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> // for block_sync_page
  26 #include <linux/workqueue.h>
  27 #include "crc32c.h"
  28 #include "ctree.h"
  29 #include "disk-io.h"
  30 #include "transaction.h"
  31 #include "btrfs_inode.h"
  32 #include "volumes.h"
  33 #include "print-tree.h"
  34
  35 #if 0
  36 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
  37 {
  38         if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
  39                 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
  40                        (unsigned long long)extent_buffer_blocknr(buf),
  41                        (unsigned long long)btrfs_header_blocknr(buf));
  42                 return 1;
  43         }
  44         return 0;
  45 }
  46 #endif
  47
  48 static struct extent_io_ops btree_extent_io_ops;
  49 static struct workqueue_struct *end_io_workqueue;
  50 static struct workqueue_struct *async_submit_workqueue;
  51
  52 struct end_io_wq {
  53         struct bio *bio;
  54         bio_end_io_t *end_io;
  55         void *private;
  56         struct btrfs_fs_info *info;
  57         int error;
  58         int metadata;
  59         struct list_head list;
  60 };
  61
  62 struct async_submit_bio {
  63         struct inode *inode;
  64         struct bio *bio;
  65         struct list_head list;
  66         extent_submit_bio_hook_t *submit_bio_hook;
  67         int rw;
  68         int mirror_num;
  69 };
  70
  71 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
  72                                     size_t page_offset, u64 start, u64 len,
  73                                     int create)
  74 {
  75         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
  76         struct extent_map *em;
  77         int ret;
  78
  79 again:
  80         spin_lock(&em_tree->lock);
  81         em = lookup_extent_mapping(em_tree, start, len);
  82         spin_unlock(&em_tree->lock);
  83         if (em) {
  84                 goto out;
  85         }
  86         em = alloc_extent_map(GFP_NOFS);
  87         if (!em) {
  88                 em = ERR_PTR(-ENOMEM);
  89                 goto out;
  90         }
  91         em->start = 0;
  92         em->len = i_size_read(inode);
  93         em->block_start = 0;
  94         em->bdev = inode->i_sb->s_bdev;
  95
  96         spin_lock(&em_tree->lock);
  97         ret = add_extent_mapping(em_tree, em);
  98         spin_unlock(&em_tree->lock);
  99
 100         if (ret == -EEXIST) {
 101                 free_extent_map(em);
 102                 em = NULL;
 103                 goto again;
 104         } else if (ret) {
 105                 em = ERR_PTR(ret);
 106         }
 107 out:
 108         return em;
 109 }
 110
 111 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 112 {
 113         return btrfs_crc32c(seed, data, len);
 114 }
 115
 116 void btrfs_csum_final(u32 crc, char *result)
 117 {
 118         *(__le32 *)result = ~cpu_to_le32(crc);
 119 }
 120
 121 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 122                            int verify)
 123 {
 124         char result[BTRFS_CRC32_SIZE];
 125         unsigned long len;
 126         unsigned long cur_len;
 127         unsigned long offset = BTRFS_CSUM_SIZE;
 128         char *map_token = NULL;
 129         char *kaddr;
 130         unsigned long map_start;
 131         unsigned long map_len;
 132         int err;
 133         u32 crc = ~(u32)0;
 134
 135         len = buf->len - offset;
 136         while(len > 0) {
 137                 err = map_private_extent_buffer(buf, offset, 32,
 138                                         &map_token, &kaddr,
 139                                         &map_start, &map_len, KM_USER0);
 140                 if (err) {
 141                         printk("failed to map extent buffer! %lu\n",
 142                                offset);
 143                         return 1;
 144                 }
 145                 cur_len = min(len, map_len - (offset - map_start));
 146                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 147                                       crc, cur_len);
 148                 len -= cur_len;
 149                 offset += cur_len;
 150                 unmap_extent_buffer(buf, map_token, KM_USER0);
 151         }
 152         btrfs_csum_final(crc, result);
 153
 154         if (verify) {
 155                 int from_this_trans = 0;
 156
 157                 if (root->fs_info->running_transaction &&
 158                     btrfs_header_generation(buf) ==
 159                     root->fs_info->running_transaction->transid)
 160                         from_this_trans = 1;
 161
 162                 /* FIXME, this is not good */
 163                 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
 164                         u32 val;
 165                         u32 found = 0;
 166                         memcpy(&found, result, BTRFS_CRC32_SIZE);
 167
 168                         read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
 169                         printk("btrfs: %s checksum verify failed on %llu "
 170                                "wanted %X found %X from_this_trans %d "
 171                                "level %d\n",
 172                                root->fs_info->sb->s_id,
 173                                buf->start, val, found, from_this_trans,
 174                                btrfs_header_level(buf));
 175                         return 1;
 176                 }
 177         } else {
 178                 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
 179         }
 180         return 0;
 181 }
 182
 183 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 184                                           struct extent_buffer *eb,
 185                                           u64 start)
 186 {
 187         struct extent_io_tree *io_tree;
 188         int ret;
 189         int num_copies = 0;
 190         int mirror_num = 0;
 191
 192         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 193         while (1) {
 194                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 195                                                btree_get_extent, mirror_num);
 196                 if (!ret) {
 197                         if (mirror_num)
 198 printk("good read %Lu mirror %d total %d\n", eb->start, mirror_num, num_copies);
 199                         return ret;
 200                 }
 201                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 202                                               eb->start, eb->len);
 203 printk("failed to read %Lu mirror %d total %d\n", eb->start, mirror_num, num_copies);
 204                 if (num_copies == 1) {
 205 printk("reading %Lu failed only one copy\n", eb->start);
 206                         return ret;
 207                 }
 208                 mirror_num++;
 209                 if (mirror_num > num_copies) {
 210 printk("bailing at mirror %d of %d\n", mirror_num, num_copies);
 211                         return ret;
 212                 }
 213         }
 214 printk("read extent buffer page last\n");
 215         return -EIO;
 216 }
 217
 218 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 219 {
 220         struct extent_io_tree *tree;
 221         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 222         u64 found_start;
 223         int found_level;
 224         unsigned long len;
 225         struct extent_buffer *eb;
 226         int ret;
 227
 228         tree = &BTRFS_I(page->mapping->host)->io_tree;
 229
 230         if (page->private == EXTENT_PAGE_PRIVATE)
 231                 goto out;
 232         if (!page->private)
 233                 goto out;
 234         len = page->private >> 2;
 235         if (len == 0) {
 236                 WARN_ON(1);
 237         }
 238         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 239         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE);
