         kfree(async);
 591 }
 592
 593 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 594                         int rw, struct bio *bio, int mirror_num,
 595                         unsigned long bio_flags,
 596                         u64 bio_offset,
 597                         extent_submit_bio_hook_t *submit_bio_start,
 598                         extent_submit_bio_hook_t *submit_bio_done)
 599 {
 600         struct async_submit_bio *async;
 601
 602         async = kmalloc(sizeof(*async), GFP_NOFS);
 603         if (!async)
 604                 return -ENOMEM;
 605
 606         async->inode = inode;
 607         async->rw = rw;
 608         async->bio = bio;
 609         async->mirror_num = mirror_num;
 610         async->submit_bio_start = submit_bio_start;
 611         async->submit_bio_done = submit_bio_done;
 612
 613         async->work.func = run_one_async_start;
 614         async->work.ordered_func = run_one_async_done;
 615         async->work.ordered_free = run_one_async_free;
 616
 617         async->work.flags = 0;
 618         async->bio_flags = bio_flags;
 619         async->bio_offset = bio_offset;
 620
 621         atomic_inc(&fs_info->nr_async_submits);
 622
 623         if (rw & REQ_SYNC)
 624                 btrfs_set_work_high_prio(&async->work);
 625
 626         btrfs_queue_worker(&fs_info->workers, &async->work);
 627
 628         while (atomic_read(&fs_info->async_submit_draining) &&
 629               atomic_read(&fs_info->nr_async_submits)) {
 630                 wait_event(fs_info->async_submit_wait,
 631                            (atomic_read(&fs_info->nr_async_submits) == 0));
 632         }
 633
 634         return 0;
 635 }
 636
 637 static int btree_csum_one_bio(struct bio *bio)
 638 {
 639         struct bio_vec *bvec = bio->bi_io_vec;
 640         int bio_index = 0;
 641         struct btrfs_root *root;
 642
 643         WARN_ON(bio->bi_vcnt <= 0);
 644         while (bio_index < bio->bi_vcnt) {
 645                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 646                 csum_dirty_buffer(root, bvec->bv_page);
 647                 bio_index++;
 648                 bvec++;
 649         }
 650         return 0;
 651 }
 652
 653 static int __btree_submit_bio_start(struct inode *inode, int rw,
 654                                     struct bio *bio, int mirror_num,
 655                                     unsigned long bio_flags,
 656                                     u64 bio_offset)
 657 {
 658         /*
 659          * when we're called for a write, we're already in the async
 660          * submission context.  Just jump into btrfs_map_bio
 661          */
 662         btree_csum_one_bio(bio);
 663         return 0;
 664 }
 665
 666 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 667                                  int mirror_num, unsigned long bio_flags,
 668                                  u64 bio_offset)
 669 {
 670         /*
 671          * when we're called for a write, we're already in the async
 672          * submission context.  Just jump into btrfs_map_bio
 673          */
 674         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 675 }
 676
 677 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 678                                  int mirror_num, unsigned long bio_flags,
 679                                  u64 bio_offset)
 680 {
 681         int ret;
 682
 683         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 684                                           bio, 1);
 685         BUG_ON(ret);
 686
 687         if (!(rw & REQ_WRITE)) {
 688                 /*
 689                  * called for a read, do the setup so that checksum validation
 690                  * can happen in the async kernel threads
 691                  */
 692                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 693                                      mirror_num, 0);
 694         }
 695
 696         /*
 697          * kthread helpers are used to submit writes so that checksumming
 698          * can happen in parallel across all CPUs
 699          */
 700         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 701                                    inode, rw, bio, mirror_num, 0,
 702                                    bio_offset,
 703                                    __btree_submit_bio_start,
 704                                    __btree_submit_bio_done);
 705 }
 706
 707 #ifdef CONFIG_MIGRATION
 708 static int btree_migratepage(struct address_space *mapping,
 709                         struct page *newpage, struct page *page)
 710 {
 711         /*
 712          * we can't safely write a btree page from here,
 713          * we haven't done the locking hook
 714          */
 715         if (PageDirty(page))
 716                 return -EAGAIN;
 717         /*
 718          * Buffers may be managed in a filesystem specific way.
 719          * We must have no buffers or drop them.
 720          */
 721         if (page_has_private(page) &&
 722             !try_to_release_page(page, GFP_KERNEL))
 723                 return -EAGAIN;
 724         return migrate_page(mapping, newpage, page);
 725 }
 726 #endif
 727
 728 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 729 {
 730         struct extent_io_tree *tree;
 731         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 732         struct extent_buffer *eb;
 733         int was_dirty;
 734
 735         tree = &BTRFS_I(page->mapping->host)->io_tree;
 736         if (!(current->flags & PF_MEMALLOC)) {
 737                 return extent_write_full_page(tree, page,
 738                                               btree_get_extent, wbc);
 739         }
 740
 741         redirty_page_for_writepage(wbc, page);
 742         eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
 743         WARN_ON(!eb);
 744
 745         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
 746         if (!was_dirty) {
 747                 spin_lock(&root->fs_info->delalloc_lock);
 748                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
 749                 spin_unlock(&root->fs_info->delalloc_lock);
 750         }
 751         free_extent_buffer(eb);
 752
 753         unlock_page(page);
 754         return 0;
 755 }
 756
 757 static int btree_writepages(struct address_space *mapping,
 758                             struct writeback_control *wbc)
 759 {
 760         struct extent_io_tree *tree;
 761         tree = &BTRFS_I(mapping->host)->io_tree;
 762         if (wbc->sync_mode == WB_SYNC_NONE) {
 763                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
 764                 u64 num_dirty;
 765                 unsigned long thresh = 32 * 1024 * 1024;
 766
 767                 if (wbc->for_kupdate)
 768                         return 0;
 769
 770                 /* this is a bit racy, but that's ok */
 771                 num_dirty = root->fs_info->dirty_metadata_bytes;
 772                 if (num_dirty < thresh)
 773                         return 0;
 774         }
 775         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 776 }
 777
 778 static int btree_readpage(struct file *file, struct page *page)
 779 {
 780         struct extent_io_tree *tree;
 781         tree = &BTRFS_I(page->mapping->host)->io_tree;
 782         return extent_read_full_page(tree, page, btree_get_extent);
 783 }
 784
 785 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 786 {
 787         struct extent_io_tree *tree;
 788         struct extent_map_tree *map;
 789         int ret;
 790
 791         if (PageWriteback(page) || PageDirty(page))
 792                 return 0;
 793
 794         tree = &BTRFS_I(page->mapping->host)->io_tree;
 795         map = &BTRFS_I(page->mapping->host)->extent_tree;
 796
 797         ret = try_release_extent_state(map, tree, page, gfp_flags);
 798         if (!ret)
 799                 return 0;
 800
 801         ret = try_release_extent_buffer(tree, page);
 802         if (ret == 1) {
 803                 ClearPagePrivate(page);
 804                 set_page_private(page, 0);
 805                 page_cache_release(page);
 806         }
 807
 808         return ret;
 809 }
 810
 811 static void btree_invalidatepage(struct page *page, unsigned long offset)
 812 {
 813         struct extent_io_tree *tree;
 814         tree = &BTRFS_I(page->mapping->host)->io_tree;
 815         extent_invalidatepage(tree, page, offset);
 816         btree_releasepage(page, GFP_NOFS);
 817         if (PagePrivate(page)) {
 818                 printk(KERN_WARNING "btrfs warning page private not zero "
 819                        "on page %llu\n", (unsigned long long)page_offset(page));
 820                 ClearPagePrivate(page);
 821                 set_page_private(page, 0);
 822                 page_cache_release(page);
 823         }
 824 }
 825
 826 static const struct address_space_operations btree_aops = {
 827         .readpage       = btree_readpage,
 828         .writepage      = btree_writepage,
 829         .writepages     = btree_writepages,
