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         /*
 385          * inline writerback should stay inline, only hop to the async
 386          * queue if we're pdflush
 387          */
 388         if (!current_is_pdflush())
 389                 return submit_bio_hook(inode, rw, bio, mirror_num);
 390
 391         async = kmalloc(sizeof(*async), GFP_NOFS);
 392         if (!async)
 393                 return -ENOMEM;
 394
 395         async->inode = inode;
 396         async->rw = rw;
 397         async->bio = bio;
 398         async->mirror_num = mirror_num;
 399         async->submit_bio_hook = submit_bio_hook;
 400
 401         spin_lock(&fs_info->async_submit_work_lock);
 402         list_add_tail(&async->list, &fs_info->async_submit_work_list);
 403         spin_unlock(&fs_info->async_submit_work_lock);
 404
 405         queue_work(async_submit_workqueue, &fs_info->async_submit_work);
 406         return 0;
 407 }
 408
 409 static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 410                                  int mirror_num)
 411 {
 412         struct btrfs_root *root = BTRFS_I(inode)->root;
 413         u64 offset;
 414         int ret;
 415
 416         offset = bio->bi_sector << 9;
 417
 418         if (rw & (1 << BIO_RW)) {
 419                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num);
 420         }
 421
 422         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
 423         BUG_ON(ret);
 424
 425         if (offset == BTRFS_SUPER_INFO_OFFSET) {
 426                 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
 427                 submit_bio(rw, bio);
 428                 return 0;
 429         }
 430         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num);
 431 }
 432
 433 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 434                                  int mirror_num)
 435 {
 436         if (!(rw & (1 << BIO_RW))) {
 437                 return __btree_submit_bio_hook(inode, rw, bio, mirror_num);
 438         }
 439         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 440                                    inode, rw, bio, mirror_num,
 441                                    __btree_submit_bio_hook);
 442 }
 443
 444 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 445 {
 446         struct extent_io_tree *tree;
 447         tree = &BTRFS_I(page->mapping->host)->io_tree;
 448         return extent_write_full_page(tree, page, btree_get_extent, wbc);
 449 }
 450
 451 static int btree_writepages(struct address_space *mapping,
 452                             struct writeback_control *wbc)
 453 {
 454         struct extent_io_tree *tree;
 455         tree = &BTRFS_I(mapping->host)->io_tree;
 456         if (wbc->sync_mode == WB_SYNC_NONE) {
 457                 u64 num_dirty;
 458                 u64 start = 0;
 459                 unsigned long thresh = 96 * 1024 * 1024;
 460
 461                 if (wbc->for_kupdate)
 462                         return 0;
 463
 464                 if (current_is_pdflush()) {
 465                         thresh = 96 * 1024 * 1024;
 466                 } else {
 467                         thresh = 8 * 1024 * 1024;
 468                 }
 469                 num_dirty = count_range_bits(tree, &start, (u64)-1,
 470                                              thresh, EXTENT_DIRTY);
 471                 if (num_dirty < thresh) {
 472                         return 0;
 473                 }
 474         }
 475         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 476 }
 477
 478 int btree_readpage(struct file *file, struct page *page)
 479 {
 480         struct extent_io_tree *tree;
 481         tree = &BTRFS_I(page->mapping->host)->io_tree;
 482         return extent_read_full_page(tree, page, btree_get_extent);
 483 }
 484
 485 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 486 {
 487         struct extent_io_tree *tree;
 488         struct extent_map_tree *map;
 489         int ret;
 490
 491         if (page_count(page) > 3) {
 492                 /* once for page->private, once for the caller, once
 493                  * once for the page cache
 494                  */
 495                 return 0;
 496         }
 497         tree = &BTRFS_I(page->mapping->host)->io_tree;
 498         map = &BTRFS_I(page->mapping->host)->extent_tree;
 499         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
 500         if (ret == 1) {
 501                 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
 502                 ClearPagePrivate(page);
 503                 set_page_private(page, 0);
 504                 page_cache_release(page);
 505         }
 506         return ret;
 507 }
 508
 509 static void btree_invalidatepage(struct page *page, unsigned long offset)
 510 {
 511         struct extent_io_tree *tree;
 512         tree = &BTRFS_I(page->mapping->host)->io_tree;
 513         extent_invalidatepage(tree, page, offset);
 514         btree_releasepage(page, GFP_NOFS);
 515 }
 516
 517 #if 0
 518 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 519 {
 520         struct buffer_head *bh;
 521         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 522         struct buffer_head *head;
 523         if (!page_has_buffers(page)) {
 524                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 525                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 526         }
 527         head = page_buffers(page);
 528         bh = head;
 529         do {
 530                 if (buffer_dirty(bh))
 531                         csum_tree_block(root, bh, 0);
 532                 bh = bh->b_this_page;
 533         } while (bh != head);
 534         return block_write_full_page(page, btree_get_block, wbc);
 535 }
 536 #endif
 537
 538 static struct address_space_operations btree_aops = {
 539         .readpage       = btree_readpage,
 540         .writepage      = btree_writepage,
 541         .writepages     = btree_writepages,
 542         .releasepage    = btree_releasepage,
 543         .invalidatepage = btree_invalidatepage,
 544         .sync_page      = block_sync_page,
 545 };
 546
 547 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize)
 548 {
 549         struct extent_buffer *buf = NULL;
 550         struct inode *btree_inode = root->fs_info->btree_inode;