 240         BUG_ON(ret);
 241         btrfs_clear_buffer_defrag(eb);
 242         found_start = btrfs_header_bytenr(eb);
 243         if (found_start != start) {
 244                 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
 245                        start, found_start, len);
 246                 WARN_ON(1);
 247                 goto err;
 248         }
 249         if (eb->first_page != page) {
 250                 printk("bad first page %lu %lu\n", eb->first_page->index,
 251                        page->index);
 252                 WARN_ON(1);
 253                 goto err;
 254         }
 255         if (!PageUptodate(page)) {
 256                 printk("csum not up to date page %lu\n", page->index);
 257                 WARN_ON(1);
 258                 goto err;
 259         }
 260         found_level = btrfs_header_level(eb);
 261         spin_lock(&root->fs_info->hash_lock);
 262         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
 263         spin_unlock(&root->fs_info->hash_lock);
 264         csum_tree_block(root, eb, 0);
 265 err:
 266         free_extent_buffer(eb);
 267 out:
 268         return 0;
 269 }
 270
 271 static int btree_writepage_io_hook(struct page *page, u64 start, u64 end)
 272 {
 273         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 274
 275         csum_dirty_buffer(root, page);
 276         return 0;
 277 }
 278
 279 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 280                                struct extent_state *state)
 281 {
 282         struct extent_io_tree *tree;
 283         u64 found_start;
 284         int found_level;
 285         unsigned long len;
 286         struct extent_buffer *eb;
 287         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 288         int ret = 0;
 289
 290         tree = &BTRFS_I(page->mapping->host)->io_tree;
 291         if (page->private == EXTENT_PAGE_PRIVATE)
 292                 goto out;
 293         if (!page->private)
 294                 goto out;
 295         len = page->private >> 2;
 296         if (len == 0) {
 297                 WARN_ON(1);
 298         }
 299         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 300
 301         btrfs_clear_buffer_defrag(eb);
 302         found_start = btrfs_header_bytenr(eb);
 303         if (found_start != start) {
 304 printk("bad start on %Lu found %Lu\n", eb->start, found_start);
 305                 ret = -EIO;
 306                 goto err;
 307         }
 308         if (eb->first_page != page) {
 309                 printk("bad first page %lu %lu\n", eb->first_page->index,
 310                        page->index);
 311                 WARN_ON(1);
 312                 ret = -EIO;
 313                 goto err;
 314         }
 315         found_level = btrfs_header_level(eb);
 316
 317         ret = csum_tree_block(root, eb, 1);
 318         if (ret)
 319                 ret = -EIO;
 320
 321         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 322         end = eb->start + end - 1;
 323         release_extent_buffer_tail_pages(eb);
 324 err:
 325         free_extent_buffer(eb);
 326 out:
 327         return ret;
 328 }
 329
 330 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
 331 static void end_workqueue_bio(struct bio *bio, int err)
 332 #else
 333 static int end_workqueue_bio(struct bio *bio,
 334                                    unsigned int bytes_done, int err)
 335 #endif
 336 {
 337         struct end_io_wq *end_io_wq = bio->bi_private;
 338         struct btrfs_fs_info *fs_info;
 339         unsigned long flags;
 340
 341 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 342         if (bio->bi_size)
 343                 return 1;
 344 #endif
 345
 346         fs_info = end_io_wq->info;
 347         spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
 348         end_io_wq->error = err;
 349         list_add_tail(&end_io_wq->list, &fs_info->end_io_work_list);
 350         spin_unlock_irqrestore(&fs_info->end_io_work_lock, flags);
 351         queue_work(end_io_workqueue, &fs_info->end_io_work);
 352
 353 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 354         return 0;
 355 #endif
 356 }
 357
 358 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 359                         int metadata)
 360 {
 361         struct end_io_wq *end_io_wq;
 362         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 363         if (!end_io_wq)
 364                 return -ENOMEM;
 365
 366         end_io_wq->private = bio->bi_private;
 367         end_io_wq->end_io = bio->bi_end_io;
 368         end_io_wq->info = info;
 369         end_io_wq->error = 0;
 370         end_io_wq->bio = bio;
 371         end_io_wq->metadata = metadata;
 372
 373         bio->bi_private = end_io_wq;
 374         bio->bi_end_io = end_workqueue_bio;
 375         return 0;
 376 }
 377
 378 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 379                         int rw, struct bio *bio, int mirror_num,
 380                         extent_submit_bio_hook_t *submit_bio_hook)
 381 {
 382         struct async_submit_bio *async;
 383
 384         async = kmalloc(sizeof(*async), GFP_NOFS);
 385         if (!async)
 386                 return -ENOMEM;
 387
 388         async->inode = inode;
 389         async->rw = rw;
 390         async->bio = bio;
 391         async->mirror_num = mirror_num;
 392         async->submit_bio_hook = submit_bio_hook;
 393
 394         spin_lock(&fs_info->async_submit_work_lock);
 395         list_add_tail(&async->list, &fs_info->async_submit_work_list);
 396         spin_unlock(&fs_info->async_submit_work_lock);
 397
 398         queue_work(async_submit_workqueue, &fs_info->async_submit_work);
 399         return 0;
 400 }
 401
 402 static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 403                                  int mirror_num)
 404 {
 405         struct btrfs_root *root = BTRFS_I(inode)->root;
 406         u64 offset;
 407         int ret;
 408
 409         offset = bio->bi_sector << 9;
 410
 411         if (rw & (1 << BIO_RW)) {
 412                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num);
 413         }
 414
 415         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
 416         BUG_ON(ret);
 417
 418         if (offset == BTRFS_SUPER_INFO_OFFSET) {
 419                 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
 420                 submit_bio(rw, bio);
 421                 return 0;
 422         }
 423         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num);
 424 }
 425
 426 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 427                                  int mirror_num)
 428 {
 429         if (!(rw & (1 << BIO_RW))) {
 430                 return __btree_submit_bio_hook(inode, rw, bio, mirror_num);
 431         }
 432         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 433                                    inode, rw, bio, mirror_num,
 434                                    __btree_submit_bio_hook);
 435 }
 436
 437 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 438 {
 439         struct extent_io_tree *tree;
 440         tree = &BTRFS_I(page->mapping->host)->io_tree;
 441         return extent_write_full_page(tree, page, btree_get_extent, wbc);
 442 }
 443
 444 static int btree_writepages(struct address_space *mapping,
 445                             struct writeback_control *wbc)
 446 {
 447         struct extent_io_tree *tree;
 448         tree = &BTRFS_I(mapping->host)->io_tree;
 449         if (wbc->sync_mode == WB_SYNC_NONE) {
 450                 u64 num_dirty;
 451                 u64 start = 0;
 452                 unsigned long thresh = 96 * 1024 * 1024;
 453
 454                 if (wbc->for_kupdate)
 455                         return 0;
 456
 457                 if (current_is_pdflush()) {
 458                         thresh = 96 * 1024 * 1024;
