git.oblomov.eu Git - linux-2.6/blob - 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_start;
  84         extent_submit_bio_hook_t *submit_bio_done;
  85         int rw;
  86         int mirror_num;
  87         unsigned long bio_flags;
  88         struct btrfs_work work;
  89 };
  90
  91 /*
  92  * extents on the btree inode are pretty simple, there's one extent
  93  * that covers the entire device
  94  */
  95 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
  96                                     size_t page_offset, u64 start, u64 len,
  97                                     int create)
  98 {
  99         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 100         struct extent_map *em;
 101         int ret;
 102
 103         spin_lock(&em_tree->lock);
 104         em = lookup_extent_mapping(em_tree, start, len);
 105         if (em) {
 106                 em->bdev =
 107                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 108                 spin_unlock(&em_tree->lock);
 109                 goto out;
 110         }
 111         spin_unlock(&em_tree->lock);
 112
 113         em = alloc_extent_map(GFP_NOFS);
 114         if (!em) {
 115                 em = ERR_PTR(-ENOMEM);
 116                 goto out;
 117         }
 118         em->start = 0;
 119         em->len = (u64)-1;
 120         em->block_len = (u64)-1;
 121         em->block_start = 0;
 122         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 123
 124         spin_lock(&em_tree->lock);
 125         ret = add_extent_mapping(em_tree, em);
 126         if (ret == -EEXIST) {
 127                 u64 failed_start = em->start;
 128                 u64 failed_len = em->len;
 129
 130                 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
 131                        em->start, em->len, em->block_start);
 132                 free_extent_map(em);
 133                 em = lookup_extent_mapping(em_tree, start, len);
 134                 if (em) {
 135                         printk("after failing, found %Lu %Lu %Lu\n",
 136                                em->start, em->len, em->block_start);
 137                         ret = 0;
 138                 } else {
 139                         em = lookup_extent_mapping(em_tree, failed_start,
 140                                                    failed_len);
 141                         if (em) {
 142                                 printk("double failure lookup gives us "
 143                                        "%Lu %Lu -> %Lu\n", em->start,
 144                                        em->len, em->block_start);
 145                                 free_extent_map(em);
 146                         }
 147                         ret = -EIO;
 148                 }
 149         } else if (ret) {
 150                 free_extent_map(em);
 151                 em = NULL;
 152         }
 153         spin_unlock(&em_tree->lock);
 154
 155         if (ret)
 156                 em = ERR_PTR(ret);
 157 out:
 158         return em;
 159 }
 160
 161 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 162 {
 163         return btrfs_crc32c(seed, data, len);
 164 }
 165
 166 void btrfs_csum_final(u32 crc, char *result)
 167 {
 168         *(__le32 *)result = ~cpu_to_le32(crc);
 169 }
 170
 171 /*
 172  * compute the csum for a btree block, and either verify it or write it
 173  * into the csum field of the block.
 174  */
 175 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 176                            int verify)
 177 {
 178         char result[BTRFS_CRC32_SIZE];
 179         unsigned long len;
 180         unsigned long cur_len;
 181         unsigned long offset = BTRFS_CSUM_SIZE;
 182         char *map_token = NULL;
 183         char *kaddr;
 184         unsigned long map_start;
 185         unsigned long map_len;
 186         int err;
 187         u32 crc = ~(u32)0;
 188
 189         len = buf->len - offset;
 190         while(len > 0) {
 191                 err = map_private_extent_buffer(buf, offset, 32,
 192                                         &map_token, &kaddr,
 193                                         &map_start, &map_len, KM_USER0);
 194                 if (err) {
 195                         printk("failed to map extent buffer! %lu\n",
 196                                offset);
 197                         return 1;
 198                 }
 199                 cur_len = min(len, map_len - (offset - map_start));
 200                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 201                                       crc, cur_len);
 202                 len -= cur_len;
 203                 offset += cur_len;
 204                 unmap_extent_buffer(buf, map_token, KM_USER0);
 205         }
 206         btrfs_csum_final(crc, result);
 207
 208         if (verify) {
 209                 /* FIXME, this is not good */
 210                 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
 211                         u32 val;
 212                         u32 found = 0;
 213                         memcpy(&found, result, BTRFS_CRC32_SIZE);
 214
 215                         read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
 216                         printk("btrfs: %s checksum verify failed on %llu "
 217                                "wanted %X found %X level %d\n",
 218                                root->fs_info->sb->s_id,
 219                                buf->start, val, found, btrfs_header_level(buf));
 220                         return 1;
 221                 }
 222         } else {
 223                 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
 224         }
 225         return 0;
 226 }
 227
 228 /*
 229  * we can't consider a given block up to date unless the transid of the
 230  * block matches the transid in the parent node's pointer.  This is how we
 231  * detect blocks that either didn't get written at all or got written
 232  * in the wrong place.
 233  */
 234 static int verify_parent_transid(struct extent_io_tree *io_tree,
 235                                  struct extent_buffer *eb, u64 parent_transid)
 236 {
 237         int ret;
 238
 239         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 240                 return 0;
 241
 242         lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
 243         if (extent_buffer_uptodate(io_tree, eb) &&
 244             btrfs_header_generation(eb) == parent_transid) {
 245                 ret = 0;
 246                 goto out;
 247         }
 248         printk("parent transid verify failed on %llu wanted %llu found %llu\n",
 249                (unsigned long long)eb->start,
 250                (unsigned long long)parent_transid,
 251                (unsigned long long)btrfs_header_generation(eb));
 252         ret = 1;
 253         clear_extent_buffer_uptodate(io_tree, eb);
 254 out:
 255         unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
 256                       GFP_NOFS);
 257         return ret;
 258 }
 259
 260 /*
 261  * helper to read a given tree block, doing retries as required when
 262  * the checksums don't match and we have alternate mirrors to try.
 263  */
 264 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 265                                           struct extent_buffer *eb,
 266                                           u64 start, u64 parent_transid)
 267 {
 268         struct extent_io_tree *io_tree;
 269         int ret;
 270         int num_copies = 0;
 271         int mirror_num = 0;
 272
 273         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 274         while (1) {
 275                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 276                                                btree_get_extent, mirror_num);
 277                 if (!ret &&
 278                     !verify_parent_transid(io_tree, eb, parent_transid))
 279                         return ret;
 280 printk("read extent buffer pages failed with ret %d mirror no %d\n", ret, mirror_num);
 281                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 282                                               eb->start, eb->len);
 283                 if (num_copies == 1)
 284                         return ret;
 285
 286                 mirror_num++;
 287                 if (mirror_num > num_copies)
 288                         return ret;
 289         }
 290         return -EIO;
 291 }
 292
 293 /*
 294  * checksum a dirty tree block before IO.  This has extra checks to make
 295  * sure we only fill in the checksum field in the first page of a multi-page block
 296  */
 297 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 298 {
 299         struct extent_io_tree *tree;
 300         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 301         u64 found_start;
 302         int found_level;
 303         unsigned long len;
 304         struct extent_buffer *eb;
 305         int ret;
 306
 307         tree = &BTRFS_I(page->mapping->host)->io_tree;
 308
 309         if (page->private == EXTENT_PAGE_PRIVATE)
 310                 goto out;
 311         if (!page->private)
 312                 goto out;
 313         len = page->private >> 2;
 314         if (len == 0) {
 315                 WARN_ON(1);
 316         }
 317         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 318         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 319                                              btrfs_header_generation(eb));
 320         BUG_ON(ret);
 321         found_start = btrfs_header_bytenr(eb);
 322         if (found_start != start) {
 323                 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
 324                        start, found_start, len);
 325                 WARN_ON(1);
 326                 goto err;
 327         }
 328         if (eb->first_page != page) {
 329                 printk("bad first page %lu %lu\n", eb->first_page->index,
 330                        page->index);
 331                 WARN_ON(1);
 332                 goto err;
 333         }
 334         if (!PageUptodate(page)) {
 335                 printk("csum not up to date page %lu\n", page->index);
 336                 WARN_ON(1);
 337                 goto err;
 338         }
 339         found_level = btrfs_header_level(eb);
 340
 341         csum_tree_block(root, eb, 0);
 342 err:
 343         free_extent_buffer(eb);
 344 out:
 345         return 0;
 346 }
 347
 348 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 349                                struct extent_state *state)
 350 {
 351         struct extent_io_tree *tree;
 352         u64 found_start;
 353         int found_level;
 354         unsigned long len;
 355         struct extent_buffer *eb;
 356         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 357         int ret = 0;
 358
