SAFE public projects git trees. - safe/jmp/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 "compat.h"
  31 #include "crc32c.h"
  32 #include "ctree.h"
  33 #include "disk-io.h"
  34 #include "transaction.h"
  35 #include "btrfs_inode.h"
  36 #include "volumes.h"
  37 #include "print-tree.h"
  38 #include "async-thread.h"
  39 #include "locking.h"
  40 #include "ref-cache.h"
  41 #include "tree-log.h"
  42
  43 #if 0
  44 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
  45 {
  46         if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
  47                 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
  48                        (unsigned long long)extent_buffer_blocknr(buf),
  49                        (unsigned long long)btrfs_header_blocknr(buf));
  50                 return 1;
  51         }
  52         return 0;
  53 }
  54 #endif
  55
  56 static struct extent_io_ops btree_extent_io_ops;
  57 static void end_workqueue_fn(struct btrfs_work *work);
  58
  59 /*
  60  * end_io_wq structs are used to do processing in task context when an IO is
  61  * complete.  This is used during reads to verify checksums, and it is used
  62  * by writes to insert metadata for new file extents after IO is complete.
  63  */
  64 struct end_io_wq {
  65         struct bio *bio;
  66         bio_end_io_t *end_io;
  67         void *private;
  68         struct btrfs_fs_info *info;
  69         int error;
  70         int metadata;
  71         struct list_head list;
  72         struct btrfs_work work;
  73 };
  74
  75 /*
  76  * async submit bios are used to offload expensive checksumming
  77  * onto the worker threads.  They checksum file and metadata bios
  78  * just before they are sent down the IO stack.
  79  */
  80 struct async_submit_bio {
  81         struct inode *inode;
  82         struct bio *bio;
  83         struct list_head list;
  84         extent_submit_bio_hook_t *submit_bio_start;
  85         extent_submit_bio_hook_t *submit_bio_done;
  86         int rw;
  87         int mirror_num;
  88         unsigned long bio_flags;
  89         struct btrfs_work work;
  90 };
  91
  92 /*
  93  * extents on the btree inode are pretty simple, there's one extent
  94  * that covers the entire device
  95  */
  96 static struct extent_map *btree_get_extent(struct inode *inode,
  97                 struct page *page, size_t page_offset, u64 start, u64 len,
  98                 int create)
  99 {
 100         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 101         struct extent_map *em;
 102         int ret;
 103
 104         spin_lock(&em_tree->lock);
 105         em = lookup_extent_mapping(em_tree, start, len);
 106         if (em) {
 107                 em->bdev =
 108                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 109                 spin_unlock(&em_tree->lock);
 110                 goto out;
 111         }
 112         spin_unlock(&em_tree->lock);
 113
 114         em = alloc_extent_map(GFP_NOFS);
 115         if (!em) {
 116                 em = ERR_PTR(-ENOMEM);
 117                 goto out;
 118         }
 119         em->start = 0;
 120         em->len = (u64)-1;
 121         em->block_len = (u64)-1;
 122         em->block_start = 0;
 123         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 124
 125         spin_lock(&em_tree->lock);
 126         ret = add_extent_mapping(em_tree, em);
 127         if (ret == -EEXIST) {
 128                 u64 failed_start = em->start;
 129                 u64 failed_len = em->len;
 130
 131                 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
 132                        em->start, em->len, em->block_start);
 133                 free_extent_map(em);
 134                 em = lookup_extent_mapping(em_tree, start, len);
 135                 if (em) {
 136                         printk("after failing, found %Lu %Lu %Lu\n",
 137                                em->start, em->len, em->block_start);
 138                         ret = 0;
 139                 } else {
 140                         em = lookup_extent_mapping(em_tree, failed_start,
 141                                                    failed_len);
 142                         if (em) {
 143                                 printk("double failure lookup gives us "
 144                                        "%Lu %Lu -> %Lu\n", em->start,
 145                                        em->len, em->block_start);
 146                                 free_extent_map(em);
 147                         }
 148                         ret = -EIO;
 149                 }
 150         } else if (ret) {
 151                 free_extent_map(em);
 152                 em = NULL;
 153         }
 154         spin_unlock(&em_tree->lock);
 155
 156         if (ret)
 157                 em = ERR_PTR(ret);
 158 out:
 159         return em;
 160 }
 161
 162 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 163 {
 164         return btrfs_crc32c(seed, data, len);
 165 }
 166
 167 void btrfs_csum_final(u32 crc, char *result)
 168 {
 169         *(__le32 *)result = ~cpu_to_le32(crc);
 170 }
 171
 172 /*
 173  * compute the csum for a btree block, and either verify it or write it
 174  * into the csum field of the block.
 175  */
 176 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 177                            int verify)
 178 {
 179         u16 csum_size =
 180                 btrfs_super_csum_size(&root->fs_info->super_copy);
 181         char *result = NULL;
 182         unsigned long len;
 183         unsigned long cur_len;
 184         unsigned long offset = BTRFS_CSUM_SIZE;
 185         char *map_token = NULL;
 186         char *kaddr;
 187         unsigned long map_start;
 188         unsigned long map_len;
 189         int err;
 190         u32 crc = ~(u32)0;
 191         unsigned long inline_result;
 192
 193         len = buf->len - offset;
 194         while(len > 0) {
 195                 err = map_private_extent_buffer(buf, offset, 32,
 196                                         &map_token, &kaddr,
 197                                         &map_start, &map_len, KM_USER0);
 198                 if (err) {
 199                         printk("failed to map extent buffer! %lu\n",
 200                                offset);
 201                         return 1;
 202                 }
 203                 cur_len = min(len, map_len - (offset - map_start));
 204                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 205                                       crc, cur_len);
 206                 len -= cur_len;
 207                 offset += cur_len;
 208                 unmap_extent_buffer(buf, map_token, KM_USER0);
 209         }
 210         if (csum_size > sizeof(inline_result)) {
 211                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
 212                 if (!result)
 213                         return 1;
 214         } else {
 215                 result = (char *)&inline_result;
 216         }
 217
 218         btrfs_csum_final(crc, result);
 219
 220         if (verify) {
 221                 /* FIXME, this is not good */
 222                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 223                         u32 val;
 224                         u32 found = 0;
 225                         memcpy(&found, result, csum_size);
 226
 227                         read_extent_buffer(buf, &val, 0, csum_size);
 228                         printk("btrfs: %s checksum verify failed on %llu "
 229                                "wanted %X found %X level %d\n",
 230                                root->fs_info->sb->s_id,
 231                                buf->start, val, found, btrfs_header_level(buf));
 232                         if (result != (char *)&inline_result)
 233                                 kfree(result);
 234                         return 1;
 235                 }
 236         } else {
 237                 write_extent_buffer(buf, result, 0, csum_size);
 238         }
 239         if (result != (char *)&inline_result)
 240                 kfree(result);
 241         return 0;
 242 }
 243
 244 /*
 245  * we can't consider a given block up to date unless the transid of the
 246  * block matches the transid in the parent node's pointer.  This is how we
 247  * detect blocks that either didn't get written at all or got written
 248  * in the wrong place.
 249  */
 250 static int verify_parent_transid(struct extent_io_tree *io_tree,
 251                                  struct extent_buffer *eb, u64 parent_transid)
 252 {
 253         int ret;
 254
 255         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 256                 return 0;
 257
 258         lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
 259         if (extent_buffer_uptodate(io_tree, eb) &&
 260             btrfs_header_generation(eb) == parent_transid) {
 261                 ret = 0;
 262                 goto out;
 263         }
 264         printk("parent transid verify failed on %llu wanted %llu found %llu\n",
 265                (unsigned long long)eb->start,
 266                (unsigned long long)parent_transid,
 267                (unsigned long long)btrfs_header_generation(eb));
 268         ret = 1;
 269         clear_extent_buffer_uptodate(io_tree, eb);
 270 out:
 271         unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
 272                       GFP_NOFS);
 273         return ret;
 274 }
 275
 276 /*
 277  * helper to read a given tree block, doing retries as required when
 278  * the checksums don't match and we have alternate mirrors to try.