 459                 } else {
 460                         thresh = 8 * 1024 * 1024;
 461                 }
 462                 num_dirty = count_range_bits(tree, &start, (u64)-1,
 463                                              thresh, EXTENT_DIRTY);
 464                 if (num_dirty < thresh) {
 465                         return 0;
 466                 }
 467         }
 468         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 469 }
 470
 471 int btree_readpage(struct file *file, struct page *page)
 472 {
 473         struct extent_io_tree *tree;
 474         tree = &BTRFS_I(page->mapping->host)->io_tree;
 475         return extent_read_full_page(tree, page, btree_get_extent);
 476 }
 477
 478 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 479 {
 480         struct extent_io_tree *tree;
 481         struct extent_map_tree *map;
 482         int ret;
 483
 484         if (page_count(page) > 3) {
 485                 /* once for page->private, once for the caller, once
 486                  * once for the page cache
 487                  */
 488                 return 0;
 489         }
 490         tree = &BTRFS_I(page->mapping->host)->io_tree;
 491         map = &BTRFS_I(page->mapping->host)->extent_tree;
 492         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
 493         if (ret == 1) {
 494                 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
 495                 ClearPagePrivate(page);
 496                 set_page_private(page, 0);
 497                 page_cache_release(page);
 498         }
 499         return ret;
 500 }
 501
 502 static void btree_invalidatepage(struct page *page, unsigned long offset)
 503 {
 504         struct extent_io_tree *tree;
 505         tree = &BTRFS_I(page->mapping->host)->io_tree;
 506         extent_invalidatepage(tree, page, offset);
 507         btree_releasepage(page, GFP_NOFS);
 508 }
 509
 510 #if 0
 511 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 512 {
 513         struct buffer_head *bh;
 514         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 515         struct buffer_head *head;
 516         if (!page_has_buffers(page)) {
 517                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 518                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 519         }
 520         head = page_buffers(page);
 521         bh = head;
 522         do {
 523                 if (buffer_dirty(bh))
 524                         csum_tree_block(root, bh, 0);
 525                 bh = bh->b_this_page;
 526         } while (bh != head);
 527         return block_write_full_page(page, btree_get_block, wbc);
 528 }
 529 #endif
 530
 531 static struct address_space_operations btree_aops = {
 532         .readpage       = btree_readpage,
 533         .writepage      = btree_writepage,
 534         .writepages     = btree_writepages,
 535         .releasepage    = btree_releasepage,
 536         .invalidatepage = btree_invalidatepage,
 537         .sync_page      = block_sync_page,
 538 };
 539
 540 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize)
 541 {
 542         struct extent_buffer *buf = NULL;
 543         struct inode *btree_inode = root->fs_info->btree_inode;
 544         int ret = 0;
 545
 546         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 547         if (!buf)
 548                 return 0;
 549         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 550                                  buf, 0, 0, btree_get_extent, 0);
 551         free_extent_buffer(buf);
 552         return ret;
 553 }
 554
 555 static int close_all_devices(struct btrfs_fs_info *fs_info)
 556 {
 557         struct list_head *list;
 558         struct list_head *next;
 559         struct btrfs_device *device;
 560
 561         list = &fs_info->fs_devices->devices;
 562         list_for_each(next, list) {
 563                 device = list_entry(next, struct btrfs_device, dev_list);
 564                 if (device->bdev && device->bdev != fs_info->sb->s_bdev)
 565                         close_bdev_excl(device->bdev);
 566                 device->bdev = NULL;
 567         }
 568         return 0;
 569 }
 570
 571 int btrfs_verify_block_csum(struct btrfs_root *root,
 572                             struct extent_buffer *buf)
 573 {
 574         return btrfs_buffer_uptodate(buf);
 575 }
 576
 577 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 578                                             u64 bytenr, u32 blocksize)
 579 {
 580         struct inode *btree_inode = root->fs_info->btree_inode;
 581         struct extent_buffer *eb;
 582         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 583                                 bytenr, blocksize, GFP_NOFS);
 584         return eb;
 585 }
 586
 587 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 588                                                  u64 bytenr, u32 blocksize)
 589 {
 590         struct inode *btree_inode = root->fs_info->btree_inode;
 591         struct extent_buffer *eb;
 592
 593         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 594                                  bytenr, blocksize, NULL, GFP_NOFS);
 595         return eb;
 596 }
 597
 598
 599 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 600                                       u32 blocksize)
 601 {
 602         struct extent_buffer *buf = NULL;
 603         struct inode *btree_inode = root->fs_info->btree_inode;
 604         struct extent_io_tree *io_tree;
 605         int ret;
 606
 607         io_tree = &BTRFS_I(btree_inode)->io_tree;
 608
 609         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 610         if (!buf)
 611                 return NULL;
 612
 613         ret = btree_read_extent_buffer_pages(root, buf, 0);
 614
 615         if (ret == 0) {
 616                 buf->flags |= EXTENT_UPTODATE;
 617         }
 618         return buf;
 619
 620 }
 621
 622 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 623                      struct extent_buffer *buf)
 624 {
 625         struct inode *btree_inode = root->fs_info->btree_inode;
 626         if (btrfs_header_generation(buf) ==
 627             root->fs_info->running_transaction->transid)
 628                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 629                                           buf);
 630         return 0;
 631 }
 632
 633 int wait_on_tree_block_writeback(struct btrfs_root *root,
 634                                  struct extent_buffer *buf)
 635 {
 636         struct inode *btree_inode = root->fs_info->btree_inode;
 637         wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
 638                                         buf);
 639         return 0;
 640 }
 641
 642 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 643                         u32 stripesize, struct btrfs_root *root,
 644                         struct btrfs_fs_info *fs_info,
 645                         u64 objectid)
 646 {
 647         root->node = NULL;
 648         root->inode = NULL;
 649         root->commit_root = NULL;
 650         root->sectorsize = sectorsize;
 651         root->nodesize = nodesize;
 652         root->leafsize = leafsize;
 653         root->stripesize = stripesize;
 654         root->ref_cows = 0;
 655         root->track_dirty = 0;
 656
 657         root->fs_info = fs_info;
 658         root->objectid = objectid;
 659         root->last_trans = 0;
 660         root->highest_inode = 0;
 661         root->last_inode_alloc = 0;
 662         root->name = NULL;
 663         root->in_sysfs = 0;
 664
 665         INIT_LIST_HEAD(&root->dirty_list);
 666         memset(&root->root_key, 0, sizeof(root->root_key));
 667         memset(&root->root_item, 0, sizeof(root->root_item));
 668         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 669         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 670         init_completion(&root->kobj_unregister);