 830         .releasepage    = btree_releasepage,
 831         .invalidatepage = btree_invalidatepage,
 832         .sync_page      = block_sync_page,
 833 #ifdef CONFIG_MIGRATION
 834         .migratepage    = btree_migratepage,
 835 #endif
 836 };
 837
 838 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 839                          u64 parent_transid)
 840 {
 841         struct extent_buffer *buf = NULL;
 842         struct inode *btree_inode = root->fs_info->btree_inode;
 843         int ret = 0;
 844
 845         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 846         if (!buf)
 847                 return 0;
 848         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 849                                  buf, 0, 0, btree_get_extent, 0);
 850         free_extent_buffer(buf);
 851         return ret;
 852 }
 853
 854 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 855                                             u64 bytenr, u32 blocksize)
 856 {
 857         struct inode *btree_inode = root->fs_info->btree_inode;
 858         struct extent_buffer *eb;
 859         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 860                                 bytenr, blocksize, GFP_NOFS);
 861         return eb;
 862 }
 863
 864 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 865                                                  u64 bytenr, u32 blocksize)
 866 {
 867         struct inode *btree_inode = root->fs_info->btree_inode;
 868         struct extent_buffer *eb;
 869
 870         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 871                                  bytenr, blocksize, NULL, GFP_NOFS);
 872         return eb;
 873 }
 874
 875
 876 int btrfs_write_tree_block(struct extent_buffer *buf)
 877 {
 878         return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
 879                                         buf->start + buf->len - 1);
 880 }
 881
 882 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 883 {
 884         return filemap_fdatawait_range(buf->first_page->mapping,
 885                                        buf->start, buf->start + buf->len - 1);
 886 }
 887
 888 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 889                                       u32 blocksize, u64 parent_transid)
 890 {
 891         struct extent_buffer *buf = NULL;
 892         int ret;
 893
 894         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 895         if (!buf)
 896                 return NULL;
 897
 898         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 899
 900         if (ret == 0)
 901                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
 902         return buf;
 903
 904 }
 905
 906 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 907                      struct extent_buffer *buf)
 908 {
 909         struct inode *btree_inode = root->fs_info->btree_inode;
 910         if (btrfs_header_generation(buf) ==
 911             root->fs_info->running_transaction->transid) {
 912                 btrfs_assert_tree_locked(buf);
 913
 914                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
 915                         spin_lock(&root->fs_info->delalloc_lock);
 916                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
 917                                 root->fs_info->dirty_metadata_bytes -= buf->len;
 918                         else
 919                                 WARN_ON(1);
 920                         spin_unlock(&root->fs_info->delalloc_lock);
 921                 }
 922
 923                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
 924                 btrfs_set_lock_blocking(buf);
 925                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 926                                           buf);
 927         }
 928         return 0;
 929 }
 930
 931 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 932                         u32 stripesize, struct btrfs_root *root,
 933                         struct btrfs_fs_info *fs_info,
 934                         u64 objectid)
 935 {
 936         root->node = NULL;
 937         root->commit_root = NULL;
 938         root->sectorsize = sectorsize;
 939         root->nodesize = nodesize;
 940         root->leafsize = leafsize;
 941         root->stripesize = stripesize;
 942         root->ref_cows = 0;
 943         root->track_dirty = 0;
 944         root->in_radix = 0;
 945         root->orphan_item_inserted = 0;
 946         root->orphan_cleanup_state = 0;
 947
 948         root->fs_info = fs_info;
 949         root->objectid = objectid;
 950         root->last_trans = 0;
 951         root->highest_objectid = 0;
 952         root->name = NULL;
 953         root->in_sysfs = 0;
 954         root->inode_tree = RB_ROOT;
 955         root->block_rsv = NULL;
 956         root->orphan_block_rsv = NULL;
 957
 958         INIT_LIST_HEAD(&root->dirty_list);
 959         INIT_LIST_HEAD(&root->orphan_list);
 960         INIT_LIST_HEAD(&root->root_list);
 961         spin_lock_init(&root->node_lock);
 962         spin_lock_init(&root->orphan_lock);
 963         spin_lock_init(&root->inode_lock);
 964         spin_lock_init(&root->accounting_lock);
 965         mutex_init(&root->objectid_mutex);
 966         mutex_init(&root->log_mutex);
 967         init_waitqueue_head(&root->log_writer_wait);
 968         init_waitqueue_head(&root->log_commit_wait[0]);
 969         init_waitqueue_head(&root->log_commit_wait[1]);
 970         atomic_set(&root->log_commit[0], 0);
 971         atomic_set(&root->log_commit[1], 0);
 972         atomic_set(&root->log_writers, 0);
 973         root->log_batch = 0;
 974         root->log_transid = 0;
 975         root->last_log_commit = 0;
 976         extent_io_tree_init(&root->dirty_log_pages,
 977                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 978
 979         memset(&root->root_key, 0, sizeof(root->root_key));
 980         memset(&root->root_item, 0, sizeof(root->root_item));
 981         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 982         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 983         root->defrag_trans_start = fs_info->generation;
 984         init_completion(&root->kobj_unregister);
 985         root->defrag_running = 0;
 986         root->root_key.objectid = objectid;
 987         root->anon_super.s_root = NULL;
 988         root->anon_super.s_dev = 0;
 989         INIT_LIST_HEAD(&root->anon_super.s_list);
 990         INIT_LIST_HEAD(&root->anon_super.s_instances);
 991         init_rwsem(&root->anon_super.s_umount);
 992
 993         return 0;
 994 }
 995
 996 static int find_and_setup_root(struct btrfs_root *tree_root,
 997                                struct btrfs_fs_info *fs_info,
 998                                u64 objectid,
 999                                struct btrfs_root *root)
1000 {
1001         int ret;
1002         u32 blocksize;
1003         u64 generation;
1004
1005         __setup_root(tree_root->nodesize, tree_root->leafsize,
1006                      tree_root->sectorsize, tree_root->stripesize,
1007                      root, fs_info, objectid);
1008         ret = btrfs_find_last_root(tree_root, objectid,
1009                                    &root->root_item, &root->root_key);
1010         if (ret > 0)
1011                 return -ENOENT;
1012         BUG_ON(ret);
1013
1014         generation = btrfs_root_generation(&root->root_item);
1015         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1016         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1017                                      blocksize, generation);
1018         if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1019                 free_extent_buffer(root->node);
1020                 return -EIO;
1021         }
1022         root->commit_root = btrfs_root_node(root);
1023         return 0;
1024 }
1025
1026 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1027                                          struct btrfs_fs_info *fs_info)
1028 {
1029         struct btrfs_root *root;
1030         struct btrfs_root *tree_root = fs_info->tree_root;
1031         struct extent_buffer *leaf;
1032
1033         root = kzalloc(sizeof(*root), GFP_NOFS);
1034         if (!root)
1035                 return ERR_PTR(-ENOMEM);
1036
1037         __setup_root(tree_root->nodesize, tree_root->leafsize,
1038                      tree_root->sectorsize, tree_root->stripesize,
1039                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1040
1041         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1042         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1043         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1044         /*
1045          * log trees do not get reference counted because they go away
1046          * before a real commit is actually done.  They do store pointers
1047          * to file data extents, and those reference counts still get
1048          * updated (along with back refs to the log tree).
