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                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
 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
 817                 /* ugh, clear_extent_buffer_dirty can be expensive */
 818                 btrfs_set_lock_blocking(buf);
 819
 820                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 821                                           buf);
 822         }
 823         return 0;
 824 }
 825
 826 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 827                         u32 stripesize, struct btrfs_root *root,
 828                         struct btrfs_fs_info *fs_info,
 829                         u64 objectid)
 830 {
 831         root->node = NULL;
 832         root->commit_root = NULL;
 833         root->ref_tree = NULL;
 834         root->sectorsize = sectorsize;
 835         root->nodesize = nodesize;
 836         root->leafsize = leafsize;
 837         root->stripesize = stripesize;
 838         root->ref_cows = 0;
 839         root->track_dirty = 0;
 840
 841         root->fs_info = fs_info;
 842         root->objectid = objectid;
 843         root->last_trans = 0;
 844         root->highest_inode = 0;
 845         root->last_inode_alloc = 0;
 846         root->name = NULL;
 847         root->in_sysfs = 0;
 848
 849         INIT_LIST_HEAD(&root->dirty_list);
 850         INIT_LIST_HEAD(&root->orphan_list);
 851         INIT_LIST_HEAD(&root->dead_list);
 852         spin_lock_init(&root->node_lock);
 853         spin_lock_init(&root->list_lock);
 854         mutex_init(&root->objectid_mutex);
 855         mutex_init(&root->log_mutex);
 856         init_waitqueue_head(&root->log_writer_wait);
 857         init_waitqueue_head(&root->log_commit_wait[0]);
 858         init_waitqueue_head(&root->log_commit_wait[1]);
 859         atomic_set(&root->log_commit[0], 0);
 860         atomic_set(&root->log_commit[1], 0);
 861         atomic_set(&root->log_writers, 0);
 862         root->log_batch = 0;
 863         root->log_transid = 0;
 864         extent_io_tree_init(&root->dirty_log_pages,
 865                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 866
 867         btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
 868         root->ref_tree = &root->ref_tree_struct;
 869
 870         memset(&root->root_key, 0, sizeof(root->root_key));
 871         memset(&root->root_item, 0, sizeof(root->root_item));
 872         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 873         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 874         root->defrag_trans_start = fs_info->generation;
 875         init_completion(&root->kobj_unregister);
 876         root->defrag_running = 0;
 877         root->defrag_level = 0;
 878         root->root_key.objectid = objectid;
 879         root->anon_super.s_root = NULL;
 880         root->anon_super.s_dev = 0;
 881         INIT_LIST_HEAD(&root->anon_super.s_list);
 882         INIT_LIST_HEAD(&root->anon_super.s_instances);
 883         init_rwsem(&root->anon_super.s_umount);
 884
 885         return 0;
 886 }
 887
 888 static int find_and_setup_root(struct btrfs_root *tree_root,
 889                                struct btrfs_fs_info *fs_info,
 890                                u64 objectid,
 891                                struct btrfs_root *root)
 892 {
 893         int ret;
 894         u32 blocksize;
 895         u64 generation;
 896
 897         __setup_root(tree_root->nodesize, tree_root->leafsize,
 898                      tree_root->sectorsize, tree_root->stripesize,
 899                      root, fs_info, objectid);
 900         ret = btrfs_find_last_root(tree_root, objectid,
 901                                    &root->root_item, &root->root_key);
 902         BUG_ON(ret);
 903
 904         generation = btrfs_root_generation(&root->root_item);
 905         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 906         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 907                                      blocksize, generation);
 908         BUG_ON(!root->node);
 909         return 0;
 910 }
 911
 912 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
 913                              struct btrfs_fs_info *fs_info)
 914 {
 915         struct extent_buffer *eb;
 916         struct btrfs_root *log_root_tree = fs_info->log_root_tree;
 917         u64 start = 0;
 918         u64 end = 0;
 919         int ret;
 920
 921         if (!log_root_tree)
 922                 return 0;
 923
 924         while (1) {
 925                 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
 926                                     0, &start, &end, EXTENT_DIRTY);
 927                 if (ret)
 928                         break;
 929
 930                 clear_extent_dirty(&log_root_tree->dirty_log_pages,
 931                                    start, end, GFP_NOFS);
 932         }
 933         eb = fs_info->log_root_tree->node;
 934
 935         WARN_ON(btrfs_header_level(eb) != 0);
 936         WARN_ON(btrfs_header_nritems(eb) != 0);
 937
 938         ret = btrfs_free_reserved_extent(fs_info->tree_root,
 939                                 eb->start, eb->len);
 940         BUG_ON(ret);
 941
 942         free_extent_buffer(eb);
 943         kfree(fs_info->log_root_tree);
 944         fs_info->log_root_tree = NULL;
 945         return 0;
 946 }
 947
 948 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
 949                                          struct btrfs_fs_info *fs_info)
 950 {
 951         struct btrfs_root *root;
 952         struct btrfs_root *tree_root = fs_info->tree_root;
 953         struct extent_buffer *leaf;
 954
 955         root = kzalloc(sizeof(*root), GFP_NOFS);
 956         if (!root)
 957                 return ERR_PTR(-ENOMEM);
 958
 959         __setup_root(tree_root->nodesize, tree_root->leafsize,
 960                      tree_root->sectorsize, tree_root->stripesize,
 961                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
 962
 963         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
 964         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
 965         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
 966         /*
 967          * log trees do not get reference counted because they go away
 968          * before a real commit is actually done.  They do store pointers
 969          * to file data extents, and those reference counts still get
 970          * updated (along with back refs to the log tree).
