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 static int btree_migratepage(struct address_space *mapping,
 700                         struct page *newpage, struct page *page)
 701 {
 702         /*
 703          * we can't safely write a btree page from here,
 704          * we haven't done the locking hook
 705          */
 706         if (PageDirty(page))
 707                 return -EAGAIN;
 708         /*
 709          * Buffers may be managed in a filesystem specific way.
 710          * We must have no buffers or drop them.
 711          */
 712         if (page_has_private(page) &&
 713             !try_to_release_page(page, GFP_KERNEL))
 714                 return -EAGAIN;
 715 #ifdef CONFIG_MIGRATION
 716         return migrate_page(mapping, newpage, page);
 717 #else
 718         return -ENOSYS;
 719 #endif
 720 }
 721
 722 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 723 {
 724         struct extent_io_tree *tree;
 725         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 726         struct extent_buffer *eb;
 727         int was_dirty;
 728
 729         tree = &BTRFS_I(page->mapping->host)->io_tree;
 730         if (!(current->flags & PF_MEMALLOC)) {
 731                 return extent_write_full_page(tree, page,
 732                                               btree_get_extent, wbc);
 733         }
 734
 735         redirty_page_for_writepage(wbc, page);
 736         eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
 737         WARN_ON(!eb);
 738
 739         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
 740         if (!was_dirty) {
 741                 spin_lock(&root->fs_info->delalloc_lock);
 742                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
 743                 spin_unlock(&root->fs_info->delalloc_lock);
 744         }
 745         free_extent_buffer(eb);
 746
 747         unlock_page(page);
 748         return 0;
 749 }
 750
 751 static int btree_writepages(struct address_space *mapping,
 752                             struct writeback_control *wbc)
 753 {
 754         struct extent_io_tree *tree;
 755         tree = &BTRFS_I(mapping->host)->io_tree;
 756         if (wbc->sync_mode == WB_SYNC_NONE) {
 757                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
 758                 u64 num_dirty;
 759                 unsigned long thresh = 32 * 1024 * 1024;
 760
 761                 if (wbc->for_kupdate)
 762                         return 0;
 763
 764                 /* this is a bit racy, but that's ok */
 765                 num_dirty = root->fs_info->dirty_metadata_bytes;
 766                 if (num_dirty < thresh)
 767                         return 0;
 768         }
 769         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 770 }
 771
 772 static int btree_readpage(struct file *file, struct page *page)
 773 {
 774         struct extent_io_tree *tree;
 775         tree = &BTRFS_I(page->mapping->host)->io_tree;
 776         return extent_read_full_page(tree, page, btree_get_extent);
 777 }
 778
 779 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 780 {
 781         struct extent_io_tree *tree;
 782         struct extent_map_tree *map;
 783         int ret;
 784
 785         if (PageWriteback(page) || PageDirty(page))
 786                 return 0;
 787
 788         tree = &BTRFS_I(page->mapping->host)->io_tree;
 789         map = &BTRFS_I(page->mapping->host)->extent_tree;
 790
 791         ret = try_release_extent_state(map, tree, page, gfp_flags);
 792         if (!ret)
 793                 return 0;
 794
 795         ret = try_release_extent_buffer(tree, page);
 796         if (ret == 1) {
 797                 ClearPagePrivate(page);
 798                 set_page_private(page, 0);
 799                 page_cache_release(page);
 800         }
 801
 802         return ret;
 803 }
 804
 805 static void btree_invalidatepage(struct page *page, unsigned long offset)
 806 {
 807         struct extent_io_tree *tree;
 808         tree = &BTRFS_I(page->mapping->host)->io_tree;
 809         extent_invalidatepage(tree, page, offset);
 810         btree_releasepage(page, GFP_NOFS);
 811         if (PagePrivate(page)) {
 812                 printk(KERN_WARNING "btrfs warning page private not zero "
 813                        "on page %llu\n", (unsigned long long)page_offset(page));
 814                 ClearPagePrivate(page);
 815                 set_page_private(page, 0);
 816                 page_cache_release(page);
 817         }
 818 }
 819
 820 static const struct address_space_operations btree_aops = {
 821         .readpage       = btree_readpage,
 822         .writepage      = btree_writepage,
 823         .writepages     = btree_writepages,
 824         .releasepage    = btree_releasepage,
 825         .invalidatepage = btree_invalidatepage,
 826         .sync_page      = block_sync_page,
 827 #ifdef CONFIG_MIGRATION
 828         .migratepage    = btree_migratepage,
 829 #endif
 830 };
 831
 832 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 833                          u64 parent_transid)
 834 {
 835         struct extent_buffer *buf = NULL;
 836         struct inode *btree_inode = root->fs_info->btree_inode;
 837         int ret = 0;
 838
 839         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 840         if (!buf)
 841                 return 0;
 842         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 843                                  buf, 0, 0, btree_get_extent, 0);
 844         free_extent_buffer(buf);
 845         return ret;
 846 }
 847
 848 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 849                                             u64 bytenr, u32 blocksize)
 850 {
 851         struct inode *btree_inode = root->fs_info->btree_inode;
 852         struct extent_buffer *eb;
 853         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 854                                 bytenr, blocksize, GFP_NOFS);
 855         return eb;
 856 }
 857
 858 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 859                                                  u64 bytenr, u32 blocksize)
 860 {
 861         struct inode *btree_inode = root->fs_info->btree_inode;
 862         struct extent_buffer *eb;
 863
 864         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 865                                  bytenr, blocksize, NULL, GFP_NOFS);
 866         return eb;
 867 }
 868
 869
 870 int btrfs_write_tree_block(struct extent_buffer *buf)
 871 {
 872         return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
 873                                         buf->start + buf->len - 1);
 874 }
 875
 876 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 877 {
 878         return filemap_fdatawait_range(buf->first_page->mapping,
 879                                        buf->start, buf->start + buf->len - 1);
 880 }
 881
 882 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 883                                       u32 blocksize, u64 parent_transid)
 884 {
 885         struct extent_buffer *buf = NULL;
 886         int ret;
 887
 888         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 889         if (!buf)
 890                 return NULL;
 891
 892         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 893
 894         if (ret == 0)
 895                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
 896         return buf;
 897
 898 }
 899
 900 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 901                      struct extent_buffer *buf)
 902 {
 903         struct inode *btree_inode = root->fs_info->btree_inode;
 904         if (btrfs_header_generation(buf) ==
 905             root->fs_info->running_transaction->transid) {
 906                 btrfs_assert_tree_locked(buf);
 907
 908                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
 909                         spin_lock(&root->fs_info->delalloc_lock);
 910                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
 911                                 root->fs_info->dirty_metadata_bytes -= buf->len;
 912                         else
 913                                 WARN_ON(1);
 914                         spin_unlock(&root->fs_info->delalloc_lock);
 915                 }
 916
 917                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
 918                 btrfs_set_lock_blocking(buf);
 919                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 920                                           buf);
 921         }
 922         return 0;
 923 }
 924
 925 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 926                         u32 stripesize, struct btrfs_root *root,
 927                         struct btrfs_fs_info *fs_info,
 928                         u64 objectid)
 929 {
 930         root->node = NULL;
 931         root->commit_root = NULL;
 932         root->sectorsize = sectorsize;
 933         root->nodesize = nodesize;
 934         root->leafsize = leafsize;
 935         root->stripesize = stripesize;
 936         root->ref_cows = 0;
 937         root->track_dirty = 0;
 938         root->in_radix = 0;
 939         root->orphan_item_inserted = 0;
 940         root->orphan_cleanup_state = 0;
 941
 942         root->fs_info = fs_info;
 943         root->objectid = objectid;
 944         root->last_trans = 0;
