git.oblomov.eu Git - linux-2.6/blob - fs/btrfs/disk-io.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/fs.h>
  20 #include <linux/blkdev.h>
  21 #include <linux/scatterlist.h>
  22 #include <linux/swap.h>
  23 #include <linux/radix-tree.h>
  24 #include <linux/writeback.h>
  25 #include <linux/buffer_head.h>
  26 #include <linux/workqueue.h>
  27 #include <linux/kthread.h>
  28 #include <linux/freezer.h>
  29 #include "compat.h"
  30 #include "crc32c.h"
  31 #include "ctree.h"
  32 #include "disk-io.h"
  33 #include "transaction.h"
  34 #include "btrfs_inode.h"
  35 #include "volumes.h"
  36 #include "print-tree.h"
  37 #include "async-thread.h"
  38 #include "locking.h"
  39 #include "ref-cache.h"
  40 #include "tree-log.h"
  41 #include "free-space-cache.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 /* These are used to set the lockdep class on the extent buffer locks.
  80  * The class is set by the readpage_end_io_hook after the buffer has
  81  * passed csum validation but before the pages are unlocked.
  82  *
  83  * The lockdep class is also set by btrfs_init_new_buffer on freshly
  84  * allocated blocks.
  85  *
  86  * The class is based on the level in the tree block, which allows lockdep
  87  * to know that lower nodes nest inside the locks of higher nodes.
  88  *
  89  * We also add a check to make sure the highest level of the tree is
  90  * the same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this
  91  * code needs update as well.
  92  */
  93 #ifdef CONFIG_DEBUG_LOCK_ALLOC
  94 # if BTRFS_MAX_LEVEL != 8
  95 #  error
  96 # endif
  97 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
  98 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
  99         /* leaf */
 100         "btrfs-extent-00",
 101         "btrfs-extent-01",
 102         "btrfs-extent-02",
 103         "btrfs-extent-03",
 104         "btrfs-extent-04",
 105         "btrfs-extent-05",
 106         "btrfs-extent-06",
 107         "btrfs-extent-07",
 108         /* highest possible level */
 109         "btrfs-extent-08",
 110 };
 111 #endif
 112
 113 /*
 114  * extents on the btree inode are pretty simple, there's one extent
 115  * that covers the entire device
 116  */
 117 static struct extent_map *btree_get_extent(struct inode *inode,
 118                 struct page *page, size_t page_offset, u64 start, u64 len,
 119                 int create)
 120 {
 121         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 122         struct extent_map *em;
 123         int ret;
 124
 125         spin_lock(&em_tree->lock);
 126         em = lookup_extent_mapping(em_tree, start, len);
 127         if (em) {
 128                 em->bdev =
 129                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 130                 spin_unlock(&em_tree->lock);
 131                 goto out;
 132         }
 133         spin_unlock(&em_tree->lock);
 134
 135         em = alloc_extent_map(GFP_NOFS);
 136         if (!em) {
 137                 em = ERR_PTR(-ENOMEM);
 138                 goto out;
 139         }
 140         em->start = 0;
 141         em->len = (u64)-1;
 142         em->block_len = (u64)-1;
 143         em->block_start = 0;
 144         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 145
 146         spin_lock(&em_tree->lock);
 147         ret = add_extent_mapping(em_tree, em);
 148         if (ret == -EEXIST) {
 149                 u64 failed_start = em->start;
 150                 u64 failed_len = em->len;
 151
 152                 free_extent_map(em);
 153                 em = lookup_extent_mapping(em_tree, start, len);
 154                 if (em) {
 155                         ret = 0;
 156                 } else {
 157                         em = lookup_extent_mapping(em_tree, failed_start,
 158                                                    failed_len);
 159                         ret = -EIO;
 160                 }
 161         } else if (ret) {
 162                 free_extent_map(em);
 163                 em = NULL;
 164         }
 165         spin_unlock(&em_tree->lock);
 166
 167         if (ret)
 168                 em = ERR_PTR(ret);
 169 out:
 170         return em;
 171 }
 172
 173 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 174 {
 175         return btrfs_crc32c(seed, data, len);
 176 }
 177
 178 void btrfs_csum_final(u32 crc, char *result)
 179 {
 180         *(__le32 *)result = ~cpu_to_le32(crc);
 181 }
 182
 183 /*
 184  * compute the csum for a btree block, and either verify it or write it
 185  * into the csum field of the block.
 186  */
 187 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 188                            int verify)
 189 {
 190         u16 csum_size =
 191                 btrfs_super_csum_size(&root->fs_info->super_copy);
 192         char *result = NULL;
 193         unsigned long len;
 194         unsigned long cur_len;
 195         unsigned long offset = BTRFS_CSUM_SIZE;
 196         char *map_token = NULL;
 197         char *kaddr;
 198         unsigned long map_start;
 199         unsigned long map_len;
 200         int err;
 201         u32 crc = ~(u32)0;
 202         unsigned long inline_result;
 203
 204         len = buf->len - offset;
 205         while (len > 0) {
 206                 err = map_private_extent_buffer(buf, offset, 32,
 207                                         &map_token, &kaddr,
 208                                         &map_start, &map_len, KM_USER0);
 209                 if (err)
 210                         return 1;
 211                 cur_len = min(len, map_len - (offset - map_start));
 212                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 213                                       crc, cur_len);
 214                 len -= cur_len;
 215                 offset += cur_len;
 216                 unmap_extent_buffer(buf, map_token, KM_USER0);
 217         }
 218         if (csum_size > sizeof(inline_result)) {
 219                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
 220                 if (!result)
 221                         return 1;
 222         } else {
 223                 result = (char *)&inline_result;
 224         }
 225
 226         btrfs_csum_final(crc, result);
 227
 228         if (verify) {
 229                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 230                         u32 val;
 231                         u32 found = 0;
 232                         memcpy(&found, result, csum_size);
 233
 234                         read_extent_buffer(buf, &val, 0, csum_size);
 235                         if (printk_ratelimit()) {
 236                                 printk(KERN_INFO "btrfs: %s checksum verify "
 237                                        "failed on %llu wanted %X found %X "
 238                                        "level %d\n",
 239                                        root->fs_info->sb->s_id,
 240                                        (unsigned long long)buf->start, val, found,
 241                                        btrfs_header_level(buf));
 242                         }
 243                         if (result != (char *)&inline_result)
 244                                 kfree(result);
 245                         return 1;
 246                 }
 247         } else {
 248                 write_extent_buffer(buf, result, 0, csum_size);
 249         }
 250         if (result != (char *)&inline_result)
 251                 kfree(result);
 252         return 0;
 253 }
 254
 255 /*
 256  * we can't consider a given block up to date unless the transid of the
 257  * block matches the transid in the parent node's pointer.  This is how we
 258  * detect blocks that either didn't get written at all or got written
 259  * in the wrong place.
 260  */
 261 static int verify_parent_transid(struct extent_io_tree *io_tree,
 262                                  struct extent_buffer *eb, u64 parent_transid)
 263 {
 264         int ret;
 265
 266         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 267                 return 0;
 268
 269         lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
 270         if (extent_buffer_uptodate(io_tree, eb) &&
 271             btrfs_header_generation(eb) == parent_transid) {
 272                 ret = 0;
 273                 goto out;
 274         }
 275         if (printk_ratelimit()) {
 276                 printk("parent transid verify failed on %llu wanted %llu "
 277                        "found %llu\n",
 278                        (unsigned long long)eb->start,
 279                        (unsigned long long)parent_transid,
 280                        (unsigned long long)btrfs_header_generation(eb));
 281         }
 282         ret = 1;
 283         clear_extent_buffer_uptodate(io_tree, eb);
 284 out:
 285         unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
 286                       GFP_NOFS);
 287         return ret;
 288 }
 289
 290 /*
 291  * helper to read a given tree block, doing retries as required when
 292  * the checksums don't match and we have alternate mirrors to try.
