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