fs/btrfs/disk-io.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/fs.h>
  20 #include <linux/blkdev.h>
  21 #include <linux/scatterlist.h>
  22 #include <linux/swap.h>
  23 #include <linux/radix-tree.h>
  24 #include <linux/writeback.h>
  25 #include <linux/buffer_head.h>
  26 #include <linux/workqueue.h>
  27 #include <linux/kthread.h>
  28 #include <linux/freezer.h>
  29 #include "compat.h"
  30 #include "crc32c.h"
  31 #include "ctree.h"
  32 #include "disk-io.h"
  33 #include "transaction.h"
  34 #include "btrfs_inode.h"
  35 #include "volumes.h"
  36 #include "print-tree.h"
  37 #include "async-thread.h"
  38 #include "locking.h"
  39 #include "ref-cache.h"
  40 #include "tree-log.h"
  41 #include "free-space-cache.h"
  42
  43 static struct extent_io_ops btree_extent_io_ops;
  44 static void end_workqueue_fn(struct btrfs_work *work);
  45
  46 /*
  47  * end_io_wq structs are used to do processing in task context when an IO is
  48  * complete.  This is used during reads to verify checksums, and it is used
  49  * by writes to insert metadata for new file extents after IO is complete.
  50  */
  51 struct end_io_wq {
  52         struct bio *bio;
  53         bio_end_io_t *end_io;
  54         void *private;
  55         struct btrfs_fs_info *info;
  56         int error;
  57         int metadata;
  58         struct list_head list;
  59         struct btrfs_work work;
  60 };
  61
  62 /*
  63  * async submit bios are used to offload expensive checksumming
  64  * onto the worker threads.  They checksum file and metadata bios
  65  * just before they are sent down the IO stack.
  66  */
  67 struct async_submit_bio {
  68         struct inode *inode;
  69         struct bio *bio;
  70         struct list_head list;
  71         extent_submit_bio_hook_t *submit_bio_start;
  72         extent_submit_bio_hook_t *submit_bio_done;
  73         int rw;
  74         int mirror_num;
  75         unsigned long bio_flags;
  76         struct btrfs_work work;
  77 };
  78
  79 /* These are used to set the lockdep class on the extent buffer locks.
  80  * The class is set by the readpage_end_io_hook after the buffer has
  81  * passed csum validation but before the pages are unlocked.
  82  *
  83  * The lockdep class is also set by btrfs_init_new_buffer on freshly
  84  * allocated blocks.
  85  *
  86  * The class is based on the level in the tree block, which allows lockdep
  87  * to know that lower nodes nest inside the locks of higher nodes.
  88  *
  89  * We also add a check to make sure the highest level of the tree is
  90  * the same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this
  91  * code needs update as well.
  92  */
  93 #ifdef CONFIG_DEBUG_LOCK_ALLOC
  94 # if BTRFS_MAX_LEVEL != 8
  95 #  error
  96 # endif
  97 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
  98 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
  99         /* leaf */
 100         "btrfs-extent-00",
 101         "btrfs-extent-01",
 102         "btrfs-extent-02",
 103         "btrfs-extent-03",
 104         "btrfs-extent-04",
 105         "btrfs-extent-05",
 106         "btrfs-extent-06",
 107         "btrfs-extent-07",
 108         /* highest possible level */
 109         "btrfs-extent-08",
 110 };
 111 #endif
 112
 113 /*
 114  * extents on the btree inode are pretty simple, there's one extent
 115  * that covers the entire device
 116  */
 117 static struct extent_map *btree_get_extent(struct inode *inode,
 118                 struct page *page, size_t page_offset, u64 start, u64 len,
 119                 int create)
 120 {
 121         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 122         struct extent_map *em;
 123         int ret;
 124
 125         spin_lock(&em_tree->lock);
 126         em = lookup_extent_mapping(em_tree, start, len);
 127         if (em) {
 128                 em->bdev =
 129                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 130                 spin_unlock(&em_tree->lock);
 131                 goto out;
 132         }
 133         spin_unlock(&em_tree->lock);
 134
 135         em = alloc_extent_map(GFP_NOFS);
 136         if (!em) {
 137                 em = ERR_PTR(-ENOMEM);
 138                 goto out;
 139         }
 140         em->start = 0;
 141         em->len = (u64)-1;
 142         em->block_len = (u64)-1;
 143         em->block_start = 0;
 144         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 145
 146         spin_lock(&em_tree->lock);
 147         ret = add_extent_mapping(em_tree, em);
 148         if (ret == -EEXIST) {
 149                 u64 failed_start = em->start;
 150                 u64 failed_len = em->len;
 151
 152                 free_extent_map(em);
 153                 em = lookup_extent_mapping(em_tree, start, len);
 154                 if (em) {
 155                         ret = 0;
 156                 } else {
 157                         em = lookup_extent_mapping(em_tree, failed_start,
 158                                                    failed_len);
 159                         ret = -EIO;
 160                 }
 161         } else if (ret) {
 162                 free_extent_map(em);
 163                 em = NULL;
 164         }
 165         spin_unlock(&em_tree->lock);
 166
 167         if (ret)
 168                 em = ERR_PTR(ret);
 169 out:
 170         return em;
 171 }
 172
 173 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 174 {
 175         return btrfs_crc32c(seed, data, len);
 176 }
 177
 178 void btrfs_csum_final(u32 crc, char *result)
 179 {
 180         *(__le32 *)result = ~cpu_to_le32(crc);
 181 }
 182
 183 /*
 184  * compute the csum for a btree block, and either verify it or write it
 185  * into the csum field of the block.
 186  */
 187 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 188                            int verify)
 189 {
 190         u16 csum_size =
 191                 btrfs_super_csum_size(&root->fs_info->super_copy);
 192         char *result = NULL;
 193         unsigned long len;
 194         unsigned long cur_len;
 195         unsigned long offset = BTRFS_CSUM_SIZE;
 196         char *map_token = NULL;
 197         char *kaddr;
 198         unsigned long map_start;
 199         unsigned long map_len;
 200         int err;
 201         u32 crc = ~(u32)0;
 202         unsigned long inline_result;
 203
 204         len = buf->len - offset;
 205         while (len > 0) {
 206                 err = map_private_extent_buffer(buf, offset, 32,
 207                                         &map_token, &kaddr,
 208                                         &map_start, &map_len, KM_USER0);
 209                 if (err)
 210                         return 1;
 211                 cur_len = min(len, map_len - (offset - map_start));
 212                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 213                                       crc, cur_len);
 214                 len -= cur_len;
 215                 offset += cur_len;
 216                 unmap_extent_buffer(buf, map_token, KM_USER0);
 217         }
 218         if (csum_size > sizeof(inline_result)) {
 219                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
 220                 if (!result)
 221                         return 1;
 222         } else {
 223                 result = (char *)&inline_result;
 224         }
 225
 226         btrfs_csum_final(crc, result);
 227
 228         if (verify) {
 229                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 230                         u32 val;
 231                         u32 found = 0;
 232                         memcpy(&found, result, csum_size);
 233
 234                         read_extent_buffer(buf, &val, 0, csum_size);
 235                         printk(KERN_INFO "btrfs: %s checksum verify failed "
 236                                "on %llu wanted %X found %X level %d\n",
 237                                root->fs_info->sb->s_id,
 238                                buf->start, val, found, btrfs_header_level(buf));
 239                         if (result != (char *)&inline_result)
 240                                 kfree(result);
 241                         return 1;
 242                 }
 243         } else {
 244                 write_extent_buffer(buf, result, 0, csum_size);
 245         }
 246         if (result != (char *)&inline_result)
 247                 kfree(result);
 248         return 0;
 249 }
 250
 251 /*
 252  * we can't consider a given block up to date unless the transid of the
 253  * block matches the transid in the parent node's pointer.  This is how we
 254  * detect blocks that either didn't get written at all or got written
 255  * in the wrong place.
 256  */
 257 static int verify_parent_transid(struct extent_io_tree *io_tree,
 258                                  struct extent_buffer *eb, u64 parent_transid)
 259 {
 260         int ret;
 261
 262         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 263                 return 0;
 264
 265         lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
 266         if (extent_buffer_uptodate(io_tree, eb) &&
 267             btrfs_header_generation(eb) == parent_transid) {
 268                 ret = 0;
 269                 goto out;
 270         }
 271         printk("parent transid verify failed on %llu wanted %llu found %llu\n",
 272                (unsigned long long)eb->start,
 273                (unsigned long long)parent_transid,
 274                (unsigned long long)btrfs_header_generation(eb));
 275         ret = 1;
 276         clear_extent_buffer_uptodate(io_tree, eb);
 277 out:
 278         unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
 279                       GFP_NOFS);
 280         return ret;
 281 }
 282
 283 /*
 284  * helper to read a given tree block, doing retries as required when
 285  * the checksums don't match and we have alternate mirrors to try.