 671         root->defrag_running = 0;
 672         root->defrag_level = 0;
 673         root->root_key.objectid = objectid;
 674         return 0;
 675 }
 676
 677 static int find_and_setup_root(struct btrfs_root *tree_root,
 678                                struct btrfs_fs_info *fs_info,
 679                                u64 objectid,
 680                                struct btrfs_root *root)
 681 {
 682         int ret;
 683         u32 blocksize;
 684
 685         __setup_root(tree_root->nodesize, tree_root->leafsize,
 686                      tree_root->sectorsize, tree_root->stripesize,
 687                      root, fs_info, objectid);
 688         ret = btrfs_find_last_root(tree_root, objectid,
 689                                    &root->root_item, &root->root_key);
 690         BUG_ON(ret);
 691
 692         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 693         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 694                                      blocksize);
 695         BUG_ON(!root->node);
 696         return 0;
 697 }
 698
 699 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
 700                                                struct btrfs_key *location)
 701 {
 702         struct btrfs_root *root;
 703         struct btrfs_root *tree_root = fs_info->tree_root;
 704         struct btrfs_path *path;
 705         struct extent_buffer *l;
 706         u64 highest_inode;
 707         u32 blocksize;
 708         int ret = 0;
 709
 710         root = kzalloc(sizeof(*root), GFP_NOFS);
 711         if (!root)
 712                 return ERR_PTR(-ENOMEM);
 713         if (location->offset == (u64)-1) {
 714                 ret = find_and_setup_root(tree_root, fs_info,
 715                                           location->objectid, root);
 716                 if (ret) {
 717                         kfree(root);
 718                         return ERR_PTR(ret);
 719                 }
 720                 goto insert;
 721         }
 722
 723         __setup_root(tree_root->nodesize, tree_root->leafsize,
 724                      tree_root->sectorsize, tree_root->stripesize,
 725                      root, fs_info, location->objectid);
 726
 727         path = btrfs_alloc_path();
 728         BUG_ON(!path);
 729         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 730         if (ret != 0) {
 731                 if (ret > 0)
 732                         ret = -ENOENT;
 733                 goto out;
 734         }
 735         l = path->nodes[0];
 736         read_extent_buffer(l, &root->root_item,
 737                btrfs_item_ptr_offset(l, path->slots[0]),
 738                sizeof(root->root_item));
 739         memcpy(&root->root_key, location, sizeof(*location));
 740         ret = 0;
 741 out:
 742         btrfs_release_path(root, path);
 743         btrfs_free_path(path);
 744         if (ret) {
 745                 kfree(root);
 746                 return ERR_PTR(ret);
 747         }
 748         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 749         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 750                                      blocksize);
 751         BUG_ON(!root->node);
 752 insert:
 753         root->ref_cows = 1;
 754         ret = btrfs_find_highest_inode(root, &highest_inode);
 755         if (ret == 0) {
 756                 root->highest_inode = highest_inode;
 757                 root->last_inode_alloc = highest_inode;
 758         }
 759         return root;
 760 }
 761
 762 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
 763                                         u64 root_objectid)
 764 {
 765         struct btrfs_root *root;
 766
 767         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
 768                 return fs_info->tree_root;
 769         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
 770                 return fs_info->extent_root;
 771
 772         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 773                                  (unsigned long)root_objectid);
 774         return root;
 775 }
 776
 777 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
 778                                               struct btrfs_key *location)
 779 {
 780         struct btrfs_root *root;
 781         int ret;
 782
 783         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 784                 return fs_info->tree_root;
 785         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
 786                 return fs_info->extent_root;
 787
 788         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 789                                  (unsigned long)location->objectid);
 790         if (root)
 791                 return root;
 792
 793         root = btrfs_read_fs_root_no_radix(fs_info, location);
 794         if (IS_ERR(root))
 795                 return root;
 796         ret = radix_tree_insert(&fs_info->fs_roots_radix,
 797                                 (unsigned long)root->root_key.objectid,
 798                                 root);
 799         if (ret) {
 800                 free_extent_buffer(root->node);
 801                 kfree(root);
 802                 return ERR_PTR(ret);
 803         }
 804         ret = btrfs_find_dead_roots(fs_info->tree_root,
 805                                     root->root_key.objectid, root);
 806         BUG_ON(ret);
 807
 808         return root;
 809 }
 810
 811 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 812                                       struct btrfs_key *location,
 813                                       const char *name, int namelen)
 814 {
 815         struct btrfs_root *root;
 816         int ret;
 817
 818         root = btrfs_read_fs_root_no_name(fs_info, location);
 819         if (!root)
 820                 return NULL;
 821
 822         if (root->in_sysfs)
 823                 return root;
 824
 825         ret = btrfs_set_root_name(root, name, namelen);
 826         if (ret) {
 827                 free_extent_buffer(root->node);
 828                 kfree(root);
 829                 return ERR_PTR(ret);
 830         }
 831
 832         ret = btrfs_sysfs_add_root(root);
 833         if (ret) {
 834                 free_extent_buffer(root->node);
 835                 kfree(root->name);
 836                 kfree(root);
 837                 return ERR_PTR(ret);
 838         }
 839         root->in_sysfs = 1;
 840         return root;
 841 }
 842 #if 0
 843 static int add_hasher(struct btrfs_fs_info *info, char *type) {
 844         struct btrfs_hasher *hasher;
 845
 846         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
 847         if (!hasher)
 848                 return -ENOMEM;
 849         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
 850         if (!hasher->hash_tfm) {
 851                 kfree(hasher);
 852                 return -EINVAL;
 853         }
 854         spin_lock(&info->hash_lock);
 855         list_add(&hasher->list, &info->hashers);
 856         spin_unlock(&info->hash_lock);
 857         return 0;
 858 }
 859 #endif
 860
 861 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
 862 {
 863         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
 864         int ret = 0;
 865         struct list_head *cur;
 866         struct btrfs_device *device;
 867         struct backing_dev_info *bdi;
 868
 869         list_for_each(cur, &info->fs_devices->devices) {
 870                 device = list_entry(cur, struct btrfs_device, dev_list);
 871                 bdi = blk_get_backing_dev_info(device->bdev);
 872                 if (bdi && bdi_congested(bdi, bdi_bits)) {
 873                         ret = 1;
 874                         break;
 875                 }
 876         }
 877         return ret;
 878 }
 879
 880 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
 881 {
 882         struct list_head *cur;
 883         struct btrfs_device *device;