1049          */
1050         root->ref_cows = 0;
1051
1052         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1053                                       BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1054         if (IS_ERR(leaf)) {
1055                 kfree(root);
1056                 return ERR_CAST(leaf);
1057         }
1058
1059         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1060         btrfs_set_header_bytenr(leaf, leaf->start);
1061         btrfs_set_header_generation(leaf, trans->transid);
1062         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1063         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1064         root->node = leaf;
1065
1066         write_extent_buffer(root->node, root->fs_info->fsid,
1067                             (unsigned long)btrfs_header_fsid(root->node),
1068                             BTRFS_FSID_SIZE);
1069         btrfs_mark_buffer_dirty(root->node);
1070         btrfs_tree_unlock(root->node);
1071         return root;
1072 }
1073
1074 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1075                              struct btrfs_fs_info *fs_info)
1076 {
1077         struct btrfs_root *log_root;
1078
1079         log_root = alloc_log_tree(trans, fs_info);
1080         if (IS_ERR(log_root))
1081                 return PTR_ERR(log_root);
1082         WARN_ON(fs_info->log_root_tree);
1083         fs_info->log_root_tree = log_root;
1084         return 0;
1085 }
1086
1087 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1088                        struct btrfs_root *root)
1089 {
1090         struct btrfs_root *log_root;
1091         struct btrfs_inode_item *inode_item;
1092
1093         log_root = alloc_log_tree(trans, root->fs_info);
1094         if (IS_ERR(log_root))
1095                 return PTR_ERR(log_root);
1096
1097         log_root->last_trans = trans->transid;
1098         log_root->root_key.offset = root->root_key.objectid;
1099
1100         inode_item = &log_root->root_item.inode;
1101         inode_item->generation = cpu_to_le64(1);
1102         inode_item->size = cpu_to_le64(3);
1103         inode_item->nlink = cpu_to_le32(1);
1104         inode_item->nbytes = cpu_to_le64(root->leafsize);
1105         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1106
1107         btrfs_set_root_node(&log_root->root_item, log_root->node);
1108
1109         WARN_ON(root->log_root);
1110         root->log_root = log_root;
1111         root->log_transid = 0;
1112         root->last_log_commit = 0;
1113         return 0;
1114 }
1115
1116 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1117                                                struct btrfs_key *location)
1118 {
1119         struct btrfs_root *root;
1120         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1121         struct btrfs_path *path;
1122         struct extent_buffer *l;
1123         u64 generation;
1124         u32 blocksize;
1125         int ret = 0;
1126
1127         root = kzalloc(sizeof(*root), GFP_NOFS);
1128         if (!root)
1129                 return ERR_PTR(-ENOMEM);
1130         if (location->offset == (u64)-1) {
1131                 ret = find_and_setup_root(tree_root, fs_info,
1132                                           location->objectid, root);
1133                 if (ret) {
1134                         kfree(root);
1135                         return ERR_PTR(ret);
1136                 }
1137                 goto out;
1138         }
1139
1140         __setup_root(tree_root->nodesize, tree_root->leafsize,
1141                      tree_root->sectorsize, tree_root->stripesize,
1142                      root, fs_info, location->objectid);
1143
1144         path = btrfs_alloc_path();
1145         BUG_ON(!path);
1146         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1147         if (ret == 0) {
1148                 l = path->nodes[0];
1149                 read_extent_buffer(l, &root->root_item,
1150                                 btrfs_item_ptr_offset(l, path->slots[0]),
1151                                 sizeof(root->root_item));
1152                 memcpy(&root->root_key, location, sizeof(*location));
1153         }
1154         btrfs_free_path(path);
1155         if (ret) {
1156                 kfree(root);
1157                 if (ret > 0)
1158                         ret = -ENOENT;
1159                 return ERR_PTR(ret);
1160         }
1161
1162         generation = btrfs_root_generation(&root->root_item);
1163         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1164         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1165                                      blocksize, generation);
1166         root->commit_root = btrfs_root_node(root);
1167         BUG_ON(!root->node);
1168 out:
1169         if (location->objectid != BTRFS_TREE_LOG_OBJECTID)
1170                 root->ref_cows = 1;
1171
1172         return root;
1173 }
1174
1175 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1176                                         u64 root_objectid)
1177 {
1178         struct btrfs_root *root;
1179
1180         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1181                 return fs_info->tree_root;
1182         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1183                 return fs_info->extent_root;
1184
1185         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1186                                  (unsigned long)root_objectid);
1187         return root;
1188 }
1189
1190 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1191                                               struct btrfs_key *location)
1192 {
1193         struct btrfs_root *root;
1194         int ret;
1195
1196         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1197                 return fs_info->tree_root;
1198         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1199                 return fs_info->extent_root;
1200         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1201                 return fs_info->chunk_root;
1202         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1203                 return fs_info->dev_root;
1204         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1205                 return fs_info->csum_root;
1206 again:
1207         spin_lock(&fs_info->fs_roots_radix_lock);
1208         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1209                                  (unsigned long)location->objectid);
1210         spin_unlock(&fs_info->fs_roots_radix_lock);
1211         if (root)
1212                 return root;
1213
1214         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1215         if (IS_ERR(root))
1216                 return root;
1217
1218         set_anon_super(&root->anon_super, NULL);
1219
1220         if (btrfs_root_refs(&root->root_item) == 0) {
1221                 ret = -ENOENT;
1222                 goto fail;
1223         }
1224
1225         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1226         if (ret < 0)
1227                 goto fail;
1228         if (ret == 0)
1229                 root->orphan_item_inserted = 1;
1230
1231         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1232         if (ret)
1233                 goto fail;
1234
1235         spin_lock(&fs_info->fs_roots_radix_lock);
1236         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1237                                 (unsigned long)root->root_key.objectid,
1238                                 root);
1239         if (ret == 0)
1240                 root->in_radix = 1;
1241
1242         spin_unlock(&fs_info->fs_roots_radix_lock);
1243         radix_tree_preload_end();
1244         if (ret) {
1245                 if (ret == -EEXIST) {
1246                         free_fs_root(root);
1247                         goto again;
1248                 }
1249                 goto fail;
1250         }
1251
1252         ret = btrfs_find_dead_roots(fs_info->tree_root,
1253                                     root->root_key.objectid);
1254         WARN_ON(ret);
1255         return root;
1256 fail:
1257         free_fs_root(root);
1258         return ERR_PTR(ret);
1259 }
1260
1261 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1262                                       struct btrfs_key *location,
1263                                       const char *name, int namelen)
1264 {
1265         return btrfs_read_fs_root_no_name(fs_info, location);
1266 #if 0
1267         struct btrfs_root *root;
1268         int ret;
1269
1270         root = btrfs_read_fs_root_no_name(fs_info, location);
1271         if (!root)
1272                 return NULL;
1273
1274         if (root->in_sysfs)
1275                 return root;
1276
1277         ret = btrfs_set_root_name(root, name, namelen);
1278         if (ret) {
1279                 free_extent_buffer(root->node);
1280                 kfree(root);
1281                 return ERR_PTR(ret);
1282         }
1283
1284         ret = btrfs_sysfs_add_root(root);
1285         if (ret) {
1286                 free_extent_buffer(root->node);
1287                 kfree(root->name);
1288                 kfree(root);
1289                 return ERR_PTR(ret);
1290         }
1291         root->in_sysfs = 1;
1292         return root;
1293 #endif
1294 }
1295
1296 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1297 {
1298         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1299         int ret = 0;
1300         struct btrfs_device *device;
1301         struct backing_dev_info *bdi;
1302
1303         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1304                 if (!device->bdev)
1305                         continue;
1306                 bdi = blk_get_backing_dev_info(device->bdev);
1307                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1308                         ret = 1;
1309                         break;
1310                 }
1311         }
1312         return ret;
1313 }
1314
1315 /*
1316  * this unplugs every device on the box, and it is only used when page
1317  * is null
1318  */
1319 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1320 {
1321         struct btrfs_device *device;
1322         struct btrfs_fs_info *info;
1323
1324         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1325         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1326                 if (!device->bdev)
1327                         continue;
1328
1329                 bdi = blk_get_backing_dev_info(device->bdev);
1330                 if (bdi->unplug_io_fn)
1331                         bdi->unplug_io_fn(bdi, page);
1332         }
1333 }
1334
1335 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1336 {
1337         struct inode *inode;
1338         struct extent_map_tree *em_tree;
1339         struct extent_map *em;
1340         struct address_space *mapping;
1341         u64 offset;
1342
1343         /* the generic O_DIRECT read code does this */
1344         if (1 || !page) {
1345                 __unplug_io_fn(bdi, page);
1346                 return;
1347         }
1348
1349         /*
1350          * page->mapping may change at any time.  Get a consistent copy
1351          * and use that for everything below
1352          */
1353         smp_mb();
1354         mapping = page->mapping;
1355         if (!mapping)
1356                 return;
1357
1358         inode = mapping->host;
1359
1360         /*
1361          * don't do the expensive searching for a small number of
1362          * devices
1363          */
1364         if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1365                 __unplug_io_fn(bdi, page);
1366                 return;
1367         }
1368
1369         offset = page_offset(page);
1370
1371         em_tree = &BTRFS_I(inode)->extent_tree;
1372         read_lock(&em_tree->lock);
1373         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1374         read_unlock(&em_tree->lock);
1375         if (!em) {
1376                 __unplug_io_fn(bdi, page);
1377                 return;
1378         }
1379
1380         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1381                 free_extent_map(em);
1382                 __unplug_io_fn(bdi, page);
1383                 return;
1384         }
1385         offset = offset - em->start;
1386         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1387                           em->block_start + offset, page);
1388         free_extent_map(em);
1389 }
1390
1391 /*
1392  * If this fails, caller must call bdi_destroy() to get rid of the
1393  * bdi again.
