Pileus Git - ~andy/linux/blob - 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 into btrfs_map_bio
 496          */
 497         if (rw & (1 << BIO_RW)) {
 498                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 499                                      mirror_num, 1);
 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
 532         if (current->flags & PF_MEMALLOC) {
 533                 redirty_page_for_writepage(wbc, page);
 534                 unlock_page(page);
 535                 return 0;
 536         }
 537         return extent_write_full_page(tree, page, btree_get_extent, wbc);
 538 }
 539
 540 static int btree_writepages(struct address_space *mapping,
 541                             struct writeback_control *wbc)
 542 {
 543         struct extent_io_tree *tree;
 544         tree = &BTRFS_I(mapping->host)->io_tree;
 545         if (wbc->sync_mode == WB_SYNC_NONE) {
 546                 u64 num_dirty;
 547                 u64 start = 0;
 548                 unsigned long thresh = 96 * 1024 * 1024;
 549
 550                 if (wbc->for_kupdate)
 551                         return 0;
 552
 553                 if (current_is_pdflush()) {
 554                         thresh = 96 * 1024 * 1024;
 555                 } else {
 556                         thresh = 8 * 1024 * 1024;
 557                 }
 558                 num_dirty = count_range_bits(tree, &start, (u64)-1,
 559                                              thresh, EXTENT_DIRTY);
 560                 if (num_dirty < thresh) {
 561                         return 0;
 562                 }
 563         }
 564         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 565 }
 566
 567 int btree_readpage(struct file *file, struct page *page)
 568 {
 569         struct extent_io_tree *tree;
 570         tree = &BTRFS_I(page->mapping->host)->io_tree;
 571         return extent_read_full_page(tree, page, btree_get_extent);
 572 }
 573
 574 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 575 {
 576         struct extent_io_tree *tree;
 577         struct extent_map_tree *map;
 578         int ret;
 579
 580         tree = &BTRFS_I(page->mapping->host)->io_tree;
 581         map = &BTRFS_I(page->mapping->host)->extent_tree;
 582
 583         ret = try_release_extent_state(map, tree, page, gfp_flags);
 584         if (!ret) {
 585                 return 0;
 586         }
 587
 588         ret = try_release_extent_buffer(tree, page);
 589         if (ret == 1) {
 590                 ClearPagePrivate(page);
 591                 set_page_private(page, 0);
 592                 page_cache_release(page);
 593         }
 594
 595         return ret;
 596 }
 597
 598 static void btree_invalidatepage(struct page *page, unsigned long offset)
 599 {
 600         struct extent_io_tree *tree;
 601         tree = &BTRFS_I(page->mapping->host)->io_tree;
 602         extent_invalidatepage(tree, page, offset);
 603         btree_releasepage(page, GFP_NOFS);
 604         if (PagePrivate(page)) {
 605                 printk("warning page private not zero on page %Lu\n",
 606                        page_offset(page));
 607                 ClearPagePrivate(page);
 608                 set_page_private(page, 0);
 609                 page_cache_release(page);
 610         }
 611 }
 612
 613 #if 0
 614 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 615 {
 616         struct buffer_head *bh;
 617         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 618         struct buffer_head *head;
 619         if (!page_has_buffers(page)) {
 620                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 621                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 622         }
 623         head = page_buffers(page);
 624         bh = head;
 625         do {
 626                 if (buffer_dirty(bh))
 627                         csum_tree_block(root, bh, 0);
 628                 bh = bh->b_this_page;
 629         } while (bh != head);
 630         return block_write_full_page(page, btree_get_block, wbc);
 631 }
 632 #endif
 633
 634 static struct address_space_operations btree_aops = {
 635         .readpage       = btree_readpage,
 636         .writepage      = btree_writepage,
 637         .writepages     = btree_writepages,
 638         .releasepage    = btree_releasepage,
 639         .invalidatepage = btree_invalidatepage,
 640         .sync_page      = block_sync_page,
 641 };
 642
 643 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 644                          u64 parent_transid)
 645 {
 646         struct extent_buffer *buf = NULL;
 647         struct inode *btree_inode = root->fs_info->btree_inode;
 648         int ret = 0;
 649
 650         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 651         if (!buf)
 652                 return 0;
 653         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 654                                  buf, 0, 0, btree_get_extent, 0);
 655         free_extent_buffer(buf);
 656         return ret;
 657 }
 658
 659 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 660                                             u64 bytenr, u32 blocksize)
 661 {
 662         struct inode *btree_inode = root->fs_info->btree_inode;
 663         struct extent_buffer *eb;
 664         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 665                                 bytenr, blocksize, GFP_NOFS);
 666         return eb;
 667 }
 668
 669 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 670                                                  u64 bytenr, u32 blocksize)
 671 {
 672         struct inode *btree_inode = root->fs_info->btree_inode;
 673         struct extent_buffer *eb;
 674
 675         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 676                                  bytenr, blocksize, NULL, GFP_NOFS);
 677         return eb;
 678 }
 679
 680