 286  */
 287 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 288                                           struct extent_buffer *eb,
 289                                           u64 start, u64 parent_transid)
 290 {
 291         struct extent_io_tree *io_tree;
 292         int ret;
 293         int num_copies = 0;
 294         int mirror_num = 0;
 295
 296         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 297         while (1) {
 298                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 299                                                btree_get_extent, mirror_num);
 300                 if (!ret &&
 301                     !verify_parent_transid(io_tree, eb, parent_transid))
 302                         return ret;
 303
 304                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 305                                               eb->start, eb->len);
 306                 if (num_copies == 1)
 307                         return ret;
 308
 309                 mirror_num++;
 310                 if (mirror_num > num_copies)
 311                         return ret;
 312         }
 313         return -EIO;
 314 }
 315
 316 /*
 317  * checksum a dirty tree block before IO.  This has extra checks to make sure
 318  * we only fill in the checksum field in the first page of a multi-page block
 319  */
 320
 321 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 322 {
 323         struct extent_io_tree *tree;
 324         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 325         u64 found_start;
 326         int found_level;
 327         unsigned long len;
 328         struct extent_buffer *eb;
 329         int ret;
 330
 331         tree = &BTRFS_I(page->mapping->host)->io_tree;
 332
 333         if (page->private == EXTENT_PAGE_PRIVATE)
 334                 goto out;
 335         if (!page->private)
 336                 goto out;
 337         len = page->private >> 2;
 338         WARN_ON(len == 0);
 339
 340         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 341         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 342                                              btrfs_header_generation(eb));
 343         BUG_ON(ret);
 344         found_start = btrfs_header_bytenr(eb);
 345         if (found_start != start) {
 346                 WARN_ON(1);
 347                 goto err;
 348         }
 349         if (eb->first_page != page) {
 350                 WARN_ON(1);
 351                 goto err;
 352         }
 353         if (!PageUptodate(page)) {
 354                 WARN_ON(1);
 355                 goto err;
 356         }
 357         found_level = btrfs_header_level(eb);
 358
 359         csum_tree_block(root, eb, 0);
 360 err:
 361         free_extent_buffer(eb);
 362 out:
 363         return 0;
 364 }
 365
 366 static int check_tree_block_fsid(struct btrfs_root *root,
 367                                  struct extent_buffer *eb)
 368 {
 369         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 370         u8 fsid[BTRFS_UUID_SIZE];
 371         int ret = 1;
 372
 373         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
 374                            BTRFS_FSID_SIZE);
 375         while (fs_devices) {
 376                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 377                         ret = 0;
 378                         break;
 379                 }
 380                 fs_devices = fs_devices->seed;
 381         }
 382         return ret;
 383 }
 384
 385 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 386 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
 387 {
 388         lockdep_set_class_and_name(&eb->lock,
 389                            &btrfs_eb_class[level],
 390                            btrfs_eb_name[level]);
 391 }
 392 #endif
 393
 394 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 395                                struct extent_state *state)
 396 {
 397         struct extent_io_tree *tree;
 398         u64 found_start;
 399         int found_level;
 400         unsigned long len;
 401         struct extent_buffer *eb;
 402         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 403         int ret = 0;
 404
 405         tree = &BTRFS_I(page->mapping->host)->io_tree;
 406         if (page->private == EXTENT_PAGE_PRIVATE)
 407                 goto out;
 408         if (!page->private)
 409                 goto out;
 410
 411         len = page->private >> 2;
 412         WARN_ON(len == 0);
 413
 414         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 415
 416         found_start = btrfs_header_bytenr(eb);
 417         if (found_start != start) {
 418                 printk(KERN_INFO "btrfs bad tree block start %llu %llu\n",
 419                        (unsigned long long)found_start,
 420                        (unsigned long long)eb->start);
 421                 ret = -EIO;
 422                 goto err;
 423         }
 424         if (eb->first_page != page) {
 425                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
 426                        eb->first_page->index, page->index);
 427                 WARN_ON(1);
 428                 ret = -EIO;
 429                 goto err;
 430         }
 431         if (check_tree_block_fsid(root, eb)) {
 432                 printk(KERN_INFO "btrfs bad fsid on block %llu\n",
 433                        (unsigned long long)eb->start);
 434                 ret = -EIO;
 435                 goto err;
 436         }
 437         found_level = btrfs_header_level(eb);
 438
 439         btrfs_set_buffer_lockdep_class(eb, found_level);
 440
 441         ret = csum_tree_block(root, eb, 1);
 442         if (ret)
 443                 ret = -EIO;
 444
 445         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 446         end = eb->start + end - 1;
 447 err:
 448         free_extent_buffer(eb);
 449 out:
 450         return ret;
 451 }
 452
 453 static void end_workqueue_bio(struct bio *bio, int err)
 454 {
 455         struct end_io_wq *end_io_wq = bio->bi_private;
 456         struct btrfs_fs_info *fs_info;
 457
 458         fs_info = end_io_wq->info;
 459         end_io_wq->error = err;
 460         end_io_wq->work.func = end_workqueue_fn;
 461         end_io_wq->work.flags = 0;
 462
 463         if (bio->bi_rw & (1 << BIO_RW)) {
 464                 if (end_io_wq->metadata)
 465                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
 466                                            &end_io_wq->work);
 467                 else
 468                         btrfs_queue_worker(&fs_info->endio_write_workers,
 469                                            &end_io_wq->work);
 470         } else {
 471                 if (end_io_wq->metadata)
 472                         btrfs_queue_worker(&fs_info->endio_meta_workers,
 473                                            &end_io_wq->work);
 474                 else
 475                         btrfs_queue_worker(&fs_info->endio_workers,
 476                                            &end_io_wq->work);
 477         }
 478 }
 479
 480 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 481                         int metadata)
 482 {
 483         struct end_io_wq *end_io_wq;
 484         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 485         if (!end_io_wq)
 486                 return -ENOMEM;
 487
 488         end_io_wq->private = bio->bi_private;
 489         end_io_wq->end_io = bio->bi_end_io;
 490         end_io_wq->info = info;
 491         end_io_wq->error = 0;
 492         end_io_wq->bio = bio;
 493         end_io_wq->metadata = metadata;
 494
 495         bio->bi_private = end_io_wq;
 496         bio->bi_end_io = end_workqueue_bio;
 497         return 0;
 498 }
 499
 500 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 501 {
 502         unsigned long limit = min_t(unsigned long,
 503                                     info->workers.max_workers,
 504                                     info->fs_devices->open_devices);
 505         return 256 * limit;
 506 }
 507
 508 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
 509 {
 510         return atomic_read(&info->nr_async_bios) >
 511                 btrfs_async_submit_limit(info);
 512 }
 513
 514 static void run_one_async_start(struct btrfs_work *work)
 515 {
 516         struct btrfs_fs_info *fs_info;
 517         struct async_submit_bio *async;
 518
 519         async = container_of(work, struct  async_submit_bio, work);
 520         fs_info = BTRFS_I(async->inode)->root->fs_info;
 521         async->submit_bio_start(async->inode, async->rw, async->bio,
 522                                async->mirror_num, async->bio_flags);
 523 }
 524
 525 static void run_one_async_done(struct btrfs_work *work)
 526 {
 527         struct btrfs_fs_info *fs_info;
 528         struct async_submit_bio *async;
 529         int limit;
 530
 531         async = container_of(work, struct  async_submit_bio, work);
 532         fs_info = BTRFS_I(async->inode)->root->fs_info;
 533
 534         limit = btrfs_async_submit_limit(fs_info);
 535         limit = limit * 2 / 3;
 536
 537         atomic_dec(&fs_info->nr_async_submits);
 538
 539         if (atomic_read(&fs_info->nr_async_submits) < limit &&
 540             waitqueue_active(&fs_info->async_submit_wait))
 541                 wake_up(&fs_info->async_submit_wait);
 542
 543         async->submit_bio_done(async->inode, async->rw, async->bio,
 544                                async->mirror_num, async->bio_flags);
 545 }
 546
 547 static void run_one_async_free(struct btrfs_work *work)
 548 {
 549         struct async_submit_bio *async;
 550
 551         async = container_of(work, struct  async_submit_bio, work);
 552         kfree(async);
 553 }
 554
 555 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 556                         int rw, struct bio *bio, int mirror_num,
 557                         unsigned long bio_flags,
 558                         extent_submit_bio_hook_t *submit_bio_start,
 559                         extent_submit_bio_hook_t *submit_bio_done)
 560 {
 561         struct async_submit_bio *async;
 562
 563         async = kmalloc(sizeof(*async), GFP_NOFS);
 564         if (!async)
 565                 return -ENOMEM;
 566
 567         async->inode = inode;
 568         async->rw = rw;
 569         async->bio = bio;
 570         async->mirror_num = mirror_num;
 571         async->submit_bio_start = submit_bio_start;
 572         async->submit_bio_done = submit_bio_done;
 573
 574         async->work.func = run_one_async_start;
 575         async->work.ordered_func = run_one_async_done;
 576         async->work.ordered_free = run_one_async_free;
 577
 578         async->work.flags = 0;
 579         async->bio_flags = bio_flags;
 580
 581         atomic_inc(&fs_info->nr_async_submits);
 582         btrfs_queue_worker(&fs_info->workers, &async->work);
 583 #if 0
 584         int limit = btrfs_async_submit_limit(fs_info);
 585         if (atomic_read(&fs_info->nr_async_submits) > limit) {
 586                 wait_event_timeout(fs_info->async_submit_wait,
 587                            (atomic_read(&fs_info->nr_async_submits) < limit),
 588                            HZ/10);
 589
 590                 wait_event_timeout(fs_info->async_submit_wait,
 591                            (atomic_read(&fs_info->nr_async_bios) < limit),
 592                            HZ/10);
 593         }
 594 #endif
 595         while (atomic_read(&fs_info->async_submit_draining) &&
 596               atomic_read(&fs_info->nr_async_submits)) {
 597                 wait_event(fs_info->async_submit_wait,
 598                            (atomic_read(&fs_info->nr_async_submits) == 0));
 599         }
 600
 601         return 0;
 602 }
 603
 604 static int btree_csum_one_bio(struct bio *bio)
 605 {
 606         struct bio_vec *bvec = bio->bi_io_vec;
 607         int bio_index = 0;
 608         struct btrfs_root *root;
 609
 610         WARN_ON(bio->bi_vcnt <= 0);
 611         while (bio_index < bio->bi_vcnt) {
 612                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 613                 csum_dirty_buffer(root, bvec->bv_page);
 614                 bio_index++;
 615                 bvec++;
 616         }
 617         return 0;
 618 }
 619
 620 static int __btree_submit_bio_start(struct inode *inode, int rw,
 621                                     struct bio *bio, int mirror_num,
 622                                     unsigned long bio_flags)
 623 {
 624         /*
 625          * when we're called for a write, we're already in the async
 626          * submission context.  Just jump into btrfs_map_bio
 627          */
 628         btree_csum_one_bio(bio);
 629         return 0;
 630 }
 631
 632 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 633                                  int mirror_num, unsigned long bio_flags)