 971          */
 972         root->ref_cows = 0;
 973
 974         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
 975                                       0, BTRFS_TREE_LOG_OBJECTID,
 976                                       trans->transid, 0, 0, 0);
 977         if (IS_ERR(leaf)) {
 978                 kfree(root);
 979                 return ERR_CAST(leaf);
 980         }
 981
 982         root->node = leaf;
 983         btrfs_set_header_nritems(root->node, 0);
 984         btrfs_set_header_level(root->node, 0);
 985         btrfs_set_header_bytenr(root->node, root->node->start);
 986         btrfs_set_header_generation(root->node, trans->transid);
 987         btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
 988
 989         write_extent_buffer(root->node, root->fs_info->fsid,
 990                             (unsigned long)btrfs_header_fsid(root->node),
 991                             BTRFS_FSID_SIZE);
 992         btrfs_mark_buffer_dirty(root->node);
 993         btrfs_tree_unlock(root->node);
 994         return root;
 995 }
 996
 997 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
 998                              struct btrfs_fs_info *fs_info)
 999 {
1000         struct btrfs_root *log_root;
1001
1002         log_root = alloc_log_tree(trans, fs_info);
1003         if (IS_ERR(log_root))
1004                 return PTR_ERR(log_root);
1005         WARN_ON(fs_info->log_root_tree);
1006         fs_info->log_root_tree = log_root;
1007         return 0;
1008 }
1009
1010 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1011                        struct btrfs_root *root)
1012 {
1013         struct btrfs_root *log_root;
1014         struct btrfs_inode_item *inode_item;
1015
1016         log_root = alloc_log_tree(trans, root->fs_info);
1017         if (IS_ERR(log_root))
1018                 return PTR_ERR(log_root);
1019
1020         log_root->last_trans = trans->transid;
1021         log_root->root_key.offset = root->root_key.objectid;
1022
1023         inode_item = &log_root->root_item.inode;
1024         inode_item->generation = cpu_to_le64(1);
1025         inode_item->size = cpu_to_le64(3);
1026         inode_item->nlink = cpu_to_le32(1);
1027         inode_item->nbytes = cpu_to_le64(root->leafsize);
1028         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1029
1030         btrfs_set_root_bytenr(&log_root->root_item, log_root->node->start);
1031         btrfs_set_root_generation(&log_root->root_item, trans->transid);
1032
1033         WARN_ON(root->log_root);
1034         root->log_root = log_root;
1035         root->log_transid = 0;
1036         return 0;
1037 }
1038
1039 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1040                                                struct btrfs_key *location)
1041 {
1042         struct btrfs_root *root;
1043         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1044         struct btrfs_path *path;
1045         struct extent_buffer *l;
1046         u64 highest_inode;
1047         u64 generation;
1048         u32 blocksize;
1049         int ret = 0;
1050
1051         root = kzalloc(sizeof(*root), GFP_NOFS);
1052         if (!root)
1053                 return ERR_PTR(-ENOMEM);
1054         if (location->offset == (u64)-1) {
1055                 ret = find_and_setup_root(tree_root, fs_info,
1056                                           location->objectid, root);
1057                 if (ret) {
1058                         kfree(root);
1059                         return ERR_PTR(ret);
1060                 }
1061                 goto insert;
1062         }
1063
1064         __setup_root(tree_root->nodesize, tree_root->leafsize,
1065                      tree_root->sectorsize, tree_root->stripesize,
1066                      root, fs_info, location->objectid);
1067
1068         path = btrfs_alloc_path();
1069         BUG_ON(!path);
1070         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1071         if (ret != 0) {
1072                 if (ret > 0)
1073                         ret = -ENOENT;
1074                 goto out;
1075         }
1076         l = path->nodes[0];
1077         read_extent_buffer(l, &root->root_item,
1078                btrfs_item_ptr_offset(l, path->slots[0]),
1079                sizeof(root->root_item));
1080         memcpy(&root->root_key, location, sizeof(*location));
1081         ret = 0;
1082 out:
1083         btrfs_release_path(root, path);
1084         btrfs_free_path(path);
1085         if (ret) {
1086                 kfree(root);
1087                 return ERR_PTR(ret);
1088         }
1089         generation = btrfs_root_generation(&root->root_item);
1090         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1091         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1092                                      blocksize, generation);
1093         BUG_ON(!root->node);
1094 insert:
1095         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1096                 root->ref_cows = 1;
1097                 ret = btrfs_find_highest_inode(root, &highest_inode);
1098                 if (ret == 0) {
1099                         root->highest_inode = highest_inode;
1100                         root->last_inode_alloc = highest_inode;
1101                 }
1102         }
1103         return root;
1104 }
1105
1106 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1107                                         u64 root_objectid)
1108 {
1109         struct btrfs_root *root;
1110
1111         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1112                 return fs_info->tree_root;
1113         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1114                 return fs_info->extent_root;
1115
1116         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1117                                  (unsigned long)root_objectid);
1118         return root;
1119 }
1120
1121 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1122                                               struct btrfs_key *location)
1123 {
1124         struct btrfs_root *root;
1125         int ret;
1126
1127         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1128                 return fs_info->tree_root;
1129         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1130                 return fs_info->extent_root;
1131         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1132                 return fs_info->chunk_root;
1133         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1134                 return fs_info->dev_root;
1135         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1136                 return fs_info->csum_root;
1137
1138         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1139                                  (unsigned long)location->objectid);
1140         if (root)
1141                 return root;
1142
1143         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1144         if (IS_ERR(root))
1145                 return root;
1146
1147         set_anon_super(&root->anon_super, NULL);
1148
1149         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1150                                 (unsigned long)root->root_key.objectid,
1151                                 root);
1152         if (ret) {
1153                 free_extent_buffer(root->node);
1154                 kfree(root);
1155                 return ERR_PTR(ret);
1156         }
1157         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
1158                 ret = btrfs_find_dead_roots(fs_info->tree_root,
1159                                             root->root_key.objectid, root);
1160                 BUG_ON(ret);
1161                 btrfs_orphan_cleanup(root);
1162         }
1163         return root;
1164 }
1165
1166 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1167                                       struct btrfs_key *location,
1168                                       const char *name, int namelen)
1169 {
1170         struct btrfs_root *root;
1171         int ret;
1172
1173         root = btrfs_read_fs_root_no_name(fs_info, location);
1174         if (!root)
1175                 return NULL;
1176
1177         if (root->in_sysfs)
1178                 return root;
1179
1180         ret = btrfs_set_root_name(root, name, namelen);
1181         if (ret) {
1182                 free_extent_buffer(root->node);
1183                 kfree(root);
1184                 return ERR_PTR(ret);
1185         }
1186 #if 0
1187         ret = btrfs_sysfs_add_root(root);
1188         if (ret) {
1189                 free_extent_buffer(root->node);
1190                 kfree(root->name);
1191                 kfree(root);
1192                 return ERR_PTR(ret);
1193         }
1194 #endif
1195         root->in_sysfs = 1;
1196         return root;
1197 }
1198
1199 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1200 {
1201         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1202         int ret = 0;