1394  */
1395 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1396 {
1397         int err;
1398
1399         bdi->capabilities = BDI_CAP_MAP_COPY;
1400         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1401         if (err)
1402                 return err;
1403
1404         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1405         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1406         bdi->unplug_io_data     = info;
1407         bdi->congested_fn       = btrfs_congested_fn;
1408         bdi->congested_data     = info;
1409         return 0;
1410 }
1411
1412 static int bio_ready_for_csum(struct bio *bio)
1413 {
1414         u64 length = 0;
1415         u64 buf_len = 0;
1416         u64 start = 0;
1417         struct page *page;
1418         struct extent_io_tree *io_tree = NULL;
1419         struct bio_vec *bvec;
1420         int i;
1421         int ret;
1422
1423         bio_for_each_segment(bvec, bio, i) {
1424                 page = bvec->bv_page;
1425                 if (page->private == EXTENT_PAGE_PRIVATE) {
1426                         length += bvec->bv_len;
1427                         continue;
1428                 }
1429                 if (!page->private) {
1430                         length += bvec->bv_len;
1431                         continue;
1432                 }
1433                 length = bvec->bv_len;
1434                 buf_len = page->private >> 2;
1435                 start = page_offset(page) + bvec->bv_offset;
1436                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1437         }
1438         /* are we fully contained in this bio? */
1439         if (buf_len <= length)
1440                 return 1;
1441
1442         ret = extent_range_uptodate(io_tree, start + length,
1443                                     start + buf_len - 1);
1444         return ret;
1445 }
1446
1447 /*
1448  * called by the kthread helper functions to finally call the bio end_io
1449  * functions.  This is where read checksum verification actually happens
1450  */
1451 static void end_workqueue_fn(struct btrfs_work *work)
1452 {
1453         struct bio *bio;
1454         struct end_io_wq *end_io_wq;
1455         struct btrfs_fs_info *fs_info;
1456         int error;
1457
1458         end_io_wq = container_of(work, struct end_io_wq, work);
1459         bio = end_io_wq->bio;
1460         fs_info = end_io_wq->info;
1461
1462         /* metadata bio reads are special because the whole tree block must
1463          * be checksummed at once.  This makes sure the entire block is in
1464          * ram and up to date before trying to verify things.  For
1465          * blocksize <= pagesize, it is basically a noop
1466          */
1467         if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1468             !bio_ready_for_csum(bio)) {
1469                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1470                                    &end_io_wq->work);
1471                 return;
1472         }
1473         error = end_io_wq->error;
1474         bio->bi_private = end_io_wq->private;
1475         bio->bi_end_io = end_io_wq->end_io;
1476         kfree(end_io_wq);
1477         bio_endio(bio, error);
1478 }
1479
1480 static int cleaner_kthread(void *arg)
1481 {
1482         struct btrfs_root *root = arg;
1483
1484         do {
1485                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1486
1487                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1488                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1489                         btrfs_run_delayed_iputs(root);
1490                         btrfs_clean_old_snapshots(root);
1491                         mutex_unlock(&root->fs_info->cleaner_mutex);
1492                 }
1493
1494                 if (freezing(current)) {
1495                         refrigerator();
1496                 } else {
1497                         set_current_state(TASK_INTERRUPTIBLE);
1498                         if (!kthread_should_stop())
1499                                 schedule();
1500                         __set_current_state(TASK_RUNNING);
1501                 }
1502         } while (!kthread_should_stop());
1503         return 0;
1504 }
1505
1506 static int transaction_kthread(void *arg)
1507 {
1508         struct btrfs_root *root = arg;
1509         struct btrfs_trans_handle *trans;
1510         struct btrfs_transaction *cur;
1511         u64 transid;
1512         unsigned long now;
1513         unsigned long delay;
1514         int ret;
1515
1516         do {
1517                 delay = HZ * 30;
1518                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1519                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1520
1521                 spin_lock(&root->fs_info->new_trans_lock);
1522                 cur = root->fs_info->running_transaction;
1523                 if (!cur) {
1524                         spin_unlock(&root->fs_info->new_trans_lock);
1525                         goto sleep;
1526                 }
1527
1528                 now = get_seconds();
1529                 if (!cur->blocked &&
1530                     (now < cur->start_time || now - cur->start_time < 30)) {
1531                         spin_unlock(&root->fs_info->new_trans_lock);
1532                         delay = HZ * 5;
1533                         goto sleep;
1534                 }
1535                 transid = cur->transid;
1536                 spin_unlock(&root->fs_info->new_trans_lock);
1537
1538                 trans = btrfs_join_transaction(root, 1);
1539                 if (transid == trans->transid) {
1540                         ret = btrfs_commit_transaction(trans, root);
1541                         BUG_ON(ret);
1542                 } else {
1543                         btrfs_end_transaction(trans, root);
1544                 }
1545 sleep:
1546                 wake_up_process(root->fs_info->cleaner_kthread);
1547                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1548
1549                 if (freezing(current)) {
1550                         refrigerator();
1551                 } else {
1552                         set_current_state(TASK_INTERRUPTIBLE);
1553                         if (!kthread_should_stop() &&
1554                             !btrfs_transaction_blocked(root->fs_info))
1555                                 schedule_timeout(delay);
1556                         __set_current_state(TASK_RUNNING);
1557                 }
1558         } while (!kthread_should_stop());
1559         return 0;
1560 }
1561
1562 struct btrfs_root *open_ctree(struct super_block *sb,
1563                               struct btrfs_fs_devices *fs_devices,
1564                               char *options)
1565 {
1566         u32 sectorsize;
1567         u32 nodesize;
1568         u32 leafsize;
1569         u32 blocksize;
1570         u32 stripesize;
1571         u64 generation;
1572         u64 features;
1573         struct btrfs_key location;
1574         struct buffer_head *bh;
1575         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1576                                                  GFP_NOFS);
1577         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1578                                                  GFP_NOFS);
1579         struct btrfs_root *tree_root = btrfs_sb(sb);
1580         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1581         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1582                                                 GFP_NOFS);
1583         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1584                                               GFP_NOFS);
1585         struct btrfs_root *log_tree_root;
1586
1587         int ret;
1588         int err = -EINVAL;
1589
1590         struct btrfs_super_block *disk_super;
1591
1592         if (!extent_root || !tree_root || !fs_info ||
1593             !chunk_root || !dev_root || !csum_root) {
1594                 err = -ENOMEM;
1595                 goto fail;
1596         }
1597
1598         ret = init_srcu_struct(&fs_info->subvol_srcu);
1599         if (ret) {
1600                 err = ret;
1601                 goto fail;
1602         }
1603
1604         ret = setup_bdi(fs_info, &fs_info->bdi);
1605         if (ret) {
1606                 err = ret;
1607                 goto fail_srcu;
1608         }
1609
1610         fs_info->btree_inode = new_inode(sb);
1611         if (!fs_info->btree_inode) {
1612                 err = -ENOMEM;
1613                 goto fail_bdi;
1614         }
1615
1616         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1617         INIT_LIST_HEAD(&fs_info->trans_list);
1618         INIT_LIST_HEAD(&fs_info->dead_roots);
1619         INIT_LIST_HEAD(&fs_info->delayed_iputs);
1620         INIT_LIST_HEAD(&fs_info->hashers);
1621         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1622         INIT_LIST_HEAD(&fs_info->ordered_operations);
1623         INIT_LIST_HEAD(&fs_info->caching_block_groups);
1624         spin_lock_init(&fs_info->delalloc_lock);
1625         spin_lock_init(&fs_info->new_trans_lock);
1626         spin_lock_init(&fs_info->ref_cache_lock);
1627         spin_lock_init(&fs_info->fs_roots_radix_lock);
1628         spin_lock_init(&fs_info->delayed_iput_lock);
1629
1630         init_completion(&fs_info->kobj_unregister);
1631         fs_info->tree_root = tree_root;
1632         fs_info->extent_root = extent_root;
1633         fs_info->csum_root = csum_root;
1634         fs_info->chunk_root = chunk_root;
1635         fs_info->dev_root = dev_root;
1636         fs_info->fs_devices = fs_devices;
1637         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1638         INIT_LIST_HEAD(&fs_info->space_info);
1639         btrfs_mapping_init(&fs_info->mapping_tree);
1640         btrfs_init_block_rsv(&fs_info->global_block_rsv);
1641         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1642         btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1643         btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1644         btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1645         INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1646         mutex_init(&fs_info->durable_block_rsv_mutex);
1647         atomic_set(&fs_info->nr_async_submits, 0);
1648         atomic_set(&fs_info->async_delalloc_pages, 0);
1649         atomic_set(&fs_info->async_submit_draining, 0);
1650         atomic_set(&fs_info->nr_async_bios, 0);
1651         fs_info->sb = sb;
1652         fs_info->max_inline = 8192 * 1024;
1653         fs_info->metadata_ratio = 0;
1654
1655         fs_info->thread_pool_size = min_t(unsigned long,
1656                                           num_online_cpus() + 2, 8);
1657
1658         INIT_LIST_HEAD(&fs_info->ordered_extents);
1659         spin_lock_init(&fs_info->ordered_extent_lock);
1660
1661         sb->s_blocksize = 4096;
1662         sb->s_blocksize_bits = blksize_bits(4096);
1663         sb->s_bdi = &fs_info->bdi;
1664
1665         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1666         fs_info->btree_inode->i_nlink = 1;
1667         /*
1668          * we set the i_size on the btree inode to the max possible int.