 634 {
 635         /*
 636          * when we're called for a write, we're already in the async
 637          * submission context.  Just jump into btrfs_map_bio
 638          */
 639         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 640 }
 641
 642 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 643                                  int mirror_num, unsigned long bio_flags)
 644 {
 645         int ret;
 646
 647         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 648                                           bio, 1);
 649         BUG_ON(ret);
 650
 651         if (!(rw & (1 << BIO_RW))) {
 652                 /*
 653                  * called for a read, do the setup so that checksum validation
 654                  * can happen in the async kernel threads
 655                  */
 656                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 657                                      mirror_num, 0);
 658         }
 659         /*
 660          * kthread helpers are used to submit writes so that checksumming
 661          * can happen in parallel across all CPUs
 662          */
 663         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 664                                    inode, rw, bio, mirror_num, 0,
 665                                    __btree_submit_bio_start,
 666                                    __btree_submit_bio_done);
 667 }
 668
 669 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 670 {
 671         struct extent_io_tree *tree;
 672         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 673         struct extent_buffer *eb;
 674         int was_dirty;
 675
 676         tree = &BTRFS_I(page->mapping->host)->io_tree;
 677         if (!(current->flags & PF_MEMALLOC)) {
 678                 return extent_write_full_page(tree, page,
 679                                               btree_get_extent, wbc);
 680         }
 681
 682         redirty_page_for_writepage(wbc, page);
 683         eb = btrfs_find_tree_block(root, page_offset(page),
 684                                       PAGE_CACHE_SIZE);
 685         WARN_ON(!eb);
 686
 687         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
 688         if (!was_dirty) {
 689                 spin_lock(&root->fs_info->delalloc_lock);
 690                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
 691                 spin_unlock(&root->fs_info->delalloc_lock);
 692         }
 693         free_extent_buffer(eb);
 694
 695         unlock_page(page);
 696         return 0;
 697 }
 698
 699 static int btree_writepages(struct address_space *mapping,
 700                             struct writeback_control *wbc)
 701 {
 702         struct extent_io_tree *tree;
 703         tree = &BTRFS_I(mapping->host)->io_tree;
 704         if (wbc->sync_mode == WB_SYNC_NONE) {
 705                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
 706                 u64 num_dirty;
 707                 unsigned long thresh = 32 * 1024 * 1024;
 708
 709                 if (wbc->for_kupdate)
 710                         return 0;
 711
 712                 /* this is a bit racy, but that's ok */
 713                 num_dirty = root->fs_info->dirty_metadata_bytes;
 714                 if (num_dirty < thresh)
 715                         return 0;
 716         }
 717         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 718 }
 719
 720 static int btree_readpage(struct file *file, struct page *page)
 721 {
 722         struct extent_io_tree *tree;
 723         tree = &BTRFS_I(page->mapping->host)->io_tree;
 724         return extent_read_full_page(tree, page, btree_get_extent);
 725 }
 726
 727 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 728 {
 729         struct extent_io_tree *tree;
 730         struct extent_map_tree *map;
 731         int ret;
 732
 733         if (PageWriteback(page) || PageDirty(page))
 734                 return 0;
 735
 736         tree = &BTRFS_I(page->mapping->host)->io_tree;
 737         map = &BTRFS_I(page->mapping->host)->extent_tree;
 738
 739         ret = try_release_extent_state(map, tree, page, gfp_flags);
 740         if (!ret)
 741                 return 0;
 742
 743         ret = try_release_extent_buffer(tree, page);
 744         if (ret == 1) {
 745                 ClearPagePrivate(page);
 746                 set_page_private(page, 0);
 747                 page_cache_release(page);
 748         }
 749
 750         return ret;
 751 }
 752
 753 static void btree_invalidatepage(struct page *page, unsigned long offset)
 754 {
 755         struct extent_io_tree *tree;
 756         tree = &BTRFS_I(page->mapping->host)->io_tree;
 757         extent_invalidatepage(tree, page, offset);
 758         btree_releasepage(page, GFP_NOFS);
 759         if (PagePrivate(page)) {
 760                 printk(KERN_WARNING "btrfs warning page private not zero "
 761                        "on page %llu\n", (unsigned long long)page_offset(page));
 762                 ClearPagePrivate(page);
 763                 set_page_private(page, 0);
 764                 page_cache_release(page);
 765         }
 766 }
 767
 768 #if 0
 769 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 770 {
 771         struct buffer_head *bh;
 772         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 773         struct buffer_head *head;
 774         if (!page_has_buffers(page)) {
 775                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 776                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 777         }
 778         head = page_buffers(page);
 779         bh = head;
 780         do {
 781                 if (buffer_dirty(bh))
 782                         csum_tree_block(root, bh, 0);
 783                 bh = bh->b_this_page;
 784         } while (bh != head);
 785         return block_write_full_page(page, btree_get_block, wbc);
 786 }
 787 #endif
 788
 789 static struct address_space_operations btree_aops = {
 790         .readpage       = btree_readpage,
 791         .writepage      = btree_writepage,
 792         .writepages     = btree_writepages,
 793         .releasepage    = btree_releasepage,
 794         .invalidatepage = btree_invalidatepage,
 795         .sync_page      = block_sync_page,
 796 };
 797
 798 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 799                          u64 parent_transid)
 800 {
 801         struct extent_buffer *buf = NULL;
 802         struct inode *btree_inode = root->fs_info->btree_inode;
 803         int ret = 0;
 804
 805         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 806         if (!buf)
 807                 return 0;
 808         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 809                                  buf, 0, 0, btree_get_extent, 0);
 810         free_extent_buffer(buf);
 811         return ret;
 812 }
 813
 814 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 815                                             u64 bytenr, u32 blocksize)
 816 {
 817         struct inode *btree_inode = root->fs_info->btree_inode;
 818         struct extent_buffer *eb;
 819         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 820                                 bytenr, blocksize, GFP_NOFS);
 821         return eb;
 822 }
 823
 824 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 825                                                  u64 bytenr, u32 blocksize)
 826 {
 827         struct inode *btree_inode = root->fs_info->btree_inode;
 828         struct extent_buffer *eb;
 829
 830         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 831                                  bytenr, blocksize, NULL, GFP_NOFS);
 832         return eb;
 833 }
 834
 835
 836 int btrfs_write_tree_block(struct extent_buffer *buf)
 837 {
 838         return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
 839                                       buf->start + buf->len - 1, WB_SYNC_ALL);
 840 }
 841
 842 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 843 {
 844         return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
 845                                   buf->start, buf->start + buf->len - 1);
 846 }
 847
 848 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 849                                       u32 blocksize, u64 parent_transid)
 850 {
 851         struct extent_buffer *buf = NULL;
 852         struct inode *btree_inode = root->fs_info->btree_inode;
 853         struct extent_io_tree *io_tree;
 854         int ret;
 855
 856         io_tree = &BTRFS_I(btree_inode)->io_tree;
 857
 858         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 859         if (!buf)
 860                 return NULL;
 861
 862         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 863
 864         if (ret == 0)
 865                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
 866         else
 867                 WARN_ON(1);
 868         return buf;
 869
 870 }
 871
 872 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 873                      struct extent_buffer *buf)
 874 {
 875         struct inode *btree_inode = root->fs_info->btree_inode;
 876         if (btrfs_header_generation(buf) ==
 877             root->fs_info->running_transaction->transid) {
 878                 btrfs_assert_tree_locked(buf);
 879
 880                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
 881                         spin_lock(&root->fs_info->delalloc_lock);
 882                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
 883                                 root->fs_info->dirty_metadata_bytes -= buf->len;
 884                         else
 885                                 WARN_ON(1);
 886                         spin_unlock(&root->fs_info->delalloc_lock);
 887                 }
 888
 889                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
 890                 btrfs_set_lock_blocking(buf);
 891                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 892                                           buf);
 893         }
 894         return 0;
 895 }
 896
 897 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 898                         u32 stripesize, struct btrfs_root *root,
 899                         struct btrfs_fs_info *fs_info,
 900                         u64 objectid)
 901 {
 902         root->node = NULL;
 903         root->commit_root = NULL;
 904         root->ref_tree = NULL;
 905         root->sectorsize = sectorsize;
 906         root->nodesize = nodesize;
 907         root->leafsize = leafsize;
 908         root->stripesize = stripesize;
 909         root->ref_cows = 0;
 910         root->track_dirty = 0;
 911
 912         root->fs_info = fs_info;
 913         root->objectid = objectid;
 914         root->last_trans = 0;
 915         root->highest_inode = 0;
 916         root->last_inode_alloc = 0;
 917         root->name = NULL;
 918         root->in_sysfs = 0;
 919
 920         INIT_LIST_HEAD(&root->dirty_list);
 921         INIT_LIST_HEAD(&root->orphan_list);
 922         INIT_LIST_HEAD(&root->dead_list);
 923         spin_lock_init(&root->node_lock);
 924         spin_lock_init(&root->list_lock);
 925         mutex_init(&root->objectid_mutex);
 926         mutex_init(&root->log_mutex);
 927         init_waitqueue_head(&root->log_writer_wait);
 928         init_waitqueue_head(&root->log_commit_wait[0]);
 929         init_waitqueue_head(&root->log_commit_wait[1]);
 930         atomic_set(&root->log_commit[0], 0);
 931         atomic_set(&root->log_commit[1], 0);
 932         atomic_set(&root->log_writers, 0);
 933         root->log_batch = 0;
 934         root->log_transid = 0;
 935         extent_io_tree_init(&root->dirty_log_pages,
 936                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 937
 938         btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
 939         root->ref_tree = &root->ref_tree_struct;
 940
 941         memset(&root->root_key, 0, sizeof(root->root_key));
 942         memset(&root->root_item, 0, sizeof(root->root_item));
 943         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 944         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 945         root->defrag_trans_start = fs_info->generation;
 946         init_completion(&root->kobj_unregister);
 947         root->defrag_running = 0;
 948         root->defrag_level = 0;
 949         root->root_key.objectid = objectid;
 950         root->anon_super.s_root = NULL;
 951         root->anon_super.s_dev = 0;
 952         INIT_LIST_HEAD(&root->anon_super.s_list);
 953         INIT_LIST_HEAD(&root->anon_super.s_instances);
 954         init_rwsem(&root->anon_super.s_umount);
 955
 956         return 0;
 957 }
 958
 959 static int find_and_setup_root(struct btrfs_root *tree_root,