 279  */
 280 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 281                                           struct extent_buffer *eb,
 282                                           u64 start, u64 parent_transid)
 283 {
 284         struct extent_io_tree *io_tree;
 285         int ret;
 286         int num_copies = 0;
 287         int mirror_num = 0;
 288
 289         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 290         while (1) {
 291                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 292                                                btree_get_extent, mirror_num);
 293                 if (!ret &&
 294                     !verify_parent_transid(io_tree, eb, parent_transid))
 295                         return ret;
 296 printk("read extent buffer pages failed with ret %d mirror no %d\n", ret, mirror_num);
 297                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 298                                               eb->start, eb->len);
 299                 if (num_copies == 1)
 300                         return ret;
 301
 302                 mirror_num++;
 303                 if (mirror_num > num_copies)
 304                         return ret;
 305         }
 306         return -EIO;
 307 }
 308
 309 /*
 310  * checksum a dirty tree block before IO.  This has extra checks to make
 311  * sure we only fill in the checksum field in the first page of a multi-page block
 312  */
 313 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 314 {
 315         struct extent_io_tree *tree;
 316         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 317         u64 found_start;
 318         int found_level;
 319         unsigned long len;
 320         struct extent_buffer *eb;
 321         int ret;
 322
 323         tree = &BTRFS_I(page->mapping->host)->io_tree;
 324
 325         if (page->private == EXTENT_PAGE_PRIVATE)
 326                 goto out;
 327         if (!page->private)
 328                 goto out;
 329         len = page->private >> 2;
 330         if (len == 0) {
 331                 WARN_ON(1);
 332         }
 333         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 334         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 335                                              btrfs_header_generation(eb));
 336         BUG_ON(ret);
 337         found_start = btrfs_header_bytenr(eb);
 338         if (found_start != start) {
 339                 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
 340                        start, found_start, len);
 341                 WARN_ON(1);
 342                 goto err;
 343         }
 344         if (eb->first_page != page) {
 345                 printk("bad first page %lu %lu\n", eb->first_page->index,
 346                        page->index);
 347                 WARN_ON(1);
 348                 goto err;
 349         }
 350         if (!PageUptodate(page)) {
 351                 printk("csum not up to date page %lu\n", page->index);
 352                 WARN_ON(1);
 353                 goto err;
 354         }
 355         found_level = btrfs_header_level(eb);
 356
 357         csum_tree_block(root, eb, 0);
 358 err:
 359         free_extent_buffer(eb);
 360 out:
 361         return 0;
 362 }
 363
 364 static int check_tree_block_fsid(struct btrfs_root *root,
 365                                  struct extent_buffer *eb)
 366 {
 367         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 368         u8 fsid[BTRFS_UUID_SIZE];
 369         int ret = 1;
 370
 371         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
 372                            BTRFS_FSID_SIZE);
 373         while (fs_devices) {
 374                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 375                         ret = 0;
 376                         break;
 377                 }
 378                 fs_devices = fs_devices->seed;
 379         }
 380         return ret;
 381 }
 382
 383 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 384                                struct extent_state *state)
 385 {
 386         struct extent_io_tree *tree;
 387         u64 found_start;
 388         int found_level;
 389         unsigned long len;
 390         struct extent_buffer *eb;
 391         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 392         int ret = 0;
 393
 394         tree = &BTRFS_I(page->mapping->host)->io_tree;
 395         if (page->private == EXTENT_PAGE_PRIVATE)
 396                 goto out;
 397         if (!page->private)
 398                 goto out;
 399         len = page->private >> 2;
 400         if (len == 0) {
 401                 WARN_ON(1);
 402         }
 403         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 404
 405         found_start = btrfs_header_bytenr(eb);
 406         if (found_start != start) {
 407                 printk("bad tree block start %llu %llu\n",
 408                        (unsigned long long)found_start,
 409                        (unsigned long long)eb->start);
 410                 ret = -EIO;
 411                 goto err;
 412         }
 413         if (eb->first_page != page) {
 414                 printk("bad first page %lu %lu\n", eb->first_page->index,
 415                        page->index);
 416                 WARN_ON(1);
 417                 ret = -EIO;
 418                 goto err;
 419         }
 420         if (check_tree_block_fsid(root, eb)) {
 421                 printk("bad fsid on block %Lu\n", eb->start);
 422                 ret = -EIO;
 423                 goto err;
 424         }
 425         found_level = btrfs_header_level(eb);
 426
 427         ret = csum_tree_block(root, eb, 1);
 428         if (ret)
 429                 ret = -EIO;
 430
 431         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 432         end = eb->start + end - 1;
 433 err:
 434         free_extent_buffer(eb);
 435 out:
 436         return ret;
 437 }
 438
 439 static void end_workqueue_bio(struct bio *bio, int err)
 440 {
 441         struct end_io_wq *end_io_wq = bio->bi_private;
 442         struct btrfs_fs_info *fs_info;
 443
 444         fs_info = end_io_wq->info;
 445         end_io_wq->error = err;
 446         end_io_wq->work.func = end_workqueue_fn;
 447         end_io_wq->work.flags = 0;
 448
 449         if (bio->bi_rw & (1 << BIO_RW)) {
 450                 if (end_io_wq->metadata)
 451                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
 452                                            &end_io_wq->work);
 453                 else
 454                         btrfs_queue_worker(&fs_info->endio_write_workers,
 455                                            &end_io_wq->work);
 456         } else {
 457                 if (end_io_wq->metadata)
 458                         btrfs_queue_worker(&fs_info->endio_meta_workers,
 459                                            &end_io_wq->work);
 460                 else
 461                         btrfs_queue_worker(&fs_info->endio_workers,
 462                                            &end_io_wq->work);
 463         }
 464 }
 465
 466 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 467                         int metadata)
 468 {
 469         struct end_io_wq *end_io_wq;
 470         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 471         if (!end_io_wq)
 472                 return -ENOMEM;
 473
 474         end_io_wq->private = bio->bi_private;
 475         end_io_wq->end_io = bio->bi_end_io;
 476         end_io_wq->info = info;
 477         end_io_wq->error = 0;
 478         end_io_wq->bio = bio;
 479         end_io_wq->metadata = metadata;
 480
 481         bio->bi_private = end_io_wq;
 482         bio->bi_end_io = end_workqueue_bio;
 483         return 0;
 484 }
 485
 486 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 487 {
 488         unsigned long limit = min_t(unsigned long,
 489                                     info->workers.max_workers,
 490                                     info->fs_devices->open_devices);
 491         return 256 * limit;
 492 }
 493
 494 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
 495 {
 496         return atomic_read(&info->nr_async_bios) >
 497                 btrfs_async_submit_limit(info);
 498 }
 499
 500 static void run_one_async_start(struct btrfs_work *work)
 501 {
 502         struct btrfs_fs_info *fs_info;
 503         struct async_submit_bio *async;
 504
 505         async = container_of(work, struct  async_submit_bio, work);
 506         fs_info = BTRFS_I(async->inode)->root->fs_info;
 507         async->submit_bio_start(async->inode, async->rw, async->bio,
 508                                async->mirror_num, async->bio_flags);
 509 }
 510
 511 static void run_one_async_done(struct btrfs_work *work)
 512 {
 513         struct btrfs_fs_info *fs_info;
 514         struct async_submit_bio *async;
 515         int limit;
 516
 517         async = container_of(work, struct  async_submit_bio, work);
 518         fs_info = BTRFS_I(async->inode)->root->fs_info;
 519
 520         limit = btrfs_async_submit_limit(fs_info);
 521         limit = limit * 2 / 3;
 522
 523         atomic_dec(&fs_info->nr_async_submits);
 524
 525         if (atomic_read(&fs_info->nr_async_submits) < limit &&
 526             waitqueue_active(&fs_info->async_submit_wait))
 527                 wake_up(&fs_info->async_submit_wait);
 528
 529         async->submit_bio_done(async->inode, async->rw, async->bio,
 530                                async->mirror_num, async->bio_flags);
 531 }
 532
 533 static void run_one_async_free(struct btrfs_work *work)
 534 {
 535         struct async_submit_bio *async;
 536
 537         async = container_of(work, struct  async_submit_bio, work);
 538         kfree(async);
 539 }
 540
 541 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 542                         int rw, struct bio *bio, int mirror_num,
 543                         unsigned long bio_flags,
 544                         extent_submit_bio_hook_t *submit_bio_start,
 545                         extent_submit_bio_hook_t *submit_bio_done)
 546 {
 547         struct async_submit_bio *async;
 548
 549         async = kmalloc(sizeof(*async), GFP_NOFS);
 550         if (!async)
 551                 return -ENOMEM;
 552
 553         async->inode = inode;
 554         async->rw = rw;
 555         async->bio = bio;
 556         async->mirror_num = mirror_num;
 557         async->submit_bio_start = submit_bio_start;
 558         async->submit_bio_done = submit_bio_done;
 559
 560         async->work.func = run_one_async_start;
 561         async->work.ordered_func = run_one_async_done;
 562         async->work.ordered_free = run_one_async_free;
 563
 564         async->work.flags = 0;
 565         async->bio_flags = bio_flags;
 566
 567         atomic_inc(&fs_info->nr_async_submits);
 568         btrfs_queue_worker(&fs_info->workers, &async->work);
 569 #if 0
 570         int limit = btrfs_async_submit_limit(fs_info);
 571         if (atomic_read(&fs_info->nr_async_submits) > limit) {
 572                 wait_event_timeout(fs_info->async_submit_wait,
 573                            (atomic_read(&fs_info->nr_async_submits) < limit),
 574                            HZ/10);
 575
 576                 wait_event_timeout(fs_info->async_submit_wait,
 577                            (atomic_read(&fs_info->nr_async_bios) < limit),
 578                            HZ/10);
 579         }
 580 #endif
 581         while(atomic_read(&fs_info->async_submit_draining) &&
 582               atomic_read(&fs_info->nr_async_submits)) {
 583                 wait_event(fs_info->async_submit_wait,
 584                            (atomic_read(&fs_info->nr_async_submits) == 0));
 585         }
 586
 587         return 0;
 588 }
 589
 590 static int btree_csum_one_bio(struct bio *bio)
 591 {
 592         struct bio_vec *bvec = bio->bi_io_vec;
 593         int bio_index = 0;
 594         struct btrfs_root *root;
 595
 596         WARN_ON(bio->bi_vcnt <= 0);
 597         while(bio_index < bio->bi_vcnt) {
 598                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 599                 csum_dirty_buffer(root, bvec->bv_page);