 551         int ret = 0;
 552
 553         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 554         if (!buf)
 555                 return 0;
 556         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 557                                  buf, 0, 0, btree_get_extent, 0);
 558         free_extent_buffer(buf);
 559         return ret;
 560 }
 561
 562 static int close_all_devices(struct btrfs_fs_info *fs_info)
 563 {
 564         struct list_head *list;
 565         struct list_head *next;
 566         struct btrfs_device *device;
 567
 568         list = &fs_info->fs_devices->devices;
 569         list_for_each(next, list) {
 570                 device = list_entry(next, struct btrfs_device, dev_list);
 571                 if (device->bdev && device->bdev != fs_info->sb->s_bdev)
 572                         close_bdev_excl(device->bdev);
 573                 device->bdev = NULL;
 574         }
 575         return 0;
 576 }
 577
 578 int btrfs_verify_block_csum(struct btrfs_root *root,
 579                             struct extent_buffer *buf)
 580 {
 581         return btrfs_buffer_uptodate(buf);
 582 }
 583
 584 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 585                                             u64 bytenr, u32 blocksize)
 586 {
 587         struct inode *btree_inode = root->fs_info->btree_inode;
 588         struct extent_buffer *eb;
 589         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 590                                 bytenr, blocksize, GFP_NOFS);
 591         return eb;
 592 }
 593
 594 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 595                                                  u64 bytenr, u32 blocksize)
 596 {
 597         struct inode *btree_inode = root->fs_info->btree_inode;
 598         struct extent_buffer *eb;
 599
 600         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 601                                  bytenr, blocksize, NULL, GFP_NOFS);
 602         return eb;
 603 }
 604
 605
 606 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 607                                       u32 blocksize)
 608 {
 609         struct extent_buffer *buf = NULL;
 610         struct inode *btree_inode = root->fs_info->btree_inode;
 611         struct extent_io_tree *io_tree;
 612         int ret;
 613
 614         io_tree = &BTRFS_I(btree_inode)->io_tree;
 615
 616         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 617         if (!buf)
 618                 return NULL;
 619
 620         ret = btree_read_extent_buffer_pages(root, buf, 0);
 621
 622         if (ret == 0) {
 623                 buf->flags |= EXTENT_UPTODATE;
 624         }
 625         return buf;
 626
 627 }
 628
 629 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 630                      struct extent_buffer *buf)
 631 {
 632         struct inode *btree_inode = root->fs_info->btree_inode;
 633         if (btrfs_header_generation(buf) ==
 634             root->fs_info->running_transaction->transid)
 635                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 636                                           buf);
 637         return 0;
 638 }
 639
 640 int wait_on_tree_block_writeback(struct btrfs_root *root,
 641                                  struct extent_buffer *buf)
 642 {
 643         struct inode *btree_inode = root->fs_info->btree_inode;
 644         wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
 645                                         buf);
 646         return 0;
 647 }
 648
 649 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 650                         u32 stripesize, struct btrfs_root *root,
 651                         struct btrfs_fs_info *fs_info,
 652                         u64 objectid)
 653 {
 654         root->node = NULL;
 655         root->inode = NULL;
 656         root->commit_root = NULL;
 657         root->sectorsize = sectorsize;
 658         root->nodesize = nodesize;
 659         root->leafsize = leafsize;
 660         root->stripesize = stripesize;
 661         root->ref_cows = 0;
 662         root->track_dirty = 0;
 663
 664         root->fs_info = fs_info;
 665         root->objectid = objectid;
 666         root->last_trans = 0;
 667         root->highest_inode = 0;
 668         root->last_inode_alloc = 0;
 669         root->name = NULL;
 670         root->in_sysfs = 0;
 671
 672         INIT_LIST_HEAD(&root->dirty_list);
 673         memset(&root->root_key, 0, sizeof(root->root_key));
 674         memset(&root->root_item, 0, sizeof(root->root_item));
 675         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 676         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 677         init_completion(&root->kobj_unregister);
 678         root->defrag_running = 0;
 679         root->defrag_level = 0;
 680         root->root_key.objectid = objectid;
 681         return 0;
 682 }
 683
 684 static int find_and_setup_root(struct btrfs_root *tree_root,
 685                                struct btrfs_fs_info *fs_info,
 686                                u64 objectid,
 687                                struct btrfs_root *root)
 688 {
 689         int ret;
 690         u32 blocksize;
 691
 692         __setup_root(tree_root->nodesize, tree_root->leafsize,
 693                      tree_root->sectorsize, tree_root->stripesize,
 694                      root, fs_info, objectid);
 695         ret = btrfs_find_last_root(tree_root, objectid,
 696                                    &root->root_item, &root->root_key);
 697         BUG_ON(ret);
 698
 699         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 700         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 701                                      blocksize);
 702         BUG_ON(!root->node);
 703         return 0;
 704 }
 705
 706 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
 707                                                struct btrfs_key *location)
 708 {
 709         struct btrfs_root *root;
 710         struct btrfs_root *tree_root = fs_info->tree_root;
 711         struct btrfs_path *path;
 712         struct extent_buffer *l;
 713         u64 highest_inode;
 714         u32 blocksize;
 715         int ret = 0;
 716
 717         root = kzalloc(sizeof(*root), GFP_NOFS);
 718         if (!root)
 719                 return ERR_PTR(-ENOMEM);
