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         BUG_ON(!root->node);
1011         root->commit_root = btrfs_root_node(root);
1012         return 0;
1013 }
1014
1015 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1016                                          struct btrfs_fs_info *fs_info)
1017 {
1018         struct btrfs_root *root;
1019         struct btrfs_root *tree_root = fs_info->tree_root;
1020         struct extent_buffer *leaf;
1021
1022         root = kzalloc(sizeof(*root), GFP_NOFS);
1023         if (!root)
1024                 return ERR_PTR(-ENOMEM);
1025
1026         __setup_root(tree_root->nodesize, tree_root->leafsize,
1027                      tree_root->sectorsize, tree_root->stripesize,
1028                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1029
1030         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1031         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1032         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1033         /*
1034          * log trees do not get reference counted because they go away
1035          * before a real commit is actually done.  They do store pointers
1036          * to file data extents, and those reference counts still get
1037          * updated (along with back refs to the log tree).
1038          */
1039         root->ref_cows = 0;
1040
1041         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1042                                       BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1043         if (IS_ERR(leaf)) {
1044                 kfree(root);
1045                 return ERR_CAST(leaf);
1046         }
1047
1048         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1049         btrfs_set_header_bytenr(leaf, leaf->start);
1050         btrfs_set_header_generation(leaf, trans->transid);
1051         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1052         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1053         root->node = leaf;
1054
1055         write_extent_buffer(root->node, root->fs_info->fsid,
1056                             (unsigned long)btrfs_header_fsid(root->node),
1057                             BTRFS_FSID_SIZE);
1058         btrfs_mark_buffer_dirty(root->node);
1059         btrfs_tree_unlock(root->node);
1060         return root;
1061 }
1062
1063 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1064                              struct btrfs_fs_info *fs_info)
1065 {
1066         struct btrfs_root *log_root;
1067
1068         log_root = alloc_log_tree(trans, fs_info);
1069         if (IS_ERR(log_root))
1070                 return PTR_ERR(log_root);
1071         WARN_ON(fs_info->log_root_tree);
1072         fs_info->log_root_tree = log_root;
1073         return 0;
1074 }
1075
1076 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1077                        struct btrfs_root *root)
1078 {
1079         struct btrfs_root *log_root;
1080         struct btrfs_inode_item *inode_item;
1081
1082         log_root = alloc_log_tree(trans, root->fs_info);
1083         if (IS_ERR(log_root))
1084                 return PTR_ERR(log_root);
1085
1086         log_root->last_trans = trans->transid;
1087         log_root->root_key.offset = root->root_key.objectid;
1088
1089         inode_item = &log_root->root_item.inode;
1090         inode_item->generation = cpu_to_le64(1);
1091         inode_item->size = cpu_to_le64(3);
1092         inode_item->nlink = cpu_to_le32(1);
1093         inode_item->nbytes = cpu_to_le64(root->leafsize);
1094         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1095
1096         btrfs_set_root_node(&log_root->root_item, log_root->node);
1097
1098         WARN_ON(root->log_root);
1099         root->log_root = log_root;
1100         root->log_transid = 0;
1101         root->last_log_commit = 0;
1102         return 0;
1103 }
1104
1105 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1106                                                struct btrfs_key *location)
1107 {
1108         struct btrfs_root *root;
1109         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1110         struct btrfs_path *path;
1111         struct extent_buffer *l;
1112         u64 generation;
1113         u32 blocksize;
1114         int ret = 0;
1115
1116         root = kzalloc(sizeof(*root), GFP_NOFS);
1117         if (!root)
1118                 return ERR_PTR(-ENOMEM);
1119         if (location->offset == (u64)-1) {
1120                 ret = find_and_setup_root(tree_root, fs_info,
1121                                           location->objectid, root);
1122                 if (ret) {
1123                         kfree(root);
1124                         return ERR_PTR(ret);
1125                 }
1126                 goto out;
1127         }
1128
1129         __setup_root(tree_root->nodesize, tree_root->leafsize,
1130                      tree_root->sectorsize, tree_root->stripesize,
1131                      root, fs_info, location->objectid);
1132
1133         path = btrfs_alloc_path();
1134         BUG_ON(!path);
1135         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1136         if (ret == 0) {
1137                 l = path->nodes[0];
1138                 read_extent_buffer(l, &root->root_item,
1139                                 btrfs_item_ptr_offset(l, path->slots[0]),
1140                                 sizeof(root->root_item));
1141                 memcpy(&root->root_key, location, sizeof(*location));
1142         }
1143         btrfs_free_path(path);
1144         if (ret) {
1145                 if (ret > 0)
1146                         ret = -ENOENT;
1147                 return ERR_PTR(ret);
1148         }
1149
1150         generation = btrfs_root_generation(&root->root_item);
1151         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1152         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1153                                      blocksize, generation);
1154         root->commit_root = btrfs_root_node(root);
1155         BUG_ON(!root->node);
1156 out:
1157         if (location->objectid != BTRFS_TREE_LOG_OBJECTID)
1158                 root->ref_cows = 1;
1159
1160         return root;
1161 }
1162
1163 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1164                                         u64 root_objectid)
1165 {
1166         struct btrfs_root *root;
1167
1168         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1169                 return fs_info->tree_root;
1170         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1171                 return fs_info->extent_root;
1172
1173         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1174                                  (unsigned long)root_objectid);
1175         return root;
1176 }
1177
1178 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1179                                               struct btrfs_key *location)
1180 {
1181         struct btrfs_root *root;
1182         int ret;
1183
1184         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1185                 return fs_info->tree_root;
1186         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1187                 return fs_info->extent_root;
1188         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1189                 return fs_info->chunk_root;
1190         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1191                 return fs_info->dev_root;
1192         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1193                 return fs_info->csum_root;
1194 again:
1195         spin_lock(&fs_info->fs_roots_radix_lock);
1196         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1197                                  (unsigned long)location->objectid);
1198         spin_unlock(&fs_info->fs_roots_radix_lock);
1199         if (root)
1200                 return root;
1201
1202         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1203         if (IS_ERR(root))
1204                 return root;
1205
1206         set_anon_super(&root->anon_super, NULL);
1207
1208         if (btrfs_root_refs(&root->root_item) == 0) {
1209                 ret = -ENOENT;
1210                 goto fail;
1211         }
1212
1213         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1214         if (ret < 0)
1215                 goto fail;
1216         if (ret == 0)