 945         root->highest_objectid = 0;
 946         root->name = NULL;
 947         root->in_sysfs = 0;
 948         root->inode_tree = RB_ROOT;
 949         root->block_rsv = NULL;
 950         root->orphan_block_rsv = NULL;
 951
 952         INIT_LIST_HEAD(&root->dirty_list);
 953         INIT_LIST_HEAD(&root->orphan_list);
 954         INIT_LIST_HEAD(&root->root_list);
 955         spin_lock_init(&root->node_lock);
 956         spin_lock_init(&root->orphan_lock);
 957         spin_lock_init(&root->inode_lock);
 958         spin_lock_init(&root->accounting_lock);
 959         mutex_init(&root->objectid_mutex);
 960         mutex_init(&root->log_mutex);
 961         init_waitqueue_head(&root->log_writer_wait);
 962         init_waitqueue_head(&root->log_commit_wait[0]);
 963         init_waitqueue_head(&root->log_commit_wait[1]);
 964         atomic_set(&root->log_commit[0], 0);
 965         atomic_set(&root->log_commit[1], 0);
 966         atomic_set(&root->log_writers, 0);
 967         root->log_batch = 0;
 968         root->log_transid = 0;
 969         root->last_log_commit = 0;
 970         extent_io_tree_init(&root->dirty_log_pages,
 971                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 972
 973         memset(&root->root_key, 0, sizeof(root->root_key));
 974         memset(&root->root_item, 0, sizeof(root->root_item));
 975         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 976         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 977         root->defrag_trans_start = fs_info->generation;
 978         init_completion(&root->kobj_unregister);
 979         root->defrag_running = 0;
 980         root->root_key.objectid = objectid;
 981         root->anon_super.s_root = NULL;
 982         root->anon_super.s_dev = 0;
 983         INIT_LIST_HEAD(&root->anon_super.s_list);
 984         INIT_LIST_HEAD(&root->anon_super.s_instances);
 985         init_rwsem(&root->anon_super.s_umount);
 986
 987         return 0;
 988 }
 989
 990 static int find_and_setup_root(struct btrfs_root *tree_root,
 991                                struct btrfs_fs_info *fs_info,
 992                                u64 objectid,
 993                                struct btrfs_root *root)
 994 {
 995         int ret;
 996         u32 blocksize;
 997         u64 generation;
 998
 999         __setup_root(tree_root->nodesize, tree_root->leafsize,
1000                      tree_root->sectorsize, tree_root->stripesize,
1001                      root, fs_info, objectid);
1002         ret = btrfs_find_last_root(tree_root, objectid,
1003                                    &root->root_item, &root->root_key);
1004         if (ret > 0)
1005                 return -ENOENT;
1006         BUG_ON(ret);
1007
1008         generation = btrfs_root_generation(&root->root_item);
1009         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1010         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1011                                      blocksize, generation);
1012         BUG_ON(!root->node);
1013         root->commit_root = btrfs_root_node(root);
1014         return 0;
1015 }
1016
1017 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1018                                          struct btrfs_fs_info *fs_info)
1019 {
1020         struct btrfs_root *root;
1021         struct btrfs_root *tree_root = fs_info->tree_root;
1022         struct extent_buffer *leaf;
1023
1024         root = kzalloc(sizeof(*root), GFP_NOFS);
1025         if (!root)
1026                 return ERR_PTR(-ENOMEM);
1027
1028         __setup_root(tree_root->nodesize, tree_root->leafsize,
1029                      tree_root->sectorsize, tree_root->stripesize,
1030                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1031
1032         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1033         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1034         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1035         /*
1036          * log trees do not get reference counted because they go away
1037          * before a real commit is actually done.  They do store pointers
1038          * to file data extents, and those reference counts still get
1039          * updated (along with back refs to the log tree).
1040          */
1041         root->ref_cows = 0;
1042
1043         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1044                                       BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1045         if (IS_ERR(leaf)) {
1046                 kfree(root);
1047                 return ERR_CAST(leaf);
1048         }
1049
1050         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1051         btrfs_set_header_bytenr(leaf, leaf->start);
1052         btrfs_set_header_generation(leaf, trans->transid);
1053         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1054         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1055         root->node = leaf;
1056
1057         write_extent_buffer(root->node, root->fs_info->fsid,
1058                             (unsigned long)btrfs_header_fsid(root->node),
1059                             BTRFS_FSID_SIZE);
1060         btrfs_mark_buffer_dirty(root->node);
1061         btrfs_tree_unlock(root->node);
1062         return root;
1063 }
1064
1065 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1066                              struct btrfs_fs_info *fs_info)
1067 {
1068         struct btrfs_root *log_root;
1069
1070         log_root = alloc_log_tree(trans, fs_info);
1071         if (IS_ERR(log_root))
1072                 return PTR_ERR(log_root);
1073         WARN_ON(fs_info->log_root_tree);
1074         fs_info->log_root_tree = log_root;
1075         return 0;
1076 }
1077
1078 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1079                        struct btrfs_root *root)
1080 {
1081         struct btrfs_root *log_root;
1082         struct btrfs_inode_item *inode_item;
1083
1084         log_root = alloc_log_tree(trans, root->fs_info);
1085         if (IS_ERR(log_root))
1086                 return PTR_ERR(log_root);
1087
1088         log_root->last_trans = trans->transid;
1089         log_root->root_key.offset = root->root_key.objectid;
1090
1091         inode_item = &log_root->root_item.inode;
1092         inode_item->generation = cpu_to_le64(1);
1093         inode_item->size = cpu_to_le64(3);
1094         inode_item->nlink = cpu_to_le32(1);
1095         inode_item->nbytes = cpu_to_le64(root->leafsize);
1096         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1097
1098         btrfs_set_root_node(&log_root->root_item, log_root->node);
1099
1100         WARN_ON(root->log_root);
1101         root->log_root = log_root;
1102         root->log_transid = 0;
1103         root->last_log_commit = 0;
1104         return 0;
1105 }
1106
1107 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1108                                                struct btrfs_key *location)
1109 {
1110         struct btrfs_root *root;
1111         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1112         struct btrfs_path *path;
1113         struct extent_buffer *l;
1114         u64 generation;
1115         u32 blocksize;
1116         int ret = 0;
1117
1118         root = kzalloc(sizeof(*root), GFP_NOFS);
1119         if (!root)
1120                 return ERR_PTR(-ENOMEM);
1121         if (location->offset == (u64)-1) {
1122                 ret = find_and_setup_root(tree_root, fs_info,
1123                                           location->objectid, root);
1124                 if (ret) {
1125                         kfree(root);
1126                         return ERR_PTR(ret);
1127                 }
1128                 goto out;
1129         }
1130
1131         __setup_root(tree_root->nodesize, tree_root->leafsize,
1132                      tree_root->sectorsize, tree_root->stripesize,
1133                      root, fs_info, location->objectid);
1134
1135         path = btrfs_alloc_path();
1136         BUG_ON(!path);
1137         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1138         if (ret == 0) {
1139                 l = path->nodes[0];
1140                 read_extent_buffer(l, &root->root_item,
1141                                 btrfs_item_ptr_offset(l, path->slots[0]),
1142                                 sizeof(root->root_item));
1143                 memcpy(&root->root_key, location, sizeof(*location));
1144         }
1145         btrfs_free_path(path);
1146         if (ret) {
1147                 if (ret > 0)
1148                         ret = -ENOENT;
1149                 return ERR_PTR(ret);
1150         }
1151
1152         generation = btrfs_root_generation(&root->root_item);
1153         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1154         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1155                                      blocksize, generation);
1156         root->commit_root = btrfs_root_node(root);
1157         BUG_ON(!root->node);
1158 out:
1159         if (location->objectid != BTRFS_TREE_LOG_OBJECTID)
1160                 root->ref_cows = 1;
1161
1162         return root;
1163 }
1164
1165 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1166                                         u64 root_objectid)
1167 {
1168         struct btrfs_root *root;
1169
