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

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/version.h>
  20 #include <linux/fs.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/scatterlist.h>
  23 #include <linux/swap.h>
  24 #include <linux/radix-tree.h>
  25 #include <linux/writeback.h>
  26 #include <linux/buffer_head.h> // for block_sync_page
  27 #include <linux/workqueue.h>
  28 #include <linux/kthread.h>
  29 # include <linux/freezer.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 #if 0
  43 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
  44 {
  45         if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
  46                 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
  47                        (unsigned long long)extent_buffer_blocknr(buf),
  48                        (unsigned long long)btrfs_header_blocknr(buf));
  49                 return 1;
  50         }
  51         return 0;
  52 }
  53 #endif
  54
  55 static struct extent_io_ops btree_extent_io_ops;
  56 static void end_workqueue_fn(struct btrfs_work *work);
  57
  58 /*
  59  * end_io_wq structs are used to do processing in task context when an IO is
  60  * complete.  This is used during reads to verify checksums, and it is used
  61  * by writes to insert metadata for new file extents after IO is complete.
  62  */
  63 struct end_io_wq {
  64         struct bio *bio;
  65         bio_end_io_t *end_io;
  66         void *private;
  67         struct btrfs_fs_info *info;
  68         int error;
  69         int metadata;
  70         struct list_head list;
  71         struct btrfs_work work;
  72 };
  73
  74 /*
  75  * async submit bios are used to offload expensive checksumming
  76  * onto the worker threads.  They checksum file and metadata bios
  77  * just before they are sent down the IO stack.
  78  */
  79 struct async_submit_bio {
  80         struct inode *inode;
  81         struct bio *bio;
  82         struct list_head list;
  83         extent_submit_bio_hook_t *submit_bio_start;
  84         extent_submit_bio_hook_t *submit_bio_done;
  85         int rw;
  86         int mirror_num;
  87         unsigned long bio_flags;
  88         struct btrfs_work work;
  89 };
  90
  91 /*
  92  * extents on the btree inode are pretty simple, there's one extent
  93  * that covers the entire device
  94  */
  95 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
  96                                     size_t page_offset, u64 start, u64 len,
  97                                     int create)
  98 {
  99         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 100         struct extent_map *em;
 101         int ret;
 102
 103         spin_lock(&em_tree->lock);
 104         em = lookup_extent_mapping(em_tree, start, len);
 105         if (em) {
 106                 em->bdev =
 107                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 108                 spin_unlock(&em_tree->lock);
 109                 goto out;
 110         }
 111         spin_unlock(&em_tree->lock);
 112
 113         em = alloc_extent_map(GFP_NOFS);
 114         if (!em) {
 115                 em = ERR_PTR(-ENOMEM);
 116                 goto out;
 117         }
 118         em->start = 0;
 119         em->len = (u64)-1;
 120         em->block_len = (u64)-1;
 121         em->block_start = 0;
 122         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 123
 124         spin_lock(&em_tree->lock);
 125         ret = add_extent_mapping(em_tree, em);
 126         if (ret == -EEXIST) {
 127                 u64 failed_start = em->start;
 128                 u64 failed_len = em->len;
 129
 130                 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
 131                        em->start, em->len, em->block_start);
 132                 free_extent_map(em);
 133                 em = lookup_extent_mapping(em_tree, start, len);
 134                 if (em) {
 135                         printk("after failing, found %Lu %Lu %Lu\n",
 136                                em->start, em->len, em->block_start);
 137                         ret = 0;
 138                 } else {
 139                         em = lookup_extent_mapping(em_tree, failed_start,
 140                                                    failed_len);
 141                         if (em) {
 142                                 printk("double failure lookup gives us "
 143                                        "%Lu %Lu -> %Lu\n", em->start,
 144                                        em->len, em->block_start);
 145                                 free_extent_map(em);
 146                         }
 147                         ret = -EIO;
 148                 }
 149         } else if (ret) {
 150                 free_extent_map(em);
 151                 em = NULL;
 152         }
 153         spin_unlock(&em_tree->lock);
 154
 155         if (ret)
 156                 em = ERR_PTR(ret);
 157 out:
 158         return em;
 159 }
 160
 161 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 162 {
 163         return btrfs_crc32c(seed, data, len);
 164 }
 165
 166 void btrfs_csum_final(u32 crc, char *result)
 167 {
 168         *(__le32 *)result = ~cpu_to_le32(crc);
 169 }
 170
 171 /*
 172  * compute the csum for a btree block, and either verify it or write it
 173  * into the csum field of the block.
 174  */
 175 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 176                            int verify)
 177 {
 178         char result[BTRFS_CRC32_SIZE];
 179         unsigned long len;
 180         unsigned long cur_len;
 181         unsigned long offset = BTRFS_CSUM_SIZE;
 182         char *map_token = NULL;
 183         char *kaddr;
 184         unsigned long map_start;
 185         unsigned long map_len;
 186         int err;
 187         u32 crc = ~(u32)0;
 188
 189         len = buf->len - offset;
 190         while(len > 0) {
 191                 err = map_private_extent_buffer(buf, offset, 32,
 192                                         &map_token, &kaddr,
 193                                         &map_start, &map_len, KM_USER0);
 194                 if (err) {
 195                         printk("failed to map extent buffer! %lu\n",
 196                                offset);
 197                         return 1;
 198                 }
 199                 cur_len = min(len, map_len - (offset - map_start));
 200                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 201                                       crc, cur_len);
 202                 len -= cur_len;
 203                 offset += cur_len;
 204                 unmap_extent_buffer(buf, map_token, KM_USER0);
 205         }
 206         btrfs_csum_final(crc, result);
 207
 208         if (verify) {
 209                 /* FIXME, this is not good */
 210                 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
 211                         u32 val;
 212                         u32 found = 0;
 213                         memcpy(&found, result, BTRFS_CRC32_SIZE);
 214
 215                         read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
 216                         printk("btrfs: %s checksum verify failed on %llu "
 217                                "wanted %X found %X level %d\n",
 218                                root->fs_info->sb->s_id,
 219                                buf->start, val, found, btrfs_header_level(buf));
 220                         return 1;
 221                 }
 222         } else {
 223                 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
 224         }
 225         return 0;
 226 }
 227
 228 /*
 229  * we can't consider a given block up to date unless the transid of the
 230  * block matches the transid in the parent node's pointer.  This is how we
 231  * detect blocks that either didn't get written at all or got written
 232  * in the wrong place.
 233  */
 234 static int verify_parent_transid(struct extent_io_tree *io_tree,
 235                                  struct extent_buffer *eb, u64 parent_transid)
 236 {
 237         int ret;
 238
 239         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 240                 return 0;
 241
 242         lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
 243         if (extent_buffer_uptodate(io_tree, eb) &&
 244             btrfs_header_generation(eb) == parent_transid) {
 245                 ret = 0;
 246                 goto out;
 247         }
 248         printk("parent transid verify failed on %llu wanted %llu found %llu\n",
 249                (unsigned long long)eb->start,
 250                (unsigned long long)parent_transid,
 251                (unsigned long long)btrfs_header_generation(eb));
 252         ret = 1;
 253         clear_extent_buffer_uptodate(io_tree, eb);
 254 out:
 255         unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
 256                       GFP_NOFS);
 257         return ret;
 258 }
 259
 260 /*
 261  * helper to read a given tree block, doing retries as required when
 262  * the checksums don't match and we have alternate mirrors to try.