1217                 root->orphan_item_inserted = 1;
1218
1219         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1220         if (ret)
1221                 goto fail;
1222
1223         spin_lock(&fs_info->fs_roots_radix_lock);
1224         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1225                                 (unsigned long)root->root_key.objectid,
1226                                 root);
1227         if (ret == 0)
1228                 root->in_radix = 1;
1229
1230         spin_unlock(&fs_info->fs_roots_radix_lock);
1231         radix_tree_preload_end();
1232         if (ret) {
1233                 if (ret == -EEXIST) {
1234                         free_fs_root(root);
1235                         goto again;
1236                 }
1237                 goto fail;
1238         }
1239
1240         ret = btrfs_find_dead_roots(fs_info->tree_root,
1241                                     root->root_key.objectid);
1242         WARN_ON(ret);
1243         return root;
1244 fail:
1245         free_fs_root(root);
1246         return ERR_PTR(ret);
1247 }
1248
1249 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1250                                       struct btrfs_key *location,
1251                                       const char *name, int namelen)
1252 {
1253         return btrfs_read_fs_root_no_name(fs_info, location);
1254 #if 0
1255         struct btrfs_root *root;
1256         int ret;
1257
1258         root = btrfs_read_fs_root_no_name(fs_info, location);
1259         if (!root)
1260                 return NULL;
1261
1262         if (root->in_sysfs)
1263                 return root;
1264
1265         ret = btrfs_set_root_name(root, name, namelen);
1266         if (ret) {
1267                 free_extent_buffer(root->node);
1268                 kfree(root);
1269                 return ERR_PTR(ret);
1270         }
1271
1272         ret = btrfs_sysfs_add_root(root);
1273         if (ret) {
1274                 free_extent_buffer(root->node);
1275                 kfree(root->name);
1276                 kfree(root);
1277                 return ERR_PTR(ret);
1278         }
1279         root->in_sysfs = 1;
1280         return root;
1281 #endif
1282 }
1283
1284 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1285 {
1286         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1287         int ret = 0;
1288         struct btrfs_device *device;
1289         struct backing_dev_info *bdi;
1290
1291         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1292                 if (!device->bdev)
1293                         continue;
1294                 bdi = blk_get_backing_dev_info(device->bdev);
1295                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1296                         ret = 1;
1297                         break;
1298                 }
1299         }
1300         return ret;
1301 }
1302
1303 /*
1304  * this unplugs every device on the box, and it is only used when page
1305  * is null
1306  */
1307 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1308 {
1309         struct btrfs_device *device;
1310         struct btrfs_fs_info *info;
1311
1312         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1313         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1314                 if (!device->bdev)
1315                         continue;
1316
1317                 bdi = blk_get_backing_dev_info(device->bdev);
1318                 if (bdi->unplug_io_fn)
1319                         bdi->unplug_io_fn(bdi, page);
1320         }
1321 }
1322
1323 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1324 {
1325         struct inode *inode;
1326         struct extent_map_tree *em_tree;
1327         struct extent_map *em;
1328         struct address_space *mapping;
1329         u64 offset;
1330
1331         /* the generic O_DIRECT read code does this */
1332         if (1 || !page) {
1333                 __unplug_io_fn(bdi, page);
1334                 return;
1335         }
1336
1337         /*
1338          * page->mapping may change at any time.  Get a consistent copy
1339          * and use that for everything below
1340          */
1341         smp_mb();
1342         mapping = page->mapping;
1343         if (!mapping)
1344                 return;
1345
1346         inode = mapping->host;
1347
1348         /*
1349          * don't do the expensive searching for a small number of
1350          * devices
1351          */
1352         if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1353                 __unplug_io_fn(bdi, page);
1354                 return;
1355         }
1356
1357         offset = page_offset(page);
1358
1359         em_tree = &BTRFS_I(inode)->extent_tree;
1360         read_lock(&em_tree->lock);
1361         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1362         read_unlock(&em_tree->lock);
1363         if (!em) {
1364                 __unplug_io_fn(bdi, page);
1365                 return;
1366         }
1367
1368         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1369                 free_extent_map(em);
1370                 __unplug_io_fn(bdi, page);
1371                 return;
1372         }
1373         offset = offset - em->start;
1374         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1375                           em->block_start + offset, page);
1376         free_extent_map(em);
1377 }
1378
1379 /*
1380  * If this fails, caller must call bdi_destroy() to get rid of the
1381  * bdi again.
1382  */
1383 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1384 {
1385         int err;
1386
1387         bdi->capabilities = BDI_CAP_MAP_COPY;
1388         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1389         if (err)
1390                 return err;
1391
1392         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1393         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1394         bdi->unplug_io_data     = info;
1395         bdi->congested_fn       = btrfs_congested_fn;
1396         bdi->congested_data     = info;
1397         return 0;
1398 }
1399
1400 static int bio_ready_for_csum(struct bio *bio)
1401 {
1402         u64 length = 0;
1403         u64 buf_len = 0;
1404         u64 start = 0;
1405         struct page *page;
1406         struct extent_io_tree *io_tree = NULL;
1407         struct bio_vec *bvec;
1408         int i;
1409         int ret;
1410
1411         bio_for_each_segment(bvec, bio, i) {
1412                 page = bvec->bv_page;
1413                 if (page->private == EXTENT_PAGE_PRIVATE) {
1414                         length += bvec->bv_len;
1415                         continue;
1416                 }
1417                 if (!page->private) {
1418                         length += bvec->bv_len;
1419                         continue;
1420                 }
1421                 length = bvec->bv_len;
1422                 buf_len = page->private >> 2;
1423                 start = page_offset(page) + bvec->bv_offset;
1424                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1425         }
1426         /* are we fully contained in this bio? */
1427         if (buf_len <= length)
1428                 return 1;
1429
1430         ret = extent_range_uptodate(io_tree, start + length,
1431                                     start + buf_len - 1);
1432         return ret;
1433 }
1434
1435 /*
1436  * called by the kthread helper functions to finally call the bio end_io
1437  * functions.  This is where read checksum verification actually happens
1438  */
1439 static void end_workqueue_fn(struct btrfs_work *work)
1440 {
1441         struct bio *bio;
1442         struct end_io_wq *end_io_wq;
1443         struct btrfs_fs_info *fs_info;
1444         int error;
1445
1446         end_io_wq = container_of(work, struct end_io_wq, work);
1447         bio = end_io_wq->bio;
1448         fs_info = end_io_wq->info;
1449
1450         /* metadata bio reads are special because the whole tree block must
1451          * be checksummed at once.  This makes sure the entire block is in
1452          * ram and up to date before trying to verify things.  For
1453          * blocksize <= pagesize, it is basically a noop
1454          */
1455         if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1456             !bio_ready_for_csum(bio)) {
