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 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,20)
  30 # include <linux/freezer.h>
  31 #else
  32 # include <linux/sched.h>
  33 #endif
  34 #include "crc32c.h"
  35 #include "ctree.h"
  36 #include "disk-io.h"
  37 #include "transaction.h"
  38 #include "btrfs_inode.h"
  39 #include "volumes.h"
  40 #include "print-tree.h"
  41 #include "async-thread.h"
  42 #include "locking.h"
  43 #include "ref-cache.h"
  44 #include "tree-log.h"
  45
  46 #if 0
  47 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
  48 {
  49         if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
  50                 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
  51                        (unsigned long long)extent_buffer_blocknr(buf),
  52                        (unsigned long long)btrfs_header_blocknr(buf));
  53                 return 1;
  54         }
  55         return 0;
  56 }
  57 #endif
  58
  59 static struct extent_io_ops btree_extent_io_ops;
  60 static void end_workqueue_fn(struct btrfs_work *work);
  61
  62 struct end_io_wq {
  63         struct bio *bio;
  64         bio_end_io_t *end_io;
  65         void *private;
  66         struct btrfs_fs_info *info;
  67         int error;
  68         int metadata;
  69         struct list_head list;
  70         struct btrfs_work work;
  71 };
  72
  73 struct async_submit_bio {
  74         struct inode *inode;
  75         struct bio *bio;
  76         struct list_head list;
  77         extent_submit_bio_hook_t *submit_bio_hook;
  78         int rw;
  79         int mirror_num;
  80         struct btrfs_work work;
  81 };
  82
  83 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
  84                                     size_t page_offset, u64 start, u64 len,
  85                                     int create)
  86 {
  87         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
  88         struct extent_map *em;
  89         int ret;
  90
  91         spin_lock(&em_tree->lock);
  92         em = lookup_extent_mapping(em_tree, start, len);
  93         if (em) {
  94                 em->bdev =
  95                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
  96                 spin_unlock(&em_tree->lock);
  97                 goto out;
  98         }
  99         spin_unlock(&em_tree->lock);
 100
 101         em = alloc_extent_map(GFP_NOFS);
 102         if (!em) {
 103                 em = ERR_PTR(-ENOMEM);
 104                 goto out;
 105         }
 106         em->start = 0;
 107         em->len = (u64)-1;
 108         em->block_start = 0;
 109         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 110
 111         spin_lock(&em_tree->lock);
 112         ret = add_extent_mapping(em_tree, em);
 113         if (ret == -EEXIST) {
 114                 u64 failed_start = em->start;
 115                 u64 failed_len = em->len;
 116
 117                 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
 118                        em->start, em->len, em->block_start);
 119                 free_extent_map(em);
 120                 em = lookup_extent_mapping(em_tree, start, len);
 121                 if (em) {
 122                         printk("after failing, found %Lu %Lu %Lu\n",
 123                                em->start, em->len, em->block_start);
 124                         ret = 0;
 125                 } else {
 126                         em = lookup_extent_mapping(em_tree, failed_start,
 127                                                    failed_len);
 128                         if (em) {
 129                                 printk("double failure lookup gives us "
 130                                        "%Lu %Lu -> %Lu\n", em->start,
 131                                        em->len, em->block_start);
 132                                 free_extent_map(em);
 133                         }
 134                         ret = -EIO;
 135                 }
 136         } else if (ret) {
 137                 free_extent_map(em);
 138                 em = NULL;
 139         }
 140         spin_unlock(&em_tree->lock);
 141
 142         if (ret)
 143                 em = ERR_PTR(ret);
 144 out:
 145         return em;
 146 }
 147
 148 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 149 {
 150         return btrfs_crc32c(seed, data, len);
 151 }
 152
 153 void btrfs_csum_final(u32 crc, char *result)
 154 {
 155         *(__le32 *)result = ~cpu_to_le32(crc);
 156 }
 157
 158 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 159                            int verify)
 160 {
 161         char result[BTRFS_CRC32_SIZE];
 162         unsigned long len;
 163         unsigned long cur_len;
 164         unsigned long offset = BTRFS_CSUM_SIZE;
 165         char *map_token = NULL;
 166         char *kaddr;
 167         unsigned long map_start;
 168         unsigned long map_len;
 169         int err;
 170         u32 crc = ~(u32)0;
 171
 172         len = buf->len - offset;
 173         while(len > 0) {
 174                 err = map_private_extent_buffer(buf, offset, 32,
 175                                         &map_token, &kaddr,
 176                                         &map_start, &map_len, KM_USER0);
 177                 if (err) {
 178                         printk("failed to map extent buffer! %lu\n",
 179                                offset);
 180                         return 1;
 181                 }
 182                 cur_len = min(len, map_len - (offset - map_start));
 183                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 184                                       crc, cur_len);
 185                 len -= cur_len;
 186                 offset += cur_len;
 187                 unmap_extent_buffer(buf, map_token, KM_USER0);
 188         }
 189         btrfs_csum_final(crc, result);
 190
 191         if (verify) {
 192                 /* FIXME, this is not good */
 193                 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
 194                         u32 val;
 195                         u32 found = 0;
 196                         memcpy(&found, result, BTRFS_CRC32_SIZE);
 197
 198                         read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
 199                         printk("btrfs: %s checksum verify failed on %llu "
 200                                "wanted %X found %X level %d\n",
 201                                root->fs_info->sb->s_id,
 202                                buf->start, val, found, btrfs_header_level(buf));
 203                         return 1;
 204                 }
 205         } else {
 206                 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
 207         }
 208         return 0;
 209 }
 210
 211 static int verify_parent_transid(struct extent_io_tree *io_tree,
 212                                  struct extent_buffer *eb, u64 parent_transid)
 213 {
 214         int ret;
 215
 216         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 217                 return 0;
 218
 219         lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
 220         if (extent_buffer_uptodate(io_tree, eb) &&
 221             btrfs_header_generation(eb) == parent_transid) {
 222                 ret = 0;
 223                 goto out;
 224         }
 225         printk("parent transid verify failed on %llu wanted %llu found %llu\n",
 226                (unsigned long long)eb->start,
 227                (unsigned long long)parent_transid,
 228                (unsigned long long)btrfs_header_generation(eb));
 229         ret = 1;
 230         clear_extent_buffer_uptodate(io_tree, eb);
 231 out:
 232         unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
 233                       GFP_NOFS);
 234         return ret;
 235
 236 }
 237
 238 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 239                                           struct extent_buffer *eb,
 240                                           u64 start, u64 parent_transid)
 241 {
 242         struct extent_io_tree *io_tree;
 243         int ret;
 244         int num_copies = 0;
 245         int mirror_num = 0;
 246
 247         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 248         while (1) {
 249                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 250                                                btree_get_extent, mirror_num);
 251                 if (!ret &&
 252                     !verify_parent_transid(io_tree, eb, parent_transid))
 253                         return ret;
 254 printk("read extent buffer pages failed with ret %d mirror no %d\n", ret, mirror_num);
 255                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 256                                               eb->start, eb->len);
 257                 if (num_copies == 1)
 258                         return ret;
 259
 260                 mirror_num++;
 261                 if (mirror_num > num_copies)
 262                         return ret;
 263         }
 264         return -EIO;
 265 }
 266
 267 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 268 {
 269         struct extent_io_tree *tree;
 270         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 271         u64 found_start;
 272         int found_level;
 273         unsigned long len;
 274         struct extent_buffer *eb;
 275         int ret;
 276
 277         tree = &BTRFS_I(page->mapping->host)->io_tree;
 278