 359         tree = &BTRFS_I(page->mapping->host)->io_tree;
 360         if (page->private == EXTENT_PAGE_PRIVATE)
 361                 goto out;
 362         if (!page->private)
 363                 goto out;
 364         len = page->private >> 2;
 365         if (len == 0) {
 366                 WARN_ON(1);
 367         }
 368         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 369
 370         found_start = btrfs_header_bytenr(eb);
 371         if (found_start != start) {
 372                 printk("bad tree block start %llu %llu\n",
 373                        (unsigned long long)found_start,
 374                        (unsigned long long)eb->start);
 375                 ret = -EIO;
 376                 goto err;
 377         }
 378         if (eb->first_page != page) {
 379                 printk("bad first page %lu %lu\n", eb->first_page->index,
 380                        page->index);
 381                 WARN_ON(1);
 382                 ret = -EIO;
 383                 goto err;
 384         }
 385         if (memcmp_extent_buffer(eb, root->fs_info->fsid,
 386                                  (unsigned long)btrfs_header_fsid(eb),
 387                                  BTRFS_FSID_SIZE)) {
 388                 printk("bad fsid on block %Lu\n", eb->start);
 389                 ret = -EIO;
 390                 goto err;
 391         }
 392         found_level = btrfs_header_level(eb);
 393
 394         ret = csum_tree_block(root, eb, 1);
 395         if (ret)
 396                 ret = -EIO;
 397
 398         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 399         end = eb->start + end - 1;
 400 err:
 401         free_extent_buffer(eb);
 402 out:
 403         return ret;
 404 }
 405
 406 static void end_workqueue_bio(struct bio *bio, int err)
 407 {
 408         struct end_io_wq *end_io_wq = bio->bi_private;
 409         struct btrfs_fs_info *fs_info;
 410
 411         fs_info = end_io_wq->info;
 412         end_io_wq->error = err;
 413         end_io_wq->work.func = end_workqueue_fn;
 414         end_io_wq->work.flags = 0;
 415         if (bio->bi_rw & (1 << BIO_RW))
 416                 btrfs_queue_worker(&fs_info->endio_write_workers,
 417                                    &end_io_wq->work);
 418         else
 419                 btrfs_queue_worker(&fs_info->endio_workers, &end_io_wq->work);
 420 }
 421
 422 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 423                         int metadata)
 424 {
 425         struct end_io_wq *end_io_wq;
 426         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 427         if (!end_io_wq)
 428                 return -ENOMEM;
 429
 430         end_io_wq->private = bio->bi_private;
 431         end_io_wq->end_io = bio->bi_end_io;
 432         end_io_wq->info = info;
 433         end_io_wq->error = 0;
 434         end_io_wq->bio = bio;
 435         end_io_wq->metadata = metadata;
 436
 437         bio->bi_private = end_io_wq;
 438         bio->bi_end_io = end_workqueue_bio;
 439         return 0;
 440 }
 441
 442 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 443 {
 444         unsigned long limit = min_t(unsigned long,
 445                                     info->workers.max_workers,
 446                                     info->fs_devices->open_devices);
 447         return 256 * limit;
 448 }
 449
 450 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
 451 {
 452         return atomic_read(&info->nr_async_bios) >
 453                 btrfs_async_submit_limit(info);
 454 }
 455
 456 static void run_one_async_start(struct btrfs_work *work)
 457 {
 458         struct btrfs_fs_info *fs_info;
 459         struct async_submit_bio *async;
 460
 461         async = container_of(work, struct  async_submit_bio, work);
 462         fs_info = BTRFS_I(async->inode)->root->fs_info;
 463         async->submit_bio_start(async->inode, async->rw, async->bio,
 464                                async->mirror_num, async->bio_flags);
 465 }
 466
 467 static void run_one_async_done(struct btrfs_work *work)
 468 {
 469         struct btrfs_fs_info *fs_info;
 470         struct async_submit_bio *async;
 471         int limit;
 472
 473         async = container_of(work, struct  async_submit_bio, work);
 474         fs_info = BTRFS_I(async->inode)->root->fs_info;
 475
 476         limit = btrfs_async_submit_limit(fs_info);
 477         limit = limit * 2 / 3;
 478
 479         atomic_dec(&fs_info->nr_async_submits);
 480
 481         if (atomic_read(&fs_info->nr_async_submits) < limit &&
 482             waitqueue_active(&fs_info->async_submit_wait))
 483                 wake_up(&fs_info->async_submit_wait);
 484
 485         async->submit_bio_done(async->inode, async->rw, async->bio,
 486                                async->mirror_num, async->bio_flags);
 487 }
 488
 489 static void run_one_async_free(struct btrfs_work *work)
 490 {
 491         struct async_submit_bio *async;
 492
 493         async = container_of(work, struct  async_submit_bio, work);
 494         kfree(async);
 495 }
 496
 497 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 498                         int rw, struct bio *bio, int mirror_num,
 499                         unsigned long bio_flags,
 500                         extent_submit_bio_hook_t *submit_bio_start,
 501                         extent_submit_bio_hook_t *submit_bio_done)
 502 {
 503         struct async_submit_bio *async;
 504         int limit = btrfs_async_submit_limit(fs_info);
 505
 506         async = kmalloc(sizeof(*async), GFP_NOFS);
 507         if (!async)
 508                 return -ENOMEM;
 509
 510         async->inode = inode;
 511         async->rw = rw;
 512         async->bio = bio;
 513         async->mirror_num = mirror_num;
 514         async->submit_bio_start = submit_bio_start;
 515         async->submit_bio_done = submit_bio_done;
 516
 517         async->work.func = run_one_async_start;
 518         async->work.ordered_func = run_one_async_done;
 519         async->work.ordered_free = run_one_async_free;
 520
 521         async->work.flags = 0;
 522         async->bio_flags = bio_flags;
 523
 524         while(atomic_read(&fs_info->async_submit_draining) &&
 525               atomic_read(&fs_info->nr_async_submits)) {
 526                 wait_event(fs_info->async_submit_wait,
 527                            (atomic_read(&fs_info->nr_async_submits) == 0));
 528         }
 529
 530         atomic_inc(&fs_info->nr_async_submits);
 531         btrfs_queue_worker(&fs_info->workers, &async->work);
 532
 533         if (atomic_read(&fs_info->nr_async_submits) > limit) {
 534                 wait_event_timeout(fs_info->async_submit_wait,
 535                            (atomic_read(&fs_info->nr_async_submits) < limit),
 536                            HZ/10);
 537
 538                 wait_event_timeout(fs_info->async_submit_wait,
 539                            (atomic_read(&fs_info->nr_async_bios) < limit),
 540                            HZ/10);
 541         }
 542
 543         while(atomic_read(&fs_info->async_submit_draining) &&
 544               atomic_read(&fs_info->nr_async_submits)) {
 545                 wait_event(fs_info->async_submit_wait,
 546                            (atomic_read(&fs_info->nr_async_submits) == 0));
 547         }
 548
 549         return 0;
 550 }
 551
 552 static int btree_csum_one_bio(struct bio *bio)
 553 {
 554         struct bio_vec *bvec = bio->bi_io_vec;
 555         int bio_index = 0;
 556         struct btrfs_root *root;
 557
 558         WARN_ON(bio->bi_vcnt <= 0);
 559         while(bio_index < bio->bi_vcnt) {
 560                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 561                 csum_dirty_buffer(root, bvec->bv_page);
 562                 bio_index++;
 563                 bvec++;
 564         }
 565         return 0;
 566 }
 567
 568 static int __btree_submit_bio_start(struct inode *inode, int rw,
 569                                     struct bio *bio, int mirror_num,
 570                                     unsigned long bio_flags)
 571 {
 572         /*
 573          * when we're called for a write, we're already in the async
 574          * submission context.  Just jump into btrfs_map_bio
 575          */
 576         btree_csum_one_bio(bio);
 577         return 0;
 578 }
 579
 580 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 581                                  int mirror_num, unsigned long bio_flags)
 582 {
 583         /*
 584          * when we're called for a write, we're already in the async
 585          * submission context.  Just jump into btrfs_map_bio
 586          */
 587         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 588 }
 589
 590 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 591                                  int mirror_num, unsigned long bio_flags)
 592 {
 593         /*
 594          * kthread helpers are used to submit writes so that checksumming
 595          * can happen in parallel across all CPUs
 596          */
 597         if (!(rw & (1 << BIO_RW))) {
 598                 int ret;
 599                 /*
 600                  * called for a read, do the setup so that checksum validation
 601                  * can happen in the async kernel threads
 602                  */
 603                 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 604                                           bio, 1);
 605                 BUG_ON(ret);
 606
 607                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 608                                      mirror_num, 0);
 609         }
 610         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 611                                    inode, rw, bio, mirror_num, 0,
 612                                    __btree_submit_bio_start,
 613                                    __btree_submit_bio_done);
 614 }
 615
 616 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 617 {
 618         struct extent_io_tree *tree;
 619         tree = &BTRFS_I(page->mapping->host)->io_tree;
 620
 621         if (current->flags & PF_MEMALLOC) {
 622                 redirty_page_for_writepage(wbc, page);
 623                 unlock_page(page);
 624                 return 0;
 625         }
 626         return extent_write_full_page(tree, page, btree_get_extent, wbc);
 627 }
 628
 629 static int btree_writepages(struct address_space *mapping,
 630                             struct writeback_control *wbc)
 631 {
 632         struct extent_io_tree *tree;
 633         tree = &BTRFS_I(mapping->host)->io_tree;
 634         if (wbc->sync_mode == WB_SYNC_NONE) {
 635                 u64 num_dirty;
 636                 u64 start = 0;
 637                 unsigned long thresh = 32 * 1024 * 1024;
 638
 639                 if (wbc->for_kupdate)
 640                         return 0;
 641