 681 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 682                                       u32 blocksize, u64 parent_transid)
 683 {
 684         struct extent_buffer *buf = NULL;
 685         struct inode *btree_inode = root->fs_info->btree_inode;
 686         struct extent_io_tree *io_tree;
 687         int ret;
 688
 689         io_tree = &BTRFS_I(btree_inode)->io_tree;
 690
 691         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 692         if (!buf)
 693                 return NULL;
 694
 695         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 696
 697         if (ret == 0) {
 698                 buf->flags |= EXTENT_UPTODATE;
 699         }
 700         return buf;
 701
 702 }
 703
 704 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 705                      struct extent_buffer *buf)
 706 {
 707         struct inode *btree_inode = root->fs_info->btree_inode;
 708         if (btrfs_header_generation(buf) ==
 709             root->fs_info->running_transaction->transid) {
 710                 WARN_ON(!btrfs_tree_locked(buf));
 711                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 712                                           buf);
 713         }
 714         return 0;
 715 }
 716
 717 int wait_on_tree_block_writeback(struct btrfs_root *root,
 718                                  struct extent_buffer *buf)
 719 {
 720         struct inode *btree_inode = root->fs_info->btree_inode;
 721         wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
 722                                         buf);
 723         return 0;
 724 }
 725
 726 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 727                         u32 stripesize, struct btrfs_root *root,
 728                         struct btrfs_fs_info *fs_info,
 729                         u64 objectid)
 730 {
 731         root->node = NULL;
 732         root->inode = NULL;
 733         root->commit_root = NULL;
 734         root->ref_tree = NULL;
 735         root->sectorsize = sectorsize;
 736         root->nodesize = nodesize;
 737         root->leafsize = leafsize;
 738         root->stripesize = stripesize;
 739         root->ref_cows = 0;
 740         root->track_dirty = 0;
 741
 742         root->fs_info = fs_info;
 743         root->objectid = objectid;
 744         root->last_trans = 0;
 745         root->highest_inode = 0;
 746         root->last_inode_alloc = 0;
 747         root->name = NULL;
 748         root->in_sysfs = 0;
 749
 750         INIT_LIST_HEAD(&root->dirty_list);
 751         INIT_LIST_HEAD(&root->orphan_list);
 752         INIT_LIST_HEAD(&root->dead_list);
 753         spin_lock_init(&root->node_lock);
 754         spin_lock_init(&root->list_lock);
 755         mutex_init(&root->objectid_mutex);
 756
 757         btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
 758         root->ref_tree = &root->ref_tree_struct;
 759
 760         memset(&root->root_key, 0, sizeof(root->root_key));
 761         memset(&root->root_item, 0, sizeof(root->root_item));
 762         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 763         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 764         root->defrag_trans_start = fs_info->generation;
 765         init_completion(&root->kobj_unregister);
 766         root->defrag_running = 0;
 767         root->defrag_level = 0;
 768         root->root_key.objectid = objectid;
 769         return 0;
 770 }
 771
 772 static int find_and_setup_root(struct btrfs_root *tree_root,
 773                                struct btrfs_fs_info *fs_info,
 774                                u64 objectid,
 775                                struct btrfs_root *root)
 776 {
 777         int ret;
 778         u32 blocksize;
 779
 780         __setup_root(tree_root->nodesize, tree_root->leafsize,
 781                      tree_root->sectorsize, tree_root->stripesize,
 782                      root, fs_info, objectid);
 783         ret = btrfs_find_last_root(tree_root, objectid,
 784                                    &root->root_item, &root->root_key);
 785         BUG_ON(ret);
 786
 787         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 788         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 789                                      blocksize, 0);
 790         BUG_ON(!root->node);
 791         return 0;
 792 }
 793
 794 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
 795                                                struct btrfs_key *location)
 796 {
 797         struct btrfs_root *root;
 798         struct btrfs_root *tree_root = fs_info->tree_root;
 799         struct btrfs_path *path;
 800         struct extent_buffer *l;
 801         u64 highest_inode;
 802         u32 blocksize;
 803         int ret = 0;
 804
 805         root = kzalloc(sizeof(*root), GFP_NOFS);
 806         if (!root)
 807                 return ERR_PTR(-ENOMEM);
 808         if (location->offset == (u64)-1) {
 809                 ret = find_and_setup_root(tree_root, fs_info,
 810                                           location->objectid, root);
 811                 if (ret) {
 812                         kfree(root);
 813                         return ERR_PTR(ret);
 814                 }
 815                 goto insert;
 816         }
 817
 818         __setup_root(tree_root->nodesize, tree_root->leafsize,
 819                      tree_root->sectorsize, tree_root->stripesize,
 820                      root, fs_info, location->objectid);
 821
 822         path = btrfs_alloc_path();
 823         BUG_ON(!path);
 824         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 825         if (ret != 0) {
 826                 if (ret > 0)
 827                         ret = -ENOENT;
 828                 goto out;
 829         }
 830         l = path->nodes[0];
 831         read_extent_buffer(l, &root->root_item,
 832                btrfs_item_ptr_offset(l, path->slots[0]),
 833                sizeof(root->root_item));
