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         char result[BTRFS_CRC32_SIZE];
 180         unsigned long len;
 181         unsigned long cur_len;
 182         unsigned long offset = BTRFS_CSUM_SIZE;
 183         char *map_token = NULL;
 184         char *kaddr;
 185         unsigned long map_start;
 186         unsigned long map_len;
 187         int err;
 188         u32 crc = ~(u32)0;
 189
 190         len = buf->len - offset;
 191         while(len > 0) {
 192                 err = map_private_extent_buffer(buf, offset, 32,
 193                                         &map_token, &kaddr,
 194                                         &map_start, &map_len, KM_USER0);
 195                 if (err) {
 196                         printk("failed to map extent buffer! %lu\n",
 197                                offset);
 198                         return 1;
 199                 }
 200                 cur_len = min(len, map_len - (offset - map_start));
 201                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 202                                       crc, cur_len);
 203                 len -= cur_len;
 204                 offset += cur_len;
 205                 unmap_extent_buffer(buf, map_token, KM_USER0);
 206         }
 207         btrfs_csum_final(crc, result);
 208
 209         if (verify) {
 210                 /* FIXME, this is not good */
 211                 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
 212                         u32 val;
 213                         u32 found = 0;
 214                         memcpy(&found, result, BTRFS_CRC32_SIZE);
 215
 216                         read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
 217                         printk("btrfs: %s checksum verify failed on %llu "
 218                                "wanted %X found %X level %d\n",
 219                                root->fs_info->sb->s_id,
 220                                buf->start, val, found, btrfs_header_level(buf));
 221                         return 1;
 222                 }
 223         } else {
 224                 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
 225         }
 226         return 0;
 227 }
 228
 229 /*
 230  * we can't consider a given block up to date unless the transid of the
 231  * block matches the transid in the parent node's pointer.  This is how we
 232  * detect blocks that either didn't get written at all or got written
 233  * in the wrong place.
 234  */
 235 static int verify_parent_transid(struct extent_io_tree *io_tree,
 236                                  struct extent_buffer *eb, u64 parent_transid)
 237 {
 238         int ret;
 239
 240         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 241                 return 0;
 242
 243         lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
 244         if (extent_buffer_uptodate(io_tree, eb) &&
 245             btrfs_header_generation(eb) == parent_transid) {
 246                 ret = 0;
 247                 goto out;
 248         }
 249         printk("parent transid verify failed on %llu wanted %llu found %llu\n",
 250                (unsigned long long)eb->start,
 251                (unsigned long long)parent_transid,
 252                (unsigned long long)btrfs_header_generation(eb));
 253         ret = 1;
 254         clear_extent_buffer_uptodate(io_tree, eb);
 255 out:
 256         unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
 257                       GFP_NOFS);
 258         return ret;
 259 }
 260
 261 /*
 262  * helper to read a given tree block, doing retries as required when
 263  * the checksums don't match and we have alternate mirrors to try.
 264  */
 265 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 266                                           struct extent_buffer *eb,
 267                                           u64 start, u64 parent_transid)
 268 {
 269         struct extent_io_tree *io_tree;
 270         int ret;
 271         int num_copies = 0;
 272         int mirror_num = 0;
 273
 274         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 275         while (1) {
 276                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 277                                                btree_get_extent, mirror_num);
 278                 if (!ret &&
 279                     !verify_parent_transid(io_tree, eb, parent_transid))
 280                         return ret;
 281 printk("read extent buffer pages failed with ret %d mirror no %d\n", ret, mirror_num);
 282                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 283                                               eb->start, eb->len);
 284                 if (num_copies == 1)
 285                         return ret;
 286
 287                 mirror_num++;
 288                 if (mirror_num > num_copies)
 289                         return ret;
 290         }
 291         return -EIO;
 292 }
 293
 294 /*
 295  * checksum a dirty tree block before IO.  This has extra checks to make
 296  * sure we only fill in the checksum field in the first page of a multi-page block
 297  */
 298 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 299 {
 300         struct extent_io_tree *tree;
 301         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 302         u64 found_start;
 303         int found_level;
 304         unsigned long len;
 305         struct extent_buffer *eb;
 306         int ret;
 307
 308         tree = &BTRFS_I(page->mapping->host)->io_tree;
 309
 310         if (page->private == EXTENT_PAGE_PRIVATE)
 311                 goto out;
 312         if (!page->private)
 313                 goto out;
 314         len = page->private >> 2;
 315         if (len == 0) {
 316                 WARN_ON(1);
 317         }
 318         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 319         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 320                                              btrfs_header_generation(eb));
 321         BUG_ON(ret);
 322         found_start = btrfs_header_bytenr(eb);
 323         if (found_start != start) {
 324                 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
 325                        start, found_start, len);
 326                 WARN_ON(1);
 327                 goto err;
 328         }
 329         if (eb->first_page != page) {
 330                 printk("bad first page %lu %lu\n", eb->first_page->index,
 331                        page->index);
 332                 WARN_ON(1);
 333                 goto err;
 334         }
 335         if (!PageUptodate(page)) {
 336                 printk("csum not up to date page %lu\n", page->index);
 337                 WARN_ON(1);
 338                 goto err;
 339         }
 340         found_level = btrfs_header_level(eb);
 341
 342         csum_tree_block(root, eb, 0);
 343 err:
 344         free_extent_buffer(eb);
 345 out:
 346         return 0;
 347 }
 348
 349 static int check_tree_block_fsid(struct btrfs_root *root,
 350                                  struct extent_buffer *eb)
 351 {
 352         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 353         u8 fsid[BTRFS_UUID_SIZE];
 354         int ret = 1;
 355
 356         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
 357                            BTRFS_FSID_SIZE);
 358         while (fs_devices) {
 359                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 360                         ret = 0;
 361                         break;
 362                 }
 363                 fs_devices = fs_devices->seed;
 364         }
 365         return ret;
 366 }
 367
 368 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 369                                struct extent_state *state)
 370 {
 371         struct extent_io_tree *tree;
 372         u64 found_start;
 373         int found_level;
 374         unsigned long len;
 375         struct extent_buffer *eb;
 376         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 377         int ret = 0;
 378
 379         tree = &BTRFS_I(page->mapping->host)->io_tree;
 380         if (page->private == EXTENT_PAGE_PRIVATE)
 381                 goto out;
 382         if (!page->private)
 383                 goto out;
 384         len = page->private >> 2;
 385         if (len == 0) {
 386                 WARN_ON(1);
 387         }
 388         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 389
 390         found_start = btrfs_header_bytenr(eb);
 391         if (found_start != start) {
 392                 printk("bad tree block start %llu %llu\n",
 393                        (unsigned long long)found_start,
 394                        (unsigned long long)eb->start);
 395                 ret = -EIO;
 396                 goto err;
 397         }
 398         if (eb->first_page != page) {
 399                 printk("bad first page %lu %lu\n", eb->first_page->index,
 400                        page->index);
 401                 WARN_ON(1);
 402                 ret = -EIO;
 403                 goto err;
 404         }
 405         if (check_tree_block_fsid(root, eb)) {
 406                 printk("bad fsid on block %Lu\n", eb->start);
 407                 ret = -EIO;
 408                 goto err;
 409         }
 410         found_level = btrfs_header_level(eb);
 411
 412         ret = csum_tree_block(root, eb, 1);
 413         if (ret)
 414                 ret = -EIO;
 415
 416         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 417         end = eb->start + end - 1;
 418 err:
 419         free_extent_buffer(eb);
 420 out:
 421         return ret;
 422 }
 423
 424 static void end_workqueue_bio(struct bio *bio, int err)
 425 {
 426         struct end_io_wq *end_io_wq = bio->bi_private;
 427         struct btrfs_fs_info *fs_info;
 428
 429         fs_info = end_io_wq->info;
 430         end_io_wq->error = err;
 431         end_io_wq->work.func = end_workqueue_fn;
 432         end_io_wq->work.flags = 0;
 433         if (bio->bi_rw & (1 << BIO_RW))
 434                 btrfs_queue_worker(&fs_info->endio_write_workers,
 435                                    &end_io_wq->work);
 436         else
 437                 btrfs_queue_worker(&fs_info->endio_workers, &end_io_wq->work);
 438 }
 439
