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 #include "ref-cache.h"
  44
  45 #if 0
  46 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
  47 {
  48         if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
  49                 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
  50                        (unsigned long long)extent_buffer_blocknr(buf),
  51                        (unsigned long long)btrfs_header_blocknr(buf));
  52                 return 1;
  53         }
  54         return 0;
  55 }
  56 #endif
  57
  58 static struct extent_io_ops btree_extent_io_ops;
  59 static void end_workqueue_fn(struct btrfs_work *work);
  60
  61 struct end_io_wq {
  62         struct bio *bio;
  63         bio_end_io_t *end_io;
  64         void *private;
  65         struct btrfs_fs_info *info;
  66         int error;
  67         int metadata;
  68         struct list_head list;
  69         struct btrfs_work work;
  70 };
  71
  72 struct async_submit_bio {
  73         struct inode *inode;
  74         struct bio *bio;
  75         struct list_head list;
  76         extent_submit_bio_hook_t *submit_bio_hook;
  77         int rw;
  78         int mirror_num;
  79         struct btrfs_work work;
  80 };
  81
  82 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
  83                                     size_t page_offset, u64 start, u64 len,
  84                                     int create)
  85 {
  86         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
  87         struct extent_map *em;
  88         int ret;
  89
  90         spin_lock(&em_tree->lock);
  91         em = lookup_extent_mapping(em_tree, start, len);
  92         if (em) {
  93                 em->bdev =
  94                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
  95                 spin_unlock(&em_tree->lock);
  96                 goto out;
  97         }
  98         spin_unlock(&em_tree->lock);
  99
 100         em = alloc_extent_map(GFP_NOFS);
 101         if (!em) {
 102                 em = ERR_PTR(-ENOMEM);
 103                 goto out;
 104         }
 105         em->start = 0;
 106         em->len = (u64)-1;
 107         em->block_start = 0;
 108         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 109
 110         spin_lock(&em_tree->lock);
 111         ret = add_extent_mapping(em_tree, em);
 112         if (ret == -EEXIST) {
 113                 u64 failed_start = em->start;
 114                 u64 failed_len = em->len;
 115
 116                 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
 117                        em->start, em->len, em->block_start);
 118                 free_extent_map(em);
 119                 em = lookup_extent_mapping(em_tree, start, len);
 120                 if (em) {
 121                         printk("after failing, found %Lu %Lu %Lu\n",
 122                                em->start, em->len, em->block_start);
 123                         ret = 0;
 124                 } else {
 125                         em = lookup_extent_mapping(em_tree, failed_start,
 126                                                    failed_len);
 127                         if (em) {
 128                                 printk("double failure lookup gives us "
 129                                        "%Lu %Lu -> %Lu\n", em->start,
 130                                        em->len, em->block_start);
 131                                 free_extent_map(em);
 132                         }
 133                         ret = -EIO;
 134                 }
 135         } else if (ret) {
 136                 free_extent_map(em);
 137                 em = NULL;
 138         }
 139         spin_unlock(&em_tree->lock);
 140
 141         if (ret)
 142                 em = ERR_PTR(ret);
 143 out:
 144         return em;
 145 }
 146
 147 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 148 {
 149         return btrfs_crc32c(seed, data, len);
 150 }
 151
 152 void btrfs_csum_final(u32 crc, char *result)
 153 {
 154         *(__le32 *)result = ~cpu_to_le32(crc);
 155 }
 156
 157 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 158                            int verify)
 159 {
 160         char result[BTRFS_CRC32_SIZE];
 161         unsigned long len;
 162         unsigned long cur_len;
 163         unsigned long offset = BTRFS_CSUM_SIZE;
 164         char *map_token = NULL;
 165         char *kaddr;
 166         unsigned long map_start;
 167         unsigned long map_len;
 168         int err;
 169         u32 crc = ~(u32)0;
 170
 171         len = buf->len - offset;
 172         while(len > 0) {
 173                 err = map_private_extent_buffer(buf, offset, 32,
 174                                         &map_token, &kaddr,
 175                                         &map_start, &map_len, KM_USER0);
 176                 if (err) {
 177                         printk("failed to map extent buffer! %lu\n",
 178                                offset);
 179                         return 1;
 180                 }
 181                 cur_len = min(len, map_len - (offset - map_start));
 182                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 183                                       crc, cur_len);
 184                 len -= cur_len;
 185                 offset += cur_len;
 186                 unmap_extent_buffer(buf, map_token, KM_USER0);
 187         }
 188         btrfs_csum_final(crc, result);
 189
 190         if (verify) {
 191                 /* FIXME, this is not good */
 192                 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
 193                         u32 val;
 194                         u32 found = 0;
 195                         memcpy(&found, result, BTRFS_CRC32_SIZE);
 196
 197                         read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
 198                         printk("btrfs: %s checksum verify failed on %llu "
 199                                "wanted %X found %X level %d\n",
 200                                root->fs_info->sb->s_id,
 201                                buf->start, val, found, btrfs_header_level(buf));
 202                         return 1;
 203                 }
 204         } else {
 205                 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
 206         }
 207         return 0;
 208 }
 209
 210 static int verify_parent_transid(struct extent_io_tree *io_tree,
 211                                  struct extent_buffer *eb, u64 parent_transid)
 212 {
 213         int ret;
 214
 215         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 216                 return 0;
 217
 218         lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
 219         if (extent_buffer_uptodate(io_tree, eb) &&
 220             btrfs_header_generation(eb) == parent_transid) {
 221                 ret = 0;
 222                 goto out;
 223         }
 224         printk("parent transid verify failed on %llu wanted %llu found %llu\n",
 225                (unsigned long long)eb->start,
 226                (unsigned long long)parent_transid,
 227                (unsigned long long)btrfs_header_generation(eb));
 228         ret = 1;
 229         clear_extent_buffer_uptodate(io_tree, eb);
 230 out:
 231         unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
 232                       GFP_NOFS);
 233         return ret;
 234
 235 }
 236
 237 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 238                                           struct extent_buffer *eb,
 239                                           u64 start, u64 parent_transid)
 240 {
 241         struct extent_io_tree *io_tree;
 242         int ret;
 243         int num_copies = 0;
 244         int mirror_num = 0;
 245
 246         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 247         while (1) {
 248                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 249                                                btree_get_extent, mirror_num);
 250                 if (!ret &&
 251                     !verify_parent_transid(io_tree, eb, parent_transid))
 252                         return ret;
 253
 254                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 255                                               eb->start, eb->len);
 256                 if (num_copies == 1)
 257                         return ret;
 258
 259                 mirror_num++;
 260                 if (mirror_num > num_copies)
 261                         return ret;
 262         }
 263         return -EIO;
 264 }
 265
 266 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 267 {
 268         struct extent_io_tree *tree;
 269         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 270         u64 found_start;
 271         int found_level;
 272         unsigned long len;
 273         struct extent_buffer *eb;
 274         int ret;
 275