1669          * the real end of the address space is determined by all of
1670          * the devices in the system
1671          */
1672         fs_info->btree_inode->i_size = OFFSET_MAX;
1673         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1674         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1675
1676         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1677         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1678                              fs_info->btree_inode->i_mapping,
1679                              GFP_NOFS);
1680         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1681                              GFP_NOFS);
1682
1683         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1684
1685         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1686         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1687                sizeof(struct btrfs_key));
1688         BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1689         insert_inode_hash(fs_info->btree_inode);
1690
1691         spin_lock_init(&fs_info->block_group_cache_lock);
1692         fs_info->block_group_cache_tree = RB_ROOT;
1693
1694         extent_io_tree_init(&fs_info->freed_extents[0],
1695                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1696         extent_io_tree_init(&fs_info->freed_extents[1],
1697                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1698         fs_info->pinned_extents = &fs_info->freed_extents[0];
1699         fs_info->do_barriers = 1;
1700
1701
1702         mutex_init(&fs_info->trans_mutex);
1703         mutex_init(&fs_info->ordered_operations_mutex);
1704         mutex_init(&fs_info->tree_log_mutex);
1705         mutex_init(&fs_info->chunk_mutex);
1706         mutex_init(&fs_info->transaction_kthread_mutex);
1707         mutex_init(&fs_info->cleaner_mutex);
1708         mutex_init(&fs_info->volume_mutex);
1709         init_rwsem(&fs_info->extent_commit_sem);
1710         init_rwsem(&fs_info->cleanup_work_sem);
1711         init_rwsem(&fs_info->subvol_sem);
1712
1713         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1714         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1715
1716         init_waitqueue_head(&fs_info->transaction_throttle);
1717         init_waitqueue_head(&fs_info->transaction_wait);
1718         init_waitqueue_head(&fs_info->transaction_blocked_wait);
1719         init_waitqueue_head(&fs_info->async_submit_wait);
1720
1721         __setup_root(4096, 4096, 4096, 4096, tree_root,
1722                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1723
1724         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1725         if (!bh) {
1726                 err = -EINVAL;
1727                 goto fail_iput;
1728         }
1729
1730         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1731         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1732                sizeof(fs_info->super_for_commit));
1733         brelse(bh);
1734
1735         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1736
1737         disk_super = &fs_info->super_copy;
1738         if (!btrfs_super_root(disk_super))
1739                 goto fail_iput;
1740
1741         ret = btrfs_parse_options(tree_root, options);
1742         if (ret) {
1743                 err = ret;
1744                 goto fail_iput;
1745         }
1746
1747         features = btrfs_super_incompat_flags(disk_super) &
1748                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1749         if (features) {
1750                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1751                        "unsupported optional features (%Lx).\n",
1752                        (unsigned long long)features);
1753                 err = -EINVAL;
1754                 goto fail_iput;
1755         }
1756
1757         features = btrfs_super_incompat_flags(disk_super);
1758         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1759         if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1760                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1761         btrfs_set_super_incompat_flags(disk_super, features);
1762
1763         features = btrfs_super_compat_ro_flags(disk_super) &
1764                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1765         if (!(sb->s_flags & MS_RDONLY) && features) {
1766                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1767                        "unsupported option features (%Lx).\n",
1768                        (unsigned long long)features);
1769                 err = -EINVAL;
1770                 goto fail_iput;
1771         }
1772
1773         btrfs_init_workers(&fs_info->generic_worker,
1774                            "genwork", 1, NULL);
1775
1776         btrfs_init_workers(&fs_info->workers, "worker",
1777                            fs_info->thread_pool_size,
1778                            &fs_info->generic_worker);
1779
1780         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1781                            fs_info->thread_pool_size,
1782                            &fs_info->generic_worker);
1783
1784         btrfs_init_workers(&fs_info->submit_workers, "submit",
1785                            min_t(u64, fs_devices->num_devices,
1786                            fs_info->thread_pool_size),
1787                            &fs_info->generic_worker);
1788
1789         /* a higher idle thresh on the submit workers makes it much more
1790          * likely that bios will be send down in a sane order to the
1791          * devices
1792          */
1793         fs_info->submit_workers.idle_thresh = 64;
1794
1795         fs_info->workers.idle_thresh = 16;
1796         fs_info->workers.ordered = 1;
1797
1798         fs_info->delalloc_workers.idle_thresh = 2;
1799         fs_info->delalloc_workers.ordered = 1;
1800
1801         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1802                            &fs_info->generic_worker);
1803         btrfs_init_workers(&fs_info->endio_workers, "endio",
1804                            fs_info->thread_pool_size,
1805                            &fs_info->generic_worker);
1806         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1807                            fs_info->thread_pool_size,
1808                            &fs_info->generic_worker);
1809         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1810                            "endio-meta-write", fs_info->thread_pool_size,
1811                            &fs_info->generic_worker);
1812         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1813                            fs_info->thread_pool_size,
1814                            &fs_info->generic_worker);
1815         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1816                            1, &fs_info->generic_worker);
1817
1818         /*
1819          * endios are largely parallel and should have a very
1820          * low idle thresh
1821          */
1822         fs_info->endio_workers.idle_thresh = 4;
1823         fs_info->endio_meta_workers.idle_thresh = 4;
1824
1825         fs_info->endio_write_workers.idle_thresh = 2;
1826         fs_info->endio_meta_write_workers.idle_thresh = 2;
1827
1828         btrfs_start_workers(&fs_info->workers, 1);
1829         btrfs_start_workers(&fs_info->generic_worker, 1);
1830         btrfs_start_workers(&fs_info->submit_workers, 1);
1831         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1832         btrfs_start_workers(&fs_info->fixup_workers, 1);
1833         btrfs_start_workers(&fs_info->endio_workers, 1);
1834         btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1835         btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1836         btrfs_start_workers(&fs_info->endio_write_workers, 1);
1837         btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1838
1839         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1840         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1841                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1842
1843         nodesize = btrfs_super_nodesize(disk_super);
1844         leafsize = btrfs_super_leafsize(disk_super);
1845         sectorsize = btrfs_super_sectorsize(disk_super);
1846         stripesize = btrfs_super_stripesize(disk_super);
1847         tree_root->nodesize = nodesize;
1848         tree_root->leafsize = leafsize;
1849         tree_root->sectorsize = sectorsize;
1850         tree_root->stripesize = stripesize;
1851
1852         sb->s_blocksize = sectorsize;
1853         sb->s_blocksize_bits = blksize_bits(sectorsize);
1854
1855         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1856                     sizeof(disk_super->magic))) {
1857                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1858                 goto fail_sb_buffer;
1859         }
1860
1861         mutex_lock(&fs_info->chunk_mutex);
1862         ret = btrfs_read_sys_array(tree_root);
1863         mutex_unlock(&fs_info->chunk_mutex);
1864         if (ret) {
1865                 printk(KERN_WARNING "btrfs: failed to read the system "
1866                        "array on %s\n", sb->s_id);
1867                 goto fail_sb_buffer;
1868         }
1869
1870         blocksize = btrfs_level_size(tree_root,
1871                                      btrfs_super_chunk_root_level(disk_super));
1872         generation = btrfs_super_chunk_root_generation(disk_super);
1873
1874         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1875                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1876
1877         chunk_root->node = read_tree_block(chunk_root,
1878                                            btrfs_super_chunk_root(disk_super),
1879                                            blocksize, generation);
1880         BUG_ON(!chunk_root->node);
1881         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1882                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1883                        sb->s_id);
1884                 goto fail_chunk_root;
1885         }