 834         memcpy(&root->root_key, location, sizeof(*location));
 835         ret = 0;
 836 out:
 837         btrfs_release_path(root, path);
 838         btrfs_free_path(path);
 839         if (ret) {
 840                 kfree(root);
 841                 return ERR_PTR(ret);
 842         }
 843         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 844         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 845                                      blocksize, 0);
 846         BUG_ON(!root->node);
 847 insert:
 848         root->ref_cows = 1;
 849         ret = btrfs_find_highest_inode(root, &highest_inode);
 850         if (ret == 0) {
 851                 root->highest_inode = highest_inode;
 852                 root->last_inode_alloc = highest_inode;
 853         }
 854         return root;
 855 }
 856
 857 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
 858                                         u64 root_objectid)
 859 {
 860         struct btrfs_root *root;
 861
 862         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
 863                 return fs_info->tree_root;
 864         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
 865                 return fs_info->extent_root;
 866
 867         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 868                                  (unsigned long)root_objectid);
 869         return root;
 870 }
 871
 872 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
 873                                               struct btrfs_key *location)
 874 {
 875         struct btrfs_root *root;
 876         int ret;
 877
 878         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 879                 return fs_info->tree_root;
 880         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
 881                 return fs_info->extent_root;
 882         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
 883                 return fs_info->chunk_root;
 884         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
 885                 return fs_info->dev_root;
 886
 887         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 888                                  (unsigned long)location->objectid);
 889         if (root)
 890                 return root;
 891
 892         root = btrfs_read_fs_root_no_radix(fs_info, location);
 893         if (IS_ERR(root))
 894                 return root;
 895         ret = radix_tree_insert(&fs_info->fs_roots_radix,
 896                                 (unsigned long)root->root_key.objectid,
 897                                 root);
 898         if (ret) {
 899                 free_extent_buffer(root->node);
 900                 kfree(root);
 901                 return ERR_PTR(ret);
 902         }
 903         ret = btrfs_find_dead_roots(fs_info->tree_root,
 904                                     root->root_key.objectid, root);
 905         BUG_ON(ret);
 906
 907         return root;
 908 }
 909
 910 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 911                                       struct btrfs_key *location,
 912                                       const char *name, int namelen)
 913 {
 914         struct btrfs_root *root;
 915         int ret;
 916
 917         root = btrfs_read_fs_root_no_name(fs_info, location);
 918         if (!root)
 919                 return NULL;
 920
 921         if (root->in_sysfs)
 922                 return root;
 923
 924         ret = btrfs_set_root_name(root, name, namelen);
 925         if (ret) {
 926                 free_extent_buffer(root->node);
 927                 kfree(root);
 928                 return ERR_PTR(ret);
 929         }
 930
 931         ret = btrfs_sysfs_add_root(root);
 932         if (ret) {
 933                 free_extent_buffer(root->node);
 934                 kfree(root->name);
 935                 kfree(root);
 936                 return ERR_PTR(ret);
 937         }
 938         root->in_sysfs = 1;
 939         return root;
 940 }
 941 #if 0
 942 static int add_hasher(struct btrfs_fs_info *info, char *type) {
 943         struct btrfs_hasher *hasher;
 944
 945         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
 946         if (!hasher)
 947                 return -ENOMEM;
 948         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
 949         if (!hasher->hash_tfm) {
 950                 kfree(hasher);
 951                 return -EINVAL;
 952         }
 953         spin_lock(&info->hash_lock);
 954         list_add(&hasher->list, &info->hashers);
 955         spin_unlock(&info->hash_lock);
 956         return 0;
 957 }
 958 #endif
 959
 960 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
 961 {
 962         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
 963         int ret = 0;
 964         struct list_head *cur;
 965         struct btrfs_device *device;
 966         struct backing_dev_info *bdi;
 967
 968         if ((bdi_bits & (1 << BDI_write_congested)) &&
 969             btrfs_congested_async(info, 0))
 970                 return 1;
 971
 972         list_for_each(cur, &info->fs_devices->devices) {
 973                 device = list_entry(cur, struct btrfs_device, dev_list);
 974                 if (!device->bdev)
 975                         continue;
 976                 bdi = blk_get_backing_dev_info(device->bdev);
 977                 if (bdi && bdi_congested(bdi, bdi_bits)) {
 978                         ret = 1;
 979                         break;
 980                 }
 981         }
 982         return ret;
 983 }
 984
 985 /*
 986  * this unplugs every device on the box, and it is only used when page
 987  * is null
 988  */
 989 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
 990 {
 991         struct list_head *cur;
 992         struct btrfs_device *device;
 993         struct btrfs_fs_info *info;
 994
 995         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