 600                 bio_index++;
 601                 bvec++;
 602         }
 603         return 0;
 604 }
 605
 606 static int __btree_submit_bio_start(struct inode *inode, int rw,
 607                                     struct bio *bio, int mirror_num,
 608                                     unsigned long bio_flags)
 609 {
 610         /*
 611          * when we're called for a write, we're already in the async
 612          * submission context.  Just jump into btrfs_map_bio
 613          */
 614         btree_csum_one_bio(bio);
 615         return 0;
 616 }
 617
 618 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 619                                  int mirror_num, unsigned long bio_flags)
 620 {
 621         /*
 622          * when we're called for a write, we're already in the async
 623          * submission context.  Just jump into btrfs_map_bio
 624          */
 625         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 626 }
 627
 628 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 629                                  int mirror_num, unsigned long bio_flags)
 630 {
 631         int ret;
 632
 633         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 634                                           bio, 1);
 635         BUG_ON(ret);
 636
 637         if (!(rw & (1 << BIO_RW))) {
 638                 /*
 639                  * called for a read, do the setup so that checksum validation
 640                  * can happen in the async kernel threads
 641                  */
 642                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 643                                      mirror_num, 0);
 644         }
 645         /*
 646          * kthread helpers are used to submit writes so that checksumming
 647          * can happen in parallel across all CPUs
 648          */
 649         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 650                                    inode, rw, bio, mirror_num, 0,
 651                                    __btree_submit_bio_start,
 652                                    __btree_submit_bio_done);
 653 }
 654
 655 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 656 {
 657         struct extent_io_tree *tree;
 658         tree = &BTRFS_I(page->mapping->host)->io_tree;
 659
 660         if (current->flags & PF_MEMALLOC) {
 661                 redirty_page_for_writepage(wbc, page);
 662                 unlock_page(page);
 663                 return 0;
 664         }
 665         return extent_write_full_page(tree, page, btree_get_extent, wbc);
 666 }
 667
 668 static int btree_writepages(struct address_space *mapping,
 669                             struct writeback_control *wbc)
 670 {
 671         struct extent_io_tree *tree;
 672         tree = &BTRFS_I(mapping->host)->io_tree;
 673         if (wbc->sync_mode == WB_SYNC_NONE) {
 674                 u64 num_dirty;
 675                 u64 start = 0;
 676                 unsigned long thresh = 32 * 1024 * 1024;
 677
 678                 if (wbc->for_kupdate)
 679                         return 0;
 680
 681                 num_dirty = count_range_bits(tree, &start, (u64)-1,
 682                                              thresh, EXTENT_DIRTY);
 683                 if (num_dirty < thresh) {
 684                         return 0;
 685                 }
 686         }
 687         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 688 }
 689
 690 static int btree_readpage(struct file *file, struct page *page)
 691 {
 692         struct extent_io_tree *tree;
 693         tree = &BTRFS_I(page->mapping->host)->io_tree;
 694         return extent_read_full_page(tree, page, btree_get_extent);
 695 }
 696
 697 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 698 {
 699         struct extent_io_tree *tree;
 700         struct extent_map_tree *map;
 701         int ret;
 702
 703         if (PageWriteback(page) || PageDirty(page))
 704             return 0;
 705
 706         tree = &BTRFS_I(page->mapping->host)->io_tree;
 707         map = &BTRFS_I(page->mapping->host)->extent_tree;
 708
 709         ret = try_release_extent_state(map, tree, page, gfp_flags);
 710         if (!ret) {
 711                 return 0;
 712         }
 713
 714         ret = try_release_extent_buffer(tree, page);
 715         if (ret == 1) {
 716                 ClearPagePrivate(page);
 717                 set_page_private(page, 0);
 718                 page_cache_release(page);
 719         }
 720
 721         return ret;
 722 }
 723
 724 static void btree_invalidatepage(struct page *page, unsigned long offset)
 725 {
 726         struct extent_io_tree *tree;
 727         tree = &BTRFS_I(page->mapping->host)->io_tree;
 728         extent_invalidatepage(tree, page, offset);
 729         btree_releasepage(page, GFP_NOFS);
 730         if (PagePrivate(page)) {
 731                 printk("warning page private not zero on page %Lu\n",
 732                        page_offset(page));
 733                 ClearPagePrivate(page);
 734                 set_page_private(page, 0);
 735                 page_cache_release(page);
 736         }
 737 }
 738
 739 #if 0
 740 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 741 {
 742         struct buffer_head *bh;
 743         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 744         struct buffer_head *head;
 745         if (!page_has_buffers(page)) {
 746                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 747                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 748         }
 749         head = page_buffers(page);
 750         bh = head;
 751         do {
 752                 if (buffer_dirty(bh))
 753                         csum_tree_block(root, bh, 0);
 754                 bh = bh->b_this_page;
 755         } while (bh != head);
 756         return block_write_full_page(page, btree_get_block, wbc);
 757 }
 758 #endif
 759
 760 static struct address_space_operations btree_aops = {
 761         .readpage       = btree_readpage,
 762         .writepage      = btree_writepage,
 763         .writepages     = btree_writepages,
 764         .releasepage    = btree_releasepage,
 765         .invalidatepage = btree_invalidatepage,
 766         .sync_page      = block_sync_page,
 767 };
 768
 769 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 770                          u64 parent_transid)
 771 {
 772         struct extent_buffer *buf = NULL;
 773         struct inode *btree_inode = root->fs_info->btree_inode;
 774         int ret = 0;
 775
 776         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 777         if (!buf)
 778                 return 0;
 779         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 780                                  buf, 0, 0, btree_get_extent, 0);
 781         free_extent_buffer(buf);
 782         return ret;
 783 }
 784
 785 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 786                                             u64 bytenr, u32 blocksize)
 787 {
 788         struct inode *btree_inode = root->fs_info->btree_inode;
 789         struct extent_buffer *eb;
 790         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 791                                 bytenr, blocksize, GFP_NOFS);
 792         return eb;
 793 }
 794
 795 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 796                                                  u64 bytenr, u32 blocksize)
 797 {
 798         struct inode *btree_inode = root->fs_info->btree_inode;
 799         struct extent_buffer *eb;
 800
 801         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 802                                  bytenr, blocksize, NULL, GFP_NOFS);
 803         return eb;
 804 }
 805
 806
 807 int btrfs_write_tree_block(struct extent_buffer *buf)
 808 {
 809         return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
 810                                       buf->start + buf->len - 1, WB_SYNC_ALL);
 811 }
 812
 813 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 814 {
 815         return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
 816                                   buf->start, buf->start + buf->len -1);
 817 }
 818
 819 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 820                                       u32 blocksize, u64 parent_transid)
 821 {
 822         struct extent_buffer *buf = NULL;
 823         struct inode *btree_inode = root->fs_info->btree_inode;
 824         struct extent_io_tree *io_tree;
 825         int ret;
 826
 827         io_tree = &BTRFS_I(btree_inode)->io_tree;
 828
 829         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 830         if (!buf)
 831                 return NULL;
 832
 833         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 834
 835         if (ret == 0) {
 836                 buf->flags |= EXTENT_UPTODATE;
 837         } else {
 838                 WARN_ON(1);
 839         }
 840         return buf;
 841
 842 }
 843
 844 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 845                      struct extent_buffer *buf)
 846 {
 847         struct inode *btree_inode = root->fs_info->btree_inode;
 848         if (btrfs_header_generation(buf) ==
 849             root->fs_info->running_transaction->transid) {
 850                 WARN_ON(!btrfs_tree_locked(buf));
 851                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 852                                           buf);
 853         }
 854         return 0;
 855 }
 856
 857 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 858                         u32 stripesize, struct btrfs_root *root,
 859                         struct btrfs_fs_info *fs_info,
 860                         u64 objectid)
 861 {
 862         root->node = NULL;
 863         root->commit_root = NULL;
 864         root->ref_tree = NULL;
 865         root->sectorsize = sectorsize;
 866         root->nodesize = nodesize;
 867         root->leafsize = leafsize;
 868         root->stripesize = stripesize;
 869         root->ref_cows = 0;
 870         root->track_dirty = 0;
 871
 872         root->fs_info = fs_info;
 873         root->objectid = objectid;
 874         root->last_trans = 0;
 875         root->highest_inode = 0;
 876         root->last_inode_alloc = 0;
 877         root->name = NULL;
 878         root->in_sysfs = 0;
 879
 880         INIT_LIST_HEAD(&root->dirty_list);
 881         INIT_LIST_HEAD(&root->orphan_list);
 882         INIT_LIST_HEAD(&root->dead_list);
 883         spin_lock_init(&root->node_lock);
 884         spin_lock_init(&root->list_lock);
 885         mutex_init(&root->objectid_mutex);
 886         mutex_init(&root->log_mutex);
 887         extent_io_tree_init(&root->dirty_log_pages,
 888                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 889
 890         btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
 891         root->ref_tree = &root->ref_tree_struct;
 892
 893         memset(&root->root_key, 0, sizeof(root->root_key));
 894         memset(&root->root_item, 0, sizeof(root->root_item));
 895         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 896         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 897         root->defrag_trans_start = fs_info->generation;
 898         init_completion(&root->kobj_unregister);
 899         root->defrag_running = 0;
 900         root->defrag_level = 0;
 901         root->root_key.objectid = objectid;
 902         root->anon_super.s_root = NULL;
 903         root->anon_super.s_dev = 0;
 904         INIT_LIST_HEAD(&root->anon_super.s_list);
 905         INIT_LIST_HEAD(&root->anon_super.s_instances);
 906         init_rwsem(&root->anon_super.s_umount);
 907
 908         return 0;
 909 }
 910
 911 static int find_and_setup_root(struct btrfs_root *tree_root,
 912                                struct btrfs_fs_info *fs_info,
 913                                u64 objectid,
 914                                struct btrfs_root *root)
 915 {
 916         int ret;
 917         u32 blocksize;
 918         u64 generation;
 919
 920         __setup_root(tree_root->nodesize, tree_root->leafsize,