 960                                struct btrfs_fs_info *fs_info,
 961                                u64 objectid,
 962                                struct btrfs_root *root)
 963 {
 964         int ret;
 965         u32 blocksize;
 966         u64 generation;
 967
 968         __setup_root(tree_root->nodesize, tree_root->leafsize,
 969                      tree_root->sectorsize, tree_root->stripesize,
 970                      root, fs_info, objectid);
 971         ret = btrfs_find_last_root(tree_root, objectid,
 972                                    &root->root_item, &root->root_key);
 973         BUG_ON(ret);
 974
 975         generation = btrfs_root_generation(&root->root_item);
 976         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 977         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 978                                      blocksize, generation);
 979         BUG_ON(!root->node);
 980         return 0;
 981 }
 982
 983 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
 984                              struct btrfs_fs_info *fs_info)
 985 {
 986         struct extent_buffer *eb;
 987         struct btrfs_root *log_root_tree = fs_info->log_root_tree;
 988         u64 start = 0;
 989         u64 end = 0;
 990         int ret;
 991
 992         if (!log_root_tree)
 993                 return 0;
 994
 995         while (1) {
 996                 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
 997                                     0, &start, &end, EXTENT_DIRTY);
 998                 if (ret)
 999                         break;
1000
1001                 clear_extent_dirty(&log_root_tree->dirty_log_pages,
1002                                    start, end, GFP_NOFS);
1003         }
1004         eb = fs_info->log_root_tree->node;
1005
1006         WARN_ON(btrfs_header_level(eb) != 0);
1007         WARN_ON(btrfs_header_nritems(eb) != 0);
1008
1009         ret = btrfs_free_reserved_extent(fs_info->tree_root,
1010                                 eb->start, eb->len);
1011         BUG_ON(ret);
1012
1013         free_extent_buffer(eb);
1014         kfree(fs_info->log_root_tree);
1015         fs_info->log_root_tree = NULL;
1016         return 0;
1017 }
1018
1019 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1020                                          struct btrfs_fs_info *fs_info)
1021 {
1022         struct btrfs_root *root;
1023         struct btrfs_root *tree_root = fs_info->tree_root;
1024         struct extent_buffer *leaf;
1025
1026         root = kzalloc(sizeof(*root), GFP_NOFS);
1027         if (!root)
1028                 return ERR_PTR(-ENOMEM);
1029
1030         __setup_root(tree_root->nodesize, tree_root->leafsize,
1031                      tree_root->sectorsize, tree_root->stripesize,
1032                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1033
1034         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1035         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1036         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1037         /*
1038          * log trees do not get reference counted because they go away
1039          * before a real commit is actually done.  They do store pointers
1040          * to file data extents, and those reference counts still get
1041          * updated (along with back refs to the log tree).
1042          */
1043         root->ref_cows = 0;
1044
1045         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1046                                       0, BTRFS_TREE_LOG_OBJECTID,
1047                                       trans->transid, 0, 0, 0);
1048         if (IS_ERR(leaf)) {
1049                 kfree(root);
1050                 return ERR_CAST(leaf);
1051         }
1052
1053         root->node = leaf;
1054         btrfs_set_header_nritems(root->node, 0);
1055         btrfs_set_header_level(root->node, 0);
1056         btrfs_set_header_bytenr(root->node, root->node->start);
1057         btrfs_set_header_generation(root->node, trans->transid);
1058         btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
1059
1060         write_extent_buffer(root->node, root->fs_info->fsid,
1061                             (unsigned long)btrfs_header_fsid(root->node),
1062                             BTRFS_FSID_SIZE);
1063         btrfs_mark_buffer_dirty(root->node);
1064         btrfs_tree_unlock(root->node);
1065         return root;
1066 }
1067
1068 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1069                              struct btrfs_fs_info *fs_info)
1070 {
1071         struct btrfs_root *log_root;
1072
1073         log_root = alloc_log_tree(trans, fs_info);
1074         if (IS_ERR(log_root))
1075                 return PTR_ERR(log_root);
1076         WARN_ON(fs_info->log_root_tree);
1077         fs_info->log_root_tree = log_root;
1078         return 0;
1079 }
1080
1081 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1082                        struct btrfs_root *root)
1083 {
1084         struct btrfs_root *log_root;
1085         struct btrfs_inode_item *inode_item;
1086
1087         log_root = alloc_log_tree(trans, root->fs_info);
1088         if (IS_ERR(log_root))
1089                 return PTR_ERR(log_root);
1090
1091         log_root->last_trans = trans->transid;
1092         log_root->root_key.offset = root->root_key.objectid;
1093
1094         inode_item = &log_root->root_item.inode;
1095         inode_item->generation = cpu_to_le64(1);
1096         inode_item->size = cpu_to_le64(3);
1097         inode_item->nlink = cpu_to_le32(1);
1098         inode_item->nbytes = cpu_to_le64(root->leafsize);
1099         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1100
1101         btrfs_set_root_bytenr(&log_root->root_item, log_root->node->start);
1102         btrfs_set_root_generation(&log_root->root_item, trans->transid);
1103
1104         WARN_ON(root->log_root);
1105         root->log_root = log_root;
1106         root->log_transid = 0;
1107         return 0;
1108 }
1109
1110 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1111                                                struct btrfs_key *location)
1112 {
1113         struct btrfs_root *root;
1114         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1115         struct btrfs_path *path;
1116         struct extent_buffer *l;
1117         u64 highest_inode;
1118         u64 generation;
1119         u32 blocksize;
1120         int ret = 0;
1121
1122         root = kzalloc(sizeof(*root), GFP_NOFS);
1123         if (!root)
1124                 return ERR_PTR(-ENOMEM);
1125         if (location->offset == (u64)-1) {
1126                 ret = find_and_setup_root(tree_root, fs_info,
1127                                           location->objectid, root);
1128                 if (ret) {
1129                         kfree(root);
1130                         return ERR_PTR(ret);
1131                 }
1132                 goto insert;
1133         }
1134
1135         __setup_root(tree_root->nodesize, tree_root->leafsize,
1136                      tree_root->sectorsize, tree_root->stripesize,
1137                      root, fs_info, location->objectid);
1138
1139         path = btrfs_alloc_path();
1140         BUG_ON(!path);
1141         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1142         if (ret != 0) {
1143                 if (ret > 0)
1144                         ret = -ENOENT;
1145                 goto out;
1146         }
1147         l = path->nodes[0];
1148         read_extent_buffer(l, &root->root_item,
1149                btrfs_item_ptr_offset(l, path->slots[0]),
1150                sizeof(root->root_item));
1151         memcpy(&root->root_key, location, sizeof(*location));
1152         ret = 0;
1153 out:
1154         btrfs_release_path(root, path);
1155         btrfs_free_path(path);
1156         if (ret) {
1157                 kfree(root);
1158                 return ERR_PTR(ret);
1159         }
1160         generation = btrfs_root_generation(&root->root_item);
1161         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1162         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1163                                      blocksize, generation);
1164         BUG_ON(!root->node);
1165 insert:
1166         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1167                 root->ref_cows = 1;
1168                 ret = btrfs_find_highest_inode(root, &highest_inode);
1169                 if (ret == 0) {
1170                         root->highest_inode = highest_inode;
1171                         root->last_inode_alloc = highest_inode;
1172                 }
1173         }
1174         return root;
1175 }
1176
1177 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1178                                         u64 root_objectid)
1179 {
1180         struct btrfs_root *root;
1181
1182         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1183                 return fs_info->tree_root;
1184         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1185                 return fs_info->extent_root;
1186
1187         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1188                                  (unsigned long)root_objectid);
1189         return root;
1190 }
1191
1192 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1193                                               struct btrfs_key *location)
1194 {
1195         struct btrfs_root *root;
1196         int ret;
1197
1198         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1199                 return fs_info->tree_root;
1200         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1201                 return fs_info->extent_root;
1202         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1203                 return fs_info->chunk_root;
1204         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1205                 return fs_info->dev_root;
1206         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1207                 return fs_info->csum_root;
1208
1209         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1210                                  (unsigned long)location->objectid);
1211         if (root)
1212                 return root;
1213
1214         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1215         if (IS_ERR(root))
1216                 return root;
1217
1218         set_anon_super(&root->anon_super, NULL);
1219
1220         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1221                                 (unsigned long)root->root_key.objectid,
1222                                 root);
1223         if (ret) {
1224                 free_extent_buffer(root->node);
1225                 kfree(root);
1226                 return ERR_PTR(ret);
1227         }
1228         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
1229                 ret = btrfs_find_dead_roots(fs_info->tree_root,
1230                                             root->root_key.objectid, root);
1231                 BUG_ON(ret);
1232                 btrfs_orphan_cleanup(root);
1233         }
1234         return root;
1235 }
1236
1237 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1238                                       struct btrfs_key *location,
1239                                       const char *name, int namelen)
1240 {
1241         struct btrfs_root *root;
1242         int ret;
1243
1244         root = btrfs_read_fs_root_no_name(fs_info, location);
1245         if (!root)
1246                 return NULL;
1247
1248         if (root->in_sysfs)
1249                 return root;
1250
1251         ret = btrfs_set_root_name(root, name, namelen);
1252         if (ret) {
1253                 free_extent_buffer(root->node);
1254                 kfree(root);
1255                 return ERR_PTR(ret);
1256         }
1257 #if 0
1258         ret = btrfs_sysfs_add_root(root);
1259         if (ret) {
1260                 free_extent_buffer(root->node);
1261                 kfree(root->name);
1262                 kfree(root);
1263                 return ERR_PTR(ret);
1264         }
1265 #endif
1266         root->in_sysfs = 1;
1267         return root;
1268 }
1269
1270 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1271 {
1272         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1273         int ret = 0;
1274         struct btrfs_device *device;
1275         struct backing_dev_info *bdi;
1276 #if 0
1277         if ((bdi_bits & (1 << BDI_write_congested)) &&
1278             btrfs_congested_async(info, 0))
1279                 return 1;
1280 #endif
1281         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1282                 if (!device->bdev)
1283                         continue;
1284                 bdi = blk_get_backing_dev_info(device->bdev);