 293  */
 294 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 295                                           struct extent_buffer *eb,
 296                                           u64 start, u64 parent_transid)
 297 {
 298         struct extent_io_tree *io_tree;
 299         int ret;
 300         int num_copies = 0;
 301         int mirror_num = 0;
 302
 303         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 304         while (1) {
 305                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 306                                                btree_get_extent, mirror_num);
 307                 if (!ret &&
 308                     !verify_parent_transid(io_tree, eb, parent_transid))
 309                         return ret;
 310
 311                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 312                                               eb->start, eb->len);
 313                 if (num_copies == 1)
 314                         return ret;
 315
 316                 mirror_num++;
 317                 if (mirror_num > num_copies)
 318                         return ret;
 319         }
 320         return -EIO;
 321 }
 322
 323 /*
 324  * checksum a dirty tree block before IO.  This has extra checks to make sure
 325  * we only fill in the checksum field in the first page of a multi-page block
 326  */
 327
 328 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 329 {
 330         struct extent_io_tree *tree;
 331         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 332         u64 found_start;
 333         int found_level;
 334         unsigned long len;
 335         struct extent_buffer *eb;
 336         int ret;
 337
 338         tree = &BTRFS_I(page->mapping->host)->io_tree;
 339
 340         if (page->private == EXTENT_PAGE_PRIVATE)
 341                 goto out;
 342         if (!page->private)
 343                 goto out;
 344         len = page->private >> 2;
 345         WARN_ON(len == 0);
 346
 347         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 348         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 349                                              btrfs_header_generation(eb));
 350         BUG_ON(ret);
 351         found_start = btrfs_header_bytenr(eb);
 352         if (found_start != start) {
 353                 WARN_ON(1);
 354                 goto err;
 355         }
 356         if (eb->first_page != page) {
 357                 WARN_ON(1);
 358                 goto err;
 359         }
 360         if (!PageUptodate(page)) {
 361                 WARN_ON(1);
 362                 goto err;
 363         }
 364         found_level = btrfs_header_level(eb);
 365
 366         csum_tree_block(root, eb, 0);
 367 err:
 368         free_extent_buffer(eb);
 369 out:
 370         return 0;
 371 }
 372
 373 static int check_tree_block_fsid(struct btrfs_root *root,
 374                                  struct extent_buffer *eb)
 375 {
 376         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 377         u8 fsid[BTRFS_UUID_SIZE];
 378         int ret = 1;
 379
 380         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
 381                            BTRFS_FSID_SIZE);
 382         while (fs_devices) {
 383                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 384                         ret = 0;
 385                         break;
 386                 }
 387                 fs_devices = fs_devices->seed;
 388         }
 389         return ret;
 390 }
 391
 392 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 393 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
 394 {
 395         lockdep_set_class_and_name(&eb->lock,
 396                            &btrfs_eb_class[level],
 397                            btrfs_eb_name[level]);
 398 }
 399 #endif
 400
 401 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 402                                struct extent_state *state)
 403 {
 404         struct extent_io_tree *tree;
 405         u64 found_start;
 406         int found_level;
 407         unsigned long len;
 408         struct extent_buffer *eb;
 409         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 410         int ret = 0;
 411
 412         tree = &BTRFS_I(page->mapping->host)->io_tree;
 413         if (page->private == EXTENT_PAGE_PRIVATE)
 414                 goto out;
 415         if (!page->private)
 416                 goto out;
 417
 418         len = page->private >> 2;
 419         WARN_ON(len == 0);
 420
 421         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 422
 423         found_start = btrfs_header_bytenr(eb);
 424         if (found_start != start) {
 425                 if (printk_ratelimit()) {
 426                         printk(KERN_INFO "btrfs bad tree block start "
 427                                "%llu %llu\n",
 428                                (unsigned long long)found_start,
 429                                (unsigned long long)eb->start);
 430                 }
 431                 ret = -EIO;
 432                 goto err;
 433         }
 434         if (eb->first_page != page) {
 435                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
 436                        eb->first_page->index, page->index);
 437                 WARN_ON(1);
 438                 ret = -EIO;
 439                 goto err;
 440         }
 441         if (check_tree_block_fsid(root, eb)) {
 442                 if (printk_ratelimit()) {
 443                         printk(KERN_INFO "btrfs bad fsid on block %llu\n",
 444                                (unsigned long long)eb->start);
 445                 }
 446                 ret = -EIO;
 447                 goto err;
 448         }
 449         found_level = btrfs_header_level(eb);
 450
 451         btrfs_set_buffer_lockdep_class(eb, found_level);
 452
 453         ret = csum_tree_block(root, eb, 1);
 454         if (ret)
 455                 ret = -EIO;
 456
 457         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 458         end = eb->start + end - 1;
 459 err:
 460         free_extent_buffer(eb);
 461 out:
 462         return ret;
 463 }
 464
 465 static void end_workqueue_bio(struct bio *bio, int err)
 466 {
 467         struct end_io_wq *end_io_wq = bio->bi_private;
 468         struct btrfs_fs_info *fs_info;
 469
 470         fs_info = end_io_wq->info;
 471         end_io_wq->error = err;
 472         end_io_wq->work.func = end_workqueue_fn;
 473         end_io_wq->work.flags = 0;
 474
 475         if (bio->bi_rw & (1 << BIO_RW)) {
 476                 if (end_io_wq->metadata)
 477                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
 478                                            &end_io_wq->work);
 479                 else
 480                         btrfs_queue_worker(&fs_info->endio_write_workers,
 481                                            &end_io_wq->work);
 482         } else {
 483                 if (end_io_wq->metadata)
 484                         btrfs_queue_worker(&fs_info->endio_meta_workers,
 485                                            &end_io_wq->work);
 486                 else
 487                         btrfs_queue_worker(&fs_info->endio_workers,
 488                                            &end_io_wq->work);
 489         }
 490 }
 491
 492 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 493                         int metadata)
 494 {
 495         struct end_io_wq *end_io_wq;
 496         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 497         if (!end_io_wq)
 498                 return -ENOMEM;
 499
 500         end_io_wq->private = bio->bi_private;
 501         end_io_wq->end_io = bio->bi_end_io;
 502         end_io_wq->info = info;
 503         end_io_wq->error = 0;
 504         end_io_wq->bio = bio;
 505         end_io_wq->metadata = metadata;
 506
 507         bio->bi_private = end_io_wq;
 508         bio->bi_end_io = end_workqueue_bio;
 509         return 0;
 510 }
 511
 512 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 513 {
 514         unsigned long limit = min_t(unsigned long,
 515                                     info->workers.max_workers,
 516                                     info->fs_devices->open_devices);
 517         return 256 * limit;
 518 }
 519
 520 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
 521 {
 522         return atomic_read(&info->nr_async_bios) >
 523                 btrfs_async_submit_limit(info);
 524 }
 525
 526 static void run_one_async_start(struct btrfs_work *work)
 527 {
 528         struct btrfs_fs_info *fs_info;
 529         struct async_submit_bio *async;
 530
 531         async = container_of(work, struct  async_submit_bio, work);
 532         fs_info = BTRFS_I(async->inode)->root->fs_info;
 533         async->submit_bio_start(async->inode, async->rw, async->bio,
 534                                async->mirror_num, async->bio_flags);
 535 }
 536
 537 static void run_one_async_done(struct btrfs_work *work)
 538 {
 539         struct btrfs_fs_info *fs_info;
 540         struct async_submit_bio *async;
 541         int limit;
 542
 543         async = container_of(work, struct  async_submit_bio, work);
 544         fs_info = BTRFS_I(async->inode)->root->fs_info;
 545
 546         limit = btrfs_async_submit_limit(fs_info);
 547         limit = limit * 2 / 3;
 548
 549         atomic_dec(&fs_info->nr_async_submits);
 550
 551         if (atomic_read(&fs_info->nr_async_submits) < limit &&
 552             waitqueue_active(&fs_info->async_submit_wait))
 553                 wake_up(&fs_info->async_submit_wait);
 554
 555         async->submit_bio_done(async->inode, async->rw, async->bio,
 556                                async->mirror_num, async->bio_flags);
 557 }
 558
 559 static void run_one_async_free(struct btrfs_work *work)
 560 {
 561         struct async_submit_bio *async;
 562
 563         async = container_of(work, struct  async_submit_bio, work);
 564         kfree(async);
 565 }
 566
 567 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 568                         int rw, struct bio *bio, int mirror_num,
 569                         unsigned long bio_flags,
 570                         extent_submit_bio_hook_t *submit_bio_start,
 571                         extent_submit_bio_hook_t *submit_bio_done)
 572 {
 573         struct async_submit_bio *async;
 574
 575         async = kmalloc(sizeof(*async), GFP_NOFS);
 576         if (!async)
 577                 return -ENOMEM;
 578
 579         async->inode = inode;
 580         async->rw = rw;
 581         async->bio = bio;
 582         async->mirror_num = mirror_num;
 583         async->submit_bio_start = submit_bio_start;
 584         async->submit_bio_done = submit_bio_done;
 585
 586         async->work.func = run_one_async_start;
 587         async->work.ordered_func = run_one_async_done;
 588         async->work.ordered_free = run_one_async_free;
 589
 590         async->work.flags = 0;
 591         async->bio_flags = bio_flags;
 592
 593         atomic_inc(&fs_info->nr_async_submits);
 594
 595         if (rw & (1 << BIO_RW_SYNCIO))
 596                 btrfs_set_work_high_prio(&async->work);
 597
 598         btrfs_queue_worker(&fs_info->workers, &async->work);
 599
 600         while (atomic_read(&fs_info->async_submit_draining) &&
 601               atomic_read(&fs_info->nr_async_submits)) {
 602                 wait_event(fs_info->async_submit_wait,
 603                            (atomic_read(&fs_info->nr_async_submits) == 0));
 604         }
 605
 606         return 0;
 607 }
 608
 609 static int btree_csum_one_bio(struct bio *bio)
 610 {
 611         struct bio_vec *bvec = bio->bi_io_vec;
 612         int bio_index = 0;
 613         struct btrfs_root *root;
 614
 615         WARN_ON(bio->bi_vcnt <= 0);
 616         while (bio_index < bio->bi_vcnt) {
 617                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 618                 csum_dirty_buffer(root, bvec->bv_page);
 619                 bio_index++;
 620                 bvec++;
 621         }
 622         return 0;
 623 }
 624
 625 static int __btree_submit_bio_start(struct inode *inode, int rw,
 626                                     struct bio *bio, int mirror_num,
 627                                     unsigned long bio_flags)
 628 {
 629         /*
 630          * when we're called for a write, we're already in the async
 631          * submission context.  Just jump into btrfs_map_bio
 632          */
 633         btree_csum_one_bio(bio);
 634         return 0;
 635 }
 636
 637 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 638                                  int mirror_num, unsigned long bio_flags)
 639 {
 640         /*
 641          * when we're called for a write, we're already in the async
 642          * submission context.  Just jump into btrfs_map_bio
 643          */
 644         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 645 }
 646
 647 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 648                                  int mirror_num, unsigned long bio_flags)
 649 {
 650         int ret;
 651
 652         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 653                                           bio, 1);
 654         BUG_ON(ret);
 655
 656         if (!(rw & (1 << BIO_RW))) {
 657                 /*
 658                  * called for a read, do the setup so that checksum validation
 659                  * can happen in the async kernel threads
 660                  */
 661                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 662                                      mirror_num, 0);
 663         }
 664
 665         /*
 666          * kthread helpers are used to submit writes so that checksumming
 667          * can happen in parallel across all CPUs
 668          */
 669         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 670                                    inode, rw, bio, mirror_num, 0,
 671                                    __btree_submit_bio_start,