 276         tree = &BTRFS_I(page->mapping->host)->io_tree;
 277
 278         if (page->private == EXTENT_PAGE_PRIVATE)
 279                 goto out;
 280         if (!page->private)
 281                 goto out;
 282         len = page->private >> 2;
 283         if (len == 0) {
 284                 WARN_ON(1);
 285         }
 286         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 287         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 288                                              btrfs_header_generation(eb));
 289         BUG_ON(ret);
 290         found_start = btrfs_header_bytenr(eb);
 291         if (found_start != start) {
 292                 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
 293                        start, found_start, len);
 294                 WARN_ON(1);
 295                 goto err;
 296         }
 297         if (eb->first_page != page) {
 298                 printk("bad first page %lu %lu\n", eb->first_page->index,
 299                        page->index);
 300                 WARN_ON(1);
 301                 goto err;
 302         }
 303         if (!PageUptodate(page)) {
 304                 printk("csum not up to date page %lu\n", page->index);
 305                 WARN_ON(1);
 306                 goto err;
 307         }
 308         found_level = btrfs_header_level(eb);
 309         spin_lock(&root->fs_info->hash_lock);
 310         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
 311         spin_unlock(&root->fs_info->hash_lock);
 312         csum_tree_block(root, eb, 0);
 313 err:
 314         free_extent_buffer(eb);
 315 out:
 316         return 0;
 317 }
 318
 319 static int btree_writepage_io_hook(struct page *page, u64 start, u64 end)
 320 {
 321         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 322
 323         csum_dirty_buffer(root, page);
 324         return 0;
 325 }
 326
 327 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 328                                struct extent_state *state)
 329 {
 330         struct extent_io_tree *tree;
 331         u64 found_start;
 332         int found_level;
 333         unsigned long len;
 334         struct extent_buffer *eb;
 335         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 336         int ret = 0;
 337
 338         tree = &BTRFS_I(page->mapping->host)->io_tree;
 339         if (page->private == EXTENT_PAGE_PRIVATE)
 340                 goto out;
 341         if (!page->private)
 342                 goto out;
 343         len = page->private >> 2;
 344         if (len == 0) {
 345                 WARN_ON(1);
 346         }
 347         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 348
 349         found_start = btrfs_header_bytenr(eb);
 350         if (found_start != start) {
 351                 ret = -EIO;
 352                 goto err;
 353         }
 354         if (eb->first_page != page) {
 355                 printk("bad first page %lu %lu\n", eb->first_page->index,
 356                        page->index);
 357                 WARN_ON(1);
 358                 ret = -EIO;
 359                 goto err;
 360         }
 361         if (memcmp_extent_buffer(eb, root->fs_info->fsid,
 362                                  (unsigned long)btrfs_header_fsid(eb),
 363                                  BTRFS_FSID_SIZE)) {
 364                 printk("bad fsid on block %Lu\n", eb->start);
 365                 ret = -EIO;
 366                 goto err;
 367         }
 368         found_level = btrfs_header_level(eb);
 369
 370         ret = csum_tree_block(root, eb, 1);
 371         if (ret)
 372                 ret = -EIO;
 373
 374         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 375         end = eb->start + end - 1;
 376 err:
 377         free_extent_buffer(eb);
 378 out:
 379         return ret;
 380 }
 381
 382 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
 383 static void end_workqueue_bio(struct bio *bio, int err)
 384 #else
 385 static int end_workqueue_bio(struct bio *bio,
 386                                    unsigned int bytes_done, int err)
 387 #endif
 388 {
 389         struct end_io_wq *end_io_wq = bio->bi_private;
 390         struct btrfs_fs_info *fs_info;
 391
 392 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 393         if (bio->bi_size)
 394                 return 1;
 395 #endif
 396
 397         fs_info = end_io_wq->info;
 398         end_io_wq->error = err;
 399         end_io_wq->work.func = end_workqueue_fn;
 400         end_io_wq->work.flags = 0;
 401         if (bio->bi_rw & (1 << BIO_RW))
 402                 btrfs_queue_worker(&fs_info->endio_write_workers,
 403                                    &end_io_wq->work);
 404         else
 405                 btrfs_queue_worker(&fs_info->endio_workers, &end_io_wq->work);
 406
 407 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 408         return 0;
 409 #endif
 410 }
 411
 412 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 413                         int metadata)
 414 {
 415         struct end_io_wq *end_io_wq;
 416         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 417         if (!end_io_wq)
 418                 return -ENOMEM;
 419
 420         end_io_wq->private = bio->bi_private;
 421         end_io_wq->end_io = bio->bi_end_io;
 422         end_io_wq->info = info;
 423         end_io_wq->error = 0;
 424         end_io_wq->bio = bio;
 425         end_io_wq->metadata = metadata;
 426
 427         bio->bi_private = end_io_wq;
 428         bio->bi_end_io = end_workqueue_bio;
 429         return 0;
 430 }
 431
 432 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
 433 {
 434         int limit = 256 * info->fs_devices->open_devices;
 435
 436         if (iodone)
 437                 limit = (limit * 3) / 2;
 438         if (atomic_read(&info->nr_async_submits) > limit)
 439                 return 1;
 440
 441         return atomic_read(&info->nr_async_bios) > limit;
 442 }
 443
 444 static void run_one_async_submit(struct btrfs_work *work)
 445 {
 446         struct btrfs_fs_info *fs_info;
 447         struct async_submit_bio *async;
 448
 449         async = container_of(work, struct  async_submit_bio, work);
 450         fs_info = BTRFS_I(async->inode)->root->fs_info;
 451         atomic_dec(&fs_info->nr_async_submits);
 452
 453         if ((async->bio->bi_rw & (1 << BIO_RW)) &&
 454             !btrfs_congested_async(fs_info, 1)) {
 455                 clear_bdi_congested(&fs_info->bdi, WRITE);
 456         }
 457         async->submit_bio_hook(async->inode, async->rw, async->bio,
 458                                async->mirror_num);
 459         kfree(async);
 460 }
 461
 462 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 463                         int rw, struct bio *bio, int mirror_num,
 464                         extent_submit_bio_hook_t *submit_bio_hook)
 465 {
 466         struct async_submit_bio *async;
 467
 468         async = kmalloc(sizeof(*async), GFP_NOFS);
 469         if (!async)
 470                 return -ENOMEM;
 471
 472         async->inode = inode;
 473         async->rw = rw;
 474         async->bio = bio;
 475         async->mirror_num = mirror_num;
 476         async->submit_bio_hook = submit_bio_hook;
 477         async->work.func = run_one_async_submit;
 478         async->work.flags = 0;
 479         atomic_inc(&fs_info->nr_async_submits);
 480         btrfs_queue_worker(&fs_info->workers, &async->work);
 481         return 0;
 482 }
 483
 484 static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 485                                  int mirror_num)
 486 {
 487         struct btrfs_root *root = BTRFS_I(inode)->root;
 488         u64 offset;
 489         int ret;
 490
 491         offset = bio->bi_sector << 9;
 492
 493         /*
 494          * when we're called for a write, we're already in the async
 495          * submission context.  Just jump ingo btrfs_map_bio
 496          */
 497         if (rw & (1 << BIO_RW)) {
 498                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 499                                      mirror_num, 0);
 500         }
 501
 502         /*
 503          * called for a read, do the setup so that checksum validation
 504          * can happen in the async kernel threads
 505          */
 506         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
 507         BUG_ON(ret);
 508
 509         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 510 }
 511
 512 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 513                                  int mirror_num)
 514 {
 515         /*
 516          * kthread helpers are used to submit writes so that checksumming
 517          * can happen in parallel across all CPUs
 518          */
 519         if (!(rw & (1 << BIO_RW))) {
 520                 return __btree_submit_bio_hook(inode, rw, bio, mirror_num);
 521         }