1285                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1286                         ret = 1;
1287                         break;
1288                 }
1289         }
1290         return ret;
1291 }
1292
1293 /*
1294  * this unplugs every device on the box, and it is only used when page
1295  * is null
1296  */
1297 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1298 {
1299         struct btrfs_device *device;
1300         struct btrfs_fs_info *info;
1301
1302         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1303         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1304                 if (!device->bdev)
1305                         continue;
1306
1307                 bdi = blk_get_backing_dev_info(device->bdev);
1308                 if (bdi->unplug_io_fn)
1309                         bdi->unplug_io_fn(bdi, page);
1310         }
1311 }
1312
1313 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1314 {
1315         struct inode *inode;
1316         struct extent_map_tree *em_tree;
1317         struct extent_map *em;
1318         struct address_space *mapping;
1319         u64 offset;
1320
1321         /* the generic O_DIRECT read code does this */
1322         if (1 || !page) {
1323                 __unplug_io_fn(bdi, page);
1324                 return;
1325         }
1326
1327         /*
1328          * page->mapping may change at any time.  Get a consistent copy
1329          * and use that for everything below
1330          */
1331         smp_mb();
1332         mapping = page->mapping;
1333         if (!mapping)
1334                 return;
1335
1336         inode = mapping->host;
1337
1338         /*
1339          * don't do the expensive searching for a small number of
1340          * devices
1341          */
1342         if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1343                 __unplug_io_fn(bdi, page);
1344                 return;
1345         }
1346
1347         offset = page_offset(page);
1348
1349         em_tree = &BTRFS_I(inode)->extent_tree;
1350         spin_lock(&em_tree->lock);
1351         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1352         spin_unlock(&em_tree->lock);
1353         if (!em) {
1354                 __unplug_io_fn(bdi, page);
1355                 return;
1356         }
1357
1358         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1359                 free_extent_map(em);
1360                 __unplug_io_fn(bdi, page);
1361                 return;
1362         }
1363         offset = offset - em->start;
1364         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1365                           em->block_start + offset, page);
1366         free_extent_map(em);
1367 }
1368
1369 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1370 {
1371         bdi_init(bdi);
1372         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1373         bdi->state              = 0;
1374         bdi->capabilities       = default_backing_dev_info.capabilities;
1375         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1376         bdi->unplug_io_data     = info;
1377         bdi->congested_fn       = btrfs_congested_fn;
1378         bdi->congested_data     = info;
1379         return 0;
1380 }
1381
1382 static int bio_ready_for_csum(struct bio *bio)
1383 {
1384         u64 length = 0;
1385         u64 buf_len = 0;
1386         u64 start = 0;
1387         struct page *page;
1388         struct extent_io_tree *io_tree = NULL;
1389         struct btrfs_fs_info *info = NULL;
1390         struct bio_vec *bvec;
1391         int i;
1392         int ret;
1393
1394         bio_for_each_segment(bvec, bio, i) {
1395                 page = bvec->bv_page;
1396                 if (page->private == EXTENT_PAGE_PRIVATE) {
1397                         length += bvec->bv_len;
1398                         continue;
1399                 }
1400                 if (!page->private) {
1401                         length += bvec->bv_len;
1402                         continue;
1403                 }
1404                 length = bvec->bv_len;
1405                 buf_len = page->private >> 2;
1406                 start = page_offset(page) + bvec->bv_offset;
1407                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1408                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1409         }
1410         /* are we fully contained in this bio? */
1411         if (buf_len <= length)
1412                 return 1;
1413
1414         ret = extent_range_uptodate(io_tree, start + length,
1415                                     start + buf_len - 1);
1416         return ret;
1417 }
1418
1419 /*
1420  * called by the kthread helper functions to finally call the bio end_io
1421  * functions.  This is where read checksum verification actually happens
1422  */
1423 static void end_workqueue_fn(struct btrfs_work *work)
1424 {
1425         struct bio *bio;
1426         struct end_io_wq *end_io_wq;
1427         struct btrfs_fs_info *fs_info;
1428         int error;
1429
1430         end_io_wq = container_of(work, struct end_io_wq, work);
1431         bio = end_io_wq->bio;
1432         fs_info = end_io_wq->info;
1433
1434         /* metadata bio reads are special because the whole tree block must
1435          * be checksummed at once.  This makes sure the entire block is in
1436          * ram and up to date before trying to verify things.  For
1437          * blocksize <= pagesize, it is basically a noop
1438          */
1439         if (!(bio->bi_rw & (1 << BIO_RW)) && end_io_wq->metadata &&
1440             !bio_ready_for_csum(bio)) {
1441                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1442                                    &end_io_wq->work);
1443                 return;
1444         }
1445         error = end_io_wq->error;
1446         bio->bi_private = end_io_wq->private;
1447         bio->bi_end_io = end_io_wq->end_io;
1448         kfree(end_io_wq);
1449         bio_endio(bio, error);
1450 }
1451
1452 static int cleaner_kthread(void *arg)
1453 {
1454         struct btrfs_root *root = arg;
1455
1456         do {
1457                 smp_mb();
1458                 if (root->fs_info->closing)
1459                         break;
1460
1461                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1462                 mutex_lock(&root->fs_info->cleaner_mutex);
1463                 btrfs_clean_old_snapshots(root);
1464                 mutex_unlock(&root->fs_info->cleaner_mutex);
1465
1466                 if (freezing(current)) {
1467                         refrigerator();
1468                 } else {
1469                         smp_mb();
1470                         if (root->fs_info->closing)
1471                                 break;
1472                         set_current_state(TASK_INTERRUPTIBLE);
1473                         schedule();
1474                         __set_current_state(TASK_RUNNING);
1475                 }
1476         } while (!kthread_should_stop());
1477         return 0;
1478 }
1479
1480 static int transaction_kthread(void *arg)
1481 {
1482         struct btrfs_root *root = arg;
1483         struct btrfs_trans_handle *trans;
1484         struct btrfs_transaction *cur;
1485         unsigned long now;
1486         unsigned long delay;
1487         int ret;
1488
1489         do {
1490                 smp_mb();
1491                 if (root->fs_info->closing)
1492                         break;
1493
1494                 delay = HZ * 30;
1495                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1496                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1497
1498                 mutex_lock(&root->fs_info->trans_mutex);
1499                 cur = root->fs_info->running_transaction;
1500                 if (!cur) {
1501                         mutex_unlock(&root->fs_info->trans_mutex);
1502                         goto sleep;
1503                 }
1504
1505                 now = get_seconds();
1506                 if (now < cur->start_time || now - cur->start_time < 30) {
1507                         mutex_unlock(&root->fs_info->trans_mutex);
1508                         delay = HZ * 5;
1509                         goto sleep;
1510                 }
1511                 mutex_unlock(&root->fs_info->trans_mutex);
1512                 trans = btrfs_start_transaction(root, 1);
1513                 ret = btrfs_commit_transaction(trans, root);
1514
1515 sleep:
1516                 wake_up_process(root->fs_info->cleaner_kthread);
1517                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1518
1519                 if (freezing(current)) {
1520                         refrigerator();
1521                 } else {
1522                         if (root->fs_info->closing)
1523                                 break;
1524                         set_current_state(TASK_INTERRUPTIBLE);
1525                         schedule_timeout(delay);
1526                         __set_current_state(TASK_RUNNING);
1527                 }
1528         } while (!kthread_should_stop());
1529         return 0;
1530 }
1531
1532 struct btrfs_root *open_ctree(struct super_block *sb,
1533                               struct btrfs_fs_devices *fs_devices,
1534                               char *options)
1535 {
1536         u32 sectorsize;
1537         u32 nodesize;
1538         u32 leafsize;
1539         u32 blocksize;
1540         u32 stripesize;
1541         u64 generation;
1542         u64 features;
1543         struct btrfs_key location;
1544         struct buffer_head *bh;
1545         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1546                                                  GFP_NOFS);
1547         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1548                                                  GFP_NOFS);
1549         struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1550                                                GFP_NOFS);
1551         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1552                                                 GFP_NOFS);
1553         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1554                                                 GFP_NOFS);
1555         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1556                                               GFP_NOFS);
1557         struct btrfs_root *log_tree_root;
1558
1559         int ret;
1560         int err = -EINVAL;
1561
1562         struct btrfs_super_block *disk_super;
1563
1564         if (!extent_root || !tree_root || !fs_info ||
1565             !chunk_root || !dev_root || !csum_root) {
1566                 err = -ENOMEM;
1567                 goto fail;
1568         }
1569         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1570         INIT_LIST_HEAD(&fs_info->trans_list);
1571         INIT_LIST_HEAD(&fs_info->dead_roots);
1572         INIT_LIST_HEAD(&fs_info->hashers);
1573         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1574         INIT_LIST_HEAD(&fs_info->ordered_operations);
1575         spin_lock_init(&fs_info->delalloc_lock);
1576         spin_lock_init(&fs_info->new_trans_lock);
1577         spin_lock_init(&fs_info->ref_cache_lock);
1578
1579         init_completion(&fs_info->kobj_unregister);
1580         fs_info->tree_root = tree_root;
1581         fs_info->extent_root = extent_root;
1582         fs_info->csum_root = csum_root;
1583         fs_info->chunk_root = chunk_root;
1584         fs_info->dev_root = dev_root;
1585         fs_info->fs_devices = fs_devices;
1586         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1587         INIT_LIST_HEAD(&fs_info->space_info);
1588         btrfs_mapping_init(&fs_info->mapping_tree);
1589         atomic_set(&fs_info->nr_async_submits, 0);
1590         atomic_set(&fs_info->async_delalloc_pages, 0);
1591         atomic_set(&fs_info->async_submit_draining, 0);
1592         atomic_set(&fs_info->nr_async_bios, 0);
1593         atomic_set(&fs_info->throttles, 0);
1594         atomic_set(&fs_info->throttle_gen, 0);
1595         fs_info->sb = sb;
1596         fs_info->max_extent = (u64)-1;
1597         fs_info->max_inline = 8192 * 1024;
1598         setup_bdi(fs_info, &fs_info->bdi);
1599         fs_info->btree_inode = new_inode(sb);
1600         fs_info->btree_inode->i_ino = 1;
1601         fs_info->btree_inode->i_nlink = 1;
1602
1603         fs_info->thread_pool_size = min_t(unsigned long,
1604                                           num_online_cpus() + 2, 8);
1605
1606         INIT_LIST_HEAD(&fs_info->ordered_extents);
1607         spin_lock_init(&fs_info->ordered_extent_lock);
1608
1609         sb->s_blocksize = 4096;
1610         sb->s_blocksize_bits = blksize_bits(4096);
1611
1612         /*
1613          * we set the i_size on the btree inode to the max possible int.