 720         if (location->offset == (u64)-1) {
 721                 ret = find_and_setup_root(tree_root, fs_info,
 722                                           location->objectid, root);
 723                 if (ret) {
 724                         kfree(root);
 725                         return ERR_PTR(ret);
 726                 }
 727                 goto insert;
 728         }
 729
 730         __setup_root(tree_root->nodesize, tree_root->leafsize,
 731                      tree_root->sectorsize, tree_root->stripesize,
 732                      root, fs_info, location->objectid);
 733
 734         path = btrfs_alloc_path();
 735         BUG_ON(!path);
 736         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 737         if (ret != 0) {
 738                 if (ret > 0)
 739                         ret = -ENOENT;
 740                 goto out;
 741         }
 742         l = path->nodes[0];
 743         read_extent_buffer(l, &root->root_item,
 744                btrfs_item_ptr_offset(l, path->slots[0]),
 745                sizeof(root->root_item));
 746         memcpy(&root->root_key, location, sizeof(*location));
 747         ret = 0;
 748 out:
 749         btrfs_release_path(root, path);
 750         btrfs_free_path(path);
 751         if (ret) {
 752                 kfree(root);
 753                 return ERR_PTR(ret);
 754         }
 755         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 756         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 757                                      blocksize);
 758         BUG_ON(!root->node);
 759 insert:
 760         root->ref_cows = 1;
 761         ret = btrfs_find_highest_inode(root, &highest_inode);
 762         if (ret == 0) {
 763                 root->highest_inode = highest_inode;
 764                 root->last_inode_alloc = highest_inode;
 765         }
 766         return root;
 767 }
 768
 769 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
 770                                         u64 root_objectid)
 771 {
 772         struct btrfs_root *root;
 773
 774         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
 775                 return fs_info->tree_root;
 776         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
 777                 return fs_info->extent_root;
 778
 779         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 780                                  (unsigned long)root_objectid);
 781         return root;
 782 }
 783
 784 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
 785                                               struct btrfs_key *location)
 786 {
 787         struct btrfs_root *root;
 788         int ret;
 789
 790         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 791                 return fs_info->tree_root;
 792         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
 793                 return fs_info->extent_root;
 794
 795         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 796                                  (unsigned long)location->objectid);
 797         if (root)
 798                 return root;
 799
 800         root = btrfs_read_fs_root_no_radix(fs_info, location);
 801         if (IS_ERR(root))
 802                 return root;
 803         ret = radix_tree_insert(&fs_info->fs_roots_radix,
 804                                 (unsigned long)root->root_key.objectid,
 805                                 root);
 806         if (ret) {
 807                 free_extent_buffer(root->node);
 808                 kfree(root);
 809                 return ERR_PTR(ret);
 810         }
 811         ret = btrfs_find_dead_roots(fs_info->tree_root,
 812                                     root->root_key.objectid, root);
 813         BUG_ON(ret);
 814
 815         return root;
 816 }
 817
 818 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 819                                       struct btrfs_key *location,
 820                                       const char *name, int namelen)
 821 {
 822         struct btrfs_root *root;
 823         int ret;
 824
 825         root = btrfs_read_fs_root_no_name(fs_info, location);
 826         if (!root)
 827                 return NULL;
 828
 829         if (root->in_sysfs)
 830                 return root;
 831
 832         ret = btrfs_set_root_name(root, name, namelen);
 833         if (ret) {
 834                 free_extent_buffer(root->node);
 835                 kfree(root);
 836                 return ERR_PTR(ret);
 837         }
 838
 839         ret = btrfs_sysfs_add_root(root);
 840         if (ret) {
 841                 free_extent_buffer(root->node);
 842                 kfree(root->name);
 843                 kfree(root);
 844                 return ERR_PTR(ret);
 845         }
 846         root->in_sysfs = 1;
 847         return root;
 848 }
 849 #if 0
 850 static int add_hasher(struct btrfs_fs_info *info, char *type) {
 851         struct btrfs_hasher *hasher;
 852
 853         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
 854         if (!hasher)
 855                 return -ENOMEM;
 856         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
 857         if (!hasher->hash_tfm) {
 858                 kfree(hasher);
 859                 return -EINVAL;
 860         }
 861         spin_lock(&info->hash_lock);
 862         list_add(&hasher->list, &info->hashers);
 863         spin_unlock(&info->hash_lock);
 864         return 0;
 865 }
 866 #endif
 867
 868 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
 869 {
 870         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
 871         int ret = 0;
 872         struct list_head *cur;
 873         struct btrfs_device *device;
 874         struct backing_dev_info *bdi;
 875
 876         list_for_each(cur, &info->fs_devices->devices) {
 877                 device = list_entry(cur, struct btrfs_device, dev_list);
 878                 bdi = blk_get_backing_dev_info(device->bdev);
 879                 if (bdi && bdi_congested(bdi, bdi_bits)) {
 880                         ret = 1;
 881                         break;
 882                 }
 883         }
 884         return ret;
 885 }
 886
 887 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
 888 {
 889         struct list_head *cur;
 890         struct btrfs_device *device;
 891         struct btrfs_fs_info *info;