 522         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 523                                    inode, rw, bio, mirror_num,
 524                                    __btree_submit_bio_hook);
 525 }
 526
 527 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 528 {
 529         struct extent_io_tree *tree;
 530         tree = &BTRFS_I(page->mapping->host)->io_tree;
 531         return extent_write_full_page(tree, page, btree_get_extent, wbc);
 532 }
 533
 534 static int btree_writepages(struct address_space *mapping,
 535                             struct writeback_control *wbc)
 536 {
 537         struct extent_io_tree *tree;
 538         tree = &BTRFS_I(mapping->host)->io_tree;
 539         if (wbc->sync_mode == WB_SYNC_NONE) {
 540                 u64 num_dirty;
 541                 u64 start = 0;
 542                 unsigned long thresh = 96 * 1024 * 1024;
 543
 544                 if (wbc->for_kupdate)
 545                         return 0;
 546
 547                 if (current_is_pdflush()) {
 548                         thresh = 96 * 1024 * 1024;
 549                 } else {
 550                         thresh = 8 * 1024 * 1024;
 551                 }
 552                 num_dirty = count_range_bits(tree, &start, (u64)-1,
 553                                              thresh, EXTENT_DIRTY);
 554                 if (num_dirty < thresh) {
 555                         return 0;
 556                 }
 557         }
 558         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 559 }
 560
 561 int btree_readpage(struct file *file, struct page *page)
 562 {
 563         struct extent_io_tree *tree;
 564         tree = &BTRFS_I(page->mapping->host)->io_tree;
 565         return extent_read_full_page(tree, page, btree_get_extent);
 566 }
 567
 568 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 569 {
 570         struct extent_io_tree *tree;
 571         struct extent_map_tree *map;
 572         int ret;
 573
 574         tree = &BTRFS_I(page->mapping->host)->io_tree;
 575         map = &BTRFS_I(page->mapping->host)->extent_tree;
 576
 577         ret = try_release_extent_state(map, tree, page, gfp_flags);
 578         if (!ret) {
 579                 return 0;
 580         }
 581
 582         ret = try_release_extent_buffer(tree, page);
 583         if (ret == 1) {
 584                 ClearPagePrivate(page);
 585                 set_page_private(page, 0);
 586                 page_cache_release(page);
 587         }
 588
 589         return ret;
 590 }
 591
 592 static void btree_invalidatepage(struct page *page, unsigned long offset)
 593 {
 594         struct extent_io_tree *tree;
 595         tree = &BTRFS_I(page->mapping->host)->io_tree;
 596         extent_invalidatepage(tree, page, offset);
 597         btree_releasepage(page, GFP_NOFS);
 598         if (PagePrivate(page)) {
 599                 printk("warning page private not zero on page %Lu\n",
 600                        page_offset(page));
 601                 ClearPagePrivate(page);
 602                 set_page_private(page, 0);
 603                 page_cache_release(page);
 604         }
 605 }
 606
 607 #if 0
 608 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 609 {
 610         struct buffer_head *bh;
 611         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 612         struct buffer_head *head;
 613         if (!page_has_buffers(page)) {
 614                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 615                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 616         }
 617         head = page_buffers(page);
 618         bh = head;
 619         do {
 620                 if (buffer_dirty(bh))
 621                         csum_tree_block(root, bh, 0);
 622                 bh = bh->b_this_page;
 623         } while (bh != head);
 624         return block_write_full_page(page, btree_get_block, wbc);
 625 }
 626 #endif
 627
 628 static struct address_space_operations btree_aops = {
 629         .readpage       = btree_readpage,
 630         .writepage      = btree_writepage,
 631         .writepages     = btree_writepages,
 632         .releasepage    = btree_releasepage,
 633         .invalidatepage = btree_invalidatepage,
 634         .sync_page      = block_sync_page,
 635 };
 636
 637 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 638                          u64 parent_transid)
 639 {
 640         struct extent_buffer *buf = NULL;
 641         struct inode *btree_inode = root->fs_info->btree_inode;
 642         int ret = 0;
 643
 644         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 645         if (!buf)
 646                 return 0;
 647         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 648                                  buf, 0, 0, btree_get_extent, 0);
 649         free_extent_buffer(buf);
 650         return ret;
 651 }
 652
 653 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 654                                             u64 bytenr, u32 blocksize)
 655 {
 656         struct inode *btree_inode = root->fs_info->btree_inode;
 657         struct extent_buffer *eb;
 658         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 659                                 bytenr, blocksize, GFP_NOFS);
 660         return eb;
 661 }
 662
 663 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 664                                                  u64 bytenr, u32 blocksize)
 665 {
 666         struct inode *btree_inode = root->fs_info->btree_inode;
 667         struct extent_buffer *eb;
 668
 669         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 670                                  bytenr, blocksize, NULL, GFP_NOFS);
 671         return eb;
 672 }
 673
 674
 675 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 676                                       u32 blocksize, u64 parent_transid)
 677 {
 678         struct extent_buffer *buf = NULL;
 679         struct inode *btree_inode = root->fs_info->btree_inode;
 680         struct extent_io_tree *io_tree;
 681         int ret;
 682
 683         io_tree = &BTRFS_I(btree_inode)->io_tree;
 684
 685         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 686         if (!buf)
 687                 return NULL;
 688
 689         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 690
 691         if (ret == 0) {
 692                 buf->flags |= EXTENT_UPTODATE;
 693         }
 694         return buf;
 695
 696 }
 697
 698 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 699                      struct extent_buffer *buf)
 700 {
 701         struct inode *btree_inode = root->fs_info->btree_inode;
 702         if (btrfs_header_generation(buf) ==
 703             root->fs_info->running_transaction->transid) {
 704                 WARN_ON(!btrfs_tree_locked(buf));
 705                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 706                                           buf);
 707         }
 708         return 0;
 709 }
 710
 711 int wait_on_tree_block_writeback(struct btrfs_root *root,
 712                                  struct extent_buffer *buf)
 713 {
 714         struct inode *btree_inode = root->fs_info->btree_inode;
 715         wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
 716                                         buf);
 717         return 0;
 718 }
 719
 720 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 721                         u32 stripesize, struct btrfs_root *root,
 722                         struct btrfs_fs_info *fs_info,
 723                         u64 objectid)
 724 {
 725         root->node = NULL;
 726         root->inode = NULL;
 727         root->commit_root = NULL;
 728         root->ref_tree = NULL;
 729         root->sectorsize = sectorsize;
 730         root->nodesize = nodesize;
 731         root->leafsize = leafsize;
 732         root->stripesize = stripesize;
 733         root->ref_cows = 0;
 734         root->track_dirty = 0;
 735
 736         root->fs_info = fs_info;
 737         root->objectid = objectid;
 738         root->last_trans = 0;
 739         root->highest_inode = 0;
 740         root->last_inode_alloc = 0;
 741         root->name = NULL;
 742         root->in_sysfs = 0;
 743
 744         INIT_LIST_HEAD(&root->dirty_list);
 745         INIT_LIST_HEAD(&root->orphan_list);
 746         INIT_LIST_HEAD(&root->dead_list);
 747         spin_lock_init(&root->node_lock);
 748         spin_lock_init(&root->list_lock);
 749         mutex_init(&root->objectid_mutex);
 750
 751         btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
 752         root->ref_tree = &root->ref_tree_struct;
 753
 754         memset(&root->root_key, 0, sizeof(root->root_key));
 755         memset(&root->root_item, 0, sizeof(root->root_item));
 756         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 757         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 758         root->defrag_trans_start = fs_info->generation;