1614          * the real end of the address space is determined by all of
1615          * the devices in the system
1616          */
1617         fs_info->btree_inode->i_size = OFFSET_MAX;
1618         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1619         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1620
1621         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1622                              fs_info->btree_inode->i_mapping,
1623                              GFP_NOFS);
1624         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1625                              GFP_NOFS);
1626
1627         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1628
1629         spin_lock_init(&fs_info->block_group_cache_lock);
1630         fs_info->block_group_cache_tree.rb_node = NULL;
1631
1632         extent_io_tree_init(&fs_info->pinned_extents,
1633                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1634         fs_info->do_barriers = 1;
1635
1636         INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
1637         btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
1638         btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
1639
1640         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1641         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1642                sizeof(struct btrfs_key));
1643         insert_inode_hash(fs_info->btree_inode);
1644
1645         mutex_init(&fs_info->trans_mutex);
1646         mutex_init(&fs_info->ordered_operations_mutex);
1647         mutex_init(&fs_info->tree_log_mutex);
1648         mutex_init(&fs_info->drop_mutex);
1649         mutex_init(&fs_info->chunk_mutex);
1650         mutex_init(&fs_info->transaction_kthread_mutex);
1651         mutex_init(&fs_info->cleaner_mutex);
1652         mutex_init(&fs_info->volume_mutex);
1653         mutex_init(&fs_info->tree_reloc_mutex);
1654
1655         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1656         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1657
1658         init_waitqueue_head(&fs_info->transaction_throttle);
1659         init_waitqueue_head(&fs_info->transaction_wait);
1660         init_waitqueue_head(&fs_info->async_submit_wait);
1661
1662         __setup_root(4096, 4096, 4096, 4096, tree_root,
1663                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1664
1665
1666         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1667         if (!bh)
1668                 goto fail_iput;
1669
1670         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1671         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1672                sizeof(fs_info->super_for_commit));
1673         brelse(bh);
1674
1675         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1676
1677         disk_super = &fs_info->super_copy;
1678         if (!btrfs_super_root(disk_super))
1679                 goto fail_iput;
1680
1681         ret = btrfs_parse_options(tree_root, options);
1682         if (ret) {
1683                 err = ret;
1684                 goto fail_iput;
1685         }
1686
1687         features = btrfs_super_incompat_flags(disk_super) &
1688                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1689         if (features) {
1690                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1691                        "unsupported optional features (%Lx).\n",
1692                        features);
1693                 err = -EINVAL;
1694                 goto fail_iput;
1695         }
1696
1697         features = btrfs_super_compat_ro_flags(disk_super) &
1698                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1699         if (!(sb->s_flags & MS_RDONLY) && features) {
1700                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1701                        "unsupported option features (%Lx).\n",
1702                        features);
1703                 err = -EINVAL;
1704                 goto fail_iput;
1705         }
1706
1707         /*
1708          * we need to start all the end_io workers up front because the
1709          * queue work function gets called at interrupt time, and so it
1710          * cannot dynamically grow.
1711          */
1712         btrfs_init_workers(&fs_info->workers, "worker",
1713                            fs_info->thread_pool_size);
1714
1715         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1716                            fs_info->thread_pool_size);
1717
1718         btrfs_init_workers(&fs_info->submit_workers, "submit",
1719                            min_t(u64, fs_devices->num_devices,
1720                            fs_info->thread_pool_size));
1721
1722         /* a higher idle thresh on the submit workers makes it much more
1723          * likely that bios will be send down in a sane order to the
1724          * devices
1725          */
1726         fs_info->submit_workers.idle_thresh = 64;
1727
1728         fs_info->workers.idle_thresh = 16;
1729         fs_info->workers.ordered = 1;
1730
1731         fs_info->delalloc_workers.idle_thresh = 2;
1732         fs_info->delalloc_workers.ordered = 1;
1733
1734         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1735         btrfs_init_workers(&fs_info->endio_workers, "endio",
1736                            fs_info->thread_pool_size);
1737         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1738                            fs_info->thread_pool_size);
1739         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1740                            "endio-meta-write", fs_info->thread_pool_size);
1741         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1742                            fs_info->thread_pool_size);
1743
1744         /*
1745          * endios are largely parallel and should have a very
1746          * low idle thresh
1747          */
1748         fs_info->endio_workers.idle_thresh = 4;
1749         fs_info->endio_meta_workers.idle_thresh = 4;
1750
1751         fs_info->endio_write_workers.idle_thresh = 64;
1752         fs_info->endio_meta_write_workers.idle_thresh = 64;
1753
1754         btrfs_start_workers(&fs_info->workers, 1);
1755         btrfs_start_workers(&fs_info->submit_workers, 1);
1756         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1757         btrfs_start_workers(&fs_info->fixup_workers, 1);
1758         btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1759         btrfs_start_workers(&fs_info->endio_meta_workers,
1760                             fs_info->thread_pool_size);
1761         btrfs_start_workers(&fs_info->endio_meta_write_workers,
1762                             fs_info->thread_pool_size);
1763         btrfs_start_workers(&fs_info->endio_write_workers,
1764                             fs_info->thread_pool_size);
1765
1766         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1767         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1768                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1769
1770         nodesize = btrfs_super_nodesize(disk_super);
1771         leafsize = btrfs_super_leafsize(disk_super);
1772         sectorsize = btrfs_super_sectorsize(disk_super);
1773         stripesize = btrfs_super_stripesize(disk_super);
1774         tree_root->nodesize = nodesize;
1775         tree_root->leafsize = leafsize;
1776         tree_root->sectorsize = sectorsize;
1777         tree_root->stripesize = stripesize;
1778
1779         sb->s_blocksize = sectorsize;
1780         sb->s_blocksize_bits = blksize_bits(sectorsize);
1781
1782         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1783                     sizeof(disk_super->magic))) {
1784                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1785                 goto fail_sb_buffer;
1786         }
1787
1788         mutex_lock(&fs_info->chunk_mutex);
1789         ret = btrfs_read_sys_array(tree_root);
1790         mutex_unlock(&fs_info->chunk_mutex);
1791         if (ret) {
1792                 printk(KERN_WARNING "btrfs: failed to read the system "
1793                        "array on %s\n", sb->s_id);
1794                 goto fail_sys_array;
1795         }
1796
1797         blocksize = btrfs_level_size(tree_root,
1798                                      btrfs_super_chunk_root_level(disk_super));
1799         generation = btrfs_super_chunk_root_generation(disk_super);
1800
1801         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1802                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1803
1804         chunk_root->node = read_tree_block(chunk_root,
1805                                            btrfs_super_chunk_root(disk_super),
1806                                            blocksize, generation);
1807         BUG_ON(!chunk_root->node);
1808
1809         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1810            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1811            BTRFS_UUID_SIZE);
1812
1813         mutex_lock(&fs_info->chunk_mutex);
1814         ret = btrfs_read_chunk_tree(chunk_root);
1815         mutex_unlock(&fs_info->chunk_mutex);
1816         if (ret) {
1817                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1818                        sb->s_id);
1819                 goto fail_chunk_root;
1820         }
1821
1822         btrfs_close_extra_devices(fs_devices);
1823
1824         blocksize = btrfs_level_size(tree_root,
1825                                      btrfs_super_root_level(disk_super));
1826         generation = btrfs_super_generation(disk_super);
1827
1828         tree_root->node = read_tree_block(tree_root,
1829                                           btrfs_super_root(disk_super),
1830                                           blocksize, generation);
1831         if (!tree_root->node)
1832                 goto fail_chunk_root;
1833
1834
1835         ret = find_and_setup_root(tree_root, fs_info,
1836                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1837         if (ret)
1838                 goto fail_tree_root;
1839         extent_root->track_dirty = 1;
1840
1841         ret = find_and_setup_root(tree_root, fs_info,
1842                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1843         dev_root->track_dirty = 1;
1844         if (ret)
1845                 goto fail_extent_root;
1846
1847         ret = find_and_setup_root(tree_root, fs_info,
1848                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
1849         if (ret)
1850                 goto fail_extent_root;
1851
1852         csum_root->track_dirty = 1;
1853
1854         btrfs_read_block_groups(extent_root);
1855
1856         fs_info->generation = generation;
1857         fs_info->last_trans_committed = generation;
1858         fs_info->data_alloc_profile = (u64)-1;
1859         fs_info->metadata_alloc_profile = (u64)-1;
1860         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1861         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1862                                                "btrfs-cleaner");
1863         if (IS_ERR(fs_info->cleaner_kthread))
1864                 goto fail_csum_root;
1865
1866         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1867                                                    tree_root,
1868                                                    "btrfs-transaction");
1869         if (IS_ERR(fs_info->transaction_kthread))
1870                 goto fail_cleaner;
1871
1872         if (btrfs_super_log_root(disk_super) != 0) {
1873                 u64 bytenr = btrfs_super_log_root(disk_super);
1874
1875                 if (fs_devices->rw_devices == 0) {
1876                         printk(KERN_WARNING "Btrfs log replay required "
1877                                "on RO media\n");
1878                         err = -EIO;