1203         struct btrfs_device *device;
1204         struct backing_dev_info *bdi;
1205 #if 0
1206         if ((bdi_bits & (1 << BDI_write_congested)) &&
1207             btrfs_congested_async(info, 0))
1208                 return 1;
1209 #endif
1210         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1211                 if (!device->bdev)
1212                         continue;
1213                 bdi = blk_get_backing_dev_info(device->bdev);
1214                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1215                         ret = 1;
1216                         break;
1217                 }
1218         }
1219         return ret;
1220 }
1221
1222 /*
1223  * this unplugs every device on the box, and it is only used when page
1224  * is null
1225  */
1226 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1227 {
1228         struct btrfs_device *device;
1229         struct btrfs_fs_info *info;
1230
1231         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1232         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1233                 if (!device->bdev)
1234                         continue;
1235
1236                 bdi = blk_get_backing_dev_info(device->bdev);
1237                 if (bdi->unplug_io_fn)
1238                         bdi->unplug_io_fn(bdi, page);
1239         }
1240 }
1241
1242 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1243 {
1244         struct inode *inode;
1245         struct extent_map_tree *em_tree;
1246         struct extent_map *em;
1247         struct address_space *mapping;
1248         u64 offset;
1249
1250         /* the generic O_DIRECT read code does this */
1251         if (1 || !page) {
1252                 __unplug_io_fn(bdi, page);
1253                 return;
1254         }
1255
1256         /*
1257          * page->mapping may change at any time.  Get a consistent copy
1258          * and use that for everything below
1259          */
1260         smp_mb();
1261         mapping = page->mapping;
1262         if (!mapping)
1263                 return;
1264
1265         inode = mapping->host;
1266
1267         /*
1268          * don't do the expensive searching for a small number of
1269          * devices
1270          */
1271         if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1272                 __unplug_io_fn(bdi, page);
1273                 return;
1274         }
1275
1276         offset = page_offset(page);
1277
1278         em_tree = &BTRFS_I(inode)->extent_tree;
1279         spin_lock(&em_tree->lock);
1280         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1281         spin_unlock(&em_tree->lock);
1282         if (!em) {
1283                 __unplug_io_fn(bdi, page);
1284                 return;
1285         }
1286
1287         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1288                 free_extent_map(em);
1289                 __unplug_io_fn(bdi, page);
1290                 return;
1291         }
1292         offset = offset - em->start;
1293         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1294                           em->block_start + offset, page);
1295         free_extent_map(em);
1296 }
1297
1298 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1299 {
1300         bdi_init(bdi);
1301         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1302         bdi->state              = 0;
1303         bdi->capabilities       = default_backing_dev_info.capabilities;
1304         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1305         bdi->unplug_io_data     = info;
1306         bdi->congested_fn       = btrfs_congested_fn;
1307         bdi->congested_data     = info;
1308         return 0;
1309 }
1310
1311 static int bio_ready_for_csum(struct bio *bio)
1312 {
1313         u64 length = 0;
1314         u64 buf_len = 0;
1315         u64 start = 0;
1316         struct page *page;
1317         struct extent_io_tree *io_tree = NULL;
1318         struct btrfs_fs_info *info = NULL;
1319         struct bio_vec *bvec;
1320         int i;
1321         int ret;
1322
1323         bio_for_each_segment(bvec, bio, i) {
1324                 page = bvec->bv_page;
1325                 if (page->private == EXTENT_PAGE_PRIVATE) {
1326                         length += bvec->bv_len;
1327                         continue;
1328                 }
1329                 if (!page->private) {
1330                         length += bvec->bv_len;
1331                         continue;
1332                 }
1333                 length = bvec->bv_len;
1334                 buf_len = page->private >> 2;
1335                 start = page_offset(page) + bvec->bv_offset;
1336                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1337                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1338         }
1339         /* are we fully contained in this bio? */
1340         if (buf_len <= length)
1341                 return 1;
1342
1343         ret = extent_range_uptodate(io_tree, start + length,
1344                                     start + buf_len - 1);
1345         if (ret == 1)
1346                 return ret;
1347         return ret;
1348 }
1349
1350 /*
1351  * called by the kthread helper functions to finally call the bio end_io
1352  * functions.  This is where read checksum verification actually happens
1353  */
1354 static void end_workqueue_fn(struct btrfs_work *work)
1355 {
1356         struct bio *bio;
1357         struct end_io_wq *end_io_wq;
1358         struct btrfs_fs_info *fs_info;
1359         int error;
1360
1361         end_io_wq = container_of(work, struct end_io_wq, work);
1362         bio = end_io_wq->bio;
1363         fs_info = end_io_wq->info;
1364
1365         /* metadata bio reads are special because the whole tree block must
1366          * be checksummed at once.  This makes sure the entire block is in
1367          * ram and up to date before trying to verify things.  For
1368          * blocksize <= pagesize, it is basically a noop
1369          */
1370         if (!(bio->bi_rw & (1 << BIO_RW)) && end_io_wq->metadata &&
1371             !bio_ready_for_csum(bio)) {
1372                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1373                                    &end_io_wq->work);
1374                 return;
1375         }
1376         error = end_io_wq->error;
1377         bio->bi_private = end_io_wq->private;
1378         bio->bi_end_io = end_io_wq->end_io;
1379         kfree(end_io_wq);
1380         bio_endio(bio, error);
1381 }
1382
1383 static int cleaner_kthread(void *arg)
1384 {
1385         struct btrfs_root *root = arg;
1386
1387         do {
1388                 smp_mb();
1389                 if (root->fs_info->closing)
1390                         break;
1391
1392                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1393                 mutex_lock(&root->fs_info->cleaner_mutex);
1394                 btrfs_clean_old_snapshots(root);
1395                 mutex_unlock(&root->fs_info->cleaner_mutex);
1396
1397                 if (freezing(current)) {
1398                         refrigerator();
1399                 } else {
1400                         smp_mb();
1401                         if (root->fs_info->closing)
1402                                 break;
1403                         set_current_state(TASK_INTERRUPTIBLE);
1404                         schedule();
1405                         __set_current_state(TASK_RUNNING);
1406                 }
1407         } while (!kthread_should_stop());
1408         return 0;
1409 }
1410
1411 static int transaction_kthread(void *arg)
1412 {
1413         struct btrfs_root *root = arg;
1414         struct btrfs_trans_handle *trans;
1415         struct btrfs_transaction *cur;
1416         unsigned long now;
1417         unsigned long delay;
1418         int ret;
1419
1420         do {
1421                 smp_mb();
1422                 if (root->fs_info->closing)
1423                         break;
1424
1425                 delay = HZ * 30;
1426                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1427                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1428
1429                 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1430                         printk(KERN_INFO "btrfs: total reference cache "
1431                                "size %llu\n",
1432                                root->fs_info->total_ref_cache_size);
1433                 }
1434
1435                 mutex_lock(&root->fs_info->trans_mutex);
1436                 cur = root->fs_info->running_transaction;
1437                 if (!cur) {
1438                         mutex_unlock(&root->fs_info->trans_mutex);
1439                         goto sleep;
1440                 }
1441
1442                 now = get_seconds();
1443                 if (now < cur->start_time || now - cur->start_time < 30) {
1444                         mutex_unlock(&root->fs_info->trans_mutex);
1445                         delay = HZ * 5;