 440 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 441                         int metadata)
 442 {
 443         struct end_io_wq *end_io_wq;
 444         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 445         if (!end_io_wq)
 446                 return -ENOMEM;
 447
 448         end_io_wq->private = bio->bi_private;
 449         end_io_wq->end_io = bio->bi_end_io;
 450         end_io_wq->info = info;
 451         end_io_wq->error = 0;
 452         end_io_wq->bio = bio;
 453         end_io_wq->metadata = metadata;
 454
 455         bio->bi_private = end_io_wq;
 456         bio->bi_end_io = end_workqueue_bio;
 457         return 0;
 458 }
 459
 460 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 461 {
 462         unsigned long limit = min_t(unsigned long,
 463                                     info->workers.max_workers,
 464                                     info->fs_devices->open_devices);
 465         return 256 * limit;
 466 }
 467
 468 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
 469 {
 470         return atomic_read(&info->nr_async_bios) >
 471                 btrfs_async_submit_limit(info);
 472 }
 473
 474 static void run_one_async_start(struct btrfs_work *work)
 475 {
 476         struct btrfs_fs_info *fs_info;
 477         struct async_submit_bio *async;
 478
 479         async = container_of(work, struct  async_submit_bio, work);
 480         fs_info = BTRFS_I(async->inode)->root->fs_info;
 481         async->submit_bio_start(async->inode, async->rw, async->bio,
 482                                async->mirror_num, async->bio_flags);
 483 }
 484
 485 static void run_one_async_done(struct btrfs_work *work)
 486 {
 487         struct btrfs_fs_info *fs_info;
 488         struct async_submit_bio *async;
 489         int limit;
 490
 491         async = container_of(work, struct  async_submit_bio, work);
 492         fs_info = BTRFS_I(async->inode)->root->fs_info;
 493
 494         limit = btrfs_async_submit_limit(fs_info);
 495         limit = limit * 2 / 3;
 496
 497         atomic_dec(&fs_info->nr_async_submits);
 498
 499         if (atomic_read(&fs_info->nr_async_submits) < limit &&
 500             waitqueue_active(&fs_info->async_submit_wait))
 501                 wake_up(&fs_info->async_submit_wait);
 502
 503         async->submit_bio_done(async->inode, async->rw, async->bio,
 504                                async->mirror_num, async->bio_flags);
 505 }
 506
 507 static void run_one_async_free(struct btrfs_work *work)
 508 {
 509         struct async_submit_bio *async;
 510
 511         async = container_of(work, struct  async_submit_bio, work);
 512         kfree(async);
 513 }
 514
 515 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 516                         int rw, struct bio *bio, int mirror_num,
 517                         unsigned long bio_flags,
 518                         extent_submit_bio_hook_t *submit_bio_start,
 519                         extent_submit_bio_hook_t *submit_bio_done)
 520 {
 521         struct async_submit_bio *async;
 522
 523         async = kmalloc(sizeof(*async), GFP_NOFS);
 524         if (!async)
 525                 return -ENOMEM;
 526
 527         async->inode = inode;
 528         async->rw = rw;
 529         async->bio = bio;
 530         async->mirror_num = mirror_num;
 531         async->submit_bio_start = submit_bio_start;
 532         async->submit_bio_done = submit_bio_done;
 533
 534         async->work.func = run_one_async_start;
 535         async->work.ordered_func = run_one_async_done;
 536         async->work.ordered_free = run_one_async_free;
 537
 538         async->work.flags = 0;
 539         async->bio_flags = bio_flags;
 540
 541         atomic_inc(&fs_info->nr_async_submits);
 542         btrfs_queue_worker(&fs_info->workers, &async->work);
 543 #if 0
 544         int limit = btrfs_async_submit_limit(fs_info);
 545         if (atomic_read(&fs_info->nr_async_submits) > limit) {
 546                 wait_event_timeout(fs_info->async_submit_wait,
 547                            (atomic_read(&fs_info->nr_async_submits) < limit),
 548                            HZ/10);
 549
 550                 wait_event_timeout(fs_info->async_submit_wait,
 551                            (atomic_read(&fs_info->nr_async_bios) < limit),
 552                            HZ/10);
 553         }
 554 #endif
 555         while(atomic_read(&fs_info->async_submit_draining) &&
 556               atomic_read(&fs_info->nr_async_submits)) {
 557                 wait_event(fs_info->async_submit_wait,
 558                            (atomic_read(&fs_info->nr_async_submits) == 0));
 559         }
 560
 561         return 0;
 562 }
 563
 564 static int btree_csum_one_bio(struct bio *bio)
 565 {
 566         struct bio_vec *bvec = bio->bi_io_vec;
 567         int bio_index = 0;
 568         struct btrfs_root *root;
 569
 570         WARN_ON(bio->bi_vcnt <= 0);
 571         while(bio_index < bio->bi_vcnt) {
 572                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 573                 csum_dirty_buffer(root, bvec->bv_page);
 574                 bio_index++;
 575                 bvec++;
 576         }
 577         return 0;
 578 }
 579
 580 static int __btree_submit_bio_start(struct inode *inode, int rw,
 581                                     struct bio *bio, int mirror_num,
 582                                     unsigned long bio_flags)
 583 {
 584         /*
 585          * when we're called for a write, we're already in the async
 586          * submission context.  Just jump into btrfs_map_bio
 587          */
 588         btree_csum_one_bio(bio);
 589         return 0;
 590 }
 591
 592 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 593                                  int mirror_num, unsigned long bio_flags)
 594 {
 595         /*
 596          * when we're called for a write, we're already in the async
 597          * submission context.  Just jump into btrfs_map_bio
 598          */
 599         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 600 }
 601
 602 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 603                                  int mirror_num, unsigned long bio_flags)
 604 {
 605         /*
 606          * kthread helpers are used to submit writes so that checksumming
 607          * can happen in parallel across all CPUs
 608          */
 609         if (!(rw & (1 << BIO_RW))) {
 610                 int ret;
 611                 /*
 612                  * called for a read, do the setup so that checksum validation
 613                  * can happen in the async kernel threads
 614                  */
 615                 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 616                                           bio, 1);
 617                 BUG_ON(ret);
 618
 619                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 620                                      mirror_num, 0);
 621         }
 622         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 623                                    inode, rw, bio, mirror_num, 0,
 624                                    __btree_submit_bio_start,
 625                                    __btree_submit_bio_done);
 626 }
 627
 628 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 629 {
 630         struct extent_io_tree *tree;
 631         tree = &BTRFS_I(page->mapping->host)->io_tree;
 632
 633         if (current->flags & PF_MEMALLOC) {
 634                 redirty_page_for_writepage(wbc, page);
 635                 unlock_page(page);
 636                 return 0;
 637         }
 638         return extent_write_full_page(tree, page, btree_get_extent, wbc);
 639 }
 640
 641 static int btree_writepages(struct address_space *mapping,
 642                             struct writeback_control *wbc)
 643 {
 644         struct extent_io_tree *tree;
 645         tree = &BTRFS_I(mapping->host)->io_tree;
 646         if (wbc->sync_mode == WB_SYNC_NONE) {
 647                 u64 num_dirty;
 648                 u64 start = 0;
 649                 unsigned long thresh = 32 * 1024 * 1024;
 650
 651                 if (wbc->for_kupdate)
 652                         return 0;
 653
 654                 num_dirty = count_range_bits(tree, &start, (u64)-1,
 655                                              thresh, EXTENT_DIRTY);
 656                 if (num_dirty < thresh) {
 657                         return 0;
 658                 }
 659         }
 660         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 661 }
 662
 663 static int btree_readpage(struct file *file, struct page *page)
 664 {
 665         struct extent_io_tree *tree;
 666         tree = &BTRFS_I(page->mapping->host)->io_tree;
 667         return extent_read_full_page(tree, page, btree_get_extent);
 668 }
 669
 670 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 671 {
 672         struct extent_io_tree *tree;
 673         struct extent_map_tree *map;
 674         int ret;
 675
 676         if (PageWriteback(page) || PageDirty(page))
 677             return 0;
 678
 679         tree = &BTRFS_I(page->mapping->host)->io_tree;
 680         map = &BTRFS_I(page->mapping->host)->extent_tree;
 681
 682         ret = try_release_extent_state(map, tree, page, gfp_flags);
 683         if (!ret) {
 684                 return 0;
 685         }
 686
 687         ret = try_release_extent_buffer(tree, page);
 688         if (ret == 1) {
 689                 ClearPagePrivate(page);
 690                 set_page_private(page, 0);
 691                 page_cache_release(page);
 692         }
 693
 694         return ret;
 695 }
 696
 697 static void btree_invalidatepage(struct page *page, unsigned long offset)
 698 {
 699         struct extent_io_tree *tree;
 700         tree = &BTRFS_I(page->mapping->host)->io_tree;
 701         extent_invalidatepage(tree, page, offset);
 702         btree_releasepage(page, GFP_NOFS);
 703         if (PagePrivate(page)) {
 704                 printk("warning page private not zero on page %Lu\n",
 705                        page_offset(page));
 706                 ClearPagePrivate(page);
 707                 set_page_private(page, 0);
 708                 page_cache_release(page);
 709         }
 710 }
 711
 712 #if 0
 713 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 714 {
