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
  42 static struct extent_io_ops btree_extent_io_ops;
  43 static void end_workqueue_fn(struct btrfs_work *work);
  44
  45 /*
  46  * end_io_wq structs are used to do processing in task context when an IO is
  47  * complete.  This is used during reads to verify checksums, and it is used
  48  * by writes to insert metadata for new file extents after IO is complete.
  49  */
  50 struct end_io_wq {
  51         struct bio *bio;
  52         bio_end_io_t *end_io;
  53         void *private;
  54         struct btrfs_fs_info *info;
  55         int error;
  56         int metadata;
  57         struct list_head list;
  58         struct btrfs_work work;
  59 };
  60
  61 /*
  62  * async submit bios are used to offload expensive checksumming
  63  * onto the worker threads.  They checksum file and metadata bios
  64  * just before they are sent down the IO stack.
  65  */
  66 struct async_submit_bio {
  67         struct inode *inode;
  68         struct bio *bio;
  69         struct list_head list;
  70         extent_submit_bio_hook_t *submit_bio_start;
  71         extent_submit_bio_hook_t *submit_bio_done;
  72         int rw;
  73         int mirror_num;
  74         unsigned long bio_flags;
  75         struct btrfs_work work;
  76 };
  77
  78 /* These are used to set the lockdep class on the extent buffer locks.
  79  * The class is set by the readpage_end_io_hook after the buffer has
  80  * passed csum validation but before the pages are unlocked.
  81  *
  82  * The lockdep class is also set by btrfs_init_new_buffer on freshly
  83  * allocated blocks.
  84  *
  85  * The class is based on the level in the tree block, which allows lockdep
  86  * to know that lower nodes nest inside the locks of higher nodes.
  87  *
  88  * We also add a check to make sure the highest level of the tree is
  89  * the same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this
  90  * code needs update as well.
  91  */
  92 #ifdef CONFIG_DEBUG_LOCK_ALLOC
  93 # if BTRFS_MAX_LEVEL != 8
  94 #  error
  95 # endif
  96 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
  97 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
  98         /* leaf */
  99         "btrfs-extent-00",
 100         "btrfs-extent-01",
 101         "btrfs-extent-02",
 102         "btrfs-extent-03",
 103         "btrfs-extent-04",
 104         "btrfs-extent-05",
 105         "btrfs-extent-06",
 106         "btrfs-extent-07",
 107         /* highest possible level */
 108         "btrfs-extent-08",
 109 };
 110 #endif
 111
 112 /*
 113  * extents on the btree inode are pretty simple, there's one extent
 114  * that covers the entire device
 115  */
 116 static struct extent_map *btree_get_extent(struct inode *inode,
 117                 struct page *page, size_t page_offset, u64 start, u64 len,
 118                 int create)
 119 {
 120         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 121         struct extent_map *em;
 122         int ret;
 123
 124         spin_lock(&em_tree->lock);
 125         em = lookup_extent_mapping(em_tree, start, len);
 126         if (em) {
 127                 em->bdev =
 128                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 129                 spin_unlock(&em_tree->lock);
 130                 goto out;
 131         }
 132         spin_unlock(&em_tree->lock);
 133
 134         em = alloc_extent_map(GFP_NOFS);
 135         if (!em) {
 136                 em = ERR_PTR(-ENOMEM);
 137                 goto out;
 138         }
 139         em->start = 0;
 140         em->len = (u64)-1;
 141         em->block_len = (u64)-1;
 142         em->block_start = 0;
 143         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 144
 145         spin_lock(&em_tree->lock);
 146         ret = add_extent_mapping(em_tree, em);
 147         if (ret == -EEXIST) {
 148                 u64 failed_start = em->start;
 149                 u64 failed_len = em->len;
 150
 151                 free_extent_map(em);
 152                 em = lookup_extent_mapping(em_tree, start, len);
 153                 if (em) {
 154                         ret = 0;
 155                 } else {
 156                         em = lookup_extent_mapping(em_tree, failed_start,
 157                                                    failed_len);
 158                         ret = -EIO;
 159                 }
 160         } else if (ret) {
 161                 free_extent_map(em);
 162                 em = NULL;
 163         }
 164         spin_unlock(&em_tree->lock);
 165
 166         if (ret)
 167                 em = ERR_PTR(ret);
 168 out:
 169         return em;
 170 }
 171
 172 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 173 {
 174         return btrfs_crc32c(seed, data, len);
 175 }
 176
 177 void btrfs_csum_final(u32 crc, char *result)
 178 {
 179         *(__le32 *)result = ~cpu_to_le32(crc);
 180 }
 181
 182 /*
 183  * compute the csum for a btree block, and either verify it or write it
 184  * into the csum field of the block.
 185  */
 186 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 187                            int verify)
 188 {
 189         u16 csum_size =
 190                 btrfs_super_csum_size(&root->fs_info->super_copy);
 191         char *result = NULL;
 192         unsigned long len;
 193         unsigned long cur_len;
 194         unsigned long offset = BTRFS_CSUM_SIZE;
 195         char *map_token = NULL;
 196         char *kaddr;
 197         unsigned long map_start;
 198         unsigned long map_len;
 199         int err;
 200         u32 crc = ~(u32)0;
 201         unsigned long inline_result;
 202
 203         len = buf->len - offset;
 204         while (len > 0) {
 205                 err = map_private_extent_buffer(buf, offset, 32,
 206                                         &map_token, &kaddr,
 207                                         &map_start, &map_len, KM_USER0);
 208                 if (err)
 209                         return 1;
 210                 cur_len = min(len, map_len - (offset - map_start));
 211                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 212                                       crc, cur_len);
 213                 len -= cur_len;
 214                 offset += cur_len;
 215                 unmap_extent_buffer(buf, map_token, KM_USER0);
 216         }
 217         if (csum_size > sizeof(inline_result)) {
 218                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
 219                 if (!result)
 220                         return 1;
 221         } else {
 222                 result = (char *)&inline_result;
 223         }
 224
 225         btrfs_csum_final(crc, result);
 226
 227         if (verify) {
 228                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 229                         u32 val;
 230                         u32 found = 0;
 231                         memcpy(&found, result, csum_size);
 232
 233                         read_extent_buffer(buf, &val, 0, csum_size);
 234                         printk(KERN_INFO "btrfs: %s checksum verify failed "
 235                                "on %llu wanted %X found %X level %d\n",
 236                                root->fs_info->sb->s_id,
 237                                buf->start, val, found, btrfs_header_level(buf));
 238                         if (result != (char *)&inline_result)
 239                                 kfree(result);
 240                         return 1;
 241                 }
 242         } else {
 243                 write_extent_buffer(buf, result, 0, csum_size);
 244         }
 245         if (result != (char *)&inline_result)
 246                 kfree(result);
 247         return 0;
 248 }
 249
 250 /*
 251  * we can't consider a given block up to date unless the transid of the
 252  * block matches the transid in the parent node's pointer.  This is how we
 253  * detect blocks that either didn't get written at all or got written
 254  * in the wrong place.
 255  */
 256 static int verify_parent_transid(struct extent_io_tree *io_tree,
 257                                  struct extent_buffer *eb, u64 parent_transid)
 258 {
 259         int ret;
 260
 261         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 262                 return 0;
 263
 264         lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
 265         if (extent_buffer_uptodate(io_tree, eb) &&
 266             btrfs_header_generation(eb) == parent_transid) {
 267                 ret = 0;
 268                 goto out;
 269         }
 270         printk("parent transid verify failed on %llu wanted %llu found %llu\n",
 271                (unsigned long long)eb->start,
 272                (unsigned long long)parent_transid,
 273                (unsigned long long)btrfs_header_generation(eb));
 274         ret = 1;
 275         clear_extent_buffer_uptodate(io_tree, eb);
 276 out:
 277         unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
 278                       GFP_NOFS);
 279         return ret;
 280 }
 281
 282 /*
 283  * helper to read a given tree block, doing retries as required when
 284  * the checksums don't match and we have alternate mirrors to try.