 279         if (page->private == EXTENT_PAGE_PRIVATE)
 280                 goto out;
 281         if (!page->private)
 282                 goto out;
 283         len = page->private >> 2;
 284         if (len == 0) {
 285                 WARN_ON(1);
 286         }
 287         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 288         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 289                                              btrfs_header_generation(eb));
 290         BUG_ON(ret);
 291         found_start = btrfs_header_bytenr(eb);
 292         if (found_start != start) {
 293                 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
 294                        start, found_start, len);
 295                 WARN_ON(1);
 296                 goto err;
 297         }
 298         if (eb->first_page != page) {
 299                 printk("bad first page %lu %lu\n", eb->first_page->index,
 300                        page->index);
 301                 WARN_ON(1);
 302                 goto err;
 303         }
 304         if (!PageUptodate(page)) {
 305                 printk("csum not up to date page %lu\n", page->index);
 306                 WARN_ON(1);
 307                 goto err;
 308         }
 309         found_level = btrfs_header_level(eb);
 310
 311         csum_tree_block(root, eb, 0);
 312 err:
 313         free_extent_buffer(eb);
 314 out:
 315         return 0;
 316 }
 317
 318 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 319                                struct extent_state *state)
 320 {
 321         struct extent_io_tree *tree;
 322         u64 found_start;
 323         int found_level;
 324         unsigned long len;
 325         struct extent_buffer *eb;
 326         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 327         int ret = 0;
 328
 329         tree = &BTRFS_I(page->mapping->host)->io_tree;
 330         if (page->private == EXTENT_PAGE_PRIVATE)
 331                 goto out;
 332         if (!page->private)
 333                 goto out;
 334         len = page->private >> 2;
 335         if (len == 0) {
 336                 WARN_ON(1);
 337         }
 338         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 339
 340         found_start = btrfs_header_bytenr(eb);
 341         if (found_start != start) {
 342                 printk("bad tree block start %llu %llu\n",
 343                        (unsigned long long)found_start,
 344                        (unsigned long long)eb->start);
 345                 ret = -EIO;
 346                 goto err;
 347         }
 348         if (eb->first_page != page) {
 349                 printk("bad first page %lu %lu\n", eb->first_page->index,
 350                        page->index);
 351                 WARN_ON(1);
 352                 ret = -EIO;
 353                 goto err;
 354         }
 355         if (memcmp_extent_buffer(eb, root->fs_info->fsid,
 356                                  (unsigned long)btrfs_header_fsid(eb),
 357                                  BTRFS_FSID_SIZE)) {
 358                 printk("bad fsid on block %Lu\n", eb->start);
 359                 ret = -EIO;
 360                 goto err;
 361         }
 362         found_level = btrfs_header_level(eb);
 363
 364         ret = csum_tree_block(root, eb, 1);
 365         if (ret)
 366                 ret = -EIO;
 367
 368         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 369         end = eb->start + end - 1;
 370 err:
 371         free_extent_buffer(eb);
 372 out:
 373         return ret;
 374 }
 375
 376 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
 377 static void end_workqueue_bio(struct bio *bio, int err)
 378 #else
 379 static int end_workqueue_bio(struct bio *bio,
 380                                    unsigned int bytes_done, int err)
 381 #endif
 382 {
 383         struct end_io_wq *end_io_wq = bio->bi_private;
 384         struct btrfs_fs_info *fs_info;
 385
 386 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 387         if (bio->bi_size)
 388                 return 1;
 389 #endif
 390
 391         fs_info = end_io_wq->info;
 392         end_io_wq->error = err;
 393         end_io_wq->work.func = end_workqueue_fn;
 394         end_io_wq->work.flags = 0;
 395         if (bio->bi_rw & (1 << BIO_RW))
 396                 btrfs_queue_worker(&fs_info->endio_write_workers,
 397                                    &end_io_wq->work);
 398         else
 399                 btrfs_queue_worker(&fs_info->endio_workers, &end_io_wq->work);
 400
 401 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 402         return 0;
 403 #endif
 404 }
 405
 406 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 407                         int metadata)
 408 {
 409         struct end_io_wq *end_io_wq;
 410         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 411         if (!end_io_wq)
 412                 return -ENOMEM;
 413
 414         end_io_wq->private = bio->bi_private;
 415         end_io_wq->end_io = bio->bi_end_io;
 416         end_io_wq->info = info;
 417         end_io_wq->error = 0;
 418         end_io_wq->bio = bio;
 419         end_io_wq->metadata = metadata;
 420
 421         bio->bi_private = end_io_wq;
 422         bio->bi_end_io = end_workqueue_bio;
 423         return 0;
 424 }
 425
 426 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 427 {
 428         unsigned long limit = min_t(unsigned long,
 429                                     info->workers.max_workers,
 430                                     info->fs_devices->open_devices);
 431         return 256 * limit;
 432 }
 433
 434 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
 435 {
 436         return atomic_read(&info->nr_async_bios) >
 437                 btrfs_async_submit_limit(info);
 438 }
 439
 440 static void run_one_async_submit(struct btrfs_work *work)
 441 {
 442         struct btrfs_fs_info *fs_info;
 443         struct async_submit_bio *async;
 444         int limit;
 445
 446         async = container_of(work, struct  async_submit_bio, work);
 447         fs_info = BTRFS_I(async->inode)->root->fs_info;
 448
 449         limit = btrfs_async_submit_limit(fs_info);
 450         limit = limit * 2 / 3;
 451
 452         atomic_dec(&fs_info->nr_async_submits);
 453
 454         if (atomic_read(&fs_info->nr_async_submits) < limit &&
 455             waitqueue_active(&fs_info->async_submit_wait))
 456                 wake_up(&fs_info->async_submit_wait);
 457
 458         async->submit_bio_hook(async->inode, async->rw, async->bio,
 459                                async->mirror_num);
 460         kfree(async);
 461 }
 462
 463 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 464                         int rw, struct bio *bio, int mirror_num,
 465                         extent_submit_bio_hook_t *submit_bio_hook)
 466 {
 467         struct async_submit_bio *async;
 468         int limit = btrfs_async_submit_limit(fs_info);
 469
 470         async = kmalloc(sizeof(*async), GFP_NOFS);
 471         if (!async)
 472                 return -ENOMEM;
 473
 474         async->inode = inode;
 475         async->rw = rw;
 476         async->bio = bio;
 477         async->mirror_num = mirror_num;
 478         async->submit_bio_hook = submit_bio_hook;
 479         async->work.func = run_one_async_submit;
 480         async->work.flags = 0;
 481         atomic_inc(&fs_info->nr_async_submits);
 482         btrfs_queue_worker(&fs_info->workers, &async->work);
 483
 484         if (atomic_read(&fs_info->nr_async_submits) > limit) {
 485                 wait_event_timeout(fs_info->async_submit_wait,
 486                            (atomic_read(&fs_info->nr_async_submits) < limit),
 487                            HZ/10);
 488
 489                 wait_event_timeout(fs_info->async_submit_wait,
 490                            (atomic_read(&fs_info->nr_async_bios) < limit),
 491                            HZ/10);
 492         }
 493         return 0;
 494 }
 495
 496 static int btree_csum_one_bio(struct bio *bio)
 497 {
 498         struct bio_vec *bvec = bio->bi_io_vec;
 499         int bio_index = 0;
 500         struct btrfs_root *root;
 501
 502         WARN_ON(bio->bi_vcnt <= 0);
 503         while(bio_index < bio->bi_vcnt) {
 504                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 505                 csum_dirty_buffer(root, bvec->bv_page);
 506                 bio_index++;
 507                 bvec++;
 508         }
 509         return 0;
 510 }
 511
 512 static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 513                                  int mirror_num)
 514 {
 515         struct btrfs_root *root = BTRFS_I(inode)->root;
 516         u64 offset;
 517         int ret;
 518
 519         offset = bio->bi_sector << 9;
 520
 521         /*
 522          * when we're called for a write, we're already in the async
 523          * submission context.  Just jump into btrfs_map_bio
 524          */
 525         if (rw & (1 << BIO_RW)) {
 526                 btree_csum_one_bio(bio);
 527                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 528                                      mirror_num, 1);
 529         }
 530
 531         /*
 532          * called for a read, do the setup so that checksum validation
 533          * can happen in the async kernel threads
 534          */
 535         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
 536         BUG_ON(ret);
 537
 538         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 539 }
