fs/btrfs/disk-io.c

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