1446                         goto sleep;
1447                 }
1448                 mutex_unlock(&root->fs_info->trans_mutex);
1449                 trans = btrfs_start_transaction(root, 1);
1450                 ret = btrfs_commit_transaction(trans, root);
1451 sleep:
1452                 wake_up_process(root->fs_info->cleaner_kthread);
1453                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1454
1455                 if (freezing(current)) {
1456                         refrigerator();
1457                 } else {
1458                         if (root->fs_info->closing)
1459                                 break;
1460                         set_current_state(TASK_INTERRUPTIBLE);
1461                         schedule_timeout(delay);
1462                         __set_current_state(TASK_RUNNING);
1463                 }
1464         } while (!kthread_should_stop());
1465         return 0;
1466 }
1467
1468 struct btrfs_root *open_ctree(struct super_block *sb,
1469                               struct btrfs_fs_devices *fs_devices,
1470                               char *options)
1471 {
1472         u32 sectorsize;
1473         u32 nodesize;
1474         u32 leafsize;
1475         u32 blocksize;
1476         u32 stripesize;
1477         u64 generation;
1478         u64 features;
1479         struct btrfs_key location;
1480         struct buffer_head *bh;
1481         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1482                                                  GFP_NOFS);
1483         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1484                                                  GFP_NOFS);
1485         struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1486                                                GFP_NOFS);
1487         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1488                                                 GFP_NOFS);
1489         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1490                                                 GFP_NOFS);
1491         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1492                                               GFP_NOFS);
1493         struct btrfs_root *log_tree_root;
1494
1495         int ret;
1496         int err = -EINVAL;
1497
1498         struct btrfs_super_block *disk_super;
1499
1500         if (!extent_root || !tree_root || !fs_info ||
1501             !chunk_root || !dev_root || !csum_root) {
1502                 err = -ENOMEM;
1503                 goto fail;
1504         }
1505         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1506         INIT_LIST_HEAD(&fs_info->trans_list);
1507         INIT_LIST_HEAD(&fs_info->dead_roots);
1508         INIT_LIST_HEAD(&fs_info->hashers);
1509         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1510         spin_lock_init(&fs_info->delalloc_lock);
1511         spin_lock_init(&fs_info->new_trans_lock);
1512         spin_lock_init(&fs_info->ref_cache_lock);
1513
1514         init_completion(&fs_info->kobj_unregister);
1515         fs_info->tree_root = tree_root;
1516         fs_info->extent_root = extent_root;
1517         fs_info->csum_root = csum_root;
1518         fs_info->chunk_root = chunk_root;
1519         fs_info->dev_root = dev_root;
1520         fs_info->fs_devices = fs_devices;
1521         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1522         INIT_LIST_HEAD(&fs_info->space_info);
1523         btrfs_mapping_init(&fs_info->mapping_tree);
1524         atomic_set(&fs_info->nr_async_submits, 0);
1525         atomic_set(&fs_info->async_delalloc_pages, 0);
1526         atomic_set(&fs_info->async_submit_draining, 0);
1527         atomic_set(&fs_info->nr_async_bios, 0);
1528         atomic_set(&fs_info->throttles, 0);
1529         atomic_set(&fs_info->throttle_gen, 0);
1530         fs_info->sb = sb;
1531         fs_info->max_extent = (u64)-1;
1532         fs_info->max_inline = 8192 * 1024;
1533         setup_bdi(fs_info, &fs_info->bdi);
1534         fs_info->btree_inode = new_inode(sb);
1535         fs_info->btree_inode->i_ino = 1;
1536         fs_info->btree_inode->i_nlink = 1;
1537
1538         fs_info->thread_pool_size = min_t(unsigned long,
1539                                           num_online_cpus() + 2, 8);
1540
1541         INIT_LIST_HEAD(&fs_info->ordered_extents);
1542         spin_lock_init(&fs_info->ordered_extent_lock);
1543
1544         sb->s_blocksize = 4096;
1545         sb->s_blocksize_bits = blksize_bits(4096);
1546
1547         /*
1548          * we set the i_size on the btree inode to the max possible int.
1549          * the real end of the address space is determined by all of
1550          * the devices in the system
1551          */
1552         fs_info->btree_inode->i_size = OFFSET_MAX;
1553         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1554         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1555
1556         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1557                              fs_info->btree_inode->i_mapping,
1558                              GFP_NOFS);
1559         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1560                              GFP_NOFS);
1561
1562         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1563
1564         spin_lock_init(&fs_info->block_group_cache_lock);
1565         fs_info->block_group_cache_tree.rb_node = NULL;
1566
1567         extent_io_tree_init(&fs_info->pinned_extents,
1568                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1569         extent_io_tree_init(&fs_info->pending_del,
1570                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1571         extent_io_tree_init(&fs_info->extent_ins,
1572                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1573         fs_info->do_barriers = 1;
1574
1575         INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
1576         btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
1577         btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
1578
1579         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1580         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1581                sizeof(struct btrfs_key));
1582         insert_inode_hash(fs_info->btree_inode);
1583
1584         mutex_init(&fs_info->trans_mutex);
1585         mutex_init(&fs_info->tree_log_mutex);
1586         mutex_init(&fs_info->drop_mutex);
1587         mutex_init(&fs_info->extent_ins_mutex);
1588         mutex_init(&fs_info->pinned_mutex);
1589         mutex_init(&fs_info->chunk_mutex);
1590         mutex_init(&fs_info->transaction_kthread_mutex);
1591         mutex_init(&fs_info->cleaner_mutex);
1592         mutex_init(&fs_info->volume_mutex);
1593         mutex_init(&fs_info->tree_reloc_mutex);
1594         init_waitqueue_head(&fs_info->transaction_throttle);
1595         init_waitqueue_head(&fs_info->transaction_wait);
1596         init_waitqueue_head(&fs_info->async_submit_wait);
1597
1598         __setup_root(4096, 4096, 4096, 4096, tree_root,
1599                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1600
1601
1602         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1603         if (!bh)
1604                 goto fail_iput;
1605
1606         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1607         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1608                sizeof(fs_info->super_for_commit));
1609         brelse(bh);
1610
1611         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1612
1613         disk_super = &fs_info->super_copy;
1614         if (!btrfs_super_root(disk_super))
1615                 goto fail_iput;
1616
1617         ret = btrfs_parse_options(tree_root, options);
1618         if (ret) {
1619                 err = ret;
1620                 goto fail_iput;
1621         }
1622
1623         features = btrfs_super_incompat_flags(disk_super) &
1624                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1625         if (features) {
1626                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1627                        "unsupported optional features (%Lx).\n",
1628                        features);
1629                 err = -EINVAL;
1630                 goto fail_iput;
1631         }
1632
1633         features = btrfs_super_compat_ro_flags(disk_super) &
1634                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1635         if (!(sb->s_flags & MS_RDONLY) && features) {
1636                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1637                        "unsupported option features (%Lx).\n",
1638                        features);
1639                 err = -EINVAL;
1640                 goto fail_iput;
1641         }
1642
1643         /*
1644          * we need to start all the end_io workers up front because the
1645          * queue work function gets called at interrupt time, and so it
1646          * cannot dynamically grow.