1170         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1171                 return fs_info->tree_root;
1172         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1173                 return fs_info->extent_root;
1174
1175         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1176                                  (unsigned long)root_objectid);
1177         return root;
1178 }
1179
1180 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1181                                               struct btrfs_key *location)
1182 {
1183         struct btrfs_root *root;
1184         int ret;
1185
1186         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1187                 return fs_info->tree_root;
1188         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1189                 return fs_info->extent_root;
1190         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1191                 return fs_info->chunk_root;
1192         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1193                 return fs_info->dev_root;
1194         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1195                 return fs_info->csum_root;
1196 again:
1197         spin_lock(&fs_info->fs_roots_radix_lock);
1198         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1199                                  (unsigned long)location->objectid);
1200         spin_unlock(&fs_info->fs_roots_radix_lock);
1201         if (root)
1202                 return root;
1203
1204         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1205         if (IS_ERR(root))
1206                 return root;
1207
1208         set_anon_super(&root->anon_super, NULL);
1209
1210         if (btrfs_root_refs(&root->root_item) == 0) {
1211                 ret = -ENOENT;
1212                 goto fail;
1213         }
1214
1215         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1216         if (ret < 0)
1217                 goto fail;
1218         if (ret == 0)
1219                 root->orphan_item_inserted = 1;
1220
1221         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1222         if (ret)
1223                 goto fail;
1224
1225         spin_lock(&fs_info->fs_roots_radix_lock);
1226         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1227                                 (unsigned long)root->root_key.objectid,
1228                                 root);
1229         if (ret == 0)
1230                 root->in_radix = 1;
1231
1232         spin_unlock(&fs_info->fs_roots_radix_lock);
1233         radix_tree_preload_end();
1234         if (ret) {
1235                 if (ret == -EEXIST) {
1236                         free_fs_root(root);
1237                         goto again;
1238                 }
1239                 goto fail;
1240         }
1241
1242         ret = btrfs_find_dead_roots(fs_info->tree_root,
1243                                     root->root_key.objectid);
1244         WARN_ON(ret);
1245         return root;
1246 fail:
1247         free_fs_root(root);
1248         return ERR_PTR(ret);
1249 }
1250
1251 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1252                                       struct btrfs_key *location,
1253                                       const char *name, int namelen)
1254 {
1255         return btrfs_read_fs_root_no_name(fs_info, location);
1256 #if 0
1257         struct btrfs_root *root;
1258         int ret;
1259
1260         root = btrfs_read_fs_root_no_name(fs_info, location);
1261         if (!root)
1262                 return NULL;
1263
1264         if (root->in_sysfs)
1265                 return root;
1266
1267         ret = btrfs_set_root_name(root, name, namelen);
1268         if (ret) {
1269                 free_extent_buffer(root->node);
1270                 kfree(root);
1271                 return ERR_PTR(ret);
1272         }
1273
1274         ret = btrfs_sysfs_add_root(root);
1275         if (ret) {
1276                 free_extent_buffer(root->node);
1277                 kfree(root->name);
1278                 kfree(root);
1279                 return ERR_PTR(ret);
1280         }
1281         root->in_sysfs = 1;
1282         return root;
1283 #endif
1284 }
1285
1286 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1287 {
1288         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1289         int ret = 0;
1290         struct btrfs_device *device;
1291         struct backing_dev_info *bdi;
1292
1293         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1294                 if (!device->bdev)
1295                         continue;
1296                 bdi = blk_get_backing_dev_info(device->bdev);
1297                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1298                         ret = 1;
1299                         break;
1300                 }
1301         }
1302         return ret;
1303 }
1304
1305 /*
1306  * this unplugs every device on the box, and it is only used when page
1307  * is null
1308  */
1309 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1310 {
1311         struct btrfs_device *device;
1312         struct btrfs_fs_info *info;
1313
1314         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1315         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1316                 if (!device->bdev)
1317                         continue;
1318
1319                 bdi = blk_get_backing_dev_info(device->bdev);
1320                 if (bdi->unplug_io_fn)
1321                         bdi->unplug_io_fn(bdi, page);
1322         }
1323 }
1324
1325 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1326 {
1327         struct inode *inode;
1328         struct extent_map_tree *em_tree;
1329         struct extent_map *em;
1330         struct address_space *mapping;
1331         u64 offset;
1332
1333         /* the generic O_DIRECT read code does this */
1334         if (1 || !page) {
1335                 __unplug_io_fn(bdi, page);
1336                 return;
1337         }
1338
1339         /*
1340          * page->mapping may change at any time.  Get a consistent copy
1341          * and use that for everything below
1342          */
1343         smp_mb();
1344         mapping = page->mapping;
1345         if (!mapping)
1346                 return;
1347
1348         inode = mapping->host;
1349
1350         /*
1351          * don't do the expensive searching for a small number of
1352          * devices
1353          */
1354         if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1355                 __unplug_io_fn(bdi, page);
1356                 return;
1357         }
1358
1359         offset = page_offset(page);
1360
1361         em_tree = &BTRFS_I(inode)->extent_tree;
1362         read_lock(&em_tree->lock);
1363         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1364         read_unlock(&em_tree->lock);
1365         if (!em) {
1366                 __unplug_io_fn(bdi, page);
1367                 return;
1368         }
1369
1370         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1371                 free_extent_map(em);
1372                 __unplug_io_fn(bdi, page);
1373                 return;
1374         }
1375         offset = offset - em->start;
1376         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1377                           em->block_start + offset, page);
1378         free_extent_map(em);
1379 }
1380
1381 /*
1382  * If this fails, caller must call bdi_destroy() to get rid of the
1383  * bdi again.
1384  */
1385 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1386 {
1387         int err;
1388
1389         bdi->capabilities = BDI_CAP_MAP_COPY;
1390         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1391         if (err)
1392                 return err;
1393
1394         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1395         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1396         bdi->unplug_io_data     = info;
1397         bdi->congested_fn       = btrfs_congested_fn;
1398         bdi->congested_data     = info;
1399         return 0;
1400 }
1401
1402 static int bio_ready_for_csum(struct bio *bio)
1403 {
1404         u64 length = 0;
1405         u64 buf_len = 0;
1406         u64 start = 0;
1407         struct page *page;
1408         struct extent_io_tree *io_tree = NULL;
1409         struct bio_vec *bvec;
1410         int i;
1411         int ret;
1412
1413         bio_for_each_segment(bvec, bio, i) {
1414                 page = bvec->bv_page;
1415                 if (page->private == EXTENT_PAGE_PRIVATE) {
1416                         length += bvec->bv_len;
1417                         continue;
1418                 }
1419                 if (!page->private) {
1420                         length += bvec->bv_len;
1421                         continue;
1422                 }
1423                 length = bvec->bv_len;
1424                 buf_len = page->private >> 2;
1425                 start = page_offset(page) + bvec->bv_offset;
1426                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1427         }
1428         /* are we fully contained in this bio? */
1429         if (buf_len <= length)
1430                 return 1;
1431
1432         ret = extent_range_uptodate(io_tree, start + length,
1433                                     start + buf_len - 1);
1434         return ret;
1435 }
1436
1437 /*
