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/version.h>
  20 #include <linux/fs.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/scatterlist.h>
  23 #include <linux/swap.h>
  24 #include <linux/radix-tree.h>
  25 #include <linux/writeback.h>
  26 #include <linux/buffer_head.h> // for block_sync_page
  27 #include <linux/workqueue.h>
  28 #include <linux/kthread.h>
  29 # include <linux/freezer.h>
  30 #include "crc32c.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 "ref-cache.h"
  40 #include "tree-log.h"
  41
  42 #if 0
  43 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
  44 {
  45         if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
  46                 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
  47                        (unsigned long long)extent_buffer_blocknr(buf),
  48                        (unsigned long long)btrfs_header_blocknr(buf));
  49                 return 1;
  50         }
  51         return 0;
  52 }
  53 #endif
  54
  55 static struct extent_io_ops btree_extent_io_ops;
  56 static void end_workqueue_fn(struct btrfs_work *work);
  57
  58 /*
  59  * end_io_wq structs are used to do processing in task context when an IO is
  60  * complete.  This is used during reads to verify checksums, and it is used
  61  * by writes to insert metadata for new file extents after IO is complete.
  62  */
  63 struct end_io_wq {
  64         struct bio *bio;
  65         bio_end_io_t *end_io;
  66         void *private;
  67         struct btrfs_fs_info *info;
  68         int error;
  69         int metadata;
  70         struct list_head list;
  71         struct btrfs_work work;
  72 };
  73
  74 /*
  75  * async submit bios are used to offload expensive checksumming
  76  * onto the worker threads.  They checksum file and metadata bios
  77  * just before they are sent down the IO stack.
  78  */
  79 struct async_submit_bio {
  80         struct inode *inode;
  81         struct bio *bio;
  82         struct list_head list;
  83         extent_submit_bio_hook_t *submit_bio_hook;
  84         int rw;
  85         int mirror_num;
  86         struct btrfs_work work;
  87 };
  88
  89 /*
  90  * extents on the btree inode are pretty simple, there's one extent
  91  * that covers the entire device
  92  */
  93 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
  94                                     size_t page_offset, u64 start, u64 len,
  95                                     int create)
  96 {
  97         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
  98         struct extent_map *em;
  99         int ret;
 100
 101         spin_lock(&em_tree->lock);
 102         em = lookup_extent_mapping(em_tree, start, len);
 103         if (em) {
 104                 em->bdev =
 105                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 106                 spin_unlock(&em_tree->lock);
 107                 goto out;
 108         }
 109         spin_unlock(&em_tree->lock);
 110
 111         em = alloc_extent_map(GFP_NOFS);
 112         if (!em) {
 113                 em = ERR_PTR(-ENOMEM);
 114                 goto out;
 115         }
 116         em->start = 0;
 117         em->len = (u64)-1;
 118         em->block_start = 0;
 119         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 120
 121         spin_lock(&em_tree->lock);
 122         ret = add_extent_mapping(em_tree, em);
 123         if (ret == -EEXIST) {
 124                 u64 failed_start = em->start;
 125                 u64 failed_len = em->len;
 126
 127                 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
 128                        em->start, em->len, em->block_start);
 129                 free_extent_map(em);
 130                 em = lookup_extent_mapping(em_tree, start, len);
 131                 if (em) {
 132                         printk("after failing, found %Lu %Lu %Lu\n",
 133                                em->start, em->len, em->block_start);
 134                         ret = 0;
 135                 } else {
 136                         em = lookup_extent_mapping(em_tree, failed_start,
 137                                                    failed_len);
 138                         if (em) {
 139                                 printk("double failure lookup gives us "
 140                                        "%Lu %Lu -> %Lu\n", em->start,
 141                                        em->len, em->block_start);
 142                                 free_extent_map(em);
 143                         }
 144                         ret = -EIO;
 145                 }
 146         } else if (ret) {
 147                 free_extent_map(em);
 148                 em = NULL;
 149         }
 150         spin_unlock(&em_tree->lock);
 151
 152         if (ret)
 153                 em = ERR_PTR(ret);
 154 out:
 155         return em;
 156 }
 157
 158 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 159 {
 160         return btrfs_crc32c(seed, data, len);
 161 }
 162
 163 void btrfs_csum_final(u32 crc, char *result)
 164 {
 165         *(__le32 *)result = ~cpu_to_le32(crc);
 166 }
 167
 168 /*
 169  * compute the csum for a btree block, and either verify it or write it
 170  * into the csum field of the block.
 171  */
 172 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 173                            int verify)
 174 {
 175         char result[BTRFS_CRC32_SIZE];
 176         unsigned long len;
 177         unsigned long cur_len;
 178         unsigned long offset = BTRFS_CSUM_SIZE;
 179         char *map_token = NULL;
 180         char *kaddr;
 181         unsigned long map_start;
 182         unsigned long map_len;
 183         int err;
 184         u32 crc = ~(u32)0;
 185
 186         len = buf->len - offset;
 187         while(len > 0) {
 188                 err = map_private_extent_buffer(buf, offset, 32,
 189                                         &map_token, &kaddr,
 190                                         &map_start, &map_len, KM_USER0);
 191                 if (err) {
 192                         printk("failed to map extent buffer! %lu\n",
 193                                offset);
 194                         return 1;
 195                 }
 196                 cur_len = min(len, map_len - (offset - map_start));
 197                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 198                                       crc, cur_len);
 199                 len -= cur_len;
 200                 offset += cur_len;
 201                 unmap_extent_buffer(buf, map_token, KM_USER0);
 202         }
 203         btrfs_csum_final(crc, result);
 204
 205         if (verify) {
 206                 /* FIXME, this is not good */
 207                 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
 208                         u32 val;
 209                         u32 found = 0;
 210                         memcpy(&found, result, BTRFS_CRC32_SIZE);
 211
 212                         read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
 213                         printk("btrfs: %s checksum verify failed on %llu "
 214                                "wanted %X found %X level %d\n",
 215                                root->fs_info->sb->s_id,
 216                                buf->start, val, found, btrfs_header_level(buf));
 217                         return 1;
 218                 }
 219         } else {
 220                 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
 221         }
 222         return 0;
 223 }
 224
 225 /*
 226  * we can't consider a given block up to date unless the transid of the
 227  * block matches the transid in the parent node's pointer.  This is how we
 228  * detect blocks that either didn't get written at all or got written
 229  * in the wrong place.
 230  */
 231 static int verify_parent_transid(struct extent_io_tree *io_tree,
 232                                  struct extent_buffer *eb, u64 parent_transid)
 233 {
 234         int ret;
 235
 236         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 237                 return 0;
 238
 239         lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
 240         if (extent_buffer_uptodate(io_tree, eb) &&
 241             btrfs_header_generation(eb) == parent_transid) {
 242                 ret = 0;
 243                 goto out;
 244         }
 245         printk("parent transid verify failed on %llu wanted %llu found %llu\n",
 246                (unsigned long long)eb->start,
 247                (unsigned long long)parent_transid,
 248                (unsigned long long)btrfs_header_generation(eb));
 249         ret = 1;
 250         clear_extent_buffer_uptodate(io_tree, eb);
 251 out:
 252         unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
 253                       GFP_NOFS);
 254         return ret;
 255 }
 256
 257 /*
 258  * helper to read a given tree block, doing retries as required when
 259  * the checksums don't match and we have alternate mirrors to try.