 921                      tree_root->sectorsize, tree_root->stripesize,
 922                      root, fs_info, objectid);
 923         ret = btrfs_find_last_root(tree_root, objectid,
 924                                    &root->root_item, &root->root_key);
 925         BUG_ON(ret);
 926
 927         generation = btrfs_root_generation(&root->root_item);
 928         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 929         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 930                                      blocksize, generation);
 931         BUG_ON(!root->node);
 932         return 0;
 933 }
 934
 935 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
 936                              struct btrfs_fs_info *fs_info)
 937 {
 938         struct extent_buffer *eb;
 939         struct btrfs_root *log_root_tree = fs_info->log_root_tree;
 940         u64 start = 0;
 941         u64 end = 0;
 942         int ret;
 943
 944         if (!log_root_tree)
 945                 return 0;
 946
 947         while(1) {
 948                 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
 949                                     0, &start, &end, EXTENT_DIRTY);
 950                 if (ret)
 951                         break;
 952
 953                 clear_extent_dirty(&log_root_tree->dirty_log_pages,
 954                                    start, end, GFP_NOFS);
 955         }
 956         eb = fs_info->log_root_tree->node;
 957
 958         WARN_ON(btrfs_header_level(eb) != 0);
 959         WARN_ON(btrfs_header_nritems(eb) != 0);
 960
 961         ret = btrfs_free_reserved_extent(fs_info->tree_root,
 962                                 eb->start, eb->len);
 963         BUG_ON(ret);
 964
 965         free_extent_buffer(eb);
 966         kfree(fs_info->log_root_tree);
 967         fs_info->log_root_tree = NULL;
 968         return 0;
 969 }
 970
 971 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
 972                              struct btrfs_fs_info *fs_info)
 973 {
 974         struct btrfs_root *root;
 975         struct btrfs_root *tree_root = fs_info->tree_root;
 976
 977         root = kzalloc(sizeof(*root), GFP_NOFS);
 978         if (!root)
 979                 return -ENOMEM;
 980
 981         __setup_root(tree_root->nodesize, tree_root->leafsize,
 982                      tree_root->sectorsize, tree_root->stripesize,
 983                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
 984
 985         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
 986         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
 987         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
 988         root->ref_cows = 0;
 989
 990         root->node = btrfs_alloc_free_block(trans, root, root->leafsize,
 991                                             0, BTRFS_TREE_LOG_OBJECTID,
 992                                             trans->transid, 0, 0, 0);
 993
 994         btrfs_set_header_nritems(root->node, 0);
 995         btrfs_set_header_level(root->node, 0);
 996         btrfs_set_header_bytenr(root->node, root->node->start);
 997         btrfs_set_header_generation(root->node, trans->transid);
 998         btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
 999
1000         write_extent_buffer(root->node, root->fs_info->fsid,
1001                             (unsigned long)btrfs_header_fsid(root->node),
1002                             BTRFS_FSID_SIZE);
1003         btrfs_mark_buffer_dirty(root->node);
1004         btrfs_tree_unlock(root->node);
1005         fs_info->log_root_tree = root;
1006         return 0;
1007 }
1008
1009 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1010                                                struct btrfs_key *location)
1011 {
1012         struct btrfs_root *root;
1013         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1014         struct btrfs_path *path;
1015         struct extent_buffer *l;
1016         u64 highest_inode;
1017         u64 generation;
1018         u32 blocksize;
1019         int ret = 0;
1020
1021         root = kzalloc(sizeof(*root), GFP_NOFS);
1022         if (!root)
1023                 return ERR_PTR(-ENOMEM);
1024         if (location->offset == (u64)-1) {
1025                 ret = find_and_setup_root(tree_root, fs_info,
1026                                           location->objectid, root);
1027                 if (ret) {
1028                         kfree(root);
1029                         return ERR_PTR(ret);
1030                 }
1031                 goto insert;
1032         }
1033
1034         __setup_root(tree_root->nodesize, tree_root->leafsize,
1035                      tree_root->sectorsize, tree_root->stripesize,
1036                      root, fs_info, location->objectid);
1037
1038         path = btrfs_alloc_path();
1039         BUG_ON(!path);
1040         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1041         if (ret != 0) {
1042                 if (ret > 0)
1043                         ret = -ENOENT;
1044                 goto out;
1045         }
1046         l = path->nodes[0];
1047         read_extent_buffer(l, &root->root_item,
1048                btrfs_item_ptr_offset(l, path->slots[0]),
1049                sizeof(root->root_item));
1050         memcpy(&root->root_key, location, sizeof(*location));
1051         ret = 0;
1052 out:
1053         btrfs_release_path(root, path);
1054         btrfs_free_path(path);
1055         if (ret) {
1056                 kfree(root);
1057                 return ERR_PTR(ret);
1058         }
1059         generation = btrfs_root_generation(&root->root_item);
1060         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1061         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1062                                      blocksize, generation);
1063         BUG_ON(!root->node);
1064 insert:
1065         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1066                 root->ref_cows = 1;
1067                 ret = btrfs_find_highest_inode(root, &highest_inode);
1068                 if (ret == 0) {
1069                         root->highest_inode = highest_inode;
1070                         root->last_inode_alloc = highest_inode;
1071                 }
1072         }
1073         return root;
1074 }
1075
1076 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1077                                         u64 root_objectid)
1078 {
1079         struct btrfs_root *root;
1080
1081         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1082                 return fs_info->tree_root;
1083         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1084                 return fs_info->extent_root;
1085
1086         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1087                                  (unsigned long)root_objectid);
1088         return root;
1089 }
1090
1091 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1092                                               struct btrfs_key *location)
1093 {
1094         struct btrfs_root *root;
1095         int ret;
1096
1097         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1098                 return fs_info->tree_root;
1099         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1100                 return fs_info->extent_root;
1101         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1102                 return fs_info->chunk_root;
1103         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1104                 return fs_info->dev_root;
1105         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1106                 return fs_info->csum_root;
1107
1108         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1109                                  (unsigned long)location->objectid);
1110         if (root)
1111                 return root;
1112
1113         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1114         if (IS_ERR(root))
1115                 return root;
1116
1117         set_anon_super(&root->anon_super, NULL);
1118
1119         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1120                                 (unsigned long)root->root_key.objectid,
1121                                 root);
1122         if (ret) {
1123                 free_extent_buffer(root->node);
1124                 kfree(root);
1125                 return ERR_PTR(ret);
1126         }
1127         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
1128                 ret = btrfs_find_dead_roots(fs_info->tree_root,
1129                                             root->root_key.objectid, root);
1130                 BUG_ON(ret);
1131                 btrfs_orphan_cleanup(root);
1132         }
1133         return root;
1134 }
1135
1136 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1137                                       struct btrfs_key *location,
1138                                       const char *name, int namelen)
1139 {
1140         struct btrfs_root *root;
1141         int ret;
1142
1143         root = btrfs_read_fs_root_no_name(fs_info, location);
1144         if (!root)
1145                 return NULL;
1146
1147         if (root->in_sysfs)
1148                 return root;
1149
1150         ret = btrfs_set_root_name(root, name, namelen);
1151         if (ret) {
1152                 free_extent_buffer(root->node);
1153                 kfree(root);
1154                 return ERR_PTR(ret);
1155         }
1156 #if 0
1157         ret = btrfs_sysfs_add_root(root);
1158         if (ret) {
1159                 free_extent_buffer(root->node);
1160                 kfree(root->name);
1161                 kfree(root);
1162                 return ERR_PTR(ret);
1163         }
1164 #endif
1165         root->in_sysfs = 1;
1166         return root;
1167 }
1168 #if 0
1169 static int add_hasher(struct btrfs_fs_info *info, char *type) {
1170         struct btrfs_hasher *hasher;
1171
1172         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
1173         if (!hasher)
1174                 return -ENOMEM;
1175         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
1176         if (!hasher->hash_tfm) {
1177                 kfree(hasher);
1178                 return -EINVAL;
1179         }
1180         spin_lock(&info->hash_lock);
1181         list_add(&hasher->list, &info->hashers);
1182         spin_unlock(&info->hash_lock);
1183         return 0;
1184 }
1185 #endif
1186
1187 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1188 {
1189         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1190         int ret = 0;
1191         struct list_head *cur;
1192         struct btrfs_device *device;
1193         struct backing_dev_info *bdi;
1194 #if 0
1195         if ((bdi_bits & (1 << BDI_write_congested)) &&
1196             btrfs_congested_async(info, 0))
1197                 return 1;
1198 #endif
1199         list_for_each(cur, &info->fs_devices->devices) {
1200                 device = list_entry(cur, struct btrfs_device, dev_list);
1201                 if (!device->bdev)
1202                         continue;
1203                 bdi = blk_get_backing_dev_info(device->bdev);
1204                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1205                         ret = 1;
1206                         break;
1207                 }
1208         }
1209         return ret;
1210 }
1211
1212 /*
1213  * this unplugs every device on the box, and it is only used when page
1214  * is null
1215  */
1216 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1217 {
1218         struct list_head *cur;
1219         struct btrfs_device *device;
1220         struct btrfs_fs_info *info;
1221
1222         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1223         list_for_each(cur, &info->fs_devices->devices) {
1224                 device = list_entry(cur, struct btrfs_device, dev_list);
1225                 if (!device->bdev)
1226                         continue;
1227
1228                 bdi = blk_get_backing_dev_info(device->bdev);
1229                 if (bdi->unplug_io_fn) {
1230                         bdi->unplug_io_fn(bdi, page);
1231                 }
1232         }
1233 }
1234