 263  */
 264 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 265                                           struct extent_buffer *eb,
 266                                           u64 start, u64 parent_transid)
 267 {
 268         struct extent_io_tree *io_tree;
 269         int ret;
 270         int num_copies = 0;
 271         int mirror_num = 0;
 272
 273         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 274         while (1) {
 275                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 276                                                btree_get_extent, mirror_num);
 277                 if (!ret &&
 278                     !verify_parent_transid(io_tree, eb, parent_transid))
 279                         return ret;
 280 printk("read extent buffer pages failed with ret %d mirror no %d\n", ret, mirror_num);
 281                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 282                                               eb->start, eb->len);
 283                 if (num_copies == 1)
 284                         return ret;
 285
 286                 mirror_num++;
 287                 if (mirror_num > num_copies)
 288                         return ret;
 289         }
 290         return -EIO;
 291 }
 292
 293 /*
 294  * checksum a dirty tree block before IO.  This has extra checks to make
 295  * sure we only fill in the checksum field in the first page of a multi-page block
 296  */
 297 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 298 {
 299         struct extent_io_tree *tree;
 300         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 301         u64 found_start;
 302         int found_level;
 303         unsigned long len;
 304         struct extent_buffer *eb;
 305         int ret;
 306
 307         tree = &BTRFS_I(page->mapping->host)->io_tree;
 308
 309         if (page->private == EXTENT_PAGE_PRIVATE)
 310                 goto out;
 311         if (!page->private)
 312                 goto out;
 313         len = page->private >> 2;
 314         if (len == 0) {
 315                 WARN_ON(1);
 316         }
 317         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 318         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 319                                              btrfs_header_generation(eb));
 320         BUG_ON(ret);
 321         found_start = btrfs_header_bytenr(eb);
 322         if (found_start != start) {
 323                 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
 324                        start, found_start, len);
 325                 WARN_ON(1);
 326                 goto err;
 327         }
 328         if (eb->first_page != page) {
 329                 printk("bad first page %lu %lu\n", eb->first_page->index,
 330                        page->index);
 331                 WARN_ON(1);
 332                 goto err;
 333         }
 334         if (!PageUptodate(page)) {
 335                 printk("csum not up to date page %lu\n", page->index);
 336                 WARN_ON(1);
 337                 goto err;
 338         }
 339         found_level = btrfs_header_level(eb);
 340
 341         csum_tree_block(root, eb, 0);
 342 err:
 343         free_extent_buffer(eb);
 344 out:
 345         return 0;
 346 }
 347
 348 static int check_tree_block_fsid(struct btrfs_root *root,
 349                                  struct extent_buffer *eb)
 350 {
 351         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 352         u8 fsid[BTRFS_UUID_SIZE];
 353         int ret = 1;
 354
 355         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
 356                            BTRFS_FSID_SIZE);
 357         while (fs_devices) {
 358                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 359                         ret = 0;
 360                         break;
 361                 }
 362                 fs_devices = fs_devices->seed;
 363         }
 364         return ret;
 365 }
 366
 367 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 368                                struct extent_state *state)
 369 {
 370         struct extent_io_tree *tree;
 371         u64 found_start;
 372         int found_level;
 373         unsigned long len;
 374         struct extent_buffer *eb;
 375         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 376         int ret = 0;
 377
 378         tree = &BTRFS_I(page->mapping->host)->io_tree;
 379         if (page->private == EXTENT_PAGE_PRIVATE)
 380                 goto out;
 381         if (!page->private)
 382                 goto out;
 383         len = page->private >> 2;
 384         if (len == 0) {
 385                 WARN_ON(1);
 386         }
 387         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 388
 389         found_start = btrfs_header_bytenr(eb);
 390         if (found_start != start) {
 391                 printk("bad tree block start %llu %llu\n",
 392                        (unsigned long long)found_start,
 393                        (unsigned long long)eb->start);
 394                 ret = -EIO;
 395                 goto err;
 396         }
 397         if (eb->first_page != page) {
 398                 printk("bad first page %lu %lu\n", eb->first_page->index,
 399                        page->index);
 400                 WARN_ON(1);
 401                 ret = -EIO;
 402                 goto err;
 403         }
 404         if (check_tree_block_fsid(root, eb)) {
 405                 printk("bad fsid on block %Lu\n", eb->start);
 406                 ret = -EIO;
 407                 goto err;
 408         }
 409         found_level = btrfs_header_level(eb);
 410
 411         ret = csum_tree_block(root, eb, 1);
 412         if (ret)
 413                 ret = -EIO;
 414
 415         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 416         end = eb->start + end - 1;
 417 err:
 418         free_extent_buffer(eb);
 419 out:
 420         return ret;
 421 }
 422
 423 static void end_workqueue_bio(struct bio *bio, int err)
 424 {
 425         struct end_io_wq *end_io_wq = bio->bi_private;
 426         struct btrfs_fs_info *fs_info;
 427
 428         fs_info = end_io_wq->info;
 429         end_io_wq->error = err;
 430         end_io_wq->work.func = end_workqueue_fn;
 431         end_io_wq->work.flags = 0;
 432         if (bio->bi_rw & (1 << BIO_RW))
 433                 btrfs_queue_worker(&fs_info->endio_write_workers,
 434                                    &end_io_wq->work);
 435         else
 436                 btrfs_queue_worker(&fs_info->endio_workers, &end_io_wq->work);
 437 }
 438
 439 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 440                         int metadata)
 441 {
 442         struct end_io_wq *end_io_wq;
 443         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 444         if (!end_io_wq)
 445                 return -ENOMEM;
 446
 447         end_io_wq->private = bio->bi_private;
 448         end_io_wq->end_io = bio->bi_end_io;
 449         end_io_wq->info = info;
 450         end_io_wq->error = 0;
 451         end_io_wq->bio = bio;
 452         end_io_wq->metadata = metadata;
 453
 454         bio->bi_private = end_io_wq;
 455         bio->bi_end_io = end_workqueue_bio;
 456         return 0;
 457 }
 458
 459 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 460 {
 461         unsigned long limit = min_t(unsigned long,
 462                                     info->workers.max_workers,
 463                                     info->fs_devices->open_devices);
 464         return 256 * limit;
 465 }
 466
 467 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
 468 {
 469         return atomic_read(&info->nr_async_bios) >
 470                 btrfs_async_submit_limit(info);
 471 }
 472
 473 static void run_one_async_start(struct btrfs_work *work)
 474 {
 475         struct btrfs_fs_info *fs_info;
 476         struct async_submit_bio *async;
 477
 478         async = container_of(work, struct  async_submit_bio, work);
 479         fs_info = BTRFS_I(async->inode)->root->fs_info;
 480         async->submit_bio_start(async->inode, async->rw, async->bio,
 481                                async->mirror_num, async->bio_flags);
 482 }
 483
 484 static void run_one_async_done(struct btrfs_work *work)
 485 {
 486         struct btrfs_fs_info *fs_info;
 487         struct async_submit_bio *async;
 488         int limit;
 489
 490         async = container_of(work, struct  async_submit_bio, work);
 491         fs_info = BTRFS_I(async->inode)->root->fs_info;
 492
 493         limit = btrfs_async_submit_limit(fs_info);
 494         limit = limit * 2 / 3;
 495
 496         atomic_dec(&fs_info->nr_async_submits);
 497
 498         if (atomic_read(&fs_info->nr_async_submits) < limit &&
 499             waitqueue_active(&fs_info->async_submit_wait))
 500                 wake_up(&fs_info->async_submit_wait);
 501
 502         async->submit_bio_done(async->inode, async->rw, async->bio,
 503                                async->mirror_num, async->bio_flags);
 504 }
 505
 506 static void run_one_async_free(struct btrfs_work *work)
 507 {
 508         struct async_submit_bio *async;
 509
 510         async = container_of(work, struct  async_submit_bio, work);
 511         kfree(async);
 512 }
 513
 514 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 515                         int rw, struct bio *bio, int mirror_num,
 516                         unsigned long bio_flags,
 517                         extent_submit_bio_hook_t *submit_bio_start,
 518                         extent_submit_bio_hook_t *submit_bio_done)
 519 {
 520         struct async_submit_bio *async;
 521         int limit = btrfs_async_submit_limit(fs_info);
 522
 523         async = kmalloc(sizeof(*async), GFP_NOFS);
 524         if (!async)
 525                 return -ENOMEM;
 526
 527         async->inode = inode;
 528         async->rw = rw;
 529         async->bio = bio;
 530         async->mirror_num = mirror_num;
 531         async->submit_bio_start = submit_bio_start;
 532         async->submit_bio_done = submit_bio_done;
 533
 534         async->work.func = run_one_async_start;
 535         async->work.ordered_func = run_one_async_done;
 536         async->work.ordered_free = run_one_async_free;
 537
 538         async->work.flags = 0;
 539         async->bio_flags = bio_flags;
 540
 541         while(atomic_read(&fs_info->async_submit_draining) &&
 542               atomic_read(&fs_info->nr_async_submits)) {
 543                 wait_event(fs_info->async_submit_wait,
 544                            (atomic_read(&fs_info->nr_async_submits) == 0));
 545         }
 546
 547         atomic_inc(&fs_info->nr_async_submits);
 548         btrfs_queue_worker(&fs_info->workers, &async->work);
 549
 550         if (atomic_read(&fs_info->nr_async_submits) > limit) {
 551                 wait_event_timeout(fs_info->async_submit_wait,