1457                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1458                                    &end_io_wq->work);
1459                 return;
1460         }
1461         error = end_io_wq->error;
1462         bio->bi_private = end_io_wq->private;
1463         bio->bi_end_io = end_io_wq->end_io;
1464         kfree(end_io_wq);
1465         bio_endio(bio, error);
1466 }
1467
1468 static int cleaner_kthread(void *arg)
1469 {
1470         struct btrfs_root *root = arg;
1471
1472         do {
1473                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1474
1475                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1476                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1477                         btrfs_run_delayed_iputs(root);
1478                         btrfs_clean_old_snapshots(root);
1479                         mutex_unlock(&root->fs_info->cleaner_mutex);
1480                 }
1481
1482                 if (freezing(current)) {
1483                         refrigerator();
1484                 } else {
1485                         set_current_state(TASK_INTERRUPTIBLE);
1486                         if (!kthread_should_stop())
1487                                 schedule();
1488                         __set_current_state(TASK_RUNNING);
1489                 }
1490         } while (!kthread_should_stop());
1491         return 0;
1492 }
1493
1494 static int transaction_kthread(void *arg)
1495 {
1496         struct btrfs_root *root = arg;
1497         struct btrfs_trans_handle *trans;
1498         struct btrfs_transaction *cur;
1499         u64 transid;
1500         unsigned long now;
1501         unsigned long delay;
1502         int ret;
1503
1504         do {
1505                 delay = HZ * 30;
1506                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1507                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1508
1509                 spin_lock(&root->fs_info->new_trans_lock);
1510                 cur = root->fs_info->running_transaction;
1511                 if (!cur) {
1512                         spin_unlock(&root->fs_info->new_trans_lock);
1513                         goto sleep;
1514                 }
1515
1516                 now = get_seconds();
1517                 if (!cur->blocked &&
1518                     (now < cur->start_time || now - cur->start_time < 30)) {
1519                         spin_unlock(&root->fs_info->new_trans_lock);
1520                         delay = HZ * 5;
1521                         goto sleep;
1522                 }
1523                 transid = cur->transid;
1524                 spin_unlock(&root->fs_info->new_trans_lock);
1525
1526                 trans = btrfs_join_transaction(root, 1);
1527                 if (transid == trans->transid) {
1528                         ret = btrfs_commit_transaction(trans, root);
1529                         BUG_ON(ret);
1530                 } else {
1531                         btrfs_end_transaction(trans, root);
1532                 }
1533 sleep:
1534                 wake_up_process(root->fs_info->cleaner_kthread);
1535                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1536
1537                 if (freezing(current)) {
1538                         refrigerator();
1539                 } else {
1540                         set_current_state(TASK_INTERRUPTIBLE);
1541                         if (!kthread_should_stop() &&
1542                             !btrfs_transaction_blocked(root->fs_info))
1543                                 schedule_timeout(delay);
1544                         __set_current_state(TASK_RUNNING);
1545                 }
1546         } while (!kthread_should_stop());
1547         return 0;
1548 }
1549
1550 struct btrfs_root *open_ctree(struct super_block *sb,
1551                               struct btrfs_fs_devices *fs_devices,
1552                               char *options)
1553 {
1554         u32 sectorsize;
1555         u32 nodesize;
1556         u32 leafsize;
1557         u32 blocksize;
1558         u32 stripesize;
1559         u64 generation;
1560         u64 features;
1561         struct btrfs_key location;
1562         struct buffer_head *bh;
1563         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1564                                                  GFP_NOFS);
1565         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1566                                                  GFP_NOFS);
1567         struct btrfs_root *tree_root = btrfs_sb(sb);
1568         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1569         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1570                                                 GFP_NOFS);
1571         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1572                                               GFP_NOFS);
1573         struct btrfs_root *log_tree_root;
1574
1575         int ret;
1576         int err = -EINVAL;
1577
1578         struct btrfs_super_block *disk_super;
1579
1580         if (!extent_root || !tree_root || !fs_info ||
1581             !chunk_root || !dev_root || !csum_root) {
1582                 err = -ENOMEM;
1583                 goto fail;
1584         }
1585
1586         ret = init_srcu_struct(&fs_info->subvol_srcu);
1587         if (ret) {
1588                 err = ret;
1589                 goto fail;
1590         }
1591
1592         ret = setup_bdi(fs_info, &fs_info->bdi);
1593         if (ret) {
1594                 err = ret;
1595                 goto fail_srcu;
1596         }
1597
1598         fs_info->btree_inode = new_inode(sb);
1599         if (!fs_info->btree_inode) {
1600                 err = -ENOMEM;
1601                 goto fail_bdi;
1602         }
1603
1604         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1605         INIT_LIST_HEAD(&fs_info->trans_list);
1606         INIT_LIST_HEAD(&fs_info->dead_roots);
1607         INIT_LIST_HEAD(&fs_info->delayed_iputs);
1608         INIT_LIST_HEAD(&fs_info->hashers);
1609         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1610         INIT_LIST_HEAD(&fs_info->ordered_operations);
1611         INIT_LIST_HEAD(&fs_info->caching_block_groups);
1612         spin_lock_init(&fs_info->delalloc_lock);
1613         spin_lock_init(&fs_info->new_trans_lock);
1614         spin_lock_init(&fs_info->ref_cache_lock);
1615         spin_lock_init(&fs_info->fs_roots_radix_lock);
1616         spin_lock_init(&fs_info->delayed_iput_lock);
1617
1618         init_completion(&fs_info->kobj_unregister);
1619         fs_info->tree_root = tree_root;
1620         fs_info->extent_root = extent_root;
1621         fs_info->csum_root = csum_root;
1622         fs_info->chunk_root = chunk_root;
1623         fs_info->dev_root = dev_root;
1624         fs_info->fs_devices = fs_devices;
1625         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1626         INIT_LIST_HEAD(&fs_info->space_info);
1627         btrfs_mapping_init(&fs_info->mapping_tree);
1628         btrfs_init_block_rsv(&fs_info->global_block_rsv);
1629         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1630         btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1631         btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1632         btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1633         INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1634         mutex_init(&fs_info->durable_block_rsv_mutex);
1635         atomic_set(&fs_info->nr_async_submits, 0);
1636         atomic_set(&fs_info->async_delalloc_pages, 0);
1637         atomic_set(&fs_info->async_submit_draining, 0);
1638         atomic_set(&fs_info->nr_async_bios, 0);
1639         fs_info->sb = sb;
1640         fs_info->max_inline = 8192 * 1024;
1641         fs_info->metadata_ratio = 0;
1642
1643         fs_info->thread_pool_size = min_t(unsigned long,
1644                                           num_online_cpus() + 2, 8);
1645
1646         INIT_LIST_HEAD(&fs_info->ordered_extents);
1647         spin_lock_init(&fs_info->ordered_extent_lock);
1648
1649         sb->s_blocksize = 4096;
1650         sb->s_blocksize_bits = blksize_bits(4096);
1651         sb->s_bdi = &fs_info->bdi;
1652
1653         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1654         fs_info->btree_inode->i_nlink = 1;
1655         /*
1656          * we set the i_size on the btree inode to the max possible int.