 996         list_for_each(cur, &info->fs_devices->devices) {
 997                 device = list_entry(cur, struct btrfs_device, dev_list);
 998                 bdi = blk_get_backing_dev_info(device->bdev);
 999                 if (bdi->unplug_io_fn) {
1000                         bdi->unplug_io_fn(bdi, page);
1001                 }
1002         }
1003 }
1004
1005 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1006 {
1007         struct inode *inode;
1008         struct extent_map_tree *em_tree;
1009         struct extent_map *em;
1010         struct address_space *mapping;
1011         u64 offset;
1012
1013         /* the generic O_DIRECT read code does this */
1014         if (!page) {
1015                 __unplug_io_fn(bdi, page);
1016                 return;
1017         }
1018
1019         /*
1020          * page->mapping may change at any time.  Get a consistent copy
1021          * and use that for everything below
1022          */
1023         smp_mb();
1024         mapping = page->mapping;
1025         if (!mapping)
1026                 return;
1027
1028         inode = mapping->host;
1029         offset = page_offset(page);
1030
1031         em_tree = &BTRFS_I(inode)->extent_tree;
1032         spin_lock(&em_tree->lock);
1033         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1034         spin_unlock(&em_tree->lock);
1035         if (!em) {
1036                 __unplug_io_fn(bdi, page);
1037                 return;
1038         }
1039
1040         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1041                 free_extent_map(em);
1042                 __unplug_io_fn(bdi, page);
1043                 return;
1044         }
1045         offset = offset - em->start;
1046         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1047                           em->block_start + offset, page);
1048         free_extent_map(em);
1049 }
1050
1051 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1052 {
1053 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1054         bdi_init(bdi);
1055 #endif
1056         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1057         bdi->state              = 0;
1058         bdi->capabilities       = default_backing_dev_info.capabilities;
1059         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1060         bdi->unplug_io_data     = info;
1061         bdi->congested_fn       = btrfs_congested_fn;
1062         bdi->congested_data     = info;
1063         return 0;
1064 }
1065
1066 static int bio_ready_for_csum(struct bio *bio)
1067 {
1068         u64 length = 0;
1069         u64 buf_len = 0;
1070         u64 start = 0;
1071         struct page *page;
1072         struct extent_io_tree *io_tree = NULL;
1073         struct btrfs_fs_info *info = NULL;
1074         struct bio_vec *bvec;
1075         int i;
1076         int ret;
1077
1078         bio_for_each_segment(bvec, bio, i) {
1079                 page = bvec->bv_page;
1080                 if (page->private == EXTENT_PAGE_PRIVATE) {
1081                         length += bvec->bv_len;
1082                         continue;
1083                 }
1084                 if (!page->private) {
1085                         length += bvec->bv_len;
1086                         continue;
1087                 }
1088                 length = bvec->bv_len;
1089                 buf_len = page->private >> 2;
1090                 start = page_offset(page) + bvec->bv_offset;
1091                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1092                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1093         }
1094         /* are we fully contained in this bio? */
1095         if (buf_len <= length)
1096                 return 1;
1097
1098         ret = extent_range_uptodate(io_tree, start + length,
1099                                     start + buf_len - 1);
1100         if (ret == 1)
1101                 return ret;
1102         return ret;
1103 }
1104
1105 /*
1106  * called by the kthread helper functions to finally call the bio end_io
1107  * functions.  This is where read checksum verification actually happens
1108  */
1109 static void end_workqueue_fn(struct btrfs_work *work)
1110 {
1111         struct bio *bio;
1112         struct end_io_wq *end_io_wq;
1113         struct btrfs_fs_info *fs_info;
1114         int error;
1115
1116         end_io_wq = container_of(work, struct end_io_wq, work);
1117         bio = end_io_wq->bio;
1118         fs_info = end_io_wq->info;
1119
1120         /* metadata bios are special because the whole tree block must
1121          * be checksummed at once.  This makes sure the entire block is in
1122          * ram and up to date before trying to verify things.  For
1123          * blocksize <= pagesize, it is basically a noop
1124          */
1125         if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1126                 btrfs_queue_worker(&fs_info->endio_workers,
1127                                    &end_io_wq->work);
1128                 return;
1129         }
1130         error = end_io_wq->error;
1131         bio->bi_private = end_io_wq->private;
1132         bio->bi_end_io = end_io_wq->end_io;
1133         kfree(end_io_wq);
1134 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1135         bio_endio(bio, bio->bi_size, error);
1136 #else
1137         bio_endio(bio, error);
1138 #endif
1139 }
1140
1141 static int cleaner_kthread(void *arg)
1142 {
1143         struct btrfs_root *root = arg;
1144
1145         do {
1146                 smp_mb();
1147                 if (root->fs_info->closing)
1148                         break;
1149
1150                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1151                 mutex_lock(&root->fs_info->cleaner_mutex);
1152                 btrfs_clean_old_snapshots(root);
1153                 mutex_unlock(&root->fs_info->cleaner_mutex);
1154
1155                 if (freezing(current)) {
1156                         refrigerator();