 672                                    __btree_submit_bio_done);
 673 }
 674
 675 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 676 {
 677         struct extent_io_tree *tree;
 678         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 679         struct extent_buffer *eb;
 680         int was_dirty;
 681
 682         tree = &BTRFS_I(page->mapping->host)->io_tree;
 683         if (!(current->flags & PF_MEMALLOC)) {
 684                 return extent_write_full_page(tree, page,
 685                                               btree_get_extent, wbc);
 686         }
 687
 688         redirty_page_for_writepage(wbc, page);
 689         eb = btrfs_find_tree_block(root, page_offset(page),
 690                                       PAGE_CACHE_SIZE);
 691         WARN_ON(!eb);
 692
 693         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
 694         if (!was_dirty) {
 695                 spin_lock(&root->fs_info->delalloc_lock);
 696                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
 697                 spin_unlock(&root->fs_info->delalloc_lock);
 698         }
 699         free_extent_buffer(eb);
 700
 701         unlock_page(page);
 702         return 0;
 703 }
 704
 705 static int btree_writepages(struct address_space *mapping,
 706                             struct writeback_control *wbc)
 707 {
 708         struct extent_io_tree *tree;
 709         tree = &BTRFS_I(mapping->host)->io_tree;
 710         if (wbc->sync_mode == WB_SYNC_NONE) {
 711                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
 712                 u64 num_dirty;
 713                 unsigned long thresh = 32 * 1024 * 1024;
 714
 715                 if (wbc->for_kupdate)
 716                         return 0;
 717
 718                 /* this is a bit racy, but that's ok */
 719                 num_dirty = root->fs_info->dirty_metadata_bytes;
 720                 if (num_dirty < thresh)
 721                         return 0;
 722         }
 723         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 724 }
 725
 726 static int btree_readpage(struct file *file, struct page *page)
 727 {
 728         struct extent_io_tree *tree;
 729         tree = &BTRFS_I(page->mapping->host)->io_tree;
 730         return extent_read_full_page(tree, page, btree_get_extent);
 731 }
 732
 733 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 734 {
 735         struct extent_io_tree *tree;
 736         struct extent_map_tree *map;
 737         int ret;
 738
 739         if (PageWriteback(page) || PageDirty(page))
 740                 return 0;
 741
 742         tree = &BTRFS_I(page->mapping->host)->io_tree;
 743         map = &BTRFS_I(page->mapping->host)->extent_tree;
 744
 745         ret = try_release_extent_state(map, tree, page, gfp_flags);
 746         if (!ret)
 747                 return 0;
 748
 749         ret = try_release_extent_buffer(tree, page);
 750         if (ret == 1) {
 751                 ClearPagePrivate(page);
 752                 set_page_private(page, 0);
 753                 page_cache_release(page);
 754         }
 755
 756         return ret;
 757 }
 758
 759 static void btree_invalidatepage(struct page *page, unsigned long offset)
 760 {
 761         struct extent_io_tree *tree;
 762         tree = &BTRFS_I(page->mapping->host)->io_tree;
 763         extent_invalidatepage(tree, page, offset);
 764         btree_releasepage(page, GFP_NOFS);
 765         if (PagePrivate(page)) {
 766                 printk(KERN_WARNING "btrfs warning page private not zero "
 767                        "on page %llu\n", (unsigned long long)page_offset(page));
 768                 ClearPagePrivate(page);
 769                 set_page_private(page, 0);
 770                 page_cache_release(page);
 771         }
 772 }
 773
 774 static struct address_space_operations btree_aops = {
 775         .readpage       = btree_readpage,
 776         .writepage      = btree_writepage,
 777         .writepages     = btree_writepages,
 778         .releasepage    = btree_releasepage,
 779         .invalidatepage = btree_invalidatepage,
 780         .sync_page      = block_sync_page,
 781 };
 782
 783 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 784                          u64 parent_transid)
 785 {
 786         struct extent_buffer *buf = NULL;
 787         struct inode *btree_inode = root->fs_info->btree_inode;
 788         int ret = 0;
 789
 790         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 791         if (!buf)
 792                 return 0;
 793         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 794                                  buf, 0, 0, btree_get_extent, 0);
 795         free_extent_buffer(buf);
 796         return ret;
 797 }
 798
 799 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 800                                             u64 bytenr, u32 blocksize)
 801 {
 802         struct inode *btree_inode = root->fs_info->btree_inode;
 803         struct extent_buffer *eb;
 804         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 805                                 bytenr, blocksize, GFP_NOFS);
 806         return eb;
 807 }
 808
 809 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 810                                                  u64 bytenr, u32 blocksize)
 811 {
 812         struct inode *btree_inode = root->fs_info->btree_inode;
 813         struct extent_buffer *eb;
 814
 815         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 816                                  bytenr, blocksize, NULL, GFP_NOFS);
 817         return eb;
 818 }
 819
 820
 821 int btrfs_write_tree_block(struct extent_buffer *buf)
 822 {
 823         return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
 824                                       buf->start + buf->len - 1, WB_SYNC_ALL);
 825 }
 826
 827 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 828 {
 829         return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
 830                                   buf->start, buf->start + buf->len - 1);
 831 }
 832
 833 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 834                                       u32 blocksize, u64 parent_transid)
 835 {
 836         struct extent_buffer *buf = NULL;
 837         struct inode *btree_inode = root->fs_info->btree_inode;
 838         struct extent_io_tree *io_tree;
 839         int ret;
 840
 841         io_tree = &BTRFS_I(btree_inode)->io_tree;
 842
 843         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 844         if (!buf)
 845                 return NULL;
 846
 847         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 848
 849         if (ret == 0)
 850                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
 851         return buf;
 852
 853 }
 854
 855 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 856                      struct extent_buffer *buf)
 857 {
 858         struct inode *btree_inode = root->fs_info->btree_inode;
 859         if (btrfs_header_generation(buf) ==
 860             root->fs_info->running_transaction->transid) {
 861                 btrfs_assert_tree_locked(buf);
 862
 863                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
 864                         spin_lock(&root->fs_info->delalloc_lock);
 865                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
 866                                 root->fs_info->dirty_metadata_bytes -= buf->len;
 867                         else
 868                                 WARN_ON(1);
 869                         spin_unlock(&root->fs_info->delalloc_lock);
 870                 }
 871
 872                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
 873                 btrfs_set_lock_blocking(buf);
 874                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 875                                           buf);
 876         }
 877         return 0;
 878 }
 879
 880 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 881                         u32 stripesize, struct btrfs_root *root,
 882                         struct btrfs_fs_info *fs_info,
 883                         u64 objectid)
 884 {
 885         root->node = NULL;
 886         root->commit_root = NULL;
 887         root->ref_tree = NULL;
 888         root->sectorsize = sectorsize;
 889         root->nodesize = nodesize;
 890         root->leafsize = leafsize;
 891         root->stripesize = stripesize;
 892         root->ref_cows = 0;
 893         root->track_dirty = 0;
 894
 895         root->fs_info = fs_info;
 896         root->objectid = objectid;
 897         root->last_trans = 0;
 898         root->highest_inode = 0;
 899         root->last_inode_alloc = 0;
 900         root->name = NULL;
 901         root->in_sysfs = 0;
 902
 903         INIT_LIST_HEAD(&root->dirty_list);
 904         INIT_LIST_HEAD(&root->orphan_list);
 905         INIT_LIST_HEAD(&root->dead_list);
 906         spin_lock_init(&root->node_lock);
 907         spin_lock_init(&root->list_lock);
 908         mutex_init(&root->objectid_mutex);
 909         mutex_init(&root->log_mutex);
 910         init_waitqueue_head(&root->log_writer_wait);
 911         init_waitqueue_head(&root->log_commit_wait[0]);
 912         init_waitqueue_head(&root->log_commit_wait[1]);
 913         atomic_set(&root->log_commit[0], 0);
 914         atomic_set(&root->log_commit[1], 0);
 915         atomic_set(&root->log_writers, 0);
 916         root->log_batch = 0;
 917         root->log_transid = 0;
 918         extent_io_tree_init(&root->dirty_log_pages,
 919                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 920
 921         btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
 922         root->ref_tree = &root->ref_tree_struct;
 923
 924         memset(&root->root_key, 0, sizeof(root->root_key));
 925         memset(&root->root_item, 0, sizeof(root->root_item));
 926         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 927         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 928         root->defrag_trans_start = fs_info->generation;
 929         init_completion(&root->kobj_unregister);
 930         root->defrag_running = 0;
 931         root->defrag_level = 0;
 932         root->root_key.objectid = objectid;
 933         root->anon_super.s_root = NULL;
 934         root->anon_super.s_dev = 0;
 935         INIT_LIST_HEAD(&root->anon_super.s_list);
 936         INIT_LIST_HEAD(&root->anon_super.s_instances);
 937         init_rwsem(&root->anon_super.s_umount);
 938
 939         return 0;
 940 }
 941
 942 static int find_and_setup_root(struct btrfs_root *tree_root,
 943                                struct btrfs_fs_info *fs_info,
 944                                u64 objectid,
 945                                struct btrfs_root *root)
 946 {
 947         int ret;
 948         u32 blocksize;
 949         u64 generation;
 950
 951         __setup_root(tree_root->nodesize, tree_root->leafsize,
 952                      tree_root->sectorsize, tree_root->stripesize,
 953                      root, fs_info, objectid);
 954         ret = btrfs_find_last_root(tree_root, objectid,
 955                                    &root->root_item, &root->root_key);
 956         BUG_ON(ret);
 957
 958         generation = btrfs_root_generation(&root->root_item);
 959         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 960         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 961                                      blocksize, generation);
 962         BUG_ON(!root->node);
 963         return 0;
 964 }
 965
 966 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
 967                              struct btrfs_fs_info *fs_info)
 968 {
 969         struct extent_buffer *eb;
 970         struct btrfs_root *log_root_tree = fs_info->log_root_tree;
 971         u64 start = 0;
 972         u64 end = 0;
 973         int ret;
 974
 975         if (!log_root_tree)
 976                 return 0;
 977
 978         while (1) {
 979                 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
 980                                     0, &start, &end, EXTENT_DIRTY);
 981                 if (ret)
 982                         break;
 983
 984                 clear_extent_dirty(&log_root_tree->dirty_log_pages,
 985                                    start, end, GFP_NOFS);
 986         }
 987         eb = fs_info->log_root_tree->node;
 988
 989         WARN_ON(btrfs_header_level(eb) != 0);
 990         WARN_ON(btrfs_header_nritems(eb) != 0);
 991
 992         ret = btrfs_free_reserved_extent(fs_info->tree_root,
 993                                 eb->start, eb->len);
 994         BUG_ON(ret);
 995
 996         free_extent_buffer(eb);
 997         kfree(fs_info->log_root_tree);
 998         fs_info->log_root_tree = NULL;
 999         return 0;
1000 }
1001
1002 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1003                                          struct btrfs_fs_info *fs_info)
1004 {
1005         struct btrfs_root *root;
1006         struct btrfs_root *tree_root = fs_info->tree_root;
1007         struct extent_buffer *leaf;
1008
1009         root = kzalloc(sizeof(*root), GFP_NOFS);
1010         if (!root)
1011                 return ERR_PTR(-ENOMEM);
1012
1013         __setup_root(tree_root->nodesize, tree_root->leafsize,
1014                      tree_root->sectorsize, tree_root->stripesize,
1015                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1016
1017         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1018         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1019         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1020         /*
1021          * log trees do not get reference counted because they go away
1022          * before a real commit is actually done.  They do store pointers
1023          * to file data extents, and those reference counts still get
1024          * updated (along with back refs to the log tree).
