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