 540
 541 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 542                                  int mirror_num)
 543 {
 544         /*
 545          * kthread helpers are used to submit writes so that checksumming
 546          * can happen in parallel across all CPUs
 547          */
 548         if (!(rw & (1 << BIO_RW))) {
 549                 return __btree_submit_bio_hook(inode, rw, bio, mirror_num);
 550         }
 551         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 552                                    inode, rw, bio, mirror_num,
 553                                    __btree_submit_bio_hook);
 554 }
 555
 556 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 557 {
 558         struct extent_io_tree *tree;
 559         tree = &BTRFS_I(page->mapping->host)->io_tree;
 560
 561         if (current->flags & PF_MEMALLOC) {
 562                 redirty_page_for_writepage(wbc, page);
 563                 unlock_page(page);
 564                 return 0;
 565         }
 566         return extent_write_full_page(tree, page, btree_get_extent, wbc);
 567 }
 568
 569 static int btree_writepages(struct address_space *mapping,
 570                             struct writeback_control *wbc)
 571 {
 572         struct extent_io_tree *tree;
 573         tree = &BTRFS_I(mapping->host)->io_tree;
 574         if (wbc->sync_mode == WB_SYNC_NONE) {
 575                 u64 num_dirty;
 576                 u64 start = 0;
 577                 unsigned long thresh = 8 * 1024 * 1024;
 578
 579                 if (wbc->for_kupdate)
 580                         return 0;
 581
 582                 num_dirty = count_range_bits(tree, &start, (u64)-1,
 583                                              thresh, EXTENT_DIRTY);
 584                 if (num_dirty < thresh) {
 585                         return 0;
 586                 }
 587         }
 588         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 589 }
 590
 591 int btree_readpage(struct file *file, struct page *page)
 592 {
 593         struct extent_io_tree *tree;
 594         tree = &BTRFS_I(page->mapping->host)->io_tree;
 595         return extent_read_full_page(tree, page, btree_get_extent);
 596 }
 597
 598 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 599 {
 600         struct extent_io_tree *tree;
 601         struct extent_map_tree *map;
 602         int ret;
 603
 604         if (PageWriteback(page) || PageDirty(page))
 605             return 0;
 606
 607         tree = &BTRFS_I(page->mapping->host)->io_tree;
 608         map = &BTRFS_I(page->mapping->host)->extent_tree;
 609
 610         ret = try_release_extent_state(map, tree, page, gfp_flags);
 611         if (!ret) {
 612                 return 0;
 613         }
 614
 615         ret = try_release_extent_buffer(tree, page);
 616         if (ret == 1) {
 617                 ClearPagePrivate(page);
 618                 set_page_private(page, 0);
 619                 page_cache_release(page);
 620         }
 621
 622         return ret;
 623 }
 624
 625 static void btree_invalidatepage(struct page *page, unsigned long offset)
 626 {
 627         struct extent_io_tree *tree;
 628         tree = &BTRFS_I(page->mapping->host)->io_tree;
 629         extent_invalidatepage(tree, page, offset);
 630         btree_releasepage(page, GFP_NOFS);
 631         if (PagePrivate(page)) {
 632                 printk("warning page private not zero on page %Lu\n",
 633                        page_offset(page));
 634                 ClearPagePrivate(page);
 635                 set_page_private(page, 0);
 636                 page_cache_release(page);
 637         }
 638 }
 639
 640 #if 0
 641 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 642 {
 643         struct buffer_head *bh;
 644         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 645         struct buffer_head *head;
 646         if (!page_has_buffers(page)) {
 647                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 648                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 649         }
 650         head = page_buffers(page);
 651         bh = head;
 652         do {
 653                 if (buffer_dirty(bh))
 654                         csum_tree_block(root, bh, 0);
 655                 bh = bh->b_this_page;
 656         } while (bh != head);
 657         return block_write_full_page(page, btree_get_block, wbc);
 658 }
 659 #endif
 660
 661 static struct address_space_operations btree_aops = {
 662         .readpage       = btree_readpage,
 663         .writepage      = btree_writepage,
 664         .writepages     = btree_writepages,
 665         .releasepage    = btree_releasepage,
 666         .invalidatepage = btree_invalidatepage,
 667         .sync_page      = block_sync_page,
 668 };
 669
 670 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 671                          u64 parent_transid)
 672 {
 673         struct extent_buffer *buf = NULL;
 674         struct inode *btree_inode = root->fs_info->btree_inode;
 675         int ret = 0;
 676
 677         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 678         if (!buf)
 679                 return 0;
 680         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 681                                  buf, 0, 0, btree_get_extent, 0);
 682         free_extent_buffer(buf);
 683         return ret;
 684 }
 685
 686 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 687                                             u64 bytenr, u32 blocksize)
 688 {
 689         struct inode *btree_inode = root->fs_info->btree_inode;
 690         struct extent_buffer *eb;
 691         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 692                                 bytenr, blocksize, GFP_NOFS);
 693         return eb;
 694 }
 695
 696 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 697                                                  u64 bytenr, u32 blocksize)
 698 {
 699         struct inode *btree_inode = root->fs_info->btree_inode;
 700         struct extent_buffer *eb;
 701
 702         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 703                                  bytenr, blocksize, NULL, GFP_NOFS);
 704         return eb;
 705 }
 706
 707
 708 int btrfs_write_tree_block(struct extent_buffer *buf)
 709 {
 710         return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
 711                                       buf->start + buf->len - 1, WB_SYNC_NONE);
 712 }
 713
 714 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 715 {
 716         return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
 717                                   buf->start, buf->start + buf->len -1);
 718 }
 719
 720 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 721                                       u32 blocksize, u64 parent_transid)
 722 {
 723         struct extent_buffer *buf = NULL;
 724         struct inode *btree_inode = root->fs_info->btree_inode;
 725         struct extent_io_tree *io_tree;
 726         int ret;
 727
 728         io_tree = &BTRFS_I(btree_inode)->io_tree;
 729
 730         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 731         if (!buf)
 732                 return NULL;
 733
 734         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 735
 736         if (ret == 0) {
 737                 buf->flags |= EXTENT_UPTODATE;
 738         } else {
 739                 WARN_ON(1);
 740         }
 741         return buf;
 742
 743 }
 744
 745 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 746                      struct extent_buffer *buf)
 747 {
 748         struct inode *btree_inode = root->fs_info->btree_inode;
 749         if (btrfs_header_generation(buf) ==
 750             root->fs_info->running_transaction->transid) {
 751                 WARN_ON(!btrfs_tree_locked(buf));
 752                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 753                                           buf);
 754         }
 755         return 0;
 756 }
 757
 758 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 759                         u32 stripesize, struct btrfs_root *root,
 760                         struct btrfs_fs_info *fs_info,
 761                         u64 objectid)
 762 {
 763         root->node = NULL;
 764         root->inode = NULL;
 765         root->commit_root = NULL;
 766         root->ref_tree = NULL;
 767         root->sectorsize = sectorsize;
 768         root->nodesize = nodesize;
 769         root->leafsize = leafsize;
 770         root->stripesize = stripesize;
 771         root->ref_cows = 0;
 772         root->track_dirty = 0;
 773
 774         root->fs_info = fs_info;
 775         root->objectid = objectid;
 776         root->last_trans = 0;
 777         root->highest_inode = 0;
 778         root->last_inode_alloc = 0;
 779         root->name = NULL;
 780         root->in_sysfs = 0;
 781
 782         INIT_LIST_HEAD(&root->dirty_list);
 783         INIT_LIST_HEAD(&root->orphan_list);
 784         INIT_LIST_HEAD(&root->dead_list);
 785         spin_lock_init(&root->node_lock);
 786         spin_lock_init(&root->list_lock);
 787         mutex_init(&root->objectid_mutex);
 788         mutex_init(&root->log_mutex);
 789         extent_io_tree_init(&root->dirty_log_pages,
 790                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 791
 792         btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
 793         root->ref_tree = &root->ref_tree_struct;
 794