 260  */
 261 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 262                                           struct extent_buffer *eb,
 263                                           u64 start, u64 parent_transid)
 264 {
 265         struct extent_io_tree *io_tree;
 266         int ret;
 267         int num_copies = 0;
 268         int mirror_num = 0;
 269
 270         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 271         while (1) {
 272                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 273                                                btree_get_extent, mirror_num);
 274                 if (!ret &&
 275                     !verify_parent_transid(io_tree, eb, parent_transid))
 276                         return ret;
 277 printk("read extent buffer pages failed with ret %d mirror no %d\n", ret, mirror_num);
 278                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 279                                               eb->start, eb->len);
 280                 if (num_copies == 1)
 281                         return ret;
 282
 283                 mirror_num++;
 284                 if (mirror_num > num_copies)
 285                         return ret;
 286         }
 287         return -EIO;
 288 }
 289
 290 /*
 291  * checksum a dirty tree block before IO.  This has extra checks to make
 292  * sure we only fill in the checksum field in the first page of a multi-page block
 293  */
 294 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 295 {
 296         struct extent_io_tree *tree;
 297         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 298         u64 found_start;
 299         int found_level;
 300         unsigned long len;
 301         struct extent_buffer *eb;
 302         int ret;
 303
 304         tree = &BTRFS_I(page->mapping->host)->io_tree;
 305
 306         if (page->private == EXTENT_PAGE_PRIVATE)
 307                 goto out;
 308         if (!page->private)
 309                 goto out;
 310         len = page->private >> 2;
 311         if (len == 0) {
 312                 WARN_ON(1);
 313         }
 314         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 315         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 316                                              btrfs_header_generation(eb));
 317         BUG_ON(ret);
 318         found_start = btrfs_header_bytenr(eb);
 319         if (found_start != start) {
 320                 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
 321                        start, found_start, len);
 322                 WARN_ON(1);
 323                 goto err;
 324         }
 325         if (eb->first_page != page) {
 326                 printk("bad first page %lu %lu\n", eb->first_page->index,
 327                        page->index);
 328                 WARN_ON(1);
 329                 goto err;
 330         }
 331         if (!PageUptodate(page)) {
 332                 printk("csum not up to date page %lu\n", page->index);
 333                 WARN_ON(1);
 334                 goto err;
 335         }
 336         found_level = btrfs_header_level(eb);
 337
 338         csum_tree_block(root, eb, 0);
 339 err:
 340         free_extent_buffer(eb);
 341 out:
 342         return 0;
 343 }
 344
 345 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 346                                struct extent_state *state)
 347 {
 348         struct extent_io_tree *tree;
 349         u64 found_start;
 350         int found_level;
 351         unsigned long len;
 352         struct extent_buffer *eb;
 353         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 354         int ret = 0;
 355
 356         tree = &BTRFS_I(page->mapping->host)->io_tree;
 357         if (page->private == EXTENT_PAGE_PRIVATE)
 358                 goto out;
 359         if (!page->private)
 360                 goto out;
 361         len = page->private >> 2;
 362         if (len == 0) {
 363                 WARN_ON(1);
 364         }
 365         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 366
 367         found_start = btrfs_header_bytenr(eb);
 368         if (found_start != start) {
 369                 printk("bad tree block start %llu %llu\n",
 370                        (unsigned long long)found_start,
 371                        (unsigned long long)eb->start);
 372                 ret = -EIO;
 373                 goto err;
 374         }
 375         if (eb->first_page != page) {
 376                 printk("bad first page %lu %lu\n", eb->first_page->index,
 377                        page->index);
 378                 WARN_ON(1);
 379                 ret = -EIO;
 380                 goto err;
 381         }
 382         if (memcmp_extent_buffer(eb, root->fs_info->fsid,
 383                                  (unsigned long)btrfs_header_fsid(eb),
 384                                  BTRFS_FSID_SIZE)) {
 385                 printk("bad fsid on block %Lu\n", eb->start);
 386                 ret = -EIO;
 387                 goto err;
 388         }
 389         found_level = btrfs_header_level(eb);
 390
 391         ret = csum_tree_block(root, eb, 1);
 392         if (ret)
 393                 ret = -EIO;
 394
 395         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 396         end = eb->start + end - 1;
 397 err:
 398         free_extent_buffer(eb);
 399 out:
 400         return ret;
 401 }
 402
 403 static void end_workqueue_bio(struct bio *bio, int err)
 404 {
 405         struct end_io_wq *end_io_wq = bio->bi_private;
 406         struct btrfs_fs_info *fs_info;
 407
 408         fs_info = end_io_wq->info;
 409         end_io_wq->error = err;
 410         end_io_wq->work.func = end_workqueue_fn;
 411         end_io_wq->work.flags = 0;
 412         if (bio->bi_rw & (1 << BIO_RW))
 413                 btrfs_queue_worker(&fs_info->endio_write_workers,
 414                                    &end_io_wq->work);
 415         else
 416                 btrfs_queue_worker(&fs_info->endio_workers, &end_io_wq->work);
 417 }
 418
 419 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 420                         int metadata)
 421 {
 422         struct end_io_wq *end_io_wq;
 423         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 424         if (!end_io_wq)
 425                 return -ENOMEM;
 426
 427         end_io_wq->private = bio->bi_private;
 428         end_io_wq->end_io = bio->bi_end_io;
 429         end_io_wq->info = info;
 430         end_io_wq->error = 0;
 431         end_io_wq->bio = bio;
 432         end_io_wq->metadata = metadata;
 433
 434         bio->bi_private = end_io_wq;
 435         bio->bi_end_io = end_workqueue_bio;
 436         return 0;
 437 }
 438
 439 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 440 {
 441         unsigned long limit = min_t(unsigned long,
 442                                     info->workers.max_workers,
 443                                     info->fs_devices->open_devices);
 444         return 256 * limit;
 445 }
 446
 447 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
 448 {
 449         return atomic_read(&info->nr_async_bios) >
 450                 btrfs_async_submit_limit(info);
 451 }
 452
 453 static void run_one_async_submit(struct btrfs_work *work)
 454 {
 455         struct btrfs_fs_info *fs_info;
 456         struct async_submit_bio *async;
 457         int limit;
 458
 459         async = container_of(work, struct  async_submit_bio, work);
 460         fs_info = BTRFS_I(async->inode)->root->fs_info;
 461
 462         limit = btrfs_async_submit_limit(fs_info);
 463         limit = limit * 2 / 3;
 464
 465         atomic_dec(&fs_info->nr_async_submits);
 466
 467         if (atomic_read(&fs_info->nr_async_submits) < limit &&
 468             waitqueue_active(&fs_info->async_submit_wait))
 469                 wake_up(&fs_info->async_submit_wait);
 470
 471         async->submit_bio_hook(async->inode, async->rw, async->bio,
 472                                async->mirror_num);
 473         kfree(async);
 474 }
 475
 476 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 477                         int rw, struct bio *bio, int mirror_num,
 478                         extent_submit_bio_hook_t *submit_bio_hook)
 479 {
 480         struct async_submit_bio *async;
 481         int limit = btrfs_async_submit_limit(fs_info);
 482
 483         async = kmalloc(sizeof(*async), GFP_NOFS);
 484         if (!async)
 485                 return -ENOMEM;
 486
 487         async->inode = inode;
 488         async->rw = rw;
 489         async->bio = bio;
 490         async->mirror_num = mirror_num;
 491         async->submit_bio_hook = submit_bio_hook;
 492         async->work.func = run_one_async_submit;
 493         async->work.flags = 0;
 494
 495         while(atomic_read(&fs_info->async_submit_draining) &&
 496               atomic_read(&fs_info->nr_async_submits)) {
 497                 wait_event(fs_info->async_submit_wait,
 498                            (atomic_read(&fs_info->nr_async_submits) == 0));
 499         }
 500
 501         atomic_inc(&fs_info->nr_async_submits);
 502         btrfs_queue_worker(&fs_info->workers, &async->work);
 503
 504         if (atomic_read(&fs_info->nr_async_submits) > limit) {
 505                 wait_event_timeout(fs_info->async_submit_wait,
 506                            (atomic_read(&fs_info->nr_async_submits) < limit),
 507                            HZ/10);
 508
 509                 wait_event_timeout(fs_info->async_submit_wait,
 510                            (atomic_read(&fs_info->nr_async_bios) < limit),
 511                            HZ/10);
 512         }
 513         return 0;
 514 }
 515
 516 static int btree_csum_one_bio(struct bio *bio)
 517 {
 518         struct bio_vec *bvec = bio->bi_io_vec;
 519         int bio_index = 0;
 520         struct btrfs_root *root;
 521
 522         WARN_ON(bio->bi_vcnt <= 0);