 884         struct btrfs_fs_info *info;
 885
 886         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
 887         list_for_each(cur, &info->fs_devices->devices) {
 888                 device = list_entry(cur, struct btrfs_device, dev_list);
 889                 bdi = blk_get_backing_dev_info(device->bdev);
 890                 if (bdi->unplug_io_fn) {
 891                         bdi->unplug_io_fn(bdi, page);
 892                 }
 893         }
 894 }
 895
 896 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
 897 {
 898 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
 899         bdi_init(bdi);
 900 #endif
 901         bdi->ra_pages   = default_backing_dev_info.ra_pages * 4;
 902         bdi->state              = 0;
 903         bdi->capabilities       = default_backing_dev_info.capabilities;
 904         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
 905         bdi->unplug_io_data     = info;
 906         bdi->congested_fn       = btrfs_congested_fn;
 907         bdi->congested_data     = info;
 908         return 0;
 909 }
 910
 911 static int bio_ready_for_csum(struct bio *bio)
 912 {
 913         u64 length = 0;
 914         u64 buf_len = 0;
 915         u64 start = 0;
 916         struct page *page;
 917         struct extent_io_tree *io_tree = NULL;
 918         struct btrfs_fs_info *info = NULL;
 919         struct bio_vec *bvec;
 920         int i;
 921         int ret;
 922
 923         bio_for_each_segment(bvec, bio, i) {
 924                 page = bvec->bv_page;
 925                 if (page->private == EXTENT_PAGE_PRIVATE) {
 926                         length += bvec->bv_len;
 927                         continue;
 928                 }
 929                 if (!page->private) {
 930                         length += bvec->bv_len;
 931                         continue;
 932                 }
 933                 length = bvec->bv_len;
 934                 buf_len = page->private >> 2;
 935                 start = page_offset(page) + bvec->bv_offset;
 936                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
 937                 info = BTRFS_I(page->mapping->host)->root->fs_info;
 938         }
 939         /* are we fully contained in this bio? */
 940         if (buf_len <= length)
 941                 return 1;
 942
 943         ret = extent_range_uptodate(io_tree, start + length,
 944                                     start + buf_len - 1);
 945         if (ret == 1)
 946                 return ret;
 947         return ret;
 948 }
 949
 950 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 951 static void btrfs_end_io_csum(void *p)
 952 #else
 953 static void btrfs_end_io_csum(struct work_struct *work)
 954 #endif
 955 {
 956 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 957         struct btrfs_fs_info *fs_info = p;
 958 #else
 959         struct btrfs_fs_info *fs_info = container_of(work,
 960                                                      struct btrfs_fs_info,
 961                                                      end_io_work);
 962 #endif
 963         unsigned long flags;
 964         struct end_io_wq *end_io_wq;
 965         struct bio *bio;
 966         struct list_head *next;
 967         int error;
 968         int was_empty;
 969
 970         while(1) {
 971                 spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
 972                 if (list_empty(&fs_info->end_io_work_list)) {
 973                         spin_unlock_irqrestore(&fs_info->end_io_work_lock,
 974                                                flags);
 975                         return;
 976                 }
 977                 next = fs_info->end_io_work_list.next;
 978                 list_del(next);
 979                 spin_unlock_irqrestore(&fs_info->end_io_work_lock, flags);
 980
 981                 end_io_wq = list_entry(next, struct end_io_wq, list);
 982
 983                 bio = end_io_wq->bio;
 984                 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
 985                         spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
 986                         was_empty = list_empty(&fs_info->end_io_work_list);
 987                         list_add_tail(&end_io_wq->list,
 988                                       &fs_info->end_io_work_list);
 989                         spin_unlock_irqrestore(&fs_info->end_io_work_lock,
 990                                                flags);
 991                         if (was_empty)
 992                                 return;
 993                         continue;
 994                 }
 995                 error = end_io_wq->error;
 996                 bio->bi_private = end_io_wq->private;
 997                 bio->bi_end_io = end_io_wq->end_io;
 998                 kfree(end_io_wq);
 999 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1000                 bio_endio(bio, bio->bi_size, error);
1001 #else
1002                 bio_endio(bio, error);
1003 #endif
1004         }
1005 }
1006
1007 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1008 static void btrfs_async_submit_work(void *p)
1009 #else
1010 static void btrfs_async_submit_work(struct work_struct *work)
1011 #endif
1012 {
1013 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1014         struct btrfs_fs_info *fs_info = p;
1015 #else
1016         struct btrfs_fs_info *fs_info = container_of(work,
1017                                                      struct btrfs_fs_info,
1018                                                      async_submit_work);
1019 #endif
1020         struct async_submit_bio *async;
1021         struct list_head *next;
1022
1023         while(1) {
1024                 spin_lock(&fs_info->async_submit_work_lock);
1025                 if (list_empty(&fs_info->async_submit_work_list)) {
1026                         spin_unlock(&fs_info->async_submit_work_lock);
1027                         return;
1028                 }
1029                 next = fs_info->async_submit_work_list.next;
1030                 list_del(next);
1031                 spin_unlock(&fs_info->async_submit_work_lock);
1032
1033                 async = list_entry(next, struct async_submit_bio, list);
1034                 async->submit_bio_hook(async->inode, async->rw, async->bio,
1035                                        async->mirror_num);
1036                 kfree(async);
1037         }
1038 }
1039
1040 struct btrfs_root *open_ctree(struct super_block *sb,
1041                               struct btrfs_fs_devices *fs_devices)
1042 {
1043         u32 sectorsize;
1044         u32 nodesize;
1045         u32 leafsize;
1046         u32 blocksize;
1047         u32 stripesize;
1048         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
1049                                                  GFP_NOFS);
1050         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
1051                                                GFP_NOFS);
1052         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1053                                                 GFP_NOFS);
1054         struct btrfs_root *chunk_root = kmalloc(sizeof(struct btrfs_root),
1055                                                 GFP_NOFS);
1056         struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
1057                                               GFP_NOFS);
1058         int ret;
1059         int err = -EINVAL;
1060         struct btrfs_super_block *disk_super;
1061
1062         if (!extent_root || !tree_root || !fs_info) {
1063                 err = -ENOMEM;
1064                 goto fail;
1065         }
1066         end_io_workqueue = create_workqueue("btrfs-end-io");
1067         BUG_ON(!end_io_workqueue);
1068         async_submit_workqueue = create_workqueue("btrfs-async-submit");
1069
1070         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1071         INIT_LIST_HEAD(&fs_info->trans_list);