 795         memset(&root->root_key, 0, sizeof(root->root_key));
 796         memset(&root->root_item, 0, sizeof(root->root_item));
 797         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 798         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 799         root->defrag_trans_start = fs_info->generation;
 800         init_completion(&root->kobj_unregister);
 801         root->defrag_running = 0;
 802         root->defrag_level = 0;
 803         root->root_key.objectid = objectid;
 804         return 0;
 805 }
 806
 807 static int find_and_setup_root(struct btrfs_root *tree_root,
 808                                struct btrfs_fs_info *fs_info,
 809                                u64 objectid,
 810                                struct btrfs_root *root)
 811 {
 812         int ret;
 813         u32 blocksize;
 814
 815         __setup_root(tree_root->nodesize, tree_root->leafsize,
 816                      tree_root->sectorsize, tree_root->stripesize,
 817                      root, fs_info, objectid);
 818         ret = btrfs_find_last_root(tree_root, objectid,
 819                                    &root->root_item, &root->root_key);
 820         BUG_ON(ret);
 821
 822         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 823         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 824                                      blocksize, 0);
 825         BUG_ON(!root->node);
 826         return 0;
 827 }
 828
 829 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
 830                              struct btrfs_fs_info *fs_info)
 831 {
 832         struct extent_buffer *eb;
 833         struct btrfs_root *log_root_tree = fs_info->log_root_tree;
 834         u64 start = 0;
 835         u64 end = 0;
 836         int ret;
 837
 838         if (!log_root_tree)
 839                 return 0;
 840
 841         while(1) {
 842                 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
 843                                     0, &start, &end, EXTENT_DIRTY);
 844                 if (ret)
 845                         break;
 846
 847                 clear_extent_dirty(&log_root_tree->dirty_log_pages,
 848                                    start, end, GFP_NOFS);
 849         }
 850         eb = fs_info->log_root_tree->node;
 851
 852         WARN_ON(btrfs_header_level(eb) != 0);
 853         WARN_ON(btrfs_header_nritems(eb) != 0);
 854
 855         ret = btrfs_free_reserved_extent(fs_info->tree_root,
 856                                 eb->start, eb->len);
 857         BUG_ON(ret);
 858
 859         free_extent_buffer(eb);
 860         kfree(fs_info->log_root_tree);
 861         fs_info->log_root_tree = NULL;
 862         return 0;
 863 }
 864
 865 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
 866                              struct btrfs_fs_info *fs_info)
 867 {
 868         struct btrfs_root *root;
 869         struct btrfs_root *tree_root = fs_info->tree_root;
 870
 871         root = kzalloc(sizeof(*root), GFP_NOFS);
 872         if (!root)
 873                 return -ENOMEM;
 874
 875         __setup_root(tree_root->nodesize, tree_root->leafsize,
 876                      tree_root->sectorsize, tree_root->stripesize,
 877                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
 878
 879         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
 880         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
 881         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
 882         root->ref_cows = 0;
 883
 884         root->node = btrfs_alloc_free_block(trans, root, root->leafsize,
 885                                             0, BTRFS_TREE_LOG_OBJECTID,
 886                                             trans->transid, 0, 0, 0);
 887
 888         btrfs_set_header_nritems(root->node, 0);
 889         btrfs_set_header_level(root->node, 0);
 890         btrfs_set_header_bytenr(root->node, root->node->start);
 891         btrfs_set_header_generation(root->node, trans->transid);
 892         btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
 893
 894         write_extent_buffer(root->node, root->fs_info->fsid,
 895                             (unsigned long)btrfs_header_fsid(root->node),
 896                             BTRFS_FSID_SIZE);
 897         btrfs_mark_buffer_dirty(root->node);
 898         btrfs_tree_unlock(root->node);
 899         fs_info->log_root_tree = root;
 900         return 0;
 901 }
 902
 903 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
 904                                                struct btrfs_key *location)
 905 {
 906         struct btrfs_root *root;
 907         struct btrfs_fs_info *fs_info = tree_root->fs_info;
 908         struct btrfs_path *path;
 909         struct extent_buffer *l;
 910         u64 highest_inode;
 911         u32 blocksize;
 912         int ret = 0;
 913
 914         root = kzalloc(sizeof(*root), GFP_NOFS);
 915         if (!root)
 916                 return ERR_PTR(-ENOMEM);
 917         if (location->offset == (u64)-1) {
 918                 ret = find_and_setup_root(tree_root, fs_info,
 919                                           location->objectid, root);
 920                 if (ret) {
 921                         kfree(root);
 922                         return ERR_PTR(ret);
 923                 }
 924                 goto insert;
 925         }
 926
 927         __setup_root(tree_root->nodesize, tree_root->leafsize,
 928                      tree_root->sectorsize, tree_root->stripesize,
 929                      root, fs_info, location->objectid);
 930
 931         path = btrfs_alloc_path();
 932         BUG_ON(!path);
 933         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 934         if (ret != 0) {
 935                 if (ret > 0)
 936                         ret = -ENOENT;
 937                 goto out;
 938         }
 939         l = path->nodes[0];
 940         read_extent_buffer(l, &root->root_item,
 941                btrfs_item_ptr_offset(l, path->slots[0]),
 942                sizeof(root->root_item));
 943         memcpy(&root->root_key, location, sizeof(*location));
 944         ret = 0;
 945 out:
 946         btrfs_release_path(root, path);
 947         btrfs_free_path(path);
 948         if (ret) {
 949                 kfree(root);
 950                 return ERR_PTR(ret);
 951         }
 952         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 953         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 954                                      blocksize, 0);
 955         BUG_ON(!root->node);
 956 insert:
 957         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
 958                 root->ref_cows = 1;
 959                 ret = btrfs_find_highest_inode(root, &highest_inode);
 960                 if (ret == 0) {
 961                         root->highest_inode = highest_inode;
 962                         root->last_inode_alloc = highest_inode;
 963                 }
 964         }
 965         return root;
 966 }
 967
 968 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
 969                                         u64 root_objectid)
 970 {
 971         struct btrfs_root *root;
 972
 973         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
 974                 return fs_info->tree_root;
 975         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
 976                 return fs_info->extent_root;
 977
 978         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 979                                  (unsigned long)root_objectid);
 980         return root;
 981 }
 982
 983 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
 984                                               struct btrfs_key *location)
 985 {
 986         struct btrfs_root *root;
 987         int ret;
 988
 989         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 990                 return fs_info->tree_root;
 991         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
 992                 return fs_info->extent_root;
 993         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
 994                 return fs_info->chunk_root;
 995         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
 996                 return fs_info->dev_root;
 997
 998         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 999                                  (unsigned long)location->objectid);
1000         if (root)
1001                 return root;
1002
1003         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1004         if (IS_ERR(root))
1005                 return root;
1006         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1007                                 (unsigned long)root->root_key.objectid,
1008                                 root);
1009         if (ret) {
1010                 free_extent_buffer(root->node);
1011                 kfree(root);
1012                 return ERR_PTR(ret);
1013         }
1014         ret = btrfs_find_dead_roots(fs_info->tree_root,
1015                                     root->root_key.objectid, root);
1016         BUG_ON(ret);
1017
1018         return root;
1019 }
1020
1021 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1022                                       struct btrfs_key *location,
1023                                       const char *name, int namelen)
1024 {
1025         struct btrfs_root *root;
1026         int ret;
1027
1028         root = btrfs_read_fs_root_no_name(fs_info, location);
1029         if (!root)