 642                 num_dirty = count_range_bits(tree, &start, (u64)-1,
 643                                              thresh, EXTENT_DIRTY);
 644                 if (num_dirty < thresh) {
 645                         return 0;
 646                 }
 647         }
 648         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 649 }
 650
 651 int btree_readpage(struct file *file, struct page *page)
 652 {
 653         struct extent_io_tree *tree;
 654         tree = &BTRFS_I(page->mapping->host)->io_tree;
 655         return extent_read_full_page(tree, page, btree_get_extent);
 656 }
 657
 658 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 659 {
 660         struct extent_io_tree *tree;
 661         struct extent_map_tree *map;
 662         int ret;
 663
 664         if (PageWriteback(page) || PageDirty(page))
 665             return 0;
 666
 667         tree = &BTRFS_I(page->mapping->host)->io_tree;
 668         map = &BTRFS_I(page->mapping->host)->extent_tree;
 669
 670         ret = try_release_extent_state(map, tree, page, gfp_flags);
 671         if (!ret) {
 672                 return 0;
 673         }
 674
 675         ret = try_release_extent_buffer(tree, page);
 676         if (ret == 1) {
 677                 ClearPagePrivate(page);
 678                 set_page_private(page, 0);
 679                 page_cache_release(page);
 680         }
 681
 682         return ret;
 683 }
 684
 685 static void btree_invalidatepage(struct page *page, unsigned long offset)
 686 {
 687         struct extent_io_tree *tree;
 688         tree = &BTRFS_I(page->mapping->host)->io_tree;
 689         extent_invalidatepage(tree, page, offset);
 690         btree_releasepage(page, GFP_NOFS);
 691         if (PagePrivate(page)) {
 692                 printk("warning page private not zero on page %Lu\n",
 693                        page_offset(page));
 694                 ClearPagePrivate(page);
 695                 set_page_private(page, 0);
 696                 page_cache_release(page);
 697         }
 698 }
 699
 700 #if 0
 701 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 702 {
 703         struct buffer_head *bh;
 704         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 705         struct buffer_head *head;
 706         if (!page_has_buffers(page)) {
 707                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 708                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 709         }
 710         head = page_buffers(page);
 711         bh = head;
 712         do {
 713                 if (buffer_dirty(bh))
 714                         csum_tree_block(root, bh, 0);
 715                 bh = bh->b_this_page;
 716         } while (bh != head);
 717         return block_write_full_page(page, btree_get_block, wbc);
 718 }
 719 #endif
 720
 721 static struct address_space_operations btree_aops = {
 722         .readpage       = btree_readpage,
 723         .writepage      = btree_writepage,
 724         .writepages     = btree_writepages,
 725         .releasepage    = btree_releasepage,
 726         .invalidatepage = btree_invalidatepage,
 727         .sync_page      = block_sync_page,
 728 };
 729
 730 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 731                          u64 parent_transid)
 732 {
 733         struct extent_buffer *buf = NULL;
 734         struct inode *btree_inode = root->fs_info->btree_inode;
 735         int ret = 0;
 736
 737         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 738         if (!buf)
 739                 return 0;
 740         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 741                                  buf, 0, 0, btree_get_extent, 0);
 742         free_extent_buffer(buf);
 743         return ret;
 744 }
 745
 746 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 747                                             u64 bytenr, u32 blocksize)
 748 {
 749         struct inode *btree_inode = root->fs_info->btree_inode;
 750         struct extent_buffer *eb;
 751         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 752                                 bytenr, blocksize, GFP_NOFS);
 753         return eb;
 754 }
 755
 756 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 757                                                  u64 bytenr, u32 blocksize)
 758 {
 759         struct inode *btree_inode = root->fs_info->btree_inode;
 760         struct extent_buffer *eb;
 761
 762         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 763                                  bytenr, blocksize, NULL, GFP_NOFS);
 764         return eb;
 765 }
 766
 767
 768 int btrfs_write_tree_block(struct extent_buffer *buf)
 769 {
 770         return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
 771                                       buf->start + buf->len - 1, WB_SYNC_ALL);
 772 }
 773
 774 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 775 {
 776         return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
 777                                   buf->start, buf->start + buf->len -1);
 778 }
 779
 780 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 781                                       u32 blocksize, u64 parent_transid)
 782 {
 783         struct extent_buffer *buf = NULL;
 784         struct inode *btree_inode = root->fs_info->btree_inode;
 785         struct extent_io_tree *io_tree;
 786         int ret;
 787
 788         io_tree = &BTRFS_I(btree_inode)->io_tree;
 789
 790         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 791         if (!buf)
 792                 return NULL;
 793
 794         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 795
 796         if (ret == 0) {
 797                 buf->flags |= EXTENT_UPTODATE;
 798         } else {
 799                 WARN_ON(1);
 800         }
 801         return buf;
 802
 803 }
 804
 805 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 806                      struct extent_buffer *buf)
 807 {
 808         struct inode *btree_inode = root->fs_info->btree_inode;
 809         if (btrfs_header_generation(buf) ==
 810             root->fs_info->running_transaction->transid) {
 811                 WARN_ON(!btrfs_tree_locked(buf));
 812                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 813                                           buf);
 814         }
 815         return 0;
 816 }
 817
 818 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 819                         u32 stripesize, struct btrfs_root *root,
 820                         struct btrfs_fs_info *fs_info,
 821                         u64 objectid)
 822 {
 823         root->node = NULL;
 824         root->inode = NULL;
 825         root->commit_root = NULL;
 826         root->ref_tree = NULL;
 827         root->sectorsize = sectorsize;
 828         root->nodesize = nodesize;
 829         root->leafsize = leafsize;
 830         root->stripesize = stripesize;
 831         root->ref_cows = 0;
 832         root->track_dirty = 0;
 833
 834         root->fs_info = fs_info;
 835         root->objectid = objectid;
 836         root->last_trans = 0;
 837         root->highest_inode = 0;
 838         root->last_inode_alloc = 0;
 839         root->name = NULL;
 840         root->in_sysfs = 0;
 841
 842         INIT_LIST_HEAD(&root->dirty_list);
 843         INIT_LIST_HEAD(&root->orphan_list);
 844         INIT_LIST_HEAD(&root->dead_list);
 845         spin_lock_init(&root->node_lock);
 846         spin_lock_init(&root->list_lock);
 847         mutex_init(&root->objectid_mutex);
 848         mutex_init(&root->log_mutex);
 849         extent_io_tree_init(&root->dirty_log_pages,
 850                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 851
 852         btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
 853         root->ref_tree = &root->ref_tree_struct;
 854
 855         memset(&root->root_key, 0, sizeof(root->root_key));
 856         memset(&root->root_item, 0, sizeof(root->root_item));
 857         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 858         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 859         root->defrag_trans_start = fs_info->generation;
 860         init_completion(&root->kobj_unregister);
 861         root->defrag_running = 0;
 862         root->defrag_level = 0;
 863         root->root_key.objectid = objectid;
 864         return 0;
 865 }
 866
 867 static int find_and_setup_root(struct btrfs_root *tree_root,
 868                                struct btrfs_fs_info *fs_info,
 869                                u64 objectid,
 870                                struct btrfs_root *root)
 871 {
 872         int ret;
 873         u32 blocksize;
 874         u64 generation;
 875
 876         __setup_root(tree_root->nodesize, tree_root->leafsize,
 877                      tree_root->sectorsize, tree_root->stripesize,
 878                      root, fs_info, objectid);
 879         ret = btrfs_find_last_root(tree_root, objectid,
 880                                    &root->root_item, &root->root_key);
 881         BUG_ON(ret);
 882
 883         generation = btrfs_root_generation(&root->root_item);
 884         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 885         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 886                                      blocksize, generation);
 887         BUG_ON(!root->node);
 888         return 0;
 889 }
 890
 891 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
 892                              struct btrfs_fs_info *fs_info)
 893 {
 894         struct extent_buffer *eb;
 895         struct btrfs_root *log_root_tree = fs_info->log_root_tree;
 896         u64 start = 0;
 897         u64 end = 0;
 898         int ret;
 899
 900         if (!log_root_tree)
 901                 return 0;
 902
 903         while(1) {
 904                 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
 905                                     0, &start, &end, EXTENT_DIRTY);
 906                 if (ret)
 907                         break;
 908
 909                 clear_extent_dirty(&log_root_tree->dirty_log_pages,
 910                                    start, end, GFP_NOFS);
 911         }
 912         eb = fs_info->log_root_tree->node;
 913
 914         WARN_ON(btrfs_header_level(eb) != 0);
 915         WARN_ON(btrfs_header_nritems(eb) != 0);
 916
 917         ret = btrfs_free_reserved_extent(fs_info->tree_root,
 918                                 eb->start, eb->len);
 919         BUG_ON(ret);
 920
 921         free_extent_buffer(eb);
 922         kfree(fs_info->log_root_tree);
 923         fs_info->log_root_tree = NULL;
 924         return 0;
 925 }
 926
 927 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
 928                              struct btrfs_fs_info *fs_info)
 929 {
 930         struct btrfs_root *root;
