fs/btrfs/disk-io.c

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