1030                 return NULL;
1031
1032         if (root->in_sysfs)
1033                 return root;
1034
1035         ret = btrfs_set_root_name(root, name, namelen);
1036         if (ret) {
1037                 free_extent_buffer(root->node);
1038                 kfree(root);
1039                 return ERR_PTR(ret);
1040         }
1041
1042         ret = btrfs_sysfs_add_root(root);
1043         if (ret) {
1044                 free_extent_buffer(root->node);
1045                 kfree(root->name);
1046                 kfree(root);
1047                 return ERR_PTR(ret);
1048         }
1049         root->in_sysfs = 1;
1050         return root;
1051 }
1052 #if 0
1053 static int add_hasher(struct btrfs_fs_info *info, char *type) {
1054         struct btrfs_hasher *hasher;
1055
1056         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
1057         if (!hasher)
1058                 return -ENOMEM;
1059         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
1060         if (!hasher->hash_tfm) {
1061                 kfree(hasher);
1062                 return -EINVAL;
1063         }
1064         spin_lock(&info->hash_lock);
1065         list_add(&hasher->list, &info->hashers);
1066         spin_unlock(&info->hash_lock);
1067         return 0;
1068 }
1069 #endif
1070
1071 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1072 {
1073         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1074         int ret = 0;
1075         struct list_head *cur;
1076         struct btrfs_device *device;
1077         struct backing_dev_info *bdi;
1078
1079         if ((bdi_bits & (1 << BDI_write_congested)) &&
1080             btrfs_congested_async(info, 0))
1081                 return 1;
1082
1083         list_for_each(cur, &info->fs_devices->devices) {
1084                 device = list_entry(cur, struct btrfs_device, dev_list);
1085                 if (!device->bdev)
1086                         continue;
1087                 bdi = blk_get_backing_dev_info(device->bdev);
1088                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1089                         ret = 1;
1090                         break;
1091                 }
1092         }
1093         return ret;
1094 }
1095
1096 /*
1097  * this unplugs every device on the box, and it is only used when page
1098  * is null
1099  */
1100 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1101 {
1102         struct list_head *cur;
1103         struct btrfs_device *device;
1104         struct btrfs_fs_info *info;
1105
1106         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1107         list_for_each(cur, &info->fs_devices->devices) {
1108                 device = list_entry(cur, struct btrfs_device, dev_list);
1109                 bdi = blk_get_backing_dev_info(device->bdev);
1110                 if (bdi->unplug_io_fn) {
1111                         bdi->unplug_io_fn(bdi, page);
1112                 }
1113         }
1114 }
1115
1116 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1117 {
1118         struct inode *inode;
1119         struct extent_map_tree *em_tree;
1120         struct extent_map *em;
1121         struct address_space *mapping;
1122         u64 offset;
1123
1124         /* the generic O_DIRECT read code does this */
1125         if (!page) {
1126                 __unplug_io_fn(bdi, page);
1127                 return;
1128         }
1129
1130         /*
1131          * page->mapping may change at any time.  Get a consistent copy
1132          * and use that for everything below
1133          */
1134         smp_mb();
1135         mapping = page->mapping;
1136         if (!mapping)
1137                 return;
1138
1139         inode = mapping->host;
1140         offset = page_offset(page);
1141
1142         em_tree = &BTRFS_I(inode)->extent_tree;
1143         spin_lock(&em_tree->lock);
1144         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1145         spin_unlock(&em_tree->lock);
1146         if (!em) {
1147                 __unplug_io_fn(bdi, page);
1148                 return;
1149         }
1150
1151         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1152                 free_extent_map(em);
1153                 __unplug_io_fn(bdi, page);
1154                 return;
1155         }
1156         offset = offset - em->start;
1157         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1158                           em->block_start + offset, page);
1159         free_extent_map(em);
1160 }
1161
1162 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1163 {
1164 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1165         bdi_init(bdi);
1166 #endif
1167         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1168         bdi->state              = 0;
1169         bdi->capabilities       = default_backing_dev_info.capabilities;
1170         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1171         bdi->unplug_io_data     = info;
1172         bdi->congested_fn       = btrfs_congested_fn;
1173         bdi->congested_data     = info;
1174         return 0;
1175 }
1176
1177 static int bio_ready_for_csum(struct bio *bio)
1178 {
1179         u64 length = 0;
1180         u64 buf_len = 0;
1181         u64 start = 0;
1182         struct page *page;
1183         struct extent_io_tree *io_tree = NULL;
1184         struct btrfs_fs_info *info = NULL;
1185         struct bio_vec *bvec;
1186         int i;
1187         int ret;
1188
1189         bio_for_each_segment(bvec, bio, i) {
1190                 page = bvec->bv_page;
1191                 if (page->private == EXTENT_PAGE_PRIVATE) {
1192                         length += bvec->bv_len;
1193                         continue;
1194                 }
1195                 if (!page->private) {
1196                         length += bvec->bv_len;
1197                         continue;
1198                 }
1199                 length = bvec->bv_len;
1200                 buf_len = page->private >> 2;
1201                 start = page_offset(page) + bvec->bv_offset;
1202                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1203                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1204         }
1205         /* are we fully contained in this bio? */
1206         if (buf_len <= length)
1207                 return 1;
1208
1209         ret = extent_range_uptodate(io_tree, start + length,
1210                                     start + buf_len - 1);
1211         if (ret == 1)
1212                 return ret;
1213         return ret;
1214 }
1215
1216 /*
1217  * called by the kthread helper functions to finally call the bio end_io
1218  * functions.  This is where read checksum verification actually happens
1219  */
1220 static void end_workqueue_fn(struct btrfs_work *work)
1221 {
1222         struct bio *bio;
1223         struct end_io_wq *end_io_wq;
1224         struct btrfs_fs_info *fs_info;
1225         int error;
1226
1227         end_io_wq = container_of(work, struct end_io_wq, work);
1228         bio = end_io_wq->bio;
1229         fs_info = end_io_wq->info;
1230
1231         /* metadata bios are special because the whole tree block must
1232          * be checksummed at once.  This makes sure the entire block is in
1233          * ram and up to date before trying to verify things.  For
1234          * blocksize <= pagesize, it is basically a noop
1235          */
1236         if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1237                 btrfs_queue_worker(&fs_info->endio_workers,
1238                                    &end_io_wq->work);
1239                 return;
1240         }
1241         error = end_io_wq->error;
1242         bio->bi_private = end_io_wq->private;
1243         bio->bi_end_io = end_io_wq->end_io;
1244         kfree(end_io_wq);
1245 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1246         bio_endio(bio, bio->bi_size, error);
1247 #else
1248         bio_endio(bio, error);
1249 #endif
1250 }
1251
1252 static int cleaner_kthread(void *arg)
1253 {
1254         struct btrfs_root *root = arg;
1255
1256         do {
1257                 smp_mb();
1258                 if (root->fs_info->closing)
1259                         break;
1260
1261                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1262                 mutex_lock(&root->fs_info->cleaner_mutex);
1263                 btrfs_clean_old_snapshots(root);
1264                 mutex_unlock(&root->fs_info->cleaner_mutex);
1265
1266                 if (freezing(current)) {
1267                         refrigerator();
1268                 } else {
1269                         smp_mb();
1270                         if (root->fs_info->closing)
1271                                 break;
1272                         set_current_state(TASK_INTERRUPTIBLE);
1273                         schedule();
1274                         __set_current_state(TASK_RUNNING);
1275                 }
1276         } while (!kthread_should_stop());
1277         return 0;
1278 }
1279
1280 static int transaction_kthread(void *arg)
1281 {
1282         struct btrfs_root *root = arg;
1283         struct btrfs_trans_handle *trans;
1284         struct btrfs_transaction *cur;
1285         unsigned long now;
1286         unsigned long delay;
1287         int ret;
1288
1289         do {
1290                 smp_mb();
1291                 if (root->fs_info->closing)
1292                         break;
1293
1294                 delay = HZ * 30;