1438  * called by the kthread helper functions to finally call the bio end_io
1439  * functions.  This is where read checksum verification actually happens
1440  */
1441 static void end_workqueue_fn(struct btrfs_work *work)
1442 {
1443         struct bio *bio;
1444         struct end_io_wq *end_io_wq;
1445         struct btrfs_fs_info *fs_info;
1446         int error;
1447
1448         end_io_wq = container_of(work, struct end_io_wq, work);
1449         bio = end_io_wq->bio;
1450         fs_info = end_io_wq->info;
1451
1452         /* metadata bio reads are special because the whole tree block must
1453          * be checksummed at once.  This makes sure the entire block is in
1454          * ram and up to date before trying to verify things.  For
1455          * blocksize <= pagesize, it is basically a noop
1456          */
1457         if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1458             !bio_ready_for_csum(bio)) {
1459                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1460                                    &end_io_wq->work);
1461                 return;
1462         }
1463         error = end_io_wq->error;
1464         bio->bi_private = end_io_wq->private;
1465         bio->bi_end_io = end_io_wq->end_io;
1466         kfree(end_io_wq);
1467         bio_endio(bio, error);
1468 }
1469
1470 static int cleaner_kthread(void *arg)
1471 {
1472         struct btrfs_root *root = arg;
1473
1474         do {
1475                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1476
1477                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1478                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1479                         btrfs_run_delayed_iputs(root);
1480                         btrfs_clean_old_snapshots(root);
1481                         mutex_unlock(&root->fs_info->cleaner_mutex);
1482                 }
1483
1484                 if (freezing(current)) {
1485                         refrigerator();
1486                 } else {
1487                         set_current_state(TASK_INTERRUPTIBLE);
1488                         if (!kthread_should_stop())
1489                                 schedule();
1490                         __set_current_state(TASK_RUNNING);
1491                 }
1492         } while (!kthread_should_stop());
1493         return 0;
1494 }
1495
1496 static int transaction_kthread(void *arg)
1497 {
1498         struct btrfs_root *root = arg;
1499         struct btrfs_trans_handle *trans;
1500         struct btrfs_transaction *cur;
1501         u64 transid;
1502         unsigned long now;
1503         unsigned long delay;
1504         int ret;
1505
1506         do {
1507                 delay = HZ * 30;
1508                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1509                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1510
1511                 spin_lock(&root->fs_info->new_trans_lock);
1512                 cur = root->fs_info->running_transaction;
1513                 if (!cur) {
1514                         spin_unlock(&root->fs_info->new_trans_lock);
1515                         goto sleep;
1516                 }
1517
1518                 now = get_seconds();
1519                 if (!cur->blocked &&
1520                     (now < cur->start_time || now - cur->start_time < 30)) {
1521                         spin_unlock(&root->fs_info->new_trans_lock);
1522                         delay = HZ * 5;
1523                         goto sleep;
1524                 }
1525                 transid = cur->transid;
1526                 spin_unlock(&root->fs_info->new_trans_lock);
1527
1528                 trans = btrfs_join_transaction(root, 1);
1529                 if (transid == trans->transid) {
1530                         ret = btrfs_commit_transaction(trans, root);
1531                         BUG_ON(ret);
1532                 } else {
1533                         btrfs_end_transaction(trans, root);
1534                 }
1535 sleep:
1536                 wake_up_process(root->fs_info->cleaner_kthread);
1537                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1538
1539                 if (freezing(current)) {
1540                         refrigerator();
1541                 } else {
1542                         set_current_state(TASK_INTERRUPTIBLE);
1543                         if (!kthread_should_stop() &&
1544                             !btrfs_transaction_blocked(root->fs_info))
1545                                 schedule_timeout(delay);
1546                         __set_current_state(TASK_RUNNING);
1547                 }
1548         } while (!kthread_should_stop());
1549         return 0;
1550 }
1551
1552 struct btrfs_root *open_ctree(struct super_block *sb,
1553                               struct btrfs_fs_devices *fs_devices,
1554                               char *options)
1555 {
1556         u32 sectorsize;
1557         u32 nodesize;
1558         u32 leafsize;
1559         u32 blocksize;
1560         u32 stripesize;
1561         u64 generation;
1562         u64 features;
1563         struct btrfs_key location;
1564         struct buffer_head *bh;
1565         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1566                                                  GFP_NOFS);
1567         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1568                                                  GFP_NOFS);
1569         struct btrfs_root *tree_root = btrfs_sb(sb);
1570         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1571         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1572                                                 GFP_NOFS);
1573         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1574                                               GFP_NOFS);
1575         struct btrfs_root *log_tree_root;
1576
1577         int ret;
1578         int err = -EINVAL;
1579
1580         struct btrfs_super_block *disk_super;
1581
1582         if (!extent_root || !tree_root || !fs_info ||
1583             !chunk_root || !dev_root || !csum_root) {
1584                 err = -ENOMEM;
1585                 goto fail;
1586         }
1587
1588         ret = init_srcu_struct(&fs_info->subvol_srcu);
1589         if (ret) {
1590                 err = ret;
1591                 goto fail;
1592         }
1593
1594         ret = setup_bdi(fs_info, &fs_info->bdi);
1595         if (ret) {
1596                 err = ret;
1597                 goto fail_srcu;
1598         }
1599
1600         fs_info->btree_inode = new_inode(sb);
1601         if (!fs_info->btree_inode) {
1602                 err = -ENOMEM;
1603                 goto fail_bdi;
1604         }
1605
1606         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1607         INIT_LIST_HEAD(&fs_info->trans_list);
1608         INIT_LIST_HEAD(&fs_info->dead_roots);
1609         INIT_LIST_HEAD(&fs_info->delayed_iputs);
1610         INIT_LIST_HEAD(&fs_info->hashers);
1611         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1612         INIT_LIST_HEAD(&fs_info->ordered_operations);
1613         INIT_LIST_HEAD(&fs_info->caching_block_groups);
1614         spin_lock_init(&fs_info->delalloc_lock);
1615         spin_lock_init(&fs_info->new_trans_lock);
1616         spin_lock_init(&fs_info->ref_cache_lock);
1617         spin_lock_init(&fs_info->fs_roots_radix_lock);
1618         spin_lock_init(&fs_info->delayed_iput_lock);
1619
1620         init_completion(&fs_info->kobj_unregister);
1621         fs_info->tree_root = tree_root;
1622         fs_info->extent_root = extent_root;
1623         fs_info->csum_root = csum_root;
1624         fs_info->chunk_root = chunk_root;
1625         fs_info->dev_root = dev_root;
1626         fs_info->fs_devices = fs_devices;
1627         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1628         INIT_LIST_HEAD(&fs_info->space_info);
1629         btrfs_mapping_init(&fs_info->mapping_tree);
1630         btrfs_init_block_rsv(&fs_info->global_block_rsv);
1631         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1632         btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1633         btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1634         btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1635         INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1636         mutex_init(&fs_info->durable_block_rsv_mutex);
1637         atomic_set(&fs_info->nr_async_submits, 0);
1638         atomic_set(&fs_info->async_delalloc_pages, 0);
1639         atomic_set(&fs_info->async_submit_draining, 0);
1640         atomic_set(&fs_info->nr_async_bios, 0);
1641         fs_info->sb = sb;
1642         fs_info->max_inline = 8192 * 1024;
1643         fs_info->metadata_ratio = 0;
1644
1645         fs_info->thread_pool_size = min_t(unsigned long,
1646                                           num_online_cpus() + 2, 8);
1647
1648         INIT_LIST_HEAD(&fs_info->ordered_extents);
1649         spin_lock_init(&fs_info->ordered_extent_lock);
1650
1651         sb->s_blocksize = 4096;
1652         sb->s_blocksize_bits = blksize_bits(4096);
1653         sb->s_bdi = &fs_info->bdi;
1654
1655         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1656         fs_info->btree_inode->i_nlink = 1;
1657         /*
1658          * we set the i_size on the btree inode to the max possible int.