 523         while(bio_index < bio->bi_vcnt) {
 524                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 525                 csum_dirty_buffer(root, bvec->bv_page);
 526                 bio_index++;
 527                 bvec++;
 528         }
 529         return 0;
 530 }
 531
 532 static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 533                                  int mirror_num)
 534 {
 535         struct btrfs_root *root = BTRFS_I(inode)->root;
 536         int ret;
 537
 538         /*
 539          * when we're called for a write, we're already in the async
 540          * submission context.  Just jump into btrfs_map_bio
 541          */
 542         if (rw & (1 << BIO_RW)) {
 543                 btree_csum_one_bio(bio);
 544                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 545                                      mirror_num, 1);
 546         }
 547
 548         /*
 549          * called for a read, do the setup so that checksum validation
 550          * can happen in the async kernel threads
 551          */
 552         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
 553         BUG_ON(ret);
 554
 555         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 556 }
 557
 558 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 559                                  int mirror_num)
 560 {
 561         /*
 562          * kthread helpers are used to submit writes so that checksumming
 563          * can happen in parallel across all CPUs
 564          */
 565         if (!(rw & (1 << BIO_RW))) {
 566                 return __btree_submit_bio_hook(inode, rw, bio, mirror_num);
 567         }
 568         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 569                                    inode, rw, bio, mirror_num,
 570                                    __btree_submit_bio_hook);
 571 }
 572
 573 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 574 {
 575         struct extent_io_tree *tree;
 576         tree = &BTRFS_I(page->mapping->host)->io_tree;
 577
 578         if (current->flags & PF_MEMALLOC) {
 579                 redirty_page_for_writepage(wbc, page);
 580                 unlock_page(page);
 581                 return 0;
 582         }
 583         return extent_write_full_page(tree, page, btree_get_extent, wbc);
 584 }
 585
 586 static int btree_writepages(struct address_space *mapping,
 587                             struct writeback_control *wbc)
 588 {
 589         struct extent_io_tree *tree;
 590         tree = &BTRFS_I(mapping->host)->io_tree;
 591         if (wbc->sync_mode == WB_SYNC_NONE) {
 592                 u64 num_dirty;
 593                 u64 start = 0;
 594                 unsigned long thresh = 32 * 1024 * 1024;
 595
 596                 if (wbc->for_kupdate)
 597                         return 0;
 598
 599                 num_dirty = count_range_bits(tree, &start, (u64)-1,
 600                                              thresh, EXTENT_DIRTY);
 601                 if (num_dirty < thresh) {
 602                         return 0;
 603                 }
 604         }
 605         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 606 }
 607
 608 int btree_readpage(struct file *file, struct page *page)
 609 {
 610         struct extent_io_tree *tree;
 611         tree = &BTRFS_I(page->mapping->host)->io_tree;
 612         return extent_read_full_page(tree, page, btree_get_extent);
 613 }
 614
 615 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 616 {
 617         struct extent_io_tree *tree;
 618         struct extent_map_tree *map;
 619         int ret;
 620
 621         if (PageWriteback(page) || PageDirty(page))
 622             return 0;
 623
 624         tree = &BTRFS_I(page->mapping->host)->io_tree;
 625         map = &BTRFS_I(page->mapping->host)->extent_tree;
 626
 627         ret = try_release_extent_state(map, tree, page, gfp_flags);
 628         if (!ret) {
 629                 return 0;
 630         }
 631
 632         ret = try_release_extent_buffer(tree, page);
 633         if (ret == 1) {
 634                 ClearPagePrivate(page);
 635                 set_page_private(page, 0);
 636                 page_cache_release(page);
 637         }
 638
 639         return ret;
 640 }
 641
 642 static void btree_invalidatepage(struct page *page, unsigned long offset)
 643 {
 644         struct extent_io_tree *tree;
 645         tree = &BTRFS_I(page->mapping->host)->io_tree;
 646         extent_invalidatepage(tree, page, offset);
 647         btree_releasepage(page, GFP_NOFS);
 648         if (PagePrivate(page)) {
 649                 printk("warning page private not zero on page %Lu\n",
 650                        page_offset(page));
 651                 ClearPagePrivate(page);
 652                 set_page_private(page, 0);
 653                 page_cache_release(page);
 654         }
 655 }
 656
 657 #if 0
 658 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 659 {
 660         struct buffer_head *bh;
 661         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 662         struct buffer_head *head;
 663         if (!page_has_buffers(page)) {
 664                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 665                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 666         }
 667         head = page_buffers(page);
 668         bh = head;
 669         do {
 670                 if (buffer_dirty(bh))
 671                         csum_tree_block(root, bh, 0);
 672                 bh = bh->b_this_page;
 673         } while (bh != head);
 674         return block_write_full_page(page, btree_get_block, wbc);
 675 }
 676 #endif
 677
 678 static struct address_space_operations btree_aops = {
 679         .readpage       = btree_readpage,
 680         .writepage      = btree_writepage,
 681         .writepages     = btree_writepages,
 682         .releasepage    = btree_releasepage,
 683         .invalidatepage = btree_invalidatepage,
 684         .sync_page      = block_sync_page,
 685 };
 686
 687 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 688                          u64 parent_transid)
 689 {
 690         struct extent_buffer *buf = NULL;
 691         struct inode *btree_inode = root->fs_info->btree_inode;
 692         int ret = 0;
 693
 694         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 695         if (!buf)
 696                 return 0;
 697         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 698                                  buf, 0, 0, btree_get_extent, 0);
 699         free_extent_buffer(buf);
 700         return ret;
 701 }
 702
 703 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 704                                             u64 bytenr, u32 blocksize)
 705 {
 706         struct inode *btree_inode = root->fs_info->btree_inode;
 707         struct extent_buffer *eb;
 708         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 709                                 bytenr, blocksize, GFP_NOFS);
 710         return eb;
 711 }
 712
 713 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 714                                                  u64 bytenr, u32 blocksize)
 715 {
 716         struct inode *btree_inode = root->fs_info->btree_inode;
 717         struct extent_buffer *eb;
 718
 719         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 720                                  bytenr, blocksize, NULL, GFP_NOFS);
 721         return eb;
 722 }
 723
 724
 725 int btrfs_write_tree_block(struct extent_buffer *buf)
 726 {
 727         return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
 728                                       buf->start + buf->len - 1, WB_SYNC_ALL);
 729 }
 730
 731 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 732 {
 733         return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
 734                                   buf->start, buf->start + buf->len -1);
 735 }
 736
 737 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 738                                       u32 blocksize, u64 parent_transid)
 739 {
 740         struct extent_buffer *buf = NULL;
 741         struct inode *btree_inode = root->fs_info->btree_inode;
 742         struct extent_io_tree *io_tree;
 743         int ret;
 744
 745         io_tree = &BTRFS_I(btree_inode)->io_tree;
 746
 747         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 748         if (!buf)
 749                 return NULL;
 750
 751         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 752
 753         if (ret == 0) {
 754                 buf->flags |= EXTENT_UPTODATE;
 755         } else {
 756                 WARN_ON(1);
 757         }
 758         return buf;
 759
 760 }
 761
 762 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 763                      struct extent_buffer *buf)
 764 {
 765         struct inode *btree_inode = root->fs_info->btree_inode;
 766         if (btrfs_header_generation(buf) ==
 767             root->fs_info->running_transaction->transid) {
 768                 WARN_ON(!btrfs_tree_locked(buf));
 769                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 770                                           buf);
 771         }
 772         return 0;
 773 }
 774
 775 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 776                         u32 stripesize, struct btrfs_root *root,
 777                         struct btrfs_fs_info *fs_info,
 778                         u64 objectid)
 779 {
 780         root->node = NULL;
 781         root->inode = NULL;
 782         root->commit_root = NULL;
 783         root->ref_tree = NULL;
 784         root->sectorsize = sectorsize;
 785         root->nodesize = nodesize;
 786         root->leafsize = leafsize;
 787         root->stripesize = stripesize;
 788         root->ref_cows = 0;
 789         root->track_dirty = 0;
 790