 285  */
 286 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 287                                           struct extent_buffer *eb,
 288                                           u64 start, u64 parent_transid)
 289 {
 290         struct extent_io_tree *io_tree;
 291         int ret;
 292         int num_copies = 0;
 293         int mirror_num = 0;
 294
 295         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 296         while (1) {
 297                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 298                                                btree_get_extent, mirror_num);
 299                 if (!ret &&
 300                     !verify_parent_transid(io_tree, eb, parent_transid))
 301                         return ret;
 302
 303                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 304                                               eb->start, eb->len);
 305                 if (num_copies == 1)
 306                         return ret;
 307
 308                 mirror_num++;
 309                 if (mirror_num > num_copies)
 310                         return ret;
 311         }
 312         return -EIO;
 313 }
 314
 315 /*
 316  * checksum a dirty tree block before IO.  This has extra checks to make sure
 317  * we only fill in the checksum field in the first page of a multi-page block
 318  */
 319
 320 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 321 {
 322         struct extent_io_tree *tree;
 323         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 324         u64 found_start;
 325         int found_level;
 326         unsigned long len;
 327         struct extent_buffer *eb;
 328         int ret;
 329
 330         tree = &BTRFS_I(page->mapping->host)->io_tree;
 331
 332         if (page->private == EXTENT_PAGE_PRIVATE)
 333                 goto out;
 334         if (!page->private)
 335                 goto out;
 336         len = page->private >> 2;
 337         WARN_ON(len == 0);
 338
 339         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 340         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 341                                              btrfs_header_generation(eb));
 342         BUG_ON(ret);
 343         found_start = btrfs_header_bytenr(eb);
 344         if (found_start != start) {
 345                 WARN_ON(1);
 346                 goto err;
 347         }
 348         if (eb->first_page != page) {
 349                 WARN_ON(1);
 350                 goto err;
 351         }
 352         if (!PageUptodate(page)) {
 353                 WARN_ON(1);
 354                 goto err;
 355         }
 356         found_level = btrfs_header_level(eb);
 357
 358         csum_tree_block(root, eb, 0);
 359 err:
 360         free_extent_buffer(eb);
 361 out:
 362         return 0;
 363 }
 364
 365 static int check_tree_block_fsid(struct btrfs_root *root,
 366                                  struct extent_buffer *eb)
 367 {
 368         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 369         u8 fsid[BTRFS_UUID_SIZE];
 370         int ret = 1;
 371
 372         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
 373                            BTRFS_FSID_SIZE);
 374         while (fs_devices) {
 375                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 376                         ret = 0;
 377                         break;
 378                 }
 379                 fs_devices = fs_devices->seed;
 380         }
 381         return ret;
 382 }
 383
 384 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 385 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
 386 {
 387         lockdep_set_class_and_name(&eb->lock,
 388                            &btrfs_eb_class[level],
 389                            btrfs_eb_name[level]);
 390 }
 391 #endif
 392
 393 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 394                                struct extent_state *state)
 395 {
 396         struct extent_io_tree *tree;
 397         u64 found_start;
 398         int found_level;
 399         unsigned long len;
 400         struct extent_buffer *eb;
 401         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 402         int ret = 0;
 403
 404         tree = &BTRFS_I(page->mapping->host)->io_tree;
 405         if (page->private == EXTENT_PAGE_PRIVATE)
 406                 goto out;
 407         if (!page->private)
 408                 goto out;
 409
 410         len = page->private >> 2;
 411         WARN_ON(len == 0);
 412
 413         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 414
 415         found_start = btrfs_header_bytenr(eb);
 416         if (found_start != start) {
 417                 printk(KERN_INFO "btrfs bad tree block start %llu %llu\n",
 418                        (unsigned long long)found_start,
 419                        (unsigned long long)eb->start);
 420                 ret = -EIO;
 421                 goto err;
 422         }
 423         if (eb->first_page != page) {
 424                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
 425                        eb->first_page->index, page->index);
 426                 WARN_ON(1);
 427                 ret = -EIO;
 428                 goto err;
 429         }
 430         if (check_tree_block_fsid(root, eb)) {
 431                 printk(KERN_INFO "btrfs bad fsid on block %llu\n",
 432                        (unsigned long long)eb->start);
 433                 ret = -EIO;
 434                 goto err;
 435         }
 436         found_level = btrfs_header_level(eb);
 437
 438         btrfs_set_buffer_lockdep_class(eb, found_level);
 439
 440         ret = csum_tree_block(root, eb, 1);
 441         if (ret)
 442                 ret = -EIO;
 443
 444         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 445         end = eb->start + end - 1;
 446 err:
 447         free_extent_buffer(eb);
 448 out:
 449         return ret;
 450 }
 451
 452 static void end_workqueue_bio(struct bio *bio, int err)
 453 {
 454         struct end_io_wq *end_io_wq = bio->bi_private;
 455         struct btrfs_fs_info *fs_info;
 456
 457         fs_info = end_io_wq->info;
 458         end_io_wq->error = err;
 459         end_io_wq->work.func = end_workqueue_fn;
 460         end_io_wq->work.flags = 0;
 461
 462         if (bio->bi_rw & (1 << BIO_RW)) {
 463                 if (end_io_wq->metadata)
 464                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
 465                                            &end_io_wq->work);
 466                 else
 467                         btrfs_queue_worker(&fs_info->endio_write_workers,
 468                                            &end_io_wq->work);
 469         } else {
 470                 if (end_io_wq->metadata)
 471                         btrfs_queue_worker(&fs_info->endio_meta_workers,
 472                                            &end_io_wq->work);
 473                 else
 474                         btrfs_queue_worker(&fs_info->endio_workers,
 475                                            &end_io_wq->work);
 476         }
 477 }
 478
 479 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 480                         int metadata)
 481 {
 482         struct end_io_wq *end_io_wq;
 483         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 484         if (!end_io_wq)
 485                 return -ENOMEM;
 486
 487         end_io_wq->private = bio->bi_private;
 488         end_io_wq->end_io = bio->bi_end_io;
 489         end_io_wq->info = info;
 490         end_io_wq->error = 0;
 491         end_io_wq->bio = bio;
 492         end_io_wq->metadata = metadata;
 493
 494         bio->bi_private = end_io_wq;
 495         bio->bi_end_io = end_workqueue_bio;
 496         return 0;
 497 }
 498
 499 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 500 {
 501         unsigned long limit = min_t(unsigned long,
 502                                     info->workers.max_workers,
 503                                     info->fs_devices->open_devices);
 504         return 256 * limit;
 505 }
 506
 507 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
 508 {
 509         return atomic_read(&info->nr_async_bios) >
 510                 btrfs_async_submit_limit(info);
 511 }
 512
 513 static void run_one_async_start(struct btrfs_work *work)
 514 {
 515         struct btrfs_fs_info *fs_info;
 516         struct async_submit_bio *async;
 517
 518         async = container_of(work, struct  async_submit_bio, work);
 519         fs_info = BTRFS_I(async->inode)->root->fs_info;
 520         async->submit_bio_start(async->inode, async->rw, async->bio,
 521                                async->mirror_num, async->bio_flags);
 522 }
 523
 524 static void run_one_async_done(struct btrfs_work *work)
 525 {
 526         struct btrfs_fs_info *fs_info;
 527         struct async_submit_bio *async;
 528         int limit;
 529
 530         async = container_of(work, struct  async_submit_bio, work);
 531         fs_info = BTRFS_I(async->inode)->root->fs_info;
 532
 533         limit = btrfs_async_submit_limit(fs_info);
 534         limit = limit * 2 / 3;
 535
 536         atomic_dec(&fs_info->nr_async_submits);
 537
 538         if (atomic_read(&fs_info->nr_async_submits) < limit &&
 539             waitqueue_active(&fs_info->async_submit_wait))
 540                 wake_up(&fs_info->async_submit_wait);
 541
 542         async->submit_bio_done(async->inode, async->rw, async->bio,
 543                                async->mirror_num, async->bio_flags);
 544 }
 545
 546 static void run_one_async_free(struct btrfs_work *work)
 547 {
 548         struct async_submit_bio *async;
 549
 550         async = container_of(work, struct  async_submit_bio, work);
 551         kfree(async);
 552 }
 553
 554 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 555                         int rw, struct bio *bio, int mirror_num,
 556                         unsigned long bio_flags,
 557                         extent_submit_bio_hook_t *submit_bio_start,
 558                         extent_submit_bio_hook_t *submit_bio_done)
 559 {
 560         struct async_submit_bio *async;
 561
 562         async = kmalloc(sizeof(*async), GFP_NOFS);
 563         if (!async)
 564                 return -ENOMEM;
 565
 566         async->inode = inode;
 567         async->rw = rw;
 568         async->bio = bio;
 569         async->mirror_num = mirror_num;
 570         async->submit_bio_start = submit_bio_start;
 571         async->submit_bio_done = submit_bio_done;
 572
 573         async->work.func = run_one_async_start;
 574         async->work.ordered_func = run_one_async_done;
 575         async->work.ordered_free = run_one_async_free;
 576
 577         async->work.flags = 0;
 578         async->bio_flags = bio_flags;
 579
 580         atomic_inc(&fs_info->nr_async_submits);
 581         btrfs_queue_worker(&fs_info->workers, &async->work);
 582 #if 0
 583         int limit = btrfs_async_submit_limit(fs_info);
 584         if (atomic_read(&fs_info->nr_async_submits) > limit) {
 585                 wait_event_timeout(fs_info->async_submit_wait,
 586                            (atomic_read(&fs_info->nr_async_submits) < limit),
 587                            HZ/10);
 588
 589                 wait_event_timeout(fs_info->async_submit_wait,
 590                            (atomic_read(&fs_info->nr_async_bios) < limit),
 591                            HZ/10);
 592         }
 593 #endif
 594         while (atomic_read(&fs_info->async_submit_draining) &&
 595               atomic_read(&fs_info->nr_async_submits)) {
 596                 wait_event(fs_info->async_submit_wait,
 597                            (atomic_read(&fs_info->nr_async_submits) == 0));
 598         }
 599
 600         return 0;
 601 }
 602
 603 static int btree_csum_one_bio(struct bio *bio)
 604 {
 605         struct bio_vec *bvec = bio->bi_io_vec;
 606         int bio_index = 0;
 607         struct btrfs_root *root;
 608
 609         WARN_ON(bio->bi_vcnt <= 0);
 610         while (bio_index < bio->bi_vcnt) {
 611                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 612                 csum_dirty_buffer(root, bvec->bv_page);
 613                 bio_index++;
 614                 bvec++;
 615         }
 616         return 0;
 617 }
 618
 619 static int __btree_submit_bio_start(struct inode *inode, int rw,
 620                                     struct bio *bio, int mirror_num,
 621                                     unsigned long bio_flags)
 622 {
 623         /*
 624          * when we're called for a write, we're already in the async
 625          * submission context.  Just jump into btrfs_map_bio
 626          */
 627         btree_csum_one_bio(bio);
 628         return 0;
 629 }
 630
 631 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 632                                  int mirror_num, unsigned long bio_flags)
 633 {
 634         /*
 635          * when we're called for a write, we're already in the async
 636          * submission context.  Just jump into btrfs_map_bio
 637          */
 638         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 639 }
 640
 641 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 642                                  int mirror_num, unsigned long bio_flags)
 643 {
 644         int ret;
 645
 646         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 647                                           bio, 1);
 648         BUG_ON(ret);
 649
 650         if (!(rw & (1 << BIO_RW))) {
 651                 /*
 652                  * called for a read, do the setup so that checksum validation
 653                  * can happen in the async kernel threads
 654                  */
 655                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 656                                      mirror_num, 0);
 657         }
 658         /*
 659          * kthread helpers are used to submit writes so that checksumming
 660          * can happen in parallel across all CPUs
 661          */
 662         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 663                                    inode, rw, bio, mirror_num, 0,
 664                                    __btree_submit_bio_start,
 665                                    __btree_submit_bio_done);
 666 }
 667
 668 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 669 {
 670         struct extent_io_tree *tree;
 671         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 672         struct extent_buffer *eb;