1647          */
1648         btrfs_init_workers(&fs_info->workers, "worker",
1649                            fs_info->thread_pool_size);
1650
1651         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1652                            fs_info->thread_pool_size);
1653
1654         btrfs_init_workers(&fs_info->submit_workers, "submit",
1655                            min_t(u64, fs_devices->num_devices,
1656                            fs_info->thread_pool_size));
1657
1658         /* a higher idle thresh on the submit workers makes it much more
1659          * likely that bios will be send down in a sane order to the
1660          * devices
1661          */
1662         fs_info->submit_workers.idle_thresh = 64;
1663
1664         fs_info->workers.idle_thresh = 16;
1665         fs_info->workers.ordered = 1;
1666
1667         fs_info->delalloc_workers.idle_thresh = 2;
1668         fs_info->delalloc_workers.ordered = 1;
1669
1670         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1671         btrfs_init_workers(&fs_info->endio_workers, "endio",
1672                            fs_info->thread_pool_size);
1673         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1674                            fs_info->thread_pool_size);
1675         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1676                            "endio-meta-write", fs_info->thread_pool_size);
1677         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1678                            fs_info->thread_pool_size);
1679
1680         /*
1681          * endios are largely parallel and should have a very
1682          * low idle thresh
1683          */
1684         fs_info->endio_workers.idle_thresh = 4;
1685         fs_info->endio_meta_workers.idle_thresh = 4;
1686
1687         fs_info->endio_write_workers.idle_thresh = 64;
1688         fs_info->endio_meta_write_workers.idle_thresh = 64;
1689
1690         btrfs_start_workers(&fs_info->workers, 1);
1691         btrfs_start_workers(&fs_info->submit_workers, 1);
1692         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1693         btrfs_start_workers(&fs_info->fixup_workers, 1);
1694         btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1695         btrfs_start_workers(&fs_info->endio_meta_workers,
1696                             fs_info->thread_pool_size);
1697         btrfs_start_workers(&fs_info->endio_meta_write_workers,
1698                             fs_info->thread_pool_size);
1699         btrfs_start_workers(&fs_info->endio_write_workers,
1700                             fs_info->thread_pool_size);
1701
1702         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1703         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1704                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1705
1706         nodesize = btrfs_super_nodesize(disk_super);
1707         leafsize = btrfs_super_leafsize(disk_super);
1708         sectorsize = btrfs_super_sectorsize(disk_super);
1709         stripesize = btrfs_super_stripesize(disk_super);
1710         tree_root->nodesize = nodesize;
1711         tree_root->leafsize = leafsize;
1712         tree_root->sectorsize = sectorsize;
1713         tree_root->stripesize = stripesize;
1714
1715         sb->s_blocksize = sectorsize;
1716         sb->s_blocksize_bits = blksize_bits(sectorsize);
1717
1718         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1719                     sizeof(disk_super->magic))) {
1720                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1721                 goto fail_sb_buffer;
1722         }
1723
1724         mutex_lock(&fs_info->chunk_mutex);
1725         ret = btrfs_read_sys_array(tree_root);
1726         mutex_unlock(&fs_info->chunk_mutex);
1727         if (ret) {
1728                 printk(KERN_WARNING "btrfs: failed to read the system "
1729                        "array on %s\n", sb->s_id);
1730                 goto fail_sys_array;
1731         }
1732
1733         blocksize = btrfs_level_size(tree_root,
1734                                      btrfs_super_chunk_root_level(disk_super));
1735         generation = btrfs_super_chunk_root_generation(disk_super);
1736
1737         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1738                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1739
1740         chunk_root->node = read_tree_block(chunk_root,
1741                                            btrfs_super_chunk_root(disk_super),
1742                                            blocksize, generation);
1743         BUG_ON(!chunk_root->node);
1744
1745         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1746            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1747            BTRFS_UUID_SIZE);
1748
1749         mutex_lock(&fs_info->chunk_mutex);
1750         ret = btrfs_read_chunk_tree(chunk_root);
1751         mutex_unlock(&fs_info->chunk_mutex);
1752         if (ret) {
1753                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1754                        sb->s_id);
1755                 goto fail_chunk_root;
1756         }
1757
1758         btrfs_close_extra_devices(fs_devices);
1759
1760         blocksize = btrfs_level_size(tree_root,
1761                                      btrfs_super_root_level(disk_super));
1762         generation = btrfs_super_generation(disk_super);
1763
1764         tree_root->node = read_tree_block(tree_root,
1765                                           btrfs_super_root(disk_super),
1766                                           blocksize, generation);
1767         if (!tree_root->node)
1768                 goto fail_chunk_root;
1769
1770
1771         ret = find_and_setup_root(tree_root, fs_info,
1772                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1773         if (ret)
1774                 goto fail_tree_root;
1775         extent_root->track_dirty = 1;
1776
1777         ret = find_and_setup_root(tree_root, fs_info,
1778                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1779         dev_root->track_dirty = 1;
1780
1781         if (ret)
1782                 goto fail_extent_root;