1657          * the real end of the address space is determined by all of
1658          * the devices in the system
1659          */
1660         fs_info->btree_inode->i_size = OFFSET_MAX;
1661         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1662         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1663
1664         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1665         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1666                              fs_info->btree_inode->i_mapping,
1667                              GFP_NOFS);
1668         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1669                              GFP_NOFS);
1670
1671         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1672
1673         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1674         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1675                sizeof(struct btrfs_key));
1676         BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1677         insert_inode_hash(fs_info->btree_inode);
1678
1679         spin_lock_init(&fs_info->block_group_cache_lock);
1680         fs_info->block_group_cache_tree = RB_ROOT;
1681
1682         extent_io_tree_init(&fs_info->freed_extents[0],
1683                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1684         extent_io_tree_init(&fs_info->freed_extents[1],
1685                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1686         fs_info->pinned_extents = &fs_info->freed_extents[0];
1687         fs_info->do_barriers = 1;
1688
1689
1690         mutex_init(&fs_info->trans_mutex);
1691         mutex_init(&fs_info->ordered_operations_mutex);
1692         mutex_init(&fs_info->tree_log_mutex);
1693         mutex_init(&fs_info->chunk_mutex);
1694         mutex_init(&fs_info->transaction_kthread_mutex);
1695         mutex_init(&fs_info->cleaner_mutex);
1696         mutex_init(&fs_info->volume_mutex);
1697         init_rwsem(&fs_info->extent_commit_sem);
1698         init_rwsem(&fs_info->cleanup_work_sem);
1699         init_rwsem(&fs_info->subvol_sem);
1700
1701         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1702         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1703
1704         init_waitqueue_head(&fs_info->transaction_throttle);
1705         init_waitqueue_head(&fs_info->transaction_wait);
1706         init_waitqueue_head(&fs_info->transaction_blocked_wait);
1707         init_waitqueue_head(&fs_info->async_submit_wait);
1708
1709         __setup_root(4096, 4096, 4096, 4096, tree_root,
1710                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1711
1712         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1713         if (!bh)
1714                 goto fail_iput;
1715
1716         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1717         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1718                sizeof(fs_info->super_for_commit));
1719         brelse(bh);
1720
1721         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1722
1723         disk_super = &fs_info->super_copy;
1724         if (!btrfs_super_root(disk_super))
1725                 goto fail_iput;
1726
1727         ret = btrfs_parse_options(tree_root, options);
1728         if (ret) {
1729                 err = ret;
1730                 goto fail_iput;
1731         }
1732
1733         features = btrfs_super_incompat_flags(disk_super) &
1734                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1735         if (features) {
1736                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1737                        "unsupported optional features (%Lx).\n",
1738                        (unsigned long long)features);
1739                 err = -EINVAL;
1740                 goto fail_iput;
1741         }
1742
1743         features = btrfs_super_incompat_flags(disk_super);
1744         if (!(features & BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF)) {
1745                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1746                 btrfs_set_super_incompat_flags(disk_super, features);
1747         }
1748
1749         features = btrfs_super_compat_ro_flags(disk_super) &
1750                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1751         if (!(sb->s_flags & MS_RDONLY) && features) {
1752                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1753                        "unsupported option features (%Lx).\n",
1754                        (unsigned long long)features);
1755                 err = -EINVAL;
1756                 goto fail_iput;
1757         }
1758
1759         btrfs_init_workers(&fs_info->generic_worker,
1760                            "genwork", 1, NULL);
1761
1762         btrfs_init_workers(&fs_info->workers, "worker",
1763                            fs_info->thread_pool_size,
1764                            &fs_info->generic_worker);
1765
1766         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1767                            fs_info->thread_pool_size,
1768                            &fs_info->generic_worker);
1769
1770         btrfs_init_workers(&fs_info->submit_workers, "submit",
1771                            min_t(u64, fs_devices->num_devices,
1772                            fs_info->thread_pool_size),
1773                            &fs_info->generic_worker);
1774
1775         /* a higher idle thresh on the submit workers makes it much more
1776          * likely that bios will be send down in a sane order to the
1777          * devices
1778          */
1779         fs_info->submit_workers.idle_thresh = 64;
1780
1781         fs_info->workers.idle_thresh = 16;
1782         fs_info->workers.ordered = 1;
1783
1784         fs_info->delalloc_workers.idle_thresh = 2;
1785         fs_info->delalloc_workers.ordered = 1;
1786
1787         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1788                            &fs_info->generic_worker);
1789         btrfs_init_workers(&fs_info->endio_workers, "endio",
1790                            fs_info->thread_pool_size,
1791                            &fs_info->generic_worker);
1792         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1793                            fs_info->thread_pool_size,
1794                            &fs_info->generic_worker);
1795         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1796                            "endio-meta-write", fs_info->thread_pool_size,
1797                            &fs_info->generic_worker);
1798         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1799                            fs_info->thread_pool_size,
1800                            &fs_info->generic_worker);
1801         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1802                            1, &fs_info->generic_worker);
1803
1804         /*
1805          * endios are largely parallel and should have a very
1806          * low idle thresh
1807          */
1808         fs_info->endio_workers.idle_thresh = 4;
1809         fs_info->endio_meta_workers.idle_thresh = 4;
1810
1811         fs_info->endio_write_workers.idle_thresh = 2;
1812         fs_info->endio_meta_write_workers.idle_thresh = 2;
1813
1814         btrfs_start_workers(&fs_info->workers, 1);
1815         btrfs_start_workers(&fs_info->generic_worker, 1);
1816         btrfs_start_workers(&fs_info->submit_workers, 1);
1817         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1818         btrfs_start_workers(&fs_info->fixup_workers, 1);
1819         btrfs_start_workers(&fs_info->endio_workers, 1);
1820         btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1821         btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1822         btrfs_start_workers(&fs_info->endio_write_workers, 1);
1823         btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1824
1825         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1826         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1827                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1828
1829         nodesize = btrfs_super_nodesize(disk_super);
1830         leafsize = btrfs_super_leafsize(disk_super);