 673         int was_dirty;
 674
 675         tree = &BTRFS_I(page->mapping->host)->io_tree;
 676         if (!(current->flags & PF_MEMALLOC)) {
 677                 return extent_write_full_page(tree, page,
 678                                               btree_get_extent, wbc);
 679         }
 680
 681         redirty_page_for_writepage(wbc, page);
 682         eb = btrfs_find_tree_block(root, page_offset(page),
 683                                       PAGE_CACHE_SIZE);
 684         WARN_ON(!eb);
 685
 686         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
 687         if (!was_dirty) {
 688                 spin_lock(&root->fs_info->delalloc_lock);
 689                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
 690                 spin_unlock(&root->fs_info->delalloc_lock);
 691         }
 692         free_extent_buffer(eb);
 693
 694         unlock_page(page);
 695         return 0;
 696 }
 697
 698 static int btree_writepages(struct address_space *mapping,
 699                             struct writeback_control *wbc)
 700 {
 701         struct extent_io_tree *tree;
 702         tree = &BTRFS_I(mapping->host)->io_tree;
 703         if (wbc->sync_mode == WB_SYNC_NONE) {
 704                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
 705                 u64 num_dirty;
 706                 unsigned long thresh = 32 * 1024 * 1024;
 707
 708                 if (wbc->for_kupdate)
 709                         return 0;
 710
 711                 /* this is a bit racy, but that's ok */
 712                 num_dirty = root->fs_info->dirty_metadata_bytes;
 713                 if (num_dirty < thresh)
 714                         return 0;
 715         }
 716         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 717 }
 718
 719 static int btree_readpage(struct file *file, struct page *page)
 720 {
 721         struct extent_io_tree *tree;
 722         tree = &BTRFS_I(page->mapping->host)->io_tree;
 723         return extent_read_full_page(tree, page, btree_get_extent);
 724 }
 725
 726 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 727 {
 728         struct extent_io_tree *tree;
 729         struct extent_map_tree *map;
 730         int ret;
 731
 732         if (PageWriteback(page) || PageDirty(page))
 733                 return 0;
 734
 735         tree = &BTRFS_I(page->mapping->host)->io_tree;
 736         map = &BTRFS_I(page->mapping->host)->extent_tree;
 737
 738         ret = try_release_extent_state(map, tree, page, gfp_flags);
 739         if (!ret)
 740                 return 0;
 741
 742         ret = try_release_extent_buffer(tree, page);
 743         if (ret == 1) {
 744                 ClearPagePrivate(page);
 745                 set_page_private(page, 0);
 746                 page_cache_release(page);
 747         }
 748
 749         return ret;
 750 }
 751
 752 static void btree_invalidatepage(struct page *page, unsigned long offset)
 753 {
 754         struct extent_io_tree *tree;
 755         tree = &BTRFS_I(page->mapping->host)->io_tree;
 756         extent_invalidatepage(tree, page, offset);
 757         btree_releasepage(page, GFP_NOFS);
 758         if (PagePrivate(page)) {
 759                 printk(KERN_WARNING "btrfs warning page private not zero "
 760                        "on page %llu\n", (unsigned long long)page_offset(page));
 761                 ClearPagePrivate(page);
 762                 set_page_private(page, 0);
 763                 page_cache_release(page);
 764         }
 765 }
 766
 767 #if 0
 768 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 769 {
 770         struct buffer_head *bh;
 771         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 772         struct buffer_head *head;
 773         if (!page_has_buffers(page)) {
 774                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 775                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 776         }
 777         head = page_buffers(page);
 778         bh = head;
 779         do {
 780                 if (buffer_dirty(bh))
 781                         csum_tree_block(root, bh, 0);
 782                 bh = bh->b_this_page;
 783         } while (bh != head);
 784         return block_write_full_page(page, btree_get_block, wbc);
 785 }
 786 #endif
 787
 788 static struct address_space_operations btree_aops = {
 789         .readpage       = btree_readpage,
 790         .writepage      = btree_writepage,
 791         .writepages     = btree_writepages,
 792         .releasepage    = btree_releasepage,
 793         .invalidatepage = btree_invalidatepage,
 794         .sync_page      = block_sync_page,
 795 };
 796
 797 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 798                          u64 parent_transid)
 799 {
 800         struct extent_buffer *buf = NULL;
 801         struct inode *btree_inode = root->fs_info->btree_inode;
 802         int ret = 0;
 803
 804         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 805         if (!buf)
 806                 return 0;
 807         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 808                                  buf, 0, 0, btree_get_extent, 0);
 809         free_extent_buffer(buf);
 810         return ret;
 811 }
 812
 813 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 814                                             u64 bytenr, u32 blocksize)
 815 {
 816         struct inode *btree_inode = root->fs_info->btree_inode;
 817         struct extent_buffer *eb;
 818         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 819                                 bytenr, blocksize, GFP_NOFS);
 820         return eb;
 821 }
 822
 823 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 824                                                  u64 bytenr, u32 blocksize)
 825 {
 826         struct inode *btree_inode = root->fs_info->btree_inode;
 827         struct extent_buffer *eb;
 828
 829         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 830                                  bytenr, blocksize, NULL, GFP_NOFS);
 831         return eb;
 832 }
 833
 834
 835 int btrfs_write_tree_block(struct extent_buffer *buf)
 836 {
 837         return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
 838                                       buf->start + buf->len - 1, WB_SYNC_ALL);
 839 }
 840
 841 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 842 {
 843         return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
 844                                   buf->start, buf->start + buf->len - 1);
 845 }
 846
 847 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 848                                       u32 blocksize, u64 parent_transid)
 849 {
 850         struct extent_buffer *buf = NULL;
 851         struct inode *btree_inode = root->fs_info->btree_inode;
 852         struct extent_io_tree *io_tree;
 853         int ret;
 854
 855         io_tree = &BTRFS_I(btree_inode)->io_tree;
 856
 857         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 858         if (!buf)
 859                 return NULL;
 860
 861         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 862
 863         if (ret == 0)
 864                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
 865         else
 866                 WARN_ON(1);
 867         return buf;
 868
 869 }
 870
 871 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 872                      struct extent_buffer *buf)
 873 {
 874         struct inode *btree_inode = root->fs_info->btree_inode;
 875         if (btrfs_header_generation(buf) ==
 876             root->fs_info->running_transaction->transid) {
 877                 btrfs_assert_tree_locked(buf);
 878
 879                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
 880                         spin_lock(&root->fs_info->delalloc_lock);
 881                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
 882                                 root->fs_info->dirty_metadata_bytes -= buf->len;
 883                         else
 884                                 WARN_ON(1);
 885                         spin_unlock(&root->fs_info->delalloc_lock);
 886                 }
 887
 888                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
 889                 btrfs_set_lock_blocking(buf);
 890                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 891                                           buf);
 892         }
 893         return 0;
 894 }
 895
 896 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 897                         u32 stripesize, struct btrfs_root *root,
 898                         struct btrfs_fs_info *fs_info,
 899                         u64 objectid)
 900 {
 901         root->node = NULL;
 902         root->commit_root = NULL;
 903         root->ref_tree = NULL;
 904         root->sectorsize = sectorsize;
 905         root->nodesize = nodesize;
 906         root->leafsize = leafsize;
 907         root->stripesize = stripesize;
 908         root->ref_cows = 0;
 909         root->track_dirty = 0;
 910
 911         root->fs_info = fs_info;
 912         root->objectid = objectid;
 913         root->last_trans = 0;
 914         root->highest_inode = 0;
 915         root->last_inode_alloc = 0;
 916         root->name = NULL;
 917         root->in_sysfs = 0;
 918
 919         INIT_LIST_HEAD(&root->dirty_list);
 920         INIT_LIST_HEAD(&root->orphan_list);
 921         INIT_LIST_HEAD(&root->dead_list);
 922         spin_lock_init(&root->node_lock);
 923         spin_lock_init(&root->list_lock);
 924         mutex_init(&root->objectid_mutex);
 925         mutex_init(&root->log_mutex);
 926         init_waitqueue_head(&root->log_writer_wait);
 927         init_waitqueue_head(&root->log_commit_wait[0]);
 928         init_waitqueue_head(&root->log_commit_wait[1]);
 929         atomic_set(&root->log_commit[0], 0);
 930         atomic_set(&root->log_commit[1], 0);
 931         atomic_set(&root->log_writers, 0);
 932         root->log_batch = 0;
 933         root->log_transid = 0;
 934         extent_io_tree_init(&root->dirty_log_pages,
 935                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 936
 937         btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
 938         root->ref_tree = &root->ref_tree_struct;
 939
 940         memset(&root->root_key, 0, sizeof(root->root_key));
 941         memset(&root->root_item, 0, sizeof(root->root_item));
 942         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 943         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 944         root->defrag_trans_start = fs_info->generation;
 945         init_completion(&root->kobj_unregister);
 946         root->defrag_running = 0;
 947         root->defrag_level = 0;
 948         root->root_key.objectid = objectid;
 949         root->anon_super.s_root = NULL;
 950         root->anon_super.s_dev = 0;
 951         INIT_LIST_HEAD(&root->anon_super.s_list);
 952         INIT_LIST_HEAD(&root->anon_super.s_instances);
 953         init_rwsem(&root->anon_super.s_umount);
 954
 955         return 0;
 956 }
 957
 958 static int find_and_setup_root(struct btrfs_root *tree_root,
 959                                struct btrfs_fs_info *fs_info,
 960                                u64 objectid,
 961                                struct btrfs_root *root)
 962 {
 963         int ret;
 964         u32 blocksize;
 965         u64 generation;
 966
 967         __setup_root(tree_root->nodesize, tree_root->leafsize,
 968                      tree_root->sectorsize, tree_root->stripesize,
 969                      root, fs_info, objectid);
 970         ret = btrfs_find_last_root(tree_root, objectid,
 971                                    &root->root_item, &root->root_key);
 972         BUG_ON(ret);
 973
 974         generation = btrfs_root_generation(&root->root_item);
 975         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 976         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 977                                      blocksize, generation);
 978         BUG_ON(!root->node);
 979         return 0;
 980 }
 981
 982 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
 983                              struct btrfs_fs_info *fs_info)
 984 {
 985         struct extent_buffer *eb;
 986         struct btrfs_root *log_root_tree = fs_info->log_root_tree;
 987         u64 start = 0;
 988         u64 end = 0;
 989         int ret;
 990
 991         if (!log_root_tree)
 992                 return 0;
 993
 994         while (1) {
 995                 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
 996                                     0, &start, &end, EXTENT_DIRTY);
 997                 if (ret)
 998                         break;
 999
1000                 clear_extent_dirty(&log_root_tree->dirty_log_pages,
1001                                    start, end, GFP_NOFS);
1002         }
1003         eb = fs_info->log_root_tree->node;
1004
1005         WARN_ON(btrfs_header_level(eb) != 0);
1006         WARN_ON(btrfs_header_nritems(eb) != 0);
1007
1008         ret = btrfs_free_reserved_extent(fs_info->tree_root,
1009                                 eb->start, eb->len);
1010         BUG_ON(ret);
1011
1012         free_extent_buffer(eb);
1013         kfree(fs_info->log_root_tree);
1014         fs_info->log_root_tree = NULL;
1015         return 0;
1016 }
1017
1018 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1019                                          struct btrfs_fs_info *fs_info)
1020 {
1021         struct btrfs_root *root;
1022         struct btrfs_root *tree_root = fs_info->tree_root;
1023         struct extent_buffer *leaf;
1024
1025         root = kzalloc(sizeof(*root), GFP_NOFS);
1026         if (!root)
1027                 return ERR_PTR(-ENOMEM);
1028
1029         __setup_root(tree_root->nodesize, tree_root->leafsize,
1030                      tree_root->sectorsize, tree_root->stripesize,
1031                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1032
1033         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1034         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1035         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1036         /*
1037          * log trees do not get reference counted because they go away
1038          * before a real commit is actually done.  They do store pointers
1039          * to file data extents, and those reference counts still get
1040          * updated (along with back refs to the log tree).