 552                            (atomic_read(&fs_info->nr_async_submits) < limit),
 553                            HZ/10);
 554
 555                 wait_event_timeout(fs_info->async_submit_wait,
 556                            (atomic_read(&fs_info->nr_async_bios) < limit),
 557                            HZ/10);
 558         }
 559
 560         while(atomic_read(&fs_info->async_submit_draining) &&
 561               atomic_read(&fs_info->nr_async_submits)) {
 562                 wait_event(fs_info->async_submit_wait,
 563                            (atomic_read(&fs_info->nr_async_submits) == 0));
 564         }
 565
 566         return 0;
 567 }
 568
 569 static int btree_csum_one_bio(struct bio *bio)
 570 {
 571         struct bio_vec *bvec = bio->bi_io_vec;
 572         int bio_index = 0;
 573         struct btrfs_root *root;
 574
 575         WARN_ON(bio->bi_vcnt <= 0);
 576         while(bio_index < bio->bi_vcnt) {
 577                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 578                 csum_dirty_buffer(root, bvec->bv_page);
 579                 bio_index++;
 580                 bvec++;
 581         }
 582         return 0;
 583 }
 584
 585 static int __btree_submit_bio_start(struct inode *inode, int rw,
 586                                     struct bio *bio, int mirror_num,
 587                                     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         btree_csum_one_bio(bio);
 594         return 0;
 595 }
 596
 597 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 598                                  int mirror_num, unsigned long bio_flags)
 599 {
 600         /*
 601          * when we're called for a write, we're already in the async
 602          * submission context.  Just jump into btrfs_map_bio
 603          */
 604         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 605 }
 606
 607 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 608                                  int mirror_num, unsigned long bio_flags)
 609 {
 610         /*
 611          * kthread helpers are used to submit writes so that checksumming
 612          * can happen in parallel across all CPUs
 613          */
 614         if (!(rw & (1 << BIO_RW))) {
 615                 int ret;
 616                 /*
 617                  * called for a read, do the setup so that checksum validation
 618                  * can happen in the async kernel threads
 619                  */
 620                 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 621                                           bio, 1);
 622                 BUG_ON(ret);
 623
 624                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 625                                      mirror_num, 0);
 626         }
 627         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 628                                    inode, rw, bio, mirror_num, 0,
 629                                    __btree_submit_bio_start,
 630                                    __btree_submit_bio_done);
 631 }
 632
 633 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 634 {
 635         struct extent_io_tree *tree;
 636         tree = &BTRFS_I(page->mapping->host)->io_tree;
 637
 638         if (current->flags & PF_MEMALLOC) {
 639                 redirty_page_for_writepage(wbc, page);
 640                 unlock_page(page);
 641                 return 0;
 642         }
 643         return extent_write_full_page(tree, page, btree_get_extent, wbc);
 644 }
 645
 646 static int btree_writepages(struct address_space *mapping,
 647                             struct writeback_control *wbc)
 648 {
 649         struct extent_io_tree *tree;
 650         tree = &BTRFS_I(mapping->host)->io_tree;
 651         if (wbc->sync_mode == WB_SYNC_NONE) {
 652                 u64 num_dirty;
 653                 u64 start = 0;
 654                 unsigned long thresh = 32 * 1024 * 1024;
 655
 656                 if (wbc->for_kupdate)
 657                         return 0;
 658
 659                 num_dirty = count_range_bits(tree, &start, (u64)-1,
 660                                              thresh, EXTENT_DIRTY);
 661                 if (num_dirty < thresh) {
 662                         return 0;
 663                 }
 664         }
 665         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 666 }
 667
 668 int btree_readpage(struct file *file, struct page *page)
 669 {
 670         struct extent_io_tree *tree;
 671         tree = &BTRFS_I(page->mapping->host)->io_tree;
 672         return extent_read_full_page(tree, page, btree_get_extent);
 673 }
 674
 675 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 676 {
 677         struct extent_io_tree *tree;
 678         struct extent_map_tree *map;
 679         int ret;
 680
 681         if (PageWriteback(page) || PageDirty(page))
 682             return 0;
 683
 684         tree = &BTRFS_I(page->mapping->host)->io_tree;
 685         map = &BTRFS_I(page->mapping->host)->extent_tree;
 686
 687         ret = try_release_extent_state(map, tree, page, gfp_flags);
 688         if (!ret) {
 689                 return 0;
 690         }
 691
 692         ret = try_release_extent_buffer(tree, page);
 693         if (ret == 1) {
 694                 ClearPagePrivate(page);
 695                 set_page_private(page, 0);
 696                 page_cache_release(page);
 697         }
 698
 699         return ret;
 700 }
 701
 702 static void btree_invalidatepage(struct page *page, unsigned long offset)
 703 {
 704         struct extent_io_tree *tree;
 705         tree = &BTRFS_I(page->mapping->host)->io_tree;
 706         extent_invalidatepage(tree, page, offset);
 707         btree_releasepage(page, GFP_NOFS);
 708         if (PagePrivate(page)) {
 709                 printk("warning page private not zero on page %Lu\n",
 710                        page_offset(page));
 711                 ClearPagePrivate(page);
 712                 set_page_private(page, 0);
 713                 page_cache_release(page);
 714         }
 715 }
 716
 717 #if 0
 718 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 719 {
 720         struct buffer_head *bh;
 721         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 722         struct buffer_head *head;
 723         if (!page_has_buffers(page)) {
 724                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 725                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 726         }
 727         head = page_buffers(page);
 728         bh = head;
 729         do {
 730                 if (buffer_dirty(bh))
 731                         csum_tree_block(root, bh, 0);
 732                 bh = bh->b_this_page;
 733         } while (bh != head);
 734         return block_write_full_page(page, btree_get_block, wbc);
 735 }
 736 #endif
 737
 738 static struct address_space_operations btree_aops = {
 739         .readpage       = btree_readpage,
 740         .writepage      = btree_writepage,
 741         .writepages     = btree_writepages,
 742         .releasepage    = btree_releasepage,
 743         .invalidatepage = btree_invalidatepage,
 744         .sync_page      = block_sync_page,
 745 };
 746
 747 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 748                          u64 parent_transid)
 749 {
 750         struct extent_buffer *buf = NULL;
 751         struct inode *btree_inode = root->fs_info->btree_inode;
 752         int ret = 0;
 753
 754         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 755         if (!buf)
 756                 return 0;
 757         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 758                                  buf, 0, 0, btree_get_extent, 0);
 759         free_extent_buffer(buf);
 760         return ret;
 761 }
 762
 763 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 764                                             u64 bytenr, u32 blocksize)
 765 {
 766         struct inode *btree_inode = root->fs_info->btree_inode;
 767         struct extent_buffer *eb;
 768         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 769                                 bytenr, blocksize, GFP_NOFS);
 770         return eb;
 771 }
 772
 773 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 774                                                  u64 bytenr, u32 blocksize)
 775 {
 776         struct inode *btree_inode = root->fs_info->btree_inode;
 777         struct extent_buffer *eb;
 778
 779         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 780                                  bytenr, blocksize, NULL, GFP_NOFS);
 781         return eb;
 782 }
 783
 784
 785 int btrfs_write_tree_block(struct extent_buffer *buf)
 786 {
 787         return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
 788                                       buf->start + buf->len - 1, WB_SYNC_ALL);
 789 }
 790
 791 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 792 {
 793         return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
 794                                   buf->start, buf->start + buf->len -1);
 795 }
 796
 797 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 798                                       u32 blocksize, u64 parent_transid)
 799 {
 800         struct extent_buffer *buf = NULL;
 801         struct inode *btree_inode = root->fs_info->btree_inode;
 802         struct extent_io_tree *io_tree;
 803         int ret;
 804
 805         io_tree = &BTRFS_I(btree_inode)->io_tree;
 806
 807         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 808         if (!buf)
 809                 return NULL;
 810
 811         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 812
 813         if (ret == 0) {
 814                 buf->flags |= EXTENT_UPTODATE;
 815         } else {
 816                 WARN_ON(1);
 817         }
 818         return buf;
 819
 820 }
 821
 822 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 823                      struct extent_buffer *buf)
 824 {
 825         struct inode *btree_inode = root->fs_info->btree_inode;
 826         if (btrfs_header_generation(buf) ==
 827             root->fs_info->running_transaction->transid) {
 828                 WARN_ON(!btrfs_tree_locked(buf));
 829                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 830                                           buf);
 831         }
 832         return 0;
 833 }
 834
 835 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 836                         u32 stripesize, struct btrfs_root *root,
 837                         struct btrfs_fs_info *fs_info,
 838                         u64 objectid)
 839 {
 840         root->node = NULL;
 841         root->inode = NULL;
 842         root->commit_root = NULL;
 843         root->ref_tree = NULL;
 844         root->sectorsize = sectorsize;
 845         root->nodesize = nodesize;