1783
1784         ret = find_and_setup_root(tree_root, fs_info,
1785                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
1786         if (ret)
1787                 goto fail_extent_root;
1788
1789         csum_root->track_dirty = 1;
1790
1791         btrfs_read_block_groups(extent_root);
1792
1793         fs_info->generation = generation;
1794         fs_info->last_trans_committed = generation;
1795         fs_info->data_alloc_profile = (u64)-1;
1796         fs_info->metadata_alloc_profile = (u64)-1;
1797         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1798         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1799                                                "btrfs-cleaner");
1800         if (IS_ERR(fs_info->cleaner_kthread))
1801                 goto fail_csum_root;
1802
1803         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1804                                                    tree_root,
1805                                                    "btrfs-transaction");
1806         if (IS_ERR(fs_info->transaction_kthread))
1807                 goto fail_cleaner;
1808
1809         if (btrfs_super_log_root(disk_super) != 0) {
1810                 u64 bytenr = btrfs_super_log_root(disk_super);
1811
1812                 if (fs_devices->rw_devices == 0) {
1813                         printk(KERN_WARNING "Btrfs log replay required "
1814                                "on RO media\n");
1815                         err = -EIO;
1816                         goto fail_trans_kthread;
1817                 }
1818                 blocksize =
1819                      btrfs_level_size(tree_root,
1820                                       btrfs_super_log_root_level(disk_super));
1821
1822                 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1823                                                       GFP_NOFS);
1824
1825                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1826                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1827
1828                 log_tree_root->node = read_tree_block(tree_root, bytenr,
1829                                                       blocksize,
1830                                                       generation + 1);
1831                 ret = btrfs_recover_log_trees(log_tree_root);
1832                 BUG_ON(ret);
1833
1834                 if (sb->s_flags & MS_RDONLY) {
1835                         ret =  btrfs_commit_super(tree_root);
1836                         BUG_ON(ret);
1837                 }
1838         }
1839
1840         if (!(sb->s_flags & MS_RDONLY)) {
1841                 ret = btrfs_cleanup_reloc_trees(tree_root);
1842                 BUG_ON(ret);
1843         }
1844
1845         location.objectid = BTRFS_FS_TREE_OBJECTID;
1846         location.type = BTRFS_ROOT_ITEM_KEY;
1847         location.offset = (u64)-1;
1848
1849         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1850         if (!fs_info->fs_root)
1851                 goto fail_trans_kthread;
1852         return tree_root;
1853
1854 fail_trans_kthread:
1855         kthread_stop(fs_info->transaction_kthread);
1856 fail_cleaner:
1857         kthread_stop(fs_info->cleaner_kthread);
1858
1859         /*
1860          * make sure we're done with the btree inode before we stop our
1861          * kthreads
1862          */
1863         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1864         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1865
1866 fail_csum_root:
1867         free_extent_buffer(csum_root->node);
1868 fail_extent_root:
1869         free_extent_buffer(extent_root->node);
1870 fail_tree_root:
1871         free_extent_buffer(tree_root->node);
1872 fail_chunk_root:
1873         free_extent_buffer(chunk_root->node);
1874 fail_sys_array:
1875         free_extent_buffer(dev_root->node);
1876 fail_sb_buffer:
1877         btrfs_stop_workers(&fs_info->fixup_workers);
1878         btrfs_stop_workers(&fs_info->delalloc_workers);
1879         btrfs_stop_workers(&fs_info->workers);
1880         btrfs_stop_workers(&fs_info->endio_workers);
1881         btrfs_stop_workers(&fs_info->endio_meta_workers);
1882         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
1883         btrfs_stop_workers(&fs_info->endio_write_workers);
1884         btrfs_stop_workers(&fs_info->submit_workers);
1885 fail_iput:
1886         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1887         iput(fs_info->btree_inode);
1888
1889         btrfs_close_devices(fs_info->fs_devices);
1890         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1891         bdi_destroy(&fs_info->bdi);
1892
1893 fail:
1894         kfree(extent_root);
1895         kfree(tree_root);
1896         kfree(fs_info);
1897         kfree(chunk_root);
1898         kfree(dev_root);
1899         kfree(csum_root);
1900         return ERR_PTR(err);
1901 }
1902
1903 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1904 {
1905         char b[BDEVNAME_SIZE];
1906
1907         if (uptodate) {
1908                 set_buffer_uptodate(bh);
1909         } else {
1910                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1911                         printk(KERN_WARNING "lost page write due to "
1912                                         "I/O error on %s\n",
1913                                        bdevname(bh->b_bdev, b));
1914                 }
1915                 /* note, we dont' set_buffer_write_io_error because we have
1916                  * our own ways of dealing with the IO errors
1917                  */
1918                 clear_buffer_uptodate(bh);
1919         }
1920         unlock_buffer(bh);
1921         put_bh(bh);
1922 }
1923
1924 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
1925 {
1926         struct buffer_head *bh;
1927         struct buffer_head *latest = NULL;
1928         struct btrfs_super_block *super;
1929         int i;
1930         u64 transid = 0;
1931         u64 bytenr;
1932
1933         /* we would like to check all the supers, but that would make
1934          * a btrfs mount succeed after a mkfs from a different FS.