1041          */
1042         root->ref_cows = 0;
1043
1044         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1045                                       0, BTRFS_TREE_LOG_OBJECTID,
1046                                       trans->transid, 0, 0, 0);
1047         if (IS_ERR(leaf)) {
1048                 kfree(root);
1049                 return ERR_CAST(leaf);
1050         }
1051
1052         root->node = leaf;
1053         btrfs_set_header_nritems(root->node, 0);
1054         btrfs_set_header_level(root->node, 0);
1055         btrfs_set_header_bytenr(root->node, root->node->start);
1056         btrfs_set_header_generation(root->node, trans->transid);
1057         btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
1058
1059         write_extent_buffer(root->node, root->fs_info->fsid,
1060                             (unsigned long)btrfs_header_fsid(root->node),
1061                             BTRFS_FSID_SIZE);
1062         btrfs_mark_buffer_dirty(root->node);
1063         btrfs_tree_unlock(root->node);
1064         return root;
1065 }
1066
1067 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1068                              struct btrfs_fs_info *fs_info)
1069 {
1070         struct btrfs_root *log_root;
1071
1072         log_root = alloc_log_tree(trans, fs_info);
1073         if (IS_ERR(log_root))
1074                 return PTR_ERR(log_root);
1075         WARN_ON(fs_info->log_root_tree);
1076         fs_info->log_root_tree = log_root;
1077         return 0;
1078 }
1079
1080 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1081                        struct btrfs_root *root)
1082 {
1083         struct btrfs_root *log_root;
1084         struct btrfs_inode_item *inode_item;
1085
1086         log_root = alloc_log_tree(trans, root->fs_info);
1087         if (IS_ERR(log_root))
1088                 return PTR_ERR(log_root);
1089
1090         log_root->last_trans = trans->transid;
1091         log_root->root_key.offset = root->root_key.objectid;
1092
1093         inode_item = &log_root->root_item.inode;
1094         inode_item->generation = cpu_to_le64(1);
1095         inode_item->size = cpu_to_le64(3);
1096         inode_item->nlink = cpu_to_le32(1);
1097         inode_item->nbytes = cpu_to_le64(root->leafsize);
1098         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1099
1100         btrfs_set_root_bytenr(&log_root->root_item, log_root->node->start);
1101         btrfs_set_root_generation(&log_root->root_item, trans->transid);
1102
1103         WARN_ON(root->log_root);
1104         root->log_root = log_root;
1105         root->log_transid = 0;
1106         return 0;
1107 }
1108
1109 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1110                                                struct btrfs_key *location)
1111 {
1112         struct btrfs_root *root;
1113         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1114         struct btrfs_path *path;
1115         struct extent_buffer *l;
1116         u64 highest_inode;
1117         u64 generation;
1118         u32 blocksize;
1119         int ret = 0;
1120
1121         root = kzalloc(sizeof(*root), GFP_NOFS);
1122         if (!root)
1123                 return ERR_PTR(-ENOMEM);
1124         if (location->offset == (u64)-1) {
1125                 ret = find_and_setup_root(tree_root, fs_info,
1126                                           location->objectid, root);
1127                 if (ret) {
1128                         kfree(root);
1129                         return ERR_PTR(ret);
1130                 }
1131                 goto insert;
1132         }
1133
1134         __setup_root(tree_root->nodesize, tree_root->leafsize,
1135                      tree_root->sectorsize, tree_root->stripesize,
1136                      root, fs_info, location->objectid);
1137
1138         path = btrfs_alloc_path();
1139         BUG_ON(!path);
1140         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1141         if (ret != 0) {
1142                 if (ret > 0)
1143                         ret = -ENOENT;
1144                 goto out;
1145         }
1146         l = path->nodes[0];
1147         read_extent_buffer(l, &root->root_item,
1148                btrfs_item_ptr_offset(l, path->slots[0]),
1149                sizeof(root->root_item));
1150         memcpy(&root->root_key, location, sizeof(*location));
1151         ret = 0;
1152 out:
1153         btrfs_release_path(root, path);
1154         btrfs_free_path(path);
1155         if (ret) {
1156                 kfree(root);
1157                 return ERR_PTR(ret);
1158         }
1159         generation = btrfs_root_generation(&root->root_item);
1160         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1161         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1162                                      blocksize, generation);
1163         BUG_ON(!root->node);
1164 insert:
1165         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1166                 root->ref_cows = 1;
1167                 ret = btrfs_find_highest_inode(root, &highest_inode);
1168                 if (ret == 0) {
1169                         root->highest_inode = highest_inode;
1170                         root->last_inode_alloc = highest_inode;
1171                 }
1172         }
1173         return root;
1174 }
1175
1176 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1177                                         u64 root_objectid)
1178 {
1179         struct btrfs_root *root;
1180
1181         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1182                 return fs_info->tree_root;
1183         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1184                 return fs_info->extent_root;
1185
1186         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1187                                  (unsigned long)root_objectid);
1188         return root;
1189 }
1190
1191 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1192                                               struct btrfs_key *location)
1193 {
1194         struct btrfs_root *root;
1195         int ret;
1196
1197         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1198                 return fs_info->tree_root;
1199         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1200                 return fs_info->extent_root;
1201         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1202                 return fs_info->chunk_root;
1203         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1204                 return fs_info->dev_root;
1205         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1206                 return fs_info->csum_root;
1207
1208         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1209                                  (unsigned long)location->objectid);
1210         if (root)
1211                 return root;
1212
1213         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1214         if (IS_ERR(root))
1215                 return root;
1216
1217         set_anon_super(&root->anon_super, NULL);
1218
1219         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1220                                 (unsigned long)root->root_key.objectid,
1221                                 root);
1222         if (ret) {
1223                 free_extent_buffer(root->node);
1224                 kfree(root);
1225                 return ERR_PTR(ret);
1226         }
1227         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
1228                 ret = btrfs_find_dead_roots(fs_info->tree_root,
1229                                             root->root_key.objectid, root);
1230                 BUG_ON(ret);
1231                 btrfs_orphan_cleanup(root);
1232         }
1233         return root;
1234 }
1235
1236 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1237                                       struct btrfs_key *location,
1238                                       const char *name, int namelen)
1239 {
1240         struct btrfs_root *root;
1241         int ret;
1242
1243         root = btrfs_read_fs_root_no_name(fs_info, location);
1244         if (!root)
1245                 return NULL;
1246
1247         if (root->in_sysfs)
1248                 return root;
1249
1250         ret = btrfs_set_root_name(root, name, namelen);
1251         if (ret) {
1252                 free_extent_buffer(root->node);
1253                 kfree(root);
1254                 return ERR_PTR(ret);
1255         }
1256 #if 0
1257         ret = btrfs_sysfs_add_root(root);
1258         if (ret) {
1259                 free_extent_buffer(root->node);
1260                 kfree(root->name);
1261                 kfree(root);
1262                 return ERR_PTR(ret);
1263         }
1264 #endif
1265         root->in_sysfs = 1;
1266         return root;
1267 }
1268
1269 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1270 {
1271         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1272         int ret = 0;
1273         struct btrfs_device *device;
1274         struct backing_dev_info *bdi;
1275 #if 0
1276         if ((bdi_bits & (1 << BDI_write_congested)) &&
1277             btrfs_congested_async(info, 0))
1278                 return 1;
1279 #endif
1280         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1281                 if (!device->bdev)
1282                         continue;
1283                 bdi = blk_get_backing_dev_info(device->bdev);
1284                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1285                         ret = 1;
1286                         break;
1287                 }
1288         }
1289         return ret;
1290 }
1291
1292 /*
1293  * this unplugs every device on the box, and it is only used when page
1294  * is null
1295  */
1296 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1297 {
1298         struct btrfs_device *device;
1299         struct btrfs_fs_info *info;
1300
1301         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1302         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1303                 if (!device->bdev)
1304                         continue;
1305
1306                 bdi = blk_get_backing_dev_info(device->bdev);
1307                 if (bdi->unplug_io_fn)
1308                         bdi->unplug_io_fn(bdi, page);
1309         }
1310 }
1311
1312 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1313 {
1314         struct inode *inode;
1315         struct extent_map_tree *em_tree;
1316         struct extent_map *em;
1317         struct address_space *mapping;
1318         u64 offset;
1319
1320         /* the generic O_DIRECT read code does this */
1321         if (1 || !page) {
1322                 __unplug_io_fn(bdi, page);
1323                 return;
1324         }
1325
1326         /*
1327          * page->mapping may change at any time.  Get a consistent copy
1328          * and use that for everything below
1329          */
1330         smp_mb();
1331         mapping = page->mapping;
1332         if (!mapping)
1333                 return;
1334
1335         inode = mapping->host;
1336
1337         /*
1338          * don't do the expensive searching for a small number of