 846         root->leafsize = leafsize;
 847         root->stripesize = stripesize;
 848         root->ref_cows = 0;
 849         root->track_dirty = 0;
 850
 851         root->fs_info = fs_info;
 852         root->objectid = objectid;
 853         root->last_trans = 0;
 854         root->highest_inode = 0;
 855         root->last_inode_alloc = 0;
 856         root->name = NULL;
 857         root->in_sysfs = 0;
 858
 859         INIT_LIST_HEAD(&root->dirty_list);
 860         INIT_LIST_HEAD(&root->orphan_list);
 861         INIT_LIST_HEAD(&root->dead_list);
 862         spin_lock_init(&root->node_lock);
 863         spin_lock_init(&root->list_lock);
 864         mutex_init(&root->objectid_mutex);
 865         mutex_init(&root->log_mutex);
 866         extent_io_tree_init(&root->dirty_log_pages,
 867                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 868
 869         btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
 870         root->ref_tree = &root->ref_tree_struct;
 871
 872         memset(&root->root_key, 0, sizeof(root->root_key));
 873         memset(&root->root_item, 0, sizeof(root->root_item));
 874         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 875         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 876         root->defrag_trans_start = fs_info->generation;
 877         init_completion(&root->kobj_unregister);
 878         root->defrag_running = 0;
 879         root->defrag_level = 0;
 880         root->root_key.objectid = objectid;
 881         return 0;
 882 }
 883
 884 static int find_and_setup_root(struct btrfs_root *tree_root,
 885                                struct btrfs_fs_info *fs_info,
 886                                u64 objectid,
 887                                struct btrfs_root *root)
 888 {
 889         int ret;
 890         u32 blocksize;
 891         u64 generation;
 892
 893         __setup_root(tree_root->nodesize, tree_root->leafsize,
 894                      tree_root->sectorsize, tree_root->stripesize,
 895                      root, fs_info, objectid);
 896         ret = btrfs_find_last_root(tree_root, objectid,
 897                                    &root->root_item, &root->root_key);
 898         BUG_ON(ret);
 899
 900         generation = btrfs_root_generation(&root->root_item);
 901         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 902         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 903                                      blocksize, generation);
 904         BUG_ON(!root->node);
 905         return 0;
 906 }
 907
 908 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
 909                              struct btrfs_fs_info *fs_info)
 910 {
 911         struct extent_buffer *eb;
 912         struct btrfs_root *log_root_tree = fs_info->log_root_tree;
 913         u64 start = 0;
 914         u64 end = 0;
 915         int ret;
 916
 917         if (!log_root_tree)
 918                 return 0;
 919
 920         while(1) {
 921                 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
 922                                     0, &start, &end, EXTENT_DIRTY);
 923                 if (ret)
 924                         break;
 925
 926                 clear_extent_dirty(&log_root_tree->dirty_log_pages,
 927                                    start, end, GFP_NOFS);
 928         }
 929         eb = fs_info->log_root_tree->node;
 930
 931         WARN_ON(btrfs_header_level(eb) != 0);
 932         WARN_ON(btrfs_header_nritems(eb) != 0);
 933
 934         ret = btrfs_free_reserved_extent(fs_info->tree_root,
 935                                 eb->start, eb->len);
 936         BUG_ON(ret);
 937
 938         free_extent_buffer(eb);
 939         kfree(fs_info->log_root_tree);
 940         fs_info->log_root_tree = NULL;
 941         return 0;
 942 }
 943
 944 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
 945                              struct btrfs_fs_info *fs_info)
 946 {
 947         struct btrfs_root *root;
 948         struct btrfs_root *tree_root = fs_info->tree_root;
 949
 950         root = kzalloc(sizeof(*root), GFP_NOFS);
 951         if (!root)
 952                 return -ENOMEM;
 953
 954         __setup_root(tree_root->nodesize, tree_root->leafsize,
 955                      tree_root->sectorsize, tree_root->stripesize,
 956                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
 957
 958         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
 959         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
 960         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
 961         root->ref_cows = 0;
 962
 963         root->node = btrfs_alloc_free_block(trans, root, root->leafsize,
 964                                             0, BTRFS_TREE_LOG_OBJECTID,
 965                                             trans->transid, 0, 0, 0);
 966
 967         btrfs_set_header_nritems(root->node, 0);
 968         btrfs_set_header_level(root->node, 0);
 969         btrfs_set_header_bytenr(root->node, root->node->start);
 970         btrfs_set_header_generation(root->node, trans->transid);
 971         btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
 972
 973         write_extent_buffer(root->node, root->fs_info->fsid,
 974                             (unsigned long)btrfs_header_fsid(root->node),
 975                             BTRFS_FSID_SIZE);
 976         btrfs_mark_buffer_dirty(root->node);
 977         btrfs_tree_unlock(root->node);
 978         fs_info->log_root_tree = root;
 979         return 0;
 980 }
 981
 982 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
 983                                                struct btrfs_key *location)
 984 {
 985         struct btrfs_root *root;
 986         struct btrfs_fs_info *fs_info = tree_root->fs_info;
 987         struct btrfs_path *path;
 988         struct extent_buffer *l;
 989         u64 highest_inode;
 990         u64 generation;
 991         u32 blocksize;
 992         int ret = 0;
 993
 994         root = kzalloc(sizeof(*root), GFP_NOFS);
 995         if (!root)
 996                 return ERR_PTR(-ENOMEM);
 997         if (location->offset == (u64)-1) {
 998                 ret = find_and_setup_root(tree_root, fs_info,
 999                                           location->objectid, root);
1000                 if (ret) {
1001                         kfree(root);
1002                         return ERR_PTR(ret);
1003                 }
1004                 goto insert;
1005         }
1006
1007         __setup_root(tree_root->nodesize, tree_root->leafsize,
1008                      tree_root->sectorsize, tree_root->stripesize,
1009                      root, fs_info, location->objectid);
1010
1011         path = btrfs_alloc_path();
1012         BUG_ON(!path);
1013         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1014         if (ret != 0) {
1015                 if (ret > 0)
1016                         ret = -ENOENT;
1017                 goto out;
1018         }
1019         l = path->nodes[0];
1020         read_extent_buffer(l, &root->root_item,
1021                btrfs_item_ptr_offset(l, path->slots[0]),
1022                sizeof(root->root_item));
1023         memcpy(&root->root_key, location, sizeof(*location));
1024         ret = 0;
1025 out:
1026         btrfs_release_path(root, path);
1027         btrfs_free_path(path);
1028         if (ret) {
1029                 kfree(root);
1030                 return ERR_PTR(ret);
1031         }
1032         generation = btrfs_root_generation(&root->root_item);
1033         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1034         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1035                                      blocksize, generation);
1036         BUG_ON(!root->node);
1037 insert:
1038         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1039                 root->ref_cows = 1;
1040                 ret = btrfs_find_highest_inode(root, &highest_inode);
1041                 if (ret == 0) {
1042                         root->highest_inode = highest_inode;
1043                         root->last_inode_alloc = highest_inode;
1044                 }
1045         }
1046         return root;
1047 }
1048
1049 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1050                                         u64 root_objectid)
1051 {
1052         struct btrfs_root *root;
1053
1054         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1055                 return fs_info->tree_root;
1056         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1057                 return fs_info->extent_root;
1058
1059         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1060                                  (unsigned long)root_objectid);
1061         return root;
1062 }
1063
1064 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1065                                               struct btrfs_key *location)
1066 {
1067         struct btrfs_root *root;
1068         int ret;
1069
1070         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1071                 return fs_info->tree_root;
1072         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1073                 return fs_info->extent_root;
1074         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1075                 return fs_info->chunk_root;
1076         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1077                 return fs_info->dev_root;
1078
1079         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1080                                  (unsigned long)location->objectid);
1081         if (root)
1082                 return root;
1083
1084         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1085         if (IS_ERR(root))
1086                 return root;
1087         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1088                                 (unsigned long)root->root_key.objectid,
1089                                 root);
1090         if (ret) {
1091                 free_extent_buffer(root->node);
1092                 kfree(root);
1093                 return ERR_PTR(ret);
1094         }
1095         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
1096                 ret = btrfs_find_dead_roots(fs_info->tree_root,
1097                                             root->root_key.objectid, root);
1098                 BUG_ON(ret);
1099                 btrfs_orphan_cleanup(root);
1100         }
1101         return root;
1102 }
1103
1104 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1105                                       struct btrfs_key *location,
1106                                       const char *name, int namelen)
1107 {
1108         struct btrfs_root *root;
1109         int ret;
1110
1111         root = btrfs_read_fs_root_no_name(fs_info, location);
1112         if (!root)