1879                         goto fail_trans_kthread;
1880                 }
1881                 blocksize =
1882                      btrfs_level_size(tree_root,
1883                                       btrfs_super_log_root_level(disk_super));
1884
1885                 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1886                                                       GFP_NOFS);
1887
1888                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1889                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1890
1891                 log_tree_root->node = read_tree_block(tree_root, bytenr,
1892                                                       blocksize,
1893                                                       generation + 1);
1894                 ret = btrfs_recover_log_trees(log_tree_root);
1895                 BUG_ON(ret);
1896
1897                 if (sb->s_flags & MS_RDONLY) {
1898                         ret =  btrfs_commit_super(tree_root);
1899                         BUG_ON(ret);
1900                 }
1901         }
1902
1903         if (!(sb->s_flags & MS_RDONLY)) {
1904                 ret = btrfs_cleanup_reloc_trees(tree_root);
1905                 BUG_ON(ret);
1906         }
1907
1908         location.objectid = BTRFS_FS_TREE_OBJECTID;
1909         location.type = BTRFS_ROOT_ITEM_KEY;
1910         location.offset = (u64)-1;
1911
1912         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1913         if (!fs_info->fs_root)
1914                 goto fail_trans_kthread;
1915         return tree_root;
1916
1917 fail_trans_kthread:
1918         kthread_stop(fs_info->transaction_kthread);
1919 fail_cleaner:
1920         kthread_stop(fs_info->cleaner_kthread);
1921
1922         /*
1923          * make sure we're done with the btree inode before we stop our
1924          * kthreads
1925          */
1926         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1927         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1928
1929 fail_csum_root:
1930         free_extent_buffer(csum_root->node);
1931 fail_extent_root:
1932         free_extent_buffer(extent_root->node);
1933 fail_tree_root:
1934         free_extent_buffer(tree_root->node);
1935 fail_chunk_root:
1936         free_extent_buffer(chunk_root->node);
1937 fail_sys_array:
1938         free_extent_buffer(dev_root->node);
1939 fail_sb_buffer:
1940         btrfs_stop_workers(&fs_info->fixup_workers);
1941         btrfs_stop_workers(&fs_info->delalloc_workers);
1942         btrfs_stop_workers(&fs_info->workers);
1943         btrfs_stop_workers(&fs_info->endio_workers);
1944         btrfs_stop_workers(&fs_info->endio_meta_workers);
1945         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
1946         btrfs_stop_workers(&fs_info->endio_write_workers);
1947         btrfs_stop_workers(&fs_info->submit_workers);
1948 fail_iput:
1949         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1950         iput(fs_info->btree_inode);
1951
1952         btrfs_close_devices(fs_info->fs_devices);
1953         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1954         bdi_destroy(&fs_info->bdi);
1955
1956 fail:
1957         kfree(extent_root);
1958         kfree(tree_root);
1959         kfree(fs_info);
1960         kfree(chunk_root);
1961         kfree(dev_root);
1962         kfree(csum_root);
1963         return ERR_PTR(err);
1964 }
1965
1966 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1967 {
1968         char b[BDEVNAME_SIZE];
1969
1970         if (uptodate) {
1971                 set_buffer_uptodate(bh);
1972         } else {
1973                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1974                         printk(KERN_WARNING "lost page write due to "
1975                                         "I/O error on %s\n",
1976                                        bdevname(bh->b_bdev, b));
1977                 }
1978                 /* note, we dont' set_buffer_write_io_error because we have
1979                  * our own ways of dealing with the IO errors
1980                  */
1981                 clear_buffer_uptodate(bh);
1982         }
1983         unlock_buffer(bh);
1984         put_bh(bh);
1985 }
1986
1987 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
1988 {
1989         struct buffer_head *bh;
1990         struct buffer_head *latest = NULL;
1991         struct btrfs_super_block *super;
1992         int i;
1993         u64 transid = 0;
1994         u64 bytenr;
1995
1996         /* we would like to check all the supers, but that would make
1997          * a btrfs mount succeed after a mkfs from a different FS.
1998          * So, we need to add a special mount option to scan for
1999          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2000          */
2001         for (i = 0; i < 1; i++) {
2002                 bytenr = btrfs_sb_offset(i);
2003                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2004                         break;
2005                 bh = __bread(bdev, bytenr / 4096, 4096);
2006                 if (!bh)
2007                         continue;
2008
2009                 super = (struct btrfs_super_block *)bh->b_data;
2010                 if (btrfs_super_bytenr(super) != bytenr ||
2011                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2012                             sizeof(super->magic))) {
2013                         brelse(bh);
2014                         continue;
2015                 }
2016
2017                 if (!latest || btrfs_super_generation(super) > transid) {
2018                         brelse(latest);
2019                         latest = bh;
2020                         transid = btrfs_super_generation(super);
2021                 } else {
2022                         brelse(bh);
2023                 }
2024         }
2025         return latest;
2026 }
2027
2028 static int write_dev_supers(struct btrfs_device *device,
2029                             struct btrfs_super_block *sb,
2030                             int do_barriers, int wait, int max_mirrors)
2031 {
2032         struct buffer_head *bh;
2033         int i;
2034         int ret;
2035         int errors = 0;
2036         u32 crc;
2037         u64 bytenr;
2038         int last_barrier = 0;
2039
2040         if (max_mirrors == 0)
2041                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2042
2043         /* make sure only the last submit_bh does a barrier */
2044         if (do_barriers) {
2045                 for (i = 0; i < max_mirrors; i++) {
2046                         bytenr = btrfs_sb_offset(i);
2047                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2048                             device->total_bytes)
2049                                 break;
2050                         last_barrier = i;
2051                 }
2052         }
2053
2054         for (i = 0; i < max_mirrors; i++) {
2055                 bytenr = btrfs_sb_offset(i);
2056                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2057                         break;
2058
2059                 if (wait) {
2060                         bh = __find_get_block(device->bdev, bytenr / 4096,
2061                                               BTRFS_SUPER_INFO_SIZE);
2062                         BUG_ON(!bh);
2063                         brelse(bh);
2064                         wait_on_buffer(bh);
2065                         if (buffer_uptodate(bh)) {
2066                                 brelse(bh);
2067                                 continue;
2068                         }
2069                 } else {
2070                         btrfs_set_super_bytenr(sb, bytenr);
2071
2072                         crc = ~(u32)0;
2073                         crc = btrfs_csum_data(NULL, (char *)sb +
2074                                               BTRFS_CSUM_SIZE, crc,
2075                                               BTRFS_SUPER_INFO_SIZE -
2076                                               BTRFS_CSUM_SIZE);
2077                         btrfs_csum_final(crc, sb->csum);
2078
2079                         bh = __getblk(device->bdev, bytenr / 4096,
2080                                       BTRFS_SUPER_INFO_SIZE);
2081                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2082
2083                         set_buffer_uptodate(bh);
2084                         get_bh(bh);
2085                         lock_buffer(bh);
2086                         bh->b_end_io = btrfs_end_buffer_write_sync;
2087                 }
2088
2089                 if (i == last_barrier && do_barriers && device->barriers) {
2090                         ret = submit_bh(WRITE_BARRIER, bh);
2091                         if (ret == -EOPNOTSUPP) {
2092                                 printk("btrfs: disabling barriers on dev %s\n",
2093                                        device->name);
2094                                 set_buffer_uptodate(bh);
2095                                 device->barriers = 0;
2096                                 get_bh(bh);
2097                                 lock_buffer(bh);
2098                                 ret = submit_bh(WRITE, bh);
2099                         }
2100                 } else {
2101                         ret = submit_bh(WRITE, bh);
2102                 }
2103
2104                 if (!ret && wait) {
2105                         wait_on_buffer(bh);
2106                         if (!buffer_uptodate(bh))
2107                                 errors++;
2108                 } else if (ret) {
2109                         errors++;
2110                 }
2111                 if (wait)
2112                         brelse(bh);
2113         }
2114         return errors < i ? 0 : -1;
2115 }
2116
2117 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2118 {
2119         struct list_head *head = &root->fs_info->fs_devices->devices;
2120         struct btrfs_device *dev;
2121         struct btrfs_super_block *sb;
2122         struct btrfs_dev_item *dev_item;
2123         int ret;
2124         int do_barriers;
2125         int max_errors;
2126         int total_errors = 0;
2127         u64 flags;
2128
2129         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2130         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2131
2132         sb = &root->fs_info->super_for_commit;
2133         dev_item = &sb->dev_item;
2134         list_for_each_entry(dev, head, dev_list) {
2135                 if (!dev->bdev) {
2136                         total_errors++;
2137                         continue;
2138                 }
2139                 if (!dev->in_fs_metadata || !dev->writeable)
2140                         continue;
2141
2142                 btrfs_set_stack_device_generation(dev_item, 0);
2143                 btrfs_set_stack_device_type(dev_item, dev->type);
2144                 btrfs_set_stack_device_id(dev_item, dev->devid);
2145                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2146                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2147                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2148                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2149                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2150                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2151                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2152
2153                 flags = btrfs_super_flags(sb);
2154                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2155
2156                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2157                 if (ret)
2158                         total_errors++;
2159         }
2160         if (total_errors > max_errors) {
2161                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2162                        total_errors);