 892
 893         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
 894         list_for_each(cur, &info->fs_devices->devices) {
 895                 device = list_entry(cur, struct btrfs_device, dev_list);
 896                 bdi = blk_get_backing_dev_info(device->bdev);
 897                 if (bdi->unplug_io_fn) {
 898                         bdi->unplug_io_fn(bdi, page);
 899                 }
 900         }
 901 }
 902
 903 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
 904 {
 905 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
 906         bdi_init(bdi);
 907 #endif
 908         bdi->ra_pages   = default_backing_dev_info.ra_pages * 4;
 909         bdi->state              = 0;
 910         bdi->capabilities       = default_backing_dev_info.capabilities;
 911         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
 912         bdi->unplug_io_data     = info;
 913         bdi->congested_fn       = btrfs_congested_fn;
 914         bdi->congested_data     = info;
 915         return 0;
 916 }
 917
 918 static int bio_ready_for_csum(struct bio *bio)
 919 {
 920         u64 length = 0;
 921         u64 buf_len = 0;
 922         u64 start = 0;
 923         struct page *page;
 924         struct extent_io_tree *io_tree = NULL;
 925         struct btrfs_fs_info *info = NULL;
 926         struct bio_vec *bvec;
 927         int i;
 928         int ret;
 929
 930         bio_for_each_segment(bvec, bio, i) {
 931                 page = bvec->bv_page;
 932                 if (page->private == EXTENT_PAGE_PRIVATE) {
 933                         length += bvec->bv_len;
 934                         continue;
 935                 }
 936                 if (!page->private) {
 937                         length += bvec->bv_len;
 938                         continue;
 939                 }
 940                 length = bvec->bv_len;
 941                 buf_len = page->private >> 2;
 942                 start = page_offset(page) + bvec->bv_offset;
 943                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
 944                 info = BTRFS_I(page->mapping->host)->root->fs_info;
 945         }
 946         /* are we fully contained in this bio? */
 947         if (buf_len <= length)
 948                 return 1;
 949
 950         ret = extent_range_uptodate(io_tree, start + length,
 951                                     start + buf_len - 1);
 952         if (ret == 1)
 953                 return ret;
 954         return ret;
 955 }
 956
 957 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 958 static void btrfs_end_io_csum(void *p)
 959 #else
 960 static void btrfs_end_io_csum(struct work_struct *work)
 961 #endif
 962 {
 963 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 964         struct btrfs_fs_info *fs_info = p;
 965 #else
 966         struct btrfs_fs_info *fs_info = container_of(work,
 967                                                      struct btrfs_fs_info,
 968                                                      end_io_work);
 969 #endif
 970         unsigned long flags;
 971         struct end_io_wq *end_io_wq;
 972         struct bio *bio;
 973         struct list_head *next;
 974         int error;
 975         int was_empty;
 976
 977         while(1) {
 978                 spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
 979                 if (list_empty(&fs_info->end_io_work_list)) {
 980                         spin_unlock_irqrestore(&fs_info->end_io_work_lock,
 981                                                flags);
 982                         return;
 983                 }
 984                 next = fs_info->end_io_work_list.next;
 985                 list_del(next);
 986                 spin_unlock_irqrestore(&fs_info->end_io_work_lock, flags);
 987
 988                 end_io_wq = list_entry(next, struct end_io_wq, list);
 989
 990                 bio = end_io_wq->bio;
 991                 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
 992                         spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
 993                         was_empty = list_empty(&fs_info->end_io_work_list);
 994                         list_add_tail(&end_io_wq->list,
 995                                       &fs_info->end_io_work_list);
 996                         spin_unlock_irqrestore(&fs_info->end_io_work_lock,
 997                                                flags);
 998                         if (was_empty)
 999                                 return;
1000                         continue;
1001                 }
1002                 error = end_io_wq->error;
1003                 bio->bi_private = end_io_wq->private;
1004                 bio->bi_end_io = end_io_wq->end_io;
1005                 kfree(end_io_wq);
1006 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1007                 bio_endio(bio, bio->bi_size, error);
1008 #else
1009                 bio_endio(bio, error);
1010 #endif
1011         }
1012 }
1013
1014 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1015 static void btrfs_async_submit_work(void *p)
1016 #else
1017 static void btrfs_async_submit_work(struct work_struct *work)
1018 #endif
1019 {
1020 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1021         struct btrfs_fs_info *fs_info = p;
1022 #else
1023         struct btrfs_fs_info *fs_info = container_of(work,
1024                                                      struct btrfs_fs_info,
1025                                                      async_submit_work);
1026 #endif
1027         struct async_submit_bio *async;
1028         struct list_head *next;
1029
1030         while(1) {
1031                 spin_lock(&fs_info->async_submit_work_lock);
1032                 if (list_empty(&fs_info->async_submit_work_list)) {
1033                         spin_unlock(&fs_info->async_submit_work_lock);
1034                         return;
1035                 }
1036                 next = fs_info->async_submit_work_list.next;
1037                 list_del(next);
1038                 spin_unlock(&fs_info->async_submit_work_lock);
1039
1040                 async = list_entry(next, struct async_submit_bio, list);
1041                 async->submit_bio_hook(async->inode, async->rw, async->bio,
1042                                        async->mirror_num);
1043                 kfree(async);