1235 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1236 {
1237         struct inode *inode;
1238         struct extent_map_tree *em_tree;
1239         struct extent_map *em;
1240         struct address_space *mapping;
1241         u64 offset;
1242
1243         /* the generic O_DIRECT read code does this */
1244         if (1 || !page) {
1245                 __unplug_io_fn(bdi, page);
1246                 return;
1247         }
1248
1249         /*
1250          * page->mapping may change at any time.  Get a consistent copy
1251          * and use that for everything below
1252          */
1253         smp_mb();
1254         mapping = page->mapping;
1255         if (!mapping)
1256                 return;
1257
1258         inode = mapping->host;
1259
1260         /*
1261          * don't do the expensive searching for a small number of
1262          * devices
1263          */
1264         if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1265                 __unplug_io_fn(bdi, page);
1266                 return;
1267         }
1268
1269         offset = page_offset(page);
1270
1271         em_tree = &BTRFS_I(inode)->extent_tree;
1272         spin_lock(&em_tree->lock);
1273         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1274         spin_unlock(&em_tree->lock);
1275         if (!em) {
1276                 __unplug_io_fn(bdi, page);
1277                 return;
1278         }
1279
1280         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1281                 free_extent_map(em);
1282                 __unplug_io_fn(bdi, page);
1283                 return;
1284         }
1285         offset = offset - em->start;
1286         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1287                           em->block_start + offset, page);
1288         free_extent_map(em);
1289 }
1290
1291 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1292 {
1293         bdi_init(bdi);
1294         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1295         bdi->state              = 0;
1296         bdi->capabilities       = default_backing_dev_info.capabilities;
1297         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1298         bdi->unplug_io_data     = info;
1299         bdi->congested_fn       = btrfs_congested_fn;
1300         bdi->congested_data     = info;
1301         return 0;
1302 }
1303
1304 static int bio_ready_for_csum(struct bio *bio)
1305 {
1306         u64 length = 0;
1307         u64 buf_len = 0;
1308         u64 start = 0;
1309         struct page *page;
1310         struct extent_io_tree *io_tree = NULL;
1311         struct btrfs_fs_info *info = NULL;
1312         struct bio_vec *bvec;
1313         int i;
1314         int ret;
1315
1316         bio_for_each_segment(bvec, bio, i) {
1317                 page = bvec->bv_page;
1318                 if (page->private == EXTENT_PAGE_PRIVATE) {
1319                         length += bvec->bv_len;
1320                         continue;
1321                 }
1322                 if (!page->private) {
1323                         length += bvec->bv_len;
1324                         continue;
1325                 }
1326                 length = bvec->bv_len;
1327                 buf_len = page->private >> 2;
1328                 start = page_offset(page) + bvec->bv_offset;
1329                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1330                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1331         }
1332         /* are we fully contained in this bio? */
1333         if (buf_len <= length)
1334                 return 1;
1335
1336         ret = extent_range_uptodate(io_tree, start + length,
1337                                     start + buf_len - 1);
1338         if (ret == 1)
1339                 return ret;
1340         return ret;
1341 }
1342
1343 /*
1344  * called by the kthread helper functions to finally call the bio end_io
1345  * functions.  This is where read checksum verification actually happens
1346  */
1347 static void end_workqueue_fn(struct btrfs_work *work)
1348 {
1349         struct bio *bio;
1350         struct end_io_wq *end_io_wq;
1351         struct btrfs_fs_info *fs_info;
1352         int error;
1353
1354         end_io_wq = container_of(work, struct end_io_wq, work);
1355         bio = end_io_wq->bio;
1356         fs_info = end_io_wq->info;
1357
1358         /* metadata bio reads are special because the whole tree block must
1359          * be checksummed at once.  This makes sure the entire block is in
1360          * ram and up to date before trying to verify things.  For
1361          * blocksize <= pagesize, it is basically a noop
1362          */
1363         if (!(bio->bi_rw & (1 << BIO_RW)) && end_io_wq->metadata &&
1364             !bio_ready_for_csum(bio)) {
1365                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1366                                    &end_io_wq->work);
1367                 return;
1368         }
1369         error = end_io_wq->error;
1370         bio->bi_private = end_io_wq->private;
1371         bio->bi_end_io = end_io_wq->end_io;
1372         kfree(end_io_wq);
1373         bio_endio(bio, error);
1374 }
1375
1376 static int cleaner_kthread(void *arg)
1377 {
1378         struct btrfs_root *root = arg;
1379
1380         do {
1381                 smp_mb();
1382                 if (root->fs_info->closing)
1383                         break;
1384
1385                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1386                 mutex_lock(&root->fs_info->cleaner_mutex);
1387                 btrfs_clean_old_snapshots(root);
1388                 mutex_unlock(&root->fs_info->cleaner_mutex);
1389
1390                 if (freezing(current)) {
1391                         refrigerator();
1392                 } else {
1393                         smp_mb();
1394                         if (root->fs_info->closing)
1395                                 break;
1396                         set_current_state(TASK_INTERRUPTIBLE);
1397                         schedule();
1398                         __set_current_state(TASK_RUNNING);
1399                 }
1400         } while (!kthread_should_stop());
1401         return 0;
1402 }
1403
1404 static int transaction_kthread(void *arg)
1405 {
1406         struct btrfs_root *root = arg;
1407         struct btrfs_trans_handle *trans;
1408         struct btrfs_transaction *cur;
1409         unsigned long now;
1410         unsigned long delay;
1411         int ret;
1412
1413         do {
1414                 smp_mb();
1415                 if (root->fs_info->closing)
1416                         break;
1417
1418                 delay = HZ * 30;
1419                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1420                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1421
1422                 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1423                         printk("btrfs: total reference cache size %Lu\n",
1424                                 root->fs_info->total_ref_cache_size);
1425                 }
1426
1427                 mutex_lock(&root->fs_info->trans_mutex);
1428                 cur = root->fs_info->running_transaction;
1429                 if (!cur) {
1430                         mutex_unlock(&root->fs_info->trans_mutex);
1431                         goto sleep;
1432                 }
1433
1434                 now = get_seconds();
1435                 if (now < cur->start_time || now - cur->start_time < 30) {
1436                         mutex_unlock(&root->fs_info->trans_mutex);
1437                         delay = HZ * 5;
1438                         goto sleep;
1439                 }
1440                 mutex_unlock(&root->fs_info->trans_mutex);
1441                 trans = btrfs_start_transaction(root, 1);
1442                 ret = btrfs_commit_transaction(trans, root);
1443 sleep:
1444                 wake_up_process(root->fs_info->cleaner_kthread);
1445                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1446
1447                 if (freezing(current)) {
1448                         refrigerator();
1449                 } else {
1450                         if (root->fs_info->closing)
1451                                 break;
1452                         set_current_state(TASK_INTERRUPTIBLE);
1453                         schedule_timeout(delay);
1454                         __set_current_state(TASK_RUNNING);
1455                 }
1456         } while (!kthread_should_stop());
1457         return 0;
1458 }
1459
1460 struct btrfs_root *open_ctree(struct super_block *sb,
1461                               struct btrfs_fs_devices *fs_devices,
1462                               char *options)
1463 {
1464         u32 sectorsize;
1465         u32 nodesize;
1466         u32 leafsize;
1467         u32 blocksize;
1468         u32 stripesize;
1469         u64 generation;
1470         u64 features;
1471         struct btrfs_key location;
1472         struct buffer_head *bh;
1473         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1474                                                  GFP_NOFS);
1475         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1476                                                  GFP_NOFS);
1477         struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1478                                                GFP_NOFS);
1479         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1480                                                 GFP_NOFS);
1481         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1482                                                 GFP_NOFS);
1483         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1484                                               GFP_NOFS);
1485         struct btrfs_root *log_tree_root;
1486
1487         int ret;
1488         int err = -EINVAL;
1489
1490         struct btrfs_super_block *disk_super;
1491
1492         if (!extent_root || !tree_root || !fs_info ||
1493             !chunk_root || !dev_root || !csum_root) {
1494                 err = -ENOMEM;
1495                 goto fail;
1496         }
1497         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1498         INIT_LIST_HEAD(&fs_info->trans_list);
1499         INIT_LIST_HEAD(&fs_info->dead_roots);
1500         INIT_LIST_HEAD(&fs_info->hashers);
1501         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1502         spin_lock_init(&fs_info->hash_lock);
1503         spin_lock_init(&fs_info->delalloc_lock);
1504         spin_lock_init(&fs_info->new_trans_lock);
1505         spin_lock_init(&fs_info->ref_cache_lock);
1506
1507         init_completion(&fs_info->kobj_unregister);
1508         fs_info->tree_root = tree_root;
1509         fs_info->extent_root = extent_root;
1510         fs_info->csum_root = csum_root;
1511         fs_info->chunk_root = chunk_root;
1512         fs_info->dev_root = dev_root;
1513         fs_info->fs_devices = fs_devices;
1514         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1515         INIT_LIST_HEAD(&fs_info->space_info);
1516         btrfs_mapping_init(&fs_info->mapping_tree);
1517         atomic_set(&fs_info->nr_async_submits, 0);
1518         atomic_set(&fs_info->async_delalloc_pages, 0);
1519         atomic_set(&fs_info->async_submit_draining, 0);
1520         atomic_set(&fs_info->nr_async_bios, 0);
1521         atomic_set(&fs_info->throttles, 0);
1522         atomic_set(&fs_info->throttle_gen, 0);
1523         fs_info->sb = sb;
1524         fs_info->max_extent = (u64)-1;
1525         fs_info->max_inline = 8192 * 1024;
1526         setup_bdi(fs_info, &fs_info->bdi);
1527         fs_info->btree_inode = new_inode(sb);
1528         fs_info->btree_inode->i_ino = 1;
1529         fs_info->btree_inode->i_nlink = 1;
1530
1531         fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1532
1533         INIT_LIST_HEAD(&fs_info->ordered_extents);
1534         spin_lock_init(&fs_info->ordered_extent_lock);
1535
1536         sb->s_blocksize = 4096;
1537         sb->s_blocksize_bits = blksize_bits(4096);
1538
1539         /*
1540          * we set the i_size on the btree inode to the max possible int.