1659          * the real end of the address space is determined by all of
1660          * the devices in the system
1661          */
1662         fs_info->btree_inode->i_size = OFFSET_MAX;
1663         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1664         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1665
1666         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1667         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1668                              fs_info->btree_inode->i_mapping,
1669                              GFP_NOFS);
1670         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1671                              GFP_NOFS);
1672
1673         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1674
1675         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1676         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1677                sizeof(struct btrfs_key));
1678         BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1679         insert_inode_hash(fs_info->btree_inode);
1680
1681         spin_lock_init(&fs_info->block_group_cache_lock);
1682         fs_info->block_group_cache_tree = RB_ROOT;
1683
1684         extent_io_tree_init(&fs_info->freed_extents[0],
1685                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1686         extent_io_tree_init(&fs_info->freed_extents[1],
1687                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1688         fs_info->pinned_extents = &fs_info->freed_extents[0];
1689         fs_info->do_barriers = 1;
1690
1691
1692         mutex_init(&fs_info->trans_mutex);
1693         mutex_init(&fs_info->ordered_operations_mutex);
1694         mutex_init(&fs_info->tree_log_mutex);
1695         mutex_init(&fs_info->chunk_mutex);
1696         mutex_init(&fs_info->transaction_kthread_mutex);
1697         mutex_init(&fs_info->cleaner_mutex);
1698         mutex_init(&fs_info->volume_mutex);
1699         init_rwsem(&fs_info->extent_commit_sem);
1700         init_rwsem(&fs_info->cleanup_work_sem);
1701         init_rwsem(&fs_info->subvol_sem);
1702
1703         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1704         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1705
1706         init_waitqueue_head(&fs_info->transaction_throttle);
1707         init_waitqueue_head(&fs_info->transaction_wait);
1708         init_waitqueue_head(&fs_info->transaction_blocked_wait);
1709         init_waitqueue_head(&fs_info->async_submit_wait);
1710
1711         __setup_root(4096, 4096, 4096, 4096, tree_root,
1712                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1713
1714         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1715         if (!bh)
1716                 goto fail_iput;
1717
1718         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1719         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1720                sizeof(fs_info->super_for_commit));
1721         brelse(bh);
1722
1723         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1724
1725         disk_super = &fs_info->super_copy;
1726         if (!btrfs_super_root(disk_super))
1727                 goto fail_iput;
1728
1729         ret = btrfs_parse_options(tree_root, options);
1730         if (ret) {
1731                 err = ret;
1732                 goto fail_iput;
1733         }
1734
1735         features = btrfs_super_incompat_flags(disk_super) &
1736                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1737         if (features) {
1738                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1739                        "unsupported optional features (%Lx).\n",
1740                        (unsigned long long)features);
1741                 err = -EINVAL;
1742                 goto fail_iput;
1743         }
1744
1745         features = btrfs_super_incompat_flags(disk_super);
1746         if (!(features & BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF)) {
1747                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1748                 btrfs_set_super_incompat_flags(disk_super, features);
1749         }
1750
1751         features = btrfs_super_compat_ro_flags(disk_super) &
1752                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1753         if (!(sb->s_flags & MS_RDONLY) && features) {
1754                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1755                        "unsupported option features (%Lx).\n",
1756                        (unsigned long long)features);
1757                 err = -EINVAL;
1758                 goto fail_iput;
1759         }
1760
1761         btrfs_init_workers(&fs_info->generic_worker,
1762                            "genwork", 1, NULL);
1763
1764         btrfs_init_workers(&fs_info->workers, "worker",
1765                            fs_info->thread_pool_size,
1766                            &fs_info->generic_worker);
1767
1768         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1769                            fs_info->thread_pool_size,
1770                            &fs_info->generic_worker);
1771
1772         btrfs_init_workers(&fs_info->submit_workers, "submit",
1773                            min_t(u64, fs_devices->num_devices,
1774                            fs_info->thread_pool_size),
1775                            &fs_info->generic_worker);
1776
1777         /* a higher idle thresh on the submit workers makes it much more
1778          * likely that bios will be send down in a sane order to the
1779          * devices
1780          */
1781         fs_info->submit_workers.idle_thresh = 64;
1782
1783         fs_info->workers.idle_thresh = 16;
1784         fs_info->workers.ordered = 1;
1785
1786         fs_info->delalloc_workers.idle_thresh = 2;
1787         fs_info->delalloc_workers.ordered = 1;
1788
1789         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1790                            &fs_info->generic_worker);
1791         btrfs_init_workers(&fs_info->endio_workers, "endio",
1792                            fs_info->thread_pool_size,
1793                            &fs_info->generic_worker);
1794         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1795                            fs_info->thread_pool_size,
1796                            &fs_info->generic_worker);
1797         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1798                            "endio-meta-write", fs_info->thread_pool_size,
1799                            &fs_info->generic_worker);
1800         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1801                            fs_info->thread_pool_size,
1802                            &fs_info->generic_worker);
1803         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1804                            1, &fs_info->generic_worker);
1805
1806         /*
1807          * endios are largely parallel and should have a very
1808          * low idle thresh
1809          */
1810         fs_info->endio_workers.idle_thresh = 4;
1811         fs_info->endio_meta_workers.idle_thresh = 4;
1812
1813         fs_info->endio_write_workers.idle_thresh = 2;
1814         fs_info->endio_meta_write_workers.idle_thresh = 2;
1815
1816         btrfs_start_workers(&fs_info->workers, 1);
1817         btrfs_start_workers(&fs_info->generic_worker, 1);
1818         btrfs_start_workers(&fs_info->submit_workers, 1);
1819         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1820         btrfs_start_workers(&fs_info->fixup_workers, 1);
1821         btrfs_start_workers(&fs_info->endio_workers, 1);
1822         btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1823         btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1824         btrfs_start_workers(&fs_info->endio_write_workers, 1);
1825         btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1826
1827         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1828         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1829                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1830
1831         nodesize = btrfs_super_nodesize(disk_super);
1832         leafsize = btrfs_super_leafsize(disk_super);
1833         sectorsize = btrfs_super_sectorsize(disk_super);
1834         stripesize = btrfs_super_stripesize(disk_super);
1835         tree_root->nodesize = nodesize;
1836         tree_root->leafsize = leafsize;
1837         tree_root->sectorsize = sectorsize;
1838         tree_root->stripesize = stripesize;
1839
1840         sb->s_blocksize = sectorsize;
1841         sb->s_blocksize_bits = blksize_bits(sectorsize);
1842
1843         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1844                     sizeof(disk_super->magic))) {
1845                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1846                 goto fail_sb_buffer;
1847         }
1848
1849         mutex_lock(&fs_info->chunk_mutex);
1850         ret = btrfs_read_sys_array(tree_root);
1851         mutex_unlock(&fs_info->chunk_mutex);
1852         if (ret) {
1853                 printk(KERN_WARNING "btrfs: failed to read the system "
1854                        "array on %s\n", sb->s_id);
1855                 goto fail_sb_buffer;
1856         }
1857
1858         blocksize = btrfs_level_size(tree_root,
1859                                      btrfs_super_chunk_root_level(disk_super));
1860         generation = btrfs_super_chunk_root_generation(disk_super);
1861
1862         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1863                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1864
1865         chunk_root->node = read_tree_block(chunk_root,
1866                                            btrfs_super_chunk_root(disk_super),
1867                                            blocksize, generation);
1868         BUG_ON(!chunk_root->node);
1869         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1870                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1871                        sb->s_id);
1872                 goto fail_chunk_root;
1873         }