 931         struct btrfs_root *tree_root = fs_info->tree_root;
 932
 933         root = kzalloc(sizeof(*root), GFP_NOFS);
 934         if (!root)
 935                 return -ENOMEM;
 936
 937         __setup_root(tree_root->nodesize, tree_root->leafsize,
 938                      tree_root->sectorsize, tree_root->stripesize,
 939                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
 940
 941         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
 942         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
 943         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
 944         root->ref_cows = 0;
 945
 946         root->node = btrfs_alloc_free_block(trans, root, root->leafsize,
 947                                             0, BTRFS_TREE_LOG_OBJECTID,
 948                                             trans->transid, 0, 0, 0);
 949
 950         btrfs_set_header_nritems(root->node, 0);
 951         btrfs_set_header_level(root->node, 0);
 952         btrfs_set_header_bytenr(root->node, root->node->start);
 953         btrfs_set_header_generation(root->node, trans->transid);
 954         btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
 955
 956         write_extent_buffer(root->node, root->fs_info->fsid,
 957                             (unsigned long)btrfs_header_fsid(root->node),
 958                             BTRFS_FSID_SIZE);
 959         btrfs_mark_buffer_dirty(root->node);
 960         btrfs_tree_unlock(root->node);
 961         fs_info->log_root_tree = root;
 962         return 0;
 963 }
 964
 965 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
 966                                                struct btrfs_key *location)
 967 {
 968         struct btrfs_root *root;
 969         struct btrfs_fs_info *fs_info = tree_root->fs_info;
 970         struct btrfs_path *path;
 971         struct extent_buffer *l;
 972         u64 highest_inode;
 973         u64 generation;
 974         u32 blocksize;
 975         int ret = 0;
 976
 977         root = kzalloc(sizeof(*root), GFP_NOFS);
 978         if (!root)
 979                 return ERR_PTR(-ENOMEM);
 980         if (location->offset == (u64)-1) {
 981                 ret = find_and_setup_root(tree_root, fs_info,
 982                                           location->objectid, root);
 983                 if (ret) {
 984                         kfree(root);
 985                         return ERR_PTR(ret);
 986                 }
 987                 goto insert;
 988         }
 989
 990         __setup_root(tree_root->nodesize, tree_root->leafsize,
 991                      tree_root->sectorsize, tree_root->stripesize,
 992                      root, fs_info, location->objectid);
 993
 994         path = btrfs_alloc_path();
 995         BUG_ON(!path);
 996         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 997         if (ret != 0) {
 998                 if (ret > 0)
 999                         ret = -ENOENT;
1000                 goto out;
1001         }
1002         l = path->nodes[0];
1003         read_extent_buffer(l, &root->root_item,
1004                btrfs_item_ptr_offset(l, path->slots[0]),
1005                sizeof(root->root_item));
1006         memcpy(&root->root_key, location, sizeof(*location));
1007         ret = 0;
1008 out:
1009         btrfs_release_path(root, path);
1010         btrfs_free_path(path);
1011         if (ret) {
1012                 kfree(root);
1013                 return ERR_PTR(ret);
1014         }
1015         generation = btrfs_root_generation(&root->root_item);
1016         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1017         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1018                                      blocksize, generation);
1019         BUG_ON(!root->node);
1020 insert:
1021         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1022                 root->ref_cows = 1;
1023                 ret = btrfs_find_highest_inode(root, &highest_inode);
1024                 if (ret == 0) {
1025                         root->highest_inode = highest_inode;
1026                         root->last_inode_alloc = highest_inode;
1027                 }
1028         }
1029         return root;
1030 }
1031
1032 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1033                                         u64 root_objectid)
1034 {
1035         struct btrfs_root *root;
1036
1037         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1038                 return fs_info->tree_root;
1039         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1040                 return fs_info->extent_root;
1041
1042         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1043                                  (unsigned long)root_objectid);
1044         return root;
1045 }
1046
1047 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1048                                               struct btrfs_key *location)
1049 {
1050         struct btrfs_root *root;
1051         int ret;
1052
1053         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1054                 return fs_info->tree_root;
1055         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1056                 return fs_info->extent_root;
1057         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1058                 return fs_info->chunk_root;
1059         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1060                 return fs_info->dev_root;
1061
1062         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1063                                  (unsigned long)location->objectid);
1064         if (root)
1065                 return root;
1066
1067         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1068         if (IS_ERR(root))
1069                 return root;
1070         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1071                                 (unsigned long)root->root_key.objectid,
1072                                 root);
1073         if (ret) {
1074                 free_extent_buffer(root->node);
1075                 kfree(root);
1076                 return ERR_PTR(ret);
1077         }
1078         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
1079                 ret = btrfs_find_dead_roots(fs_info->tree_root,
1080                                             root->root_key.objectid, root);
1081                 BUG_ON(ret);
1082                 btrfs_orphan_cleanup(root);
1083         }
1084         return root;
1085 }
1086
1087 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1088                                       struct btrfs_key *location,
1089                                       const char *name, int namelen)
1090 {
1091         struct btrfs_root *root;
1092         int ret;
1093
1094         root = btrfs_read_fs_root_no_name(fs_info, location);
1095         if (!root)
1096                 return NULL;
1097
1098         if (root->in_sysfs)
1099                 return root;
1100
1101         ret = btrfs_set_root_name(root, name, namelen);
1102         if (ret) {
1103                 free_extent_buffer(root->node);
1104                 kfree(root);
1105                 return ERR_PTR(ret);
1106         }
1107
1108         ret = btrfs_sysfs_add_root(root);
1109         if (ret) {
1110                 free_extent_buffer(root->node);
1111                 kfree(root->name);
1112                 kfree(root);
1113                 return ERR_PTR(ret);
1114         }
1115         root->in_sysfs = 1;
1116         return root;
1117 }
1118 #if 0
1119 static int add_hasher(struct btrfs_fs_info *info, char *type) {
1120         struct btrfs_hasher *hasher;
1121
1122         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
1123         if (!hasher)
1124                 return -ENOMEM;
1125         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
1126         if (!hasher->hash_tfm) {
1127                 kfree(hasher);
1128                 return -EINVAL;
1129         }
1130         spin_lock(&info->hash_lock);
1131         list_add(&hasher->list, &info->hashers);
1132         spin_unlock(&info->hash_lock);
1133         return 0;
1134 }
1135 #endif
1136
1137 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1138 {
1139         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1140         int ret = 0;
1141         struct list_head *cur;
1142         struct btrfs_device *device;
1143         struct backing_dev_info *bdi;
1144 #if 0
1145         if ((bdi_bits & (1 << BDI_write_congested)) &&
1146             btrfs_congested_async(info, 0))
1147                 return 1;
1148 #endif
1149         list_for_each(cur, &info->fs_devices->devices) {
1150                 device = list_entry(cur, struct btrfs_device, dev_list);
1151                 if (!device->bdev)
1152                         continue;
1153                 bdi = blk_get_backing_dev_info(device->bdev);
1154                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1155                         ret = 1;
1156                         break;
1157                 }
1158         }
1159         return ret;
1160 }
1161
1162 /*
1163  * this unplugs every device on the box, and it is only used when page
1164  * is null
1165  */
1166 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1167 {
1168         struct list_head *cur;
1169         struct btrfs_device *device;
1170         struct btrfs_fs_info *info;
1171
1172         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1173         list_for_each(cur, &info->fs_devices->devices) {
1174                 device = list_entry(cur, struct btrfs_device, dev_list);
1175                 bdi = blk_get_backing_dev_info(device->bdev);
1176                 if (bdi->unplug_io_fn) {
1177                         bdi->unplug_io_fn(bdi, page);
1178                 }
1179         }
1180 }
1181
1182 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1183 {
1184         struct inode *inode;
1185         struct extent_map_tree *em_tree;
1186         struct extent_map *em;
1187         struct address_space *mapping;
1188         u64 offset;
1189
1190         /* the generic O_DIRECT read code does this */
1191         if (!page) {
1192                 __unplug_io_fn(bdi, page);
1193                 return;
1194         }
1195
1196         /*
1197          * page->mapping may change at any time.  Get a consistent copy
1198          * and use that for everything below
1199          */
1200         smp_mb();
1201         mapping = page->mapping;
1202         if (!mapping)
1203                 return;
1204
1205         inode = mapping->host;
1206
1207         /*
1208          * don't do the expensive searching for a small number of
1209          * devices
1210          */
1211         if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1212                 __unplug_io_fn(bdi, page);