1339          * devices
1340          */
1341         if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1342                 __unplug_io_fn(bdi, page);
1343                 return;
1344         }
1345
1346         offset = page_offset(page);
1347
1348         em_tree = &BTRFS_I(inode)->extent_tree;
1349         spin_lock(&em_tree->lock);
1350         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1351         spin_unlock(&em_tree->lock);
1352         if (!em) {
1353                 __unplug_io_fn(bdi, page);
1354                 return;
1355         }
1356
1357         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1358                 free_extent_map(em);
1359                 __unplug_io_fn(bdi, page);
1360                 return;
1361         }
1362         offset = offset - em->start;
1363         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1364                           em->block_start + offset, page);
1365         free_extent_map(em);
1366 }
1367
1368 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1369 {
1370         bdi_init(bdi);
1371         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1372         bdi->state              = 0;
1373         bdi->capabilities       = default_backing_dev_info.capabilities;
1374         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1375         bdi->unplug_io_data     = info;
1376         bdi->congested_fn       = btrfs_congested_fn;
1377         bdi->congested_data     = info;
1378         return 0;
1379 }
1380
1381 static int bio_ready_for_csum(struct bio *bio)
1382 {
1383         u64 length = 0;
1384         u64 buf_len = 0;
1385         u64 start = 0;
1386         struct page *page;
1387         struct extent_io_tree *io_tree = NULL;
1388         struct btrfs_fs_info *info = NULL;
1389         struct bio_vec *bvec;
1390         int i;
1391         int ret;
1392
1393         bio_for_each_segment(bvec, bio, i) {
1394                 page = bvec->bv_page;
1395                 if (page->private == EXTENT_PAGE_PRIVATE) {
1396                         length += bvec->bv_len;
1397                         continue;
1398                 }
1399                 if (!page->private) {
1400                         length += bvec->bv_len;
1401                         continue;
1402                 }
1403                 length = bvec->bv_len;
1404                 buf_len = page->private >> 2;
1405                 start = page_offset(page) + bvec->bv_offset;
1406                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1407                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1408         }
1409         /* are we fully contained in this bio? */
1410         if (buf_len <= length)
1411                 return 1;
1412
1413         ret = extent_range_uptodate(io_tree, start + length,
1414                                     start + buf_len - 1);
1415         if (ret == 1)
1416                 return ret;
1417         return ret;
1418 }
1419
1420 /*
1421  * called by the kthread helper functions to finally call the bio end_io
1422  * functions.  This is where read checksum verification actually happens
1423  */
1424 static void end_workqueue_fn(struct btrfs_work *work)
1425 {
1426         struct bio *bio;
1427         struct end_io_wq *end_io_wq;
1428         struct btrfs_fs_info *fs_info;
1429         int error;
1430
1431         end_io_wq = container_of(work, struct end_io_wq, work);
1432         bio = end_io_wq->bio;
1433         fs_info = end_io_wq->info;
1434
1435         /* metadata bio reads are special because the whole tree block must
1436          * be checksummed at once.  This makes sure the entire block is in
1437          * ram and up to date before trying to verify things.  For
1438          * blocksize <= pagesize, it is basically a noop
1439          */
1440         if (!(bio->bi_rw & (1 << BIO_RW)) && end_io_wq->metadata &&
1441             !bio_ready_for_csum(bio)) {
1442                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1443                                    &end_io_wq->work);
1444                 return;
1445         }
1446         error = end_io_wq->error;
1447         bio->bi_private = end_io_wq->private;
1448         bio->bi_end_io = end_io_wq->end_io;
1449         kfree(end_io_wq);
1450         bio_endio(bio, error);
1451 }
1452
1453 static int cleaner_kthread(void *arg)
1454 {
1455         struct btrfs_root *root = arg;
1456
1457         do {
1458                 smp_mb();
1459                 if (root->fs_info->closing)
1460                         break;
1461
1462                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1463                 mutex_lock(&root->fs_info->cleaner_mutex);
1464                 btrfs_clean_old_snapshots(root);
1465                 mutex_unlock(&root->fs_info->cleaner_mutex);
1466
1467                 if (freezing(current)) {
1468                         refrigerator();
1469                 } else {
1470                         smp_mb();
1471                         if (root->fs_info->closing)
1472                                 break;
1473                         set_current_state(TASK_INTERRUPTIBLE);
1474                         schedule();
1475                         __set_current_state(TASK_RUNNING);
1476                 }
1477         } while (!kthread_should_stop());
1478         return 0;
1479 }
1480
1481 static int transaction_kthread(void *arg)
1482 {
1483         struct btrfs_root *root = arg;
1484         struct btrfs_trans_handle *trans;
1485         struct btrfs_transaction *cur;
1486         unsigned long now;
1487         unsigned long delay;
1488         int ret;
1489
1490         do {
1491                 smp_mb();
1492                 if (root->fs_info->closing)
1493                         break;
1494
1495                 delay = HZ * 30;
1496                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1497                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1498
1499                 mutex_lock(&root->fs_info->trans_mutex);
1500                 cur = root->fs_info->running_transaction;
1501                 if (!cur) {
1502                         mutex_unlock(&root->fs_info->trans_mutex);
1503                         goto sleep;
1504                 }
1505
1506                 now = get_seconds();
1507                 if (now < cur->start_time || now - cur->start_time < 30) {
1508                         mutex_unlock(&root->fs_info->trans_mutex);
1509                         delay = HZ * 5;
1510                         goto sleep;
1511                 }
1512                 mutex_unlock(&root->fs_info->trans_mutex);
1513                 trans = btrfs_start_transaction(root, 1);
1514                 ret = btrfs_commit_transaction(trans, root);
1515
1516 sleep:
1517                 wake_up_process(root->fs_info->cleaner_kthread);
1518                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1519
1520                 if (freezing(current)) {
1521                         refrigerator();
1522                 } else {
1523                         if (root->fs_info->closing)
1524                                 break;
1525                         set_current_state(TASK_INTERRUPTIBLE);
1526                         schedule_timeout(delay);
1527                         __set_current_state(TASK_RUNNING);
1528                 }
1529         } while (!kthread_should_stop());
1530         return 0;
1531 }
1532
1533 struct btrfs_root *open_ctree(struct super_block *sb,
1534                               struct btrfs_fs_devices *fs_devices,
1535                               char *options)
1536 {
1537         u32 sectorsize;
1538         u32 nodesize;
1539         u32 leafsize;
1540         u32 blocksize;
1541         u32 stripesize;
1542         u64 generation;
1543         u64 features;
1544         struct btrfs_key location;
1545         struct buffer_head *bh;
1546         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1547                                                  GFP_NOFS);
1548         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1549                                                  GFP_NOFS);
1550         struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1551                                                GFP_NOFS);
1552         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1553                                                 GFP_NOFS);
1554         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1555                                                 GFP_NOFS);
1556         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1557                                               GFP_NOFS);
1558         struct btrfs_root *log_tree_root;
1559
1560         int ret;
1561         int err = -EINVAL;
1562
1563         struct btrfs_super_block *disk_super;
1564
1565         if (!extent_root || !tree_root || !fs_info ||
1566             !chunk_root || !dev_root || !csum_root) {
1567                 err = -ENOMEM;
1568                 goto fail;
1569         }
1570         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1571         INIT_LIST_HEAD(&fs_info->trans_list);
1572         INIT_LIST_HEAD(&fs_info->dead_roots);
1573         INIT_LIST_HEAD(&fs_info->hashers);
1574         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1575         INIT_LIST_HEAD(&fs_info->ordered_operations);
1576         spin_lock_init(&fs_info->delalloc_lock);
1577         spin_lock_init(&fs_info->new_trans_lock);
1578         spin_lock_init(&fs_info->ref_cache_lock);
1579
1580         init_completion(&fs_info->kobj_unregister);
1581         fs_info->tree_root = tree_root;
1582         fs_info->extent_root = extent_root;
1583         fs_info->csum_root = csum_root;
1584         fs_info->chunk_root = chunk_root;
1585         fs_info->dev_root = dev_root;
1586         fs_info->fs_devices = fs_devices;
1587         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1588         INIT_LIST_HEAD(&fs_info->space_info);
1589         btrfs_mapping_init(&fs_info->mapping_tree);
1590         atomic_set(&fs_info->nr_async_submits, 0);
1591         atomic_set(&fs_info->async_delalloc_pages, 0);
1592         atomic_set(&fs_info->async_submit_draining, 0);
1593         atomic_set(&fs_info->nr_async_bios, 0);
1594         atomic_set(&fs_info->throttles, 0);
1595         atomic_set(&fs_info->throttle_gen, 0);
1596         fs_info->sb = sb;
1597         fs_info->max_extent = (u64)-1;
1598         fs_info->max_inline = 8192 * 1024;
1599         setup_bdi(fs_info, &fs_info->bdi);
1600         fs_info->btree_inode = new_inode(sb);
1601         fs_info->btree_inode->i_ino = 1;
1602         fs_info->btree_inode->i_nlink = 1;
1603
1604         fs_info->thread_pool_size = min_t(unsigned long,
1605                                           num_online_cpus() + 2, 8);
1606
1607         INIT_LIST_HEAD(&fs_info->ordered_extents);
1608         spin_lock_init(&fs_info->ordered_extent_lock);
1609
1610         sb->s_blocksize = 4096;
1611         sb->s_blocksize_bits = blksize_bits(4096);
1612
1613         /*
1614          * we set the i_size on the btree inode to the max possible int.