1025          */
1026         root->ref_cows = 0;
1027
1028         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1029                                       0, BTRFS_TREE_LOG_OBJECTID,
1030                                       trans->transid, 0, 0, 0);
1031         if (IS_ERR(leaf)) {
1032                 kfree(root);
1033                 return ERR_CAST(leaf);
1034         }
1035
1036         root->node = leaf;
1037         btrfs_set_header_nritems(root->node, 0);
1038         btrfs_set_header_level(root->node, 0);
1039         btrfs_set_header_bytenr(root->node, root->node->start);
1040         btrfs_set_header_generation(root->node, trans->transid);
1041         btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
1042
1043         write_extent_buffer(root->node, root->fs_info->fsid,
1044                             (unsigned long)btrfs_header_fsid(root->node),
1045                             BTRFS_FSID_SIZE);
1046         btrfs_mark_buffer_dirty(root->node);
1047         btrfs_tree_unlock(root->node);
1048         return root;
1049 }
1050
1051 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1052                              struct btrfs_fs_info *fs_info)
1053 {
1054         struct btrfs_root *log_root;
1055
1056         log_root = alloc_log_tree(trans, fs_info);
1057         if (IS_ERR(log_root))
1058                 return PTR_ERR(log_root);
1059         WARN_ON(fs_info->log_root_tree);
1060         fs_info->log_root_tree = log_root;
1061         return 0;
1062 }
1063
1064 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1065                        struct btrfs_root *root)
1066 {
1067         struct btrfs_root *log_root;
1068         struct btrfs_inode_item *inode_item;
1069
1070         log_root = alloc_log_tree(trans, root->fs_info);
1071         if (IS_ERR(log_root))
1072                 return PTR_ERR(log_root);
1073
1074         log_root->last_trans = trans->transid;
1075         log_root->root_key.offset = root->root_key.objectid;
1076
1077         inode_item = &log_root->root_item.inode;
1078         inode_item->generation = cpu_to_le64(1);
1079         inode_item->size = cpu_to_le64(3);
1080         inode_item->nlink = cpu_to_le32(1);
1081         inode_item->nbytes = cpu_to_le64(root->leafsize);
1082         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1083
1084         btrfs_set_root_bytenr(&log_root->root_item, log_root->node->start);
1085         btrfs_set_root_generation(&log_root->root_item, trans->transid);
1086
1087         WARN_ON(root->log_root);
1088         root->log_root = log_root;
1089         root->log_transid = 0;
1090         return 0;
1091 }
1092
1093 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1094                                                struct btrfs_key *location)
1095 {
1096         struct btrfs_root *root;
1097         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1098         struct btrfs_path *path;
1099         struct extent_buffer *l;
1100         u64 highest_inode;
1101         u64 generation;
1102         u32 blocksize;
1103         int ret = 0;
1104
1105         root = kzalloc(sizeof(*root), GFP_NOFS);
1106         if (!root)
1107                 return ERR_PTR(-ENOMEM);
1108         if (location->offset == (u64)-1) {
1109                 ret = find_and_setup_root(tree_root, fs_info,
1110                                           location->objectid, root);
1111                 if (ret) {
1112                         kfree(root);
1113                         return ERR_PTR(ret);
1114                 }
1115                 goto insert;
1116         }
1117
1118         __setup_root(tree_root->nodesize, tree_root->leafsize,
1119                      tree_root->sectorsize, tree_root->stripesize,
1120                      root, fs_info, location->objectid);
1121
1122         path = btrfs_alloc_path();
1123         BUG_ON(!path);
1124         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1125         if (ret != 0) {
1126                 if (ret > 0)
1127                         ret = -ENOENT;
1128                 goto out;
1129         }
1130         l = path->nodes[0];
1131         read_extent_buffer(l, &root->root_item,
1132                btrfs_item_ptr_offset(l, path->slots[0]),
1133                sizeof(root->root_item));
1134         memcpy(&root->root_key, location, sizeof(*location));
1135         ret = 0;
1136 out:
1137         btrfs_release_path(root, path);
1138         btrfs_free_path(path);
1139         if (ret) {
1140                 kfree(root);
1141                 return ERR_PTR(ret);
1142         }
1143         generation = btrfs_root_generation(&root->root_item);
1144         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1145         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1146                                      blocksize, generation);
1147         BUG_ON(!root->node);
1148 insert:
1149         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1150                 root->ref_cows = 1;
1151                 ret = btrfs_find_highest_inode(root, &highest_inode);
1152                 if (ret == 0) {
1153                         root->highest_inode = highest_inode;
1154                         root->last_inode_alloc = highest_inode;
1155                 }
1156         }
1157         return root;
1158 }
1159
1160 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1161                                         u64 root_objectid)
1162 {
1163         struct btrfs_root *root;
1164
1165         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1166                 return fs_info->tree_root;
1167         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1168                 return fs_info->extent_root;
1169
1170         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1171                                  (unsigned long)root_objectid);
1172         return root;
1173 }
1174
1175 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1176                                               struct btrfs_key *location)
1177 {
1178         struct btrfs_root *root;
1179         int ret;
1180
1181         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1182                 return fs_info->tree_root;
1183         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1184                 return fs_info->extent_root;
1185         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1186                 return fs_info->chunk_root;
1187         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1188                 return fs_info->dev_root;
1189         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1190                 return fs_info->csum_root;
1191
1192         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1193                                  (unsigned long)location->objectid);
1194         if (root)
1195                 return root;
1196
1197         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1198         if (IS_ERR(root))
1199                 return root;
1200
1201         set_anon_super(&root->anon_super, NULL);
1202
1203         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1204                                 (unsigned long)root->root_key.objectid,
1205                                 root);
1206         if (ret) {
1207                 free_extent_buffer(root->node);
1208                 kfree(root);
1209                 return ERR_PTR(ret);
1210         }
1211         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
1212                 ret = btrfs_find_dead_roots(fs_info->tree_root,
1213                                             root->root_key.objectid, root);
1214                 BUG_ON(ret);
1215                 btrfs_orphan_cleanup(root);
1216         }
1217         return root;
1218 }
1219
1220 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1221                                       struct btrfs_key *location,
1222                                       const char *name, int namelen)
1223 {
1224         struct btrfs_root *root;
1225         int ret;
1226
1227         root = btrfs_read_fs_root_no_name(fs_info, location);
1228         if (!root)
1229                 return NULL;
1230
1231         if (root->in_sysfs)
1232                 return root;
1233
1234         ret = btrfs_set_root_name(root, name, namelen);
1235         if (ret) {
1236                 free_extent_buffer(root->node);
1237                 kfree(root);
1238                 return ERR_PTR(ret);
1239         }
1240 #if 0
1241         ret = btrfs_sysfs_add_root(root);
1242         if (ret) {
1243                 free_extent_buffer(root->node);
1244                 kfree(root->name);
1245                 kfree(root);
1246                 return ERR_PTR(ret);
1247         }
1248 #endif
1249         root->in_sysfs = 1;
1250         return root;
1251 }
1252
1253 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1254 {
1255         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1256         int ret = 0;
1257         struct btrfs_device *device;
1258         struct backing_dev_info *bdi;
1259
1260         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1261                 if (!device->bdev)
1262                         continue;
1263                 bdi = blk_get_backing_dev_info(device->bdev);
1264                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1265                         ret = 1;
1266                         break;
1267                 }
1268         }
1269         return ret;
1270 }
1271
1272 /*
1273  * this unplugs every device on the box, and it is only used when page
1274  * is null
1275  */
1276 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1277 {
1278         struct btrfs_device *device;
1279         struct btrfs_fs_info *info;
1280
1281         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1282         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1283                 if (!device->bdev)
1284                         continue;
1285
1286                 bdi = blk_get_backing_dev_info(device->bdev);
1287                 if (bdi->unplug_io_fn)
1288                         bdi->unplug_io_fn(bdi, page);
1289         }
1290 }
1291
1292 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1293 {
1294         struct inode *inode;
1295         struct extent_map_tree *em_tree;
1296         struct extent_map *em;
1297         struct address_space *mapping;
1298         u64 offset;
1299
1300         /* the generic O_DIRECT read code does this */
1301         if (1 || !page) {
1302                 __unplug_io_fn(bdi, page);
1303                 return;
1304         }
1305
1306         /*
1307          * page->mapping may change at any time.  Get a consistent copy
1308          * and use that for everything below
1309          */
1310         smp_mb();
1311         mapping = page->mapping;
1312         if (!mapping)
1313                 return;
1314
1315         inode = mapping->host;
1316
1317         /*
1318          * don't do the expensive searching for a small number of
1319          * devices
1320          */