1213                 return;
1214         }
1215
1216         offset = page_offset(page);
1217
1218         em_tree = &BTRFS_I(inode)->extent_tree;
1219         spin_lock(&em_tree->lock);
1220         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1221         spin_unlock(&em_tree->lock);
1222         if (!em) {
1223                 __unplug_io_fn(bdi, page);
1224                 return;
1225         }
1226
1227         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1228                 free_extent_map(em);
1229                 __unplug_io_fn(bdi, page);
1230                 return;
1231         }
1232         offset = offset - em->start;
1233         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1234                           em->block_start + offset, page);
1235         free_extent_map(em);
1236 }
1237
1238 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1239 {
1240         bdi_init(bdi);
1241         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1242         bdi->state              = 0;
1243         bdi->capabilities       = default_backing_dev_info.capabilities;
1244         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1245         bdi->unplug_io_data     = info;
1246         bdi->congested_fn       = btrfs_congested_fn;
1247         bdi->congested_data     = info;
1248         return 0;
1249 }
1250
1251 static int bio_ready_for_csum(struct bio *bio)
1252 {
1253         u64 length = 0;
1254         u64 buf_len = 0;
1255         u64 start = 0;
1256         struct page *page;
1257         struct extent_io_tree *io_tree = NULL;
1258         struct btrfs_fs_info *info = NULL;
1259         struct bio_vec *bvec;
1260         int i;
1261         int ret;
1262
1263         bio_for_each_segment(bvec, bio, i) {
1264                 page = bvec->bv_page;
1265                 if (page->private == EXTENT_PAGE_PRIVATE) {
1266                         length += bvec->bv_len;
1267                         continue;
1268                 }
1269                 if (!page->private) {
1270                         length += bvec->bv_len;
1271                         continue;
1272                 }
1273                 length = bvec->bv_len;
1274                 buf_len = page->private >> 2;
1275                 start = page_offset(page) + bvec->bv_offset;
1276                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1277                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1278         }
1279         /* are we fully contained in this bio? */
1280         if (buf_len <= length)
1281                 return 1;
1282
1283         ret = extent_range_uptodate(io_tree, start + length,
1284                                     start + buf_len - 1);
1285         if (ret == 1)
1286                 return ret;
1287         return ret;
1288 }
1289
1290 /*
1291  * called by the kthread helper functions to finally call the bio end_io
1292  * functions.  This is where read checksum verification actually happens
1293  */
1294 static void end_workqueue_fn(struct btrfs_work *work)
1295 {
1296         struct bio *bio;
1297         struct end_io_wq *end_io_wq;
1298         struct btrfs_fs_info *fs_info;
1299         int error;
1300
1301         end_io_wq = container_of(work, struct end_io_wq, work);
1302         bio = end_io_wq->bio;
1303         fs_info = end_io_wq->info;
1304
1305         /* metadata bios are special because the whole tree block must
1306          * be checksummed at once.  This makes sure the entire block is in
1307          * ram and up to date before trying to verify things.  For
1308          * blocksize <= pagesize, it is basically a noop
1309          */
1310         if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1311                 btrfs_queue_worker(&fs_info->endio_workers,
1312                                    &end_io_wq->work);
1313                 return;
1314         }
1315         error = end_io_wq->error;
1316         bio->bi_private = end_io_wq->private;
1317         bio->bi_end_io = end_io_wq->end_io;
1318         kfree(end_io_wq);
1319         bio_endio(bio, error);
1320 }
1321
1322 static int cleaner_kthread(void *arg)
1323 {
1324         struct btrfs_root *root = arg;
1325
1326         do {
1327                 smp_mb();
1328                 if (root->fs_info->closing)
1329                         break;
1330
1331                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1332                 mutex_lock(&root->fs_info->cleaner_mutex);
1333                 btrfs_clean_old_snapshots(root);
1334                 mutex_unlock(&root->fs_info->cleaner_mutex);
1335
1336                 if (freezing(current)) {
1337                         refrigerator();
1338                 } else {
1339                         smp_mb();
1340                         if (root->fs_info->closing)
1341                                 break;
1342                         set_current_state(TASK_INTERRUPTIBLE);
1343                         schedule();
1344                         __set_current_state(TASK_RUNNING);
1345                 }
1346         } while (!kthread_should_stop());
1347         return 0;
1348 }
1349
1350 static int transaction_kthread(void *arg)
1351 {
1352         struct btrfs_root *root = arg;
1353         struct btrfs_trans_handle *trans;
1354         struct btrfs_transaction *cur;
1355         unsigned long now;
1356         unsigned long delay;
1357         int ret;
1358
1359         do {
1360                 smp_mb();
1361                 if (root->fs_info->closing)
1362                         break;
1363
1364                 delay = HZ * 30;
1365                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1366                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1367
1368                 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1369                         printk("btrfs: total reference cache size %Lu\n",
1370                                 root->fs_info->total_ref_cache_size);
1371                 }
1372
1373                 mutex_lock(&root->fs_info->trans_mutex);
1374                 cur = root->fs_info->running_transaction;
1375                 if (!cur) {
1376                         mutex_unlock(&root->fs_info->trans_mutex);
1377                         goto sleep;
1378                 }
1379
1380                 now = get_seconds();
1381                 if (now < cur->start_time || now - cur->start_time < 30) {
1382                         mutex_unlock(&root->fs_info->trans_mutex);
1383                         delay = HZ * 5;
1384                         goto sleep;
1385                 }
1386                 mutex_unlock(&root->fs_info->trans_mutex);
1387                 trans = btrfs_start_transaction(root, 1);
1388                 ret = btrfs_commit_transaction(trans, root);
1389 sleep:
1390                 wake_up_process(root->fs_info->cleaner_kthread);
1391                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1392
1393                 if (freezing(current)) {
1394                         refrigerator();
1395                 } else {
1396                         if (root->fs_info->closing)
1397                                 break;
1398                         set_current_state(TASK_INTERRUPTIBLE);
1399                         schedule_timeout(delay);
1400                         __set_current_state(TASK_RUNNING);
1401                 }
1402         } while (!kthread_should_stop());
1403         return 0;
1404 }
1405
1406 struct btrfs_root *open_ctree(struct super_block *sb,
1407                               struct btrfs_fs_devices *fs_devices,
1408                               char *options)
1409 {
1410         u32 sectorsize;
1411         u32 nodesize;
1412         u32 leafsize;
1413         u32 blocksize;
1414         u32 stripesize;
1415         u64 generation;
1416         struct buffer_head *bh;
1417         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1418                                                  GFP_NOFS);
1419         struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1420                                                GFP_NOFS);
1421         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1422                                                 GFP_NOFS);
1423         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1424                                                 GFP_NOFS);
1425         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1426                                               GFP_NOFS);
1427         struct btrfs_root *log_tree_root;
1428
1429         int ret;
1430         int err = -EINVAL;
1431
1432         struct btrfs_super_block *disk_super;
1433
1434         if (!extent_root || !tree_root || !fs_info ||
1435             !chunk_root || !dev_root) {
1436                 err = -ENOMEM;
1437                 goto fail;
1438         }
1439         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1440         INIT_LIST_HEAD(&fs_info->trans_list);
1441         INIT_LIST_HEAD(&fs_info->dead_roots);
1442         INIT_LIST_HEAD(&fs_info->hashers);
1443         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1444         spin_lock_init(&fs_info->hash_lock);
1445         spin_lock_init(&fs_info->delalloc_lock);
1446         spin_lock_init(&fs_info->new_trans_lock);
1447         spin_lock_init(&fs_info->ref_cache_lock);
1448
1449         init_completion(&fs_info->kobj_unregister);
1450         fs_info->tree_root = tree_root;
1451         fs_info->extent_root = extent_root;
1452         fs_info->chunk_root = chunk_root;
1453         fs_info->dev_root = dev_root;
1454         fs_info->fs_devices = fs_devices;
1455         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1456         INIT_LIST_HEAD(&fs_info->space_info);
1457         btrfs_mapping_init(&fs_info->mapping_tree);
1458         atomic_set(&fs_info->nr_async_submits, 0);
1459         atomic_set(&fs_info->async_delalloc_pages, 0);
1460         atomic_set(&fs_info->async_submit_draining, 0);
1461         atomic_set(&fs_info->nr_async_bios, 0);
1462         atomic_set(&fs_info->throttles, 0);
1463         atomic_set(&fs_info->throttle_gen, 0);
1464         fs_info->sb = sb;
1465         fs_info->max_extent = (u64)-1;
1466         fs_info->max_inline = 8192 * 1024;
1467         setup_bdi(fs_info, &fs_info->bdi);
1468         fs_info->btree_inode = new_inode(sb);
1469         fs_info->btree_inode->i_ino = 1;
1470         fs_info->btree_inode->i_nlink = 1;
1471
1472         fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1473
1474         INIT_LIST_HEAD(&fs_info->ordered_extents);
1475         spin_lock_init(&fs_info->ordered_extent_lock);
1476
1477         sb->s_blocksize = 4096;
1478         sb->s_blocksize_bits = blksize_bits(4096);
1479
1480         /*
1481          * we set the i_size on the btree inode to the max possible int.
