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