1886         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1887         chunk_root->commit_root = btrfs_root_node(chunk_root);
1888
1889         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1890            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1891            BTRFS_UUID_SIZE);
1892
1893         mutex_lock(&fs_info->chunk_mutex);
1894         ret = btrfs_read_chunk_tree(chunk_root);
1895         mutex_unlock(&fs_info->chunk_mutex);
1896         if (ret) {
1897                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1898                        sb->s_id);
1899                 goto fail_chunk_root;
1900         }
1901
1902         btrfs_close_extra_devices(fs_devices);
1903
1904         blocksize = btrfs_level_size(tree_root,
1905                                      btrfs_super_root_level(disk_super));
1906         generation = btrfs_super_generation(disk_super);
1907
1908         tree_root->node = read_tree_block(tree_root,
1909                                           btrfs_super_root(disk_super),
1910                                           blocksize, generation);
1911         if (!tree_root->node)
1912                 goto fail_chunk_root;
1913         if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1914                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1915                        sb->s_id);
1916                 goto fail_tree_root;
1917         }
1918         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1919         tree_root->commit_root = btrfs_root_node(tree_root);
1920
1921         ret = find_and_setup_root(tree_root, fs_info,
1922                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1923         if (ret)
1924                 goto fail_tree_root;
1925         extent_root->track_dirty = 1;
1926
1927         ret = find_and_setup_root(tree_root, fs_info,
1928                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1929         if (ret)
1930                 goto fail_extent_root;
1931         dev_root->track_dirty = 1;
1932
1933         ret = find_and_setup_root(tree_root, fs_info,
1934                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
1935         if (ret)
1936                 goto fail_dev_root;
1937
1938         csum_root->track_dirty = 1;
1939
1940         fs_info->generation = generation;
1941         fs_info->last_trans_committed = generation;
1942         fs_info->data_alloc_profile = (u64)-1;
1943         fs_info->metadata_alloc_profile = (u64)-1;
1944         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1945
1946         ret = btrfs_read_block_groups(extent_root);
1947         if (ret) {
1948                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
1949                 goto fail_block_groups;
1950         }
1951
1952         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1953                                                "btrfs-cleaner");
1954         if (IS_ERR(fs_info->cleaner_kthread))
1955                 goto fail_block_groups;
1956
1957         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1958                                                    tree_root,
1959                                                    "btrfs-transaction");
1960         if (IS_ERR(fs_info->transaction_kthread))
1961                 goto fail_cleaner;
1962
1963         if (!btrfs_test_opt(tree_root, SSD) &&
1964             !btrfs_test_opt(tree_root, NOSSD) &&
1965             !fs_info->fs_devices->rotating) {
1966                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
1967                        "mode\n");
1968                 btrfs_set_opt(fs_info->mount_opt, SSD);
1969         }
1970
1971         if (btrfs_super_log_root(disk_super) != 0) {
1972                 u64 bytenr = btrfs_super_log_root(disk_super);
1973
1974                 if (fs_devices->rw_devices == 0) {
1975                         printk(KERN_WARNING "Btrfs log replay required "
1976                                "on RO media\n");
1977                         err = -EIO;
1978                         goto fail_trans_kthread;
1979                 }
1980                 blocksize =
1981                      btrfs_level_size(tree_root,
1982                                       btrfs_super_log_root_level(disk_super));
1983
1984                 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1985                 if (!log_tree_root) {
1986                         err = -ENOMEM;
1987                         goto fail_trans_kthread;
1988                 }
1989
1990                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1991                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1992
1993                 log_tree_root->node = read_tree_block(tree_root, bytenr,
1994                                                       blocksize,
1995                                                       generation + 1);
1996                 ret = btrfs_recover_log_trees(log_tree_root);
1997                 BUG_ON(ret);
1998
1999                 if (sb->s_flags & MS_RDONLY) {
2000                         ret =  btrfs_commit_super(tree_root);
2001                         BUG_ON(ret);
2002                 }
2003         }
2004
2005         ret = btrfs_find_orphan_roots(tree_root);
2006         BUG_ON(ret);
2007
2008         if (!(sb->s_flags & MS_RDONLY)) {
2009                 ret = btrfs_cleanup_fs_roots(fs_info);
2010                 BUG_ON(ret);
2011
2012                 ret = btrfs_recover_relocation(tree_root);
2013                 if (ret < 0) {
2014                         printk(KERN_WARNING
2015                                "btrfs: failed to recover relocation\n");
2016                         err = -EINVAL;
2017                         goto fail_trans_kthread;
2018                 }
2019         }
2020
2021         location.objectid = BTRFS_FS_TREE_OBJECTID;
2022         location.type = BTRFS_ROOT_ITEM_KEY;
2023         location.offset = (u64)-1;
2024
2025         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2026         if (!fs_info->fs_root)
2027                 goto fail_trans_kthread;
2028         if (IS_ERR(fs_info->fs_root)) {
2029                 err = PTR_ERR(fs_info->fs_root);
2030                 goto fail_trans_kthread;
2031         }
2032
2033         if (!(sb->s_flags & MS_RDONLY)) {
2034                 down_read(&fs_info->cleanup_work_sem);
2035                 btrfs_orphan_cleanup(fs_info->fs_root);
2036                 btrfs_orphan_cleanup(fs_info->tree_root);
2037                 up_read(&fs_info->cleanup_work_sem);
2038         }
2039
2040         return tree_root;
2041
2042 fail_trans_kthread:
2043         kthread_stop(fs_info->transaction_kthread);
2044 fail_cleaner:
2045         kthread_stop(fs_info->cleaner_kthread);
2046
2047         /*
2048          * make sure we're done with the btree inode before we stop our
2049          * kthreads
2050          */
2051         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2052         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2053
2054 fail_block_groups:
2055         btrfs_free_block_groups(fs_info);
2056         free_extent_buffer(csum_root->node);
2057         free_extent_buffer(csum_root->commit_root);
2058 fail_dev_root:
2059         free_extent_buffer(dev_root->node);
2060         free_extent_buffer(dev_root->commit_root);
2061 fail_extent_root:
2062         free_extent_buffer(extent_root->node);
2063         free_extent_buffer(extent_root->commit_root);
2064 fail_tree_root:
2065         free_extent_buffer(tree_root->node);
2066         free_extent_buffer(tree_root->commit_root);
2067 fail_chunk_root:
2068         free_extent_buffer(chunk_root->node);
2069         free_extent_buffer(chunk_root->commit_root);
2070 fail_sb_buffer:
2071         btrfs_stop_workers(&fs_info->generic_worker);
2072         btrfs_stop_workers(&fs_info->fixup_workers);
2073         btrfs_stop_workers(&fs_info->delalloc_workers);
2074         btrfs_stop_workers(&fs_info->workers);
2075         btrfs_stop_workers(&fs_info->endio_workers);
2076         btrfs_stop_workers(&fs_info->endio_meta_workers);
2077         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2078         btrfs_stop_workers(&fs_info->endio_write_workers);
2079         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2080         btrfs_stop_workers(&fs_info->submit_workers);
2081 fail_iput:
2082         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2083         iput(fs_info->btree_inode);
2084
2085         btrfs_close_devices(fs_info->fs_devices);
2086         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2087 fail_bdi:
2088         bdi_destroy(&fs_info->bdi);
2089 fail_srcu:
2090         cleanup_srcu_struct(&fs_info->subvol_srcu);
2091 fail:
2092         kfree(extent_root);
2093         kfree(tree_root);
2094         kfree(fs_info);
2095         kfree(chunk_root);
2096         kfree(dev_root);
2097         kfree(csum_root);
2098         return ERR_PTR(err);
2099 }
2100
2101 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2102 {
2103         char b[BDEVNAME_SIZE];
2104
2105         if (uptodate) {
2106                 set_buffer_uptodate(bh);
2107         } else {
2108                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
2109                         printk(KERN_WARNING "lost page write due to "
2110                                         "I/O error on %s\n",
2111                                        bdevname(bh->b_bdev, b));
2112                 }
2113                 /* note, we dont' set_buffer_write_io_error because we have
2114                  * our own ways of dealing with the IO errors
2115                  */
2116                 clear_buffer_uptodate(bh);
2117         }
2118         unlock_buffer(bh);
2119         put_bh(bh);
2120 }
2121
2122 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2123 {
2124         struct buffer_head *bh;
2125         struct buffer_head *latest = NULL;
2126         struct btrfs_super_block *super;
2127         int i;
2128         u64 transid = 0;
2129         u64 bytenr;
2130
2131         /* we would like to check all the supers, but that would make
2132          * a btrfs mount succeed after a mkfs from a different FS.