 715         struct buffer_head *bh;
 716         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 717         struct buffer_head *head;
 718         if (!page_has_buffers(page)) {
 719                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 720                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 721         }
 722         head = page_buffers(page);
 723         bh = head;
 724         do {
 725                 if (buffer_dirty(bh))
 726                         csum_tree_block(root, bh, 0);
 727                 bh = bh->b_this_page;
 728         } while (bh != head);
 729         return block_write_full_page(page, btree_get_block, wbc);
 730 }
 731 #endif
 732
 733 static struct address_space_operations btree_aops = {
 734         .readpage       = btree_readpage,
 735         .writepage      = btree_writepage,
 736         .writepages     = btree_writepages,
 737         .releasepage    = btree_releasepage,
 738         .invalidatepage = btree_invalidatepage,
 739         .sync_page      = block_sync_page,
 740 };
 741
 742 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 743                          u64 parent_transid)
 744 {
 745         struct extent_buffer *buf = NULL;
 746         struct inode *btree_inode = root->fs_info->btree_inode;
 747         int ret = 0;
 748
 749         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 750         if (!buf)
 751                 return 0;
 752         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 753                                  buf, 0, 0, btree_get_extent, 0);
 754         free_extent_buffer(buf);
 755         return ret;
 756 }
 757
 758 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 759                                             u64 bytenr, u32 blocksize)
 760 {
 761         struct inode *btree_inode = root->fs_info->btree_inode;
 762         struct extent_buffer *eb;
 763         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 764                                 bytenr, blocksize, GFP_NOFS);
 765         return eb;
 766 }
 767
 768 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 769                                                  u64 bytenr, u32 blocksize)
 770 {
 771         struct inode *btree_inode = root->fs_info->btree_inode;
 772         struct extent_buffer *eb;
 773
 774         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 775                                  bytenr, blocksize, NULL, GFP_NOFS);
 776         return eb;
 777 }
 778
 779
 780 int btrfs_write_tree_block(struct extent_buffer *buf)
 781 {
 782         return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
 783                                       buf->start + buf->len - 1, WB_SYNC_ALL);
 784 }
 785
 786 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 787 {
 788         return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
 789                                   buf->start, buf->start + buf->len -1);
 790 }
 791
 792 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 793                                       u32 blocksize, u64 parent_transid)
 794 {
 795         struct extent_buffer *buf = NULL;
 796         struct inode *btree_inode = root->fs_info->btree_inode;
 797         struct extent_io_tree *io_tree;
 798         int ret;
 799
 800         io_tree = &BTRFS_I(btree_inode)->io_tree;
 801
 802         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 803         if (!buf)
 804                 return NULL;
 805
 806         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 807
 808         if (ret == 0) {
 809                 buf->flags |= EXTENT_UPTODATE;
 810         } else {
 811                 WARN_ON(1);
 812         }
 813         return buf;
 814
 815 }
 816
 817 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 818                      struct extent_buffer *buf)
 819 {
 820         struct inode *btree_inode = root->fs_info->btree_inode;
 821         if (btrfs_header_generation(buf) ==
 822             root->fs_info->running_transaction->transid) {
 823                 WARN_ON(!btrfs_tree_locked(buf));
 824                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 825                                           buf);
 826         }
 827         return 0;
 828 }
 829
 830 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 831                         u32 stripesize, struct btrfs_root *root,
 832                         struct btrfs_fs_info *fs_info,
 833                         u64 objectid)
 834 {
 835         root->node = NULL;
 836         root->commit_root = NULL;
 837         root->ref_tree = NULL;
 838         root->sectorsize = sectorsize;
 839         root->nodesize = nodesize;
 840         root->leafsize = leafsize;
 841         root->stripesize = stripesize;
 842         root->ref_cows = 0;
 843         root->track_dirty = 0;
 844
 845         root->fs_info = fs_info;
 846         root->objectid = objectid;
 847         root->last_trans = 0;
 848         root->highest_inode = 0;
 849         root->last_inode_alloc = 0;
 850         root->name = NULL;
 851         root->in_sysfs = 0;
 852
 853         INIT_LIST_HEAD(&root->dirty_list);
 854         INIT_LIST_HEAD(&root->orphan_list);
 855         INIT_LIST_HEAD(&root->dead_list);
 856         spin_lock_init(&root->node_lock);
 857         spin_lock_init(&root->list_lock);
 858         mutex_init(&root->objectid_mutex);
 859         mutex_init(&root->log_mutex);
 860         extent_io_tree_init(&root->dirty_log_pages,
 861                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 862
 863         btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
 864         root->ref_tree = &root->ref_tree_struct;
 865
 866         memset(&root->root_key, 0, sizeof(root->root_key));
 867         memset(&root->root_item, 0, sizeof(root->root_item));
 868         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 869         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 870         root->defrag_trans_start = fs_info->generation;
 871         init_completion(&root->kobj_unregister);
 872         root->defrag_running = 0;
 873         root->defrag_level = 0;
 874         root->root_key.objectid = objectid;
 875         root->anon_super.s_root = NULL;
 876         root->anon_super.s_dev = 0;
 877         INIT_LIST_HEAD(&root->anon_super.s_list);
 878         INIT_LIST_HEAD(&root->anon_super.s_instances);
 879         init_rwsem(&root->anon_super.s_umount);
 880
 881         return 0;
 882 }
 883
 884 static int find_and_setup_root(struct btrfs_root *tree_root,
 885                                struct btrfs_fs_info *fs_info,
 886                                u64 objectid,
 887                                struct btrfs_root *root)
 888 {
 889         int ret;
 890         u32 blocksize;
 891         u64 generation;
 892
 893         __setup_root(tree_root->nodesize, tree_root->leafsize,
 894                      tree_root->sectorsize, tree_root->stripesize,
 895                      root, fs_info, objectid);
 896         ret = btrfs_find_last_root(tree_root, objectid,
 897                                    &root->root_item, &root->root_key);
 898         BUG_ON(ret);
 899
 900         generation = btrfs_root_generation(&root->root_item);
 901         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 902         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 903                                      blocksize, generation);
 904         BUG_ON(!root->node);
 905         return 0;
 906 }
 907
 908 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
 909                              struct btrfs_fs_info *fs_info)
 910 {
 911         struct extent_buffer *eb;
 912         struct btrfs_root *log_root_tree = fs_info->log_root_tree;
 913         u64 start = 0;
 914         u64 end = 0;
 915         int ret;
 916
 917         if (!log_root_tree)
 918                 return 0;
 919
 920         while(1) {
 921                 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
 922                                     0, &start, &end, EXTENT_DIRTY);
 923                 if (ret)
 924                         break;
 925
 926                 clear_extent_dirty(&log_root_tree->dirty_log_pages,
 927                                    start, end, GFP_NOFS);
 928         }
 929         eb = fs_info->log_root_tree->node;
 930
 931         WARN_ON(btrfs_header_level(eb) != 0);
 932         WARN_ON(btrfs_header_nritems(eb) != 0);
 933
 934         ret = btrfs_free_reserved_extent(fs_info->tree_root,
 935                                 eb->start, eb->len);
 936         BUG_ON(ret);
 937
 938         free_extent_buffer(eb);
 939         kfree(fs_info->log_root_tree);
 940         fs_info->log_root_tree = NULL;
 941         return 0;
 942 }
 943
 944 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
 945                              struct btrfs_fs_info *fs_info)
 946 {
 947         struct btrfs_root *root;
 948         struct btrfs_root *tree_root = fs_info->tree_root;
 949
 950         root = kzalloc(sizeof(*root), GFP_NOFS);
 951         if (!root)
 952                 return -ENOMEM;
 953
 954         __setup_root(tree_root->nodesize, tree_root->leafsize,
 955                      tree_root->sectorsize, tree_root->stripesize,
 956                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
 957
 958         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
 959         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
 960         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
 961         root->ref_cows = 0;
 962
 963         root->node = btrfs_alloc_free_block(trans, root, root->leafsize,
 964                                             0, BTRFS_TREE_LOG_OBJECTID,
 965                                             trans->transid, 0, 0, 0);
 966
 967         btrfs_set_header_nritems(root->node, 0);
 968         btrfs_set_header_level(root->node, 0);
 969         btrfs_set_header_bytenr(root->node, root->node->start);
 970         btrfs_set_header_generation(root->node, trans->transid);
 971         btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
 972
 973         write_extent_buffer(root->node, root->fs_info->fsid,