 791         root->fs_info = fs_info;
 792         root->objectid = objectid;
 793         root->last_trans = 0;
 794         root->highest_inode = 0;
 795         root->last_inode_alloc = 0;
 796         root->name = NULL;
 797         root->in_sysfs = 0;
 798
 799         INIT_LIST_HEAD(&root->dirty_list);
 800         INIT_LIST_HEAD(&root->orphan_list);
 801         INIT_LIST_HEAD(&root->dead_list);
 802         spin_lock_init(&root->node_lock);
 803         spin_lock_init(&root->list_lock);
 804         mutex_init(&root->objectid_mutex);
 805         mutex_init(&root->log_mutex);
 806         extent_io_tree_init(&root->dirty_log_pages,
 807                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 808
 809         btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
 810         root->ref_tree = &root->ref_tree_struct;
 811
 812         memset(&root->root_key, 0, sizeof(root->root_key));
 813         memset(&root->root_item, 0, sizeof(root->root_item));
 814         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 815         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 816         root->defrag_trans_start = fs_info->generation;
 817         init_completion(&root->kobj_unregister);
 818         root->defrag_running = 0;
 819         root->defrag_level = 0;
 820         root->root_key.objectid = objectid;
 821         return 0;
 822 }
 823
 824 static int find_and_setup_root(struct btrfs_root *tree_root,
 825                                struct btrfs_fs_info *fs_info,
 826                                u64 objectid,
 827                                struct btrfs_root *root)
 828 {
 829         int ret;
 830         u32 blocksize;
 831
 832         __setup_root(tree_root->nodesize, tree_root->leafsize,
 833                      tree_root->sectorsize, tree_root->stripesize,
 834                      root, fs_info, objectid);
 835         ret = btrfs_find_last_root(tree_root, objectid,
 836                                    &root->root_item, &root->root_key);
 837         BUG_ON(ret);
 838
 839         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 840         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 841                                      blocksize, 0);
 842         BUG_ON(!root->node);
 843         return 0;
 844 }
 845
 846 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
 847                              struct btrfs_fs_info *fs_info)
 848 {
 849         struct extent_buffer *eb;
 850         struct btrfs_root *log_root_tree = fs_info->log_root_tree;
 851         u64 start = 0;
 852         u64 end = 0;
 853         int ret;
 854
 855         if (!log_root_tree)
 856                 return 0;
 857
 858         while(1) {
 859                 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
 860                                     0, &start, &end, EXTENT_DIRTY);
 861                 if (ret)
 862                         break;
 863
 864                 clear_extent_dirty(&log_root_tree->dirty_log_pages,
 865                                    start, end, GFP_NOFS);
 866         }
 867         eb = fs_info->log_root_tree->node;
 868
 869         WARN_ON(btrfs_header_level(eb) != 0);
 870         WARN_ON(btrfs_header_nritems(eb) != 0);
 871
 872         ret = btrfs_free_reserved_extent(fs_info->tree_root,
 873                                 eb->start, eb->len);
 874         BUG_ON(ret);
 875
 876         free_extent_buffer(eb);
 877         kfree(fs_info->log_root_tree);
 878         fs_info->log_root_tree = NULL;
 879         return 0;
 880 }
 881
 882 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
 883                              struct btrfs_fs_info *fs_info)
 884 {
 885         struct btrfs_root *root;
 886         struct btrfs_root *tree_root = fs_info->tree_root;
 887
 888         root = kzalloc(sizeof(*root), GFP_NOFS);
 889         if (!root)
 890                 return -ENOMEM;
 891
 892         __setup_root(tree_root->nodesize, tree_root->leafsize,
 893                      tree_root->sectorsize, tree_root->stripesize,
 894                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
 895
 896         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
 897         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
 898         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
 899         root->ref_cows = 0;
 900
 901         root->node = btrfs_alloc_free_block(trans, root, root->leafsize,
 902                                             0, BTRFS_TREE_LOG_OBJECTID,
 903                                             trans->transid, 0, 0, 0);
 904
 905         btrfs_set_header_nritems(root->node, 0);
 906         btrfs_set_header_level(root->node, 0);
 907         btrfs_set_header_bytenr(root->node, root->node->start);
 908         btrfs_set_header_generation(root->node, trans->transid);
 909         btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
 910
 911         write_extent_buffer(root->node, root->fs_info->fsid,
 912                             (unsigned long)btrfs_header_fsid(root->node),
 913                             BTRFS_FSID_SIZE);
 914         btrfs_mark_buffer_dirty(root->node);
 915         btrfs_tree_unlock(root->node);
 916         fs_info->log_root_tree = root;
 917         return 0;
 918 }
 919
 920 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
 921                                                struct btrfs_key *location)
 922 {
 923         struct btrfs_root *root;
 924         struct btrfs_fs_info *fs_info = tree_root->fs_info;
 925         struct btrfs_path *path;
 926         struct extent_buffer *l;
 927         u64 highest_inode;
 928         u32 blocksize;
 929         int ret = 0;
 930
 931         root = kzalloc(sizeof(*root), GFP_NOFS);
 932         if (!root)
 933                 return ERR_PTR(-ENOMEM);
 934         if (location->offset == (u64)-1) {
 935                 ret = find_and_setup_root(tree_root, fs_info,
 936                                           location->objectid, root);
 937                 if (ret) {
 938                         kfree(root);
 939                         return ERR_PTR(ret);
 940                 }
 941                 goto insert;
 942         }
 943
 944         __setup_root(tree_root->nodesize, tree_root->leafsize,
 945                      tree_root->sectorsize, tree_root->stripesize,
 946                      root, fs_info, location->objectid);
 947
 948         path = btrfs_alloc_path();
 949         BUG_ON(!path);
 950         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 951         if (ret != 0) {
 952                 if (ret > 0)
 953                         ret = -ENOENT;
 954                 goto out;
 955         }
 956         l = path->nodes[0];
 957         read_extent_buffer(l, &root->root_item,
 958                btrfs_item_ptr_offset(l, path->slots[0]),
 959                sizeof(root->root_item));
 960         memcpy(&root->root_key, location, sizeof(*location));
 961         ret = 0;
 962 out:
 963         btrfs_release_path(root, path);
 964         btrfs_free_path(path);
 965         if (ret) {
 966                 kfree(root);
 967                 return ERR_PTR(ret);
 968         }
 969         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 970         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 971                                      blocksize, 0);
 972         BUG_ON(!root->node);
 973 insert:
 974         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
 975                 root->ref_cows = 1;
 976                 ret = btrfs_find_highest_inode(root, &highest_inode);
 977                 if (ret == 0) {
 978                         root->highest_inode = highest_inode;
 979                         root->last_inode_alloc = highest_inode;
 980                 }
 981         }
 982         return root;
 983 }
 984
 985 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
 986                                         u64 root_objectid)
 987 {
 988         struct btrfs_root *root;
 989
 990         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
 991                 return fs_info->tree_root;
 992         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
 993                 return fs_info->extent_root;
 994
 995         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 996                                  (unsigned long)root_objectid);
 997         return root;
 998 }
 999
1000 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1001                                               struct btrfs_key *location)
1002 {
1003         struct btrfs_root *root;
1004         int ret;
1005
1006         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1007                 return fs_info->tree_root;
1008         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1009                 return fs_info->extent_root;
1010         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1011                 return fs_info->chunk_root;
1012         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1013                 return fs_info->dev_root;
1014
1015         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1016                                  (unsigned long)location->objectid);
1017         if (root)
1018                 return root;
1019
1020         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1021         if (IS_ERR(root))
1022                 return root;
1023         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1024                                 (unsigned long)root->root_key.objectid,
1025                                 root);
1026         if (ret) {
1027                 free_extent_buffer(root->node);
1028                 kfree(root);
1029                 return ERR_PTR(ret);
1030         }
1031         ret = btrfs_find_dead_roots(fs_info->tree_root,