1157                 } else {
1158                         smp_mb();
1159                         if (root->fs_info->closing)
1160                                 break;
1161                         set_current_state(TASK_INTERRUPTIBLE);
1162                         schedule();
1163                         __set_current_state(TASK_RUNNING);
1164                 }
1165         } while (!kthread_should_stop());
1166         return 0;
1167 }
1168
1169 static int transaction_kthread(void *arg)
1170 {
1171         struct btrfs_root *root = arg;
1172         struct btrfs_trans_handle *trans;
1173         struct btrfs_transaction *cur;
1174         unsigned long now;
1175         unsigned long delay;
1176         int ret;
1177
1178         do {
1179                 smp_mb();
1180                 if (root->fs_info->closing)
1181                         break;
1182
1183                 delay = HZ * 30;
1184                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1185                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1186
1187                 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1188                         printk("btrfs: total reference cache size %Lu\n",
1189                                 root->fs_info->total_ref_cache_size);
1190                 }
1191
1192                 mutex_lock(&root->fs_info->trans_mutex);
1193                 cur = root->fs_info->running_transaction;
1194                 if (!cur) {
1195                         mutex_unlock(&root->fs_info->trans_mutex);
1196                         goto sleep;
1197                 }
1198
1199                 now = get_seconds();
1200                 if (now < cur->start_time || now - cur->start_time < 30) {
1201                         mutex_unlock(&root->fs_info->trans_mutex);
1202                         delay = HZ * 5;
1203                         goto sleep;
1204                 }
1205                 mutex_unlock(&root->fs_info->trans_mutex);
1206                 trans = btrfs_start_transaction(root, 1);
1207                 ret = btrfs_commit_transaction(trans, root);
1208 sleep:
1209                 wake_up_process(root->fs_info->cleaner_kthread);
1210                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1211
1212                 if (freezing(current)) {
1213                         refrigerator();
1214                 } else {
1215                         if (root->fs_info->closing)
1216                                 break;
1217                         set_current_state(TASK_INTERRUPTIBLE);
1218                         schedule_timeout(delay);
1219                         __set_current_state(TASK_RUNNING);
1220                 }
1221         } while (!kthread_should_stop());
1222         return 0;
1223 }
1224
1225 struct btrfs_root *open_ctree(struct super_block *sb,
1226                               struct btrfs_fs_devices *fs_devices,
1227                               char *options)
1228 {
1229         u32 sectorsize;
1230         u32 nodesize;
1231         u32 leafsize;
1232         u32 blocksize;
1233         u32 stripesize;
1234         struct buffer_head *bh;
1235         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
1236                                                  GFP_NOFS);
1237         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
1238                                                GFP_NOFS);
1239         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1240                                                 GFP_NOFS);
1241         struct btrfs_root *chunk_root = kmalloc(sizeof(struct btrfs_root),
1242                                                 GFP_NOFS);
1243         struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
1244                                               GFP_NOFS);
1245         int ret;
1246         int err = -EINVAL;
1247
1248         struct btrfs_super_block *disk_super;
1249
1250         if (!extent_root || !tree_root || !fs_info) {
1251                 err = -ENOMEM;
1252                 goto fail;
1253         }
1254         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1255         INIT_LIST_HEAD(&fs_info->trans_list);
1256         INIT_LIST_HEAD(&fs_info->dead_roots);
1257         INIT_LIST_HEAD(&fs_info->hashers);
1258         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1259         spin_lock_init(&fs_info->hash_lock);
1260         spin_lock_init(&fs_info->delalloc_lock);
1261         spin_lock_init(&fs_info->new_trans_lock);
1262         spin_lock_init(&fs_info->ref_cache_lock);
1263
1264         init_completion(&fs_info->kobj_unregister);
1265         fs_info->tree_root = tree_root;
1266         fs_info->extent_root = extent_root;
1267         fs_info->chunk_root = chunk_root;
1268         fs_info->dev_root = dev_root;
1269         fs_info->fs_devices = fs_devices;
1270         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1271         INIT_LIST_HEAD(&fs_info->space_info);
1272         btrfs_mapping_init(&fs_info->mapping_tree);
1273         atomic_set(&fs_info->nr_async_submits, 0);
1274         atomic_set(&fs_info->nr_async_bios, 0);
1275         atomic_set(&fs_info->throttles, 0);
1276         atomic_set(&fs_info->throttle_gen, 0);
1277         fs_info->sb = sb;
1278         fs_info->max_extent = (u64)-1;
1279         fs_info->max_inline = 8192 * 1024;
1280         setup_bdi(fs_info, &fs_info->bdi);
1281         fs_info->btree_inode = new_inode(sb);
1282         fs_info->btree_inode->i_ino = 1;
1283         fs_info->btree_inode->i_nlink = 1;
1284         fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1285
1286         INIT_LIST_HEAD(&fs_info->ordered_extents);
1287         spin_lock_init(&fs_info->ordered_extent_lock);
1288
1289         sb->s_blocksize = 4096;
1290         sb->s_blocksize_bits = blksize_bits(4096);
1291
1292         /*
1293          * we set the i_size on the btree inode to the max possible int.