 974                             (unsigned long)btrfs_header_fsid(root->node),
 975                             BTRFS_FSID_SIZE);
 976         btrfs_mark_buffer_dirty(root->node);
 977         btrfs_tree_unlock(root->node);
 978         fs_info->log_root_tree = root;
 979         return 0;
 980 }
 981
 982 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
 983                                                struct btrfs_key *location)
 984 {
 985         struct btrfs_root *root;
 986         struct btrfs_fs_info *fs_info = tree_root->fs_info;
 987         struct btrfs_path *path;
 988         struct extent_buffer *l;
 989         u64 highest_inode;
 990         u64 generation;
 991         u32 blocksize;
 992         int ret = 0;
 993
 994         root = kzalloc(sizeof(*root), GFP_NOFS);
 995         if (!root)
 996                 return ERR_PTR(-ENOMEM);
 997         if (location->offset == (u64)-1) {
 998                 ret = find_and_setup_root(tree_root, fs_info,
 999                                           location->objectid, root);
1000                 if (ret) {
1001                         kfree(root);
1002                         return ERR_PTR(ret);
1003                 }
1004                 goto insert;
1005         }
1006
1007         __setup_root(tree_root->nodesize, tree_root->leafsize,
1008                      tree_root->sectorsize, tree_root->stripesize,
1009                      root, fs_info, location->objectid);
1010
1011         path = btrfs_alloc_path();
1012         BUG_ON(!path);
1013         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1014         if (ret != 0) {
1015                 if (ret > 0)
1016                         ret = -ENOENT;
1017                 goto out;
1018         }
1019         l = path->nodes[0];
1020         read_extent_buffer(l, &root->root_item,
1021                btrfs_item_ptr_offset(l, path->slots[0]),
1022                sizeof(root->root_item));
1023         memcpy(&root->root_key, location, sizeof(*location));
1024         ret = 0;
1025 out:
1026         btrfs_release_path(root, path);
1027         btrfs_free_path(path);
1028         if (ret) {
1029                 kfree(root);
1030                 return ERR_PTR(ret);
1031         }
1032         generation = btrfs_root_generation(&root->root_item);
1033         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1034         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1035                                      blocksize, generation);
1036         BUG_ON(!root->node);
1037 insert:
1038         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1039                 root->ref_cows = 1;
1040                 ret = btrfs_find_highest_inode(root, &highest_inode);
1041                 if (ret == 0) {
1042                         root->highest_inode = highest_inode;
1043                         root->last_inode_alloc = highest_inode;
1044                 }
1045         }
1046         return root;
1047 }
1048
1049 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1050                                         u64 root_objectid)
1051 {
1052         struct btrfs_root *root;
1053
1054         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1055                 return fs_info->tree_root;
1056         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1057                 return fs_info->extent_root;
1058
1059         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1060                                  (unsigned long)root_objectid);
1061         return root;
1062 }
1063
1064 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1065                                               struct btrfs_key *location)
1066 {
1067         struct btrfs_root *root;
1068         int ret;
1069
1070         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1071                 return fs_info->tree_root;
1072         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1073                 return fs_info->extent_root;
1074         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1075                 return fs_info->chunk_root;
1076         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1077                 return fs_info->dev_root;
1078
1079         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1080                                  (unsigned long)location->objectid);
1081         if (root)
1082                 return root;
1083
1084         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1085         if (IS_ERR(root))
1086                 return root;
1087
1088         set_anon_super(&root->anon_super, NULL);
1089
1090         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1091                                 (unsigned long)root->root_key.objectid,
1092                                 root);
1093         if (ret) {
1094                 free_extent_buffer(root->node);
1095                 kfree(root);
1096                 return ERR_PTR(ret);
1097         }
1098         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
1099                 ret = btrfs_find_dead_roots(fs_info->tree_root,
1100                                             root->root_key.objectid, root);
1101                 BUG_ON(ret);
1102                 btrfs_orphan_cleanup(root);
1103         }
1104         return root;
1105 }
1106
1107 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1108                                       struct btrfs_key *location,
1109                                       const char *name, int namelen)
1110 {
1111         struct btrfs_root *root;
1112         int ret;
1113
1114         root = btrfs_read_fs_root_no_name(fs_info, location);
1115         if (!root)
1116                 return NULL;
1117
1118         if (root->in_sysfs)
1119                 return root;
1120
1121         ret = btrfs_set_root_name(root, name, namelen);
1122         if (ret) {
1123                 free_extent_buffer(root->node);
1124                 kfree(root);
1125                 return ERR_PTR(ret);
1126         }
1127 #if 0
1128         ret = btrfs_sysfs_add_root(root);
1129         if (ret) {
1130                 free_extent_buffer(root->node);
1131                 kfree(root->name);
1132                 kfree(root);
1133                 return ERR_PTR(ret);
1134         }
1135 #endif
1136         root->in_sysfs = 1;
1137         return root;
1138 }
1139 #if 0
1140 static int add_hasher(struct btrfs_fs_info *info, char *type) {
1141         struct btrfs_hasher *hasher;
1142
1143         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
1144         if (!hasher)
1145                 return -ENOMEM;
1146         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
1147         if (!hasher->hash_tfm) {
1148                 kfree(hasher);
1149                 return -EINVAL;
1150         }
1151         spin_lock(&info->hash_lock);
1152         list_add(&hasher->list, &info->hashers);
1153         spin_unlock(&info->hash_lock);
1154         return 0;
1155 }
1156 #endif
1157
1158 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1159 {
1160         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1161         int ret = 0;
1162         struct list_head *cur;
1163         struct btrfs_device *device;
1164         struct backing_dev_info *bdi;
1165 #if 0
1166         if ((bdi_bits & (1 << BDI_write_congested)) &&
1167             btrfs_congested_async(info, 0))
1168                 return 1;
1169 #endif
1170         list_for_each(cur, &info->fs_devices->devices) {
1171                 device = list_entry(cur, struct btrfs_device, dev_list);
1172                 if (!device->bdev)
1173                         continue;
1174                 bdi = blk_get_backing_dev_info(device->bdev);
1175                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1176                         ret = 1;
1177                         break;
1178                 }
1179         }
1180         return ret;
1181 }
1182
1183 /*
1184  * this unplugs every device on the box, and it is only used when page
1185  * is null
1186  */
1187 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1188 {
1189         struct list_head *cur;
1190         struct btrfs_device *device;
1191         struct btrfs_fs_info *info;
1192
1193         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1194         list_for_each(cur, &info->fs_devices->devices) {
1195                 device = list_entry(cur, struct btrfs_device, dev_list);
1196                 bdi = blk_get_backing_dev_info(device->bdev);
1197                 if (bdi->unplug_io_fn) {
1198                         bdi->unplug_io_fn(bdi, page);
1199                 }
1200         }
1201 }
1202
1203 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1204 {
1205         struct inode *inode;
1206         struct extent_map_tree *em_tree;
1207         struct extent_map *em;
1208         struct address_space *mapping;
1209         u64 offset;
1210
1211         /* the generic O_DIRECT read code does this */
1212         if (1 || !page) {
1213                 __unplug_io_fn(bdi, page);
1214                 return;
1215         }
1216
1217         /*
1218          * page->mapping may change at any time.  Get a consistent copy
1219          * and use that for everything below
1220          */
1221         smp_mb();
1222         mapping = page->mapping;
1223         if (!mapping)
1224                 return;
1225
1226         inode = mapping->host;
1227
1228         /*
1229          * don't do the expensive searching for a small number of
1230          * devices
1231          */
1232         if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1233                 __unplug_io_fn(bdi, page);
1234                 return;
1235         }
1236
1237         offset = page_offset(page);
1238
1239         em_tree = &BTRFS_I(inode)->extent_tree;
1240         spin_lock(&em_tree->lock);
1241         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1242         spin_unlock(&em_tree->lock);
1243         if (!em) {
1244                 __unplug_io_fn(bdi, page);
1245                 return;
1246         }
1247
1248         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1249                 free_extent_map(em);
1250                 __unplug_io_fn(bdi, page);
1251                 return;
1252         }