1113                 return NULL;
1114
1115         if (root->in_sysfs)
1116                 return root;
1117
1118         ret = btrfs_set_root_name(root, name, namelen);
1119         if (ret) {
1120                 free_extent_buffer(root->node);
1121                 kfree(root);
1122                 return ERR_PTR(ret);
1123         }
1124
1125         ret = btrfs_sysfs_add_root(root);
1126         if (ret) {
1127                 free_extent_buffer(root->node);
1128                 kfree(root->name);
1129                 kfree(root);
1130                 return ERR_PTR(ret);
1131         }
1132         root->in_sysfs = 1;
1133         return root;
1134 }
1135 #if 0
1136 static int add_hasher(struct btrfs_fs_info *info, char *type) {
1137         struct btrfs_hasher *hasher;
1138
1139         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
1140         if (!hasher)
1141                 return -ENOMEM;
1142         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
1143         if (!hasher->hash_tfm) {
1144                 kfree(hasher);
1145                 return -EINVAL;
1146         }
1147         spin_lock(&info->hash_lock);
1148         list_add(&hasher->list, &info->hashers);
1149         spin_unlock(&info->hash_lock);
1150         return 0;
1151 }
1152 #endif
1153
1154 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1155 {
1156         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1157         int ret = 0;
1158         struct list_head *cur;
1159         struct btrfs_device *device;
1160         struct backing_dev_info *bdi;
1161 #if 0
1162         if ((bdi_bits & (1 << BDI_write_congested)) &&
1163             btrfs_congested_async(info, 0))
1164                 return 1;
1165 #endif
1166         list_for_each(cur, &info->fs_devices->devices) {
1167                 device = list_entry(cur, struct btrfs_device, dev_list);
1168                 if (!device->bdev)
1169                         continue;
1170                 bdi = blk_get_backing_dev_info(device->bdev);
1171                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1172                         ret = 1;
1173                         break;
1174                 }
1175         }
1176         return ret;
1177 }
1178
1179 /*
1180  * this unplugs every device on the box, and it is only used when page
1181  * is null
1182  */
1183 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1184 {
1185         struct list_head *cur;
1186         struct btrfs_device *device;
1187         struct btrfs_fs_info *info;
1188
1189         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1190         list_for_each(cur, &info->fs_devices->devices) {
1191                 device = list_entry(cur, struct btrfs_device, dev_list);
1192                 bdi = blk_get_backing_dev_info(device->bdev);
1193                 if (bdi->unplug_io_fn) {
1194                         bdi->unplug_io_fn(bdi, page);
1195                 }
1196         }
1197 }
1198
1199 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1200 {
1201         struct inode *inode;
1202         struct extent_map_tree *em_tree;
1203         struct extent_map *em;
1204         struct address_space *mapping;
1205         u64 offset;
1206
1207         /* the generic O_DIRECT read code does this */
1208         if (!page) {
1209                 __unplug_io_fn(bdi, page);
1210                 return;
1211         }
1212
1213         /*
1214          * page->mapping may change at any time.  Get a consistent copy
1215          * and use that for everything below
1216          */
1217         smp_mb();
1218         mapping = page->mapping;
1219         if (!mapping)
1220                 return;
1221
1222         inode = mapping->host;
1223
1224         /*
1225          * don't do the expensive searching for a small number of
1226          * devices
1227          */
1228         if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1229                 __unplug_io_fn(bdi, page);
1230                 return;
1231         }
1232
1233         offset = page_offset(page);
1234
1235         em_tree = &BTRFS_I(inode)->extent_tree;
1236         spin_lock(&em_tree->lock);
1237         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1238         spin_unlock(&em_tree->lock);
1239         if (!em) {
1240                 __unplug_io_fn(bdi, page);
1241                 return;
1242         }
1243
1244         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1245                 free_extent_map(em);
1246                 __unplug_io_fn(bdi, page);
1247                 return;
1248         }
1249         offset = offset - em->start;
1250         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1251                           em->block_start + offset, page);
1252         free_extent_map(em);
1253 }
1254
1255 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1256 {
1257         bdi_init(bdi);
1258         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1259         bdi->state              = 0;
1260         bdi->capabilities       = default_backing_dev_info.capabilities;
1261         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1262         bdi->unplug_io_data     = info;
1263         bdi->congested_fn       = btrfs_congested_fn;
1264         bdi->congested_data     = info;
1265         return 0;
1266 }
1267
1268 static int bio_ready_for_csum(struct bio *bio)
1269 {
1270         u64 length = 0;
1271         u64 buf_len = 0;
1272         u64 start = 0;
1273         struct page *page;
1274         struct extent_io_tree *io_tree = NULL;
1275         struct btrfs_fs_info *info = NULL;
1276         struct bio_vec *bvec;
1277         int i;
1278         int ret;
1279
1280         bio_for_each_segment(bvec, bio, i) {
1281                 page = bvec->bv_page;
1282                 if (page->private == EXTENT_PAGE_PRIVATE) {
1283                         length += bvec->bv_len;
1284                         continue;
1285                 }
1286                 if (!page->private) {
1287                         length += bvec->bv_len;
1288                         continue;
1289                 }
1290                 length = bvec->bv_len;
1291                 buf_len = page->private >> 2;
1292                 start = page_offset(page) + bvec->bv_offset;
1293                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1294                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1295         }
1296         /* are we fully contained in this bio? */
1297         if (buf_len <= length)
1298                 return 1;
1299
1300         ret = extent_range_uptodate(io_tree, start + length,
1301                                     start + buf_len - 1);
1302         if (ret == 1)
1303                 return ret;
1304         return ret;
1305 }
1306
1307 /*
1308  * called by the kthread helper functions to finally call the bio end_io
1309  * functions.  This is where read checksum verification actually happens
1310  */
1311 static void end_workqueue_fn(struct btrfs_work *work)
1312 {
1313         struct bio *bio;
1314         struct end_io_wq *end_io_wq;
1315         struct btrfs_fs_info *fs_info;
1316         int error;
1317
1318         end_io_wq = container_of(work, struct end_io_wq, work);
1319         bio = end_io_wq->bio;
1320         fs_info = end_io_wq->info;
1321
1322         /* metadata bios are special because the whole tree block must
1323          * be checksummed at once.  This makes sure the entire block is in
1324          * ram and up to date before trying to verify things.  For
1325          * blocksize <= pagesize, it is basically a noop
1326          */
1327         if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1328                 btrfs_queue_worker(&fs_info->endio_workers,
1329                                    &end_io_wq->work);
1330                 return;
1331         }
1332         error = end_io_wq->error;
1333         bio->bi_private = end_io_wq->private;
1334         bio->bi_end_io = end_io_wq->end_io;
1335         kfree(end_io_wq);
1336         bio_endio(bio, error);
1337 }
1338
1339 static int cleaner_kthread(void *arg)
1340 {
1341         struct btrfs_root *root = arg;
1342
1343         do {
1344                 smp_mb();
1345                 if (root->fs_info->closing)
1346                         break;
1347
1348                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1349                 mutex_lock(&root->fs_info->cleaner_mutex);
1350                 btrfs_clean_old_snapshots(root);
1351                 mutex_unlock(&root->fs_info->cleaner_mutex);
1352
1353                 if (freezing(current)) {
1354                         refrigerator();
1355                 } else {
1356                         smp_mb();
1357                         if (root->fs_info->closing)
1358                                 break;
1359                         set_current_state(TASK_INTERRUPTIBLE);
1360                         schedule();
1361                         __set_current_state(TASK_RUNNING);
1362                 }
1363         } while (!kthread_should_stop());
1364         return 0;
1365 }
1366
1367 static int transaction_kthread(void *arg)
1368 {
1369         struct btrfs_root *root = arg;
1370         struct btrfs_trans_handle *trans;
1371         struct btrfs_transaction *cur;
1372         unsigned long now;
1373         unsigned long delay;
1374         int ret;
1375
1376         do {
1377                 smp_mb();
1378                 if (root->fs_info->closing)
1379                         break;
1380
1381                 delay = HZ * 30;
1382                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1383                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1384
1385                 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1386                         printk("btrfs: total reference cache size %Lu\n",
1387                                 root->fs_info->total_ref_cache_size);
1388                 }
1389
1390                 mutex_lock(&root->fs_info->trans_mutex);
1391                 cur = root->fs_info->running_transaction;
1392                 if (!cur) {
1393                         mutex_unlock(&root->fs_info->trans_mutex);
1394                         goto sleep;
1395                 }
1396
1397                 now = get_seconds();
1398                 if (now < cur->start_time || now - cur->start_time < 30) {
1399                         mutex_unlock(&root->fs_info->trans_mutex);
1400                         delay = HZ * 5;
1401                         goto sleep;
1402                 }
1403                 mutex_unlock(&root->fs_info->trans_mutex);
1404                 trans = btrfs_start_transaction(root, 1);
1405                 ret = btrfs_commit_transaction(trans, root);
1406 sleep:
1407                 wake_up_process(root->fs_info->cleaner_kthread);
1408                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1409