1321         if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1322                 __unplug_io_fn(bdi, page);
1323                 return;
1324         }
1325
1326         offset = page_offset(page);
1327
1328         em_tree = &BTRFS_I(inode)->extent_tree;
1329         spin_lock(&em_tree->lock);
1330         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1331         spin_unlock(&em_tree->lock);
1332         if (!em) {
1333                 __unplug_io_fn(bdi, page);
1334                 return;
1335         }
1336
1337         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1338                 free_extent_map(em);
1339                 __unplug_io_fn(bdi, page);
1340                 return;
1341         }
1342         offset = offset - em->start;
1343         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1344                           em->block_start + offset, page);
1345         free_extent_map(em);
1346 }
1347
1348 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1349 {
1350         bdi_init(bdi);
1351         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1352         bdi->state              = 0;
1353         bdi->capabilities       = default_backing_dev_info.capabilities;
1354         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1355         bdi->unplug_io_data     = info;
1356         bdi->congested_fn       = btrfs_congested_fn;
1357         bdi->congested_data     = info;
1358         return 0;
1359 }
1360
1361 static int bio_ready_for_csum(struct bio *bio)
1362 {
1363         u64 length = 0;
1364         u64 buf_len = 0;
1365         u64 start = 0;
1366         struct page *page;
1367         struct extent_io_tree *io_tree = NULL;
1368         struct btrfs_fs_info *info = NULL;
1369         struct bio_vec *bvec;
1370         int i;
1371         int ret;
1372
1373         bio_for_each_segment(bvec, bio, i) {
1374                 page = bvec->bv_page;
1375                 if (page->private == EXTENT_PAGE_PRIVATE) {
1376                         length += bvec->bv_len;
1377                         continue;
1378                 }
1379                 if (!page->private) {
1380                         length += bvec->bv_len;
1381                         continue;
1382                 }
1383                 length = bvec->bv_len;
1384                 buf_len = page->private >> 2;
1385                 start = page_offset(page) + bvec->bv_offset;
1386                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1387                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1388         }
1389         /* are we fully contained in this bio? */
1390         if (buf_len <= length)
1391                 return 1;
1392
1393         ret = extent_range_uptodate(io_tree, start + length,
1394                                     start + buf_len - 1);
1395         return ret;
1396 }
1397
1398 /*
1399  * called by the kthread helper functions to finally call the bio end_io
1400  * functions.  This is where read checksum verification actually happens
1401  */
1402 static void end_workqueue_fn(struct btrfs_work *work)
1403 {
1404         struct bio *bio;
1405         struct end_io_wq *end_io_wq;
1406         struct btrfs_fs_info *fs_info;
1407         int error;
1408
1409         end_io_wq = container_of(work, struct end_io_wq, work);
1410         bio = end_io_wq->bio;
1411         fs_info = end_io_wq->info;
1412
1413         /* metadata bio reads are special because the whole tree block must
1414          * be checksummed at once.  This makes sure the entire block is in
1415          * ram and up to date before trying to verify things.  For
1416          * blocksize <= pagesize, it is basically a noop
1417          */
1418         if (!(bio->bi_rw & (1 << BIO_RW)) && end_io_wq->metadata &&
1419             !bio_ready_for_csum(bio)) {
1420                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1421                                    &end_io_wq->work);
1422                 return;
1423         }
1424         error = end_io_wq->error;
1425         bio->bi_private = end_io_wq->private;
1426         bio->bi_end_io = end_io_wq->end_io;
1427         kfree(end_io_wq);
1428         bio_endio(bio, error);
1429 }
1430
1431 static int cleaner_kthread(void *arg)
1432 {
1433         struct btrfs_root *root = arg;
1434
1435         do {
1436                 smp_mb();
1437                 if (root->fs_info->closing)
1438                         break;
1439
1440                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1441                 mutex_lock(&root->fs_info->cleaner_mutex);
1442                 btrfs_clean_old_snapshots(root);
1443                 mutex_unlock(&root->fs_info->cleaner_mutex);
1444
1445                 if (freezing(current)) {
1446                         refrigerator();
1447                 } else {
1448                         smp_mb();
1449                         if (root->fs_info->closing)
1450                                 break;
1451                         set_current_state(TASK_INTERRUPTIBLE);
1452                         schedule();
1453                         __set_current_state(TASK_RUNNING);
1454                 }
1455         } while (!kthread_should_stop());
1456         return 0;
1457 }
1458
1459 static int transaction_kthread(void *arg)
1460 {
1461         struct btrfs_root *root = arg;
1462         struct btrfs_trans_handle *trans;
1463         struct btrfs_transaction *cur;
1464         unsigned long now;
1465         unsigned long delay;
1466         int ret;
1467
1468         do {
1469                 smp_mb();
1470                 if (root->fs_info->closing)
1471                         break;
1472
1473                 delay = HZ * 30;
1474                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1475                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1476
1477                 mutex_lock(&root->fs_info->trans_mutex);
1478                 cur = root->fs_info->running_transaction;
1479                 if (!cur) {
1480                         mutex_unlock(&root->fs_info->trans_mutex);
1481                         goto sleep;
1482                 }
1483
1484                 now = get_seconds();
1485                 if (now < cur->start_time || now - cur->start_time < 30) {
1486                         mutex_unlock(&root->fs_info->trans_mutex);
1487                         delay = HZ * 5;
1488                         goto sleep;
1489                 }
1490                 mutex_unlock(&root->fs_info->trans_mutex);
1491                 trans = btrfs_start_transaction(root, 1);
1492                 ret = btrfs_commit_transaction(trans, root);
1493
1494 sleep:
1495                 wake_up_process(root->fs_info->cleaner_kthread);
1496                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1497
1498                 if (freezing(current)) {
1499                         refrigerator();
1500                 } else {
1501                         if (root->fs_info->closing)
1502                                 break;
1503                         set_current_state(TASK_INTERRUPTIBLE);
1504                         schedule_timeout(delay);
1505                         __set_current_state(TASK_RUNNING);
1506                 }
1507         } while (!kthread_should_stop());
1508         return 0;
1509 }
1510
1511 struct btrfs_root *open_ctree(struct super_block *sb,
1512                               struct btrfs_fs_devices *fs_devices,
1513                               char *options)
1514 {
1515         u32 sectorsize;
1516         u32 nodesize;
1517         u32 leafsize;
1518         u32 blocksize;
1519         u32 stripesize;
1520         u64 generation;
1521         u64 features;
1522         struct btrfs_key location;
1523         struct buffer_head *bh;
1524         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1525                                                  GFP_NOFS);
1526         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1527                                                  GFP_NOFS);
1528         struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1529                                                GFP_NOFS);
1530         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1531                                                 GFP_NOFS);
1532         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1533                                                 GFP_NOFS);
1534         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1535                                               GFP_NOFS);
1536         struct btrfs_root *log_tree_root;
1537
1538         int ret;
1539         int err = -EINVAL;
1540
1541         struct btrfs_super_block *disk_super;
1542
1543         if (!extent_root || !tree_root || !fs_info ||
1544             !chunk_root || !dev_root || !csum_root) {
1545                 err = -ENOMEM;
1546                 goto fail;
1547         }
1548         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1549         INIT_LIST_HEAD(&fs_info->trans_list);
1550         INIT_LIST_HEAD(&fs_info->dead_roots);
1551         INIT_LIST_HEAD(&fs_info->hashers);
1552         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1553         INIT_LIST_HEAD(&fs_info->ordered_operations);
1554         spin_lock_init(&fs_info->delalloc_lock);
1555         spin_lock_init(&fs_info->new_trans_lock);
1556         spin_lock_init(&fs_info->ref_cache_lock);
1557
1558         init_completion(&fs_info->kobj_unregister);
1559         fs_info->tree_root = tree_root;
1560         fs_info->extent_root = extent_root;
1561         fs_info->csum_root = csum_root;
1562         fs_info->chunk_root = chunk_root;
1563         fs_info->dev_root = dev_root;
1564         fs_info->fs_devices = fs_devices;
1565         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1566         INIT_LIST_HEAD(&fs_info->space_info);
1567         btrfs_mapping_init(&fs_info->mapping_tree);
1568         atomic_set(&fs_info->nr_async_submits, 0);
1569         atomic_set(&fs_info->async_delalloc_pages, 0);
1570         atomic_set(&fs_info->async_submit_draining, 0);
1571         atomic_set(&fs_info->nr_async_bios, 0);
1572         atomic_set(&fs_info->throttles, 0);
1573         atomic_set(&fs_info->throttle_gen, 0);
1574         fs_info->sb = sb;
1575         fs_info->max_extent = (u64)-1;
1576         fs_info->max_inline = 8192 * 1024;
1577         setup_bdi(fs_info, &fs_info->bdi);
1578         fs_info->btree_inode = new_inode(sb);
1579         fs_info->btree_inode->i_ino = 1;
1580         fs_info->btree_inode->i_nlink = 1;
1581         fs_info->metadata_ratio = 8;
1582
1583         fs_info->thread_pool_size = min_t(unsigned long,
1584                                           num_online_cpus() + 2, 8);
1585
1586         INIT_LIST_HEAD(&fs_info->ordered_extents);
1587         spin_lock_init(&fs_info->ordered_extent_lock);
1588
1589         sb->s_blocksize = 4096;
1590         sb->s_blocksize_bits = blksize_bits(4096);
1591
1592         /*
1593          * we set the i_size on the btree inode to the max possible int.