1935          * So, we need to add a special mount option to scan for
1936          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1937          */
1938         for (i = 0; i < 1; i++) {
1939                 bytenr = btrfs_sb_offset(i);
1940                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
1941                         break;
1942                 bh = __bread(bdev, bytenr / 4096, 4096);
1943                 if (!bh)
1944                         continue;
1945
1946                 super = (struct btrfs_super_block *)bh->b_data;
1947                 if (btrfs_super_bytenr(super) != bytenr ||
1948                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
1949                             sizeof(super->magic))) {
1950                         brelse(bh);
1951                         continue;
1952                 }
1953
1954                 if (!latest || btrfs_super_generation(super) > transid) {
1955                         brelse(latest);
1956                         latest = bh;
1957                         transid = btrfs_super_generation(super);
1958                 } else {
1959                         brelse(bh);
1960                 }
1961         }
1962         return latest;
1963 }
1964
1965 static int write_dev_supers(struct btrfs_device *device,
1966                             struct btrfs_super_block *sb,
1967                             int do_barriers, int wait, int max_mirrors)
1968 {
1969         struct buffer_head *bh;
1970         int i;
1971         int ret;
1972         int errors = 0;
1973         u32 crc;
1974         u64 bytenr;
1975         int last_barrier = 0;
1976
1977         if (max_mirrors == 0)
1978                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
1979
1980         /* make sure only the last submit_bh does a barrier */
1981         if (do_barriers) {
1982                 for (i = 0; i < max_mirrors; i++) {
1983                         bytenr = btrfs_sb_offset(i);
1984                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1985                             device->total_bytes)
1986                                 break;
1987                         last_barrier = i;
1988                 }
1989         }
1990
1991         for (i = 0; i < max_mirrors; i++) {
1992                 bytenr = btrfs_sb_offset(i);
1993                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
1994                         break;
1995
1996                 if (wait) {
1997                         bh = __find_get_block(device->bdev, bytenr / 4096,
1998                                               BTRFS_SUPER_INFO_SIZE);
1999                         BUG_ON(!bh);
2000                         brelse(bh);
2001                         wait_on_buffer(bh);
2002                         if (buffer_uptodate(bh)) {
2003                                 brelse(bh);
2004                                 continue;
2005                         }
2006                 } else {
2007                         btrfs_set_super_bytenr(sb, bytenr);
2008
2009                         crc = ~(u32)0;
2010                         crc = btrfs_csum_data(NULL, (char *)sb +
2011                                               BTRFS_CSUM_SIZE, crc,
2012                                               BTRFS_SUPER_INFO_SIZE -
2013                                               BTRFS_CSUM_SIZE);
2014                         btrfs_csum_final(crc, sb->csum);
2015
2016                         bh = __getblk(device->bdev, bytenr / 4096,
2017                                       BTRFS_SUPER_INFO_SIZE);
2018                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2019
2020                         set_buffer_uptodate(bh);
2021                         get_bh(bh);
2022                         lock_buffer(bh);
2023                         bh->b_end_io = btrfs_end_buffer_write_sync;
2024                 }
2025
2026                 if (i == last_barrier && do_barriers && device->barriers) {
2027                         ret = submit_bh(WRITE_BARRIER, bh);
2028                         if (ret == -EOPNOTSUPP) {
2029                                 printk("btrfs: disabling barriers on dev %s\n",
2030                                        device->name);
2031                                 set_buffer_uptodate(bh);
2032                                 device->barriers = 0;
2033                                 get_bh(bh);
2034                                 lock_buffer(bh);
2035                                 ret = submit_bh(WRITE, bh);
2036                         }
2037                 } else {
2038                         ret = submit_bh(WRITE, bh);
2039                 }
2040
2041                 if (!ret && wait) {
2042                         wait_on_buffer(bh);
2043                         if (!buffer_uptodate(bh))
2044                                 errors++;
2045                 } else if (ret) {
2046                         errors++;
2047                 }
2048                 if (wait)
2049                         brelse(bh);
2050         }
2051         return errors < i ? 0 : -1;
2052 }
2053
2054 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2055 {
2056         struct list_head *head = &root->fs_info->fs_devices->devices;
2057         struct btrfs_device *dev;
2058         struct btrfs_super_block *sb;
2059         struct btrfs_dev_item *dev_item;
2060         int ret;
2061         int do_barriers;
2062         int max_errors;
2063         int total_errors = 0;
2064         u64 flags;
2065
2066         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2067         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2068
2069         sb = &root->fs_info->super_for_commit;
2070         dev_item = &sb->dev_item;
2071         list_for_each_entry(dev, head, dev_list) {
2072                 if (!dev->bdev) {
2073                         total_errors++;
2074                         continue;
2075                 }
2076                 if (!dev->in_fs_metadata || !dev->writeable)
2077                         continue;
2078
2079                 btrfs_set_stack_device_generation(dev_item, 0);
2080                 btrfs_set_stack_device_type(dev_item, dev->type);
2081                 btrfs_set_stack_device_id(dev_item, dev->devid);
2082                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2083                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2084                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2085                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2086                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2087                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2088                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2089
2090                 flags = btrfs_super_flags(sb);
2091                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2092
2093                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2094                 if (ret)
2095                         total_errors++;
2096         }
2097         if (total_errors > max_errors) {
2098                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2099                        total_errors);
2100                 BUG();
2101         }
2102
2103         total_errors = 0;
2104         list_for_each_entry(dev, head, dev_list) {
2105                 if (!dev->bdev)
2106                         continue;
2107                 if (!dev->in_fs_metadata || !dev->writeable)
2108                         continue;
2109
2110                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2111                 if (ret)
2112                         total_errors++;
2113         }
2114         if (total_errors > max_errors) {
2115                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2116                        total_errors);
2117                 BUG();
2118         }
2119         return 0;
2120 }
2121
2122 int write_ctree_super(struct btrfs_trans_handle *trans,
2123                       struct btrfs_root *root, int max_mirrors)
2124 {
2125         int ret;
2126
2127         ret = write_all_supers(root, max_mirrors);
2128         return ret;
2129 }
2130
2131 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2132 {
2133         radix_tree_delete(&fs_info->fs_roots_radix,
2134                           (unsigned long)root->root_key.objectid);
2135         if (root->anon_super.s_dev) {
2136                 down_write(&root->anon_super.s_umount);
2137                 kill_anon_super(&root->anon_super);
2138         }
2139         if (root->node)
2140                 free_extent_buffer(root->node);
2141         if (root->commit_root)
2142                 free_extent_buffer(root->commit_root);
2143         kfree(root->name);
2144         kfree(root);
2145         return 0;
2146 }
2147
2148 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2149 {
2150         int ret;
2151         struct btrfs_root *gang[8];
2152         int i;
2153
2154         while (1) {
2155                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2156                                              (void **)gang, 0,