1032                                     root->root_key.objectid, root);
1033         BUG_ON(ret);
1034
1035         return root;
1036 }
1037
1038 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1039                                       struct btrfs_key *location,
1040                                       const char *name, int namelen)
1041 {
1042         struct btrfs_root *root;
1043         int ret;
1044
1045         root = btrfs_read_fs_root_no_name(fs_info, location);
1046         if (!root)
1047                 return NULL;
1048
1049         if (root->in_sysfs)
1050                 return root;
1051
1052         ret = btrfs_set_root_name(root, name, namelen);
1053         if (ret) {
1054                 free_extent_buffer(root->node);
1055                 kfree(root);
1056                 return ERR_PTR(ret);
1057         }
1058
1059         ret = btrfs_sysfs_add_root(root);
1060         if (ret) {
1061                 free_extent_buffer(root->node);
1062                 kfree(root->name);
1063                 kfree(root);
1064                 return ERR_PTR(ret);
1065         }
1066         root->in_sysfs = 1;
1067         return root;
1068 }
1069 #if 0
1070 static int add_hasher(struct btrfs_fs_info *info, char *type) {
1071         struct btrfs_hasher *hasher;
1072
1073         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
1074         if (!hasher)
1075                 return -ENOMEM;
1076         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
1077         if (!hasher->hash_tfm) {
1078                 kfree(hasher);
1079                 return -EINVAL;
1080         }
1081         spin_lock(&info->hash_lock);
1082         list_add(&hasher->list, &info->hashers);
1083         spin_unlock(&info->hash_lock);
1084         return 0;
1085 }
1086 #endif
1087
1088 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1089 {
1090         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1091         int ret = 0;
1092         struct list_head *cur;
1093         struct btrfs_device *device;
1094         struct backing_dev_info *bdi;
1095
1096         if ((bdi_bits & (1 << BDI_write_congested)) &&
1097             btrfs_congested_async(info, 0))
1098                 return 1;
1099
1100         list_for_each(cur, &info->fs_devices->devices) {
1101                 device = list_entry(cur, struct btrfs_device, dev_list);
1102                 if (!device->bdev)
1103                         continue;
1104                 bdi = blk_get_backing_dev_info(device->bdev);
1105                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1106                         ret = 1;
1107                         break;
1108                 }
1109         }
1110         return ret;
1111 }
1112
1113 /*
1114  * this unplugs every device on the box, and it is only used when page
1115  * is null
1116  */
1117 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1118 {
1119         struct list_head *cur;
1120         struct btrfs_device *device;
1121         struct btrfs_fs_info *info;
1122
1123         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1124         list_for_each(cur, &info->fs_devices->devices) {
1125                 device = list_entry(cur, struct btrfs_device, dev_list);
1126                 bdi = blk_get_backing_dev_info(device->bdev);
1127                 if (bdi->unplug_io_fn) {
1128                         bdi->unplug_io_fn(bdi, page);
1129                 }
1130         }
1131 }
1132
1133 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1134 {
1135         struct inode *inode;
1136         struct extent_map_tree *em_tree;
1137         struct extent_map *em;
1138         struct address_space *mapping;
1139         u64 offset;
1140
1141         /* the generic O_DIRECT read code does this */
1142         if (!page) {
1143                 __unplug_io_fn(bdi, page);
1144                 return;
1145         }
1146
1147         /*
1148          * page->mapping may change at any time.  Get a consistent copy
1149          * and use that for everything below
1150          */
1151         smp_mb();
1152         mapping = page->mapping;
1153         if (!mapping)
1154                 return;
1155
1156         inode = mapping->host;
1157         offset = page_offset(page);
1158
1159         em_tree = &BTRFS_I(inode)->extent_tree;
1160         spin_lock(&em_tree->lock);
1161         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1162         spin_unlock(&em_tree->lock);
1163         if (!em) {
1164                 __unplug_io_fn(bdi, page);
1165                 return;
1166         }
1167
1168         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1169                 free_extent_map(em);
1170                 __unplug_io_fn(bdi, page);
1171                 return;
1172         }
1173         offset = offset - em->start;
1174         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1175                           em->block_start + offset, page);
1176         free_extent_map(em);
1177 }
1178
1179 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1180 {
1181         bdi_init(bdi);
1182         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1183         bdi->state              = 0;
1184         bdi->capabilities       = default_backing_dev_info.capabilities;
1185         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1186         bdi->unplug_io_data     = info;
1187         bdi->congested_fn       = btrfs_congested_fn;
1188         bdi->congested_data     = info;
1189         return 0;
1190 }
1191
1192 static int bio_ready_for_csum(struct bio *bio)
1193 {
1194         u64 length = 0;
1195         u64 buf_len = 0;
1196         u64 start = 0;
1197         struct page *page;
1198         struct extent_io_tree *io_tree = NULL;
1199         struct btrfs_fs_info *info = NULL;
1200         struct bio_vec *bvec;
1201         int i;
1202         int ret;
1203
1204         bio_for_each_segment(bvec, bio, i) {
1205                 page = bvec->bv_page;
1206                 if (page->private == EXTENT_PAGE_PRIVATE) {
1207                         length += bvec->bv_len;
1208                         continue;
1209                 }
1210                 if (!page->private) {
1211                         length += bvec->bv_len;
1212                         continue;
1213                 }
1214                 length = bvec->bv_len;
1215                 buf_len = page->private >> 2;
1216                 start = page_offset(page) + bvec->bv_offset;
1217                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1218                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1219         }
1220         /* are we fully contained in this bio? */
1221         if (buf_len <= length)
1222                 return 1;
1223
1224         ret = extent_range_uptodate(io_tree, start + length,
1225                                     start + buf_len - 1);
1226         if (ret == 1)
1227                 return ret;
1228         return ret;
1229 }
1230
1231 /*
1232  * called by the kthread helper functions to finally call the bio end_io
1233  * functions.  This is where read checksum verification actually happens
1234  */
1235 static void end_workqueue_fn(struct btrfs_work *work)
1236 {
1237         struct bio *bio;
1238         struct end_io_wq *end_io_wq;
1239         struct btrfs_fs_info *fs_info;
1240         int error;
1241
1242         end_io_wq = container_of(work, struct end_io_wq, work);
1243         bio = end_io_wq->bio;
1244         fs_info = end_io_wq->info;
1245
1246         /* metadata bios are special because the whole tree block must
1247          * be checksummed at once.  This makes sure the entire block is in
1248          * ram and up to date before trying to verify things.  For
1249          * blocksize <= pagesize, it is basically a noop
1250          */
1251         if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1252                 btrfs_queue_worker(&fs_info->endio_workers,
1253                                    &end_io_wq->work);
1254                 return;
1255         }
1256         error = end_io_wq->error;
1257         bio->bi_private = end_io_wq->private;
1258         bio->bi_end_io = end_io_wq->end_io;
1259         kfree(end_io_wq);
1260         bio_endio(bio, error);
1261 }
1262
1263 static int cleaner_kthread(void *arg)
1264 {
1265         struct btrfs_root *root = arg;
1266
1267         do {
1268                 smp_mb();
1269                 if (root->fs_info->closing)
1270                         break;
1271
1272                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1273                 mutex_lock(&root->fs_info->cleaner_mutex);
1274                 btrfs_clean_old_snapshots(root);
1275                 mutex_unlock(&root->fs_info->cleaner_mutex);
1276
1277                 if (freezing(current)) {
1278                         refrigerator();
1279                 } else {
1280                         smp_mb();
1281                         if (root->fs_info->closing)
1282                                 break;
1283                         set_current_state(TASK_INTERRUPTIBLE);
1284                         schedule();
1285                         __set_current_state(TASK_RUNNING);
1286                 }
1287         } while (!kthread_should_stop());
1288         return 0;
1289 }
1290
1291 static int transaction_kthread(void *arg)
1292 {
1293         struct btrfs_root *root = arg;
1294         struct btrfs_trans_handle *trans;
1295         struct btrfs_transaction *cur;
1296         unsigned long now;
1297         unsigned long delay;
1298         int ret;
1299
1300         do {
1301                 smp_mb();
1302                 if (root->fs_info->closing)
1303                         break;
1304
1305                 delay = HZ * 30;
1306                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1307                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1308