2133          * So, we need to add a special mount option to scan for
2134          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2135          */
2136         for (i = 0; i < 1; i++) {
2137                 bytenr = btrfs_sb_offset(i);
2138                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2139                         break;
2140                 bh = __bread(bdev, bytenr / 4096, 4096);
2141                 if (!bh)
2142                         continue;
2143
2144                 super = (struct btrfs_super_block *)bh->b_data;
2145                 if (btrfs_super_bytenr(super) != bytenr ||
2146                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2147                             sizeof(super->magic))) {
2148                         brelse(bh);
2149                         continue;
2150                 }
2151
2152                 if (!latest || btrfs_super_generation(super) > transid) {
2153                         brelse(latest);
2154                         latest = bh;
2155                         transid = btrfs_super_generation(super);
2156                 } else {
2157                         brelse(bh);
2158                 }
2159         }
2160         return latest;
2161 }
2162
2163 /*
2164  * this should be called twice, once with wait == 0 and
2165  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2166  * we write are pinned.
2167  *
2168  * They are released when wait == 1 is done.
2169  * max_mirrors must be the same for both runs, and it indicates how
2170  * many supers on this one device should be written.
2171  *
2172  * max_mirrors == 0 means to write them all.
2173  */
2174 static int write_dev_supers(struct btrfs_device *device,
2175                             struct btrfs_super_block *sb,
2176                             int do_barriers, int wait, int max_mirrors)
2177 {
2178         struct buffer_head *bh;
2179         int i;
2180         int ret;
2181         int errors = 0;
2182         u32 crc;
2183         u64 bytenr;
2184         int last_barrier = 0;
2185
2186         if (max_mirrors == 0)
2187                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2188
2189         /* make sure only the last submit_bh does a barrier */
2190         if (do_barriers) {
2191                 for (i = 0; i < max_mirrors; i++) {
2192                         bytenr = btrfs_sb_offset(i);
2193                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2194                             device->total_bytes)
2195                                 break;
2196                         last_barrier = i;
2197                 }
2198         }
2199
2200         for (i = 0; i < max_mirrors; i++) {
2201                 bytenr = btrfs_sb_offset(i);
2202                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2203                         break;
2204
2205                 if (wait) {
2206                         bh = __find_get_block(device->bdev, bytenr / 4096,
2207                                               BTRFS_SUPER_INFO_SIZE);
2208                         BUG_ON(!bh);
2209                         wait_on_buffer(bh);
2210                         if (!buffer_uptodate(bh))
2211                                 errors++;
2212
2213                         /* drop our reference */
2214                         brelse(bh);
2215
2216                         /* drop the reference from the wait == 0 run */
2217                         brelse(bh);
2218                         continue;
2219                 } else {
2220                         btrfs_set_super_bytenr(sb, bytenr);
2221
2222                         crc = ~(u32)0;
2223                         crc = btrfs_csum_data(NULL, (char *)sb +
2224                                               BTRFS_CSUM_SIZE, crc,
2225                                               BTRFS_SUPER_INFO_SIZE -
2226                                               BTRFS_CSUM_SIZE);
2227                         btrfs_csum_final(crc, sb->csum);
2228
2229                         /*
2230                          * one reference for us, and we leave it for the
2231                          * caller
2232                          */
2233                         bh = __getblk(device->bdev, bytenr / 4096,
2234                                       BTRFS_SUPER_INFO_SIZE);
2235                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2236
2237                         /* one reference for submit_bh */
2238                         get_bh(bh);
2239
2240                         set_buffer_uptodate(bh);
2241                         lock_buffer(bh);
2242                         bh->b_end_io = btrfs_end_buffer_write_sync;
2243                 }
2244
2245                 if (i == last_barrier && do_barriers && device->barriers) {
2246                         ret = submit_bh(WRITE_BARRIER, bh);
2247                         if (ret == -EOPNOTSUPP) {
2248                                 printk("btrfs: disabling barriers on dev %s\n",
2249                                        device->name);
2250                                 set_buffer_uptodate(bh);
2251                                 device->barriers = 0;
2252                                 /* one reference for submit_bh */
2253                                 get_bh(bh);
2254                                 lock_buffer(bh);
2255                                 ret = submit_bh(WRITE_SYNC, bh);
2256                         }
2257                 } else {
2258                         ret = submit_bh(WRITE_SYNC, bh);
2259                 }
2260
2261                 if (ret)
2262                         errors++;
2263         }
2264         return errors < i ? 0 : -1;
2265 }
2266
2267 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2268 {
2269         struct list_head *head;
2270         struct btrfs_device *dev;
2271         struct btrfs_super_block *sb;
2272         struct btrfs_dev_item *dev_item;
2273         int ret;
2274         int do_barriers;
2275         int max_errors;
2276         int total_errors = 0;
2277         u64 flags;
2278
2279         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2280         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2281
2282         sb = &root->fs_info->super_for_commit;
2283         dev_item = &sb->dev_item;
2284
2285         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2286         head = &root->fs_info->fs_devices->devices;
2287         list_for_each_entry(dev, head, dev_list) {
2288                 if (!dev->bdev) {
2289                         total_errors++;
2290                         continue;
2291                 }
2292                 if (!dev->in_fs_metadata || !dev->writeable)
2293                         continue;
2294
2295                 btrfs_set_stack_device_generation(dev_item, 0);
2296                 btrfs_set_stack_device_type(dev_item, dev->type);
2297                 btrfs_set_stack_device_id(dev_item, dev->devid);
2298                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2299                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2300                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2301                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2302                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2303                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2304                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2305
2306                 flags = btrfs_super_flags(sb);
2307                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2308
2309                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2310                 if (ret)
2311                         total_errors++;
2312         }
2313         if (total_errors > max_errors) {
2314                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2315                        total_errors);
2316                 BUG();
2317         }
2318
2319         total_errors = 0;
2320         list_for_each_entry(dev, head, dev_list) {
2321                 if (!dev->bdev)
2322                         continue;
2323                 if (!dev->in_fs_metadata || !dev->writeable)
2324                         continue;
2325
2326                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2327                 if (ret)
2328                         total_errors++;
2329         }
2330         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2331         if (total_errors > max_errors) {
2332                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2333                        total_errors);
2334                 BUG();
2335         }
2336         return 0;
2337 }
2338
2339 int write_ctree_super(struct btrfs_trans_handle *trans,
2340                       struct btrfs_root *root, int max_mirrors)
2341 {
2342         int ret;
2343
2344         ret = write_all_supers(root, max_mirrors);
2345         return ret;
2346 }
2347
2348 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2349 {
2350         spin_lock(&fs_info->fs_roots_radix_lock);
2351         radix_tree_delete(&fs_info->fs_roots_radix,
2352                           (unsigned long)root->root_key.objectid);
2353         spin_unlock(&fs_info->fs_roots_radix_lock);
2354
2355         if (btrfs_root_refs(&root->root_item) == 0)
2356                 synchronize_srcu(&fs_info->subvol_srcu);
2357
2358         free_fs_root(root);
2359         return 0;
2360 }
2361
2362 static void free_fs_root(struct btrfs_root *root)
2363 {
2364         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2365         if (root->anon_super.s_dev) {
2366                 down_write(&root->anon_super.s_umount);
2367                 kill_anon_super(&root->anon_super);
2368         }
2369         free_extent_buffer(root->node);
2370         free_extent_buffer(root->commit_root);
2371         kfree(root->name);
2372         kfree(root);
2373 }
2374
2375 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2376 {
2377         int ret;
2378         struct btrfs_root *gang[8];
2379         int i;
2380
2381         while (!list_empty(&fs_info->dead_roots)) {
2382                 gang[0] = list_entry(fs_info->dead_roots.next,
2383                                      struct btrfs_root, root_list);
2384                 list_del(&gang[0]->root_list);
2385
