fs/btrfs/disk-io.c

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