1831         sectorsize = btrfs_super_sectorsize(disk_super);
1832         stripesize = btrfs_super_stripesize(disk_super);
1833         tree_root->nodesize = nodesize;
1834         tree_root->leafsize = leafsize;
1835         tree_root->sectorsize = sectorsize;
1836         tree_root->stripesize = stripesize;
1837
1838         sb->s_blocksize = sectorsize;
1839         sb->s_blocksize_bits = blksize_bits(sectorsize);
1840
1841         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1842                     sizeof(disk_super->magic))) {
1843                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1844                 goto fail_sb_buffer;
1845         }
1846
1847         mutex_lock(&fs_info->chunk_mutex);
1848         ret = btrfs_read_sys_array(tree_root);
1849         mutex_unlock(&fs_info->chunk_mutex);
1850         if (ret) {
1851                 printk(KERN_WARNING "btrfs: failed to read the system "
1852                        "array on %s\n", sb->s_id);
1853                 goto fail_sb_buffer;
1854         }
1855
1856         blocksize = btrfs_level_size(tree_root,
1857                                      btrfs_super_chunk_root_level(disk_super));
1858         generation = btrfs_super_chunk_root_generation(disk_super);
1859
1860         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1861                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1862
1863         chunk_root->node = read_tree_block(chunk_root,
1864                                            btrfs_super_chunk_root(disk_super),
1865                                            blocksize, generation);
1866         BUG_ON(!chunk_root->node);
1867         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1868                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1869                        sb->s_id);
1870                 goto fail_chunk_root;
1871         }
1872         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1873         chunk_root->commit_root = btrfs_root_node(chunk_root);
1874
1875         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1876            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1877            BTRFS_UUID_SIZE);
1878
1879         mutex_lock(&fs_info->chunk_mutex);
1880         ret = btrfs_read_chunk_tree(chunk_root);
1881         mutex_unlock(&fs_info->chunk_mutex);
1882         if (ret) {
1883                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1884                        sb->s_id);
1885                 goto fail_chunk_root;
1886         }
1887
1888         btrfs_close_extra_devices(fs_devices);
1889
1890         blocksize = btrfs_level_size(tree_root,
1891                                      btrfs_super_root_level(disk_super));
1892         generation = btrfs_super_generation(disk_super);
1893
1894         tree_root->node = read_tree_block(tree_root,
1895                                           btrfs_super_root(disk_super),
1896                                           blocksize, generation);
1897         if (!tree_root->node)
1898                 goto fail_chunk_root;
1899         if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1900                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1901                        sb->s_id);
1902                 goto fail_tree_root;
1903         }
1904         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1905         tree_root->commit_root = btrfs_root_node(tree_root);
1906
1907         ret = find_and_setup_root(tree_root, fs_info,
1908                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1909         if (ret)
1910                 goto fail_tree_root;
1911         extent_root->track_dirty = 1;
1912
1913         ret = find_and_setup_root(tree_root, fs_info,
1914                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1915         if (ret)
1916                 goto fail_extent_root;
1917         dev_root->track_dirty = 1;
1918
1919         ret = find_and_setup_root(tree_root, fs_info,
1920                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
1921         if (ret)
1922                 goto fail_dev_root;
1923
1924         csum_root->track_dirty = 1;
1925
1926         fs_info->generation = generation;
1927         fs_info->last_trans_committed = generation;
1928         fs_info->data_alloc_profile = (u64)-1;
1929         fs_info->metadata_alloc_profile = (u64)-1;
1930         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1931
1932         ret = btrfs_read_block_groups(extent_root);
1933         if (ret) {
1934                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
1935                 goto fail_block_groups;
1936         }
1937
1938         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1939                                                "btrfs-cleaner");
1940         if (IS_ERR(fs_info->cleaner_kthread))
1941                 goto fail_block_groups;
1942
1943         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1944                                                    tree_root,
1945                                                    "btrfs-transaction");
1946         if (IS_ERR(fs_info->transaction_kthread))
1947                 goto fail_cleaner;
1948
1949         if (!btrfs_test_opt(tree_root, SSD) &&
1950             !btrfs_test_opt(tree_root, NOSSD) &&
1951             !fs_info->fs_devices->rotating) {
1952                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
1953                        "mode\n");
1954                 btrfs_set_opt(fs_info->mount_opt, SSD);
1955         }
1956
1957         if (btrfs_super_log_root(disk_super) != 0) {
1958                 u64 bytenr = btrfs_super_log_root(disk_super);
1959
1960                 if (fs_devices->rw_devices == 0) {
1961                         printk(KERN_WARNING "Btrfs log replay required "
1962                                "on RO media\n");
1963                         err = -EIO;
1964                         goto fail_trans_kthread;
1965                 }
1966                 blocksize =
1967                      btrfs_level_size(tree_root,
1968                                       btrfs_super_log_root_level(disk_super));
1969
1970                 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1971                 if (!log_tree_root) {
1972                         err = -ENOMEM;
1973                         goto fail_trans_kthread;
1974                 }
1975
1976                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1977                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1978
1979                 log_tree_root->node = read_tree_block(tree_root, bytenr,
1980                                                       blocksize,
1981                                                       generation + 1);
1982                 ret = btrfs_recover_log_trees(log_tree_root);
1983                 BUG_ON(ret);
1984
1985                 if (sb->s_flags & MS_RDONLY) {
1986                         ret =  btrfs_commit_super(tree_root);
1987                         BUG_ON(ret);
1988                 }
1989         }
1990
1991         ret = btrfs_find_orphan_roots(tree_root);
1992         BUG_ON(ret);
1993
1994         if (!(sb->s_flags & MS_RDONLY)) {
1995                 ret = btrfs_cleanup_fs_roots(fs_info);
1996                 BUG_ON(ret);
1997
1998                 ret = btrfs_recover_relocation(tree_root);
1999                 if (ret < 0) {
2000                         printk(KERN_WARNING
2001                                "btrfs: failed to recover relocation\n");
2002                         err = -EINVAL;
2003                         goto fail_trans_kthread;
2004                 }
2005         }
2006
2007         location.objectid = BTRFS_FS_TREE_OBJECTID;
2008         location.type = BTRFS_ROOT_ITEM_KEY;
2009         location.offset = (u64)-1;
2010
2011         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2012         if (!fs_info->fs_root)
2013                 goto fail_trans_kthread;
2014         if (IS_ERR(fs_info->fs_root)) {
2015                 err = PTR_ERR(fs_info->fs_root);
2016                 goto fail_trans_kthread;
2017         }
2018