1044         }
1045 }
1046
1047 struct btrfs_root *open_ctree(struct super_block *sb,
1048                               struct btrfs_fs_devices *fs_devices)
1049 {
1050         u32 sectorsize;
1051         u32 nodesize;
1052         u32 leafsize;
1053         u32 blocksize;
1054         u32 stripesize;
1055         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
1056                                                  GFP_NOFS);
1057         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
1058                                                GFP_NOFS);
1059         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1060                                                 GFP_NOFS);
1061         struct btrfs_root *chunk_root = kmalloc(sizeof(struct btrfs_root),
1062                                                 GFP_NOFS);
1063         struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
1064                                               GFP_NOFS);
1065         int ret;
1066         int err = -EINVAL;
1067         struct btrfs_super_block *disk_super;
1068
1069         if (!extent_root || !tree_root || !fs_info) {
1070                 err = -ENOMEM;
1071                 goto fail;
1072         }
1073         end_io_workqueue = create_workqueue("btrfs-end-io");
1074         BUG_ON(!end_io_workqueue);
1075         async_submit_workqueue = create_workqueue("btrfs-async-submit");
1076
1077         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1078         INIT_LIST_HEAD(&fs_info->trans_list);
1079         INIT_LIST_HEAD(&fs_info->dead_roots);
1080         INIT_LIST_HEAD(&fs_info->hashers);
1081         INIT_LIST_HEAD(&fs_info->end_io_work_list);
1082         INIT_LIST_HEAD(&fs_info->async_submit_work_list);
1083         spin_lock_init(&fs_info->hash_lock);
1084         spin_lock_init(&fs_info->end_io_work_lock);
1085         spin_lock_init(&fs_info->async_submit_work_lock);
1086         spin_lock_init(&fs_info->delalloc_lock);
1087         spin_lock_init(&fs_info->new_trans_lock);
1088
1089         init_completion(&fs_info->kobj_unregister);
1090         sb_set_blocksize(sb, BTRFS_SUPER_INFO_SIZE);
1091         fs_info->tree_root = tree_root;
1092         fs_info->extent_root = extent_root;
1093         fs_info->chunk_root = chunk_root;
1094         fs_info->dev_root = dev_root;
1095         fs_info->fs_devices = fs_devices;
1096         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1097         INIT_LIST_HEAD(&fs_info->space_info);
1098         btrfs_mapping_init(&fs_info->mapping_tree);
1099         fs_info->sb = sb;
1100         fs_info->max_extent = (u64)-1;
1101         fs_info->max_inline = 8192 * 1024;
1102         setup_bdi(fs_info, &fs_info->bdi);
1103         fs_info->btree_inode = new_inode(sb);
1104         fs_info->btree_inode->i_ino = 1;
1105         fs_info->btree_inode->i_nlink = 1;
1106         fs_info->btree_inode->i_size = sb->s_bdev->bd_inode->i_size;
1107         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1108         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1109
1110         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1111                              fs_info->btree_inode->i_mapping,
1112                              GFP_NOFS);
1113         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1114                              GFP_NOFS);
1115
1116         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1117
1118         extent_io_tree_init(&fs_info->free_space_cache,
1119                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1120         extent_io_tree_init(&fs_info->block_group_cache,
1121                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1122         extent_io_tree_init(&fs_info->pinned_extents,
1123                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1124         extent_io_tree_init(&fs_info->pending_del,
1125                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1126         extent_io_tree_init(&fs_info->extent_ins,
1127                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1128         fs_info->do_barriers = 1;
1129
1130 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1131         INIT_WORK(&fs_info->end_io_work, btrfs_end_io_csum, fs_info);
1132         INIT_WORK(&fs_info->async_submit_work, btrfs_async_submit_work,
1133                   fs_info);
1134         INIT_WORK(&fs_info->trans_work, btrfs_transaction_cleaner, fs_info);
1135 #else
1136         INIT_WORK(&fs_info->end_io_work, btrfs_end_io_csum);
1137         INIT_WORK(&fs_info->async_submit_work, btrfs_async_submit_work);
1138         INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
1139 #endif
1140         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1141         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1142                sizeof(struct btrfs_key));
1143         insert_inode_hash(fs_info->btree_inode);
1144         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1145
1146         mutex_init(&fs_info->trans_mutex);
1147         mutex_init(&fs_info->fs_mutex);
1148
1149 #if 0
1150         ret = add_hasher(fs_info, "crc32c");
1151         if (ret) {
1152                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1153                 err = -ENOMEM;
1154                 goto fail_iput;
1155         }
1156 #endif
1157         __setup_root(4096, 4096, 4096, 4096, tree_root,
1158                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1159
1160         fs_info->sb_buffer = read_tree_block(tree_root,
1161                                              BTRFS_SUPER_INFO_OFFSET,
1162                                              4096);
1163
1164         if (!fs_info->sb_buffer)
1165                 goto fail_iput;
1166
1167         read_extent_buffer(fs_info->sb_buffer, &fs_info->super_copy, 0,
1168                            sizeof(fs_info->super_copy));
1169
1170         read_extent_buffer(fs_info->sb_buffer, fs_info->fsid,
1171                            (unsigned long)btrfs_super_fsid(fs_info->sb_buffer),
1172                            BTRFS_FSID_SIZE);
1173
1174         disk_super = &fs_info->super_copy;
1175         if (!btrfs_super_root(disk_super))
1176                 goto fail_sb_buffer;
1177
1178         if (btrfs_super_num_devices(disk_super) != fs_devices->num_devices) {