1410                 if (freezing(current)) {
1411                         refrigerator();
1412                 } else {
1413                         if (root->fs_info->closing)
1414                                 break;
1415                         set_current_state(TASK_INTERRUPTIBLE);
1416                         schedule_timeout(delay);
1417                         __set_current_state(TASK_RUNNING);
1418                 }
1419         } while (!kthread_should_stop());
1420         return 0;
1421 }
1422
1423 struct btrfs_root *open_ctree(struct super_block *sb,
1424                               struct btrfs_fs_devices *fs_devices,
1425                               char *options)
1426 {
1427         u32 sectorsize;
1428         u32 nodesize;
1429         u32 leafsize;
1430         u32 blocksize;
1431         u32 stripesize;
1432         u64 generation;
1433         struct buffer_head *bh;
1434         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1435                                                  GFP_NOFS);
1436         struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1437                                                GFP_NOFS);
1438         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1439                                                 GFP_NOFS);
1440         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1441                                                 GFP_NOFS);
1442         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1443                                               GFP_NOFS);
1444         struct btrfs_root *log_tree_root;
1445
1446         int ret;
1447         int err = -EINVAL;
1448
1449         struct btrfs_super_block *disk_super;
1450
1451         if (!extent_root || !tree_root || !fs_info ||
1452             !chunk_root || !dev_root) {
1453                 err = -ENOMEM;
1454                 goto fail;
1455         }
1456         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1457         INIT_LIST_HEAD(&fs_info->trans_list);
1458         INIT_LIST_HEAD(&fs_info->dead_roots);
1459         INIT_LIST_HEAD(&fs_info->hashers);
1460         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1461         spin_lock_init(&fs_info->hash_lock);
1462         spin_lock_init(&fs_info->delalloc_lock);
1463         spin_lock_init(&fs_info->new_trans_lock);
1464         spin_lock_init(&fs_info->ref_cache_lock);
1465
1466         init_completion(&fs_info->kobj_unregister);
1467         fs_info->tree_root = tree_root;
1468         fs_info->extent_root = extent_root;
1469         fs_info->chunk_root = chunk_root;
1470         fs_info->dev_root = dev_root;
1471         fs_info->fs_devices = fs_devices;
1472         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1473         INIT_LIST_HEAD(&fs_info->space_info);
1474         btrfs_mapping_init(&fs_info->mapping_tree);
1475         atomic_set(&fs_info->nr_async_submits, 0);
1476         atomic_set(&fs_info->async_delalloc_pages, 0);
1477         atomic_set(&fs_info->async_submit_draining, 0);
1478         atomic_set(&fs_info->nr_async_bios, 0);
1479         atomic_set(&fs_info->throttles, 0);
1480         atomic_set(&fs_info->throttle_gen, 0);
1481         fs_info->sb = sb;
1482         fs_info->max_extent = (u64)-1;
1483         fs_info->max_inline = 8192 * 1024;
1484         setup_bdi(fs_info, &fs_info->bdi);
1485         fs_info->btree_inode = new_inode(sb);
1486         fs_info->btree_inode->i_ino = 1;
1487         fs_info->btree_inode->i_nlink = 1;
1488
1489         fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1490
1491         INIT_LIST_HEAD(&fs_info->ordered_extents);
1492         spin_lock_init(&fs_info->ordered_extent_lock);
1493
1494         sb->s_blocksize = 4096;
1495         sb->s_blocksize_bits = blksize_bits(4096);
1496
1497         /*
1498          * we set the i_size on the btree inode to the max possible int.
1499          * the real end of the address space is determined by all of
1500          * the devices in the system
1501          */
1502         fs_info->btree_inode->i_size = OFFSET_MAX;
1503         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1504         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1505
1506         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1507                              fs_info->btree_inode->i_mapping,
1508                              GFP_NOFS);
1509         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1510                              GFP_NOFS);
1511
1512         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1513
1514         spin_lock_init(&fs_info->block_group_cache_lock);
1515         fs_info->block_group_cache_tree.rb_node = NULL;
1516
1517         extent_io_tree_init(&fs_info->pinned_extents,
1518                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1519         extent_io_tree_init(&fs_info->pending_del,
1520                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1521         extent_io_tree_init(&fs_info->extent_ins,
1522                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1523         fs_info->do_barriers = 1;
1524
1525         INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
1526         btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
1527         btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
1528
1529         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1530         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1531                sizeof(struct btrfs_key));
1532         insert_inode_hash(fs_info->btree_inode);
1533
1534         mutex_init(&fs_info->trans_mutex);
1535         mutex_init(&fs_info->tree_log_mutex);
1536         mutex_init(&fs_info->drop_mutex);
1537         mutex_init(&fs_info->extent_ins_mutex);
1538         mutex_init(&fs_info->pinned_mutex);
1539         mutex_init(&fs_info->chunk_mutex);
1540         mutex_init(&fs_info->transaction_kthread_mutex);
1541         mutex_init(&fs_info->cleaner_mutex);
1542         mutex_init(&fs_info->volume_mutex);
1543         mutex_init(&fs_info->tree_reloc_mutex);
1544         init_waitqueue_head(&fs_info->transaction_throttle);
1545         init_waitqueue_head(&fs_info->transaction_wait);
1546         init_waitqueue_head(&fs_info->async_submit_wait);
1547         init_waitqueue_head(&fs_info->tree_log_wait);
1548         atomic_set(&fs_info->tree_log_commit, 0);
1549         atomic_set(&fs_info->tree_log_writers, 0);
1550         fs_info->tree_log_transid = 0;
1551
1552 #if 0
1553         ret = add_hasher(fs_info, "crc32c");
1554         if (ret) {
1555                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1556                 err = -ENOMEM;
1557                 goto fail_iput;
1558         }
1559 #endif
1560         __setup_root(4096, 4096, 4096, 4096, tree_root,
1561                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1562
1563
1564         bh = __bread(fs_devices->latest_bdev,
1565                      BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1566         if (!bh)
1567                 goto fail_iput;
1568
1569         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1570         brelse(bh);
1571
1572         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1573
1574         disk_super = &fs_info->super_copy;
1575         if (!btrfs_super_root(disk_super))
1576                 goto fail_sb_buffer;
1577
1578         ret = btrfs_parse_options(tree_root, options);
1579         if (ret) {
1580                 err = ret;
1581                 goto fail_sb_buffer;
1582         }
1583
1584         /*
1585          * we need to start all the end_io workers up front because the
1586          * queue work function gets called at interrupt time, and so it
1587          * cannot dynamically grow.
1588          */
1589         btrfs_init_workers(&fs_info->workers, "worker",
1590                            fs_info->thread_pool_size);
1591
1592         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1593                            fs_info->thread_pool_size);
1594
1595         btrfs_init_workers(&fs_info->submit_workers, "submit",
1596                            min_t(u64, fs_devices->num_devices,
1597                            fs_info->thread_pool_size));
1598
1599         /* a higher idle thresh on the submit workers makes it much more
1600          * likely that bios will be send down in a sane order to the
1601          * devices
1602          */
1603         fs_info->submit_workers.idle_thresh = 64;
1604
1605         fs_info->workers.idle_thresh = 16;
1606         fs_info->workers.ordered = 1;
1607
1608         fs_info->delalloc_workers.idle_thresh = 2;
1609         fs_info->delalloc_workers.ordered = 1;
1610
1611         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1612         btrfs_init_workers(&fs_info->endio_workers, "endio",
1613                            fs_info->thread_pool_size);
1614         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1615                            fs_info->thread_pool_size);
1616
1617         /*
1618          * endios are largely parallel and should have a very
1619          * low idle thresh
1620          */
1621         fs_info->endio_workers.idle_thresh = 4;
1622         fs_info->endio_write_workers.idle_thresh = 64;
1623
1624         btrfs_start_workers(&fs_info->workers, 1);
1625         btrfs_start_workers(&fs_info->submit_workers, 1);
1626         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1627         btrfs_start_workers(&fs_info->fixup_workers, 1);
1628         btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1629         btrfs_start_workers(&fs_info->endio_write_workers,
1630                             fs_info->thread_pool_size);
1631
1632         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1633         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1634                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1635
1636         nodesize = btrfs_super_nodesize(disk_super);
1637         leafsize = btrfs_super_leafsize(disk_super);
1638         sectorsize = btrfs_super_sectorsize(disk_super);
1639         stripesize = btrfs_super_stripesize(disk_super);
1640         tree_root->nodesize = nodesize;
1641         tree_root->leafsize = leafsize;
1642         tree_root->sectorsize = sectorsize;
1643         tree_root->stripesize = stripesize;
1644
1645         sb->s_blocksize = sectorsize;
1646         sb->s_blocksize_bits = blksize_bits(sectorsize);
1647
1648         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1649                     sizeof(disk_super->magic))) {
1650                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1651                 goto fail_sb_buffer;
1652         }
1653
1654         mutex_lock(&fs_info->chunk_mutex);
1655         ret = btrfs_read_sys_array(tree_root);