1541          * the real end of the address space is determined by all of
1542          * the devices in the system
1543          */
1544         fs_info->btree_inode->i_size = OFFSET_MAX;
1545         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1546         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1547
1548         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1549                              fs_info->btree_inode->i_mapping,
1550                              GFP_NOFS);
1551         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1552                              GFP_NOFS);
1553
1554         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1555
1556         spin_lock_init(&fs_info->block_group_cache_lock);
1557         fs_info->block_group_cache_tree.rb_node = NULL;
1558
1559         extent_io_tree_init(&fs_info->pinned_extents,
1560                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1561         extent_io_tree_init(&fs_info->pending_del,
1562                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1563         extent_io_tree_init(&fs_info->extent_ins,
1564                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1565         fs_info->do_barriers = 1;
1566
1567         INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
1568         btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
1569         btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
1570
1571         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1572         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1573                sizeof(struct btrfs_key));
1574         insert_inode_hash(fs_info->btree_inode);
1575
1576         mutex_init(&fs_info->trans_mutex);
1577         mutex_init(&fs_info->tree_log_mutex);
1578         mutex_init(&fs_info->drop_mutex);
1579         mutex_init(&fs_info->extent_ins_mutex);
1580         mutex_init(&fs_info->pinned_mutex);
1581         mutex_init(&fs_info->chunk_mutex);
1582         mutex_init(&fs_info->transaction_kthread_mutex);
1583         mutex_init(&fs_info->cleaner_mutex);
1584         mutex_init(&fs_info->volume_mutex);
1585         mutex_init(&fs_info->tree_reloc_mutex);
1586         init_waitqueue_head(&fs_info->transaction_throttle);
1587         init_waitqueue_head(&fs_info->transaction_wait);
1588         init_waitqueue_head(&fs_info->async_submit_wait);
1589         init_waitqueue_head(&fs_info->tree_log_wait);
1590         atomic_set(&fs_info->tree_log_commit, 0);
1591         atomic_set(&fs_info->tree_log_writers, 0);
1592         fs_info->tree_log_transid = 0;
1593
1594 #if 0
1595         ret = add_hasher(fs_info, "crc32c");
1596         if (ret) {
1597                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1598                 err = -ENOMEM;
1599                 goto fail_iput;
1600         }
1601 #endif
1602         __setup_root(4096, 4096, 4096, 4096, tree_root,
1603                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1604
1605
1606         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1607         if (!bh)
1608                 goto fail_iput;
1609
1610         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1611         brelse(bh);
1612
1613         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1614
1615         disk_super = &fs_info->super_copy;
1616         if (!btrfs_super_root(disk_super))
1617                 goto fail_iput;
1618
1619         ret = btrfs_parse_options(tree_root, options);
1620         if (ret) {
1621                 err = ret;
1622                 goto fail_iput;
1623         }
1624
1625         features = btrfs_super_incompat_flags(disk_super) &
1626                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1627         if (features) {
1628                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1629                        "unsupported optional features (%Lx).\n",
1630                        features);
1631                 err = -EINVAL;
1632                 goto fail_iput;
1633         }
1634
1635         features = btrfs_super_compat_ro_flags(disk_super) &
1636                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1637         if (!(sb->s_flags & MS_RDONLY) && features) {
1638                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1639                        "unsupported option features (%Lx).\n",
1640                        features);
1641                 err = -EINVAL;
1642                 goto fail_iput;
1643         }
1644
1645         /*
1646          * we need to start all the end_io workers up front because the
1647          * queue work function gets called at interrupt time, and so it
1648          * cannot dynamically grow.
1649          */
1650         btrfs_init_workers(&fs_info->workers, "worker",
1651                            fs_info->thread_pool_size);
1652
1653         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1654                            fs_info->thread_pool_size);
1655
1656         btrfs_init_workers(&fs_info->submit_workers, "submit",
1657                            min_t(u64, fs_devices->num_devices,
1658                            fs_info->thread_pool_size));
1659
1660         /* a higher idle thresh on the submit workers makes it much more
1661          * likely that bios will be send down in a sane order to the
1662          * devices
1663          */
1664         fs_info->submit_workers.idle_thresh = 64;
1665
1666         fs_info->workers.idle_thresh = 16;
1667         fs_info->workers.ordered = 1;
1668
1669         fs_info->delalloc_workers.idle_thresh = 2;
1670         fs_info->delalloc_workers.ordered = 1;
1671
1672         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1673         btrfs_init_workers(&fs_info->endio_workers, "endio",
1674                            fs_info->thread_pool_size);
1675         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1676                            fs_info->thread_pool_size);
1677         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1678                            "endio-meta-write", fs_info->thread_pool_size);
1679         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1680                            fs_info->thread_pool_size);
1681
1682         /*
1683          * endios are largely parallel and should have a very
1684          * low idle thresh
1685          */
1686         fs_info->endio_workers.idle_thresh = 4;
1687         fs_info->endio_write_workers.idle_thresh = 64;
1688         fs_info->endio_meta_write_workers.idle_thresh = 64;
1689
1690         btrfs_start_workers(&fs_info->workers, 1);
1691         btrfs_start_workers(&fs_info->submit_workers, 1);
1692         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1693         btrfs_start_workers(&fs_info->fixup_workers, 1);
1694         btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1695         btrfs_start_workers(&fs_info->endio_meta_workers,
1696                             fs_info->thread_pool_size);
1697         btrfs_start_workers(&fs_info->endio_meta_write_workers,
1698                             fs_info->thread_pool_size);
1699         btrfs_start_workers(&fs_info->endio_write_workers,
1700                             fs_info->thread_pool_size);
1701
1702         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1703         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1704                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1705
1706         nodesize = btrfs_super_nodesize(disk_super);
1707         leafsize = btrfs_super_leafsize(disk_super);
1708         sectorsize = btrfs_super_sectorsize(disk_super);
1709         stripesize = btrfs_super_stripesize(disk_super);
1710         tree_root->nodesize = nodesize;
1711         tree_root->leafsize = leafsize;
1712         tree_root->sectorsize = sectorsize;
1713         tree_root->stripesize = stripesize;
1714
1715         sb->s_blocksize = sectorsize;
1716         sb->s_blocksize_bits = blksize_bits(sectorsize);
1717
1718         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1719                     sizeof(disk_super->magic))) {
1720                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1721                 goto fail_sb_buffer;
1722         }
1723
1724         mutex_lock(&fs_info->chunk_mutex);
1725         ret = btrfs_read_sys_array(tree_root);
1726         mutex_unlock(&fs_info->chunk_mutex);
1727         if (ret) {
1728                 printk("btrfs: failed to read the system array on %s\n",
1729                        sb->s_id);
1730                 goto fail_sys_array;
1731         }
1732
1733         blocksize = btrfs_level_size(tree_root,
1734                                      btrfs_super_chunk_root_level(disk_super));
1735         generation = btrfs_super_chunk_root_generation(disk_super);
1736
1737         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1738                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1739
1740         chunk_root->node = read_tree_block(chunk_root,
1741                                            btrfs_super_chunk_root(disk_super),
1742                                            blocksize, generation);
1743         BUG_ON(!chunk_root->node);
1744
1745         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1746                  (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1747                  BTRFS_UUID_SIZE);
1748
1749         mutex_lock(&fs_info->chunk_mutex);
1750         ret = btrfs_read_chunk_tree(chunk_root);
1751         mutex_unlock(&fs_info->chunk_mutex);
1752         if (ret) {
1753                 printk("btrfs: failed to read chunk tree on %s\n", sb->s_id);
1754                 goto fail_chunk_root;
1755         }
1756
1757         btrfs_close_extra_devices(fs_devices);
1758
1759         blocksize = btrfs_level_size(tree_root,
1760                                      btrfs_super_root_level(disk_super));
1761         generation = btrfs_super_generation(disk_super);
1762
1763         tree_root->node = read_tree_block(tree_root,
1764                                           btrfs_super_root(disk_super),
1765                                           blocksize, generation);
1766         if (!tree_root->node)
1767                 goto fail_chunk_root;
1768
1769
1770         ret = find_and_setup_root(tree_root, fs_info,
1771                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1772         if (ret)
1773                 goto fail_tree_root;
1774         extent_root->track_dirty = 1;
1775
1776         ret = find_and_setup_root(tree_root, fs_info,
1777                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1778         dev_root->track_dirty = 1;
1779
1780         if (ret)
1781                 goto fail_extent_root;
1782
1783         ret = find_and_setup_root(tree_root, fs_info,
1784                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
1785         if (ret)
1786                 goto fail_extent_root;
1787
1788         csum_root->track_dirty = 1;
1789
1790         btrfs_read_block_groups(extent_root);
1791
1792         fs_info->generation = generation + 1;
1793         fs_info->last_trans_committed = generation;
1794         fs_info->data_alloc_profile = (u64)-1;
1795         fs_info->metadata_alloc_profile = (u64)-1;
1796         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1797         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1798                                                "btrfs-cleaner");
1799         if (!fs_info->cleaner_kthread)
1800                 goto fail_csum_root;
1801
1802         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1803                                                    tree_root,
1804                                                    "btrfs-transaction");