1253         offset = offset - em->start;
1254         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1255                           em->block_start + offset, page);
1256         free_extent_map(em);
1257 }
1258
1259 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1260 {
1261         bdi_init(bdi);
1262         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1263         bdi->state              = 0;
1264         bdi->capabilities       = default_backing_dev_info.capabilities;
1265         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1266         bdi->unplug_io_data     = info;
1267         bdi->congested_fn       = btrfs_congested_fn;
1268         bdi->congested_data     = info;
1269         return 0;
1270 }
1271
1272 static int bio_ready_for_csum(struct bio *bio)
1273 {
1274         u64 length = 0;
1275         u64 buf_len = 0;
1276         u64 start = 0;
1277         struct page *page;
1278         struct extent_io_tree *io_tree = NULL;
1279         struct btrfs_fs_info *info = NULL;
1280         struct bio_vec *bvec;
1281         int i;
1282         int ret;
1283
1284         bio_for_each_segment(bvec, bio, i) {
1285                 page = bvec->bv_page;
1286                 if (page->private == EXTENT_PAGE_PRIVATE) {
1287                         length += bvec->bv_len;
1288                         continue;
1289                 }
1290                 if (!page->private) {
1291                         length += bvec->bv_len;
1292                         continue;
1293                 }
1294                 length = bvec->bv_len;
1295                 buf_len = page->private >> 2;
1296                 start = page_offset(page) + bvec->bv_offset;
1297                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1298                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1299         }
1300         /* are we fully contained in this bio? */
1301         if (buf_len <= length)
1302                 return 1;
1303
1304         ret = extent_range_uptodate(io_tree, start + length,
1305                                     start + buf_len - 1);
1306         if (ret == 1)
1307                 return ret;
1308         return ret;
1309 }
1310
1311 /*
1312  * called by the kthread helper functions to finally call the bio end_io
1313  * functions.  This is where read checksum verification actually happens
1314  */
1315 static void end_workqueue_fn(struct btrfs_work *work)
1316 {
1317         struct bio *bio;
1318         struct end_io_wq *end_io_wq;
1319         struct btrfs_fs_info *fs_info;
1320         int error;
1321
1322         end_io_wq = container_of(work, struct end_io_wq, work);
1323         bio = end_io_wq->bio;
1324         fs_info = end_io_wq->info;
1325
1326         /* metadata bios are special because the whole tree block must
1327          * be checksummed at once.  This makes sure the entire block is in
1328          * ram and up to date before trying to verify things.  For
1329          * blocksize <= pagesize, it is basically a noop
1330          */
1331         if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1332                 btrfs_queue_worker(&fs_info->endio_workers,
1333                                    &end_io_wq->work);
1334                 return;
1335         }
1336         error = end_io_wq->error;
1337         bio->bi_private = end_io_wq->private;
1338         bio->bi_end_io = end_io_wq->end_io;
1339         kfree(end_io_wq);
1340         bio_endio(bio, error);
1341 }
1342
1343 static int cleaner_kthread(void *arg)
1344 {
1345         struct btrfs_root *root = arg;
1346
1347         do {
1348                 smp_mb();
1349                 if (root->fs_info->closing)
1350                         break;
1351
1352                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1353                 mutex_lock(&root->fs_info->cleaner_mutex);
1354                 btrfs_clean_old_snapshots(root);
1355                 mutex_unlock(&root->fs_info->cleaner_mutex);
1356
1357                 if (freezing(current)) {
1358                         refrigerator();
1359                 } else {
1360                         smp_mb();
1361                         if (root->fs_info->closing)
1362                                 break;
1363                         set_current_state(TASK_INTERRUPTIBLE);
1364                         schedule();
1365                         __set_current_state(TASK_RUNNING);
1366                 }
1367         } while (!kthread_should_stop());
1368         return 0;
1369 }
1370
1371 static int transaction_kthread(void *arg)
1372 {
1373         struct btrfs_root *root = arg;
1374         struct btrfs_trans_handle *trans;
1375         struct btrfs_transaction *cur;
1376         unsigned long now;
1377         unsigned long delay;
1378         int ret;
1379
1380         do {
1381                 smp_mb();
1382                 if (root->fs_info->closing)
1383                         break;
1384
1385                 delay = HZ * 30;
1386                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1387                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1388
1389                 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1390                         printk("btrfs: total reference cache size %Lu\n",
1391                                 root->fs_info->total_ref_cache_size);
1392                 }
1393
1394                 mutex_lock(&root->fs_info->trans_mutex);
1395                 cur = root->fs_info->running_transaction;
1396                 if (!cur) {
1397                         mutex_unlock(&root->fs_info->trans_mutex);
1398                         goto sleep;
1399                 }
1400
1401                 now = get_seconds();
1402                 if (now < cur->start_time || now - cur->start_time < 30) {
1403                         mutex_unlock(&root->fs_info->trans_mutex);
1404                         delay = HZ * 5;
1405                         goto sleep;
1406                 }
1407                 mutex_unlock(&root->fs_info->trans_mutex);
1408                 trans = btrfs_start_transaction(root, 1);
1409                 ret = btrfs_commit_transaction(trans, root);
1410 sleep:
1411                 wake_up_process(root->fs_info->cleaner_kthread);
1412                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1413
1414                 if (freezing(current)) {
1415                         refrigerator();
1416                 } else {
1417                         if (root->fs_info->closing)
1418                                 break;
1419                         set_current_state(TASK_INTERRUPTIBLE);
1420                         schedule_timeout(delay);
1421                         __set_current_state(TASK_RUNNING);
1422                 }
1423         } while (!kthread_should_stop());
1424         return 0;
1425 }
1426
1427 struct btrfs_root *open_ctree(struct super_block *sb,
1428                               struct btrfs_fs_devices *fs_devices,
1429                               char *options)
1430 {
1431         u32 sectorsize;
1432         u32 nodesize;
1433         u32 leafsize;
1434         u32 blocksize;
1435         u32 stripesize;
1436         u64 generation;
1437         struct btrfs_key location;
1438         struct buffer_head *bh;
1439         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1440                                                  GFP_NOFS);
1441         struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1442                                                GFP_NOFS);
1443         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1444                                                 GFP_NOFS);
1445         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1446                                                 GFP_NOFS);
1447         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1448                                               GFP_NOFS);
1449         struct btrfs_root *log_tree_root;
1450
1451         int ret;
1452         int err = -EINVAL;
1453
1454         struct btrfs_super_block *disk_super;
1455
1456         if (!extent_root || !tree_root || !fs_info ||
1457             !chunk_root || !dev_root) {
1458                 err = -ENOMEM;
1459                 goto fail;
1460         }
1461         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1462         INIT_LIST_HEAD(&fs_info->trans_list);
1463         INIT_LIST_HEAD(&fs_info->dead_roots);
1464         INIT_LIST_HEAD(&fs_info->hashers);
1465         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1466         spin_lock_init(&fs_info->hash_lock);
1467         spin_lock_init(&fs_info->delalloc_lock);
1468         spin_lock_init(&fs_info->new_trans_lock);
1469         spin_lock_init(&fs_info->ref_cache_lock);
1470
1471         init_completion(&fs_info->kobj_unregister);
1472         fs_info->tree_root = tree_root;
1473         fs_info->extent_root = extent_root;
1474         fs_info->chunk_root = chunk_root;
1475         fs_info->dev_root = dev_root;
1476         fs_info->fs_devices = fs_devices;
1477         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1478         INIT_LIST_HEAD(&fs_info->space_info);
1479         btrfs_mapping_init(&fs_info->mapping_tree);
1480         atomic_set(&fs_info->nr_async_submits, 0);
1481         atomic_set(&fs_info->async_delalloc_pages, 0);
1482         atomic_set(&fs_info->async_submit_draining, 0);
1483         atomic_set(&fs_info->nr_async_bios, 0);
1484         atomic_set(&fs_info->throttles, 0);
1485         atomic_set(&fs_info->throttle_gen, 0);
1486         fs_info->sb = sb;
1487         fs_info->max_extent = (u64)-1;
1488         fs_info->max_inline = 8192 * 1024;
1489         setup_bdi(fs_info, &fs_info->bdi);
1490         fs_info->btree_inode = new_inode(sb);
1491         fs_info->btree_inode->i_ino = 1;
1492         fs_info->btree_inode->i_nlink = 1;
1493
1494         fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1495
1496         INIT_LIST_HEAD(&fs_info->ordered_extents);
1497         spin_lock_init(&fs_info->ordered_extent_lock);
1498
1499         sb->s_blocksize = 4096;
1500         sb->s_blocksize_bits = blksize_bits(4096);
1501
1502         /*
1503          * we set the i_size on the btree inode to the max possible int.