 759         init_completion(&root->kobj_unregister);
 760         root->defrag_running = 0;
 761         root->defrag_level = 0;
 762         root->root_key.objectid = objectid;
 763         return 0;
 764 }
 765
 766 static int find_and_setup_root(struct btrfs_root *tree_root,
 767                                struct btrfs_fs_info *fs_info,
 768                                u64 objectid,
 769                                struct btrfs_root *root)
 770 {
 771         int ret;
 772         u32 blocksize;
 773
 774         __setup_root(tree_root->nodesize, tree_root->leafsize,
 775                      tree_root->sectorsize, tree_root->stripesize,
 776                      root, fs_info, objectid);
 777         ret = btrfs_find_last_root(tree_root, objectid,
 778                                    &root->root_item, &root->root_key);
 779         BUG_ON(ret);
 780
 781         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 782         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 783                                      blocksize, 0);
 784         BUG_ON(!root->node);
 785         return 0;
 786 }
 787
 788 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
 789                                                struct btrfs_key *location)
 790 {
 791         struct btrfs_root *root;
 792         struct btrfs_root *tree_root = fs_info->tree_root;
 793         struct btrfs_path *path;
 794         struct extent_buffer *l;
 795         u64 highest_inode;
 796         u32 blocksize;
 797         int ret = 0;
 798
 799         root = kzalloc(sizeof(*root), GFP_NOFS);
 800         if (!root)
 801                 return ERR_PTR(-ENOMEM);
 802         if (location->offset == (u64)-1) {
 803                 ret = find_and_setup_root(tree_root, fs_info,
 804                                           location->objectid, root);
 805                 if (ret) {
 806                         kfree(root);
 807                         return ERR_PTR(ret);
 808                 }
 809                 goto insert;
 810         }
 811
 812         __setup_root(tree_root->nodesize, tree_root->leafsize,
 813                      tree_root->sectorsize, tree_root->stripesize,
 814                      root, fs_info, location->objectid);
 815
 816         path = btrfs_alloc_path();
 817         BUG_ON(!path);
 818         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 819         if (ret != 0) {
 820                 if (ret > 0)
 821                         ret = -ENOENT;
 822                 goto out;
 823         }
 824         l = path->nodes[0];
 825         read_extent_buffer(l, &root->root_item,
 826                btrfs_item_ptr_offset(l, path->slots[0]),
 827                sizeof(root->root_item));
 828         memcpy(&root->root_key, location, sizeof(*location));
 829         ret = 0;
 830 out:
 831         btrfs_release_path(root, path);
 832         btrfs_free_path(path);
 833         if (ret) {
 834                 kfree(root);
 835                 return ERR_PTR(ret);
 836         }
 837         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 838         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 839                                      blocksize, 0);
 840         BUG_ON(!root->node);
 841 insert:
 842         root->ref_cows = 1;
 843         ret = btrfs_find_highest_inode(root, &highest_inode);
 844         if (ret == 0) {
 845                 root->highest_inode = highest_inode;
 846                 root->last_inode_alloc = highest_inode;
 847         }
 848         return root;
 849 }
 850
 851 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
 852                                         u64 root_objectid)
 853 {
 854         struct btrfs_root *root;
 855
 856         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
 857                 return fs_info->tree_root;
 858         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
 859                 return fs_info->extent_root;
 860
 861         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 862                                  (unsigned long)root_objectid);
 863         return root;
 864 }
 865
 866 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
 867                                               struct btrfs_key *location)
 868 {
 869         struct btrfs_root *root;
 870         int ret;
 871
 872         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 873                 return fs_info->tree_root;
 874         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
 875                 return fs_info->extent_root;
 876         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
 877                 return fs_info->chunk_root;
 878         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
 879                 return fs_info->dev_root;
 880
 881         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 882                                  (unsigned long)location->objectid);
 883         if (root)
 884                 return root;
 885
 886         root = btrfs_read_fs_root_no_radix(fs_info, location);
 887         if (IS_ERR(root))
 888                 return root;
 889         ret = radix_tree_insert(&fs_info->fs_roots_radix,
 890                                 (unsigned long)root->root_key.objectid,
 891                                 root);
 892         if (ret) {
 893                 free_extent_buffer(root->node);
 894                 kfree(root);
 895                 return ERR_PTR(ret);
 896         }
 897         ret = btrfs_find_dead_roots(fs_info->tree_root,
 898                                     root->root_key.objectid, root);
 899         BUG_ON(ret);
 900
 901         return root;
 902 }
 903
 904 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 905                                       struct btrfs_key *location,
 906                                       const char *name, int namelen)
 907 {
 908         struct btrfs_root *root;
 909         int ret;
 910
 911         root = btrfs_read_fs_root_no_name(fs_info, location);
 912         if (!root)
 913                 return NULL;
 914
 915         if (root->in_sysfs)
 916                 return root;
 917
 918         ret = btrfs_set_root_name(root, name, namelen);
 919         if (ret) {
 920                 free_extent_buffer(root->node);
 921                 kfree(root);
 922                 return ERR_PTR(ret);
 923         }
 924
 925         ret = btrfs_sysfs_add_root(root);
 926         if (ret) {
 927                 free_extent_buffer(root->node);
 928                 kfree(root->name);
 929                 kfree(root);
 930                 return ERR_PTR(ret);
 931         }
 932         root->in_sysfs = 1;
 933         return root;
 934 }
 935 #if 0
 936 static int add_hasher(struct btrfs_fs_info *info, char *type) {
 937         struct btrfs_hasher *hasher;
 938
 939         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
 940         if (!hasher)
 941                 return -ENOMEM;
 942         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
 943         if (!hasher->hash_tfm) {
 944                 kfree(hasher);
 945                 return -EINVAL;
 946         }
 947         spin_lock(&info->hash_lock);
 948         list_add(&hasher->list, &info->hashers);
 949         spin_unlock(&info->hash_lock);
 950         return 0;
 951 }
 952 #endif
 953
 954 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
 955 {
 956         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
 957         int ret = 0;
 958         struct list_head *cur;
 959         struct btrfs_device *device;
 960         struct backing_dev_info *bdi;
 961
 962         if ((bdi_bits & (1 << BDI_write_congested)) &&
 963             btrfs_congested_async(info, 0))
 964                 return 1;
 965
 966         list_for_each(cur, &info->fs_devices->devices) {
 967                 device = list_entry(cur, struct btrfs_device, dev_list);
 968                 if (!device->bdev)
 969                         continue;
 970                 bdi = blk_get_backing_dev_info(device->bdev);
 971                 if (bdi && bdi_congested(bdi, bdi_bits)) {
 972                         ret = 1;
 973                         break;
 974                 }
 975         }
 976         return ret;
 977 }
 978
 979 /*
 980  * this unplugs every device on the box, and it is only used when page
 981  * is null
 982  */
 983 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
 984 {
 985         struct list_head *cur;
 986         struct btrfs_device *device;
 987         struct btrfs_fs_info *info;
 988
 989         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
 990         list_for_each(cur, &info->fs_devices->devices) {
 991                 device = list_entry(cur, struct btrfs_device, dev_list);
 992                 bdi = blk_get_backing_dev_info(device->bdev);
 993                 if (bdi->unplug_io_fn) {