2019         if (!(sb->s_flags & MS_RDONLY)) {
2020                 down_read(&fs_info->cleanup_work_sem);
2021                 btrfs_orphan_cleanup(fs_info->fs_root);
2022                 btrfs_orphan_cleanup(fs_info->tree_root);
2023                 up_read(&fs_info->cleanup_work_sem);
2024         }
2025
2026         return tree_root;
2027
2028 fail_trans_kthread:
2029         kthread_stop(fs_info->transaction_kthread);
2030 fail_cleaner:
2031         kthread_stop(fs_info->cleaner_kthread);
2032
2033         /*
2034          * make sure we're done with the btree inode before we stop our
2035          * kthreads
2036          */
2037         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2038         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2039
2040 fail_block_groups:
2041         btrfs_free_block_groups(fs_info);
2042         free_extent_buffer(csum_root->node);
2043         free_extent_buffer(csum_root->commit_root);
2044 fail_dev_root:
2045         free_extent_buffer(dev_root->node);
2046         free_extent_buffer(dev_root->commit_root);
2047 fail_extent_root:
2048         free_extent_buffer(extent_root->node);
2049         free_extent_buffer(extent_root->commit_root);
2050 fail_tree_root:
2051         free_extent_buffer(tree_root->node);
2052         free_extent_buffer(tree_root->commit_root);
2053 fail_chunk_root:
2054         free_extent_buffer(chunk_root->node);
2055         free_extent_buffer(chunk_root->commit_root);
2056 fail_sb_buffer:
2057         btrfs_stop_workers(&fs_info->generic_worker);
2058         btrfs_stop_workers(&fs_info->fixup_workers);
2059         btrfs_stop_workers(&fs_info->delalloc_workers);
2060         btrfs_stop_workers(&fs_info->workers);
2061         btrfs_stop_workers(&fs_info->endio_workers);
2062         btrfs_stop_workers(&fs_info->endio_meta_workers);
2063         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2064         btrfs_stop_workers(&fs_info->endio_write_workers);
2065         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2066         btrfs_stop_workers(&fs_info->submit_workers);
2067 fail_iput:
2068         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2069         iput(fs_info->btree_inode);
2070
2071         btrfs_close_devices(fs_info->fs_devices);
2072         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2073 fail_bdi:
2074         bdi_destroy(&fs_info->bdi);
2075 fail_srcu:
2076         cleanup_srcu_struct(&fs_info->subvol_srcu);
2077 fail:
2078         kfree(extent_root);
2079         kfree(tree_root);
2080         kfree(fs_info);
2081         kfree(chunk_root);
2082         kfree(dev_root);
2083         kfree(csum_root);
2084         return ERR_PTR(err);
2085 }
2086
2087 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2088 {
2089         char b[BDEVNAME_SIZE];
2090
2091         if (uptodate) {
2092                 set_buffer_uptodate(bh);
2093         } else {
2094                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
2095                         printk(KERN_WARNING "lost page write due to "
2096                                         "I/O error on %s\n",
2097                                        bdevname(bh->b_bdev, b));
2098                 }
2099                 /* note, we dont' set_buffer_write_io_error because we have
2100                  * our own ways of dealing with the IO errors
2101                  */
2102                 clear_buffer_uptodate(bh);
2103         }
2104         unlock_buffer(bh);
2105         put_bh(bh);
2106 }
2107
2108 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2109 {
2110         struct buffer_head *bh;
2111         struct buffer_head *latest = NULL;
2112         struct btrfs_super_block *super;
2113         int i;
2114         u64 transid = 0;
2115         u64 bytenr;
2116
2117         /* we would like to check all the supers, but that would make
2118          * a btrfs mount succeed after a mkfs from a different FS.
2119          * So, we need to add a special mount option to scan for
2120          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2121          */
2122         for (i = 0; i < 1; i++) {
2123                 bytenr = btrfs_sb_offset(i);
2124                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2125                         break;
2126                 bh = __bread(bdev, bytenr / 4096, 4096);
2127                 if (!bh)
2128                         continue;
2129
2130                 super = (struct btrfs_super_block *)bh->b_data;
2131                 if (btrfs_super_bytenr(super) != bytenr ||
2132                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2133                             sizeof(super->magic))) {
2134                         brelse(bh);
2135                         continue;
2136                 }
2137
2138                 if (!latest || btrfs_super_generation(super) > transid) {
2139                         brelse(latest);
2140                         latest = bh;
2141                         transid = btrfs_super_generation(super);
2142                 } else {
2143                         brelse(bh);
2144                 }
2145         }
2146         return latest;
2147 }
2148
2149 /*
2150  * this should be called twice, once with wait == 0 and
2151  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2152  * we write are pinned.
2153  *
2154  * They are released when wait == 1 is done.
2155  * max_mirrors must be the same for both runs, and it indicates how
2156  * many supers on this one device should be written.
2157  *
2158  * max_mirrors == 0 means to write them all.
2159  */
2160 static int write_dev_supers(struct btrfs_device *device,
2161                             struct btrfs_super_block *sb,
2162                             int do_barriers, int wait, int max_mirrors)
2163 {
2164         struct buffer_head *bh;
2165         int i;
2166         int ret;
2167         int errors = 0;
2168         u32 crc;
2169         u64 bytenr;
2170         int last_barrier = 0;
2171
2172         if (max_mirrors == 0)
2173                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2174
2175         /* make sure only the last submit_bh does a barrier */
2176         if (do_barriers) {
2177                 for (i = 0; i < max_mirrors; i++) {
2178                         bytenr = btrfs_sb_offset(i);
2179                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2180                             device->total_bytes)
2181                                 break;
2182                         last_barrier = i;
2183                 }
2184         }
2185
2186         for (i = 0; i < max_mirrors; i++) {
2187                 bytenr = btrfs_sb_offset(i);
2188                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2189                         break;
2190
2191                 if (wait) {
2192                         bh = __find_get_block(device->bdev, bytenr / 4096,
2193                                               BTRFS_SUPER_INFO_SIZE);
2194                         BUG_ON(!bh);
2195                         wait_on_buffer(bh);
2196                         if (!buffer_uptodate(bh))
2197                                 errors++;
2198
2199                         /* drop our reference */
2200                         brelse(bh);
2201
2202                         /* drop the reference from the wait == 0 run */
2203                         brelse(bh);
2204                         continue;
2205                 } else {
2206                         btrfs_set_super_bytenr(sb, bytenr);
2207
2208                         crc = ~(u32)0;
2209                         crc = btrfs_csum_data(NULL, (char *)sb +
2210                                               BTRFS_CSUM_SIZE, crc,
2211                                               BTRFS_SUPER_INFO_SIZE -
2212                                               BTRFS_CSUM_SIZE);
2213                         btrfs_csum_final(crc, sb->csum);
2214
2215                         /*
2216                          * one reference for us, and we leave it for the
2217                          * caller
2218                          */
2219                         bh = __getblk(device->bdev, bytenr / 4096,