2157                                              ARRAY_SIZE(gang));
2158                 if (!ret)
2159                         break;
2160                 for (i = 0; i < ret; i++)
2161                         btrfs_free_fs_root(fs_info, gang[i]);
2162         }
2163         return 0;
2164 }
2165
2166 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2167 {
2168         u64 root_objectid = 0;
2169         struct btrfs_root *gang[8];
2170         int i;
2171         int ret;
2172
2173         while (1) {
2174                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2175                                              (void **)gang, root_objectid,
2176                                              ARRAY_SIZE(gang));
2177                 if (!ret)
2178                         break;
2179                 for (i = 0; i < ret; i++) {
2180                         root_objectid = gang[i]->root_key.objectid;
2181                         ret = btrfs_find_dead_roots(fs_info->tree_root,
2182                                                     root_objectid, gang[i]);
2183                         BUG_ON(ret);
2184                         btrfs_orphan_cleanup(gang[i]);
2185                 }
2186                 root_objectid++;
2187         }
2188         return 0;
2189 }
2190
2191 int btrfs_commit_super(struct btrfs_root *root)
2192 {
2193         struct btrfs_trans_handle *trans;
2194         int ret;
2195
2196         mutex_lock(&root->fs_info->cleaner_mutex);
2197         btrfs_clean_old_snapshots(root);
2198         mutex_unlock(&root->fs_info->cleaner_mutex);
2199         trans = btrfs_start_transaction(root, 1);
2200         ret = btrfs_commit_transaction(trans, root);
2201         BUG_ON(ret);
2202         /* run commit again to drop the original snapshot */
2203         trans = btrfs_start_transaction(root, 1);
2204         btrfs_commit_transaction(trans, root);
2205         ret = btrfs_write_and_wait_transaction(NULL, root);
2206         BUG_ON(ret);
2207
2208         ret = write_ctree_super(NULL, root, 0);
2209         return ret;
2210 }
2211
2212 int close_ctree(struct btrfs_root *root)
2213 {
2214         struct btrfs_fs_info *fs_info = root->fs_info;
2215         int ret;
2216
2217         fs_info->closing = 1;
2218         smp_mb();
2219
2220         kthread_stop(root->fs_info->transaction_kthread);
2221         kthread_stop(root->fs_info->cleaner_kthread);
2222
2223         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2224                 ret =  btrfs_commit_super(root);
2225                 if (ret)
2226                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2227         }
2228
2229         if (fs_info->delalloc_bytes) {
2230                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2231                        fs_info->delalloc_bytes);
2232         }
2233         if (fs_info->total_ref_cache_size) {
2234                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2235                        (unsigned long long)fs_info->total_ref_cache_size);
2236         }
2237
2238         if (fs_info->extent_root->node)
2239                 free_extent_buffer(fs_info->extent_root->node);
2240
2241         if (fs_info->tree_root->node)
2242                 free_extent_buffer(fs_info->tree_root->node);
2243
2244         if (root->fs_info->chunk_root->node)
2245                 free_extent_buffer(root->fs_info->chunk_root->node);
2246
2247         if (root->fs_info->dev_root->node)
2248                 free_extent_buffer(root->fs_info->dev_root->node);
2249
2250         if (root->fs_info->csum_root->node)
2251                 free_extent_buffer(root->fs_info->csum_root->node);
2252
2253         btrfs_free_block_groups(root->fs_info);
2254
2255         del_fs_roots(fs_info);
2256
2257         iput(fs_info->btree_inode);
2258
2259         btrfs_stop_workers(&fs_info->fixup_workers);
2260         btrfs_stop_workers(&fs_info->delalloc_workers);
2261         btrfs_stop_workers(&fs_info->workers);
2262         btrfs_stop_workers(&fs_info->endio_workers);
2263         btrfs_stop_workers(&fs_info->endio_meta_workers);
2264         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2265         btrfs_stop_workers(&fs_info->endio_write_workers);
2266         btrfs_stop_workers(&fs_info->submit_workers);
2267
2268 #if 0
2269         while (!list_empty(&fs_info->hashers)) {
2270                 struct btrfs_hasher *hasher;
2271                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
2272                                     hashers);
2273                 list_del(&hasher->hashers);
2274                 crypto_free_hash(&fs_info->hash_tfm);
2275                 kfree(hasher);
2276         }
2277 #endif
2278         btrfs_close_devices(fs_info->fs_devices);
2279         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2280
2281         bdi_destroy(&fs_info->bdi);
2282
2283         kfree(fs_info->extent_root);
2284         kfree(fs_info->tree_root);
2285         kfree(fs_info->chunk_root);
2286         kfree(fs_info->dev_root);
2287         kfree(fs_info->csum_root);
2288         return 0;
2289 }
2290
2291 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2292 {
2293         int ret;
2294         struct inode *btree_inode = buf->first_page->mapping->host;
2295
2296         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2297         if (!ret)
2298                 return ret;
2299
2300         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2301                                     parent_transid);
2302         return !ret;
2303 }
2304
2305 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2306 {
2307         struct inode *btree_inode = buf->first_page->mapping->host;
2308         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2309                                           buf);
2310 }
2311
2312 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2313 {
2314         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2315         u64 transid = btrfs_header_generation(buf);
2316         struct inode *btree_inode = root->fs_info->btree_inode;
2317
2318         btrfs_set_lock_blocking(buf);
2319
2320         WARN_ON(!btrfs_tree_locked(buf));
2321         if (transid != root->fs_info->generation) {
2322                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2323                        "found %llu running %llu\n",
2324                         (unsigned long long)buf->start,
2325                         (unsigned long long)transid,
2326                         (unsigned long long)root->fs_info->generation);
2327                 WARN_ON(1);
2328         }
2329         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
2330 }
2331
2332 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2333 {
2334         /*
2335          * looks as though older kernels can get into trouble with
2336          * this code, they end up stuck in balance_dirty_pages forever
2337          */
2338         struct extent_io_tree *tree;
2339         u64 num_dirty;
2340         u64 start = 0;
2341         unsigned long thresh = 32 * 1024 * 1024;
2342         tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
2343
2344         if (current_is_pdflush() || current->flags & PF_MEMALLOC)
2345                 return;
2346
2347         num_dirty = count_range_bits(tree, &start, (u64)-1,
2348                                      thresh, EXTENT_DIRTY);
2349         if (num_dirty > thresh) {
2350                 balance_dirty_pages_ratelimited_nr(
2351                                    root->fs_info->btree_inode->i_mapping, 1);
2352         }
2353         return;
2354 }
2355
2356 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2357 {
2358         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2359         int ret;
2360         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2361         if (ret == 0)
2362                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2363         return ret;
2364 }
2365
2366 int btree_lock_page_hook(struct page *page)
2367 {
2368         struct inode *inode = page->mapping->host;
2369         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2370         struct extent_buffer *eb;
2371         unsigned long len;
2372         u64 bytenr = page_offset(page);
2373
2374         if (page->private == EXTENT_PAGE_PRIVATE)
2375                 goto out;
2376
2377         len = page->private >> 2;
2378         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2379         if (!eb)
2380                 goto out;
2381
2382         btrfs_tree_lock(eb);
2383         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2384         btrfs_tree_unlock(eb);
2385         free_extent_buffer(eb);
2386 out:
2387         lock_page(page);
2388         return 0;
2389 }
2390
2391 static struct extent_io_ops btree_extent_io_ops = {
2392         .write_cache_pages_lock_hook = btree_lock_page_hook,
2393         .readpage_end_io_hook = btree_readpage_end_io_hook,
2394         .submit_bio_hook = btree_submit_bio_hook,
2395         /* note we're sharing with inode.c for the merge bio hook */
2396         .merge_bio_hook = btrfs_merge_bio_hook,
2397 };