 994                         bdi->unplug_io_fn(bdi, page);
 995                 }
 996         }
 997 }
 998
 999 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1000 {
1001         struct inode *inode;
1002         struct extent_map_tree *em_tree;
1003         struct extent_map *em;
1004         struct address_space *mapping;
1005         u64 offset;
1006
1007         /* the generic O_DIRECT read code does this */
1008         if (!page) {
1009                 __unplug_io_fn(bdi, page);
1010                 return;
1011         }
1012
1013         /*
1014          * page->mapping may change at any time.  Get a consistent copy
1015          * and use that for everything below
1016          */
1017         smp_mb();
1018         mapping = page->mapping;
1019         if (!mapping)
1020                 return;
1021
1022         inode = mapping->host;
1023         offset = page_offset(page);
1024
1025         em_tree = &BTRFS_I(inode)->extent_tree;
1026         spin_lock(&em_tree->lock);
1027         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1028         spin_unlock(&em_tree->lock);
1029         if (!em) {
1030                 __unplug_io_fn(bdi, page);
1031                 return;
1032         }
1033
1034         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1035                 free_extent_map(em);
1036                 __unplug_io_fn(bdi, page);
1037                 return;
1038         }
1039         offset = offset - em->start;
1040         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1041                           em->block_start + offset, page);
1042         free_extent_map(em);
1043 }
1044
1045 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1046 {
1047 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1048         bdi_init(bdi);
1049 #endif
1050         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1051         bdi->state              = 0;
1052         bdi->capabilities       = default_backing_dev_info.capabilities;
1053         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1054         bdi->unplug_io_data     = info;
1055         bdi->congested_fn       = btrfs_congested_fn;
1056         bdi->congested_data     = info;
1057         return 0;
1058 }
1059
1060 static int bio_ready_for_csum(struct bio *bio)
1061 {
1062         u64 length = 0;
1063         u64 buf_len = 0;
1064         u64 start = 0;
1065         struct page *page;
1066         struct extent_io_tree *io_tree = NULL;
1067         struct btrfs_fs_info *info = NULL;
1068         struct bio_vec *bvec;
1069         int i;
1070         int ret;
1071
1072         bio_for_each_segment(bvec, bio, i) {
1073                 page = bvec->bv_page;
1074                 if (page->private == EXTENT_PAGE_PRIVATE) {
1075                         length += bvec->bv_len;
1076                         continue;
1077                 }
1078                 if (!page->private) {
1079                         length += bvec->bv_len;
1080                         continue;
1081                 }
1082                 length = bvec->bv_len;
1083                 buf_len = page->private >> 2;
1084                 start = page_offset(page) + bvec->bv_offset;
1085                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1086                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1087         }
1088         /* are we fully contained in this bio? */
1089         if (buf_len <= length)
1090                 return 1;
1091
1092         ret = extent_range_uptodate(io_tree, start + length,
1093                                     start + buf_len - 1);
1094         if (ret == 1)
1095                 return ret;
1096         return ret;
1097 }
1098
1099 /*
1100  * called by the kthread helper functions to finally call the bio end_io
1101  * functions.  This is where read checksum verification actually happens
1102  */
1103 static void end_workqueue_fn(struct btrfs_work *work)
1104 {
1105         struct bio *bio;
1106         struct end_io_wq *end_io_wq;
1107         struct btrfs_fs_info *fs_info;
1108         int error;
1109
1110         end_io_wq = container_of(work, struct end_io_wq, work);
1111         bio = end_io_wq->bio;
1112         fs_info = end_io_wq->info;
1113
1114         /* metadata bios are special because the whole tree block must
1115          * be checksummed at once.  This makes sure the entire block is in
1116          * ram and up to date before trying to verify things.  For
1117          * blocksize <= pagesize, it is basically a noop
1118          */
1119         if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1120                 btrfs_queue_worker(&fs_info->endio_workers,
1121                                    &end_io_wq->work);
1122                 return;
1123         }
1124         error = end_io_wq->error;
1125         bio->bi_private = end_io_wq->private;
1126         bio->bi_end_io = end_io_wq->end_io;
1127         kfree(end_io_wq);
1128 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1129         bio_endio(bio, bio->bi_size, error);
1130 #else
1131         bio_endio(bio, error);
1132 #endif
1133 }
1134
1135 static int cleaner_kthread(void *arg)
1136 {
1137         struct btrfs_root *root = arg;
1138
1139         do {
1140                 smp_mb();
1141                 if (root->fs_info->closing)
1142                         break;
1143
1144                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1145                 mutex_lock(&root->fs_info->cleaner_mutex);
1146                 btrfs_clean_old_snapshots(root);
1147                 mutex_unlock(&root->fs_info->cleaner_mutex);
1148
1149                 if (freezing(current)) {
1150                         refrigerator();
1151                 } else {
1152                         smp_mb();
1153                         if (root->fs_info->closing)
1154                                 break;
1155                         set_current_state(TASK_INTERRUPTIBLE);
1156                         schedule();
1157                         __set_current_state(TASK_RUNNING);
1158                 }
1159         } while (!kthread_should_stop());
1160         return 0;
1161 }
1162
1163 static int transaction_kthread(void *arg)
1164 {
1165         struct btrfs_root *root = arg;
1166         struct btrfs_trans_handle *trans;
1167         struct btrfs_transaction *cur;
1168         unsigned long now;
1169         unsigned long delay;
1170         int ret;
1171
1172         do {
1173                 smp_mb();
1174                 if (root->fs_info->closing)
1175                         break;
1176
1177                 delay = HZ * 30;
1178                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1179                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1180
1181                 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1182                         printk("btrfs: total reference cache size %Lu\n",
1183                                 root->fs_info->total_ref_cache_size);
1184                 }
1185
1186                 mutex_lock(&root->fs_info->trans_mutex);
1187                 cur = root->fs_info->running_transaction;
1188                 if (!cur) {
1189                         mutex_unlock(&root->fs_info->trans_mutex);
1190                         goto sleep;
1191                 }
1192
1193                 now = get_seconds();
1194                 if (now < cur->start_time || now - cur->start_time < 30) {
1195                         mutex_unlock(&root->fs_info->trans_mutex);
1196                         delay = HZ * 5;
1197                         goto sleep;
1198                 }
1199                 mutex_unlock(&root->fs_info->trans_mutex);
1200                 trans = btrfs_start_transaction(root, 1);
1201                 ret = btrfs_commit_transaction(trans, root);
1202 sleep:
1203                 wake_up_process(root->fs_info->cleaner_kthread);
1204                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1205
1206                 if (freezing(current)) {
1207                         refrigerator();
1208                 } else {
1209                         if (root->fs_info->closing)
1210                                 break;
1211                         set_current_state(TASK_INTERRUPTIBLE);
1212                         schedule_timeout(delay);
1213                         __set_current_state(TASK_RUNNING);
1214                 }
1215         } while (!kthread_should_stop());
1216         return 0;
1217 }
1218
1219 struct btrfs_root *open_ctree(struct super_block *sb,
1220                               struct btrfs_fs_devices *fs_devices,
1221                               char *options)
1222 {
1223         u32 sectorsize;
1224         u32 nodesize;
1225         u32 leafsize;
1226         u32 blocksize;
1227         u32 stripesize;
1228         struct buffer_head *bh;
1229         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