1179                 printk("Btrfs: wanted %llu devices, but found %llu\n",
1180                        (unsigned long long)btrfs_super_num_devices(disk_super),
1181                        (unsigned long long)fs_devices->num_devices);
1182                 goto fail_sb_buffer;
1183         }
1184         nodesize = btrfs_super_nodesize(disk_super);
1185         leafsize = btrfs_super_leafsize(disk_super);
1186         sectorsize = btrfs_super_sectorsize(disk_super);
1187         stripesize = btrfs_super_stripesize(disk_super);
1188         tree_root->nodesize = nodesize;
1189         tree_root->leafsize = leafsize;
1190         tree_root->sectorsize = sectorsize;
1191         tree_root->stripesize = stripesize;
1192         sb_set_blocksize(sb, sectorsize);
1193
1194         i_size_write(fs_info->btree_inode,
1195                      btrfs_super_total_bytes(disk_super));
1196
1197         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1198                     sizeof(disk_super->magic))) {
1199                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1200                 goto fail_sb_buffer;
1201         }
1202
1203         mutex_lock(&fs_info->fs_mutex);
1204
1205         ret = btrfs_read_sys_array(tree_root);
1206         BUG_ON(ret);
1207
1208         blocksize = btrfs_level_size(tree_root,
1209                                      btrfs_super_chunk_root_level(disk_super));
1210
1211         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1212                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1213
1214         chunk_root->node = read_tree_block(chunk_root,
1215                                            btrfs_super_chunk_root(disk_super),
1216                                            blocksize);
1217         BUG_ON(!chunk_root->node);
1218
1219         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1220                  (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1221                  BTRFS_UUID_SIZE);
1222
1223         ret = btrfs_read_chunk_tree(chunk_root);
1224         BUG_ON(ret);
1225
1226         blocksize = btrfs_level_size(tree_root,
1227                                      btrfs_super_root_level(disk_super));
1228
1229
1230         tree_root->node = read_tree_block(tree_root,
1231                                           btrfs_super_root(disk_super),
1232                                           blocksize);
1233         if (!tree_root->node)
1234                 goto fail_sb_buffer;
1235
1236
1237         ret = find_and_setup_root(tree_root, fs_info,
1238                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1239         if (ret)
1240                 goto fail_tree_root;
1241         extent_root->track_dirty = 1;
1242
1243         ret = find_and_setup_root(tree_root, fs_info,
1244                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1245         dev_root->track_dirty = 1;
1246
1247         if (ret)
1248                 goto fail_extent_root;
1249
1250         btrfs_read_block_groups(extent_root);
1251
1252         fs_info->generation = btrfs_super_generation(disk_super) + 1;
1253         fs_info->data_alloc_profile = (u64)-1;
1254         fs_info->metadata_alloc_profile = (u64)-1;
1255         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1256
1257         mutex_unlock(&fs_info->fs_mutex);
1258         return tree_root;
1259
1260 fail_extent_root:
1261         free_extent_buffer(extent_root->node);
1262 fail_tree_root:
1263         mutex_unlock(&fs_info->fs_mutex);
1264         free_extent_buffer(tree_root->node);
1265 fail_sb_buffer:
1266         free_extent_buffer(fs_info->sb_buffer);
1267         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1268 fail_iput:
1269         iput(fs_info->btree_inode);
1270 fail:
1271         close_all_devices(fs_info);
1272         kfree(extent_root);
1273         kfree(tree_root);
1274 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
1275         bdi_destroy(&fs_info->bdi);
1276 #endif
1277         kfree(fs_info);
1278         return ERR_PTR(err);
1279 }
1280
1281 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1282 {
1283         char b[BDEVNAME_SIZE];
1284
1285         if (uptodate) {
1286                 set_buffer_uptodate(bh);
1287         } else {
1288                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1289                         printk(KERN_WARNING "lost page write due to "
1290                                         "I/O error on %s\n",
1291                                        bdevname(bh->b_bdev, b));
1292                 }
1293                 set_buffer_write_io_error(bh);
1294                 clear_buffer_uptodate(bh);
1295         }
1296         unlock_buffer(bh);
1297         put_bh(bh);
1298 }
1299
1300 int write_all_supers(struct btrfs_root *root)
1301 {
1302         struct list_head *cur;
1303         struct list_head *head = &root->fs_info->fs_devices->devices;
1304         struct btrfs_device *dev;
1305         struct extent_buffer *sb;
1306         struct btrfs_dev_item *dev_item;
1307         struct buffer_head *bh;
1308         int ret;
1309         int do_barriers;
1310
1311         do_barriers = !btrfs_test_opt(root, NOBARRIER);
1312
1313         sb = root->fs_info->sb_buffer;
1314         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
1315                                                       dev_item);
1316         list_for_each(cur, head) {
1317                 dev = list_entry(cur, struct btrfs_device, dev_list);
1318                 btrfs_set_device_type(sb, dev_item, dev->type);
1319                 btrfs_set_device_id(sb, dev_item, dev->devid);
1320                 btrfs_set_device_total_bytes(sb, dev_item, dev->total_bytes);
1321                 btrfs_set_device_bytes_used(sb, dev_item, dev->bytes_used);
1322                 btrfs_set_device_io_align(sb, dev_item, dev->io_align);
1323                 btrfs_set_device_io_width(sb, dev_item, dev->io_width);
1324                 btrfs_set_device_sector_size(sb, dev_item, dev->sector_size);
1325                 write_extent_buffer(sb, dev->uuid,
1326                                     (unsigned long)btrfs_device_uuid(dev_item),
1327                                     BTRFS_UUID_SIZE);
1328
1329                 btrfs_set_header_flag(sb, BTRFS_HEADER_FLAG_WRITTEN);