1874         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1875         chunk_root->commit_root = btrfs_root_node(chunk_root);
1876
1877         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1878            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1879            BTRFS_UUID_SIZE);
1880
1881         mutex_lock(&fs_info->chunk_mutex);
1882         ret = btrfs_read_chunk_tree(chunk_root);
1883         mutex_unlock(&fs_info->chunk_mutex);
1884         if (ret) {
1885                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1886                        sb->s_id);
1887                 goto fail_chunk_root;
1888         }
1889
1890         btrfs_close_extra_devices(fs_devices);
1891
1892         blocksize = btrfs_level_size(tree_root,
1893                                      btrfs_super_root_level(disk_super));
1894         generation = btrfs_super_generation(disk_super);
1895
1896         tree_root->node = read_tree_block(tree_root,
1897                                           btrfs_super_root(disk_super),
1898                                           blocksize, generation);
1899         if (!tree_root->node)
1900                 goto fail_chunk_root;
1901         if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1902                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1903                        sb->s_id);
1904                 goto fail_tree_root;
1905         }
1906         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1907         tree_root->commit_root = btrfs_root_node(tree_root);
1908
1909         ret = find_and_setup_root(tree_root, fs_info,
1910                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1911         if (ret)
1912                 goto fail_tree_root;
1913         extent_root->track_dirty = 1;
1914
1915         ret = find_and_setup_root(tree_root, fs_info,
1916                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1917         if (ret)
1918                 goto fail_extent_root;
1919         dev_root->track_dirty = 1;
1920
1921         ret = find_and_setup_root(tree_root, fs_info,
1922                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
1923         if (ret)
1924                 goto fail_dev_root;
1925
1926         csum_root->track_dirty = 1;
1927
1928         fs_info->generation = generation;
1929         fs_info->last_trans_committed = generation;
1930         fs_info->data_alloc_profile = (u64)-1;
1931         fs_info->metadata_alloc_profile = (u64)-1;
1932         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1933
1934         ret = btrfs_read_block_groups(extent_root);
1935         if (ret) {
1936                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
1937                 goto fail_block_groups;
1938         }
1939
1940         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1941                                                "btrfs-cleaner");
1942         if (IS_ERR(fs_info->cleaner_kthread))
1943                 goto fail_block_groups;
1944
1945         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1946                                                    tree_root,
1947                                                    "btrfs-transaction");
1948         if (IS_ERR(fs_info->transaction_kthread))
1949                 goto fail_cleaner;
1950
1951         if (!btrfs_test_opt(tree_root, SSD) &&
1952             !btrfs_test_opt(tree_root, NOSSD) &&
1953             !fs_info->fs_devices->rotating) {
1954                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
1955                        "mode\n");
1956                 btrfs_set_opt(fs_info->mount_opt, SSD);
1957         }
1958
1959         if (btrfs_super_log_root(disk_super) != 0) {
1960                 u64 bytenr = btrfs_super_log_root(disk_super);
1961
1962                 if (fs_devices->rw_devices == 0) {
1963                         printk(KERN_WARNING "Btrfs log replay required "
1964                                "on RO media\n");
1965                         err = -EIO;
1966                         goto fail_trans_kthread;
1967                 }
1968                 blocksize =
1969                      btrfs_level_size(tree_root,
1970                                       btrfs_super_log_root_level(disk_super));
1971
1972                 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1973                 if (!log_tree_root) {
1974                         err = -ENOMEM;
1975                         goto fail_trans_kthread;
1976                 }
1977
1978                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1979                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1980
1981                 log_tree_root->node = read_tree_block(tree_root, bytenr,
1982                                                       blocksize,
1983                                                       generation + 1);
1984                 ret = btrfs_recover_log_trees(log_tree_root);
1985                 BUG_ON(ret);
1986
1987                 if (sb->s_flags & MS_RDONLY) {
1988                         ret =  btrfs_commit_super(tree_root);
1989                         BUG_ON(ret);
1990                 }
1991         }
1992
1993         ret = btrfs_find_orphan_roots(tree_root);
1994         BUG_ON(ret);
1995
1996         if (!(sb->s_flags & MS_RDONLY)) {
1997                 ret = btrfs_cleanup_fs_roots(fs_info);
1998                 BUG_ON(ret);
1999
2000                 ret = btrfs_recover_relocation(tree_root);
2001                 if (ret < 0) {
2002                         printk(KERN_WARNING
2003                                "btrfs: failed to recover relocation\n");
2004                         err = -EINVAL;
2005                         goto fail_trans_kthread;
2006                 }
2007         }
2008
2009         location.objectid = BTRFS_FS_TREE_OBJECTID;
2010         location.type = BTRFS_ROOT_ITEM_KEY;
2011         location.offset = (u64)-1;
2012
2013         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2014         if (!fs_info->fs_root)
2015                 goto fail_trans_kthread;
2016         if (IS_ERR(fs_info->fs_root)) {
2017                 err = PTR_ERR(fs_info->fs_root);
2018                 goto fail_trans_kthread;
2019         }
2020
2021         if (!(sb->s_flags & MS_RDONLY)) {
2022                 down_read(&fs_info->cleanup_work_sem);
2023                 btrfs_orphan_cleanup(fs_info->fs_root);
2024                 btrfs_orphan_cleanup(fs_info->tree_root);
2025                 up_read(&fs_info->cleanup_work_sem);
2026         }
2027
2028         return tree_root;
2029
2030 fail_trans_kthread:
2031         kthread_stop(fs_info->transaction_kthread);
2032 fail_cleaner:
2033         kthread_stop(fs_info->cleaner_kthread);
2034
2035         /*
2036          * make sure we're done with the btree inode before we stop our
2037          * kthreads
2038          */
2039         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2040         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2041
2042 fail_block_groups:
2043         btrfs_free_block_groups(fs_info);
2044         free_extent_buffer(csum_root->node);
2045         free_extent_buffer(csum_root->commit_root);
2046 fail_dev_root:
2047         free_extent_buffer(dev_root->node);
2048         free_extent_buffer(dev_root->commit_root);
2049 fail_extent_root:
2050         free_extent_buffer(extent_root->node);
2051         free_extent_buffer(extent_root->commit_root);
2052 fail_tree_root:
2053         free_extent_buffer(tree_root->node);
2054         free_extent_buffer(tree_root->commit_root);
2055 fail_chunk_root:
2056         free_extent_buffer(chunk_root->node);
2057         free_extent_buffer(chunk_root->commit_root);
2058 fail_sb_buffer:
2059         btrfs_stop_workers(&fs_info->generic_worker);
2060         btrfs_stop_workers(&fs_info->fixup_workers);
2061         btrfs_stop_workers(&fs_info->delalloc_workers);
2062         btrfs_stop_workers(&fs_info->workers);
2063         btrfs_stop_workers(&fs_info->endio_workers);
2064         btrfs_stop_workers(&fs_info->endio_meta_workers);
2065         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2066         btrfs_stop_workers(&fs_info->endio_write_workers);
2067         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2068         btrfs_stop_workers(&fs_info->submit_workers);
2069 fail_iput:
2070         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2071         iput(fs_info->btree_inode);
2072
2073         btrfs_close_devices(fs_info->fs_devices);
2074         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2075 fail_bdi:
2076         bdi_destroy(&fs_info->bdi);
2077 fail_srcu:
2078         cleanup_srcu_struct(&fs_info->subvol_srcu);
2079 fail:
2080         kfree(extent_root);
2081         kfree(tree_root);
2082         kfree(fs_info);
2083         kfree(chunk_root);
2084         kfree(dev_root);
2085         kfree(csum_root);
2086         return ERR_PTR(err);
2087 }
2088
2089 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2090 {
2091         char b[BDEVNAME_SIZE];
2092
2093         if (uptodate) {
2094                 set_buffer_uptodate(bh);
2095         } else {
2096                 if (printk_ratelimit()) {
2097                         printk(KERN_WARNING "lost page write due to "
2098                                         "I/O error on %s\n",
2099                                        bdevname(bh->b_bdev, b));
2100                 }
2101                 /* note, we dont' set_buffer_write_io_error because we have
2102                  * our own ways of dealing with the IO errors
2103                  */
2104                 clear_buffer_uptodate(bh);
2105         }
2106         unlock_buffer(bh);
2107         put_bh(bh);
2108 }
2109
2110 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2111 {
2112         struct buffer_head *bh;
2113         struct buffer_head *latest = NULL;
2114         struct btrfs_super_block *super;
2115         int i;
2116         u64 transid = 0;
2117         u64 bytenr;
2118
2119         /* we would like to check all the supers, but that would make
2120          * a btrfs mount succeed after a mkfs from a different FS.
