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