1594          * the real end of the address space is determined by all of
1595          * the devices in the system
1596          */
1597         fs_info->btree_inode->i_size = OFFSET_MAX;
1598         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1599         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1600
1601         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1602                              fs_info->btree_inode->i_mapping,
1603                              GFP_NOFS);
1604         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1605                              GFP_NOFS);
1606
1607         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1608
1609         spin_lock_init(&fs_info->block_group_cache_lock);
1610         fs_info->block_group_cache_tree.rb_node = NULL;
1611
1612         extent_io_tree_init(&fs_info->pinned_extents,
1613                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1614         fs_info->do_barriers = 1;
1615
1616         INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
1617         btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
1618         btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
1619
1620         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1621         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1622                sizeof(struct btrfs_key));
1623         insert_inode_hash(fs_info->btree_inode);
1624
1625         mutex_init(&fs_info->trans_mutex);
1626         mutex_init(&fs_info->ordered_operations_mutex);
1627         mutex_init(&fs_info->tree_log_mutex);
1628         mutex_init(&fs_info->drop_mutex);
1629         mutex_init(&fs_info->chunk_mutex);
1630         mutex_init(&fs_info->transaction_kthread_mutex);
1631         mutex_init(&fs_info->cleaner_mutex);
1632         mutex_init(&fs_info->volume_mutex);
1633         mutex_init(&fs_info->tree_reloc_mutex);
1634
1635         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1636         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1637
1638         init_waitqueue_head(&fs_info->transaction_throttle);
1639         init_waitqueue_head(&fs_info->transaction_wait);
1640         init_waitqueue_head(&fs_info->async_submit_wait);
1641
1642         __setup_root(4096, 4096, 4096, 4096, tree_root,
1643                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1644
1645
1646         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1647         if (!bh)
1648                 goto fail_iput;
1649
1650         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1651         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1652                sizeof(fs_info->super_for_commit));
1653         brelse(bh);
1654
1655         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1656
1657         disk_super = &fs_info->super_copy;
1658         if (!btrfs_super_root(disk_super))
1659                 goto fail_iput;
1660
1661         ret = btrfs_parse_options(tree_root, options);
1662         if (ret) {
1663                 err = ret;
1664                 goto fail_iput;
1665         }
1666
1667         features = btrfs_super_incompat_flags(disk_super) &
1668                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1669         if (features) {
1670                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1671                        "unsupported optional features (%Lx).\n",
1672                        (unsigned long long)features);
1673                 err = -EINVAL;
1674                 goto fail_iput;
1675         }
1676
1677         features = btrfs_super_compat_ro_flags(disk_super) &
1678                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1679         if (!(sb->s_flags & MS_RDONLY) && features) {
1680                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1681                        "unsupported option features (%Lx).\n",
1682                        (unsigned long long)features);
1683                 err = -EINVAL;
1684                 goto fail_iput;
1685         }
1686
1687         /*
1688          * we need to start all the end_io workers up front because the
1689          * queue work function gets called at interrupt time, and so it
1690          * cannot dynamically grow.
1691          */
1692         btrfs_init_workers(&fs_info->workers, "worker",
1693                            fs_info->thread_pool_size);
1694
1695         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1696                            fs_info->thread_pool_size);
1697
1698         btrfs_init_workers(&fs_info->submit_workers, "submit",
1699                            min_t(u64, fs_devices->num_devices,
1700                            fs_info->thread_pool_size));
1701
1702         /* a higher idle thresh on the submit workers makes it much more
1703          * likely that bios will be send down in a sane order to the
1704          * devices
1705          */
1706         fs_info->submit_workers.idle_thresh = 64;
1707
1708         fs_info->workers.idle_thresh = 16;
1709         fs_info->workers.ordered = 1;
1710
1711         fs_info->delalloc_workers.idle_thresh = 2;
1712         fs_info->delalloc_workers.ordered = 1;
1713
1714         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1715         btrfs_init_workers(&fs_info->endio_workers, "endio",
1716                            fs_info->thread_pool_size);
1717         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1718                            fs_info->thread_pool_size);
1719         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1720                            "endio-meta-write", fs_info->thread_pool_size);
1721         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1722                            fs_info->thread_pool_size);
1723
1724         /*
1725          * endios are largely parallel and should have a very
1726          * low idle thresh
1727          */
1728         fs_info->endio_workers.idle_thresh = 4;
1729         fs_info->endio_meta_workers.idle_thresh = 4;
1730
1731         fs_info->endio_write_workers.idle_thresh = 64;
1732         fs_info->endio_meta_write_workers.idle_thresh = 64;
1733
1734         btrfs_start_workers(&fs_info->workers, 1);
1735         btrfs_start_workers(&fs_info->submit_workers, 1);
1736         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1737         btrfs_start_workers(&fs_info->fixup_workers, 1);
1738         btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1739         btrfs_start_workers(&fs_info->endio_meta_workers,
1740                             fs_info->thread_pool_size);
1741         btrfs_start_workers(&fs_info->endio_meta_write_workers,
1742                             fs_info->thread_pool_size);
1743         btrfs_start_workers(&fs_info->endio_write_workers,
1744                             fs_info->thread_pool_size);
1745
1746         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1747         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1748                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1749
1750         nodesize = btrfs_super_nodesize(disk_super);
1751         leafsize = btrfs_super_leafsize(disk_super);
1752         sectorsize = btrfs_super_sectorsize(disk_super);
1753         stripesize = btrfs_super_stripesize(disk_super);
1754         tree_root->nodesize = nodesize;
1755         tree_root->leafsize = leafsize;
1756         tree_root->sectorsize = sectorsize;
1757         tree_root->stripesize = stripesize;
1758
1759         sb->s_blocksize = sectorsize;
1760         sb->s_blocksize_bits = blksize_bits(sectorsize);
1761
1762         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1763                     sizeof(disk_super->magic))) {
1764                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1765                 goto fail_sb_buffer;
1766         }
1767
1768         mutex_lock(&fs_info->chunk_mutex);
1769         ret = btrfs_read_sys_array(tree_root);
1770         mutex_unlock(&fs_info->chunk_mutex);
1771         if (ret) {
1772                 printk(KERN_WARNING "btrfs: failed to read the system "
1773                        "array on %s\n", sb->s_id);
1774                 goto fail_sys_array;
1775         }
1776
1777         blocksize = btrfs_level_size(tree_root,
1778                                      btrfs_super_chunk_root_level(disk_super));
1779         generation = btrfs_super_chunk_root_generation(disk_super);
1780
1781         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1782                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1783
1784         chunk_root->node = read_tree_block(chunk_root,
1785                                            btrfs_super_chunk_root(disk_super),
1786                                            blocksize, generation);
1787         BUG_ON(!chunk_root->node);
1788
1789         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1790            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1791            BTRFS_UUID_SIZE);
1792
1793         mutex_lock(&fs_info->chunk_mutex);
1794         ret = btrfs_read_chunk_tree(chunk_root);
1795         mutex_unlock(&fs_info->chunk_mutex);
1796         if (ret) {
1797                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1798                        sb->s_id);
1799                 goto fail_chunk_root;
1800         }
1801
1802         btrfs_close_extra_devices(fs_devices);
1803
1804         blocksize = btrfs_level_size(tree_root,
1805                                      btrfs_super_root_level(disk_super));
1806         generation = btrfs_super_generation(disk_super);
1807
1808         tree_root->node = read_tree_block(tree_root,
1809                                           btrfs_super_root(disk_super),
1810                                           blocksize, generation);
1811         if (!tree_root->node)
1812                 goto fail_chunk_root;
1813
1814
1815         ret = find_and_setup_root(tree_root, fs_info,
1816                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1817         if (ret)
1818                 goto fail_tree_root;
1819         extent_root->track_dirty = 1;
1820
1821         ret = find_and_setup_root(tree_root, fs_info,
1822                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1823         dev_root->track_dirty = 1;
1824         if (ret)
1825                 goto fail_extent_root;
1826
1827         ret = find_and_setup_root(tree_root, fs_info,
1828                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
1829         if (ret)
1830                 goto fail_extent_root;
1831
1832         csum_root->track_dirty = 1;
1833
1834         btrfs_read_block_groups(extent_root);
1835
1836         fs_info->generation = generation;
1837         fs_info->last_trans_committed = generation;
1838         fs_info->data_alloc_profile = (u64)-1;
1839         fs_info->metadata_alloc_profile = (u64)-1;
1840         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1841         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1842                                                "btrfs-cleaner");
1843         if (IS_ERR(fs_info->cleaner_kthread))
1844                 goto fail_csum_root;
1845
1846         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1847                                                    tree_root,
1848                                                    "btrfs-transaction");
1849         if (IS_ERR(fs_info->transaction_kthread))
1850                 goto fail_cleaner;
1851
1852         if (btrfs_super_log_root(disk_super) != 0) {
1853                 u64 bytenr = btrfs_super_log_root(disk_super);
1854
1855                 if (fs_devices->rw_devices == 0) {