1072         INIT_LIST_HEAD(&fs_info->dead_roots);
1073         INIT_LIST_HEAD(&fs_info->hashers);
1074         INIT_LIST_HEAD(&fs_info->end_io_work_list);
1075         INIT_LIST_HEAD(&fs_info->async_submit_work_list);
1076         spin_lock_init(&fs_info->hash_lock);
1077         spin_lock_init(&fs_info->end_io_work_lock);
1078         spin_lock_init(&fs_info->async_submit_work_lock);
1079         spin_lock_init(&fs_info->delalloc_lock);
1080         spin_lock_init(&fs_info->new_trans_lock);
1081
1082         init_completion(&fs_info->kobj_unregister);
1083         sb_set_blocksize(sb, BTRFS_SUPER_INFO_SIZE);
1084         fs_info->tree_root = tree_root;
1085         fs_info->extent_root = extent_root;
1086         fs_info->chunk_root = chunk_root;
1087         fs_info->dev_root = dev_root;
1088         fs_info->fs_devices = fs_devices;
1089         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1090         INIT_LIST_HEAD(&fs_info->space_info);
1091         btrfs_mapping_init(&fs_info->mapping_tree);
1092         fs_info->sb = sb;
1093         fs_info->max_extent = (u64)-1;
1094         fs_info->max_inline = 8192 * 1024;
1095         setup_bdi(fs_info, &fs_info->bdi);
1096         fs_info->btree_inode = new_inode(sb);
1097         fs_info->btree_inode->i_ino = 1;
1098         fs_info->btree_inode->i_nlink = 1;
1099         fs_info->btree_inode->i_size = sb->s_bdev->bd_inode->i_size;
1100         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1101         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1102
1103         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1104                              fs_info->btree_inode->i_mapping,
1105                              GFP_NOFS);
1106         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1107                              GFP_NOFS);
1108
1109         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1110
1111         extent_io_tree_init(&fs_info->free_space_cache,
1112                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1113         extent_io_tree_init(&fs_info->block_group_cache,
1114                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1115         extent_io_tree_init(&fs_info->pinned_extents,
1116                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1117         extent_io_tree_init(&fs_info->pending_del,
1118                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1119         extent_io_tree_init(&fs_info->extent_ins,
1120                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1121         fs_info->do_barriers = 1;
1122
1123 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1124         INIT_WORK(&fs_info->end_io_work, btrfs_end_io_csum, fs_info);
1125         INIT_WORK(&fs_info->async_submit_work, btrfs_async_submit_work,
1126                   fs_info);
1127         INIT_WORK(&fs_info->trans_work, btrfs_transaction_cleaner, fs_info);
1128 #else
1129         INIT_WORK(&fs_info->end_io_work, btrfs_end_io_csum);
1130         INIT_WORK(&fs_info->async_submit_work, btrfs_async_submit_work);
1131         INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
1132 #endif
1133         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1134         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1135                sizeof(struct btrfs_key));
1136         insert_inode_hash(fs_info->btree_inode);
1137         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1138
1139         mutex_init(&fs_info->trans_mutex);
1140         mutex_init(&fs_info->fs_mutex);
1141
1142 #if 0
1143         ret = add_hasher(fs_info, "crc32c");
1144         if (ret) {
1145                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1146                 err = -ENOMEM;
1147                 goto fail_iput;
1148         }
1149 #endif
1150         __setup_root(4096, 4096, 4096, 4096, tree_root,
1151                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1152
1153         fs_info->sb_buffer = read_tree_block(tree_root,
1154                                              BTRFS_SUPER_INFO_OFFSET,
1155                                              4096);
1156
1157         if (!fs_info->sb_buffer)
1158                 goto fail_iput;
1159
1160         read_extent_buffer(fs_info->sb_buffer, &fs_info->super_copy, 0,
1161                            sizeof(fs_info->super_copy));
1162
1163         read_extent_buffer(fs_info->sb_buffer, fs_info->fsid,
1164                            (unsigned long)btrfs_super_fsid(fs_info->sb_buffer),
1165                            BTRFS_FSID_SIZE);
1166
1167         disk_super = &fs_info->super_copy;
1168         if (!btrfs_super_root(disk_super))
1169                 goto fail_sb_buffer;
1170
1171         if (btrfs_super_num_devices(disk_super) != fs_devices->num_devices) {
1172                 printk("Btrfs: wanted %llu devices, but found %llu\n",
1173                        (unsigned long long)btrfs_super_num_devices(disk_super),
1174                        (unsigned long long)fs_devices->num_devices);
1175                 goto fail_sb_buffer;
1176         }
1177         nodesize = btrfs_super_nodesize(disk_super);
1178         leafsize = btrfs_super_leafsize(disk_super);
1179         sectorsize = btrfs_super_sectorsize(disk_super);
1180         stripesize = btrfs_super_stripesize(disk_super);
1181         tree_root->nodesize = nodesize;
1182         tree_root->leafsize = leafsize;
1183         tree_root->sectorsize = sectorsize;
1184         tree_root->stripesize = stripesize;
1185         sb_set_blocksize(sb, sectorsize);
1186
1187         i_size_write(fs_info->btree_inode,
1188                      btrfs_super_total_bytes(disk_super));
1189
1190         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1191                     sizeof(disk_super->magic))) {
1192                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1193                 goto fail_sb_buffer;
1194         }
1195
1196         mutex_lock(&fs_info->fs_mutex);
1197
1198         ret = btrfs_read_sys_array(tree_root);
1199         BUG_ON(ret);
1200
1201         blocksize = btrfs_level_size(tree_root,
1202                                      btrfs_super_chunk_root_level(disk_super));
1203
1204         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1205                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1206
1207         chunk_root->node = read_tree_block(chunk_root,
1208                                            btrfs_super_chunk_root(disk_super),
1209                                            blocksize);
1210         BUG_ON(!chunk_root->node);
1211
1212         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1213                  (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1214                  BTRFS_UUID_SIZE);
1215
1216         ret = btrfs_read_chunk_tree(chunk_root);
1217         BUG_ON(ret);
1218
1219         blocksize = btrfs_level_size(tree_root,
1220                                      btrfs_super_root_level(disk_super));
1221
1222
1223         tree_root->node = read_tree_block(tree_root,
1224                                           btrfs_super_root(disk_super),
1225                                           blocksize);
1226         if (!tree_root->node)
1227                 goto fail_sb_buffer;
1228
1229
1230         ret = find_and_setup_root(tree_root, fs_info,
1231                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1232         if (ret)
1233                 goto fail_tree_root;
1234         extent_root->track_dirty = 1;
1235
1236         ret = find_and_setup_root(tree_root, fs_info,
1237                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1238         dev_root->track_dirty = 1;
1239
1240         if (ret)
1241                 goto fail_extent_root;
1242
1243         btrfs_read_block_groups(extent_root);
1244
1245         fs_info->generation = btrfs_super_generation(disk_super) + 1;