1295                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1296                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1297
1298                 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1299                         printk("btrfs: total reference cache size %Lu\n",
1300                                 root->fs_info->total_ref_cache_size);
1301                 }
1302
1303                 mutex_lock(&root->fs_info->trans_mutex);
1304                 cur = root->fs_info->running_transaction;
1305                 if (!cur) {
1306                         mutex_unlock(&root->fs_info->trans_mutex);
1307                         goto sleep;
1308                 }
1309
1310                 now = get_seconds();
1311                 if (now < cur->start_time || now - cur->start_time < 30) {
1312                         mutex_unlock(&root->fs_info->trans_mutex);
1313                         delay = HZ * 5;
1314                         goto sleep;
1315                 }
1316                 mutex_unlock(&root->fs_info->trans_mutex);
1317                 trans = btrfs_start_transaction(root, 1);
1318                 ret = btrfs_commit_transaction(trans, root);
1319 sleep:
1320                 wake_up_process(root->fs_info->cleaner_kthread);
1321                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1322
1323                 if (freezing(current)) {
1324                         refrigerator();
1325                 } else {
1326                         if (root->fs_info->closing)
1327                                 break;
1328                         set_current_state(TASK_INTERRUPTIBLE);
1329                         schedule_timeout(delay);
1330                         __set_current_state(TASK_RUNNING);
1331                 }
1332         } while (!kthread_should_stop());
1333         return 0;
1334 }
1335
1336 struct btrfs_root *open_ctree(struct super_block *sb,
1337                               struct btrfs_fs_devices *fs_devices,
1338                               char *options)
1339 {
1340         u32 sectorsize;
1341         u32 nodesize;
1342         u32 leafsize;
1343         u32 blocksize;
1344         u32 stripesize;
1345         struct buffer_head *bh;
1346         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1347                                                  GFP_NOFS);
1348         struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1349                                                GFP_NOFS);
1350         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1351                                                 GFP_NOFS);
1352         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1353                                                 GFP_NOFS);
1354         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1355                                               GFP_NOFS);
1356         struct btrfs_root *log_tree_root;
1357
1358         int ret;
1359         int err = -EINVAL;
1360
1361         struct btrfs_super_block *disk_super;
1362
1363         if (!extent_root || !tree_root || !fs_info) {
1364                 err = -ENOMEM;
1365                 goto fail;
1366         }
1367         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1368         INIT_LIST_HEAD(&fs_info->trans_list);
1369         INIT_LIST_HEAD(&fs_info->dead_roots);
1370         INIT_LIST_HEAD(&fs_info->hashers);
1371         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1372         spin_lock_init(&fs_info->hash_lock);
1373         spin_lock_init(&fs_info->delalloc_lock);
1374         spin_lock_init(&fs_info->new_trans_lock);
1375         spin_lock_init(&fs_info->ref_cache_lock);
1376
1377         init_completion(&fs_info->kobj_unregister);
1378         fs_info->tree_root = tree_root;
1379         fs_info->extent_root = extent_root;
1380         fs_info->chunk_root = chunk_root;
1381         fs_info->dev_root = dev_root;
1382         fs_info->fs_devices = fs_devices;
1383         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1384         INIT_LIST_HEAD(&fs_info->space_info);
1385         btrfs_mapping_init(&fs_info->mapping_tree);
1386         atomic_set(&fs_info->nr_async_submits, 0);
1387         atomic_set(&fs_info->nr_async_bios, 0);
1388         atomic_set(&fs_info->throttles, 0);
1389         atomic_set(&fs_info->throttle_gen, 0);
1390         fs_info->sb = sb;
1391         fs_info->max_extent = (u64)-1;
1392         fs_info->max_inline = 8192 * 1024;
1393         setup_bdi(fs_info, &fs_info->bdi);
1394         fs_info->btree_inode = new_inode(sb);
1395         fs_info->btree_inode->i_ino = 1;
1396         fs_info->btree_inode->i_nlink = 1;
1397         fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1398
1399         INIT_LIST_HEAD(&fs_info->ordered_extents);
1400         spin_lock_init(&fs_info->ordered_extent_lock);
1401
1402         sb->s_blocksize = 4096;
1403         sb->s_blocksize_bits = blksize_bits(4096);
1404
1405         /*
1406          * we set the i_size on the btree inode to the max possible int.
1407          * the real end of the address space is determined by all of
1408          * the devices in the system
1409          */
1410         fs_info->btree_inode->i_size = OFFSET_MAX;
1411         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1412         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1413
1414         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1415                              fs_info->btree_inode->i_mapping,
1416                              GFP_NOFS);
1417         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1418                              GFP_NOFS);
1419
1420         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1421
1422         spin_lock_init(&fs_info->block_group_cache_lock);
1423         fs_info->block_group_cache_tree.rb_node = NULL;
1424
1425         extent_io_tree_init(&fs_info->pinned_extents,
1426                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1427         extent_io_tree_init(&fs_info->pending_del,
1428                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1429         extent_io_tree_init(&fs_info->extent_ins,
1430                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1431         fs_info->do_barriers = 1;
1432
1433         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1434         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1435                sizeof(struct btrfs_key));
1436         insert_inode_hash(fs_info->btree_inode);
1437
1438         mutex_init(&fs_info->trans_mutex);
1439         mutex_init(&fs_info->tree_log_mutex);
1440         mutex_init(&fs_info->drop_mutex);
1441         mutex_init(&fs_info->alloc_mutex);
1442         mutex_init(&fs_info->chunk_mutex);
1443         mutex_init(&fs_info->transaction_kthread_mutex);
1444         mutex_init(&fs_info->cleaner_mutex);
1445         mutex_init(&fs_info->volume_mutex);
1446         init_waitqueue_head(&fs_info->transaction_throttle);
1447         init_waitqueue_head(&fs_info->transaction_wait);
1448         init_waitqueue_head(&fs_info->async_submit_wait);
1449         init_waitqueue_head(&fs_info->tree_log_wait);
1450         atomic_set(&fs_info->tree_log_commit, 0);
1451         atomic_set(&fs_info->tree_log_writers, 0);
1452         fs_info->tree_log_transid = 0;
1453
1454 #if 0
1455         ret = add_hasher(fs_info, "crc32c");
1456         if (ret) {
1457                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1458                 err = -ENOMEM;
1459                 goto fail_iput;
1460         }
1461 #endif
1462         __setup_root(4096, 4096, 4096, 4096, tree_root,
1463                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1464
1465
1466         bh = __bread(fs_devices->latest_bdev,
1467                      BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1468         if (!bh)
1469                 goto fail_iput;
1470
1471         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1472         brelse(bh);
1473
1474         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1475
1476         disk_super = &fs_info->super_copy;
1477         if (!btrfs_super_root(disk_super))
1478                 goto fail_sb_buffer;
1479
1480         err = btrfs_parse_options(tree_root, options);
1481         if (err)
1482                 goto fail_sb_buffer;
1483
1484         /*
1485          * we need to start all the end_io workers up front because the
1486          * queue work function gets called at interrupt time, and so it
1487          * cannot dynamically grow.
1488          */
1489         btrfs_init_workers(&fs_info->workers, "worker",
1490                            fs_info->thread_pool_size);
1491         btrfs_init_workers(&fs_info->submit_workers, "submit",
1492                            min_t(u64, fs_devices->num_devices,
1493                            fs_info->thread_pool_size));
1494
1495         /* a higher idle thresh on the submit workers makes it much more
1496          * likely that bios will be send down in a sane order to the
1497          * devices
1498          */
1499         fs_info->submit_workers.idle_thresh = 64;
1500
1501         /* fs_info->workers is responsible for checksumming file data
1502          * blocks and metadata.  Using a larger idle thresh allows each
1503          * worker thread to operate on things in roughly the order they
1504          * were sent by the writeback daemons, improving overall locality
1505          * of the IO going down the pipe.