2386                 if (gang[0]->in_radix) {
2387                         btrfs_free_fs_root(fs_info, gang[0]);
2388                 } else {
2389                         free_extent_buffer(gang[0]->node);
2390                         free_extent_buffer(gang[0]->commit_root);
2391                         kfree(gang[0]);
2392                 }
2393         }
2394
2395         while (1) {
2396                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2397                                              (void **)gang, 0,
2398                                              ARRAY_SIZE(gang));
2399                 if (!ret)
2400                         break;
2401                 for (i = 0; i < ret; i++)
2402                         btrfs_free_fs_root(fs_info, gang[i]);
2403         }
2404         return 0;
2405 }
2406
2407 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2408 {
2409         u64 root_objectid = 0;
2410         struct btrfs_root *gang[8];
2411         int i;
2412         int ret;
2413
2414         while (1) {
2415                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2416                                              (void **)gang, root_objectid,
2417                                              ARRAY_SIZE(gang));
2418                 if (!ret)
2419                         break;
2420
2421                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2422                 for (i = 0; i < ret; i++) {
2423                         root_objectid = gang[i]->root_key.objectid;
2424                         btrfs_orphan_cleanup(gang[i]);
2425                 }
2426                 root_objectid++;
2427         }
2428         return 0;
2429 }
2430
2431 int btrfs_commit_super(struct btrfs_root *root)
2432 {
2433         struct btrfs_trans_handle *trans;
2434         int ret;
2435
2436         mutex_lock(&root->fs_info->cleaner_mutex);
2437         btrfs_run_delayed_iputs(root);
2438         btrfs_clean_old_snapshots(root);
2439         mutex_unlock(&root->fs_info->cleaner_mutex);
2440
2441         /* wait until ongoing cleanup work done */
2442         down_write(&root->fs_info->cleanup_work_sem);
2443         up_write(&root->fs_info->cleanup_work_sem);
2444
2445         trans = btrfs_join_transaction(root, 1);
2446         ret = btrfs_commit_transaction(trans, root);
2447         BUG_ON(ret);
2448         /* run commit again to drop the original snapshot */
2449         trans = btrfs_join_transaction(root, 1);
2450         btrfs_commit_transaction(trans, root);
2451         ret = btrfs_write_and_wait_transaction(NULL, root);
2452         BUG_ON(ret);
2453
2454         ret = write_ctree_super(NULL, root, 0);
2455         return ret;
2456 }
2457
2458 int close_ctree(struct btrfs_root *root)
2459 {
2460         struct btrfs_fs_info *fs_info = root->fs_info;
2461         int ret;
2462
2463         fs_info->closing = 1;
2464         smp_mb();
2465
2466         btrfs_put_block_group_cache(fs_info);
2467         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2468                 ret =  btrfs_commit_super(root);
2469                 if (ret)
2470                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2471         }
2472
2473         kthread_stop(root->fs_info->transaction_kthread);
2474         kthread_stop(root->fs_info->cleaner_kthread);
2475
2476         fs_info->closing = 2;
2477         smp_mb();
2478
2479         if (fs_info->delalloc_bytes) {
2480                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2481                        (unsigned long long)fs_info->delalloc_bytes);
2482         }
2483         if (fs_info->total_ref_cache_size) {
2484                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2485                        (unsigned long long)fs_info->total_ref_cache_size);
2486         }
2487
2488         free_extent_buffer(fs_info->extent_root->node);
2489         free_extent_buffer(fs_info->extent_root->commit_root);
2490         free_extent_buffer(fs_info->tree_root->node);
2491         free_extent_buffer(fs_info->tree_root->commit_root);
2492         free_extent_buffer(root->fs_info->chunk_root->node);
2493         free_extent_buffer(root->fs_info->chunk_root->commit_root);
2494         free_extent_buffer(root->fs_info->dev_root->node);
2495         free_extent_buffer(root->fs_info->dev_root->commit_root);
2496         free_extent_buffer(root->fs_info->csum_root->node);
2497         free_extent_buffer(root->fs_info->csum_root->commit_root);
2498
2499         btrfs_free_block_groups(root->fs_info);
2500
2501         del_fs_roots(fs_info);
2502
2503         iput(fs_info->btree_inode);
2504
2505         btrfs_stop_workers(&fs_info->generic_worker);
2506         btrfs_stop_workers(&fs_info->fixup_workers);
2507         btrfs_stop_workers(&fs_info->delalloc_workers);
2508         btrfs_stop_workers(&fs_info->workers);
2509         btrfs_stop_workers(&fs_info->endio_workers);
2510         btrfs_stop_workers(&fs_info->endio_meta_workers);
2511         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2512         btrfs_stop_workers(&fs_info->endio_write_workers);
2513         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2514         btrfs_stop_workers(&fs_info->submit_workers);
2515
2516         btrfs_close_devices(fs_info->fs_devices);
2517         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2518
2519         bdi_destroy(&fs_info->bdi);
2520         cleanup_srcu_struct(&fs_info->subvol_srcu);
2521
2522         kfree(fs_info->extent_root);
2523         kfree(fs_info->tree_root);
2524         kfree(fs_info->chunk_root);
2525         kfree(fs_info->dev_root);
2526         kfree(fs_info->csum_root);
2527         return 0;
2528 }
2529
2530 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2531 {
2532         int ret;
2533         struct inode *btree_inode = buf->first_page->mapping->host;
2534
2535         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2536                                      NULL);
2537         if (!ret)
2538                 return ret;
2539
2540         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2541                                     parent_transid);
2542         return !ret;
2543 }
2544
2545 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2546 {
2547         struct inode *btree_inode = buf->first_page->mapping->host;
2548         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2549                                           buf);
2550 }
2551
2552 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2553 {
2554         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2555         u64 transid = btrfs_header_generation(buf);
2556         struct inode *btree_inode = root->fs_info->btree_inode;
2557         int was_dirty;
2558
2559         btrfs_assert_tree_locked(buf);
2560         if (transid != root->fs_info->generation) {
2561                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2562                        "found %llu running %llu\n",
2563                         (unsigned long long)buf->start,
2564                         (unsigned long long)transid,
2565                         (unsigned long long)root->fs_info->generation);
2566                 WARN_ON(1);
2567         }
2568         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2569                                             buf);
2570         if (!was_dirty) {
2571                 spin_lock(&root->fs_info->delalloc_lock);
2572                 root->fs_info->dirty_metadata_bytes += buf->len;
2573                 spin_unlock(&root->fs_info->delalloc_lock);
2574         }
2575 }
2576
2577 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2578 {
2579         /*
2580          * looks as though older kernels can get into trouble with
2581          * this code, they end up stuck in balance_dirty_pages forever
2582          */
2583         u64 num_dirty;
2584         unsigned long thresh = 32 * 1024 * 1024;
2585
2586         if (current->flags & PF_MEMALLOC)
2587                 return;
2588
2589         num_dirty = root->fs_info->dirty_metadata_bytes;
2590
2591         if (num_dirty > thresh) {
2592                 balance_dirty_pages_ratelimited_nr(
2593                                    root->fs_info->btree_inode->i_mapping, 1);
2594         }
2595         return;
2596 }
2597
2598 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2599 {
2600         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2601         int ret;
2602         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2603         if (ret == 0)
2604                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2605         return ret;
2606 }
2607
2608 int btree_lock_page_hook(struct page *page)
2609 {
2610         struct inode *inode = page->mapping->host;
2611         struct btrfs_root *root = BTRFS_I(inode)->root;
2612         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2613         struct extent_buffer *eb;
2614         unsigned long len;
2615         u64 bytenr = page_offset(page);
2616
2617         if (page->private == EXTENT_PAGE_PRIVATE)
2618                 goto out;
2619
2620         len = page->private >> 2;
2621         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2622         if (!eb)
2623                 goto out;
2624
2625         btrfs_tree_lock(eb);
2626         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2627
2628         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2629                 spin_lock(&root->fs_info->delalloc_lock);
2630                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2631                         root->fs_info->dirty_metadata_bytes -= eb->len;
2632                 else
2633                         WARN_ON(1);
2634                 spin_unlock(&root->fs_info->delalloc_lock);
2635         }
2636
2637         btrfs_tree_unlock(eb);
2638         free_extent_buffer(eb);
2639 out:
2640         lock_page(page);
2641         return 0;
2642 }
2643
2644 static struct extent_io_ops btree_extent_io_ops = {
2645         .write_cache_pages_lock_hook = btree_lock_page_hook,
2646         .readpage_end_io_hook = btree_readpage_end_io_hook,
2647         .submit_bio_hook = btree_submit_bio_hook,
2648         /* note we're sharing with inode.c for the merge bio hook */
2649         .merge_bio_hook = btrfs_merge_bio_hook,
2650 };