1246         fs_info->data_alloc_profile = (u64)-1;
1247         fs_info->metadata_alloc_profile = (u64)-1;
1248         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1249
1250         mutex_unlock(&fs_info->fs_mutex);
1251         return tree_root;
1252
1253 fail_extent_root:
1254         free_extent_buffer(extent_root->node);
1255 fail_tree_root:
1256         mutex_unlock(&fs_info->fs_mutex);
1257         free_extent_buffer(tree_root->node);
1258 fail_sb_buffer:
1259         free_extent_buffer(fs_info->sb_buffer);
1260         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1261 fail_iput:
1262         iput(fs_info->btree_inode);
1263 fail:
1264         close_all_devices(fs_info);
1265         kfree(extent_root);
1266         kfree(tree_root);
1267 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
1268         bdi_destroy(&fs_info->bdi);
1269 #endif
1270         kfree(fs_info);
1271         return ERR_PTR(err);
1272 }
1273
1274 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1275 {
1276         char b[BDEVNAME_SIZE];
1277
1278         if (uptodate) {
1279                 set_buffer_uptodate(bh);
1280         } else {
1281                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1282                         printk(KERN_WARNING "lost page write due to "
1283                                         "I/O error on %s\n",
1284                                        bdevname(bh->b_bdev, b));
1285                 }
1286                 set_buffer_write_io_error(bh);
1287                 clear_buffer_uptodate(bh);
1288         }
1289         unlock_buffer(bh);
1290         put_bh(bh);
1291 }
1292
1293 int write_all_supers(struct btrfs_root *root)
1294 {
1295         struct list_head *cur;
1296         struct list_head *head = &root->fs_info->fs_devices->devices;
1297         struct btrfs_device *dev;
1298         struct extent_buffer *sb;
1299         struct btrfs_dev_item *dev_item;
1300         struct buffer_head *bh;
1301         int ret;
1302         int do_barriers;
1303
1304         do_barriers = !btrfs_test_opt(root, NOBARRIER);
1305
1306         sb = root->fs_info->sb_buffer;
1307         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
1308                                                       dev_item);
1309         list_for_each(cur, head) {
1310                 dev = list_entry(cur, struct btrfs_device, dev_list);
1311                 btrfs_set_device_type(sb, dev_item, dev->type);
1312                 btrfs_set_device_id(sb, dev_item, dev->devid);
1313                 btrfs_set_device_total_bytes(sb, dev_item, dev->total_bytes);
1314                 btrfs_set_device_bytes_used(sb, dev_item, dev->bytes_used);
1315                 btrfs_set_device_io_align(sb, dev_item, dev->io_align);
1316                 btrfs_set_device_io_width(sb, dev_item, dev->io_width);
1317                 btrfs_set_device_sector_size(sb, dev_item, dev->sector_size);
1318                 write_extent_buffer(sb, dev->uuid,
1319                                     (unsigned long)btrfs_device_uuid(dev_item),
1320                                     BTRFS_UUID_SIZE);
1321
1322                 btrfs_set_header_flag(sb, BTRFS_HEADER_FLAG_WRITTEN);
1323                 csum_tree_block(root, sb, 0);
1324
1325                 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET /
1326                               root->fs_info->sb->s_blocksize,
1327                               BTRFS_SUPER_INFO_SIZE);
1328
1329                 read_extent_buffer(sb, bh->b_data, 0, BTRFS_SUPER_INFO_SIZE);
1330                 dev->pending_io = bh;
1331
1332                 get_bh(bh);
1333                 set_buffer_uptodate(bh);
1334                 lock_buffer(bh);
1335                 bh->b_end_io = btrfs_end_buffer_write_sync;
1336
1337                 if (do_barriers && dev->barriers) {
1338                         ret = submit_bh(WRITE_BARRIER, bh);
1339                         if (ret == -EOPNOTSUPP) {
1340                                 printk("btrfs: disabling barriers on dev %s\n",
1341                                        dev->name);
1342                                 set_buffer_uptodate(bh);
1343                                 dev->barriers = 0;
1344                                 get_bh(bh);
1345                                 lock_buffer(bh);
1346                                 ret = submit_bh(WRITE, bh);
1347                         }
1348                 } else {
1349                         ret = submit_bh(WRITE, bh);
1350                 }
1351                 BUG_ON(ret);
1352         }
1353
1354         list_for_each(cur, head) {
1355                 dev = list_entry(cur, struct btrfs_device, dev_list);
1356                 BUG_ON(!dev->pending_io);
1357                 bh = dev->pending_io;
1358                 wait_on_buffer(bh);
1359                 if (!buffer_uptodate(dev->pending_io)) {
1360                         if (do_barriers && dev->barriers) {
1361                                 printk("btrfs: disabling barriers on dev %s\n",
1362                                        dev->name);
1363                                 set_buffer_uptodate(bh);
1364                                 get_bh(bh);
1365                                 lock_buffer(bh);
1366                                 dev->barriers = 0;
1367                                 ret = submit_bh(WRITE, bh);
1368                                 BUG_ON(ret);
1369                                 wait_on_buffer(bh);
1370                                 BUG_ON(!buffer_uptodate(bh));
1371                         } else {
1372                                 BUG();
1373                         }
1374
1375                 }
1376                 dev->pending_io = NULL;
1377                 brelse(bh);
1378         }
1379         return 0;
1380 }
1381
1382 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1383                       *root)
1384 {
1385         int ret;
1386
1387         ret = write_all_supers(root);
1388 #if 0
1389         if (!btrfs_test_opt(root, NOBARRIER))
1390                 blkdev_issue_flush(sb->s_bdev, NULL);
1391         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, super);
1392         ret = sync_page_range_nolock(btree_inode, btree_inode->i_mapping,
1393                                      super->start, super->len);
1394         if (!btrfs_test_opt(root, NOBARRIER))
1395                 blkdev_issue_flush(sb->s_bdev, NULL);
1396 #endif
1397         return ret;
1398 }
1399
1400 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1401 {
1402         radix_tree_delete(&fs_info->fs_roots_radix,
1403                           (unsigned long)root->root_key.objectid);
1404         if (root->in_sysfs)
1405                 btrfs_sysfs_del_root(root);
1406         if (root->inode)
1407                 iput(root->inode);
1408         if (root->node)
1409                 free_extent_buffer(root->node);
1410         if (root->commit_root)
1411                 free_extent_buffer(root->commit_root);
1412         if (root->name)
1413                 kfree(root->name);
1414         kfree(root);
1415         return 0;
1416 }
1417
1418 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1419 {
1420         int ret;
1421         struct btrfs_root *gang[8];
1422         int i;
1423
1424         while(1) {
1425                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1426                                              (void **)gang, 0,
1427                                              ARRAY_SIZE(gang));
1428                 if (!ret)
1429                         break;
1430                 for (i = 0; i < ret; i++)
1431                         btrfs_free_fs_root(fs_info, gang[i]);
1432         }
1433         return 0;
1434 }
1435
1436 int close_ctree(struct btrfs_root *root)
1437 {
1438         int ret;
1439         struct btrfs_trans_handle *trans;
1440         struct btrfs_fs_info *fs_info = root->fs_info;
1441
1442         fs_info->closing = 1;