2220                                       BTRFS_SUPER_INFO_SIZE);
2221                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2222
2223                         /* one reference for submit_bh */
2224                         get_bh(bh);
2225
2226                         set_buffer_uptodate(bh);
2227                         lock_buffer(bh);
2228                         bh->b_end_io = btrfs_end_buffer_write_sync;
2229                 }
2230
2231                 if (i == last_barrier && do_barriers && device->barriers) {
2232                         ret = submit_bh(WRITE_BARRIER, bh);
2233                         if (ret == -EOPNOTSUPP) {
2234                                 printk("btrfs: disabling barriers on dev %s\n",
2235                                        device->name);
2236                                 set_buffer_uptodate(bh);
2237                                 device->barriers = 0;
2238                                 /* one reference for submit_bh */
2239                                 get_bh(bh);
2240                                 lock_buffer(bh);
2241                                 ret = submit_bh(WRITE_SYNC, bh);
2242                         }
2243                 } else {
2244                         ret = submit_bh(WRITE_SYNC, bh);
2245                 }
2246
2247                 if (ret)
2248                         errors++;
2249         }
2250         return errors < i ? 0 : -1;
2251 }
2252
2253 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2254 {
2255         struct list_head *head;
2256         struct btrfs_device *dev;
2257         struct btrfs_super_block *sb;
2258         struct btrfs_dev_item *dev_item;
2259         int ret;
2260         int do_barriers;
2261         int max_errors;
2262         int total_errors = 0;
2263         u64 flags;
2264
2265         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2266         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2267
2268         sb = &root->fs_info->super_for_commit;
2269         dev_item = &sb->dev_item;
2270
2271         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2272         head = &root->fs_info->fs_devices->devices;
2273         list_for_each_entry(dev, head, dev_list) {
2274                 if (!dev->bdev) {
2275                         total_errors++;
2276                         continue;
2277                 }
2278                 if (!dev->in_fs_metadata || !dev->writeable)
2279                         continue;
2280
2281                 btrfs_set_stack_device_generation(dev_item, 0);
2282                 btrfs_set_stack_device_type(dev_item, dev->type);
2283                 btrfs_set_stack_device_id(dev_item, dev->devid);
2284                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2285                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2286                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2287                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2288                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2289                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2290                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2291
2292                 flags = btrfs_super_flags(sb);
2293                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2294
2295                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2296                 if (ret)
2297                         total_errors++;
2298         }
2299         if (total_errors > max_errors) {
2300                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2301                        total_errors);
2302                 BUG();
2303         }
2304
2305         total_errors = 0;
2306         list_for_each_entry(dev, head, dev_list) {
2307                 if (!dev->bdev)
2308                         continue;
2309                 if (!dev->in_fs_metadata || !dev->writeable)
2310                         continue;
2311
2312                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2313                 if (ret)
2314                         total_errors++;
2315         }
2316         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2317         if (total_errors > max_errors) {
2318                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2319                        total_errors);
2320                 BUG();
2321         }
2322         return 0;
2323 }
2324
2325 int write_ctree_super(struct btrfs_trans_handle *trans,
2326                       struct btrfs_root *root, int max_mirrors)
2327 {
2328         int ret;
2329
2330         ret = write_all_supers(root, max_mirrors);
2331         return ret;
2332 }
2333
2334 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2335 {
2336         spin_lock(&fs_info->fs_roots_radix_lock);
2337         radix_tree_delete(&fs_info->fs_roots_radix,
2338                           (unsigned long)root->root_key.objectid);
2339         spin_unlock(&fs_info->fs_roots_radix_lock);
2340
2341         if (btrfs_root_refs(&root->root_item) == 0)
2342                 synchronize_srcu(&fs_info->subvol_srcu);
2343
2344         free_fs_root(root);
2345         return 0;
2346 }
2347
2348 static void free_fs_root(struct btrfs_root *root)
2349 {
2350         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2351         if (root->anon_super.s_dev) {
2352                 down_write(&root->anon_super.s_umount);
2353                 kill_anon_super(&root->anon_super);
2354         }
2355         free_extent_buffer(root->node);
2356         free_extent_buffer(root->commit_root);
2357         kfree(root->name);
2358         kfree(root);
2359 }
2360
2361 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2362 {
2363         int ret;
2364         struct btrfs_root *gang[8];
2365         int i;
2366
2367         while (!list_empty(&fs_info->dead_roots)) {
2368                 gang[0] = list_entry(fs_info->dead_roots.next,
2369                                      struct btrfs_root, root_list);
2370                 list_del(&gang[0]->root_list);
2371
2372                 if (gang[0]->in_radix) {
2373                         btrfs_free_fs_root(fs_info, gang[0]);
2374                 } else {
2375                         free_extent_buffer(gang[0]->node);
2376                         free_extent_buffer(gang[0]->commit_root);
2377                         kfree(gang[0]);
2378                 }
2379         }
2380
2381         while (1) {
2382                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2383                                              (void **)gang, 0,
2384                                              ARRAY_SIZE(gang));
2385                 if (!ret)
2386                         break;
2387                 for (i = 0; i < ret; i++)
2388                         btrfs_free_fs_root(fs_info, gang[i]);
2389         }
2390         return 0;
2391 }
2392
2393 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2394 {
2395         u64 root_objectid = 0;
2396         struct btrfs_root *gang[8];
2397         int i;
2398         int ret;
2399
2400         while (1) {
2401                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2402                                              (void **)gang, root_objectid,
2403                                              ARRAY_SIZE(gang));
2404                 if (!ret)
2405                         break;
2406
2407                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2408                 for (i = 0; i < ret; i++) {
2409                         root_objectid = gang[i]->root_key.objectid;
2410                         btrfs_orphan_cleanup(gang[i]);
2411                 }
2412                 root_objectid++;
2413         }
2414         return 0;
2415 }
2416
2417 int btrfs_commit_super(struct btrfs_root *root)
2418 {
2419         struct btrfs_trans_handle *trans;
2420         int ret;
2421
2422         mutex_lock(&root->fs_info->cleaner_mutex);
2423         btrfs_run_delayed_iputs(root);
2424         btrfs_clean_old_snapshots(root);
2425         mutex_unlock(&root->fs_info->cleaner_mutex);
2426
2427         /* wait until ongoing cleanup work done */
2428         down_write(&root->fs_info->cleanup_work_sem);