1615          * the real end of the address space is determined by all of
1616          * the devices in the system
1617          */
1618         fs_info->btree_inode->i_size = OFFSET_MAX;
1619         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1620         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1621
1622         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1623                              fs_info->btree_inode->i_mapping,
1624                              GFP_NOFS);
1625         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1626                              GFP_NOFS);
1627
1628         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1629
1630         spin_lock_init(&fs_info->block_group_cache_lock);
1631         fs_info->block_group_cache_tree.rb_node = NULL;
1632
1633         extent_io_tree_init(&fs_info->pinned_extents,
1634                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1635         fs_info->do_barriers = 1;
1636
1637         INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
1638         btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
1639         btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
1640
1641         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1642         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1643                sizeof(struct btrfs_key));
1644         insert_inode_hash(fs_info->btree_inode);
1645
1646         mutex_init(&fs_info->trans_mutex);
1647         mutex_init(&fs_info->ordered_operations_mutex);
1648         mutex_init(&fs_info->tree_log_mutex);
1649         mutex_init(&fs_info->drop_mutex);
1650         mutex_init(&fs_info->pinned_mutex);
1651         mutex_init(&fs_info->chunk_mutex);
1652         mutex_init(&fs_info->transaction_kthread_mutex);
1653         mutex_init(&fs_info->cleaner_mutex);
1654         mutex_init(&fs_info->volume_mutex);
1655         mutex_init(&fs_info->tree_reloc_mutex);
1656         init_waitqueue_head(&fs_info->transaction_throttle);
1657         init_waitqueue_head(&fs_info->transaction_wait);
1658         init_waitqueue_head(&fs_info->async_submit_wait);
1659
1660         __setup_root(4096, 4096, 4096, 4096, tree_root,
1661                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1662
1663
1664         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1665         if (!bh)
1666                 goto fail_iput;
1667
1668         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1669         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1670                sizeof(fs_info->super_for_commit));
1671         brelse(bh);
1672
1673         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1674
1675         disk_super = &fs_info->super_copy;
1676         if (!btrfs_super_root(disk_super))
1677                 goto fail_iput;
1678
1679         ret = btrfs_parse_options(tree_root, options);
1680         if (ret) {
1681                 err = ret;
1682                 goto fail_iput;
1683         }
1684
1685         features = btrfs_super_incompat_flags(disk_super) &
1686                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1687         if (features) {
1688                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1689                        "unsupported optional features (%Lx).\n",
1690                        features);
1691                 err = -EINVAL;
1692                 goto fail_iput;
1693         }
1694
1695         features = btrfs_super_compat_ro_flags(disk_super) &
1696                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1697         if (!(sb->s_flags & MS_RDONLY) && features) {
1698                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1699                        "unsupported option features (%Lx).\n",
1700                        features);
1701                 err = -EINVAL;
1702                 goto fail_iput;
1703         }
1704
1705         /*
1706          * we need to start all the end_io workers up front because the
1707          * queue work function gets called at interrupt time, and so it
1708          * cannot dynamically grow.
1709          */
1710         btrfs_init_workers(&fs_info->workers, "worker",
1711                            fs_info->thread_pool_size);
1712
1713         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1714                            fs_info->thread_pool_size);
1715
1716         btrfs_init_workers(&fs_info->submit_workers, "submit",
1717                            min_t(u64, fs_devices->num_devices,
1718                            fs_info->thread_pool_size));
1719
1720         /* a higher idle thresh on the submit workers makes it much more
1721          * likely that bios will be send down in a sane order to the
1722          * devices
1723          */
1724         fs_info->submit_workers.idle_thresh = 64;
1725
1726         fs_info->workers.idle_thresh = 16;
1727         fs_info->workers.ordered = 1;
1728
1729         fs_info->delalloc_workers.idle_thresh = 2;
1730         fs_info->delalloc_workers.ordered = 1;
1731
1732         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1733         btrfs_init_workers(&fs_info->endio_workers, "endio",
1734                            fs_info->thread_pool_size);
1735         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1736                            fs_info->thread_pool_size);
1737         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1738                            "endio-meta-write", fs_info->thread_pool_size);
1739         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1740                            fs_info->thread_pool_size);
1741
1742         /*
1743          * endios are largely parallel and should have a very
1744          * low idle thresh
1745          */
1746         fs_info->endio_workers.idle_thresh = 4;
1747         fs_info->endio_meta_workers.idle_thresh = 4;
1748
1749         fs_info->endio_write_workers.idle_thresh = 64;
1750         fs_info->endio_meta_write_workers.idle_thresh = 64;
1751
1752         btrfs_start_workers(&fs_info->workers, 1);
1753         btrfs_start_workers(&fs_info->submit_workers, 1);
1754         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1755         btrfs_start_workers(&fs_info->fixup_workers, 1);
1756         btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1757         btrfs_start_workers(&fs_info->endio_meta_workers,
1758                             fs_info->thread_pool_size);
1759         btrfs_start_workers(&fs_info->endio_meta_write_workers,
1760                             fs_info->thread_pool_size);
1761         btrfs_start_workers(&fs_info->endio_write_workers,
1762                             fs_info->thread_pool_size);
1763
1764         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1765         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1766                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1767
1768         nodesize = btrfs_super_nodesize(disk_super);
1769         leafsize = btrfs_super_leafsize(disk_super);
1770         sectorsize = btrfs_super_sectorsize(disk_super);
1771         stripesize = btrfs_super_stripesize(disk_super);
1772         tree_root->nodesize = nodesize;
1773         tree_root->leafsize = leafsize;
1774         tree_root->sectorsize = sectorsize;
1775         tree_root->stripesize = stripesize;
1776
1777         sb->s_blocksize = sectorsize;
1778         sb->s_blocksize_bits = blksize_bits(sectorsize);
1779
1780         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1781                     sizeof(disk_super->magic))) {
1782                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1783                 goto fail_sb_buffer;
1784         }
1785
1786         mutex_lock(&fs_info->chunk_mutex);
1787         ret = btrfs_read_sys_array(tree_root);
1788         mutex_unlock(&fs_info->chunk_mutex);
1789         if (ret) {
1790                 printk(KERN_WARNING "btrfs: failed to read the system "
1791                        "array on %s\n", sb->s_id);
1792                 goto fail_sys_array;
1793         }
1794
1795         blocksize = btrfs_level_size(tree_root,
1796                                      btrfs_super_chunk_root_level(disk_super));
1797         generation = btrfs_super_chunk_root_generation(disk_super);
1798
1799         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1800                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1801
1802         chunk_root->node = read_tree_block(chunk_root,
1803                                            btrfs_super_chunk_root(disk_super),
1804                                            blocksize, generation);
1805         BUG_ON(!chunk_root->node);
1806
1807         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1808            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1809            BTRFS_UUID_SIZE);
1810
1811         mutex_lock(&fs_info->chunk_mutex);
1812         ret = btrfs_read_chunk_tree(chunk_root);
1813         mutex_unlock(&fs_info->chunk_mutex);
1814         if (ret) {
1815                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1816                        sb->s_id);
1817                 goto fail_chunk_root;
1818         }
1819
1820         btrfs_close_extra_devices(fs_devices);
1821
1822         blocksize = btrfs_level_size(tree_root,
1823                                      btrfs_super_root_level(disk_super));
1824         generation = btrfs_super_generation(disk_super);
1825
1826         tree_root->node = read_tree_block(tree_root,
1827                                           btrfs_super_root(disk_super),
1828                                           blocksize, generation);
1829         if (!tree_root->node)
1830                 goto fail_chunk_root;
1831
1832
1833         ret = find_and_setup_root(tree_root, fs_info,
1834                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1835         if (ret)
1836                 goto fail_tree_root;
1837         extent_root->track_dirty = 1;
1838
1839         ret = find_and_setup_root(tree_root, fs_info,
1840                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1841         dev_root->track_dirty = 1;
1842         if (ret)
1843                 goto fail_extent_root;
1844
1845         ret = find_and_setup_root(tree_root, fs_info,
1846                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
1847         if (ret)
1848                 goto fail_extent_root;
1849
1850         csum_root->track_dirty = 1;
1851
1852         btrfs_read_block_groups(extent_root);
1853
1854         fs_info->generation = generation;
1855         fs_info->last_trans_committed = generation;
1856         fs_info->data_alloc_profile = (u64)-1;
1857         fs_info->metadata_alloc_profile = (u64)-1;
1858         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1859         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1860                                                "btrfs-cleaner");
1861         if (IS_ERR(fs_info->cleaner_kthread))
1862                 goto fail_csum_root;
1863
1864         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1865                                                    tree_root,
1866                                                    "btrfs-transaction");
1867         if (IS_ERR(fs_info->transaction_kthread))
1868                 goto fail_cleaner;
1869
1870         if (btrfs_super_log_root(disk_super) != 0) {
1871                 u64 bytenr = btrfs_super_log_root(disk_super);
1872
1873                 if (fs_devices->rw_devices == 0) {
1874                         printk(KERN_WARNING "Btrfs log replay required "
1875                                "on RO media\n");
1876                         err = -EIO;
1877                         goto fail_trans_kthread;