1482          * the real end of the address space is determined by all of
1483          * the devices in the system
1484          */
1485         fs_info->btree_inode->i_size = OFFSET_MAX;
1486         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1487         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1488
1489         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1490                              fs_info->btree_inode->i_mapping,
1491                              GFP_NOFS);
1492         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1493                              GFP_NOFS);
1494
1495         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1496
1497         spin_lock_init(&fs_info->block_group_cache_lock);
1498         fs_info->block_group_cache_tree.rb_node = NULL;
1499
1500         extent_io_tree_init(&fs_info->pinned_extents,
1501                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1502         extent_io_tree_init(&fs_info->pending_del,
1503                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1504         extent_io_tree_init(&fs_info->extent_ins,
1505                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1506         fs_info->do_barriers = 1;
1507
1508         INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
1509         btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
1510         btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
1511
1512         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1513         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1514                sizeof(struct btrfs_key));
1515         insert_inode_hash(fs_info->btree_inode);
1516
1517         mutex_init(&fs_info->trans_mutex);
1518         mutex_init(&fs_info->tree_log_mutex);
1519         mutex_init(&fs_info->drop_mutex);
1520         mutex_init(&fs_info->extent_ins_mutex);
1521         mutex_init(&fs_info->pinned_mutex);
1522         mutex_init(&fs_info->chunk_mutex);
1523         mutex_init(&fs_info->transaction_kthread_mutex);
1524         mutex_init(&fs_info->cleaner_mutex);
1525         mutex_init(&fs_info->volume_mutex);
1526         mutex_init(&fs_info->tree_reloc_mutex);
1527         init_waitqueue_head(&fs_info->transaction_throttle);
1528         init_waitqueue_head(&fs_info->transaction_wait);
1529         init_waitqueue_head(&fs_info->async_submit_wait);
1530         init_waitqueue_head(&fs_info->tree_log_wait);
1531         atomic_set(&fs_info->tree_log_commit, 0);
1532         atomic_set(&fs_info->tree_log_writers, 0);
1533         fs_info->tree_log_transid = 0;
1534
1535 #if 0
1536         ret = add_hasher(fs_info, "crc32c");
1537         if (ret) {
1538                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1539                 err = -ENOMEM;
1540                 goto fail_iput;
1541         }
1542 #endif
1543         __setup_root(4096, 4096, 4096, 4096, tree_root,
1544                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1545
1546
1547         bh = __bread(fs_devices->latest_bdev,
1548                      BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1549         if (!bh)
1550                 goto fail_iput;
1551
1552         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1553         brelse(bh);
1554
1555         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1556
1557         disk_super = &fs_info->super_copy;
1558         if (!btrfs_super_root(disk_super))
1559                 goto fail_sb_buffer;
1560
1561         err = btrfs_parse_options(tree_root, options);
1562         if (err)
1563                 goto fail_sb_buffer;
1564
1565         /*
1566          * we need to start all the end_io workers up front because the
1567          * queue work function gets called at interrupt time, and so it
1568          * cannot dynamically grow.
1569          */
1570         btrfs_init_workers(&fs_info->workers, "worker",
1571                            fs_info->thread_pool_size);
1572
1573         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1574                            fs_info->thread_pool_size);
1575
1576         btrfs_init_workers(&fs_info->submit_workers, "submit",
1577                            min_t(u64, fs_devices->num_devices,
1578                            fs_info->thread_pool_size));
1579
1580         /* a higher idle thresh on the submit workers makes it much more
1581          * likely that bios will be send down in a sane order to the
1582          * devices
1583          */
1584         fs_info->submit_workers.idle_thresh = 64;
1585
1586         fs_info->workers.idle_thresh = 16;
1587         fs_info->workers.ordered = 1;
1588
1589         fs_info->delalloc_workers.idle_thresh = 2;
1590         fs_info->delalloc_workers.ordered = 1;
1591
1592         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1593         btrfs_init_workers(&fs_info->endio_workers, "endio",
1594                            fs_info->thread_pool_size);
1595         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1596                            fs_info->thread_pool_size);
1597
1598         /*
1599          * endios are largely parallel and should have a very
1600          * low idle thresh
1601          */
1602         fs_info->endio_workers.idle_thresh = 4;
1603         fs_info->endio_write_workers.idle_thresh = 64;
1604
1605         btrfs_start_workers(&fs_info->workers, 1);
1606         btrfs_start_workers(&fs_info->submit_workers, 1);
1607         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1608         btrfs_start_workers(&fs_info->fixup_workers, 1);
1609         btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1610         btrfs_start_workers(&fs_info->endio_write_workers,
1611                             fs_info->thread_pool_size);
1612
1613         err = -EINVAL;
1614         if (btrfs_super_num_devices(disk_super) > fs_devices->open_devices) {
1615                 printk("Btrfs: wanted %llu devices, but found %llu\n",
1616                        (unsigned long long)btrfs_super_num_devices(disk_super),
1617                        (unsigned long long)fs_devices->open_devices);
1618                 if (btrfs_test_opt(tree_root, DEGRADED))
1619                         printk("continuing in degraded mode\n");
1620                 else {
1621                         goto fail_sb_buffer;
1622                 }
1623         }
1624
1625         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1626         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1627                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1628
1629         nodesize = btrfs_super_nodesize(disk_super);
1630         leafsize = btrfs_super_leafsize(disk_super);
1631         sectorsize = btrfs_super_sectorsize(disk_super);
1632         stripesize = btrfs_super_stripesize(disk_super);
1633         tree_root->nodesize = nodesize;
1634         tree_root->leafsize = leafsize;
1635         tree_root->sectorsize = sectorsize;
1636         tree_root->stripesize = stripesize;
1637
1638         sb->s_blocksize = sectorsize;
1639         sb->s_blocksize_bits = blksize_bits(sectorsize);
1640
1641         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1642                     sizeof(disk_super->magic))) {
1643                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1644                 goto fail_sb_buffer;
1645         }
1646
1647         mutex_lock(&fs_info->chunk_mutex);
1648         ret = btrfs_read_sys_array(tree_root);
1649         mutex_unlock(&fs_info->chunk_mutex);
1650         if (ret) {
1651                 printk("btrfs: failed to read the system array on %s\n",
1652                        sb->s_id);
1653                 goto fail_sys_array;
1654         }
1655
1656         blocksize = btrfs_level_size(tree_root,
1657                                      btrfs_super_chunk_root_level(disk_super));
1658         generation = btrfs_super_chunk_root_generation(disk_super);
1659
1660         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1661                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1662
1663         chunk_root->node = read_tree_block(chunk_root,
1664                                            btrfs_super_chunk_root(disk_super),
1665                                            blocksize, generation);
1666         BUG_ON(!chunk_root->node);
1667
1668         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1669                  (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1670                  BTRFS_UUID_SIZE);
1671
1672         mutex_lock(&fs_info->chunk_mutex);
1673         ret = btrfs_read_chunk_tree(chunk_root);
1674         mutex_unlock(&fs_info->chunk_mutex);
1675         BUG_ON(ret);
1676
1677         btrfs_close_extra_devices(fs_devices);
1678
1679         blocksize = btrfs_level_size(tree_root,
1680                                      btrfs_super_root_level(disk_super));
1681         generation = btrfs_super_generation(disk_super);
1682
1683         tree_root->node = read_tree_block(tree_root,
1684                                           btrfs_super_root(disk_super),
1685                                           blocksize, generation);
1686         if (!tree_root->node)
1687                 goto fail_sb_buffer;
1688
1689
1690         ret = find_and_setup_root(tree_root, fs_info,
1691                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1692         if (ret)
1693                 goto fail_tree_root;
1694         extent_root->track_dirty = 1;
1695
1696         ret = find_and_setup_root(tree_root, fs_info,