1294          * the real end of the address space is determined by all of
1295          * the devices in the system
1296          */
1297         fs_info->btree_inode->i_size = OFFSET_MAX;
1298         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1299         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1300
1301         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1302                              fs_info->btree_inode->i_mapping,
1303                              GFP_NOFS);
1304         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1305                              GFP_NOFS);
1306
1307         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1308
1309         extent_io_tree_init(&fs_info->free_space_cache,
1310                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1311         extent_io_tree_init(&fs_info->block_group_cache,
1312                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1313         extent_io_tree_init(&fs_info->pinned_extents,
1314                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1315         extent_io_tree_init(&fs_info->pending_del,
1316                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1317         extent_io_tree_init(&fs_info->extent_ins,
1318                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1319         fs_info->do_barriers = 1;
1320
1321         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1322         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1323                sizeof(struct btrfs_key));
1324         insert_inode_hash(fs_info->btree_inode);
1325         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1326
1327         mutex_init(&fs_info->trans_mutex);
1328         mutex_init(&fs_info->drop_mutex);
1329         mutex_init(&fs_info->alloc_mutex);
1330         mutex_init(&fs_info->chunk_mutex);
1331         mutex_init(&fs_info->transaction_kthread_mutex);
1332         mutex_init(&fs_info->cleaner_mutex);
1333         mutex_init(&fs_info->volume_mutex);
1334         init_waitqueue_head(&fs_info->transaction_throttle);
1335         init_waitqueue_head(&fs_info->transaction_wait);
1336
1337 #if 0
1338         ret = add_hasher(fs_info, "crc32c");
1339         if (ret) {
1340                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1341                 err = -ENOMEM;
1342                 goto fail_iput;
1343         }
1344 #endif
1345         __setup_root(4096, 4096, 4096, 4096, tree_root,
1346                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1347
1348
1349         bh = __bread(fs_devices->latest_bdev,
1350                      BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1351         if (!bh)
1352                 goto fail_iput;
1353
1354         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1355         brelse(bh);
1356
1357         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1358
1359         disk_super = &fs_info->super_copy;
1360         if (!btrfs_super_root(disk_super))
1361                 goto fail_sb_buffer;
1362
1363         err = btrfs_parse_options(tree_root, options);
1364         if (err)
1365                 goto fail_sb_buffer;
1366
1367         /*
1368          * we need to start all the end_io workers up front because the
1369          * queue work function gets called at interrupt time, and so it
1370          * cannot dynamically grow.
1371          */
1372         btrfs_init_workers(&fs_info->workers, "worker",
1373                            fs_info->thread_pool_size);
1374         btrfs_init_workers(&fs_info->submit_workers, "submit",
1375                            min_t(u64, fs_devices->num_devices,
1376                            fs_info->thread_pool_size));
1377
1378         /* a higher idle thresh on the submit workers makes it much more
1379          * likely that bios will be send down in a sane order to the
1380          * devices
1381          */
1382         fs_info->submit_workers.idle_thresh = 64;
1383
1384         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1385         btrfs_init_workers(&fs_info->endio_workers, "endio",
1386                            fs_info->thread_pool_size);
1387         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1388                            fs_info->thread_pool_size);
1389
1390         /*
1391          * endios are largely parallel and should have a very
1392          * low idle thresh
1393          */
1394         fs_info->endio_workers.idle_thresh = 4;
1395         fs_info->endio_write_workers.idle_thresh = 4;
1396
1397         btrfs_start_workers(&fs_info->workers, 1);
1398         btrfs_start_workers(&fs_info->submit_workers, 1);
1399         btrfs_start_workers(&fs_info->fixup_workers, 1);
1400         btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1401         btrfs_start_workers(&fs_info->endio_write_workers,
1402                             fs_info->thread_pool_size);
1403
1404         err = -EINVAL;
1405         if (btrfs_super_num_devices(disk_super) > fs_devices->open_devices) {
1406                 printk("Btrfs: wanted %llu devices, but found %llu\n",
1407                        (unsigned long long)btrfs_super_num_devices(disk_super),
1408                        (unsigned long long)fs_devices->open_devices);
1409                 if (btrfs_test_opt(tree_root, DEGRADED))
1410                         printk("continuing in degraded mode\n");
1411                 else {
1412                         goto fail_sb_buffer;
1413                 }
1414         }
1415
1416         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1417
1418         nodesize = btrfs_super_nodesize(disk_super);
1419         leafsize = btrfs_super_leafsize(disk_super);
1420         sectorsize = btrfs_super_sectorsize(disk_super);
1421         stripesize = btrfs_super_stripesize(disk_super);
1422         tree_root->nodesize = nodesize;
1423         tree_root->leafsize = leafsize;
1424         tree_root->sectorsize = sectorsize;
1425         tree_root->stripesize = stripesize;
1426
1427         sb->s_blocksize = sectorsize;
1428         sb->s_blocksize_bits = blksize_bits(sectorsize);