1656         mutex_unlock(&fs_info->chunk_mutex);
1657         if (ret) {
1658                 printk("btrfs: failed to read the system array on %s\n",
1659                        sb->s_id);
1660                 goto fail_sys_array;
1661         }
1662
1663         blocksize = btrfs_level_size(tree_root,
1664                                      btrfs_super_chunk_root_level(disk_super));
1665         generation = btrfs_super_chunk_root_generation(disk_super);
1666
1667         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1668                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1669
1670         chunk_root->node = read_tree_block(chunk_root,
1671                                            btrfs_super_chunk_root(disk_super),
1672                                            blocksize, generation);
1673         BUG_ON(!chunk_root->node);
1674
1675         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1676                  (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1677                  BTRFS_UUID_SIZE);
1678
1679         mutex_lock(&fs_info->chunk_mutex);
1680         ret = btrfs_read_chunk_tree(chunk_root);
1681         mutex_unlock(&fs_info->chunk_mutex);
1682         if (ret) {
1683                 printk("btrfs: failed to read chunk tree on %s\n", sb->s_id);
1684                 goto fail_chunk_root;
1685         }
1686
1687         btrfs_close_extra_devices(fs_devices);
1688
1689         blocksize = btrfs_level_size(tree_root,
1690                                      btrfs_super_root_level(disk_super));
1691         generation = btrfs_super_generation(disk_super);
1692
1693         tree_root->node = read_tree_block(tree_root,
1694                                           btrfs_super_root(disk_super),
1695                                           blocksize, generation);
1696         if (!tree_root->node)
1697                 goto fail_chunk_root;
1698
1699
1700         ret = find_and_setup_root(tree_root, fs_info,
1701                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1702         if (ret)
1703                 goto fail_tree_root;
1704         extent_root->track_dirty = 1;
1705
1706         ret = find_and_setup_root(tree_root, fs_info,
1707                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1708         dev_root->track_dirty = 1;
1709
1710         if (ret)
1711                 goto fail_extent_root;
1712
1713         btrfs_read_block_groups(extent_root);
1714
1715         fs_info->generation = generation + 1;
1716         fs_info->last_trans_committed = generation;
1717         fs_info->data_alloc_profile = (u64)-1;
1718         fs_info->metadata_alloc_profile = (u64)-1;
1719         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1720         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1721                                                "btrfs-cleaner");
1722         if (!fs_info->cleaner_kthread)
1723                 goto fail_extent_root;
1724
1725         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1726                                                    tree_root,
1727                                                    "btrfs-transaction");
1728         if (!fs_info->transaction_kthread)
1729                 goto fail_cleaner;
1730
1731         if (sb->s_flags & MS_RDONLY)
1732                 return tree_root;
1733
1734         if (btrfs_super_log_root(disk_super) != 0) {
1735                 u32 blocksize;
1736                 u64 bytenr = btrfs_super_log_root(disk_super);
1737
1738                 blocksize =
1739                      btrfs_level_size(tree_root,
1740                                       btrfs_super_log_root_level(disk_super));
1741
1742                 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1743                                                       GFP_NOFS);
1744
1745                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1746                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1747
1748                 log_tree_root->node = read_tree_block(tree_root, bytenr,
1749                                                       blocksize,
1750                                                       generation + 1);
1751                 ret = btrfs_recover_log_trees(log_tree_root);
1752                 BUG_ON(ret);
1753         }
1754
1755         ret = btrfs_cleanup_reloc_trees(tree_root);
1756         BUG_ON(ret);
1757
1758         return tree_root;
1759
1760 fail_cleaner:
1761         kthread_stop(fs_info->cleaner_kthread);
1762 fail_extent_root:
1763         free_extent_buffer(extent_root->node);
1764 fail_tree_root:
1765         free_extent_buffer(tree_root->node);
1766 fail_chunk_root:
1767         free_extent_buffer(chunk_root->node);
1768 fail_sys_array:
1769 fail_sb_buffer:
1770         btrfs_stop_workers(&fs_info->fixup_workers);
1771         btrfs_stop_workers(&fs_info->delalloc_workers);
1772         btrfs_stop_workers(&fs_info->workers);
1773         btrfs_stop_workers(&fs_info->endio_workers);
1774         btrfs_stop_workers(&fs_info->endio_write_workers);
1775         btrfs_stop_workers(&fs_info->submit_workers);
1776 fail_iput:
1777         iput(fs_info->btree_inode);
1778 fail:
1779         btrfs_close_devices(fs_info->fs_devices);
1780         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1781
1782         kfree(extent_root);
1783         kfree(tree_root);
1784         bdi_destroy(&fs_info->bdi);
1785         kfree(fs_info);
1786         kfree(chunk_root);
1787         kfree(dev_root);
1788         return ERR_PTR(err);
1789 }
1790
1791 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1792 {
1793         char b[BDEVNAME_SIZE];
1794
1795         if (uptodate) {
1796                 set_buffer_uptodate(bh);
1797         } else {
1798                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1799                         printk(KERN_WARNING "lost page write due to "
1800                                         "I/O error on %s\n",
1801                                        bdevname(bh->b_bdev, b));
1802                 }
1803                 /* note, we dont' set_buffer_write_io_error because we have
1804                  * our own ways of dealing with the IO errors
1805                  */
1806                 clear_buffer_uptodate(bh);
1807         }
1808         unlock_buffer(bh);
1809         put_bh(bh);
1810 }
1811
1812 int write_all_supers(struct btrfs_root *root)
1813 {
1814         struct list_head *cur;
1815         struct list_head *head = &root->fs_info->fs_devices->devices;
1816         struct btrfs_device *dev;
1817         struct btrfs_super_block *sb;
1818         struct btrfs_dev_item *dev_item;
1819         struct buffer_head *bh;
1820         int ret;
1821         int do_barriers;
1822         int max_errors;
1823         int total_errors = 0;
1824         u32 crc;
1825         u64 flags;
1826
1827         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1828         do_barriers = !btrfs_test_opt(root, NOBARRIER);
1829
1830         sb = &root->fs_info->super_for_commit;
1831         dev_item = &sb->dev_item;
1832         list_for_each(cur, head) {
1833                 dev = list_entry(cur, struct btrfs_device, dev_list);
1834                 if (!dev->bdev) {
1835                         total_errors++;
1836                         continue;
1837                 }
1838                 if (!dev->in_fs_metadata || !dev->writeable)
1839                         continue;
1840
1841                 btrfs_set_stack_device_generation(dev_item, 0);
1842                 btrfs_set_stack_device_type(dev_item, dev->type);
1843                 btrfs_set_stack_device_id(dev_item, dev->devid);
1844                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1845                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1846                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1847                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1848                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1849                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1850                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
1851                 flags = btrfs_super_flags(sb);
1852                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1853
1854
1855                 crc = ~(u32)0;
1856                 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1857                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1858                 btrfs_csum_final(crc, sb->csum);
1859
1860                 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1861                               BTRFS_SUPER_INFO_SIZE);
1862
1863                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1864                 dev->pending_io = bh;
1865
1866                 get_bh(bh);
1867                 set_buffer_uptodate(bh);
1868                 lock_buffer(bh);
1869                 bh->b_end_io = btrfs_end_buffer_write_sync;
1870
1871                 if (do_barriers && dev->barriers) {
1872                         ret = submit_bh(WRITE_BARRIER, bh);
1873                         if (ret == -EOPNOTSUPP) {
1874                                 printk("btrfs: disabling barriers on dev %s\n",
1875                                        dev->name);
1876                                 set_buffer_uptodate(bh);
1877                                 dev->barriers = 0;
1878                                 get_bh(bh);
1879                                 lock_buffer(bh);
1880                                 ret = submit_bh(WRITE, bh);
1881                         }
1882                 } else {
1883                         ret = submit_bh(WRITE, bh);
1884                 }
1885                 if (ret)
1886                         total_errors++;
1887         }
1888         if (total_errors > max_errors) {
1889                 printk("btrfs: %d errors while writing supers\n", total_errors);
1890                 BUG();
1891         }
1892         total_errors = 0;
1893
1894         list_for_each(cur, head) {
1895                 dev = list_entry(cur, struct btrfs_device, dev_list);
1896                 if (!dev->bdev)
1897                         continue;
1898                 if (!dev->in_fs_metadata || !dev->writeable)
1899                         continue;
1900
1901                 BUG_ON(!dev->pending_io);
1902                 bh = dev->pending_io;
1903                 wait_on_buffer(bh);
1904                 if (!buffer_uptodate(dev->pending_io)) {
1905                         if (do_barriers && dev->barriers) {
1906                                 printk("btrfs: disabling barriers on dev %s\n",
1907                                        dev->name);
1908                                 set_buffer_uptodate(bh);
1909                                 get_bh(bh);