1504          * the real end of the address space is determined by all of
1505          * the devices in the system
1506          */
1507         fs_info->btree_inode->i_size = OFFSET_MAX;
1508         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1509         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1510
1511         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1512                              fs_info->btree_inode->i_mapping,
1513                              GFP_NOFS);
1514         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1515                              GFP_NOFS);
1516
1517         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1518
1519         spin_lock_init(&fs_info->block_group_cache_lock);
1520         fs_info->block_group_cache_tree.rb_node = NULL;
1521
1522         extent_io_tree_init(&fs_info->pinned_extents,
1523                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1524         extent_io_tree_init(&fs_info->pending_del,
1525                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1526         extent_io_tree_init(&fs_info->extent_ins,
1527                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1528         fs_info->do_barriers = 1;
1529
1530         INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
1531         btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
1532         btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
1533
1534         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1535         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1536                sizeof(struct btrfs_key));
1537         insert_inode_hash(fs_info->btree_inode);
1538
1539         mutex_init(&fs_info->trans_mutex);
1540         mutex_init(&fs_info->tree_log_mutex);
1541         mutex_init(&fs_info->drop_mutex);
1542         mutex_init(&fs_info->extent_ins_mutex);
1543         mutex_init(&fs_info->pinned_mutex);
1544         mutex_init(&fs_info->chunk_mutex);
1545         mutex_init(&fs_info->transaction_kthread_mutex);
1546         mutex_init(&fs_info->cleaner_mutex);
1547         mutex_init(&fs_info->volume_mutex);
1548         mutex_init(&fs_info->tree_reloc_mutex);
1549         init_waitqueue_head(&fs_info->transaction_throttle);
1550         init_waitqueue_head(&fs_info->transaction_wait);
1551         init_waitqueue_head(&fs_info->async_submit_wait);
1552         init_waitqueue_head(&fs_info->tree_log_wait);
1553         atomic_set(&fs_info->tree_log_commit, 0);
1554         atomic_set(&fs_info->tree_log_writers, 0);
1555         fs_info->tree_log_transid = 0;
1556
1557 #if 0
1558         ret = add_hasher(fs_info, "crc32c");
1559         if (ret) {
1560                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1561                 err = -ENOMEM;
1562                 goto fail_iput;
1563         }
1564 #endif
1565         __setup_root(4096, 4096, 4096, 4096, tree_root,
1566                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1567
1568
1569         bh = __bread(fs_devices->latest_bdev,
1570                      BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1571         if (!bh)
1572                 goto fail_iput;
1573
1574         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1575         brelse(bh);
1576
1577         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1578
1579         disk_super = &fs_info->super_copy;
1580         if (!btrfs_super_root(disk_super))
1581                 goto fail_sb_buffer;
1582
1583         ret = btrfs_parse_options(tree_root, options);
1584         if (ret) {
1585                 err = ret;
1586                 goto fail_sb_buffer;
1587         }
1588
1589         /*
1590          * we need to start all the end_io workers up front because the
1591          * queue work function gets called at interrupt time, and so it
1592          * cannot dynamically grow.
1593          */
1594         btrfs_init_workers(&fs_info->workers, "worker",
1595                            fs_info->thread_pool_size);
1596
1597         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1598                            fs_info->thread_pool_size);
1599
1600         btrfs_init_workers(&fs_info->submit_workers, "submit",
1601                            min_t(u64, fs_devices->num_devices,
1602                            fs_info->thread_pool_size));
1603
1604         /* a higher idle thresh on the submit workers makes it much more
1605          * likely that bios will be send down in a sane order to the
1606          * devices
1607          */
1608         fs_info->submit_workers.idle_thresh = 64;
1609
1610         fs_info->workers.idle_thresh = 16;
1611         fs_info->workers.ordered = 1;
1612
1613         fs_info->delalloc_workers.idle_thresh = 2;
1614         fs_info->delalloc_workers.ordered = 1;
1615
1616         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1617         btrfs_init_workers(&fs_info->endio_workers, "endio",
1618                            fs_info->thread_pool_size);
1619         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1620                            fs_info->thread_pool_size);
1621
1622         /*
1623          * endios are largely parallel and should have a very
1624          * low idle thresh
1625          */
1626         fs_info->endio_workers.idle_thresh = 4;
1627         fs_info->endio_write_workers.idle_thresh = 64;
1628
1629         btrfs_start_workers(&fs_info->workers, 1);
1630         btrfs_start_workers(&fs_info->submit_workers, 1);
1631         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1632         btrfs_start_workers(&fs_info->fixup_workers, 1);
1633         btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1634         btrfs_start_workers(&fs_info->endio_write_workers,
1635                             fs_info->thread_pool_size);
1636
1637         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1638         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1639                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1640
1641         nodesize = btrfs_super_nodesize(disk_super);
1642         leafsize = btrfs_super_leafsize(disk_super);
1643         sectorsize = btrfs_super_sectorsize(disk_super);
1644         stripesize = btrfs_super_stripesize(disk_super);
1645         tree_root->nodesize = nodesize;
1646         tree_root->leafsize = leafsize;
1647         tree_root->sectorsize = sectorsize;
1648         tree_root->stripesize = stripesize;
1649
1650         sb->s_blocksize = sectorsize;
1651         sb->s_blocksize_bits = blksize_bits(sectorsize);
1652
1653         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1654                     sizeof(disk_super->magic))) {
1655                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1656                 goto fail_sb_buffer;
1657         }
1658
1659         mutex_lock(&fs_info->chunk_mutex);
1660         ret = btrfs_read_sys_array(tree_root);
1661         mutex_unlock(&fs_info->chunk_mutex);
1662         if (ret) {
1663                 printk("btrfs: failed to read the system array on %s\n",
1664                        sb->s_id);
1665                 goto fail_sys_array;
1666         }
1667
1668         blocksize = btrfs_level_size(tree_root,
1669                                      btrfs_super_chunk_root_level(disk_super));
1670         generation = btrfs_super_chunk_root_generation(disk_super);
1671
1672         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1673                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1674
1675         chunk_root->node = read_tree_block(chunk_root,
1676                                            btrfs_super_chunk_root(disk_super),
1677                                            blocksize, generation);
1678         BUG_ON(!chunk_root->node);
1679
1680         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1681                  (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1682                  BTRFS_UUID_SIZE);
1683
1684         mutex_lock(&fs_info->chunk_mutex);
1685         ret = btrfs_read_chunk_tree(chunk_root);
1686         mutex_unlock(&fs_info->chunk_mutex);
1687         if (ret) {
1688                 printk("btrfs: failed to read chunk tree on %s\n", sb->s_id);
1689                 goto fail_chunk_root;
1690         }
1691
1692         btrfs_close_extra_devices(fs_devices);
1693
1694         blocksize = btrfs_level_size(tree_root,
1695                                      btrfs_super_root_level(disk_super));
1696         generation = btrfs_super_generation(disk_super);
1697
1698         tree_root->node = read_tree_block(tree_root,
1699                                           btrfs_super_root(disk_super),
1700                                           blocksize, generation);
1701         if (!tree_root->node)
1702                 goto fail_chunk_root;
1703
1704
1705         ret = find_and_setup_root(tree_root, fs_info,
1706                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1707         if (ret)
1708                 goto fail_tree_root;
1709         extent_root->track_dirty = 1;
1710
1711         ret = find_and_setup_root(tree_root, fs_info,
1712                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1713         dev_root->track_dirty = 1;
1714
1715         if (ret)
1716                 goto fail_extent_root;
1717
1718         btrfs_read_block_groups(extent_root);
1719
1720         fs_info->generation = generation + 1;
1721         fs_info->last_trans_committed = generation;
1722         fs_info->data_alloc_profile = (u64)-1;
1723         fs_info->metadata_alloc_profile = (u64)-1;
1724         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1725         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1726                                                "btrfs-cleaner");
1727         if (!fs_info->cleaner_kthread)
1728                 goto fail_extent_root;
1729
1730         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1731                                                    tree_root,
