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