1910                                 lock_buffer(bh);
1911                                 dev->barriers = 0;
1912                                 ret = submit_bh(WRITE, bh);
1913                                 BUG_ON(ret);
1914                                 wait_on_buffer(bh);
1915                                 if (!buffer_uptodate(bh))
1916                                         total_errors++;
1917                         } else {
1918                                 total_errors++;
1919                         }
1920
1921                 }
1922                 dev->pending_io = NULL;
1923                 brelse(bh);
1924         }
1925         if (total_errors > max_errors) {
1926                 printk("btrfs: %d errors while writing supers\n", total_errors);
1927                 BUG();
1928         }
1929         return 0;
1930 }
1931
1932 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1933                       *root)
1934 {
1935         int ret;
1936
1937         ret = write_all_supers(root);
1938         return ret;
1939 }
1940
1941 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1942 {
1943         radix_tree_delete(&fs_info->fs_roots_radix,
1944                           (unsigned long)root->root_key.objectid);
1945         if (root->in_sysfs)
1946                 btrfs_sysfs_del_root(root);
1947         if (root->inode)
1948                 iput(root->inode);
1949         if (root->node)
1950                 free_extent_buffer(root->node);
1951         if (root->commit_root)
1952                 free_extent_buffer(root->commit_root);
1953         if (root->name)
1954                 kfree(root->name);
1955         kfree(root);
1956         return 0;
1957 }
1958
1959 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1960 {
1961         int ret;
1962         struct btrfs_root *gang[8];
1963         int i;
1964
1965         while(1) {
1966                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1967                                              (void **)gang, 0,
1968                                              ARRAY_SIZE(gang));
1969                 if (!ret)
1970                         break;
1971                 for (i = 0; i < ret; i++)
1972                         btrfs_free_fs_root(fs_info, gang[i]);
1973         }
1974         return 0;
1975 }
1976
1977 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
1978 {
1979         u64 root_objectid = 0;
1980         struct btrfs_root *gang[8];
1981         int i;
1982         int ret;
1983
1984         while (1) {
1985                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1986                                              (void **)gang, root_objectid,
1987                                              ARRAY_SIZE(gang));
1988                 if (!ret)
1989                         break;
1990                 for (i = 0; i < ret; i++) {
1991                         root_objectid = gang[i]->root_key.objectid;
1992                         ret = btrfs_find_dead_roots(fs_info->tree_root,
1993                                                     root_objectid, gang[i]);
1994                         BUG_ON(ret);
1995                         btrfs_orphan_cleanup(gang[i]);
1996                 }
1997                 root_objectid++;
1998         }
1999         return 0;
2000 }
2001
2002 int btrfs_commit_super(struct btrfs_root *root)
2003 {
2004         struct btrfs_trans_handle *trans;
2005         int ret;
2006
2007         mutex_lock(&root->fs_info->cleaner_mutex);
2008         btrfs_clean_old_snapshots(root);
2009         mutex_unlock(&root->fs_info->cleaner_mutex);
2010         trans = btrfs_start_transaction(root, 1);
2011         ret = btrfs_commit_transaction(trans, root);
2012         BUG_ON(ret);
2013         /* run commit again to drop the original snapshot */
2014         trans = btrfs_start_transaction(root, 1);
2015         btrfs_commit_transaction(trans, root);
2016         ret = btrfs_write_and_wait_transaction(NULL, root);
2017         BUG_ON(ret);
2018
2019         ret = write_ctree_super(NULL, root);
2020         return ret;
2021 }
2022
2023 int close_ctree(struct btrfs_root *root)
2024 {
2025         struct btrfs_fs_info *fs_info = root->fs_info;
2026         int ret;
2027
2028         fs_info->closing = 1;
2029         smp_mb();
2030
2031         kthread_stop(root->fs_info->transaction_kthread);
2032         kthread_stop(root->fs_info->cleaner_kthread);
2033
2034         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2035                 ret =  btrfs_commit_super(root);
2036                 if (ret) {
2037                         printk("btrfs: commit super returns %d\n", ret);
2038                 }
2039         }
2040
2041         if (fs_info->delalloc_bytes) {
2042                 printk("btrfs: at unmount delalloc count %Lu\n",
2043                        fs_info->delalloc_bytes);
2044         }
2045         if (fs_info->total_ref_cache_size) {
2046                 printk("btrfs: at umount reference cache size %Lu\n",
2047                         fs_info->total_ref_cache_size);
2048         }
2049
2050         if (fs_info->extent_root->node)
2051                 free_extent_buffer(fs_info->extent_root->node);
2052
2053         if (fs_info->tree_root->node)
2054                 free_extent_buffer(fs_info->tree_root->node);
2055
2056         if (root->fs_info->chunk_root->node);
2057                 free_extent_buffer(root->fs_info->chunk_root->node);
2058
2059         if (root->fs_info->dev_root->node);
2060                 free_extent_buffer(root->fs_info->dev_root->node);
2061
2062         btrfs_free_block_groups(root->fs_info);
2063
2064         del_fs_roots(fs_info);
2065
2066         iput(fs_info->btree_inode);
2067
2068         btrfs_stop_workers(&fs_info->fixup_workers);
2069         btrfs_stop_workers(&fs_info->delalloc_workers);
2070         btrfs_stop_workers(&fs_info->workers);
2071         btrfs_stop_workers(&fs_info->endio_workers);
2072         btrfs_stop_workers(&fs_info->endio_write_workers);
2073         btrfs_stop_workers(&fs_info->submit_workers);
2074
2075 #if 0
2076         while(!list_empty(&fs_info->hashers)) {
2077                 struct btrfs_hasher *hasher;
2078                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
2079                                     hashers);
2080                 list_del(&hasher->hashers);
2081                 crypto_free_hash(&fs_info->hash_tfm);
2082                 kfree(hasher);
2083         }
2084 #endif
2085         btrfs_close_devices(fs_info->fs_devices);
2086         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2087
2088         bdi_destroy(&fs_info->bdi);
2089
2090         kfree(fs_info->extent_root);
2091         kfree(fs_info->tree_root);
2092         kfree(fs_info->chunk_root);
2093         kfree(fs_info->dev_root);
2094         return 0;
2095 }
2096
2097 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2098 {
2099         int ret;
2100         struct inode *btree_inode = buf->first_page->mapping->host;
2101
2102         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2103         if (!ret)
2104                 return ret;
2105
2106         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2107                                     parent_transid);
2108         return !ret;
2109 }
2110
2111 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2112 {
2113         struct inode *btree_inode = buf->first_page->mapping->host;
2114         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2115                                           buf);
2116 }
2117
2118 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2119 {
2120         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2121         u64 transid = btrfs_header_generation(buf);
2122         struct inode *btree_inode = root->fs_info->btree_inode;
2123
2124         WARN_ON(!btrfs_tree_locked(buf));
2125         if (transid != root->fs_info->generation) {
2126                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
2127                         (unsigned long long)buf->start,
2128                         transid, root->fs_info->generation);
2129                 WARN_ON(1);
2130         }
2131         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
2132 }
2133
2134 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2135 {
2136         /*
2137          * looks as though older kernels can get into trouble with
2138          * this code, they end up stuck in balance_dirty_pages forever
2139          */
2140         struct extent_io_tree *tree;
2141         u64 num_dirty;
2142         u64 start = 0;
2143         unsigned long thresh = 32 * 1024 * 1024;
2144         tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
2145
2146         if (current_is_pdflush() || current->flags & PF_MEMALLOC)
2147                 return;
2148
2149         num_dirty = count_range_bits(tree, &start, (u64)-1,
2150                                      thresh, EXTENT_DIRTY);
2151         if (num_dirty > thresh) {
2152                 balance_dirty_pages_ratelimited_nr(
2153                                    root->fs_info->btree_inode->i_mapping, 1);
2154         }
2155         return;
2156 }
2157
2158 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2159 {
2160         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2161         int ret;
2162         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2163         if (ret == 0) {
2164                 buf->flags |= EXTENT_UPTODATE;
2165         }
2166         return ret;
2167 }
2168
2169 int btree_lock_page_hook(struct page *page)
2170 {
2171         struct inode *inode = page->mapping->host;
2172         struct btrfs_root *root = BTRFS_I(inode)->root;
2173         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2174         struct extent_buffer *eb;
2175         unsigned long len;
2176         u64 bytenr = page_offset(page);
2177
2178         if (page->private == EXTENT_PAGE_PRIVATE)
2179                 goto out;
2180
2181         len = page->private >> 2;
2182         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2183         if (!eb)
2184                 goto out;
2185
2186         btrfs_tree_lock(eb);
2187         spin_lock(&root->fs_info->hash_lock);
2188         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2189         spin_unlock(&root->fs_info->hash_lock);
2190         btrfs_tree_unlock(eb);
2191         free_extent_buffer(eb);
2192 out:
2193         lock_page(page);
2194         return 0;
2195 }
2196
2197 static struct extent_io_ops btree_extent_io_ops = {
2198         .write_cache_pages_lock_hook = btree_lock_page_hook,
2199         .readpage_end_io_hook = btree_readpage_end_io_hook,
2200         .submit_bio_hook = btree_submit_bio_hook,
2201         /* note we're sharing with inode.c for the merge bio hook */
2202         .merge_bio_hook = btrfs_merge_bio_hook,
2203 };