2429         up_write(&root->fs_info->cleanup_work_sem);
2430
2431         trans = btrfs_join_transaction(root, 1);
2432         ret = btrfs_commit_transaction(trans, root);
2433         BUG_ON(ret);
2434         /* run commit again to drop the original snapshot */
2435         trans = btrfs_join_transaction(root, 1);
2436         btrfs_commit_transaction(trans, root);
2437         ret = btrfs_write_and_wait_transaction(NULL, root);
2438         BUG_ON(ret);
2439
2440         ret = write_ctree_super(NULL, root, 0);
2441         return ret;
2442 }
2443
2444 int close_ctree(struct btrfs_root *root)
2445 {
2446         struct btrfs_fs_info *fs_info = root->fs_info;
2447         int ret;
2448
2449         fs_info->closing = 1;
2450         smp_mb();
2451
2452         btrfs_put_block_group_cache(fs_info);
2453         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2454                 ret =  btrfs_commit_super(root);
2455                 if (ret)
2456                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2457         }
2458
2459         kthread_stop(root->fs_info->transaction_kthread);
2460         kthread_stop(root->fs_info->cleaner_kthread);
2461
2462         fs_info->closing = 2;
2463         smp_mb();
2464
2465         if (fs_info->delalloc_bytes) {
2466                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2467                        (unsigned long long)fs_info->delalloc_bytes);
2468         }
2469         if (fs_info->total_ref_cache_size) {
2470                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2471                        (unsigned long long)fs_info->total_ref_cache_size);
2472         }
2473
2474         free_extent_buffer(fs_info->extent_root->node);
2475         free_extent_buffer(fs_info->extent_root->commit_root);
2476         free_extent_buffer(fs_info->tree_root->node);
2477         free_extent_buffer(fs_info->tree_root->commit_root);
2478         free_extent_buffer(root->fs_info->chunk_root->node);
2479         free_extent_buffer(root->fs_info->chunk_root->commit_root);
2480         free_extent_buffer(root->fs_info->dev_root->node);
2481         free_extent_buffer(root->fs_info->dev_root->commit_root);
2482         free_extent_buffer(root->fs_info->csum_root->node);
2483         free_extent_buffer(root->fs_info->csum_root->commit_root);
2484
2485         btrfs_free_block_groups(root->fs_info);
2486
2487         del_fs_roots(fs_info);
2488
2489         iput(fs_info->btree_inode);
2490
2491         btrfs_stop_workers(&fs_info->generic_worker);
2492         btrfs_stop_workers(&fs_info->fixup_workers);
2493         btrfs_stop_workers(&fs_info->delalloc_workers);
2494         btrfs_stop_workers(&fs_info->workers);
2495         btrfs_stop_workers(&fs_info->endio_workers);
2496         btrfs_stop_workers(&fs_info->endio_meta_workers);
2497         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2498         btrfs_stop_workers(&fs_info->endio_write_workers);
2499         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2500         btrfs_stop_workers(&fs_info->submit_workers);
2501
2502         btrfs_close_devices(fs_info->fs_devices);
2503         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2504
2505         bdi_destroy(&fs_info->bdi);
2506         cleanup_srcu_struct(&fs_info->subvol_srcu);
2507
2508         kfree(fs_info->extent_root);
2509         kfree(fs_info->tree_root);
2510         kfree(fs_info->chunk_root);
2511         kfree(fs_info->dev_root);
2512         kfree(fs_info->csum_root);
2513         return 0;
2514 }
2515
2516 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2517 {
2518         int ret;
2519         struct inode *btree_inode = buf->first_page->mapping->host;
2520
2521         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2522                                      NULL);
2523         if (!ret)
2524                 return ret;
2525
2526         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2527                                     parent_transid);
2528         return !ret;
2529 }
2530
2531 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2532 {
2533         struct inode *btree_inode = buf->first_page->mapping->host;
2534         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2535                                           buf);
2536 }
2537
2538 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2539 {
2540         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2541         u64 transid = btrfs_header_generation(buf);
2542         struct inode *btree_inode = root->fs_info->btree_inode;
2543         int was_dirty;
2544
2545         btrfs_assert_tree_locked(buf);
2546         if (transid != root->fs_info->generation) {
2547                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2548                        "found %llu running %llu\n",
2549                         (unsigned long long)buf->start,
2550                         (unsigned long long)transid,
2551                         (unsigned long long)root->fs_info->generation);
2552                 WARN_ON(1);
2553         }
2554         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2555                                             buf);
2556         if (!was_dirty) {
2557                 spin_lock(&root->fs_info->delalloc_lock);
2558                 root->fs_info->dirty_metadata_bytes += buf->len;
2559                 spin_unlock(&root->fs_info->delalloc_lock);
2560         }
2561 }
2562
2563 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2564 {
2565         /*
2566          * looks as though older kernels can get into trouble with
2567          * this code, they end up stuck in balance_dirty_pages forever
2568          */
2569         u64 num_dirty;
2570         unsigned long thresh = 32 * 1024 * 1024;
2571
2572         if (current->flags & PF_MEMALLOC)
2573                 return;
2574
2575         num_dirty = root->fs_info->dirty_metadata_bytes;
2576
2577         if (num_dirty > thresh) {
2578                 balance_dirty_pages_ratelimited_nr(
2579                                    root->fs_info->btree_inode->i_mapping, 1);
2580         }
2581         return;
2582 }
2583
2584 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2585 {
2586         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2587         int ret;
2588         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2589         if (ret == 0)
2590                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2591         return ret;
2592 }
2593
2594 int btree_lock_page_hook(struct page *page)
2595 {
2596         struct inode *inode = page->mapping->host;
2597         struct btrfs_root *root = BTRFS_I(inode)->root;
2598         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2599         struct extent_buffer *eb;
2600         unsigned long len;
2601         u64 bytenr = page_offset(page);
2602
2603         if (page->private == EXTENT_PAGE_PRIVATE)
2604                 goto out;
2605
2606         len = page->private >> 2;
2607         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2608         if (!eb)
2609                 goto out;
2610
2611         btrfs_tree_lock(eb);
2612         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2613
2614         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2615                 spin_lock(&root->fs_info->delalloc_lock);
2616                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2617                         root->fs_info->dirty_metadata_bytes -= eb->len;
2618                 else
2619                         WARN_ON(1);
2620                 spin_unlock(&root->fs_info->delalloc_lock);
2621         }
2622
2623         btrfs_tree_unlock(eb);
2624         free_extent_buffer(eb);
2625 out:
2626         lock_page(page);
2627         return 0;
2628 }
2629
2630 static struct extent_io_ops btree_extent_io_ops = {
2631         .write_cache_pages_lock_hook = btree_lock_page_hook,
2632         .readpage_end_io_hook = btree_readpage_end_io_hook,
2633         .submit_bio_hook = btree_submit_bio_hook,
2634         /* note we're sharing with inode.c for the merge bio hook */
2635         .merge_bio_hook = btrfs_merge_bio_hook,
2636 };