1230                                                  GFP_NOFS);
1231         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
1232                                                GFP_NOFS);
1233         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1234                                                 GFP_NOFS);
1235         struct btrfs_root *chunk_root = kmalloc(sizeof(struct btrfs_root),
1236                                                 GFP_NOFS);
1237         struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
1238                                               GFP_NOFS);
1239         int ret;
1240         int err = -EINVAL;
1241
1242         struct btrfs_super_block *disk_super;
1243
1244         if (!extent_root || !tree_root || !fs_info) {
1245                 err = -ENOMEM;
1246                 goto fail;
1247         }
1248         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1249         INIT_LIST_HEAD(&fs_info->trans_list);
1250         INIT_LIST_HEAD(&fs_info->dead_roots);
1251         INIT_LIST_HEAD(&fs_info->hashers);
1252         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1253         spin_lock_init(&fs_info->hash_lock);
1254         spin_lock_init(&fs_info->delalloc_lock);
1255         spin_lock_init(&fs_info->new_trans_lock);
1256         spin_lock_init(&fs_info->ref_cache_lock);
1257
1258         init_completion(&fs_info->kobj_unregister);
1259         fs_info->tree_root = tree_root;
1260         fs_info->extent_root = extent_root;
1261         fs_info->chunk_root = chunk_root;
1262         fs_info->dev_root = dev_root;
1263         fs_info->fs_devices = fs_devices;
1264         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1265         INIT_LIST_HEAD(&fs_info->space_info);
1266         btrfs_mapping_init(&fs_info->mapping_tree);
1267         atomic_set(&fs_info->nr_async_submits, 0);
1268         atomic_set(&fs_info->nr_async_bios, 0);
1269         atomic_set(&fs_info->throttles, 0);
1270         atomic_set(&fs_info->throttle_gen, 0);
1271         fs_info->sb = sb;
1272         fs_info->max_extent = (u64)-1;
1273         fs_info->max_inline = 8192 * 1024;
1274         setup_bdi(fs_info, &fs_info->bdi);
1275         fs_info->btree_inode = new_inode(sb);
1276         fs_info->btree_inode->i_ino = 1;
1277         fs_info->btree_inode->i_nlink = 1;
1278         fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1279
1280         INIT_LIST_HEAD(&fs_info->ordered_extents);
1281         spin_lock_init(&fs_info->ordered_extent_lock);
1282
1283         sb->s_blocksize = 4096;
1284         sb->s_blocksize_bits = blksize_bits(4096);
1285
1286         /*
1287          * we set the i_size on the btree inode to the max possible int.
1288          * the real end of the address space is determined by all of
1289          * the devices in the system
1290          */
1291         fs_info->btree_inode->i_size = OFFSET_MAX;
1292         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1293         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1294
1295         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1296                              fs_info->btree_inode->i_mapping,
1297                              GFP_NOFS);
1298         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1299                              GFP_NOFS);
1300
1301         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1302
1303         extent_io_tree_init(&fs_info->free_space_cache,
1304                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1305         extent_io_tree_init(&fs_info->block_group_cache,
1306                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1307         extent_io_tree_init(&fs_info->pinned_extents,
1308                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1309         extent_io_tree_init(&fs_info->pending_del,
1310                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1311         extent_io_tree_init(&fs_info->extent_ins,
1312                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1313         fs_info->do_barriers = 1;
1314
1315         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1316         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1317                sizeof(struct btrfs_key));
1318         insert_inode_hash(fs_info->btree_inode);
1319         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1320
1321         mutex_init(&fs_info->trans_mutex);
1322         mutex_init(&fs_info->drop_mutex);
1323         mutex_init(&fs_info->alloc_mutex);
1324         mutex_init(&fs_info->chunk_mutex);
1325         mutex_init(&fs_info->transaction_kthread_mutex);
1326         mutex_init(&fs_info->cleaner_mutex);
1327         mutex_init(&fs_info->volume_mutex);
1328         init_waitqueue_head(&fs_info->transaction_throttle);
1329         init_waitqueue_head(&fs_info->transaction_wait);
1330
1331 #if 0
1332         ret = add_hasher(fs_info, "crc32c");
1333         if (ret) {
1334                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1335                 err = -ENOMEM;
1336                 goto fail_iput;
1337         }
1338 #endif
1339         __setup_root(4096, 4096, 4096, 4096, tree_root,
1340                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1341
1342
1343         bh = __bread(fs_devices->latest_bdev,
1344                      BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1345         if (!bh)
1346                 goto fail_iput;
1347
1348         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1349         brelse(bh);
1350
1351         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1352
1353         disk_super = &fs_info->super_copy;
1354         if (!btrfs_super_root(disk_super))
1355                 goto fail_sb_buffer;
1356
1357         err = btrfs_parse_options(tree_root, options);
1358         if (err)
1359                 goto fail_sb_buffer;
1360
1361         /*
1362          * we need to start all the end_io workers up front because the
1363          * queue work function gets called at interrupt time, and so it
1364          * cannot dynamically grow.
1365          */
1366         btrfs_init_workers(&fs_info->workers, fs_info->thread_pool_size);
1367         btrfs_init_workers(&fs_info->submit_workers,
1368                            min_t(u64, fs_devices->num_devices,
1369                            fs_info->thread_pool_size));
1370
1371         /* a higher idle thresh on the submit workers makes it much more
1372          * likely that bios will be send down in a sane order to the
1373          * devices
1374          */
1375         fs_info->submit_workers.idle_thresh = 64;
1376
1377         btrfs_init_workers(&fs_info->fixup_workers, 1);
1378         btrfs_init_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1379         btrfs_init_workers(&fs_info->endio_write_workers,
1380                            fs_info->thread_pool_size);
1381
1382         /*
1383          * endios are largely parallel and should have a very
1384          * low idle thresh
1385          */
1386         fs_info->endio_workers.idle_thresh = 4;
1387         fs_info->endio_write_workers.idle_thresh = 4;
1388
1389         btrfs_start_workers(&fs_info->workers, 1);
1390         btrfs_start_workers(&fs_info->submit_workers, 1);
1391         btrfs_start_workers(&fs_info->fixup_workers, 1);
1392         btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1393         btrfs_start_workers(&fs_info->endio_write_workers,
1394                             fs_info->thread_pool_size);
1395
1396         err = -EINVAL;
1397         if (btrfs_super_num_devices(disk_super) > fs_devices->open_devices) {
1398                 printk("Btrfs: wanted %llu devices, but found %llu\n",
1399                        (unsigned long long)btrfs_super_num_devices(disk_super),
1400                        (unsigned long long)fs_devices->open_devices);
1401                 if (btrfs_test_opt(tree_root, DEGRADED))
1402                         printk("continuing in degraded mode\n");
1403                 else {
1404                         goto fail_sb_buffer;
1405                 }
1406         }
1407
1408         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1409
1410         nodesize = btrfs_super_nodesize(disk_super);
1411         leafsize = btrfs_super_leafsize(disk_super);
1412         sectorsize = btrfs_super_sectorsize(disk_super);
1413         stripesize = btrfs_super_stripesize(disk_super);
1414         tree_root->nodesize = nodesize;
1415         tree_root->leafsize = leafsize;
1416         tree_root->sectorsize = sectorsize;
1417         tree_root->stripesize = stripesize;
1418
1419         sb->s_blocksize = sectorsize;
1420         sb->s_blocksize_bits = blksize_bits(sectorsize);
1421
1422         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1423                     sizeof(disk_super->magic))) {
