fs/btrfs/disk-io.c

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