1429
1430         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1431                     sizeof(disk_super->magic))) {
1432                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1433                 goto fail_sb_buffer;
1434         }
1435
1436         mutex_lock(&fs_info->chunk_mutex);
1437         ret = btrfs_read_sys_array(tree_root);
1438         mutex_unlock(&fs_info->chunk_mutex);
1439         if (ret) {
1440                 printk("btrfs: failed to read the system array on %s\n",
1441                        sb->s_id);
1442                 goto fail_sys_array;
1443         }
1444
1445         blocksize = btrfs_level_size(tree_root,
1446                                      btrfs_super_chunk_root_level(disk_super));
1447
1448         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1449                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1450
1451         chunk_root->node = read_tree_block(chunk_root,
1452                                            btrfs_super_chunk_root(disk_super),
1453                                            blocksize, 0);
1454         BUG_ON(!chunk_root->node);
1455
1456         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1457                  (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1458                  BTRFS_UUID_SIZE);
1459
1460         mutex_lock(&fs_info->chunk_mutex);
1461         ret = btrfs_read_chunk_tree(chunk_root);
1462         mutex_unlock(&fs_info->chunk_mutex);
1463         BUG_ON(ret);
1464
1465         btrfs_close_extra_devices(fs_devices);
1466
1467         blocksize = btrfs_level_size(tree_root,
1468                                      btrfs_super_root_level(disk_super));
1469
1470
1471         tree_root->node = read_tree_block(tree_root,
1472                                           btrfs_super_root(disk_super),
1473                                           blocksize, 0);
1474         if (!tree_root->node)
1475                 goto fail_sb_buffer;
1476
1477
1478         ret = find_and_setup_root(tree_root, fs_info,
1479                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1480         if (ret)
1481                 goto fail_tree_root;
1482         extent_root->track_dirty = 1;
1483
1484         ret = find_and_setup_root(tree_root, fs_info,
1485                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1486         dev_root->track_dirty = 1;
1487
1488         if (ret)
1489                 goto fail_extent_root;
1490
1491         btrfs_read_block_groups(extent_root);
1492
1493         fs_info->generation = btrfs_super_generation(disk_super) + 1;
1494         fs_info->data_alloc_profile = (u64)-1;
1495         fs_info->metadata_alloc_profile = (u64)-1;
1496         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1497         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1498                                                "btrfs-cleaner");
1499         if (!fs_info->cleaner_kthread)
1500                 goto fail_extent_root;
1501
1502         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1503                                                    tree_root,
1504                                                    "btrfs-transaction");
1505         if (!fs_info->transaction_kthread)
1506                 goto fail_cleaner;
1507
1508
1509         return tree_root;
1510
1511 fail_cleaner:
1512         kthread_stop(fs_info->cleaner_kthread);
1513 fail_extent_root:
1514         free_extent_buffer(extent_root->node);
1515 fail_tree_root:
1516         free_extent_buffer(tree_root->node);
1517 fail_sys_array:
1518 fail_sb_buffer:
1519         btrfs_stop_workers(&fs_info->fixup_workers);
1520         btrfs_stop_workers(&fs_info->workers);
1521         btrfs_stop_workers(&fs_info->endio_workers);
1522         btrfs_stop_workers(&fs_info->endio_write_workers);
1523         btrfs_stop_workers(&fs_info->submit_workers);
1524 fail_iput:
1525         iput(fs_info->btree_inode);
1526 fail:
1527         btrfs_close_devices(fs_info->fs_devices);
1528         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1529
1530         kfree(extent_root);
1531         kfree(tree_root);
1532 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1533         bdi_destroy(&fs_info->bdi);
1534 #endif
1535         kfree(fs_info);
1536         return ERR_PTR(err);
1537 }
1538
1539 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1540 {
1541         char b[BDEVNAME_SIZE];
1542
1543         if (uptodate) {
1544                 set_buffer_uptodate(bh);
1545         } else {
1546                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1547                         printk(KERN_WARNING "lost page write due to "
1548                                         "I/O error on %s\n",
1549                                        bdevname(bh->b_bdev, b));
1550                 }
1551                 /* note, we dont' set_buffer_write_io_error because we have
1552                  * our own ways of dealing with the IO errors
1553                  */
1554                 clear_buffer_uptodate(bh);
1555         }
1556         unlock_buffer(bh);
1557         put_bh(bh);
1558 }
1559
1560 int write_all_supers(struct btrfs_root *root)
1561 {
1562         struct list_head *cur;
1563         struct list_head *head = &root->fs_info->fs_devices->devices;
1564         struct btrfs_device *dev;
1565         struct btrfs_super_block *sb;
1566         struct btrfs_dev_item *dev_item;
1567         struct buffer_head *bh;
1568         int ret;
1569         int do_barriers;
1570         int max_errors;
1571         int total_errors = 0;
1572         u32 crc;
1573         u64 flags;
1574
1575         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1576         do_barriers = !btrfs_test_opt(root, NOBARRIER);
1577
1578         sb = &root->fs_info->super_for_commit;
1579         dev_item = &sb->dev_item;
1580         list_for_each(cur, head) {
1581                 dev = list_entry(cur, struct btrfs_device, dev_list);
1582                 if (!dev->bdev) {
1583                         total_errors++;
1584                         continue;
1585                 }