1878                 }
1879                 blocksize =
1880                      btrfs_level_size(tree_root,
1881                                       btrfs_super_log_root_level(disk_super));
1882
1883                 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1884                                                       GFP_NOFS);
1885
1886                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1887                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1888
1889                 log_tree_root->node = read_tree_block(tree_root, bytenr,
1890                                                       blocksize,
1891                                                       generation + 1);
1892                 ret = btrfs_recover_log_trees(log_tree_root);
1893                 BUG_ON(ret);
1894
1895                 if (sb->s_flags & MS_RDONLY) {
1896                         ret =  btrfs_commit_super(tree_root);
1897                         BUG_ON(ret);
1898                 }
1899         }
1900
1901         if (!(sb->s_flags & MS_RDONLY)) {
1902                 ret = btrfs_cleanup_reloc_trees(tree_root);
1903                 BUG_ON(ret);
1904         }
1905
1906         location.objectid = BTRFS_FS_TREE_OBJECTID;
1907         location.type = BTRFS_ROOT_ITEM_KEY;
1908         location.offset = (u64)-1;
1909
1910         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1911         if (!fs_info->fs_root)
1912                 goto fail_trans_kthread;
1913         return tree_root;
1914
1915 fail_trans_kthread:
1916         kthread_stop(fs_info->transaction_kthread);
1917 fail_cleaner:
1918         kthread_stop(fs_info->cleaner_kthread);
1919
1920         /*
1921          * make sure we're done with the btree inode before we stop our
1922          * kthreads
1923          */
1924         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1925         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1926
1927 fail_csum_root:
1928         free_extent_buffer(csum_root->node);
1929 fail_extent_root:
1930         free_extent_buffer(extent_root->node);
1931 fail_tree_root:
1932         free_extent_buffer(tree_root->node);
1933 fail_chunk_root:
1934         free_extent_buffer(chunk_root->node);
1935 fail_sys_array:
1936         free_extent_buffer(dev_root->node);
1937 fail_sb_buffer:
1938         btrfs_stop_workers(&fs_info->fixup_workers);
1939         btrfs_stop_workers(&fs_info->delalloc_workers);
1940         btrfs_stop_workers(&fs_info->workers);
1941         btrfs_stop_workers(&fs_info->endio_workers);
1942         btrfs_stop_workers(&fs_info->endio_meta_workers);
1943         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
1944         btrfs_stop_workers(&fs_info->endio_write_workers);
1945         btrfs_stop_workers(&fs_info->submit_workers);
1946 fail_iput:
1947         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1948         iput(fs_info->btree_inode);
1949
1950         btrfs_close_devices(fs_info->fs_devices);
1951         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1952         bdi_destroy(&fs_info->bdi);
1953
1954 fail:
1955         kfree(extent_root);
1956         kfree(tree_root);
1957         kfree(fs_info);
1958         kfree(chunk_root);
1959         kfree(dev_root);
1960         kfree(csum_root);
1961         return ERR_PTR(err);
1962 }
1963
1964 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1965 {
1966         char b[BDEVNAME_SIZE];
1967
1968         if (uptodate) {
1969                 set_buffer_uptodate(bh);
1970         } else {
1971                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1972                         printk(KERN_WARNING "lost page write due to "
1973                                         "I/O error on %s\n",
1974                                        bdevname(bh->b_bdev, b));
1975                 }
1976                 /* note, we dont' set_buffer_write_io_error because we have
1977                  * our own ways of dealing with the IO errors
1978                  */
1979                 clear_buffer_uptodate(bh);
1980         }
1981         unlock_buffer(bh);
1982         put_bh(bh);
1983 }
1984
1985 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
1986 {
1987         struct buffer_head *bh;
1988         struct buffer_head *latest = NULL;
1989         struct btrfs_super_block *super;
1990         int i;
1991         u64 transid = 0;
1992         u64 bytenr;
1993
1994         /* we would like to check all the supers, but that would make
1995          * a btrfs mount succeed after a mkfs from a different FS.
1996          * So, we need to add a special mount option to scan for
1997          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1998          */
1999         for (i = 0; i < 1; i++) {
2000                 bytenr = btrfs_sb_offset(i);
2001                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2002                         break;
2003                 bh = __bread(bdev, bytenr / 4096, 4096);
2004                 if (!bh)
2005                         continue;
2006
2007                 super = (struct btrfs_super_block *)bh->b_data;
2008                 if (btrfs_super_bytenr(super) != bytenr ||
2009                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2010                             sizeof(super->magic))) {
2011                         brelse(bh);
2012                         continue;
2013                 }
2014
2015                 if (!latest || btrfs_super_generation(super) > transid) {
2016                         brelse(latest);
2017                         latest = bh;
2018                         transid = btrfs_super_generation(super);
2019                 } else {
2020                         brelse(bh);
2021                 }
2022         }
2023         return latest;
2024 }
2025
2026 static int write_dev_supers(struct btrfs_device *device,
2027                             struct btrfs_super_block *sb,
2028                             int do_barriers, int wait, int max_mirrors)
2029 {
2030         struct buffer_head *bh;
2031         int i;
2032         int ret;
2033         int errors = 0;
2034         u32 crc;
2035         u64 bytenr;
2036         int last_barrier = 0;
2037
2038         if (max_mirrors == 0)
2039                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2040
2041         /* make sure only the last submit_bh does a barrier */
2042         if (do_barriers) {
2043                 for (i = 0; i < max_mirrors; i++) {
2044                         bytenr = btrfs_sb_offset(i);
2045                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2046                             device->total_bytes)
2047                                 break;
2048                         last_barrier = i;
2049                 }
2050         }
2051
2052         for (i = 0; i < max_mirrors; i++) {
2053                 bytenr = btrfs_sb_offset(i);
2054                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2055                         break;
2056
2057                 if (wait) {
2058                         bh = __find_get_block(device->bdev, bytenr / 4096,
2059                                               BTRFS_SUPER_INFO_SIZE);
2060                         BUG_ON(!bh);
2061                         brelse(bh);
2062                         wait_on_buffer(bh);
2063                         if (buffer_uptodate(bh)) {
2064                                 brelse(bh);
2065                                 continue;
2066                         }
2067                 } else {
2068                         btrfs_set_super_bytenr(sb, bytenr);
2069
2070                         crc = ~(u32)0;
2071                         crc = btrfs_csum_data(NULL, (char *)sb +
2072                                               BTRFS_CSUM_SIZE, crc,
2073                                               BTRFS_SUPER_INFO_SIZE -
2074                                               BTRFS_CSUM_SIZE);
2075                         btrfs_csum_final(crc, sb->csum);
2076
2077                         bh = __getblk(device->bdev, bytenr / 4096,
2078                                       BTRFS_SUPER_INFO_SIZE);
2079                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2080
2081                         set_buffer_uptodate(bh);
2082                         get_bh(bh);
2083                         lock_buffer(bh);
2084                         bh->b_end_io = btrfs_end_buffer_write_sync;
2085                 }
2086
2087                 if (i == last_barrier && do_barriers && device->barriers) {
2088                         ret = submit_bh(WRITE_BARRIER, bh);
2089                         if (ret == -EOPNOTSUPP) {
2090                                 printk("btrfs: disabling barriers on dev %s\n",
2091                                        device->name);
2092                                 set_buffer_uptodate(bh);
2093                                 device->barriers = 0;
2094                                 get_bh(bh);
2095                                 lock_buffer(bh);
2096                                 ret = submit_bh(WRITE, bh);
2097                         }
2098                 } else {
2099                         ret = submit_bh(WRITE, bh);
2100                 }
2101
2102                 if (!ret && wait) {
2103                         wait_on_buffer(bh);
2104                         if (!buffer_uptodate(bh))
2105                                 errors++;
2106                 } else if (ret) {
2107                         errors++;
2108                 }
2109                 if (wait)
2110                         brelse(bh);
2111         }
2112         return errors < i ? 0 : -1;
2113 }
2114
2115 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2116 {
2117         struct list_head *head = &root->fs_info->fs_devices->devices;
2118         struct btrfs_device *dev;
2119         struct btrfs_super_block *sb;
2120         struct btrfs_dev_item *dev_item;
2121         int ret;
2122         int do_barriers;
2123         int max_errors;
2124         int total_errors = 0;
2125         u64 flags;
2126
2127         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2128         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2129
2130         sb = &root->fs_info->super_for_commit;
2131         dev_item = &sb->dev_item;
2132         list_for_each_entry(dev, head, dev_list) {
2133                 if (!dev->bdev) {
2134                         total_errors++;
2135                         continue;
2136                 }
2137                 if (!dev->in_fs_metadata || !dev->writeable)
2138                         continue;
2139
2140                 btrfs_set_stack_device_generation(dev_item, 0);
2141                 btrfs_set_stack_device_type(dev_item, dev->type);
2142                 btrfs_set_stack_device_id(dev_item, dev->devid);
2143                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2144                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2145                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2146                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2147                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2148                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2149                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2150
2151                 flags = btrfs_super_flags(sb);
2152                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2153
2154                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2155                 if (ret)
2156                         total_errors++;
2157         }
2158         if (total_errors > max_errors) {
2159                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2160                        total_errors);