1805         if (!fs_info->transaction_kthread)
1806                 goto fail_cleaner;
1807
1808         if (btrfs_super_log_root(disk_super) != 0) {
1809                 u64 bytenr = btrfs_super_log_root(disk_super);
1810
1811                 if (fs_devices->rw_devices == 0) {
1812                         printk("Btrfs log replay required on RO media\n");
1813                         err = -EIO;
1814                         goto fail_trans_kthread;
1815                 }
1816                 blocksize =
1817                      btrfs_level_size(tree_root,
1818                                       btrfs_super_log_root_level(disk_super));
1819
1820                 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1821                                                       GFP_NOFS);
1822
1823                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1824                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1825
1826                 log_tree_root->node = read_tree_block(tree_root, bytenr,
1827                                                       blocksize,
1828                                                       generation + 1);
1829                 ret = btrfs_recover_log_trees(log_tree_root);
1830                 BUG_ON(ret);
1831
1832                 if (sb->s_flags & MS_RDONLY) {
1833                         ret =  btrfs_commit_super(tree_root);
1834                         BUG_ON(ret);
1835                 }
1836         }
1837
1838         if (!(sb->s_flags & MS_RDONLY)) {
1839                 ret = btrfs_cleanup_reloc_trees(tree_root);
1840                 BUG_ON(ret);
1841         }
1842
1843         location.objectid = BTRFS_FS_TREE_OBJECTID;
1844         location.type = BTRFS_ROOT_ITEM_KEY;
1845         location.offset = (u64)-1;
1846
1847         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1848         if (!fs_info->fs_root)
1849                 goto fail_trans_kthread;
1850         return tree_root;
1851
1852 fail_trans_kthread:
1853         kthread_stop(fs_info->transaction_kthread);
1854 fail_cleaner:
1855         kthread_stop(fs_info->cleaner_kthread);
1856
1857         /*
1858          * make sure we're done with the btree inode before we stop our
1859          * kthreads
1860          */
1861         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1862         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1863
1864 fail_csum_root:
1865         free_extent_buffer(csum_root->node);
1866 fail_extent_root:
1867         free_extent_buffer(extent_root->node);
1868 fail_tree_root:
1869         free_extent_buffer(tree_root->node);
1870 fail_chunk_root:
1871         free_extent_buffer(chunk_root->node);
1872 fail_sys_array:
1873         free_extent_buffer(dev_root->node);
1874 fail_sb_buffer:
1875         btrfs_stop_workers(&fs_info->fixup_workers);
1876         btrfs_stop_workers(&fs_info->delalloc_workers);
1877         btrfs_stop_workers(&fs_info->workers);
1878         btrfs_stop_workers(&fs_info->endio_workers);
1879         btrfs_stop_workers(&fs_info->endio_meta_workers);
1880         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
1881         btrfs_stop_workers(&fs_info->endio_write_workers);
1882         btrfs_stop_workers(&fs_info->submit_workers);
1883 fail_iput:
1884         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1885         iput(fs_info->btree_inode);
1886 fail:
1887         btrfs_close_devices(fs_info->fs_devices);
1888         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1889
1890         kfree(extent_root);
1891         kfree(tree_root);
1892         bdi_destroy(&fs_info->bdi);
1893         kfree(fs_info);
1894         kfree(chunk_root);
1895         kfree(dev_root);
1896         kfree(csum_root);
1897         return ERR_PTR(err);
1898 }
1899
1900 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1901 {
1902         char b[BDEVNAME_SIZE];
1903
1904         if (uptodate) {
1905                 set_buffer_uptodate(bh);
1906         } else {
1907                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1908                         printk(KERN_WARNING "lost page write due to "
1909                                         "I/O error on %s\n",
1910                                        bdevname(bh->b_bdev, b));
1911                 }
1912                 /* note, we dont' set_buffer_write_io_error because we have
1913                  * our own ways of dealing with the IO errors
1914                  */
1915                 clear_buffer_uptodate(bh);
1916         }
1917         unlock_buffer(bh);
1918         put_bh(bh);
1919 }
1920
1921 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
1922 {
1923         struct buffer_head *bh;
1924         struct buffer_head *latest = NULL;
1925         struct btrfs_super_block *super;
1926         int i;
1927         u64 transid = 0;
1928         u64 bytenr;
1929
1930         /* we would like to check all the supers, but that would make
1931          * a btrfs mount succeed after a mkfs from a different FS.
1932          * So, we need to add a special mount option to scan for
1933          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1934          */
1935         for (i = 0; i < 1; i++) {
1936                 bytenr = btrfs_sb_offset(i);
1937                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
1938                         break;
1939                 bh = __bread(bdev, bytenr / 4096, 4096);
1940                 if (!bh)
1941                         continue;
1942
1943                 super = (struct btrfs_super_block *)bh->b_data;
1944                 if (btrfs_super_bytenr(super) != bytenr ||
1945                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
1946                             sizeof(super->magic))) {
1947                         brelse(bh);
1948                         continue;
1949                 }
1950
1951                 if (!latest || btrfs_super_generation(super) > transid) {
1952                         brelse(latest);
1953                         latest = bh;
1954                         transid = btrfs_super_generation(super);
1955                 } else {
1956                         brelse(bh);
1957                 }
1958         }
1959         return latest;
1960 }
1961
1962 static int write_dev_supers(struct btrfs_device *device,
1963                             struct btrfs_super_block *sb,
1964                             int do_barriers, int wait, int max_mirrors)
1965 {
1966         struct buffer_head *bh;
1967         int i;
1968         int ret;
1969         int errors = 0;
1970         u32 crc;
1971         u64 bytenr;
1972         int last_barrier = 0;
1973
1974         if (max_mirrors == 0)
1975                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
1976
1977         /* make sure only the last submit_bh does a barrier */
1978         if (do_barriers) {
1979                 for (i = 0; i < max_mirrors; i++) {
1980                         bytenr = btrfs_sb_offset(i);
1981                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1982                             device->total_bytes)
1983                                 break;
1984                         last_barrier = i;
1985                 }
1986         }
1987
1988         for (i = 0; i < max_mirrors; i++) {
1989                 bytenr = btrfs_sb_offset(i);
1990                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
1991                         break;
1992
1993                 if (wait) {
1994                         bh = __find_get_block(device->bdev, bytenr / 4096,
1995                                               BTRFS_SUPER_INFO_SIZE);
1996                         BUG_ON(!bh);
1997                         brelse(bh);
1998                         wait_on_buffer(bh);
1999                         if (buffer_uptodate(bh)) {
2000                                 brelse(bh);
2001                                 continue;
2002                         }
2003                 } else {
2004                         btrfs_set_super_bytenr(sb, bytenr);
2005
2006                         crc = ~(u32)0;
2007                         crc = btrfs_csum_data(NULL, (char *)sb +
2008                                               BTRFS_CSUM_SIZE, crc,
2009                                               BTRFS_SUPER_INFO_SIZE -
2010                                               BTRFS_CSUM_SIZE);
2011                         btrfs_csum_final(crc, sb->csum);
2012
2013                         bh = __getblk(device->bdev, bytenr / 4096,
2014                                       BTRFS_SUPER_INFO_SIZE);
2015                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2016
2017                         set_buffer_uptodate(bh);
2018                         get_bh(bh);
2019                         lock_buffer(bh);
2020                         bh->b_end_io = btrfs_end_buffer_write_sync;
2021                 }
2022
2023                 if (i == last_barrier && do_barriers && device->barriers) {
2024                         ret = submit_bh(WRITE_BARRIER, bh);
2025                         if (ret == -EOPNOTSUPP) {
2026                                 printk("btrfs: disabling barriers on dev %s\n",
2027                                        device->name);
2028                                 set_buffer_uptodate(bh);
2029                                 device->barriers = 0;
2030                                 get_bh(bh);
2031                                 lock_buffer(bh);
2032                                 ret = submit_bh(WRITE, bh);
2033                         }
2034                 } else {
2035                         ret = submit_bh(WRITE, bh);
2036                 }
2037
2038                 if (!ret && wait) {
2039                         wait_on_buffer(bh);
2040                         if (!buffer_uptodate(bh))
2041                                 errors++;
2042                 } else if (ret) {
2043                         errors++;
2044                 }
2045                 if (wait)
2046                         brelse(bh);
2047         }
2048         return errors < i ? 0 : -1;
2049 }
2050
2051 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2052 {
2053         struct list_head *cur;
2054         struct list_head *head = &root->fs_info->fs_devices->devices;
2055         struct btrfs_device *dev;
2056         struct btrfs_super_block *sb;
2057         struct btrfs_dev_item *dev_item;
2058         int ret;
2059         int do_barriers;
2060         int max_errors;
2061         int total_errors = 0;
2062         u64 flags;
2063
2064         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2065         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2066
2067         sb = &root->fs_info->super_for_commit;
2068         dev_item = &sb->dev_item;
2069         list_for_each(cur, head) {
2070                 dev = list_entry(cur, struct btrfs_device, dev_list);
2071                 if (!dev->bdev) {
2072                         total_errors++;
2073                         continue;
2074                 }
2075                 if (!dev->in_fs_metadata || !dev->writeable)
2076                         continue;
2077
2078                 btrfs_set_stack_device_generation(dev_item, 0);
2079                 btrfs_set_stack_device_type(dev_item, dev->type);
2080                 btrfs_set_stack_device_id(dev_item, dev->devid);
2081                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2082                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2083                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2084                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2085                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2086                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2087                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2088
