fs/btrfs/disk-io.c

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