1856                         printk(KERN_WARNING "Btrfs log replay required "
1857                                "on RO media\n");
1858                         err = -EIO;
1859                         goto fail_trans_kthread;
1860                 }
1861                 blocksize =
1862                      btrfs_level_size(tree_root,
1863                                       btrfs_super_log_root_level(disk_super));
1864
1865                 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1866                                                       GFP_NOFS);
1867
1868                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1869                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1870
1871                 log_tree_root->node = read_tree_block(tree_root, bytenr,
1872                                                       blocksize,
1873                                                       generation + 1);
1874                 ret = btrfs_recover_log_trees(log_tree_root);
1875                 BUG_ON(ret);
1876
1877                 if (sb->s_flags & MS_RDONLY) {
1878                         ret =  btrfs_commit_super(tree_root);
1879                         BUG_ON(ret);
1880                 }
1881         }
1882
1883         if (!(sb->s_flags & MS_RDONLY)) {
1884                 ret = btrfs_cleanup_reloc_trees(tree_root);
1885                 BUG_ON(ret);
1886         }
1887
1888         location.objectid = BTRFS_FS_TREE_OBJECTID;
1889         location.type = BTRFS_ROOT_ITEM_KEY;
1890         location.offset = (u64)-1;
1891
1892         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1893         if (!fs_info->fs_root)
1894                 goto fail_trans_kthread;
1895         return tree_root;
1896
1897 fail_trans_kthread:
1898         kthread_stop(fs_info->transaction_kthread);
1899 fail_cleaner:
1900         kthread_stop(fs_info->cleaner_kthread);
1901
1902         /*
1903          * make sure we're done with the btree inode before we stop our
1904          * kthreads
1905          */
1906         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1907         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1908
1909 fail_csum_root:
1910         free_extent_buffer(csum_root->node);
1911 fail_extent_root:
1912         free_extent_buffer(extent_root->node);
1913 fail_tree_root:
1914         free_extent_buffer(tree_root->node);
1915 fail_chunk_root:
1916         free_extent_buffer(chunk_root->node);
1917 fail_sys_array:
1918         free_extent_buffer(dev_root->node);
1919 fail_sb_buffer:
1920         btrfs_stop_workers(&fs_info->fixup_workers);
1921         btrfs_stop_workers(&fs_info->delalloc_workers);
1922         btrfs_stop_workers(&fs_info->workers);
1923         btrfs_stop_workers(&fs_info->endio_workers);
1924         btrfs_stop_workers(&fs_info->endio_meta_workers);
1925         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
1926         btrfs_stop_workers(&fs_info->endio_write_workers);
1927         btrfs_stop_workers(&fs_info->submit_workers);
1928 fail_iput:
1929         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1930         iput(fs_info->btree_inode);
1931
1932         btrfs_close_devices(fs_info->fs_devices);
1933         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1934         bdi_destroy(&fs_info->bdi);
1935
1936 fail:
1937         kfree(extent_root);
1938         kfree(tree_root);
1939         kfree(fs_info);
1940         kfree(chunk_root);
1941         kfree(dev_root);
1942         kfree(csum_root);
1943         return ERR_PTR(err);
1944 }
1945
1946 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1947 {
1948         char b[BDEVNAME_SIZE];
1949
1950         if (uptodate) {
1951                 set_buffer_uptodate(bh);
1952         } else {
1953                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1954                         printk(KERN_WARNING "lost page write due to "
1955                                         "I/O error on %s\n",
1956                                        bdevname(bh->b_bdev, b));
1957                 }
1958                 /* note, we dont' set_buffer_write_io_error because we have
1959                  * our own ways of dealing with the IO errors
1960                  */
1961                 clear_buffer_uptodate(bh);
1962         }
1963         unlock_buffer(bh);
1964         put_bh(bh);
1965 }
1966
1967 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
1968 {
1969         struct buffer_head *bh;
1970         struct buffer_head *latest = NULL;
1971         struct btrfs_super_block *super;
1972         int i;
1973         u64 transid = 0;
1974         u64 bytenr;
1975
1976         /* we would like to check all the supers, but that would make
1977          * a btrfs mount succeed after a mkfs from a different FS.
1978          * So, we need to add a special mount option to scan for
1979          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1980          */
1981         for (i = 0; i < 1; i++) {
1982                 bytenr = btrfs_sb_offset(i);
1983                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
1984                         break;
1985                 bh = __bread(bdev, bytenr / 4096, 4096);
1986                 if (!bh)
1987                         continue;
1988
1989                 super = (struct btrfs_super_block *)bh->b_data;
1990                 if (btrfs_super_bytenr(super) != bytenr ||
1991                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
1992                             sizeof(super->magic))) {
1993                         brelse(bh);
1994                         continue;
1995                 }
1996
1997                 if (!latest || btrfs_super_generation(super) > transid) {
1998                         brelse(latest);
1999                         latest = bh;
2000                         transid = btrfs_super_generation(super);
2001                 } else {
2002                         brelse(bh);
2003                 }
2004         }
2005         return latest;
2006 }
2007
2008 static int write_dev_supers(struct btrfs_device *device,
2009                             struct btrfs_super_block *sb,
2010                             int do_barriers, int wait, int max_mirrors)
2011 {
2012         struct buffer_head *bh;
2013         int i;
2014         int ret;
2015         int errors = 0;
2016         u32 crc;
2017         u64 bytenr;
2018         int last_barrier = 0;
2019
2020         if (max_mirrors == 0)
2021                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2022
2023         /* make sure only the last submit_bh does a barrier */
2024         if (do_barriers) {
2025                 for (i = 0; i < max_mirrors; i++) {
2026                         bytenr = btrfs_sb_offset(i);
2027                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2028                             device->total_bytes)
2029                                 break;
2030                         last_barrier = i;
2031                 }
2032         }
2033
2034         for (i = 0; i < max_mirrors; i++) {
2035                 bytenr = btrfs_sb_offset(i);
2036                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2037                         break;
2038
2039                 if (wait) {
2040                         bh = __find_get_block(device->bdev, bytenr / 4096,
2041                                               BTRFS_SUPER_INFO_SIZE);
2042                         BUG_ON(!bh);
2043                         brelse(bh);
2044                         wait_on_buffer(bh);
2045                         if (buffer_uptodate(bh)) {
2046                                 brelse(bh);
2047                                 continue;
2048                         }
2049                 } else {
2050                         btrfs_set_super_bytenr(sb, bytenr);
2051
2052                         crc = ~(u32)0;
2053                         crc = btrfs_csum_data(NULL, (char *)sb +
2054                                               BTRFS_CSUM_SIZE, crc,
2055                                               BTRFS_SUPER_INFO_SIZE -
2056                                               BTRFS_CSUM_SIZE);
2057                         btrfs_csum_final(crc, sb->csum);
2058
2059                         bh = __getblk(device->bdev, bytenr / 4096,
2060                                       BTRFS_SUPER_INFO_SIZE);
2061                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2062
2063                         set_buffer_uptodate(bh);
2064                         get_bh(bh);
2065                         lock_buffer(bh);
2066                         bh->b_end_io = btrfs_end_buffer_write_sync;
2067                 }
2068
2069                 if (i == last_barrier && do_barriers && device->barriers) {
2070                         ret = submit_bh(WRITE_BARRIER, bh);
2071                         if (ret == -EOPNOTSUPP) {
2072                                 printk("btrfs: disabling barriers on dev %s\n",
2073                                        device->name);
2074                                 set_buffer_uptodate(bh);
2075                                 device->barriers = 0;
2076                                 get_bh(bh);
2077                                 lock_buffer(bh);
2078                                 ret = submit_bh(WRITE_SYNC, bh);
2079                         }
2080                 } else {
2081                         ret = submit_bh(WRITE_SYNC, bh);
2082                 }
2083
2084                 if (!ret && wait) {
2085                         wait_on_buffer(bh);
2086                         if (!buffer_uptodate(bh))
2087                                 errors++;
2088                 } else if (ret) {
2089                         errors++;
2090                 }
2091                 if (wait)
2092                         brelse(bh);
2093         }
2094         return errors < i ? 0 : -1;
2095 }
2096
2097 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2098 {
2099         struct list_head *head = &root->fs_info->fs_devices->devices;
2100         struct btrfs_device *dev;
2101         struct btrfs_super_block *sb;
2102         struct btrfs_dev_item *dev_item;
2103         int ret;
2104         int do_barriers;
2105         int max_errors;
2106         int total_errors = 0;
2107         u64 flags;
2108
2109         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2110         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2111
2112         sb = &root->fs_info->super_for_commit;
2113         dev_item = &sb->dev_item;
2114         list_for_each_entry(dev, head, dev_list) {
2115                 if (!dev->bdev) {
2116                         total_errors++;
2117                         continue;
2118                 }
2119                 if (!dev->in_fs_metadata || !dev->writeable)
2120                         continue;
2121
2122                 btrfs_set_stack_device_generation(dev_item, 0);
2123                 btrfs_set_stack_device_type(dev_item, dev->type);
2124                 btrfs_set_stack_device_id(dev_item, dev->devid);
2125                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2126                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2127                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2128                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2129                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2130                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2131                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2132
2133                 flags = btrfs_super_flags(sb);
2134                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2135