2161                 BUG();
2162         }
2163
2164         total_errors = 0;
2165         list_for_each_entry(dev, head, dev_list) {
2166                 if (!dev->bdev)
2167                         continue;
2168                 if (!dev->in_fs_metadata || !dev->writeable)
2169                         continue;
2170
2171                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2172                 if (ret)
2173                         total_errors++;
2174         }
2175         if (total_errors > max_errors) {
2176                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2177                        total_errors);
2178                 BUG();
2179         }
2180         return 0;
2181 }
2182
2183 int write_ctree_super(struct btrfs_trans_handle *trans,
2184                       struct btrfs_root *root, int max_mirrors)
2185 {
2186         int ret;
2187
2188         ret = write_all_supers(root, max_mirrors);
2189         return ret;
2190 }
2191
2192 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2193 {
2194         radix_tree_delete(&fs_info->fs_roots_radix,
2195                           (unsigned long)root->root_key.objectid);
2196         if (root->anon_super.s_dev) {
2197                 down_write(&root->anon_super.s_umount);
2198                 kill_anon_super(&root->anon_super);
2199         }
2200         if (root->node)
2201                 free_extent_buffer(root->node);
2202         if (root->commit_root)
2203                 free_extent_buffer(root->commit_root);
2204         kfree(root->name);
2205         kfree(root);
2206         return 0;
2207 }
2208
2209 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2210 {
2211         int ret;
2212         struct btrfs_root *gang[8];
2213         int i;
2214
2215         while (1) {
2216                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2217                                              (void **)gang, 0,
2218                                              ARRAY_SIZE(gang));
2219                 if (!ret)
2220                         break;
2221                 for (i = 0; i < ret; i++)
2222                         btrfs_free_fs_root(fs_info, gang[i]);
2223         }
2224         return 0;
2225 }
2226
2227 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2228 {
2229         u64 root_objectid = 0;
2230         struct btrfs_root *gang[8];
2231         int i;
2232         int ret;
2233
2234         while (1) {
2235                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2236                                              (void **)gang, root_objectid,
2237                                              ARRAY_SIZE(gang));
2238                 if (!ret)
2239                         break;
2240                 for (i = 0; i < ret; i++) {
2241                         root_objectid = gang[i]->root_key.objectid;
2242                         ret = btrfs_find_dead_roots(fs_info->tree_root,
2243                                                     root_objectid, gang[i]);
2244                         BUG_ON(ret);
2245                         btrfs_orphan_cleanup(gang[i]);
2246                 }
2247                 root_objectid++;
2248         }
2249         return 0;
2250 }
2251
2252 int btrfs_commit_super(struct btrfs_root *root)
2253 {
2254         struct btrfs_trans_handle *trans;
2255         int ret;
2256
2257         mutex_lock(&root->fs_info->cleaner_mutex);
2258         btrfs_clean_old_snapshots(root);
2259         mutex_unlock(&root->fs_info->cleaner_mutex);
2260         trans = btrfs_start_transaction(root, 1);
2261         ret = btrfs_commit_transaction(trans, root);
2262         BUG_ON(ret);
2263         /* run commit again to drop the original snapshot */
2264         trans = btrfs_start_transaction(root, 1);
2265         btrfs_commit_transaction(trans, root);
2266         ret = btrfs_write_and_wait_transaction(NULL, root);
2267         BUG_ON(ret);
2268
2269         ret = write_ctree_super(NULL, root, 0);
2270         return ret;
2271 }
2272
2273 int close_ctree(struct btrfs_root *root)
2274 {
2275         struct btrfs_fs_info *fs_info = root->fs_info;
2276         int ret;
2277
2278         fs_info->closing = 1;
2279         smp_mb();
2280
2281         kthread_stop(root->fs_info->transaction_kthread);
2282         kthread_stop(root->fs_info->cleaner_kthread);
2283
2284         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2285                 ret =  btrfs_commit_super(root);
2286                 if (ret)
2287                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2288         }
2289
2290         if (fs_info->delalloc_bytes) {
2291                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2292                        fs_info->delalloc_bytes);
2293         }
2294         if (fs_info->total_ref_cache_size) {
2295                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2296                        (unsigned long long)fs_info->total_ref_cache_size);
2297         }
2298
2299         if (fs_info->extent_root->node)
2300                 free_extent_buffer(fs_info->extent_root->node);
2301
2302         if (fs_info->tree_root->node)
2303                 free_extent_buffer(fs_info->tree_root->node);
2304
2305         if (root->fs_info->chunk_root->node)
2306                 free_extent_buffer(root->fs_info->chunk_root->node);
2307
2308         if (root->fs_info->dev_root->node)
2309                 free_extent_buffer(root->fs_info->dev_root->node);
2310
2311         if (root->fs_info->csum_root->node)
2312                 free_extent_buffer(root->fs_info->csum_root->node);
2313
2314         btrfs_free_block_groups(root->fs_info);
2315
2316         del_fs_roots(fs_info);
2317
2318         iput(fs_info->btree_inode);
2319
2320         btrfs_stop_workers(&fs_info->fixup_workers);
2321         btrfs_stop_workers(&fs_info->delalloc_workers);
2322         btrfs_stop_workers(&fs_info->workers);
2323         btrfs_stop_workers(&fs_info->endio_workers);
2324         btrfs_stop_workers(&fs_info->endio_meta_workers);
2325         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2326         btrfs_stop_workers(&fs_info->endio_write_workers);
2327         btrfs_stop_workers(&fs_info->submit_workers);
2328
2329 #if 0
2330         while (!list_empty(&fs_info->hashers)) {
2331                 struct btrfs_hasher *hasher;
2332                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
2333                                     hashers);
2334                 list_del(&hasher->hashers);
2335                 crypto_free_hash(&fs_info->hash_tfm);
2336                 kfree(hasher);
2337         }
2338 #endif
2339         btrfs_close_devices(fs_info->fs_devices);
2340         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2341
2342         bdi_destroy(&fs_info->bdi);
2343
2344         kfree(fs_info->extent_root);
2345         kfree(fs_info->tree_root);
2346         kfree(fs_info->chunk_root);
2347         kfree(fs_info->dev_root);
2348         kfree(fs_info->csum_root);
2349         return 0;
2350 }
2351
2352 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2353 {
2354         int ret;
2355         struct inode *btree_inode = buf->first_page->mapping->host;
2356
2357         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2358         if (!ret)
2359                 return ret;
2360
2361         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2362                                     parent_transid);
2363         return !ret;
2364 }
2365
2366 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2367 {
2368         struct inode *btree_inode = buf->first_page->mapping->host;
2369         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2370                                           buf);
2371 }
2372
2373 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2374 {
2375         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2376         u64 transid = btrfs_header_generation(buf);
2377         struct inode *btree_inode = root->fs_info->btree_inode;
2378         int was_dirty;
2379
2380         btrfs_assert_tree_locked(buf);
2381         if (transid != root->fs_info->generation) {
2382                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2383                        "found %llu running %llu\n",
2384                         (unsigned long long)buf->start,
2385                         (unsigned long long)transid,
2386                         (unsigned long long)root->fs_info->generation);
2387                 WARN_ON(1);
2388         }
2389         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2390                                             buf);
2391         if (!was_dirty) {
2392                 spin_lock(&root->fs_info->delalloc_lock);
2393                 root->fs_info->dirty_metadata_bytes += buf->len;
2394                 spin_unlock(&root->fs_info->delalloc_lock);
2395         }
2396 }
2397
2398 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2399 {
2400         /*
2401          * looks as though older kernels can get into trouble with
2402          * this code, they end up stuck in balance_dirty_pages forever
2403          */
2404         struct extent_io_tree *tree;
2405         u64 num_dirty;
2406         u64 start = 0;
2407         unsigned long thresh = 32 * 1024 * 1024;
2408         tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
2409
2410         if (current->flags & PF_MEMALLOC)
2411                 return;
2412
2413         num_dirty = count_range_bits(tree, &start, (u64)-1,
2414                                      thresh, EXTENT_DIRTY);
2415         if (num_dirty > thresh) {
2416                 balance_dirty_pages_ratelimited_nr(
2417                                    root->fs_info->btree_inode->i_mapping, 1);
2418         }
2419         return;
2420 }
2421
2422 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2423 {
2424         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2425         int ret;
2426         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2427         if (ret == 0)
2428                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2429         return ret;
2430 }
2431
2432 int btree_lock_page_hook(struct page *page)
2433 {
2434         struct inode *inode = page->mapping->host;
2435         struct btrfs_root *root = BTRFS_I(inode)->root;
2436         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2437         struct extent_buffer *eb;
2438         unsigned long len;
2439         u64 bytenr = page_offset(page);
2440
2441         if (page->private == EXTENT_PAGE_PRIVATE)
2442                 goto out;
2443
2444         len = page->private >> 2;
2445         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2446         if (!eb)
2447                 goto out;
2448
2449         btrfs_tree_lock(eb);
2450         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2451
2452         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2453                 spin_lock(&root->fs_info->delalloc_lock);
2454                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2455                         root->fs_info->dirty_metadata_bytes -= eb->len;
2456                 else
2457                         WARN_ON(1);
2458                 spin_unlock(&root->fs_info->delalloc_lock);
2459         }
2460
2461         btrfs_tree_unlock(eb);
2462         free_extent_buffer(eb);
2463 out:
2464         lock_page(page);
2465         return 0;
2466 }
2467
2468 static struct extent_io_ops btree_extent_io_ops = {
2469         .write_cache_pages_lock_hook = btree_lock_page_hook,
2470         .readpage_end_io_hook = btree_readpage_end_io_hook,
2471         .submit_bio_hook = btree_submit_bio_hook,
2472         /* note we're sharing with inode.c for the merge bio hook */
2473         .merge_bio_hook = btrfs_merge_bio_hook,
2474 };