1506          */
1507         fs_info->workers.idle_thresh = 128;
1508
1509         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1510         btrfs_init_workers(&fs_info->endio_workers, "endio",
1511                            fs_info->thread_pool_size);
1512         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1513                            fs_info->thread_pool_size);
1514
1515         /*
1516          * endios are largely parallel and should have a very
1517          * low idle thresh
1518          */
1519         fs_info->endio_workers.idle_thresh = 4;
1520         fs_info->endio_write_workers.idle_thresh = 64;
1521
1522         btrfs_start_workers(&fs_info->workers, 1);
1523         btrfs_start_workers(&fs_info->submit_workers, 1);
1524         btrfs_start_workers(&fs_info->fixup_workers, 1);
1525         btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1526         btrfs_start_workers(&fs_info->endio_write_workers,
1527                             fs_info->thread_pool_size);
1528
1529         err = -EINVAL;
1530         if (btrfs_super_num_devices(disk_super) > fs_devices->open_devices) {
1531                 printk("Btrfs: wanted %llu devices, but found %llu\n",
1532                        (unsigned long long)btrfs_super_num_devices(disk_super),
1533                        (unsigned long long)fs_devices->open_devices);
1534                 if (btrfs_test_opt(tree_root, DEGRADED))
1535                         printk("continuing in degraded mode\n");
1536                 else {
1537                         goto fail_sb_buffer;
1538                 }
1539         }
1540
1541         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1542
1543         nodesize = btrfs_super_nodesize(disk_super);
1544         leafsize = btrfs_super_leafsize(disk_super);
1545         sectorsize = btrfs_super_sectorsize(disk_super);
1546         stripesize = btrfs_super_stripesize(disk_super);
1547         tree_root->nodesize = nodesize;
1548         tree_root->leafsize = leafsize;
1549         tree_root->sectorsize = sectorsize;
1550         tree_root->stripesize = stripesize;
1551
1552         sb->s_blocksize = sectorsize;
1553         sb->s_blocksize_bits = blksize_bits(sectorsize);
1554
1555         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1556                     sizeof(disk_super->magic))) {
1557                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1558                 goto fail_sb_buffer;
1559         }
1560
1561         mutex_lock(&fs_info->chunk_mutex);
1562         ret = btrfs_read_sys_array(tree_root);
1563         mutex_unlock(&fs_info->chunk_mutex);
1564         if (ret) {
1565                 printk("btrfs: failed to read the system array on %s\n",
1566                        sb->s_id);
1567                 goto fail_sys_array;
1568         }
1569
1570         blocksize = btrfs_level_size(tree_root,
1571                                      btrfs_super_chunk_root_level(disk_super));
1572
1573         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1574                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1575
1576         chunk_root->node = read_tree_block(chunk_root,
1577                                            btrfs_super_chunk_root(disk_super),
1578                                            blocksize, 0);
1579         BUG_ON(!chunk_root->node);
1580
1581         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1582                  (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1583                  BTRFS_UUID_SIZE);
1584
1585         mutex_lock(&fs_info->chunk_mutex);
1586         ret = btrfs_read_chunk_tree(chunk_root);
1587         mutex_unlock(&fs_info->chunk_mutex);
1588         BUG_ON(ret);
1589
1590         btrfs_close_extra_devices(fs_devices);
1591
1592         blocksize = btrfs_level_size(tree_root,
1593                                      btrfs_super_root_level(disk_super));
1594
1595
1596         tree_root->node = read_tree_block(tree_root,
1597                                           btrfs_super_root(disk_super),
1598                                           blocksize, 0);
1599         if (!tree_root->node)
1600                 goto fail_sb_buffer;
1601
1602
1603         ret = find_and_setup_root(tree_root, fs_info,
1604                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1605         if (ret)
1606                 goto fail_tree_root;
1607         extent_root->track_dirty = 1;
1608
1609         ret = find_and_setup_root(tree_root, fs_info,
1610                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1611         dev_root->track_dirty = 1;
1612
1613         if (ret)
1614                 goto fail_extent_root;
1615
1616         btrfs_read_block_groups(extent_root);
1617
1618         fs_info->generation = btrfs_super_generation(disk_super) + 1;
1619         fs_info->data_alloc_profile = (u64)-1;
1620         fs_info->metadata_alloc_profile = (u64)-1;
1621         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1622         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1623                                                "btrfs-cleaner");
1624         if (!fs_info->cleaner_kthread)
1625                 goto fail_extent_root;
1626
1627         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1628                                                    tree_root,
1629                                                    "btrfs-transaction");
1630         if (!fs_info->transaction_kthread)
1631                 goto fail_cleaner;
1632
1633         if (btrfs_super_log_root(disk_super) != 0) {
1634                 u32 blocksize;
1635                 u64 bytenr = btrfs_super_log_root(disk_super);
1636
1637                 blocksize =
1638                      btrfs_level_size(tree_root,
1639                                       btrfs_super_log_root_level(disk_super));
1640
1641                 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1642                                                       GFP_NOFS);
1643
1644                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1645                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1646
1647                 log_tree_root->node = read_tree_block(tree_root, bytenr,
1648                                                       blocksize, 0);
1649                 ret = btrfs_recover_log_trees(log_tree_root);
1650                 BUG_ON(ret);
1651         }
1652         fs_info->last_trans_committed = btrfs_super_generation(disk_super);
1653         return tree_root;
1654
1655 fail_cleaner:
1656         kthread_stop(fs_info->cleaner_kthread);
1657 fail_extent_root:
1658         free_extent_buffer(extent_root->node);
1659 fail_tree_root:
1660         free_extent_buffer(tree_root->node);
1661 fail_sys_array:
1662 fail_sb_buffer:
1663         btrfs_stop_workers(&fs_info->fixup_workers);
1664         btrfs_stop_workers(&fs_info->workers);
1665         btrfs_stop_workers(&fs_info->endio_workers);
1666         btrfs_stop_workers(&fs_info->endio_write_workers);
1667         btrfs_stop_workers(&fs_info->submit_workers);
1668 fail_iput:
1669         iput(fs_info->btree_inode);
1670 fail:
1671         btrfs_close_devices(fs_info->fs_devices);
1672         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1673
1674         kfree(extent_root);
1675         kfree(tree_root);
1676 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1677         bdi_destroy(&fs_info->bdi);
1678 #endif
1679         kfree(fs_info);
1680         return ERR_PTR(err);
1681 }
1682
1683 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1684 {
1685         char b[BDEVNAME_SIZE];
1686
1687         if (uptodate) {
1688                 set_buffer_uptodate(bh);
1689         } else {
1690                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1691                         printk(KERN_WARNING "lost page write due to "
1692                                         "I/O error on %s\n",
1693                                        bdevname(bh->b_bdev, b));
1694                 }
1695                 /* note, we dont' set_buffer_write_io_error because we have
1696                  * our own ways of dealing with the IO errors
1697                  */
1698                 clear_buffer_uptodate(bh);
1699         }
1700         unlock_buffer(bh);
1701         put_bh(bh);
1702 }
1703
1704 int write_all_supers(struct btrfs_root *root)
1705 {
1706         struct list_head *cur;
1707         struct list_head *head = &root->fs_info->fs_devices->devices;
1708         struct btrfs_device *dev;
1709         struct btrfs_super_block *sb;
1710         struct btrfs_dev_item *dev_item;
1711         struct buffer_head *bh;
1712         int ret;
1713         int do_barriers;
1714         int max_errors;
1715         int total_errors = 0;
1716         u32 crc;
1717         u64 flags;
1718
1719         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1720         do_barriers = !btrfs_test_opt(root, NOBARRIER);
1721
1722         sb = &root->fs_info->super_for_commit;
1723         dev_item = &sb->dev_item;
1724         list_for_each(cur, head) {
1725                 dev = list_entry(cur, struct btrfs_device, dev_list);
1726                 if (!dev->bdev) {
1727                         total_errors++;
1728                         continue;
1729                 }
1730                 if (!dev->in_fs_metadata)
1731                         continue;
1732
1733                 btrfs_set_stack_device_type(dev_item, dev->type);
1734                 btrfs_set_stack_device_id(dev_item, dev->devid);
1735                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1736                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1737                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1738                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1739                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1740                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1741                 flags = btrfs_super_flags(sb);
1742                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1743
1744
1745                 crc = ~(u32)0;
1746                 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1747                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1748                 btrfs_csum_final(crc, sb->csum);
1749
1750                 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1751                               BTRFS_SUPER_INFO_SIZE);
1752
1753                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1754                 dev->pending_io = bh;
1755
1756                 get_bh(bh);
1757                 set_buffer_uptodate(bh);
1758                 lock_buffer(bh);
1759                 bh->b_end_io = btrfs_end_buffer_write_sync;
1760
1761                 if (do_barriers && dev->barriers) {
1762                         ret = submit_bh(WRITE_BARRIER, bh);
1763                         if (ret == -EOPNOTSUPP) {
1764                                 printk("btrfs: disabling barriers on dev %s\n",