1443         btrfs_transaction_flush_work(root);
1444         mutex_lock(&fs_info->fs_mutex);
1445         btrfs_defrag_dirty_roots(root->fs_info);
1446         trans = btrfs_start_transaction(root, 1);
1447         ret = btrfs_commit_transaction(trans, root);
1448         /* run commit again to  drop the original snapshot */
1449         trans = btrfs_start_transaction(root, 1);
1450         btrfs_commit_transaction(trans, root);
1451         ret = btrfs_write_and_wait_transaction(NULL, root);
1452         BUG_ON(ret);
1453         write_ctree_super(NULL, root);
1454         mutex_unlock(&fs_info->fs_mutex);
1455
1456         if (fs_info->delalloc_bytes) {
1457                 printk("btrfs: at unmount delalloc count %Lu\n",
1458                        fs_info->delalloc_bytes);
1459         }
1460         if (fs_info->extent_root->node)
1461                 free_extent_buffer(fs_info->extent_root->node);
1462
1463         if (fs_info->tree_root->node)
1464                 free_extent_buffer(fs_info->tree_root->node);
1465
1466         if (root->fs_info->chunk_root->node);
1467                 free_extent_buffer(root->fs_info->chunk_root->node);
1468
1469         if (root->fs_info->dev_root->node);
1470                 free_extent_buffer(root->fs_info->dev_root->node);
1471
1472         free_extent_buffer(fs_info->sb_buffer);
1473
1474         btrfs_free_block_groups(root->fs_info);
1475         del_fs_roots(fs_info);
1476
1477         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1478
1479         extent_io_tree_empty_lru(&fs_info->free_space_cache);
1480         extent_io_tree_empty_lru(&fs_info->block_group_cache);
1481         extent_io_tree_empty_lru(&fs_info->pinned_extents);
1482         extent_io_tree_empty_lru(&fs_info->pending_del);
1483         extent_io_tree_empty_lru(&fs_info->extent_ins);
1484         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1485
1486         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1487         flush_workqueue(end_io_workqueue);
1488         destroy_workqueue(end_io_workqueue);
1489
1490         flush_workqueue(async_submit_workqueue);
1491         destroy_workqueue(async_submit_workqueue);
1492
1493         iput(fs_info->btree_inode);
1494 #if 0
1495         while(!list_empty(&fs_info->hashers)) {
1496                 struct btrfs_hasher *hasher;
1497                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1498                                     hashers);
1499                 list_del(&hasher->hashers);
1500                 crypto_free_hash(&fs_info->hash_tfm);
1501                 kfree(hasher);
1502         }
1503 #endif
1504         close_all_devices(fs_info);
1505         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1506
1507 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
1508         bdi_destroy(&fs_info->bdi);
1509 #endif
1510
1511         kfree(fs_info->extent_root);
1512         kfree(fs_info->tree_root);
1513         kfree(fs_info->chunk_root);
1514         kfree(fs_info->dev_root);
1515         return 0;
1516 }
1517
1518 int btrfs_buffer_uptodate(struct extent_buffer *buf)
1519 {
1520         struct inode *btree_inode = buf->first_page->mapping->host;
1521         return extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1522 }
1523
1524 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1525 {
1526         struct inode *btree_inode = buf->first_page->mapping->host;
1527         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1528                                           buf);
1529 }
1530
1531 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1532 {
1533         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1534         u64 transid = btrfs_header_generation(buf);
1535         struct inode *btree_inode = root->fs_info->btree_inode;
1536
1537         if (transid != root->fs_info->generation) {
1538                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1539                         (unsigned long long)buf->start,
1540                         transid, root->fs_info->generation);
1541                 WARN_ON(1);
1542         }
1543         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1544 }
1545
1546 void btrfs_throttle(struct btrfs_root *root)
1547 {
1548         struct backing_dev_info *bdi;
1549
1550         bdi = root->fs_info->sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
1551         if (root->fs_info->throttles && bdi_write_congested(bdi)) {
1552 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
1553                 congestion_wait(WRITE, HZ/20);
1554 #else
1555                 blk_congestion_wait(WRITE, HZ/20);
1556 #endif
1557         }
1558 }
1559
1560 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1561 {
1562         balance_dirty_pages_ratelimited_nr(
1563                                    root->fs_info->btree_inode->i_mapping, 1);
1564 }
1565
1566 void btrfs_set_buffer_defrag(struct extent_buffer *buf)
1567 {
1568         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1569         struct inode *btree_inode = root->fs_info->btree_inode;
1570         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1571                         buf->start + buf->len - 1, EXTENT_DEFRAG, GFP_NOFS);
1572 }
1573
1574 void btrfs_set_buffer_defrag_done(struct extent_buffer *buf)
1575 {
1576         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1577         struct inode *btree_inode = root->fs_info->btree_inode;
1578         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1579                         buf->start + buf->len - 1, EXTENT_DEFRAG_DONE,
1580                         GFP_NOFS);
1581 }
1582
1583 int btrfs_buffer_defrag(struct extent_buffer *buf)
1584 {
1585         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1586         struct inode *btree_inode = root->fs_info->btree_inode;
1587         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1588                      buf->start, buf->start + buf->len - 1, EXTENT_DEFRAG, 0);
1589 }
1590
1591 int btrfs_buffer_defrag_done(struct extent_buffer *buf)
1592 {
1593         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1594         struct inode *btree_inode = root->fs_info->btree_inode;
1595         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1596                      buf->start, buf->start + buf->len - 1,
1597                      EXTENT_DEFRAG_DONE, 0);
1598 }
1599
1600 int btrfs_clear_buffer_defrag_done(struct extent_buffer *buf)
1601 {
1602         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1603         struct inode *btree_inode = root->fs_info->btree_inode;
1604         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1605                      buf->start, buf->start + buf->len - 1,
1606                      EXTENT_DEFRAG_DONE, GFP_NOFS);
1607 }
1608
1609 int btrfs_clear_buffer_defrag(struct extent_buffer *buf)
1610 {
1611         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1612         struct inode *btree_inode = root->fs_info->btree_inode;
1613         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1614                      buf->start, buf->start + buf->len - 1,
1615                      EXTENT_DEFRAG, GFP_NOFS);
1616 }
1617
1618 int btrfs_read_buffer(struct extent_buffer *buf)
1619 {
1620         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1621         int ret;
1622         ret = btree_read_extent_buffer_pages(root, buf, 0);
1623         if (ret == 0) {
1624                 buf->flags |= EXTENT_UPTODATE;
1625         }
1626         return ret;
1627 }
1628
1629 static struct extent_io_ops btree_extent_io_ops = {
1630         .writepage_io_hook = btree_writepage_io_hook,
1631         .readpage_end_io_hook = btree_readpage_end_io_hook,
1632         .submit_bio_hook = btree_submit_bio_hook,
1633         /* note we're sharing with inode.c for the merge bio hook */
1634         .merge_bio_hook = btrfs_merge_bio_hook,
1635 };