2121          * So, we need to add a special mount option to scan for
2122          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2123          */
2124         for (i = 0; i < 1; i++) {
2125                 bytenr = btrfs_sb_offset(i);
2126                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2127                         break;
2128                 bh = __bread(bdev, bytenr / 4096, 4096);
2129                 if (!bh)
2130                         continue;
2131
2132                 super = (struct btrfs_super_block *)bh->b_data;
2133                 if (btrfs_super_bytenr(super) != bytenr ||
2134                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2135                             sizeof(super->magic))) {
2136                         brelse(bh);
2137                         continue;
2138                 }
2139
2140                 if (!latest || btrfs_super_generation(super) > transid) {
2141                         brelse(latest);
2142                         latest = bh;
2143                         transid = btrfs_super_generation(super);
2144                 } else {
2145                         brelse(bh);
2146                 }
2147         }
2148         return latest;
2149 }
2150
2151 /*
2152  * this should be called twice, once with wait == 0 and
2153  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2154  * we write are pinned.
2155  *
2156  * They are released when wait == 1 is done.
2157  * max_mirrors must be the same for both runs, and it indicates how
2158  * many supers on this one device should be written.
2159  *
2160  * max_mirrors == 0 means to write them all.
2161  */
2162 static int write_dev_supers(struct btrfs_device *device,
2163                             struct btrfs_super_block *sb,
2164                             int do_barriers, int wait, int max_mirrors)
2165 {
2166         struct buffer_head *bh;
2167         int i;
2168         int ret;
2169         int errors = 0;
2170         u32 crc;
2171         u64 bytenr;
2172         int last_barrier = 0;
2173
2174         if (max_mirrors == 0)
2175                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2176
2177         /* make sure only the last submit_bh does a barrier */
2178         if (do_barriers) {
2179                 for (i = 0; i < max_mirrors; i++) {
2180                         bytenr = btrfs_sb_offset(i);
2181                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2182                             device->total_bytes)
2183                                 break;
2184                         last_barrier = i;
2185                 }
2186         }
2187
2188         for (i = 0; i < max_mirrors; i++) {
2189                 bytenr = btrfs_sb_offset(i);
2190                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2191                         break;
2192
2193                 if (wait) {
2194                         bh = __find_get_block(device->bdev, bytenr / 4096,
2195                                               BTRFS_SUPER_INFO_SIZE);
2196                         BUG_ON(!bh);
2197                         wait_on_buffer(bh);
2198                         if (!buffer_uptodate(bh))
2199                                 errors++;
2200
2201                         /* drop our reference */
2202                         brelse(bh);
2203
2204                         /* drop the reference from the wait == 0 run */
2205                         brelse(bh);
2206                         continue;
2207                 } else {
2208                         btrfs_set_super_bytenr(sb, bytenr);
2209
2210                         crc = ~(u32)0;
2211                         crc = btrfs_csum_data(NULL, (char *)sb +
2212                                               BTRFS_CSUM_SIZE, crc,
2213                                               BTRFS_SUPER_INFO_SIZE -
2214                                               BTRFS_CSUM_SIZE);
2215                         btrfs_csum_final(crc, sb->csum);
2216
2217                         /*
2218                          * one reference for us, and we leave it for the
2219                          * caller
2220                          */
2221                         bh = __getblk(device->bdev, bytenr / 4096,
2222                                       BTRFS_SUPER_INFO_SIZE);
2223                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2224
2225                         /* one reference for submit_bh */
2226                         get_bh(bh);
2227
2228                         set_buffer_uptodate(bh);
2229                         lock_buffer(bh);
2230                         bh->b_end_io = btrfs_end_buffer_write_sync;
2231                 }
2232
2233                 if (i == last_barrier && do_barriers)
2234                         ret = submit_bh(WRITE_FLUSH_FUA, bh);
2235                 else
2236                         ret = submit_bh(WRITE_SYNC, bh);
2237
2238                 if (ret)
2239                         errors++;
2240         }
2241         return errors < i ? 0 : -1;
2242 }
2243
2244 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2245 {
2246         struct list_head *head;
2247         struct btrfs_device *dev;
2248         struct btrfs_super_block *sb;
2249         struct btrfs_dev_item *dev_item;
2250         int ret;
2251         int do_barriers;
2252         int max_errors;
2253         int total_errors = 0;
2254         u64 flags;
2255
2256         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2257         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2258
2259         sb = &root->fs_info->super_for_commit;
2260         dev_item = &sb->dev_item;
2261
2262         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2263         head = &root->fs_info->fs_devices->devices;
2264         list_for_each_entry(dev, head, dev_list) {
2265                 if (!dev->bdev) {
2266                         total_errors++;
2267                         continue;
2268                 }
2269                 if (!dev->in_fs_metadata || !dev->writeable)
2270                         continue;
2271
2272                 btrfs_set_stack_device_generation(dev_item, 0);
2273                 btrfs_set_stack_device_type(dev_item, dev->type);
2274                 btrfs_set_stack_device_id(dev_item, dev->devid);
2275                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2276                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2277                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2278                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2279                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2280                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2281                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2282
2283                 flags = btrfs_super_flags(sb);
2284                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2285
2286                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2287                 if (ret)
2288                         total_errors++;
2289         }
2290         if (total_errors > max_errors) {
2291                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2292                        total_errors);
2293                 BUG();
2294         }
2295
2296         total_errors = 0;
2297         list_for_each_entry(dev, head, dev_list) {
2298                 if (!dev->bdev)
2299                         continue;
2300                 if (!dev->in_fs_metadata || !dev->writeable)
2301                         continue;
2302
2303                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2304                 if (ret)
2305                         total_errors++;
2306         }
2307         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2308         if (total_errors > max_errors) {
2309                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2310                        total_errors);
2311                 BUG();
2312         }
2313         return 0;
2314 }
2315
2316 int write_ctree_super(struct btrfs_trans_handle *trans,
2317                       struct btrfs_root *root, int max_mirrors)
2318 {
2319         int ret;
2320
2321         ret = write_all_supers(root, max_mirrors);
2322         return ret;
2323 }
2324
2325 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2326 {
2327         spin_lock(&fs_info->fs_roots_radix_lock);
2328         radix_tree_delete(&fs_info->fs_roots_radix,
2329                           (unsigned long)root->root_key.objectid);
2330         spin_unlock(&fs_info->fs_roots_radix_lock);
2331
2332         if (btrfs_root_refs(&root->root_item) == 0)
2333                 synchronize_srcu(&fs_info->subvol_srcu);
2334
2335         free_fs_root(root);
2336         return 0;
2337 }
2338
2339 static void free_fs_root(struct btrfs_root *root)
2340 {
2341         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2342         if (root->anon_super.s_dev) {
2343                 down_write(&root->anon_super.s_umount);
2344                 kill_anon_super(&root->anon_super);
2345         }
2346         free_extent_buffer(root->node);
2347         free_extent_buffer(root->commit_root);
2348         kfree(root->name);
2349         kfree(root);
2350 }
2351
2352 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2353 {
2354         int ret;
2355         struct btrfs_root *gang[8];
2356         int i;
2357
2358         while (!list_empty(&fs_info->dead_roots)) {
2359                 gang[0] = list_entry(fs_info->dead_roots.next,
2360                                      struct btrfs_root, root_list);
2361                 list_del(&gang[0]->root_list);
2362
2363                 if (gang[0]->in_radix) {
2364                         btrfs_free_fs_root(fs_info, gang[0]);
2365                 } else {
2366                         free_extent_buffer(gang[0]->node);
2367                         free_extent_buffer(gang[0]->commit_root);
2368                         kfree(gang[0]);