2089                 flags = btrfs_super_flags(sb);
2090                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2091
2092                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2093                 if (ret)
2094                         total_errors++;
2095         }
2096         if (total_errors > max_errors) {
2097                 printk("btrfs: %d errors while writing supers\n", total_errors);
2098                 BUG();
2099         }
2100
2101         total_errors = 0;
2102         list_for_each(cur, head) {
2103                 dev = list_entry(cur, struct btrfs_device, dev_list);
2104                 if (!dev->bdev)
2105                         continue;
2106                 if (!dev->in_fs_metadata || !dev->writeable)
2107                         continue;
2108
2109                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2110                 if (ret)
2111                         total_errors++;
2112         }
2113         if (total_errors > max_errors) {
2114                 printk("btrfs: %d errors while writing supers\n", total_errors);
2115                 BUG();
2116         }
2117         return 0;
2118 }
2119
2120 int write_ctree_super(struct btrfs_trans_handle *trans,
2121                       struct btrfs_root *root, int max_mirrors)
2122 {
2123         int ret;
2124
2125         ret = write_all_supers(root, max_mirrors);
2126         return ret;
2127 }
2128
2129 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2130 {
2131         radix_tree_delete(&fs_info->fs_roots_radix,
2132                           (unsigned long)root->root_key.objectid);
2133         if (root->anon_super.s_dev) {
2134                 down_write(&root->anon_super.s_umount);
2135                 kill_anon_super(&root->anon_super);
2136         }
2137 #if 0
2138         if (root->in_sysfs)
2139                 btrfs_sysfs_del_root(root);
2140 #endif
2141         if (root->node)
2142                 free_extent_buffer(root->node);
2143         if (root->commit_root)
2144                 free_extent_buffer(root->commit_root);
2145         if (root->name)
2146                 kfree(root->name);
2147         kfree(root);
2148         return 0;
2149 }
2150
2151 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2152 {
2153         int ret;
2154         struct btrfs_root *gang[8];
2155         int i;
2156
2157         while(1) {
2158                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2159                                              (void **)gang, 0,
2160                                              ARRAY_SIZE(gang));
2161                 if (!ret)
2162                         break;
2163                 for (i = 0; i < ret; i++)
2164                         btrfs_free_fs_root(fs_info, gang[i]);
2165         }
2166         return 0;
2167 }
2168
2169 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2170 {
2171         u64 root_objectid = 0;
2172         struct btrfs_root *gang[8];
2173         int i;
2174         int ret;
2175
2176         while (1) {
2177                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2178                                              (void **)gang, root_objectid,
2179                                              ARRAY_SIZE(gang));
2180                 if (!ret)
2181                         break;
2182                 for (i = 0; i < ret; i++) {
2183                         root_objectid = gang[i]->root_key.objectid;
2184                         ret = btrfs_find_dead_roots(fs_info->tree_root,
2185                                                     root_objectid, gang[i]);
2186                         BUG_ON(ret);
2187                         btrfs_orphan_cleanup(gang[i]);
2188                 }
2189                 root_objectid++;
2190         }
2191         return 0;
2192 }
2193
2194 int btrfs_commit_super(struct btrfs_root *root)
2195 {
2196         struct btrfs_trans_handle *trans;
2197         int ret;
2198
2199         mutex_lock(&root->fs_info->cleaner_mutex);
2200         btrfs_clean_old_snapshots(root);
2201         mutex_unlock(&root->fs_info->cleaner_mutex);
2202         trans = btrfs_start_transaction(root, 1);
2203         ret = btrfs_commit_transaction(trans, root);
2204         BUG_ON(ret);
2205         /* run commit again to drop the original snapshot */
2206         trans = btrfs_start_transaction(root, 1);
2207         btrfs_commit_transaction(trans, root);
2208         ret = btrfs_write_and_wait_transaction(NULL, root);
2209         BUG_ON(ret);
2210
2211         ret = write_ctree_super(NULL, root, 0);
2212         return ret;
2213 }
2214
2215 int close_ctree(struct btrfs_root *root)
2216 {
2217         struct btrfs_fs_info *fs_info = root->fs_info;
2218         int ret;
2219
2220         fs_info->closing = 1;
2221         smp_mb();
2222
2223         kthread_stop(root->fs_info->transaction_kthread);
2224         kthread_stop(root->fs_info->cleaner_kthread);
2225
2226         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2227                 ret =  btrfs_commit_super(root);
2228                 if (ret) {
2229                         printk("btrfs: commit super returns %d\n", ret);
2230                 }
2231         }
2232
2233         if (fs_info->delalloc_bytes) {
2234                 printk("btrfs: at unmount delalloc count %Lu\n",
2235                        fs_info->delalloc_bytes);
2236         }
2237         if (fs_info->total_ref_cache_size) {
2238                 printk("btrfs: at umount reference cache size %Lu\n",
2239                         fs_info->total_ref_cache_size);
2240         }
2241
2242         if (fs_info->extent_root->node)
2243                 free_extent_buffer(fs_info->extent_root->node);
2244
2245         if (fs_info->tree_root->node)
2246                 free_extent_buffer(fs_info->tree_root->node);
2247
2248         if (root->fs_info->chunk_root->node);
2249                 free_extent_buffer(root->fs_info->chunk_root->node);
2250
2251         if (root->fs_info->dev_root->node);
2252                 free_extent_buffer(root->fs_info->dev_root->node);
2253
2254         if (root->fs_info->csum_root->node);
2255                 free_extent_buffer(root->fs_info->csum_root->node);
2256
2257         btrfs_free_block_groups(root->fs_info);
2258
2259         del_fs_roots(fs_info);
2260
2261         iput(fs_info->btree_inode);
2262
2263         btrfs_stop_workers(&fs_info->fixup_workers);
2264         btrfs_stop_workers(&fs_info->delalloc_workers);
2265         btrfs_stop_workers(&fs_info->workers);
2266         btrfs_stop_workers(&fs_info->endio_workers);
2267         btrfs_stop_workers(&fs_info->endio_meta_workers);
2268         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2269         btrfs_stop_workers(&fs_info->endio_write_workers);
2270         btrfs_stop_workers(&fs_info->submit_workers);
2271
2272 #if 0
2273         while(!list_empty(&fs_info->hashers)) {
2274                 struct btrfs_hasher *hasher;
2275                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
2276                                     hashers);
2277                 list_del(&hasher->hashers);
2278                 crypto_free_hash(&fs_info->hash_tfm);
2279                 kfree(hasher);
2280         }
2281 #endif
2282         btrfs_close_devices(fs_info->fs_devices);
2283         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2284
2285         bdi_destroy(&fs_info->bdi);
2286
2287         kfree(fs_info->extent_root);
2288         kfree(fs_info->tree_root);
2289         kfree(fs_info->chunk_root);
2290         kfree(fs_info->dev_root);
2291         kfree(fs_info->csum_root);
2292         return 0;
2293 }
2294
2295 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2296 {
2297         int ret;
2298         struct inode *btree_inode = buf->first_page->mapping->host;
2299
2300         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2301         if (!ret)
2302                 return ret;
2303
2304         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2305                                     parent_transid);
2306         return !ret;
2307 }
2308
2309 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2310 {
2311         struct inode *btree_inode = buf->first_page->mapping->host;
2312         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2313                                           buf);
2314 }
2315
2316 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2317 {
2318         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2319         u64 transid = btrfs_header_generation(buf);
2320         struct inode *btree_inode = root->fs_info->btree_inode;
2321
2322         WARN_ON(!btrfs_tree_locked(buf));
2323         if (transid != root->fs_info->generation) {
2324                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
2325                         (unsigned long long)buf->start,
2326                         transid, root->fs_info->generation);
2327                 WARN_ON(1);
2328         }
2329         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
2330 }
2331
2332 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2333 {
2334         /*
2335          * looks as though older kernels can get into trouble with
2336          * this code, they end up stuck in balance_dirty_pages forever
2337          */
2338         struct extent_io_tree *tree;
2339         u64 num_dirty;
2340         u64 start = 0;
2341         unsigned long thresh = 32 * 1024 * 1024;
2342         tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
2343
2344         if (current_is_pdflush() || current->flags & PF_MEMALLOC)
2345                 return;
2346
2347         num_dirty = count_range_bits(tree, &start, (u64)-1,
2348                                      thresh, EXTENT_DIRTY);
2349         if (num_dirty > thresh) {
2350                 balance_dirty_pages_ratelimited_nr(
2351                                    root->fs_info->btree_inode->i_mapping, 1);
2352         }
2353         return;
2354 }
2355
2356 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2357 {
2358         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2359         int ret;
2360         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2361         if (ret == 0) {
2362                 buf->flags |= EXTENT_UPTODATE;
2363         }
2364         return ret;
2365 }
2366
2367 int btree_lock_page_hook(struct page *page)
2368 {
2369         struct inode *inode = page->mapping->host;
2370         struct btrfs_root *root = BTRFS_I(inode)->root;
2371         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2372         struct extent_buffer *eb;
2373         unsigned long len;
2374         u64 bytenr = page_offset(page);
2375
2376         if (page->private == EXTENT_PAGE_PRIVATE)
2377                 goto out;
2378
2379         len = page->private >> 2;
2380         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2381         if (!eb)
2382                 goto out;
2383
2384         btrfs_tree_lock(eb);
2385         spin_lock(&root->fs_info->hash_lock);
2386         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2387         spin_unlock(&root->fs_info->hash_lock);
2388         btrfs_tree_unlock(eb);
2389         free_extent_buffer(eb);
2390 out:
2391         lock_page(page);
2392         return 0;
2393 }
2394
2395 static struct extent_io_ops btree_extent_io_ops = {
2396         .write_cache_pages_lock_hook = btree_lock_page_hook,
2397         .readpage_end_io_hook = btree_readpage_end_io_hook,
2398         .submit_bio_hook = btree_submit_bio_hook,
2399         /* note we're sharing with inode.c for the merge bio hook */
2400         .merge_bio_hook = btrfs_merge_bio_hook,
2401 };