1765                                        dev->name);
1766                                 set_buffer_uptodate(bh);
1767                                 dev->barriers = 0;
1768                                 get_bh(bh);
1769                                 lock_buffer(bh);
1770                                 ret = submit_bh(WRITE, bh);
1771                         }
1772                 } else {
1773                         ret = submit_bh(WRITE, bh);
1774                 }
1775                 if (ret)
1776                         total_errors++;
1777         }
1778         if (total_errors > max_errors) {
1779                 printk("btrfs: %d errors while writing supers\n", total_errors);
1780                 BUG();
1781         }
1782         total_errors = 0;
1783
1784         list_for_each(cur, head) {
1785                 dev = list_entry(cur, struct btrfs_device, dev_list);
1786                 if (!dev->bdev)
1787                         continue;
1788                 if (!dev->in_fs_metadata)
1789                         continue;
1790
1791                 BUG_ON(!dev->pending_io);
1792                 bh = dev->pending_io;
1793                 wait_on_buffer(bh);
1794                 if (!buffer_uptodate(dev->pending_io)) {
1795                         if (do_barriers && dev->barriers) {
1796                                 printk("btrfs: disabling barriers on dev %s\n",
1797                                        dev->name);
1798                                 set_buffer_uptodate(bh);
1799                                 get_bh(bh);
1800                                 lock_buffer(bh);
1801                                 dev->barriers = 0;
1802                                 ret = submit_bh(WRITE, bh);
1803                                 BUG_ON(ret);
1804                                 wait_on_buffer(bh);
1805                                 if (!buffer_uptodate(bh))
1806                                         total_errors++;
1807                         } else {
1808                                 total_errors++;
1809                         }
1810
1811                 }
1812                 dev->pending_io = NULL;
1813                 brelse(bh);
1814         }
1815         if (total_errors > max_errors) {
1816                 printk("btrfs: %d errors while writing supers\n", total_errors);
1817                 BUG();
1818         }
1819         return 0;
1820 }
1821
1822 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1823                       *root)
1824 {
1825         int ret;
1826
1827         ret = write_all_supers(root);
1828         return ret;
1829 }
1830
1831 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1832 {
1833         radix_tree_delete(&fs_info->fs_roots_radix,
1834                           (unsigned long)root->root_key.objectid);
1835         if (root->in_sysfs)
1836                 btrfs_sysfs_del_root(root);
1837         if (root->inode)
1838                 iput(root->inode);
1839         if (root->node)
1840                 free_extent_buffer(root->node);
1841         if (root->commit_root)
1842                 free_extent_buffer(root->commit_root);
1843         if (root->name)
1844                 kfree(root->name);
1845         kfree(root);
1846         return 0;
1847 }
1848
1849 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1850 {
1851         int ret;
1852         struct btrfs_root *gang[8];
1853         int i;
1854
1855         while(1) {
1856                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1857                                              (void **)gang, 0,
1858                                              ARRAY_SIZE(gang));
1859                 if (!ret)
1860                         break;
1861                 for (i = 0; i < ret; i++)
1862                         btrfs_free_fs_root(fs_info, gang[i]);
1863         }
1864         return 0;
1865 }
1866
1867 int close_ctree(struct btrfs_root *root)
1868 {
1869         int ret;
1870         struct btrfs_trans_handle *trans;
1871         struct btrfs_fs_info *fs_info = root->fs_info;
1872
1873         fs_info->closing = 1;
1874         smp_mb();
1875
1876         kthread_stop(root->fs_info->transaction_kthread);
1877         kthread_stop(root->fs_info->cleaner_kthread);
1878
1879         btrfs_clean_old_snapshots(root);
1880         trans = btrfs_start_transaction(root, 1);
1881         ret = btrfs_commit_transaction(trans, root);
1882         /* run commit again to  drop the original snapshot */
1883         trans = btrfs_start_transaction(root, 1);
1884         btrfs_commit_transaction(trans, root);
1885         ret = btrfs_write_and_wait_transaction(NULL, root);
1886         BUG_ON(ret);
1887
1888         write_ctree_super(NULL, root);
1889
1890         if (fs_info->delalloc_bytes) {
1891                 printk("btrfs: at unmount delalloc count %Lu\n",
1892                        fs_info->delalloc_bytes);
1893         }
1894         if (fs_info->total_ref_cache_size) {
1895                 printk("btrfs: at umount reference cache size %Lu\n",
1896                         fs_info->total_ref_cache_size);
1897         }
1898
1899         if (fs_info->extent_root->node)
1900                 free_extent_buffer(fs_info->extent_root->node);
1901
1902         if (fs_info->tree_root->node)
1903                 free_extent_buffer(fs_info->tree_root->node);
1904
1905         if (root->fs_info->chunk_root->node);
1906                 free_extent_buffer(root->fs_info->chunk_root->node);
1907
1908         if (root->fs_info->dev_root->node);
1909                 free_extent_buffer(root->fs_info->dev_root->node);
1910
1911         btrfs_free_block_groups(root->fs_info);
1912         fs_info->closing = 2;
1913         del_fs_roots(fs_info);
1914
1915         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1916
1917         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1918
1919         btrfs_stop_workers(&fs_info->fixup_workers);
1920         btrfs_stop_workers(&fs_info->workers);
1921         btrfs_stop_workers(&fs_info->endio_workers);
1922         btrfs_stop_workers(&fs_info->endio_write_workers);
1923         btrfs_stop_workers(&fs_info->submit_workers);
1924
1925         iput(fs_info->btree_inode);
1926 #if 0
1927         while(!list_empty(&fs_info->hashers)) {
1928                 struct btrfs_hasher *hasher;
1929                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1930                                     hashers);
1931                 list_del(&hasher->hashers);
1932                 crypto_free_hash(&fs_info->hash_tfm);
1933                 kfree(hasher);
1934         }
1935 #endif
1936         btrfs_close_devices(fs_info->fs_devices);
1937         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1938
1939 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1940         bdi_destroy(&fs_info->bdi);
1941 #endif
1942
1943         kfree(fs_info->extent_root);
1944         kfree(fs_info->tree_root);
1945         kfree(fs_info->chunk_root);
1946         kfree(fs_info->dev_root);
1947         return 0;
1948 }
1949
1950 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1951 {
1952         int ret;
1953         struct inode *btree_inode = buf->first_page->mapping->host;
1954
1955         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1956         if (!ret)
1957                 return ret;
1958
1959         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
1960                                     parent_transid);
1961         return !ret;
1962 }
1963
1964 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1965 {
1966         struct inode *btree_inode = buf->first_page->mapping->host;
1967         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1968                                           buf);
1969 }
1970
1971 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1972 {
1973         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1974         u64 transid = btrfs_header_generation(buf);
1975         struct inode *btree_inode = root->fs_info->btree_inode;
1976
1977         WARN_ON(!btrfs_tree_locked(buf));
1978         if (transid != root->fs_info->generation) {
1979                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1980                         (unsigned long long)buf->start,
1981                         transid, root->fs_info->generation);
1982                 WARN_ON(1);
1983         }
1984         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1985 }
1986
1987 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1988 {
1989         /*
1990          * looks as though older kernels can get into trouble with
1991          * this code, they end up stuck in balance_dirty_pages forever
1992          */
1993         struct extent_io_tree *tree;
1994         u64 num_dirty;
1995         u64 start = 0;
1996         unsigned long thresh = 96 * 1024 * 1024;
1997         tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
1998
1999         if (current_is_pdflush() || current->flags & PF_MEMALLOC)
2000                 return;
2001
2002         num_dirty = count_range_bits(tree, &start, (u64)-1,
2003                                      thresh, EXTENT_DIRTY);
2004         if (num_dirty > thresh) {
2005                 balance_dirty_pages_ratelimited_nr(
2006                                    root->fs_info->btree_inode->i_mapping, 1);
2007         }
2008         return;
2009 }
2010
2011 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2012 {
2013         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2014         int ret;
2015         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2016         if (ret == 0) {
2017                 buf->flags |= EXTENT_UPTODATE;
2018         }
2019         return ret;
2020 }
2021
2022 int btree_lock_page_hook(struct page *page)
2023 {
2024         struct inode *inode = page->mapping->host;
2025         struct btrfs_root *root = BTRFS_I(inode)->root;
2026         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2027         struct extent_buffer *eb;
2028         unsigned long len;
2029         u64 bytenr = page_offset(page);
2030
2031         if (page->private == EXTENT_PAGE_PRIVATE)
2032                 goto out;
2033
2034         len = page->private >> 2;
2035         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2036         if (!eb)
2037                 goto out;
2038
2039         btrfs_tree_lock(eb);
2040         spin_lock(&root->fs_info->hash_lock);
2041         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2042         spin_unlock(&root->fs_info->hash_lock);
2043         btrfs_tree_unlock(eb);
2044         free_extent_buffer(eb);
2045 out:
2046         lock_page(page);
2047         return 0;
2048 }
2049
2050 static struct extent_io_ops btree_extent_io_ops = {
2051         .write_cache_pages_lock_hook = btree_lock_page_hook,
2052         .readpage_end_io_hook = btree_readpage_end_io_hook,
2053         .submit_bio_hook = btree_submit_bio_hook,
2054         /* note we're sharing with inode.c for the merge bio hook */
2055         .merge_bio_hook = btrfs_merge_bio_hook,
2056 };