1330                 csum_tree_block(root, sb, 0);
1331
1332                 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET /
1333                               root->fs_info->sb->s_blocksize,
1334                               BTRFS_SUPER_INFO_SIZE);
1335
1336                 read_extent_buffer(sb, bh->b_data, 0, BTRFS_SUPER_INFO_SIZE);
1337                 dev->pending_io = bh;
1338
1339                 get_bh(bh);
1340                 set_buffer_uptodate(bh);
1341                 lock_buffer(bh);
1342                 bh->b_end_io = btrfs_end_buffer_write_sync;
1343
1344                 if (do_barriers && dev->barriers) {
1345                         ret = submit_bh(WRITE_BARRIER, bh);
1346                         if (ret == -EOPNOTSUPP) {
1347                                 printk("btrfs: disabling barriers on dev %s\n",
1348                                        dev->name);
1349                                 set_buffer_uptodate(bh);
1350                                 dev->barriers = 0;
1351                                 get_bh(bh);
1352                                 lock_buffer(bh);
1353                                 ret = submit_bh(WRITE, bh);
1354                         }
1355                 } else {
1356                         ret = submit_bh(WRITE, bh);
1357                 }
1358                 BUG_ON(ret);
1359         }
1360
1361         list_for_each(cur, head) {
1362                 dev = list_entry(cur, struct btrfs_device, dev_list);
1363                 BUG_ON(!dev->pending_io);
1364                 bh = dev->pending_io;
1365                 wait_on_buffer(bh);
1366                 if (!buffer_uptodate(dev->pending_io)) {
1367                         if (do_barriers && dev->barriers) {
1368                                 printk("btrfs: disabling barriers on dev %s\n",
1369                                        dev->name);
1370                                 set_buffer_uptodate(bh);
1371                                 get_bh(bh);
1372                                 lock_buffer(bh);
1373                                 dev->barriers = 0;
1374                                 ret = submit_bh(WRITE, bh);
1375                                 BUG_ON(ret);
1376                                 wait_on_buffer(bh);
1377                                 BUG_ON(!buffer_uptodate(bh));
1378                         } else {
1379                                 BUG();
1380                         }
1381
1382                 }
1383                 dev->pending_io = NULL;
1384                 brelse(bh);
1385         }
1386         return 0;
1387 }
1388
1389 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1390                       *root)
1391 {
1392         int ret;
1393
1394         ret = write_all_supers(root);
1395 #if 0
1396         if (!btrfs_test_opt(root, NOBARRIER))
1397                 blkdev_issue_flush(sb->s_bdev, NULL);
1398         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, super);
1399         ret = sync_page_range_nolock(btree_inode, btree_inode->i_mapping,
1400                                      super->start, super->len);
1401         if (!btrfs_test_opt(root, NOBARRIER))
1402                 blkdev_issue_flush(sb->s_bdev, NULL);
1403 #endif
1404         return ret;
1405 }
1406
1407 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1408 {
1409         radix_tree_delete(&fs_info->fs_roots_radix,
1410                           (unsigned long)root->root_key.objectid);
1411         if (root->in_sysfs)
1412                 btrfs_sysfs_del_root(root);
1413         if (root->inode)
1414                 iput(root->inode);
1415         if (root->node)
1416                 free_extent_buffer(root->node);
1417         if (root->commit_root)
1418                 free_extent_buffer(root->commit_root);
1419         if (root->name)
1420                 kfree(root->name);
1421         kfree(root);
1422         return 0;
1423 }
1424
1425 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1426 {
1427         int ret;
1428         struct btrfs_root *gang[8];
1429         int i;
1430
1431         while(1) {
1432                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1433                                              (void **)gang, 0,
1434                                              ARRAY_SIZE(gang));
1435                 if (!ret)
1436                         break;
1437                 for (i = 0; i < ret; i++)
1438                         btrfs_free_fs_root(fs_info, gang[i]);
1439         }
1440         return 0;
1441 }
1442
1443 int close_ctree(struct btrfs_root *root)
1444 {
1445         int ret;
1446         struct btrfs_trans_handle *trans;
1447         struct btrfs_fs_info *fs_info = root->fs_info;
1448
1449         fs_info->closing = 1;
1450         btrfs_transaction_flush_work(root);
1451         mutex_lock(&fs_info->fs_mutex);
1452         btrfs_defrag_dirty_roots(root->fs_info);
1453         trans = btrfs_start_transaction(root, 1);
1454         ret = btrfs_commit_transaction(trans, root);
1455         /* run commit again to  drop the original snapshot */
1456         trans = btrfs_start_transaction(root, 1);
1457         btrfs_commit_transaction(trans, root);
1458         ret = btrfs_write_and_wait_transaction(NULL, root);
1459         BUG_ON(ret);
1460         write_ctree_super(NULL, root);
1461         mutex_unlock(&fs_info->fs_mutex);
1462
1463         if (fs_info->delalloc_bytes) {
1464                 printk("btrfs: at unmount delalloc count %Lu\n",
1465                        fs_info->delalloc_bytes);
1466         }
1467         if (fs_info->extent_root->node)
1468                 free_extent_buffer(fs_info->extent_root->node);
1469
1470         if (fs_info->tree_root->node)
1471                 free_extent_buffer(fs_info->tree_root->node);
1472
1473         if (root->fs_info->chunk_root->node);
1474                 free_extent_buffer(root->fs_info->chunk_root->node);
1475
1476         if (root->fs_info->dev_root->node);
1477                 free_extent_buffer(root->fs_info->dev_root->node);
1478
1479         free_extent_buffer(fs_info->sb_buffer);
1480
1481         btrfs_free_block_groups(root->fs_info);
1482         del_fs_roots(fs_info);
1483
1484         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1485
1486         extent_io_tree_empty_lru(&fs_info->free_space_cache);