2163                 BUG();
2164         }
2165
2166         total_errors = 0;
2167         list_for_each_entry(dev, head, dev_list) {
2168                 if (!dev->bdev)
2169                         continue;
2170                 if (!dev->in_fs_metadata || !dev->writeable)
2171                         continue;
2172
2173                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2174                 if (ret)
2175                         total_errors++;
2176         }
2177         if (total_errors > max_errors) {
2178                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2179                        total_errors);
2180                 BUG();
2181         }
2182         return 0;
2183 }
2184
2185 int write_ctree_super(struct btrfs_trans_handle *trans,
2186                       struct btrfs_root *root, int max_mirrors)
2187 {
2188         int ret;
2189
2190         ret = write_all_supers(root, max_mirrors);
2191         return ret;
2192 }
2193
2194 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2195 {
2196         radix_tree_delete(&fs_info->fs_roots_radix,
2197                           (unsigned long)root->root_key.objectid);
2198         if (root->anon_super.s_dev) {
2199                 down_write(&root->anon_super.s_umount);
2200                 kill_anon_super(&root->anon_super);
2201         }
2202         if (root->node)
2203                 free_extent_buffer(root->node);
2204         if (root->commit_root)
2205                 free_extent_buffer(root->commit_root);
2206         kfree(root->name);
2207         kfree(root);
2208         return 0;
2209 }
2210
2211 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2212 {
2213         int ret;
2214         struct btrfs_root *gang[8];
2215         int i;
2216
2217         while (1) {
2218                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2219                                              (void **)gang, 0,
2220                                              ARRAY_SIZE(gang));
2221                 if (!ret)
2222                         break;
2223                 for (i = 0; i < ret; i++)
2224                         btrfs_free_fs_root(fs_info, gang[i]);
2225         }
2226         return 0;
2227 }
2228
2229 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2230 {
2231         u64 root_objectid = 0;
2232         struct btrfs_root *gang[8];
2233         int i;
2234         int ret;
2235
2236         while (1) {
2237                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2238                                              (void **)gang, root_objectid,
2239                                              ARRAY_SIZE(gang));
2240                 if (!ret)
2241                         break;
2242                 for (i = 0; i < ret; i++) {
2243                         root_objectid = gang[i]->root_key.objectid;
2244                         ret = btrfs_find_dead_roots(fs_info->tree_root,
2245                                                     root_objectid, gang[i]);
2246                         BUG_ON(ret);
2247                         btrfs_orphan_cleanup(gang[i]);
2248                 }
2249                 root_objectid++;
2250         }
2251         return 0;
2252 }
2253
2254 int btrfs_commit_super(struct btrfs_root *root)
2255 {
2256         struct btrfs_trans_handle *trans;
2257         int ret;
2258
2259         mutex_lock(&root->fs_info->cleaner_mutex);
2260         btrfs_clean_old_snapshots(root);
2261         mutex_unlock(&root->fs_info->cleaner_mutex);
2262         trans = btrfs_start_transaction(root, 1);
2263         ret = btrfs_commit_transaction(trans, root);
2264         BUG_ON(ret);
2265         /* run commit again to drop the original snapshot */
2266         trans = btrfs_start_transaction(root, 1);
2267         btrfs_commit_transaction(trans, root);
2268         ret = btrfs_write_and_wait_transaction(NULL, root);
2269         BUG_ON(ret);
2270
2271         ret = write_ctree_super(NULL, root, 0);
2272         return ret;
2273 }
2274
2275 int close_ctree(struct btrfs_root *root)
2276 {
2277         struct btrfs_fs_info *fs_info = root->fs_info;
2278         int ret;
2279
2280         fs_info->closing = 1;
2281         smp_mb();
2282
2283         kthread_stop(root->fs_info->transaction_kthread);
2284         kthread_stop(root->fs_info->cleaner_kthread);
2285
2286         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2287                 ret =  btrfs_commit_super(root);
2288                 if (ret)
2289                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2290         }
2291
2292         if (fs_info->delalloc_bytes) {
2293                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2294                        fs_info->delalloc_bytes);
2295         }
2296         if (fs_info->total_ref_cache_size) {
2297                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2298                        (unsigned long long)fs_info->total_ref_cache_size);
2299         }
2300
2301         if (fs_info->extent_root->node)
2302                 free_extent_buffer(fs_info->extent_root->node);
2303
2304         if (fs_info->tree_root->node)
2305                 free_extent_buffer(fs_info->tree_root->node);
2306
2307         if (root->fs_info->chunk_root->node)
2308                 free_extent_buffer(root->fs_info->chunk_root->node);
2309
2310         if (root->fs_info->dev_root->node)
2311                 free_extent_buffer(root->fs_info->dev_root->node);
2312
2313         if (root->fs_info->csum_root->node)
2314                 free_extent_buffer(root->fs_info->csum_root->node);
2315
2316         btrfs_free_block_groups(root->fs_info);
2317
2318         del_fs_roots(fs_info);
2319
2320         iput(fs_info->btree_inode);
2321
2322         btrfs_stop_workers(&fs_info->fixup_workers);
2323         btrfs_stop_workers(&fs_info->delalloc_workers);
2324         btrfs_stop_workers(&fs_info->workers);
2325         btrfs_stop_workers(&fs_info->endio_workers);
2326         btrfs_stop_workers(&fs_info->endio_meta_workers);
2327         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2328         btrfs_stop_workers(&fs_info->endio_write_workers);
2329         btrfs_stop_workers(&fs_info->submit_workers);
2330
2331 #if 0
2332         while (!list_empty(&fs_info->hashers)) {
2333                 struct btrfs_hasher *hasher;
2334                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
2335                                     hashers);
2336                 list_del(&hasher->hashers);
2337                 crypto_free_hash(&fs_info->hash_tfm);
2338                 kfree(hasher);
2339         }
2340 #endif
2341         btrfs_close_devices(fs_info->fs_devices);
2342         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2343
2344         bdi_destroy(&fs_info->bdi);
2345
2346         kfree(fs_info->extent_root);
2347         kfree(fs_info->tree_root);
2348         kfree(fs_info->chunk_root);
2349         kfree(fs_info->dev_root);
2350         kfree(fs_info->csum_root);
2351         return 0;
2352 }
2353
2354 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2355 {
2356         int ret;
2357         struct inode *btree_inode = buf->first_page->mapping->host;
2358
2359         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2360         if (!ret)
2361                 return ret;
2362
2363         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2364                                     parent_transid);
2365         return !ret;
2366 }
2367
2368 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2369 {
2370         struct inode *btree_inode = buf->first_page->mapping->host;
2371         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2372                                           buf);
2373 }
2374
2375 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2376 {
2377         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2378         u64 transid = btrfs_header_generation(buf);
2379         struct inode *btree_inode = root->fs_info->btree_inode;
2380         int was_dirty;
2381
2382         btrfs_assert_tree_locked(buf);
2383         if (transid != root->fs_info->generation) {
2384                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2385                        "found %llu running %llu\n",
2386                         (unsigned long long)buf->start,
2387                         (unsigned long long)transid,
2388                         (unsigned long long)root->fs_info->generation);
2389                 WARN_ON(1);
2390         }
2391         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2392                                             buf);
2393         if (!was_dirty) {
2394                 spin_lock(&root->fs_info->delalloc_lock);
2395                 root->fs_info->dirty_metadata_bytes += buf->len;
2396                 spin_unlock(&root->fs_info->delalloc_lock);
2397         }
2398 }
2399
2400 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2401 {
2402         /*
2403          * looks as though older kernels can get into trouble with
2404          * this code, they end up stuck in balance_dirty_pages forever
2405          */
2406         struct extent_io_tree *tree;
2407         u64 num_dirty;
2408         u64 start = 0;
2409         unsigned long thresh = 32 * 1024 * 1024;
2410         tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
2411
2412         if (current->flags & PF_MEMALLOC)
2413                 return;
2414
2415         num_dirty = count_range_bits(tree, &start, (u64)-1,
2416                                      thresh, EXTENT_DIRTY);
2417         if (num_dirty > thresh) {
2418                 balance_dirty_pages_ratelimited_nr(
2419                                    root->fs_info->btree_inode->i_mapping, 1);
2420         }
2421         return;
2422 }
2423
2424 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2425 {
2426         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2427         int ret;
2428         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2429         if (ret == 0)
2430                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2431         return ret;
2432 }
2433
2434 int btree_lock_page_hook(struct page *page)
2435 {
2436         struct inode *inode = page->mapping->host;
2437         struct btrfs_root *root = BTRFS_I(inode)->root;
2438         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2439         struct extent_buffer *eb;
2440         unsigned long len;
2441         u64 bytenr = page_offset(page);
2442
2443         if (page->private == EXTENT_PAGE_PRIVATE)
2444                 goto out;
2445
2446         len = page->private >> 2;
2447         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2448         if (!eb)
2449                 goto out;
2450
2451         btrfs_tree_lock(eb);
2452         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2453
2454         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2455                 spin_lock(&root->fs_info->delalloc_lock);
2456                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2457                         root->fs_info->dirty_metadata_bytes -= eb->len;
2458                 else
2459                         WARN_ON(1);
2460                 spin_unlock(&root->fs_info->delalloc_lock);
2461         }
2462
2463         btrfs_tree_unlock(eb);
2464         free_extent_buffer(eb);
2465 out:
2466         lock_page(page);
2467         return 0;
2468 }
2469
2470 static struct extent_io_ops btree_extent_io_ops = {
2471         .write_cache_pages_lock_hook = btree_lock_page_hook,
2472         .readpage_end_io_hook = btree_readpage_end_io_hook,
2473         .submit_bio_hook = btree_submit_bio_hook,
2474         /* note we're sharing with inode.c for the merge bio hook */
2475         .merge_bio_hook = btrfs_merge_bio_hook,
2476 };