1697                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1698         dev_root->track_dirty = 1;
1699
1700         if (ret)
1701                 goto fail_extent_root;
1702
1703         btrfs_read_block_groups(extent_root);
1704
1705         fs_info->generation = generation + 1;
1706         fs_info->last_trans_committed = generation;
1707         fs_info->data_alloc_profile = (u64)-1;
1708         fs_info->metadata_alloc_profile = (u64)-1;
1709         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1710         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1711                                                "btrfs-cleaner");
1712         if (!fs_info->cleaner_kthread)
1713                 goto fail_extent_root;
1714
1715         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1716                                                    tree_root,
1717                                                    "btrfs-transaction");
1718         if (!fs_info->transaction_kthread)
1719                 goto fail_cleaner;
1720
1721         if (sb->s_flags & MS_RDONLY)
1722                 return tree_root;
1723
1724         if (btrfs_super_log_root(disk_super) != 0) {
1725                 u32 blocksize;
1726                 u64 bytenr = btrfs_super_log_root(disk_super);
1727
1728                 blocksize =
1729                      btrfs_level_size(tree_root,
1730                                       btrfs_super_log_root_level(disk_super));
1731
1732                 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1733                                                       GFP_NOFS);
1734
1735                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1736                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1737
1738                 log_tree_root->node = read_tree_block(tree_root, bytenr,
1739                                                       blocksize,
1740                                                       generation + 1);
1741                 ret = btrfs_recover_log_trees(log_tree_root);
1742                 BUG_ON(ret);
1743         }
1744
1745         ret = btrfs_cleanup_reloc_trees(tree_root);
1746         BUG_ON(ret);
1747
1748         return tree_root;
1749
1750 fail_cleaner:
1751         kthread_stop(fs_info->cleaner_kthread);
1752 fail_extent_root:
1753         free_extent_buffer(extent_root->node);
1754 fail_tree_root:
1755         free_extent_buffer(tree_root->node);
1756 fail_sys_array:
1757 fail_sb_buffer:
1758         btrfs_stop_workers(&fs_info->fixup_workers);
1759         btrfs_stop_workers(&fs_info->delalloc_workers);
1760         btrfs_stop_workers(&fs_info->workers);
1761         btrfs_stop_workers(&fs_info->endio_workers);
1762         btrfs_stop_workers(&fs_info->endio_write_workers);
1763         btrfs_stop_workers(&fs_info->submit_workers);
1764 fail_iput:
1765         iput(fs_info->btree_inode);
1766 fail:
1767         btrfs_close_devices(fs_info->fs_devices);
1768         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1769
1770         kfree(extent_root);
1771         kfree(tree_root);
1772         bdi_destroy(&fs_info->bdi);
1773         kfree(fs_info);
1774         kfree(chunk_root);
1775         kfree(dev_root);
1776         return ERR_PTR(err);
1777 }
1778
1779 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1780 {
1781         char b[BDEVNAME_SIZE];
1782
1783         if (uptodate) {
1784                 set_buffer_uptodate(bh);
1785         } else {
1786                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1787                         printk(KERN_WARNING "lost page write due to "
1788                                         "I/O error on %s\n",
1789                                        bdevname(bh->b_bdev, b));
1790                 }
1791                 /* note, we dont' set_buffer_write_io_error because we have
1792                  * our own ways of dealing with the IO errors
1793                  */
1794                 clear_buffer_uptodate(bh);
1795         }
1796         unlock_buffer(bh);
1797         put_bh(bh);
1798 }
1799
1800 int write_all_supers(struct btrfs_root *root)
1801 {
1802         struct list_head *cur;
1803         struct list_head *head = &root->fs_info->fs_devices->devices;
1804         struct btrfs_device *dev;
1805         struct btrfs_super_block *sb;
1806         struct btrfs_dev_item *dev_item;
1807         struct buffer_head *bh;
1808         int ret;
1809         int do_barriers;
1810         int max_errors;
1811         int total_errors = 0;
1812         u32 crc;
1813         u64 flags;
1814
1815         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1816         do_barriers = !btrfs_test_opt(root, NOBARRIER);
1817
1818         sb = &root->fs_info->super_for_commit;
1819         dev_item = &sb->dev_item;
1820         list_for_each(cur, head) {
1821                 dev = list_entry(cur, struct btrfs_device, dev_list);
1822                 if (!dev->bdev) {
1823                         total_errors++;
1824                         continue;
1825                 }
1826                 if (!dev->in_fs_metadata)
1827                         continue;
1828
1829                 btrfs_set_stack_device_type(dev_item, dev->type);
1830                 btrfs_set_stack_device_id(dev_item, dev->devid);
1831                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1832                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1833                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1834                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1835                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1836                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1837                 flags = btrfs_super_flags(sb);
1838                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1839
1840
1841                 crc = ~(u32)0;
1842                 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1843                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1844                 btrfs_csum_final(crc, sb->csum);
1845
1846                 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1847                               BTRFS_SUPER_INFO_SIZE);
1848
1849                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1850                 dev->pending_io = bh;
1851
1852                 get_bh(bh);
1853                 set_buffer_uptodate(bh);
1854                 lock_buffer(bh);
1855                 bh->b_end_io = btrfs_end_buffer_write_sync;
1856
1857                 if (do_barriers && dev->barriers) {
1858                         ret = submit_bh(WRITE_BARRIER, bh);
1859                         if (ret == -EOPNOTSUPP) {
1860                                 printk("btrfs: disabling barriers on dev %s\n",
1861                                        dev->name);
1862                                 set_buffer_uptodate(bh);
1863                                 dev->barriers = 0;
1864                                 get_bh(bh);
1865                                 lock_buffer(bh);
1866                                 ret = submit_bh(WRITE, bh);
1867                         }
1868                 } else {
1869                         ret = submit_bh(WRITE, bh);
1870                 }
1871                 if (ret)
1872                         total_errors++;
1873         }
1874         if (total_errors > max_errors) {
1875                 printk("btrfs: %d errors while writing supers\n", total_errors);
1876                 BUG();
1877         }
1878         total_errors = 0;
1879
1880         list_for_each(cur, head) {
1881                 dev = list_entry(cur, struct btrfs_device, dev_list);
1882                 if (!dev->bdev)
1883                         continue;
1884                 if (!dev->in_fs_metadata)
1885                         continue;
1886
1887                 BUG_ON(!dev->pending_io);
1888                 bh = dev->pending_io;
1889                 wait_on_buffer(bh);
1890                 if (!buffer_uptodate(dev->pending_io)) {
1891                         if (do_barriers && dev->barriers) {
1892                                 printk("btrfs: disabling barriers on dev %s\n",
1893                                        dev->name);
1894                                 set_buffer_uptodate(bh);
1895                                 get_bh(bh);
1896                                 lock_buffer(bh);
1897                                 dev->barriers = 0;
1898                                 ret = submit_bh(WRITE, bh);
1899                                 BUG_ON(ret);
1900                                 wait_on_buffer(bh);
1901                                 if (!buffer_uptodate(bh))
1902                                         total_errors++;
1903                         } else {
1904                                 total_errors++;
1905                         }
1906
1907                 }
1908                 dev->pending_io = NULL;
1909                 brelse(bh);
1910         }
1911         if (total_errors > max_errors) {
1912                 printk("btrfs: %d errors while writing supers\n", total_errors);
1913                 BUG();
1914         }
1915         return 0;
1916 }
1917
1918 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1919                       *root)
1920 {
1921         int ret;
1922
1923         ret = write_all_supers(root);
1924         return ret;
1925 }
1926
1927 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1928 {
1929         radix_tree_delete(&fs_info->fs_roots_radix,
1930                           (unsigned long)root->root_key.objectid);