1309                 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1310                         printk("btrfs: total reference cache size %Lu\n",
1311                                 root->fs_info->total_ref_cache_size);
1312                 }
1313
1314                 mutex_lock(&root->fs_info->trans_mutex);
1315                 cur = root->fs_info->running_transaction;
1316                 if (!cur) {
1317                         mutex_unlock(&root->fs_info->trans_mutex);
1318                         goto sleep;
1319                 }
1320
1321                 now = get_seconds();
1322                 if (now < cur->start_time || now - cur->start_time < 30) {
1323                         mutex_unlock(&root->fs_info->trans_mutex);
1324                         delay = HZ * 5;
1325                         goto sleep;
1326                 }
1327                 mutex_unlock(&root->fs_info->trans_mutex);
1328                 trans = btrfs_start_transaction(root, 1);
1329                 ret = btrfs_commit_transaction(trans, root);
1330 sleep:
1331                 wake_up_process(root->fs_info->cleaner_kthread);
1332                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1333
1334                 if (freezing(current)) {
1335                         refrigerator();
1336                 } else {
1337                         if (root->fs_info->closing)
1338                                 break;
1339                         set_current_state(TASK_INTERRUPTIBLE);
1340                         schedule_timeout(delay);
1341                         __set_current_state(TASK_RUNNING);
1342                 }
1343         } while (!kthread_should_stop());
1344         return 0;
1345 }
1346
1347 struct btrfs_root *open_ctree(struct super_block *sb,
1348                               struct btrfs_fs_devices *fs_devices,
1349                               char *options)
1350 {
1351         u32 sectorsize;
1352         u32 nodesize;
1353         u32 leafsize;
1354         u32 blocksize;
1355         u32 stripesize;
1356         struct buffer_head *bh;
1357         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1358                                                  GFP_NOFS);
1359         struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1360                                                GFP_NOFS);
1361         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1362                                                 GFP_NOFS);
1363         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1364                                                 GFP_NOFS);
1365         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1366                                               GFP_NOFS);
1367         struct btrfs_root *log_tree_root;
1368
1369         int ret;
1370         int err = -EINVAL;
1371
1372         struct btrfs_super_block *disk_super;
1373
1374         if (!extent_root || !tree_root || !fs_info ||
1375             !chunk_root || !dev_root) {
1376                 err = -ENOMEM;
1377                 goto fail;
1378         }
1379         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1380         INIT_LIST_HEAD(&fs_info->trans_list);
1381         INIT_LIST_HEAD(&fs_info->dead_roots);
1382         INIT_LIST_HEAD(&fs_info->hashers);
1383         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1384         spin_lock_init(&fs_info->hash_lock);
1385         spin_lock_init(&fs_info->delalloc_lock);
1386         spin_lock_init(&fs_info->new_trans_lock);
1387         spin_lock_init(&fs_info->ref_cache_lock);
1388
1389         init_completion(&fs_info->kobj_unregister);
1390         fs_info->tree_root = tree_root;
1391         fs_info->extent_root = extent_root;
1392         fs_info->chunk_root = chunk_root;
1393         fs_info->dev_root = dev_root;
1394         fs_info->fs_devices = fs_devices;
1395         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1396         INIT_LIST_HEAD(&fs_info->space_info);
1397         btrfs_mapping_init(&fs_info->mapping_tree);
1398         atomic_set(&fs_info->nr_async_submits, 0);
1399         atomic_set(&fs_info->async_submit_draining, 0);
1400         atomic_set(&fs_info->nr_async_bios, 0);
1401         atomic_set(&fs_info->throttles, 0);
1402         atomic_set(&fs_info->throttle_gen, 0);
1403         fs_info->sb = sb;
1404         fs_info->max_extent = (u64)-1;
1405         fs_info->max_inline = 8192 * 1024;
1406         setup_bdi(fs_info, &fs_info->bdi);
1407         fs_info->btree_inode = new_inode(sb);
1408         fs_info->btree_inode->i_ino = 1;
1409         fs_info->btree_inode->i_nlink = 1;
1410         fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1411
1412         INIT_LIST_HEAD(&fs_info->ordered_extents);
1413         spin_lock_init(&fs_info->ordered_extent_lock);
1414
1415         sb->s_blocksize = 4096;
1416         sb->s_blocksize_bits = blksize_bits(4096);
1417
1418         /*
1419          * we set the i_size on the btree inode to the max possible int.
1420          * the real end of the address space is determined by all of
1421          * the devices in the system
1422          */
1423         fs_info->btree_inode->i_size = OFFSET_MAX;
1424         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1425         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1426
1427         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1428                              fs_info->btree_inode->i_mapping,
1429                              GFP_NOFS);
1430         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1431                              GFP_NOFS);
1432
1433         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1434
1435         spin_lock_init(&fs_info->block_group_cache_lock);
1436         fs_info->block_group_cache_tree.rb_node = NULL;
1437
1438         extent_io_tree_init(&fs_info->pinned_extents,
1439                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1440         extent_io_tree_init(&fs_info->pending_del,
1441                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1442         extent_io_tree_init(&fs_info->extent_ins,
1443                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1444         fs_info->do_barriers = 1;
1445
1446         extent_io_tree_init(&fs_info->reloc_mapping_tree,
1447                             fs_info->btree_inode->i_mapping, GFP_NOFS);
1448         INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
1449         btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
1450         btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
1451
1452         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1453         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1454                sizeof(struct btrfs_key));
1455         insert_inode_hash(fs_info->btree_inode);
1456
1457         mutex_init(&fs_info->trans_mutex);
1458         mutex_init(&fs_info->tree_log_mutex);
1459         mutex_init(&fs_info->drop_mutex);
1460         mutex_init(&fs_info->alloc_mutex);
1461         mutex_init(&fs_info->chunk_mutex);
1462         mutex_init(&fs_info->transaction_kthread_mutex);
1463         mutex_init(&fs_info->cleaner_mutex);
1464         mutex_init(&fs_info->volume_mutex);
1465         mutex_init(&fs_info->tree_reloc_mutex);
1466         init_waitqueue_head(&fs_info->transaction_throttle);
1467         init_waitqueue_head(&fs_info->transaction_wait);
1468         init_waitqueue_head(&fs_info->async_submit_wait);
1469         init_waitqueue_head(&fs_info->tree_log_wait);
1470         atomic_set(&fs_info->tree_log_commit, 0);
1471         atomic_set(&fs_info->tree_log_writers, 0);
1472         fs_info->tree_log_transid = 0;
1473
1474 #if 0
1475         ret = add_hasher(fs_info, "crc32c");
1476         if (ret) {
1477                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1478                 err = -ENOMEM;
1479                 goto fail_iput;
1480         }
1481 #endif
1482         __setup_root(4096, 4096, 4096, 4096, tree_root,
1483                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1484
1485
1486         bh = __bread(fs_devices->latest_bdev,
1487                      BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1488         if (!bh)
1489                 goto fail_iput;
1490
1491         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1492         brelse(bh);
1493
1494         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1495
1496         disk_super = &fs_info->super_copy;
1497         if (!btrfs_super_root(disk_super))
1498                 goto fail_sb_buffer;
1499
1500         err = btrfs_parse_options(tree_root, options);
1501         if (err)
1502                 goto fail_sb_buffer;
1503
1504         /*
1505          * we need to start all the end_io workers up front because the
1506          * queue work function gets called at interrupt time, and so it
1507          * cannot dynamically grow.
1508          */
1509         btrfs_init_workers(&fs_info->workers, "worker",
1510                            fs_info->thread_pool_size);
1511         btrfs_init_workers(&fs_info->submit_workers, "submit",
1512                            min_t(u64, fs_devices->num_devices,
1513                            fs_info->thread_pool_size));
1514
1515         /* a higher idle thresh on the submit workers makes it much more
1516          * likely that bios will be send down in a sane order to the
1517          * devices
1518          */
1519         fs_info->submit_workers.idle_thresh = 64;
1520
1521         /* fs_info->workers is responsible for checksumming file data
1522          * blocks and metadata.  Using a larger idle thresh allows each
1523          * worker thread to operate on things in roughly the order they
1524          * were sent by the writeback daemons, improving overall locality
1525          * of the IO going down the pipe.