1424                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1425                 goto fail_sb_buffer;
1426         }
1427
1428         mutex_lock(&fs_info->chunk_mutex);
1429         ret = btrfs_read_sys_array(tree_root);
1430         mutex_unlock(&fs_info->chunk_mutex);
1431         if (ret) {
1432                 printk("btrfs: failed to read the system array on %s\n",
1433                        sb->s_id);
1434                 goto fail_sys_array;
1435         }
1436
1437         blocksize = btrfs_level_size(tree_root,
1438                                      btrfs_super_chunk_root_level(disk_super));
1439
1440         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1441                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1442
1443         chunk_root->node = read_tree_block(chunk_root,
1444                                            btrfs_super_chunk_root(disk_super),
1445                                            blocksize, 0);
1446         BUG_ON(!chunk_root->node);
1447
1448         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1449                  (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1450                  BTRFS_UUID_SIZE);
1451
1452         mutex_lock(&fs_info->chunk_mutex);
1453         ret = btrfs_read_chunk_tree(chunk_root);
1454         mutex_unlock(&fs_info->chunk_mutex);
1455         BUG_ON(ret);
1456
1457         btrfs_close_extra_devices(fs_devices);
1458
1459         blocksize = btrfs_level_size(tree_root,
1460                                      btrfs_super_root_level(disk_super));
1461
1462
1463         tree_root->node = read_tree_block(tree_root,
1464                                           btrfs_super_root(disk_super),
1465                                           blocksize, 0);
1466         if (!tree_root->node)
1467                 goto fail_sb_buffer;
1468
1469
1470         ret = find_and_setup_root(tree_root, fs_info,
1471                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1472         if (ret)
1473                 goto fail_tree_root;
1474         extent_root->track_dirty = 1;
1475
1476         ret = find_and_setup_root(tree_root, fs_info,
1477                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1478         dev_root->track_dirty = 1;
1479
1480         if (ret)
1481                 goto fail_extent_root;
1482
1483         btrfs_read_block_groups(extent_root);
1484
1485         fs_info->generation = btrfs_super_generation(disk_super) + 1;
1486         fs_info->data_alloc_profile = (u64)-1;
1487         fs_info->metadata_alloc_profile = (u64)-1;
1488         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1489         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1490                                                "btrfs-cleaner");
1491         if (!fs_info->cleaner_kthread)
1492                 goto fail_extent_root;
1493
1494         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1495                                                    tree_root,
1496                                                    "btrfs-transaction");
1497         if (!fs_info->transaction_kthread)
1498                 goto fail_cleaner;
1499
1500
1501         return tree_root;
1502
1503 fail_cleaner:
1504         kthread_stop(fs_info->cleaner_kthread);
1505 fail_extent_root:
1506         free_extent_buffer(extent_root->node);
1507 fail_tree_root:
1508         free_extent_buffer(tree_root->node);
1509 fail_sys_array:
1510 fail_sb_buffer:
1511         btrfs_stop_workers(&fs_info->fixup_workers);
1512         btrfs_stop_workers(&fs_info->workers);
1513         btrfs_stop_workers(&fs_info->endio_workers);
1514         btrfs_stop_workers(&fs_info->endio_write_workers);
1515         btrfs_stop_workers(&fs_info->submit_workers);
1516 fail_iput:
1517         iput(fs_info->btree_inode);
1518 fail:
1519         btrfs_close_devices(fs_info->fs_devices);
1520         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1521
1522         kfree(extent_root);
1523         kfree(tree_root);
1524 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1525         bdi_destroy(&fs_info->bdi);
1526 #endif
1527         kfree(fs_info);
1528         return ERR_PTR(err);
1529 }
1530
1531 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1532 {
1533         char b[BDEVNAME_SIZE];
1534
1535         if (uptodate) {
1536                 set_buffer_uptodate(bh);
1537         } else {
1538                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1539                         printk(KERN_WARNING "lost page write due to "
1540                                         "I/O error on %s\n",
1541                                        bdevname(bh->b_bdev, b));
1542                 }
1543                 /* note, we dont' set_buffer_write_io_error because we have
1544                  * our own ways of dealing with the IO errors
1545                  */
1546                 clear_buffer_uptodate(bh);
1547         }
1548         unlock_buffer(bh);
1549         put_bh(bh);
1550 }
1551
1552 int write_all_supers(struct btrfs_root *root)
1553 {
1554         struct list_head *cur;
1555         struct list_head *head = &root->fs_info->fs_devices->devices;
1556         struct btrfs_device *dev;
1557         struct btrfs_super_block *sb;
1558         struct btrfs_dev_item *dev_item;
1559         struct buffer_head *bh;
1560         int ret;
1561         int do_barriers;
1562         int max_errors;
1563         int total_errors = 0;
1564         u32 crc;
1565         u64 flags;
1566
1567         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1568         do_barriers = !btrfs_test_opt(root, NOBARRIER);
1569
1570         sb = &root->fs_info->super_for_commit;
1571         dev_item = &sb->dev_item;
1572         list_for_each(cur, head) {
1573                 dev = list_entry(cur, struct btrfs_device, dev_list);
1574                 if (!dev->bdev) {
1575                         total_errors++;
1576                         continue;
1577                 }
1578                 if (!dev->in_fs_metadata)
1579                         continue;
1580
1581                 btrfs_set_stack_device_type(dev_item, dev->type);
1582                 btrfs_set_stack_device_id(dev_item, dev->devid);
1583                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1584                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1585                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1586                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1587                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1588                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1589                 flags = btrfs_super_flags(sb);
1590                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1591
1592
1593                 crc = ~(u32)0;
1594                 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1595                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1596                 btrfs_csum_final(crc, sb->csum);
1597
1598                 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1599                               BTRFS_SUPER_INFO_SIZE);
1600
1601                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1602                 dev->pending_io = bh;
1603
1604                 get_bh(bh);
1605                 set_buffer_uptodate(bh);
1606                 lock_buffer(bh);
1607                 bh->b_end_io = btrfs_end_buffer_write_sync;
1608
1609                 if (do_barriers && dev->barriers) {
1610                         ret = submit_bh(WRITE_BARRIER, bh);
1611                         if (ret == -EOPNOTSUPP) {
1612                                 printk("btrfs: disabling barriers on dev %s\n",
1613                                        dev->name);
1614                                 set_buffer_uptodate(bh);
1615                                 dev->barriers = 0;
1616                                 get_bh(bh);
1617                                 lock_buffer(bh);
1618                                 ret = submit_bh(WRITE, bh);
1619                         }
1620                 } else {
1621                         ret = submit_bh(WRITE, bh);
1622                 }
1623                 if (ret)
1624                         total_errors++;
1625         }
1626         if (total_errors > max_errors) {
1627                 printk("btrfs: %d errors while writing supers\n", total_errors);
1628                 BUG();
1629         }
1630         total_errors = 0;
1631
1632         list_for_each(cur, head) {
1633                 dev = list_entry(cur, struct btrfs_device, dev_list);
1634                 if (!dev->bdev)
1635                         continue;
1636                 if (!dev->in_fs_metadata)
1637                         continue;
1638
1639                 BUG_ON(!dev->pending_io);
1640                 bh = dev->pending_io;
1641                 wait_on_buffer(bh);