1931         if (root->in_sysfs)
1932                 btrfs_sysfs_del_root(root);
1933         if (root->inode)
1934                 iput(root->inode);
1935         if (root->node)
1936                 free_extent_buffer(root->node);
1937         if (root->commit_root)
1938                 free_extent_buffer(root->commit_root);
1939         if (root->name)
1940                 kfree(root->name);
1941         kfree(root);
1942         return 0;
1943 }
1944
1945 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1946 {
1947         int ret;
1948         struct btrfs_root *gang[8];
1949         int i;
1950
1951         while(1) {
1952                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1953                                              (void **)gang, 0,
1954                                              ARRAY_SIZE(gang));
1955                 if (!ret)
1956                         break;
1957                 for (i = 0; i < ret; i++)
1958                         btrfs_free_fs_root(fs_info, gang[i]);
1959         }
1960         return 0;
1961 }
1962
1963 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
1964 {
1965         u64 root_objectid = 0;
1966         struct btrfs_root *gang[8];
1967         int i;
1968         int ret;
1969
1970         while (1) {
1971                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1972                                              (void **)gang, root_objectid,
1973                                              ARRAY_SIZE(gang));
1974                 if (!ret)
1975                         break;
1976                 for (i = 0; i < ret; i++) {
1977                         root_objectid = gang[i]->root_key.objectid;
1978                         ret = btrfs_find_dead_roots(fs_info->tree_root,
1979                                                     root_objectid, gang[i]);
1980                         BUG_ON(ret);
1981                         btrfs_orphan_cleanup(gang[i]);
1982                 }
1983                 root_objectid++;
1984         }
1985         return 0;
1986 }
1987
1988 int btrfs_commit_super(struct btrfs_root *root)
1989 {
1990         struct btrfs_trans_handle *trans;
1991         int ret;
1992
1993         mutex_lock(&root->fs_info->cleaner_mutex);
1994         btrfs_clean_old_snapshots(root);
1995         mutex_unlock(&root->fs_info->cleaner_mutex);
1996         trans = btrfs_start_transaction(root, 1);
1997         ret = btrfs_commit_transaction(trans, root);
1998         BUG_ON(ret);
1999         /* run commit again to drop the original snapshot */
2000         trans = btrfs_start_transaction(root, 1);
2001         btrfs_commit_transaction(trans, root);
2002         ret = btrfs_write_and_wait_transaction(NULL, root);
2003         BUG_ON(ret);
2004
2005         ret = write_ctree_super(NULL, root);
2006         return ret;
2007 }
2008
2009 int close_ctree(struct btrfs_root *root)
2010 {
2011         struct btrfs_fs_info *fs_info = root->fs_info;
2012         int ret;
2013
2014         fs_info->closing = 1;
2015         smp_mb();
2016
2017         kthread_stop(root->fs_info->transaction_kthread);
2018         kthread_stop(root->fs_info->cleaner_kthread);
2019
2020         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2021                 ret =  btrfs_commit_super(root);
2022                 if (ret) {
2023                         printk("btrfs: commit super returns %d\n", ret);
2024                 }
2025         }
2026
2027         if (fs_info->delalloc_bytes) {
2028                 printk("btrfs: at unmount delalloc count %Lu\n",
2029                        fs_info->delalloc_bytes);
2030         }
2031         if (fs_info->total_ref_cache_size) {
2032                 printk("btrfs: at umount reference cache size %Lu\n",
2033                         fs_info->total_ref_cache_size);
2034         }
2035
2036         if (fs_info->extent_root->node)
2037                 free_extent_buffer(fs_info->extent_root->node);
2038
2039         if (fs_info->tree_root->node)
2040                 free_extent_buffer(fs_info->tree_root->node);
2041
2042         if (root->fs_info->chunk_root->node);
2043                 free_extent_buffer(root->fs_info->chunk_root->node);
2044
2045         if (root->fs_info->dev_root->node);
2046                 free_extent_buffer(root->fs_info->dev_root->node);
2047
2048         btrfs_free_block_groups(root->fs_info);
2049
2050         del_fs_roots(fs_info);
2051
2052         iput(fs_info->btree_inode);
2053
2054         btrfs_stop_workers(&fs_info->fixup_workers);
2055         btrfs_stop_workers(&fs_info->delalloc_workers);
2056         btrfs_stop_workers(&fs_info->workers);
2057         btrfs_stop_workers(&fs_info->endio_workers);
2058         btrfs_stop_workers(&fs_info->endio_write_workers);
2059         btrfs_stop_workers(&fs_info->submit_workers);
2060
2061 #if 0
2062         while(!list_empty(&fs_info->hashers)) {
2063                 struct btrfs_hasher *hasher;
2064                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
2065                                     hashers);
2066                 list_del(&hasher->hashers);
2067                 crypto_free_hash(&fs_info->hash_tfm);
2068                 kfree(hasher);
2069         }
2070 #endif
2071         btrfs_close_devices(fs_info->fs_devices);
2072         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2073
2074         bdi_destroy(&fs_info->bdi);
2075
2076         kfree(fs_info->extent_root);
2077         kfree(fs_info->tree_root);
2078         kfree(fs_info->chunk_root);
2079         kfree(fs_info->dev_root);
2080         return 0;
2081 }
2082
2083 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2084 {
2085         int ret;
2086         struct inode *btree_inode = buf->first_page->mapping->host;
2087
2088         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2089         if (!ret)
2090                 return ret;
2091
2092         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2093                                     parent_transid);
2094         return !ret;
2095 }
2096
2097 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2098 {
2099         struct inode *btree_inode = buf->first_page->mapping->host;
2100         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2101                                           buf);
2102 }
2103
2104 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2105 {
2106         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2107         u64 transid = btrfs_header_generation(buf);
2108         struct inode *btree_inode = root->fs_info->btree_inode;
2109
2110         WARN_ON(!btrfs_tree_locked(buf));
2111         if (transid != root->fs_info->generation) {
2112                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
2113                         (unsigned long long)buf->start,
2114                         transid, root->fs_info->generation);
2115                 WARN_ON(1);
2116         }
2117         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
2118 }
2119
2120 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2121 {
2122         /*
2123          * looks as though older kernels can get into trouble with
2124          * this code, they end up stuck in balance_dirty_pages forever
2125          */
2126         struct extent_io_tree *tree;
2127         u64 num_dirty;
2128         u64 start = 0;
2129         unsigned long thresh = 32 * 1024 * 1024;
2130         tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
2131
2132         if (current_is_pdflush() || current->flags & PF_MEMALLOC)
2133                 return;
2134
2135         num_dirty = count_range_bits(tree, &start, (u64)-1,
2136                                      thresh, EXTENT_DIRTY);
2137         if (num_dirty > thresh) {
2138                 balance_dirty_pages_ratelimited_nr(
2139                                    root->fs_info->btree_inode->i_mapping, 1);
2140         }
2141         return;
2142 }
2143
2144 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2145 {
2146         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2147         int ret;
2148         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2149         if (ret == 0) {
2150                 buf->flags |= EXTENT_UPTODATE;
2151         }
2152         return ret;
2153 }
2154
2155 int btree_lock_page_hook(struct page *page)
2156 {
2157         struct inode *inode = page->mapping->host;
2158         struct btrfs_root *root = BTRFS_I(inode)->root;
2159         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2160         struct extent_buffer *eb;
2161         unsigned long len;
2162         u64 bytenr = page_offset(page);
2163
2164         if (page->private == EXTENT_PAGE_PRIVATE)
2165                 goto out;
2166
2167         len = page->private >> 2;
2168         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2169         if (!eb)
2170                 goto out;
2171
2172         btrfs_tree_lock(eb);
2173         spin_lock(&root->fs_info->hash_lock);
2174         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2175         spin_unlock(&root->fs_info->hash_lock);
2176         btrfs_tree_unlock(eb);
2177         free_extent_buffer(eb);
2178 out:
2179         lock_page(page);
2180         return 0;
2181 }
2182
2183 static struct extent_io_ops btree_extent_io_ops = {
2184         .write_cache_pages_lock_hook = btree_lock_page_hook,
2185         .readpage_end_io_hook = btree_readpage_end_io_hook,
2186         .submit_bio_hook = btree_submit_bio_hook,
2187         /* note we're sharing with inode.c for the merge bio hook */
2188         .merge_bio_hook = btrfs_merge_bio_hook,
2189 };