1526          */
1527         fs_info->workers.idle_thresh = 128;
1528
1529         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1530         btrfs_init_workers(&fs_info->endio_workers, "endio",
1531                            fs_info->thread_pool_size);
1532         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1533                            fs_info->thread_pool_size);
1534
1535         /*
1536          * endios are largely parallel and should have a very
1537          * low idle thresh
1538          */
1539         fs_info->endio_workers.idle_thresh = 4;
1540         fs_info->endio_write_workers.idle_thresh = 64;
1541
1542         btrfs_start_workers(&fs_info->workers, 1);
1543         btrfs_start_workers(&fs_info->submit_workers, 1);
1544         btrfs_start_workers(&fs_info->fixup_workers, 1);
1545         btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1546         btrfs_start_workers(&fs_info->endio_write_workers,
1547                             fs_info->thread_pool_size);
1548
1549         err = -EINVAL;
1550         if (btrfs_super_num_devices(disk_super) > fs_devices->open_devices) {
1551                 printk("Btrfs: wanted %llu devices, but found %llu\n",
1552                        (unsigned long long)btrfs_super_num_devices(disk_super),
1553                        (unsigned long long)fs_devices->open_devices);
1554                 if (btrfs_test_opt(tree_root, DEGRADED))
1555                         printk("continuing in degraded mode\n");
1556                 else {
1557                         goto fail_sb_buffer;
1558                 }
1559         }
1560
1561         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1562
1563         nodesize = btrfs_super_nodesize(disk_super);
1564         leafsize = btrfs_super_leafsize(disk_super);
1565         sectorsize = btrfs_super_sectorsize(disk_super);
1566         stripesize = btrfs_super_stripesize(disk_super);
1567         tree_root->nodesize = nodesize;
1568         tree_root->leafsize = leafsize;
1569         tree_root->sectorsize = sectorsize;
1570         tree_root->stripesize = stripesize;
1571
1572         sb->s_blocksize = sectorsize;
1573         sb->s_blocksize_bits = blksize_bits(sectorsize);
1574
1575         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1576                     sizeof(disk_super->magic))) {
1577                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1578                 goto fail_sb_buffer;
1579         }
1580
1581         mutex_lock(&fs_info->chunk_mutex);
1582         ret = btrfs_read_sys_array(tree_root);
1583         mutex_unlock(&fs_info->chunk_mutex);
1584         if (ret) {
1585                 printk("btrfs: failed to read the system array on %s\n",
1586                        sb->s_id);
1587                 goto fail_sys_array;
1588         }
1589
1590         blocksize = btrfs_level_size(tree_root,
1591                                      btrfs_super_chunk_root_level(disk_super));
1592
1593         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1594                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1595
1596         chunk_root->node = read_tree_block(chunk_root,
1597                                            btrfs_super_chunk_root(disk_super),
1598                                            blocksize, 0);
1599         BUG_ON(!chunk_root->node);
1600
1601         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1602                  (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1603                  BTRFS_UUID_SIZE);
1604
1605         mutex_lock(&fs_info->chunk_mutex);
1606         ret = btrfs_read_chunk_tree(chunk_root);
1607         mutex_unlock(&fs_info->chunk_mutex);
1608         BUG_ON(ret);
1609
1610         btrfs_close_extra_devices(fs_devices);
1611
1612         blocksize = btrfs_level_size(tree_root,
1613                                      btrfs_super_root_level(disk_super));
1614
1615
1616         tree_root->node = read_tree_block(tree_root,
1617                                           btrfs_super_root(disk_super),
1618                                           blocksize, 0);
1619         if (!tree_root->node)
1620                 goto fail_sb_buffer;
1621
1622
1623         ret = find_and_setup_root(tree_root, fs_info,
1624                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1625         if (ret)
1626                 goto fail_tree_root;
1627         extent_root->track_dirty = 1;
1628
1629         ret = find_and_setup_root(tree_root, fs_info,
1630                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1631         dev_root->track_dirty = 1;
1632
1633         if (ret)
1634                 goto fail_extent_root;
1635
1636         btrfs_read_block_groups(extent_root);
1637
1638         fs_info->generation = btrfs_super_generation(disk_super) + 1;
1639         fs_info->data_alloc_profile = (u64)-1;
1640         fs_info->metadata_alloc_profile = (u64)-1;
1641         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1642         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1643                                                "btrfs-cleaner");
1644         if (!fs_info->cleaner_kthread)
1645                 goto fail_extent_root;
1646
1647         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1648                                                    tree_root,
1649                                                    "btrfs-transaction");
1650         if (!fs_info->transaction_kthread)
1651                 goto fail_cleaner;
1652
1653         if (btrfs_super_log_root(disk_super) != 0) {
1654                 u32 blocksize;
1655                 u64 bytenr = btrfs_super_log_root(disk_super);
1656
1657                 blocksize =
1658                      btrfs_level_size(tree_root,
1659                                       btrfs_super_log_root_level(disk_super));
1660
1661                 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1662                                                       GFP_NOFS);
1663
1664                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1665                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1666
1667                 log_tree_root->node = read_tree_block(tree_root, bytenr,
1668                                                       blocksize, 0);
1669                 ret = btrfs_recover_log_trees(log_tree_root);
1670                 BUG_ON(ret);
1671         }
1672
1673         ret = btrfs_cleanup_reloc_trees(tree_root);
1674         BUG_ON(ret);
1675
1676         fs_info->last_trans_committed = btrfs_super_generation(disk_super);
1677         return tree_root;
1678
1679 fail_cleaner:
1680         kthread_stop(fs_info->cleaner_kthread);
1681 fail_extent_root:
1682         free_extent_buffer(extent_root->node);
1683 fail_tree_root:
1684         free_extent_buffer(tree_root->node);
1685 fail_sys_array:
1686 fail_sb_buffer:
1687         btrfs_stop_workers(&fs_info->fixup_workers);
1688         btrfs_stop_workers(&fs_info->workers);
1689         btrfs_stop_workers(&fs_info->endio_workers);
1690         btrfs_stop_workers(&fs_info->endio_write_workers);
1691         btrfs_stop_workers(&fs_info->submit_workers);
1692 fail_iput:
1693         iput(fs_info->btree_inode);
1694 fail:
1695         btrfs_close_devices(fs_info->fs_devices);
1696         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1697
1698         kfree(extent_root);
1699         kfree(tree_root);
1700         bdi_destroy(&fs_info->bdi);
1701         kfree(fs_info);
1702         kfree(chunk_root);
1703         kfree(dev_root);
1704         return ERR_PTR(err);
1705 }
1706
1707 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1708 {
1709         char b[BDEVNAME_SIZE];
1710
1711         if (uptodate) {
1712                 set_buffer_uptodate(bh);
1713         } else {
1714                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1715                         printk(KERN_WARNING "lost page write due to "
1716                                         "I/O error on %s\n",
1717                                        bdevname(bh->b_bdev, b));
1718                 }
1719                 /* note, we dont' set_buffer_write_io_error because we have
1720                  * our own ways of dealing with the IO errors
1721                  */
1722                 clear_buffer_uptodate(bh);
1723         }
1724         unlock_buffer(bh);
1725         put_bh(bh);
1726 }
1727
1728 int write_all_supers(struct btrfs_root *root)
1729 {
1730         struct list_head *cur;
1731         struct list_head *head = &root->fs_info->fs_devices->devices;
1732         struct btrfs_device *dev;
1733         struct btrfs_super_block *sb;
1734         struct btrfs_dev_item *dev_item;
1735         struct buffer_head *bh;
1736         int ret;
1737         int do_barriers;
1738         int max_errors;
1739         int total_errors = 0;
1740         u32 crc;
1741         u64 flags;
1742
1743         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1744         do_barriers = !btrfs_test_opt(root, NOBARRIER);
1745
1746         sb = &root->fs_info->super_for_commit;
1747         dev_item = &sb->dev_item;
1748         list_for_each(cur, head) {
1749                 dev = list_entry(cur, struct btrfs_device, dev_list);
1750                 if (!dev->bdev) {
1751                         total_errors++;
1752                         continue;
1753                 }
1754                 if (!dev->in_fs_metadata)
1755                         continue;
1756
1757                 btrfs_set_stack_device_type(dev_item, dev->type);
1758                 btrfs_set_stack_device_id(dev_item, dev->devid);
1759                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1760                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1761                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1762                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1763                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1764                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1765                 flags = btrfs_super_flags(sb);
1766                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1767
1768
1769                 crc = ~(u32)0;
1770                 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1771                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1772                 btrfs_csum_final(crc, sb->csum);
1773
1774                 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1775                               BTRFS_SUPER_INFO_SIZE);
1776
1777                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1778                 dev->pending_io = bh;
1779
1780                 get_bh(bh);
1781                 set_buffer_uptodate(bh);
1782                 lock_buffer(bh);
1783                 bh->b_end_io = btrfs_end_buffer_write_sync;
1784
1785                 if (do_barriers && dev->barriers) {
1786                         ret = submit_bh(WRITE_BARRIER, bh);
1787                         if (ret == -EOPNOTSUPP) {