2369                 }
2370         }
2371
2372         while (1) {
2373                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2374                                              (void **)gang, 0,
2375                                              ARRAY_SIZE(gang));
2376                 if (!ret)
2377                         break;
2378                 for (i = 0; i < ret; i++)
2379                         btrfs_free_fs_root(fs_info, gang[i]);
2380         }
2381         return 0;
2382 }
2383
2384 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2385 {
2386         u64 root_objectid = 0;
2387         struct btrfs_root *gang[8];
2388         int i;
2389         int ret;
2390
2391         while (1) {
2392                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2393                                              (void **)gang, root_objectid,
2394                                              ARRAY_SIZE(gang));
2395                 if (!ret)
2396                         break;
2397
2398                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2399                 for (i = 0; i < ret; i++) {
2400                         root_objectid = gang[i]->root_key.objectid;
2401                         btrfs_orphan_cleanup(gang[i]);
2402                 }
2403                 root_objectid++;
2404         }
2405         return 0;
2406 }
2407
2408 int btrfs_commit_super(struct btrfs_root *root)
2409 {
2410         struct btrfs_trans_handle *trans;
2411         int ret;
2412
2413         mutex_lock(&root->fs_info->cleaner_mutex);
2414         btrfs_run_delayed_iputs(root);
2415         btrfs_clean_old_snapshots(root);
2416         mutex_unlock(&root->fs_info->cleaner_mutex);
2417
2418         /* wait until ongoing cleanup work done */
2419         down_write(&root->fs_info->cleanup_work_sem);
2420         up_write(&root->fs_info->cleanup_work_sem);
2421
2422         trans = btrfs_join_transaction(root, 1);
2423         ret = btrfs_commit_transaction(trans, root);
2424         BUG_ON(ret);
2425         /* run commit again to drop the original snapshot */
2426         trans = btrfs_join_transaction(root, 1);
2427         btrfs_commit_transaction(trans, root);
2428         ret = btrfs_write_and_wait_transaction(NULL, root);
2429         BUG_ON(ret);
2430
2431         ret = write_ctree_super(NULL, root, 0);
2432         return ret;
2433 }
2434
2435 int close_ctree(struct btrfs_root *root)
2436 {
2437         struct btrfs_fs_info *fs_info = root->fs_info;
2438         int ret;
2439
2440         fs_info->closing = 1;
2441         smp_mb();
2442
2443         btrfs_put_block_group_cache(fs_info);
2444         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2445                 ret =  btrfs_commit_super(root);
2446                 if (ret)
2447                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2448         }
2449
2450         kthread_stop(root->fs_info->transaction_kthread);
2451         kthread_stop(root->fs_info->cleaner_kthread);
2452
2453         fs_info->closing = 2;
2454         smp_mb();
2455
2456         if (fs_info->delalloc_bytes) {
2457                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2458                        (unsigned long long)fs_info->delalloc_bytes);
2459         }
2460         if (fs_info->total_ref_cache_size) {
2461                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2462                        (unsigned long long)fs_info->total_ref_cache_size);
2463         }
2464
2465         free_extent_buffer(fs_info->extent_root->node);
2466         free_extent_buffer(fs_info->extent_root->commit_root);
2467         free_extent_buffer(fs_info->tree_root->node);
2468         free_extent_buffer(fs_info->tree_root->commit_root);
2469         free_extent_buffer(root->fs_info->chunk_root->node);
2470         free_extent_buffer(root->fs_info->chunk_root->commit_root);
2471         free_extent_buffer(root->fs_info->dev_root->node);
2472         free_extent_buffer(root->fs_info->dev_root->commit_root);
2473         free_extent_buffer(root->fs_info->csum_root->node);
2474         free_extent_buffer(root->fs_info->csum_root->commit_root);
2475
2476         btrfs_free_block_groups(root->fs_info);
2477
2478         del_fs_roots(fs_info);
2479
2480         iput(fs_info->btree_inode);
2481
2482         btrfs_stop_workers(&fs_info->generic_worker);
2483         btrfs_stop_workers(&fs_info->fixup_workers);
2484         btrfs_stop_workers(&fs_info->delalloc_workers);
2485         btrfs_stop_workers(&fs_info->workers);
2486         btrfs_stop_workers(&fs_info->endio_workers);
2487         btrfs_stop_workers(&fs_info->endio_meta_workers);
2488         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2489         btrfs_stop_workers(&fs_info->endio_write_workers);
2490         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2491         btrfs_stop_workers(&fs_info->submit_workers);
2492
2493         btrfs_close_devices(fs_info->fs_devices);
2494         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2495
2496         bdi_destroy(&fs_info->bdi);
2497         cleanup_srcu_struct(&fs_info->subvol_srcu);
2498
2499         kfree(fs_info->extent_root);
2500         kfree(fs_info->tree_root);
2501         kfree(fs_info->chunk_root);
2502         kfree(fs_info->dev_root);
2503         kfree(fs_info->csum_root);
2504         return 0;
2505 }
2506
2507 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2508 {
2509         int ret;
2510         struct inode *btree_inode = buf->first_page->mapping->host;
2511
2512         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2513                                      NULL);
2514         if (!ret)
2515                 return ret;
2516
2517         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2518                                     parent_transid);
2519         return !ret;
2520 }
2521
2522 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2523 {
2524         struct inode *btree_inode = buf->first_page->mapping->host;
2525         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2526                                           buf);
2527 }
2528
2529 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2530 {
2531         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2532         u64 transid = btrfs_header_generation(buf);
2533         struct inode *btree_inode = root->fs_info->btree_inode;
2534         int was_dirty;
2535
2536         btrfs_assert_tree_locked(buf);
2537         if (transid != root->fs_info->generation) {
2538                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2539                        "found %llu running %llu\n",
2540                         (unsigned long long)buf->start,
2541                         (unsigned long long)transid,
2542                         (unsigned long long)root->fs_info->generation);
2543                 WARN_ON(1);
2544         }
2545         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2546                                             buf);
2547         if (!was_dirty) {
2548                 spin_lock(&root->fs_info->delalloc_lock);
2549                 root->fs_info->dirty_metadata_bytes += buf->len;
2550                 spin_unlock(&root->fs_info->delalloc_lock);
2551         }
2552 }
2553
2554 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2555 {
2556         /*
2557          * looks as though older kernels can get into trouble with
2558          * this code, they end up stuck in balance_dirty_pages forever
2559          */
2560         u64 num_dirty;
2561         unsigned long thresh = 32 * 1024 * 1024;
2562
2563         if (current->flags & PF_MEMALLOC)
2564                 return;
2565
2566         num_dirty = root->fs_info->dirty_metadata_bytes;
2567
2568         if (num_dirty > thresh) {
2569                 balance_dirty_pages_ratelimited_nr(
2570                                    root->fs_info->btree_inode->i_mapping, 1);
2571         }
2572         return;
2573 }
2574
2575 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2576 {
2577         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2578         int ret;
2579         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2580         if (ret == 0)
2581                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2582         return ret;
2583 }
2584
2585 int btree_lock_page_hook(struct page *page)
2586 {
2587         struct inode *inode = page->mapping->host;
2588         struct btrfs_root *root = BTRFS_I(inode)->root;
2589         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2590         struct extent_buffer *eb;
2591         unsigned long len;
2592         u64 bytenr = page_offset(page);
2593
2594         if (page->private == EXTENT_PAGE_PRIVATE)
2595                 goto out;
2596
2597         len = page->private >> 2;
2598         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2599         if (!eb)
2600                 goto out;
2601
2602         btrfs_tree_lock(eb);
2603         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2604
2605         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2606                 spin_lock(&root->fs_info->delalloc_lock);
2607                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2608                         root->fs_info->dirty_metadata_bytes -= eb->len;
2609                 else
2610                         WARN_ON(1);
2611                 spin_unlock(&root->fs_info->delalloc_lock);
2612         }
2613
2614         btrfs_tree_unlock(eb);
2615         free_extent_buffer(eb);
2616 out:
2617         lock_page(page);
2618         return 0;
2619 }
2620
2621 static struct extent_io_ops btree_extent_io_ops = {
2622         .write_cache_pages_lock_hook = btree_lock_page_hook,
2623         .readpage_end_io_hook = btree_readpage_end_io_hook,
2624         .submit_bio_hook = btree_submit_bio_hook,
2625         /* note we're sharing with inode.c for the merge bio hook */
2626         .merge_bio_hook = btrfs_merge_bio_hook,
2627 };