fs/btrfs/disk-io.c

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