1788                                 printk("btrfs: disabling barriers on dev %s\n",
1789                                        dev->name);
1790                                 set_buffer_uptodate(bh);
1791                                 dev->barriers = 0;
1792                                 get_bh(bh);
1793                                 lock_buffer(bh);
1794                                 ret = submit_bh(WRITE, bh);
1795                         }
1796                 } else {
1797                         ret = submit_bh(WRITE, bh);
1798                 }
1799                 if (ret)
1800                         total_errors++;
1801         }
1802         if (total_errors > max_errors) {
1803                 printk("btrfs: %d errors while writing supers\n", total_errors);
1804                 BUG();
1805         }
1806         total_errors = 0;
1807
1808         list_for_each(cur, head) {
1809                 dev = list_entry(cur, struct btrfs_device, dev_list);
1810                 if (!dev->bdev)
1811                         continue;
1812                 if (!dev->in_fs_metadata)
1813                         continue;
1814
1815                 BUG_ON(!dev->pending_io);
1816                 bh = dev->pending_io;
1817                 wait_on_buffer(bh);
1818                 if (!buffer_uptodate(dev->pending_io)) {
1819                         if (do_barriers && dev->barriers) {
1820                                 printk("btrfs: disabling barriers on dev %s\n",
1821                                        dev->name);
1822                                 set_buffer_uptodate(bh);
1823                                 get_bh(bh);
1824                                 lock_buffer(bh);
1825                                 dev->barriers = 0;
1826                                 ret = submit_bh(WRITE, bh);
1827                                 BUG_ON(ret);
1828                                 wait_on_buffer(bh);
1829                                 if (!buffer_uptodate(bh))
1830                                         total_errors++;
1831                         } else {
1832                                 total_errors++;
1833                         }
1834
1835                 }
1836                 dev->pending_io = NULL;
1837                 brelse(bh);
1838         }
1839         if (total_errors > max_errors) {
1840                 printk("btrfs: %d errors while writing supers\n", total_errors);
1841                 BUG();
1842         }
1843         return 0;
1844 }
1845
1846 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1847                       *root)
1848 {
1849         int ret;
1850
1851         ret = write_all_supers(root);
1852         return ret;
1853 }
1854
1855 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1856 {
1857         radix_tree_delete(&fs_info->fs_roots_radix,
1858                           (unsigned long)root->root_key.objectid);
1859         if (root->in_sysfs)
1860                 btrfs_sysfs_del_root(root);
1861         if (root->inode)
1862                 iput(root->inode);
1863         if (root->node)
1864                 free_extent_buffer(root->node);
1865         if (root->commit_root)
1866                 free_extent_buffer(root->commit_root);
1867         if (root->name)
1868                 kfree(root->name);
1869         kfree(root);
1870         return 0;
1871 }
1872
1873 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1874 {
1875         int ret;
1876         struct btrfs_root *gang[8];
1877         int i;
1878
1879         while(1) {
1880                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1881                                              (void **)gang, 0,
1882                                              ARRAY_SIZE(gang));
1883                 if (!ret)
1884                         break;
1885                 for (i = 0; i < ret; i++)
1886                         btrfs_free_fs_root(fs_info, gang[i]);
1887         }
1888         return 0;
1889 }
1890
1891 int close_ctree(struct btrfs_root *root)
1892 {
1893         int ret;
1894         struct btrfs_trans_handle *trans;
1895         struct btrfs_fs_info *fs_info = root->fs_info;
1896
1897         fs_info->closing = 1;
1898         smp_mb();
1899
1900         kthread_stop(root->fs_info->transaction_kthread);
1901         kthread_stop(root->fs_info->cleaner_kthread);
1902
1903         btrfs_clean_old_snapshots(root);
1904         trans = btrfs_start_transaction(root, 1);
1905         ret = btrfs_commit_transaction(trans, root);
1906         /* run commit again to  drop the original snapshot */
1907         trans = btrfs_start_transaction(root, 1);
1908         btrfs_commit_transaction(trans, root);
1909         ret = btrfs_write_and_wait_transaction(NULL, root);
1910         BUG_ON(ret);
1911
1912         write_ctree_super(NULL, root);
1913
1914         if (fs_info->delalloc_bytes) {
1915                 printk("btrfs: at unmount delalloc count %Lu\n",
1916                        fs_info->delalloc_bytes);
1917         }
1918         if (fs_info->total_ref_cache_size) {
1919                 printk("btrfs: at umount reference cache size %Lu\n",
1920                         fs_info->total_ref_cache_size);
1921         }
1922
1923         if (fs_info->extent_root->node)
1924                 free_extent_buffer(fs_info->extent_root->node);
1925
1926         if (fs_info->tree_root->node)
1927                 free_extent_buffer(fs_info->tree_root->node);
1928
1929         if (root->fs_info->chunk_root->node);
1930                 free_extent_buffer(root->fs_info->chunk_root->node);
1931
1932         if (root->fs_info->dev_root->node);
1933                 free_extent_buffer(root->fs_info->dev_root->node);
1934
1935         btrfs_free_block_groups(root->fs_info);
1936         fs_info->closing = 2;
1937         del_fs_roots(fs_info);
1938
1939         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1940
1941         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1942
1943         btrfs_stop_workers(&fs_info->fixup_workers);
1944         btrfs_stop_workers(&fs_info->workers);
1945         btrfs_stop_workers(&fs_info->endio_workers);
1946         btrfs_stop_workers(&fs_info->endio_write_workers);
1947         btrfs_stop_workers(&fs_info->submit_workers);
1948
1949         iput(fs_info->btree_inode);
1950 #if 0
1951         while(!list_empty(&fs_info->hashers)) {
1952                 struct btrfs_hasher *hasher;
1953                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1954                                     hashers);
1955                 list_del(&hasher->hashers);
1956                 crypto_free_hash(&fs_info->hash_tfm);
1957                 kfree(hasher);
1958         }
1959 #endif
1960         btrfs_close_devices(fs_info->fs_devices);
1961         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1962
1963         bdi_destroy(&fs_info->bdi);
1964
1965         kfree(fs_info->extent_root);
1966         kfree(fs_info->tree_root);
1967         kfree(fs_info->chunk_root);
1968         kfree(fs_info->dev_root);
1969         return 0;
1970 }
1971
1972 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1973 {
1974         int ret;
1975         struct inode *btree_inode = buf->first_page->mapping->host;
1976
1977         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1978         if (!ret)
1979                 return ret;
1980
1981         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
1982                                     parent_transid);
1983         return !ret;
1984 }
1985
1986 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1987 {
1988         struct inode *btree_inode = buf->first_page->mapping->host;
1989         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1990                                           buf);
1991 }
1992
1993 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1994 {
1995         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1996         u64 transid = btrfs_header_generation(buf);
1997         struct inode *btree_inode = root->fs_info->btree_inode;
1998
1999         WARN_ON(!btrfs_tree_locked(buf));
2000         if (transid != root->fs_info->generation) {
2001                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
2002                         (unsigned long long)buf->start,
2003                         transid, root->fs_info->generation);
2004                 WARN_ON(1);
2005         }
2006         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
2007 }
2008
2009 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2010 {
2011         /*
2012          * looks as though older kernels can get into trouble with
2013          * this code, they end up stuck in balance_dirty_pages forever
2014          */
2015         struct extent_io_tree *tree;
2016         u64 num_dirty;
2017         u64 start = 0;
2018         unsigned long thresh = 96 * 1024 * 1024;
2019         tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
2020
2021         if (current_is_pdflush() || current->flags & PF_MEMALLOC)
2022                 return;
2023
2024         num_dirty = count_range_bits(tree, &start, (u64)-1,
2025                                      thresh, EXTENT_DIRTY);
2026         if (num_dirty > thresh) {
2027                 balance_dirty_pages_ratelimited_nr(
2028                                    root->fs_info->btree_inode->i_mapping, 1);
2029         }
2030         return;
2031 }
2032
2033 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2034 {
2035         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2036         int ret;
2037         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2038         if (ret == 0) {
2039                 buf->flags |= EXTENT_UPTODATE;
2040         }
2041         return ret;
2042 }
2043
2044 int btree_lock_page_hook(struct page *page)
2045 {
2046         struct inode *inode = page->mapping->host;
2047         struct btrfs_root *root = BTRFS_I(inode)->root;
2048         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2049         struct extent_buffer *eb;
2050         unsigned long len;
2051         u64 bytenr = page_offset(page);
2052
2053         if (page->private == EXTENT_PAGE_PRIVATE)
2054                 goto out;
2055
2056         len = page->private >> 2;
2057         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2058         if (!eb)
2059                 goto out;
2060
2061         btrfs_tree_lock(eb);
2062         spin_lock(&root->fs_info->hash_lock);
2063         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2064         spin_unlock(&root->fs_info->hash_lock);
2065         btrfs_tree_unlock(eb);
2066         free_extent_buffer(eb);
2067 out:
2068         lock_page(page);
2069         return 0;
2070 }
2071
2072 static struct extent_io_ops btree_extent_io_ops = {
2073         .write_cache_pages_lock_hook = btree_lock_page_hook,
2074         .readpage_end_io_hook = btree_readpage_end_io_hook,
2075         .submit_bio_hook = btree_submit_bio_hook,
2076         /* note we're sharing with inode.c for the merge bio hook */
2077         .merge_bio_hook = btrfs_merge_bio_hook,
2078 };