2136                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2137                 if (ret)
2138                         total_errors++;
2139         }
2140         if (total_errors > max_errors) {
2141                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2142                        total_errors);
2143                 BUG();
2144         }
2145
2146         total_errors = 0;
2147         list_for_each_entry(dev, head, dev_list) {
2148                 if (!dev->bdev)
2149                         continue;
2150                 if (!dev->in_fs_metadata || !dev->writeable)
2151                         continue;
2152
2153                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2154                 if (ret)
2155                         total_errors++;
2156         }
2157         if (total_errors > max_errors) {
2158                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2159                        total_errors);
2160                 BUG();
2161         }
2162         return 0;
2163 }
2164
2165 int write_ctree_super(struct btrfs_trans_handle *trans,
2166                       struct btrfs_root *root, int max_mirrors)
2167 {
2168         int ret;
2169
2170         ret = write_all_supers(root, max_mirrors);
2171         return ret;
2172 }
2173
2174 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2175 {
2176         radix_tree_delete(&fs_info->fs_roots_radix,
2177                           (unsigned long)root->root_key.objectid);
2178         if (root->anon_super.s_dev) {
2179                 down_write(&root->anon_super.s_umount);
2180                 kill_anon_super(&root->anon_super);
2181         }
2182         if (root->node)
2183                 free_extent_buffer(root->node);
2184         if (root->commit_root)
2185                 free_extent_buffer(root->commit_root);
2186         kfree(root->name);
2187         kfree(root);
2188         return 0;
2189 }
2190
2191 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2192 {
2193         int ret;
2194         struct btrfs_root *gang[8];
2195         int i;
2196
2197         while (1) {
2198                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2199                                              (void **)gang, 0,
2200                                              ARRAY_SIZE(gang));
2201                 if (!ret)
2202                         break;
2203                 for (i = 0; i < ret; i++)
2204                         btrfs_free_fs_root(fs_info, gang[i]);
2205         }
2206         return 0;
2207 }
2208
2209 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2210 {
2211         u64 root_objectid = 0;
2212         struct btrfs_root *gang[8];
2213         int i;
2214         int ret;
2215
2216         while (1) {
2217                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2218                                              (void **)gang, root_objectid,
2219                                              ARRAY_SIZE(gang));
2220                 if (!ret)
2221                         break;
2222                 for (i = 0; i < ret; i++) {
2223                         root_objectid = gang[i]->root_key.objectid;
2224                         ret = btrfs_find_dead_roots(fs_info->tree_root,
2225                                                     root_objectid, gang[i]);
2226                         BUG_ON(ret);
2227                         btrfs_orphan_cleanup(gang[i]);
2228                 }
2229                 root_objectid++;
2230         }
2231         return 0;
2232 }
2233
2234 int btrfs_commit_super(struct btrfs_root *root)
2235 {
2236         struct btrfs_trans_handle *trans;
2237         int ret;
2238
2239         mutex_lock(&root->fs_info->cleaner_mutex);
2240         btrfs_clean_old_snapshots(root);
2241         mutex_unlock(&root->fs_info->cleaner_mutex);
2242         trans = btrfs_start_transaction(root, 1);
2243         ret = btrfs_commit_transaction(trans, root);
2244         BUG_ON(ret);
2245         /* run commit again to drop the original snapshot */
2246         trans = btrfs_start_transaction(root, 1);
2247         btrfs_commit_transaction(trans, root);
2248         ret = btrfs_write_and_wait_transaction(NULL, root);
2249         BUG_ON(ret);
2250
2251         ret = write_ctree_super(NULL, root, 0);
2252         return ret;
2253 }
2254
2255 int close_ctree(struct btrfs_root *root)
2256 {
2257         struct btrfs_fs_info *fs_info = root->fs_info;
2258         int ret;
2259
2260         fs_info->closing = 1;
2261         smp_mb();
2262
2263         kthread_stop(root->fs_info->transaction_kthread);
2264         kthread_stop(root->fs_info->cleaner_kthread);
2265
2266         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2267                 ret =  btrfs_commit_super(root);
2268                 if (ret)
2269                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2270         }
2271
2272         if (fs_info->delalloc_bytes) {
2273                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2274                        (unsigned long long)fs_info->delalloc_bytes);
2275         }
2276         if (fs_info->total_ref_cache_size) {
2277                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2278                        (unsigned long long)fs_info->total_ref_cache_size);
2279         }
2280
2281         if (fs_info->extent_root->node)
2282                 free_extent_buffer(fs_info->extent_root->node);
2283
2284         if (fs_info->tree_root->node)
2285                 free_extent_buffer(fs_info->tree_root->node);
2286
2287         if (root->fs_info->chunk_root->node)
2288                 free_extent_buffer(root->fs_info->chunk_root->node);
2289
2290         if (root->fs_info->dev_root->node)
2291                 free_extent_buffer(root->fs_info->dev_root->node);
2292
2293         if (root->fs_info->csum_root->node)
2294                 free_extent_buffer(root->fs_info->csum_root->node);
2295
2296         btrfs_free_block_groups(root->fs_info);
2297
2298         del_fs_roots(fs_info);
2299
2300         iput(fs_info->btree_inode);
2301
2302         btrfs_stop_workers(&fs_info->fixup_workers);
2303         btrfs_stop_workers(&fs_info->delalloc_workers);
2304         btrfs_stop_workers(&fs_info->workers);
2305         btrfs_stop_workers(&fs_info->endio_workers);
2306         btrfs_stop_workers(&fs_info->endio_meta_workers);
2307         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2308         btrfs_stop_workers(&fs_info->endio_write_workers);
2309         btrfs_stop_workers(&fs_info->submit_workers);
2310
2311         btrfs_close_devices(fs_info->fs_devices);
2312         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2313
2314         bdi_destroy(&fs_info->bdi);
2315
2316         kfree(fs_info->extent_root);
2317         kfree(fs_info->tree_root);
2318         kfree(fs_info->chunk_root);
2319         kfree(fs_info->dev_root);
2320         kfree(fs_info->csum_root);
2321         return 0;
2322 }
2323
2324 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2325 {
2326         int ret;
2327         struct inode *btree_inode = buf->first_page->mapping->host;
2328
2329         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2330         if (!ret)
2331                 return ret;
2332
2333         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2334                                     parent_transid);
2335         return !ret;
2336 }
2337
2338 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2339 {
2340         struct inode *btree_inode = buf->first_page->mapping->host;
2341         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2342                                           buf);
2343 }
2344
2345 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2346 {
2347         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2348         u64 transid = btrfs_header_generation(buf);
2349         struct inode *btree_inode = root->fs_info->btree_inode;
2350         int was_dirty;
2351
2352         btrfs_assert_tree_locked(buf);
2353         if (transid != root->fs_info->generation) {
2354                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2355                        "found %llu running %llu\n",
2356                         (unsigned long long)buf->start,
2357                         (unsigned long long)transid,
2358                         (unsigned long long)root->fs_info->generation);
2359                 WARN_ON(1);
2360         }
2361         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2362                                             buf);
2363         if (!was_dirty) {
2364                 spin_lock(&root->fs_info->delalloc_lock);
2365                 root->fs_info->dirty_metadata_bytes += buf->len;
2366                 spin_unlock(&root->fs_info->delalloc_lock);
2367         }
2368 }
2369
2370 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2371 {
2372         /*
2373          * looks as though older kernels can get into trouble with
2374          * this code, they end up stuck in balance_dirty_pages forever
2375          */
2376         struct extent_io_tree *tree;
2377         u64 num_dirty;
2378         u64 start = 0;
2379         unsigned long thresh = 32 * 1024 * 1024;
2380         tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
2381
2382         if (current->flags & PF_MEMALLOC)
2383                 return;
2384
2385         num_dirty = count_range_bits(tree, &start, (u64)-1,
2386                                      thresh, EXTENT_DIRTY);
2387         if (num_dirty > thresh) {
2388                 balance_dirty_pages_ratelimited_nr(
2389                                    root->fs_info->btree_inode->i_mapping, 1);
2390         }
2391         return;
2392 }
2393
2394 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2395 {
2396         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2397         int ret;
2398         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2399         if (ret == 0)
2400                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2401         return ret;
2402 }
2403
2404 int btree_lock_page_hook(struct page *page)
2405 {
2406         struct inode *inode = page->mapping->host;
2407         struct btrfs_root *root = BTRFS_I(inode)->root;
2408         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2409         struct extent_buffer *eb;
2410         unsigned long len;
2411         u64 bytenr = page_offset(page);
2412
2413         if (page->private == EXTENT_PAGE_PRIVATE)
2414                 goto out;
2415
2416         len = page->private >> 2;
2417         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2418         if (!eb)
2419                 goto out;
2420
2421         btrfs_tree_lock(eb);
2422         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2423
2424         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2425                 spin_lock(&root->fs_info->delalloc_lock);
2426                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2427                         root->fs_info->dirty_metadata_bytes -= eb->len;
2428                 else
2429                         WARN_ON(1);
2430                 spin_unlock(&root->fs_info->delalloc_lock);
2431         }
2432
2433         btrfs_tree_unlock(eb);
2434         free_extent_buffer(eb);
2435 out:
2436         lock_page(page);
2437         return 0;
2438 }
2439
2440 static struct extent_io_ops btree_extent_io_ops = {
2441         .write_cache_pages_lock_hook = btree_lock_page_hook,
2442         .readpage_end_io_hook = btree_readpage_end_io_hook,
2443         .submit_bio_hook = btree_submit_bio_hook,
2444         /* note we're sharing with inode.c for the merge bio hook */
2445         .merge_bio_hook = btrfs_merge_bio_hook,
2446 };