1487         extent_io_tree_empty_lru(&fs_info->block_group_cache);
1488         extent_io_tree_empty_lru(&fs_info->pinned_extents);
1489         extent_io_tree_empty_lru(&fs_info->pending_del);
1490         extent_io_tree_empty_lru(&fs_info->extent_ins);
1491         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1492
1493         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1494         flush_workqueue(end_io_workqueue);
1495         destroy_workqueue(end_io_workqueue);
1496
1497         flush_workqueue(async_submit_workqueue);
1498         destroy_workqueue(async_submit_workqueue);
1499
1500         iput(fs_info->btree_inode);
1501 #if 0
1502         while(!list_empty(&fs_info->hashers)) {
1503                 struct btrfs_hasher *hasher;
1504                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1505                                     hashers);
1506                 list_del(&hasher->hashers);
1507                 crypto_free_hash(&fs_info->hash_tfm);
1508                 kfree(hasher);
1509         }
1510 #endif
1511         close_all_devices(fs_info);
1512         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1513
1514 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
1515         bdi_destroy(&fs_info->bdi);
1516 #endif
1517
1518         kfree(fs_info->extent_root);
1519         kfree(fs_info->tree_root);
1520         kfree(fs_info->chunk_root);
1521         kfree(fs_info->dev_root);
1522         return 0;
1523 }
1524
1525 int btrfs_buffer_uptodate(struct extent_buffer *buf)
1526 {
1527         struct inode *btree_inode = buf->first_page->mapping->host;
1528         return extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1529 }
1530
1531 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1532 {
1533         struct inode *btree_inode = buf->first_page->mapping->host;
1534         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1535                                           buf);
1536 }
1537
1538 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1539 {
1540         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1541         u64 transid = btrfs_header_generation(buf);
1542         struct inode *btree_inode = root->fs_info->btree_inode;
1543
1544         if (transid != root->fs_info->generation) {
1545                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1546                         (unsigned long long)buf->start,
1547                         transid, root->fs_info->generation);
1548                 WARN_ON(1);
1549         }
1550         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1551 }
1552
1553 void btrfs_throttle(struct btrfs_root *root)
1554 {
1555         struct backing_dev_info *bdi;
1556
1557         bdi = root->fs_info->sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
1558         if (root->fs_info->throttles && bdi_write_congested(bdi)) {
1559 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
1560                 congestion_wait(WRITE, HZ/20);
1561 #else
1562                 blk_congestion_wait(WRITE, HZ/20);
1563 #endif
1564         }
1565 }
1566
1567 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1568 {
1569         balance_dirty_pages_ratelimited_nr(
1570                                    root->fs_info->btree_inode->i_mapping, 1);
1571 }
1572
1573 void btrfs_set_buffer_defrag(struct extent_buffer *buf)
1574 {
1575         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1576         struct inode *btree_inode = root->fs_info->btree_inode;
1577         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1578                         buf->start + buf->len - 1, EXTENT_DEFRAG, GFP_NOFS);
1579 }
1580
1581 void btrfs_set_buffer_defrag_done(struct extent_buffer *buf)
1582 {
1583         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1584         struct inode *btree_inode = root->fs_info->btree_inode;
1585         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1586                         buf->start + buf->len - 1, EXTENT_DEFRAG_DONE,
1587                         GFP_NOFS);
1588 }
1589
1590 int btrfs_buffer_defrag(struct extent_buffer *buf)
1591 {
1592         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1593         struct inode *btree_inode = root->fs_info->btree_inode;
1594         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1595                      buf->start, buf->start + buf->len - 1, EXTENT_DEFRAG, 0);
1596 }
1597
1598 int btrfs_buffer_defrag_done(struct extent_buffer *buf)
1599 {
1600         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1601         struct inode *btree_inode = root->fs_info->btree_inode;
1602         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1603                      buf->start, buf->start + buf->len - 1,
1604                      EXTENT_DEFRAG_DONE, 0);
1605 }
1606
1607 int btrfs_clear_buffer_defrag_done(struct extent_buffer *buf)
1608 {
1609         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1610         struct inode *btree_inode = root->fs_info->btree_inode;
1611         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1612                      buf->start, buf->start + buf->len - 1,
1613                      EXTENT_DEFRAG_DONE, GFP_NOFS);
1614 }
1615
1616 int btrfs_clear_buffer_defrag(struct extent_buffer *buf)
1617 {
1618         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1619         struct inode *btree_inode = root->fs_info->btree_inode;
1620         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1621                      buf->start, buf->start + buf->len - 1,
1622                      EXTENT_DEFRAG, GFP_NOFS);
1623 }
1624
1625 int btrfs_read_buffer(struct extent_buffer *buf)
1626 {
1627         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1628         int ret;
1629         ret = btree_read_extent_buffer_pages(root, buf, 0);
1630         if (ret == 0) {
1631                 buf->flags |= EXTENT_UPTODATE;
1632         }
1633         return ret;
1634 }
1635
1636 static struct extent_io_ops btree_extent_io_ops = {
1637         .writepage_io_hook = btree_writepage_io_hook,
1638         .readpage_end_io_hook = btree_readpage_end_io_hook,
1639         .submit_bio_hook = btree_submit_bio_hook,
1640         /* note we're sharing with inode.c for the merge bio hook */
1641         .merge_bio_hook = btrfs_merge_bio_hook,
1642 };