1732                                                    "btrfs-transaction");
1733         if (!fs_info->transaction_kthread)
1734                 goto fail_cleaner;
1735
1736         if (btrfs_super_log_root(disk_super) != 0) {
1737                 u64 bytenr = btrfs_super_log_root(disk_super);
1738
1739                 if (fs_devices->rw_devices == 0) {
1740                         printk("Btrfs log replay required on RO media\n");
1741                         err = -EIO;
1742                         goto fail_trans_kthread;
1743                 }
1744                 blocksize =
1745                      btrfs_level_size(tree_root,
1746                                       btrfs_super_log_root_level(disk_super));
1747
1748                 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1749                                                       GFP_NOFS);
1750
1751                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1752                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1753
1754                 log_tree_root->node = read_tree_block(tree_root, bytenr,
1755                                                       blocksize,
1756                                                       generation + 1);
1757                 ret = btrfs_recover_log_trees(log_tree_root);
1758                 BUG_ON(ret);
1759
1760                 if (sb->s_flags & MS_RDONLY) {
1761                         ret =  btrfs_commit_super(tree_root);
1762                         BUG_ON(ret);
1763                 }
1764         }
1765
1766         if (!(sb->s_flags & MS_RDONLY)) {
1767                 ret = btrfs_cleanup_reloc_trees(tree_root);
1768                 BUG_ON(ret);
1769         }
1770
1771         location.objectid = BTRFS_FS_TREE_OBJECTID;
1772         location.type = BTRFS_ROOT_ITEM_KEY;
1773         location.offset = (u64)-1;
1774
1775         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1776         if (!fs_info->fs_root)
1777                 goto fail_trans_kthread;
1778         return tree_root;
1779
1780 fail_trans_kthread:
1781         kthread_stop(fs_info->transaction_kthread);
1782 fail_cleaner:
1783         kthread_stop(fs_info->cleaner_kthread);
1784
1785         /*
1786          * make sure we're done with the btree inode before we stop our
1787          * kthreads
1788          */
1789         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1790         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1791
1792 fail_extent_root:
1793         free_extent_buffer(extent_root->node);
1794 fail_tree_root:
1795         free_extent_buffer(tree_root->node);
1796 fail_chunk_root:
1797         free_extent_buffer(chunk_root->node);
1798 fail_sys_array:
1799         free_extent_buffer(dev_root->node);
1800 fail_sb_buffer:
1801         btrfs_stop_workers(&fs_info->fixup_workers);
1802         btrfs_stop_workers(&fs_info->delalloc_workers);
1803         btrfs_stop_workers(&fs_info->workers);
1804         btrfs_stop_workers(&fs_info->endio_workers);
1805         btrfs_stop_workers(&fs_info->endio_write_workers);
1806         btrfs_stop_workers(&fs_info->submit_workers);
1807 fail_iput:
1808         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1809         iput(fs_info->btree_inode);
1810 fail:
1811         btrfs_close_devices(fs_info->fs_devices);
1812         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1813
1814         kfree(extent_root);
1815         kfree(tree_root);
1816         bdi_destroy(&fs_info->bdi);
1817         kfree(fs_info);
1818         kfree(chunk_root);
1819         kfree(dev_root);
1820         return ERR_PTR(err);
1821 }
1822
1823 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1824 {
1825         char b[BDEVNAME_SIZE];
1826
1827         if (uptodate) {
1828                 set_buffer_uptodate(bh);
1829         } else {
1830                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1831                         printk(KERN_WARNING "lost page write due to "
1832                                         "I/O error on %s\n",
1833                                        bdevname(bh->b_bdev, b));
1834                 }
1835                 /* note, we dont' set_buffer_write_io_error because we have
1836                  * our own ways of dealing with the IO errors
1837                  */
1838                 clear_buffer_uptodate(bh);
1839         }
1840         unlock_buffer(bh);
1841         put_bh(bh);
1842 }
1843
1844 static int write_all_supers(struct btrfs_root *root)
1845 {
1846         struct list_head *cur;
1847         struct list_head *head = &root->fs_info->fs_devices->devices;
1848         struct btrfs_device *dev;
1849         struct btrfs_super_block *sb;
1850         struct btrfs_dev_item *dev_item;
1851         struct buffer_head *bh;
1852         int ret;
1853         int do_barriers;
1854         int max_errors;
1855         int total_errors = 0;
1856         u32 crc;
1857         u64 flags;
1858
1859         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1860         do_barriers = !btrfs_test_opt(root, NOBARRIER);
1861
1862         sb = &root->fs_info->super_for_commit;
1863         dev_item = &sb->dev_item;
1864         list_for_each(cur, head) {
1865                 dev = list_entry(cur, struct btrfs_device, dev_list);
1866                 if (!dev->bdev) {
1867                         total_errors++;
1868                         continue;
1869                 }
1870                 if (!dev->in_fs_metadata || !dev->writeable)
1871                         continue;
1872
1873                 btrfs_set_stack_device_generation(dev_item, 0);
1874                 btrfs_set_stack_device_type(dev_item, dev->type);
1875                 btrfs_set_stack_device_id(dev_item, dev->devid);
1876                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1877                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1878                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1879                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1880                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1881                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1882                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
1883                 flags = btrfs_super_flags(sb);
1884                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1885
1886
1887                 crc = ~(u32)0;
1888                 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1889                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1890                 btrfs_csum_final(crc, sb->csum);
1891
1892                 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1893                               BTRFS_SUPER_INFO_SIZE);
1894
1895                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1896                 dev->pending_io = bh;
1897
1898                 get_bh(bh);
1899                 set_buffer_uptodate(bh);
1900                 lock_buffer(bh);
1901                 bh->b_end_io = btrfs_end_buffer_write_sync;
1902
1903                 if (do_barriers && dev->barriers) {
1904                         ret = submit_bh(WRITE_BARRIER, bh);
1905                         if (ret == -EOPNOTSUPP) {
1906                                 printk("btrfs: disabling barriers on dev %s\n",
1907                                        dev->name);
1908                                 set_buffer_uptodate(bh);
1909                                 dev->barriers = 0;
1910                                 get_bh(bh);
1911                                 lock_buffer(bh);
1912                                 ret = submit_bh(WRITE, bh);
1913                         }
1914                 } else {
1915                         ret = submit_bh(WRITE, bh);
1916                 }
1917                 if (ret)
1918                         total_errors++;
1919         }
1920         if (total_errors > max_errors) {
1921                 printk("btrfs: %d errors while writing supers\n", total_errors);
1922                 BUG();
1923         }
1924         total_errors = 0;
1925
1926         list_for_each(cur, head) {
1927                 dev = list_entry(cur, struct btrfs_device, dev_list);
1928                 if (!dev->bdev)
1929                         continue;
1930                 if (!dev->in_fs_metadata || !dev->writeable)
1931                         continue;
1932
1933                 BUG_ON(!dev->pending_io);
1934                 bh = dev->pending_io;
1935                 wait_on_buffer(bh);
1936                 if (!buffer_uptodate(dev->pending_io)) {
1937                         if (do_barriers && dev->barriers) {
1938                                 printk("btrfs: disabling barriers on dev %s\n",
1939                                        dev->name);
1940                                 set_buffer_uptodate(bh);
1941                                 get_bh(bh);
1942                                 lock_buffer(bh);
1943                                 dev->barriers = 0;
1944                                 ret = submit_bh(WRITE, bh);
1945                                 BUG_ON(ret);
1946                                 wait_on_buffer(bh);
1947                                 if (!buffer_uptodate(bh))
1948                                         total_errors++;
1949                         } else {
1950                                 total_errors++;
1951                         }
1952
1953                 }
1954                 dev->pending_io = NULL;
1955                 brelse(bh);
1956         }
1957         if (total_errors > max_errors) {
1958                 printk("btrfs: %d errors while writing supers\n", total_errors);
1959                 BUG();
1960         }
1961         return 0;
1962 }
1963
1964 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1965                       *root)
1966 {
1967         int ret;
1968
1969         ret = write_all_supers(root);