1642                 if (!buffer_uptodate(dev->pending_io)) {
1643                         if (do_barriers && dev->barriers) {
1644                                 printk("btrfs: disabling barriers on dev %s\n",
1645                                        dev->name);
1646                                 set_buffer_uptodate(bh);
1647                                 get_bh(bh);
1648                                 lock_buffer(bh);
1649                                 dev->barriers = 0;
1650                                 ret = submit_bh(WRITE, bh);
1651                                 BUG_ON(ret);
1652                                 wait_on_buffer(bh);
1653                                 if (!buffer_uptodate(bh))
1654                                         total_errors++;
1655                         } else {
1656                                 total_errors++;
1657                         }
1658
1659                 }
1660                 dev->pending_io = NULL;
1661                 brelse(bh);
1662         }
1663         if (total_errors > max_errors) {
1664                 printk("btrfs: %d errors while writing supers\n", total_errors);
1665                 BUG();
1666         }
1667         return 0;
1668 }
1669
1670 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1671                       *root)
1672 {
1673         int ret;
1674
1675         ret = write_all_supers(root);
1676         return ret;
1677 }
1678
1679 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1680 {
1681         radix_tree_delete(&fs_info->fs_roots_radix,
1682                           (unsigned long)root->root_key.objectid);
1683         if (root->in_sysfs)
1684                 btrfs_sysfs_del_root(root);
1685         if (root->inode)
1686                 iput(root->inode);
1687         if (root->node)
1688                 free_extent_buffer(root->node);
1689         if (root->commit_root)
1690                 free_extent_buffer(root->commit_root);
1691         if (root->name)
1692                 kfree(root->name);
1693         kfree(root);
1694         return 0;
1695 }
1696
1697 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1698 {
1699         int ret;
1700         struct btrfs_root *gang[8];
1701         int i;
1702
1703         while(1) {
1704                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1705                                              (void **)gang, 0,
1706                                              ARRAY_SIZE(gang));
1707                 if (!ret)
1708                         break;
1709                 for (i = 0; i < ret; i++)
1710                         btrfs_free_fs_root(fs_info, gang[i]);
1711         }
1712         return 0;
1713 }
1714
1715 int close_ctree(struct btrfs_root *root)
1716 {
1717         int ret;
1718         struct btrfs_trans_handle *trans;
1719         struct btrfs_fs_info *fs_info = root->fs_info;
1720
1721         fs_info->closing = 1;
1722         smp_mb();
1723
1724         kthread_stop(root->fs_info->transaction_kthread);
1725         kthread_stop(root->fs_info->cleaner_kthread);
1726
1727         btrfs_clean_old_snapshots(root);
1728         trans = btrfs_start_transaction(root, 1);
1729         ret = btrfs_commit_transaction(trans, root);
1730         /* run commit again to  drop the original snapshot */
1731         trans = btrfs_start_transaction(root, 1);
1732         btrfs_commit_transaction(trans, root);
1733         ret = btrfs_write_and_wait_transaction(NULL, root);
1734         BUG_ON(ret);
1735
1736         write_ctree_super(NULL, root);
1737
1738         if (fs_info->delalloc_bytes) {
1739                 printk("btrfs: at unmount delalloc count %Lu\n",
1740                        fs_info->delalloc_bytes);
1741         }
1742         if (fs_info->total_ref_cache_size) {
1743                 printk("btrfs: at umount reference cache size %Lu\n",
1744                         fs_info->total_ref_cache_size);
1745         }
1746
1747         if (fs_info->extent_root->node)
1748                 free_extent_buffer(fs_info->extent_root->node);
1749
1750         if (fs_info->tree_root->node)
1751                 free_extent_buffer(fs_info->tree_root->node);
1752
1753         if (root->fs_info->chunk_root->node);
1754                 free_extent_buffer(root->fs_info->chunk_root->node);
1755
1756         if (root->fs_info->dev_root->node);
1757                 free_extent_buffer(root->fs_info->dev_root->node);
1758
1759         btrfs_free_block_groups(root->fs_info);
1760         fs_info->closing = 2;
1761         del_fs_roots(fs_info);
1762
1763         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1764
1765         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1766
1767         btrfs_stop_workers(&fs_info->fixup_workers);
1768         btrfs_stop_workers(&fs_info->workers);
1769         btrfs_stop_workers(&fs_info->endio_workers);
1770         btrfs_stop_workers(&fs_info->endio_write_workers);
1771         btrfs_stop_workers(&fs_info->submit_workers);
1772
1773         iput(fs_info->btree_inode);
1774 #if 0
1775         while(!list_empty(&fs_info->hashers)) {
1776                 struct btrfs_hasher *hasher;
1777                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1778                                     hashers);
1779                 list_del(&hasher->hashers);
1780                 crypto_free_hash(&fs_info->hash_tfm);
1781                 kfree(hasher);
1782         }
1783 #endif
1784         btrfs_close_devices(fs_info->fs_devices);
1785         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1786
1787 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1788         bdi_destroy(&fs_info->bdi);
1789 #endif
1790
1791         kfree(fs_info->extent_root);
1792         kfree(fs_info->tree_root);
1793         kfree(fs_info->chunk_root);
1794         kfree(fs_info->dev_root);
1795         return 0;
1796 }
1797
1798 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1799 {
1800         int ret;
1801         struct inode *btree_inode = buf->first_page->mapping->host;
1802
1803         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1804         if (!ret)
1805                 return ret;
1806
1807         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
1808                                     parent_transid);
1809         return !ret;
1810 }
1811
1812 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1813 {
1814         struct inode *btree_inode = buf->first_page->mapping->host;
1815         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1816                                           buf);
1817 }
1818
1819 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1820 {
1821         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1822         u64 transid = btrfs_header_generation(buf);
1823         struct inode *btree_inode = root->fs_info->btree_inode;
1824
1825         WARN_ON(!btrfs_tree_locked(buf));
1826         if (transid != root->fs_info->generation) {
1827                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1828                         (unsigned long long)buf->start,
1829                         transid, root->fs_info->generation);
1830                 WARN_ON(1);
1831         }
1832         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1833 }
1834
1835 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1836 {
1837         /*
1838          * looks as though older kernels can get into trouble with
1839          * this code, they end up stuck in balance_dirty_pages forever
1840          */
1841         struct extent_io_tree *tree;
1842         u64 num_dirty;
1843         u64 start = 0;
1844         unsigned long thresh = 2 * 1024 * 1024;
1845         tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
1846
1847         if (current_is_pdflush())
1848                 return;
1849
1850         num_dirty = count_range_bits(tree, &start, (u64)-1,
1851                                      thresh, EXTENT_DIRTY);
1852         if (num_dirty > thresh) {
1853                 balance_dirty_pages_ratelimited_nr(
1854                                    root->fs_info->btree_inode->i_mapping, 1);
1855         }
1856         return;
1857 }
1858
1859 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
1860 {
1861         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1862         int ret;
1863         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1864         if (ret == 0) {
1865                 buf->flags |= EXTENT_UPTODATE;
1866         }
1867         return ret;
1868 }
1869
1870 static struct extent_io_ops btree_extent_io_ops = {
1871         .writepage_io_hook = btree_writepage_io_hook,
1872         .readpage_end_io_hook = btree_readpage_end_io_hook,
1873         .submit_bio_hook = btree_submit_bio_hook,
1874         /* note we're sharing with inode.c for the merge bio hook */
1875         .merge_bio_hook = btrfs_merge_bio_hook,
1876 };