1970         return ret;
1971 }
1972
1973 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1974 {
1975         radix_tree_delete(&fs_info->fs_roots_radix,
1976                           (unsigned long)root->root_key.objectid);
1977         if (root->anon_super.s_dev) {
1978                 down_write(&root->anon_super.s_umount);
1979                 kill_anon_super(&root->anon_super);
1980         }
1981 #if 0
1982         if (root->in_sysfs)
1983                 btrfs_sysfs_del_root(root);
1984 #endif
1985         if (root->node)
1986                 free_extent_buffer(root->node);
1987         if (root->commit_root)
1988                 free_extent_buffer(root->commit_root);
1989         if (root->name)
1990                 kfree(root->name);
1991         kfree(root);
1992         return 0;
1993 }
1994
1995 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1996 {
1997         int ret;
1998         struct btrfs_root *gang[8];
1999         int i;
2000
2001         while(1) {
2002                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2003                                              (void **)gang, 0,
2004                                              ARRAY_SIZE(gang));
2005                 if (!ret)
2006                         break;
2007                 for (i = 0; i < ret; i++)
2008                         btrfs_free_fs_root(fs_info, gang[i]);
2009         }
2010         return 0;
2011 }
2012
2013 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2014 {
2015         u64 root_objectid = 0;
2016         struct btrfs_root *gang[8];
2017         int i;
2018         int ret;
2019
2020         while (1) {
2021                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2022                                              (void **)gang, root_objectid,
2023                                              ARRAY_SIZE(gang));
2024                 if (!ret)
2025                         break;
2026                 for (i = 0; i < ret; i++) {
2027                         root_objectid = gang[i]->root_key.objectid;
2028                         ret = btrfs_find_dead_roots(fs_info->tree_root,
2029                                                     root_objectid, gang[i]);
2030                         BUG_ON(ret);
2031                         btrfs_orphan_cleanup(gang[i]);
2032                 }
2033                 root_objectid++;
2034         }
2035         return 0;
2036 }
2037
2038 int btrfs_commit_super(struct btrfs_root *root)
2039 {
2040         struct btrfs_trans_handle *trans;
2041         int ret;
2042
2043         mutex_lock(&root->fs_info->cleaner_mutex);
2044         btrfs_clean_old_snapshots(root);
2045         mutex_unlock(&root->fs_info->cleaner_mutex);
2046         trans = btrfs_start_transaction(root, 1);
2047         ret = btrfs_commit_transaction(trans, root);
2048         BUG_ON(ret);
2049         /* run commit again to drop the original snapshot */
2050         trans = btrfs_start_transaction(root, 1);
2051         btrfs_commit_transaction(trans, root);
2052         ret = btrfs_write_and_wait_transaction(NULL, root);
2053         BUG_ON(ret);
2054
2055         ret = write_ctree_super(NULL, root);
2056         return ret;
2057 }
2058
2059 int close_ctree(struct btrfs_root *root)
2060 {
2061         struct btrfs_fs_info *fs_info = root->fs_info;
2062         int ret;
2063
2064         fs_info->closing = 1;
2065         smp_mb();
2066
2067         kthread_stop(root->fs_info->transaction_kthread);
2068         kthread_stop(root->fs_info->cleaner_kthread);
2069
2070         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2071                 ret =  btrfs_commit_super(root);
2072                 if (ret) {
2073                         printk("btrfs: commit super returns %d\n", ret);
2074                 }
2075         }
2076
2077         if (fs_info->delalloc_bytes) {
2078                 printk("btrfs: at unmount delalloc count %Lu\n",
2079                        fs_info->delalloc_bytes);
2080         }
2081         if (fs_info->total_ref_cache_size) {
2082                 printk("btrfs: at umount reference cache size %Lu\n",
2083                         fs_info->total_ref_cache_size);
2084         }
2085
2086         if (fs_info->extent_root->node)
2087                 free_extent_buffer(fs_info->extent_root->node);
2088
2089         if (fs_info->tree_root->node)
2090                 free_extent_buffer(fs_info->tree_root->node);
2091
2092         if (root->fs_info->chunk_root->node);
2093                 free_extent_buffer(root->fs_info->chunk_root->node);
2094
2095         if (root->fs_info->dev_root->node);
2096                 free_extent_buffer(root->fs_info->dev_root->node);
2097
2098         btrfs_free_block_groups(root->fs_info);
2099
2100         del_fs_roots(fs_info);
2101
2102         iput(fs_info->btree_inode);
2103
2104         btrfs_stop_workers(&fs_info->fixup_workers);
2105         btrfs_stop_workers(&fs_info->delalloc_workers);
2106         btrfs_stop_workers(&fs_info->workers);
2107         btrfs_stop_workers(&fs_info->endio_workers);
2108         btrfs_stop_workers(&fs_info->endio_write_workers);
2109         btrfs_stop_workers(&fs_info->submit_workers);
2110
2111 #if 0
2112         while(!list_empty(&fs_info->hashers)) {
2113                 struct btrfs_hasher *hasher;
2114                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
2115                                     hashers);
2116                 list_del(&hasher->hashers);
2117                 crypto_free_hash(&fs_info->hash_tfm);
2118                 kfree(hasher);
2119         }
2120 #endif
2121         btrfs_close_devices(fs_info->fs_devices);
2122         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2123
2124         bdi_destroy(&fs_info->bdi);
2125
2126         kfree(fs_info->extent_root);
2127         kfree(fs_info->tree_root);
2128         kfree(fs_info->chunk_root);
2129         kfree(fs_info->dev_root);
2130         return 0;
2131 }
2132
2133 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2134 {
2135         int ret;
2136         struct inode *btree_inode = buf->first_page->mapping->host;
2137
2138         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2139         if (!ret)
2140                 return ret;
2141
2142         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2143                                     parent_transid);
2144         return !ret;
2145 }
2146
2147 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2148 {
2149         struct inode *btree_inode = buf->first_page->mapping->host;
2150         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2151                                           buf);
2152 }
2153
2154 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2155 {
2156         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2157         u64 transid = btrfs_header_generation(buf);
2158         struct inode *btree_inode = root->fs_info->btree_inode;
2159
2160         WARN_ON(!btrfs_tree_locked(buf));
2161         if (transid != root->fs_info->generation) {
2162                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
2163                         (unsigned long long)buf->start,
2164                         transid, root->fs_info->generation);
2165                 WARN_ON(1);
2166         }
2167         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
2168 }
2169
2170 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2171 {
2172         /*
2173          * looks as though older kernels can get into trouble with
2174          * this code, they end up stuck in balance_dirty_pages forever
2175          */
2176         struct extent_io_tree *tree;
2177         u64 num_dirty;
2178         u64 start = 0;
2179         unsigned long thresh = 32 * 1024 * 1024;
2180         tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
2181
2182         if (current_is_pdflush() || current->flags & PF_MEMALLOC)
2183                 return;
2184
2185         num_dirty = count_range_bits(tree, &start, (u64)-1,
2186                                      thresh, EXTENT_DIRTY);
2187         if (num_dirty > thresh) {
2188                 balance_dirty_pages_ratelimited_nr(
2189                                    root->fs_info->btree_inode->i_mapping, 1);
2190         }
2191         return;
2192 }
2193
2194 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2195 {
2196         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2197         int ret;
2198         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2199         if (ret == 0) {
2200                 buf->flags |= EXTENT_UPTODATE;
2201         }
2202         return ret;
2203 }
2204
2205 int btree_lock_page_hook(struct page *page)
2206 {
2207         struct inode *inode = page->mapping->host;
2208         struct btrfs_root *root = BTRFS_I(inode)->root;
2209         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2210         struct extent_buffer *eb;
2211         unsigned long len;
2212         u64 bytenr = page_offset(page);
2213
2214         if (page->private == EXTENT_PAGE_PRIVATE)
2215                 goto out;
2216
2217         len = page->private >> 2;
2218         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2219         if (!eb)
2220                 goto out;
2221
2222         btrfs_tree_lock(eb);
2223         spin_lock(&root->fs_info->hash_lock);
2224         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2225         spin_unlock(&root->fs_info->hash_lock);
2226         btrfs_tree_unlock(eb);
2227         free_extent_buffer(eb);
2228 out:
2229         lock_page(page);
2230         return 0;
2231 }
2232
2233 static struct extent_io_ops btree_extent_io_ops = {
2234         .write_cache_pages_lock_hook = btree_lock_page_hook,
2235         .readpage_end_io_hook = btree_readpage_end_io_hook,
2236         .submit_bio_hook = btree_submit_bio_hook,
2237         /* note we're sharing with inode.c for the merge bio hook */
2238         .merge_bio_hook = btrfs_merge_bio_hook,
2239 };