1586                 if (!dev->in_fs_metadata)
1587                         continue;
1588
1589                 btrfs_set_stack_device_type(dev_item, dev->type);
1590                 btrfs_set_stack_device_id(dev_item, dev->devid);
1591                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1592                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1593                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1594                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1595                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1596                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1597                 flags = btrfs_super_flags(sb);
1598                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1599
1600
1601                 crc = ~(u32)0;
1602                 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1603                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1604                 btrfs_csum_final(crc, sb->csum);
1605
1606                 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1607                               BTRFS_SUPER_INFO_SIZE);
1608
1609                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1610                 dev->pending_io = bh;
1611
1612                 get_bh(bh);
1613                 set_buffer_uptodate(bh);
1614                 lock_buffer(bh);
1615                 bh->b_end_io = btrfs_end_buffer_write_sync;
1616
1617                 if (do_barriers && dev->barriers) {
1618                         ret = submit_bh(WRITE_BARRIER, bh);
1619                         if (ret == -EOPNOTSUPP) {
1620                                 printk("btrfs: disabling barriers on dev %s\n",
1621                                        dev->name);
1622                                 set_buffer_uptodate(bh);
1623                                 dev->barriers = 0;
1624                                 get_bh(bh);
1625                                 lock_buffer(bh);
1626                                 ret = submit_bh(WRITE, bh);
1627                         }
1628                 } else {
1629                         ret = submit_bh(WRITE, bh);
1630                 }
1631                 if (ret)
1632                         total_errors++;
1633         }
1634         if (total_errors > max_errors) {
1635                 printk("btrfs: %d errors while writing supers\n", total_errors);
1636                 BUG();
1637         }
1638         total_errors = 0;
1639
1640         list_for_each(cur, head) {
1641                 dev = list_entry(cur, struct btrfs_device, dev_list);
1642                 if (!dev->bdev)
1643                         continue;
1644                 if (!dev->in_fs_metadata)
1645                         continue;
1646
1647                 BUG_ON(!dev->pending_io);
1648                 bh = dev->pending_io;
1649                 wait_on_buffer(bh);
1650                 if (!buffer_uptodate(dev->pending_io)) {
1651                         if (do_barriers && dev->barriers) {
1652                                 printk("btrfs: disabling barriers on dev %s\n",
1653                                        dev->name);
1654                                 set_buffer_uptodate(bh);
1655                                 get_bh(bh);
1656                                 lock_buffer(bh);
1657                                 dev->barriers = 0;
1658                                 ret = submit_bh(WRITE, bh);
1659                                 BUG_ON(ret);
1660                                 wait_on_buffer(bh);
1661                                 if (!buffer_uptodate(bh))
1662                                         total_errors++;
1663                         } else {
1664                                 total_errors++;
1665                         }
1666
1667                 }
1668                 dev->pending_io = NULL;
1669                 brelse(bh);
1670         }
1671         if (total_errors > max_errors) {
1672                 printk("btrfs: %d errors while writing supers\n", total_errors);
1673                 BUG();
1674         }
1675         return 0;
1676 }
1677
1678 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1679                       *root)
1680 {
1681         int ret;
1682
1683         ret = write_all_supers(root);
1684         return ret;
1685 }
1686
1687 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1688 {
1689         radix_tree_delete(&fs_info->fs_roots_radix,
1690                           (unsigned long)root->root_key.objectid);
1691         if (root->in_sysfs)
1692                 btrfs_sysfs_del_root(root);
1693         if (root->inode)
1694                 iput(root->inode);
1695         if (root->node)
1696                 free_extent_buffer(root->node);
1697         if (root->commit_root)
1698                 free_extent_buffer(root->commit_root);
1699         if (root->name)
1700                 kfree(root->name);
1701         kfree(root);
1702         return 0;
1703 }
1704
1705 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1706 {
1707         int ret;
1708         struct btrfs_root *gang[8];
1709         int i;
1710
1711         while(1) {
1712                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1713                                              (void **)gang, 0,
1714                                              ARRAY_SIZE(gang));
1715                 if (!ret)
1716                         break;
1717                 for (i = 0; i < ret; i++)
1718                         btrfs_free_fs_root(fs_info, gang[i]);
1719         }
1720         return 0;
1721 }
1722
1723 int close_ctree(struct btrfs_root *root)
1724 {
1725         int ret;
1726         struct btrfs_trans_handle *trans;
1727         struct btrfs_fs_info *fs_info = root->fs_info;
1728
1729         fs_info->closing = 1;
1730         smp_mb();
1731
1732         kthread_stop(root->fs_info->transaction_kthread);
1733         kthread_stop(root->fs_info->cleaner_kthread);
1734
1735         btrfs_clean_old_snapshots(root);
1736         trans = btrfs_start_transaction(root, 1);
1737         ret = btrfs_commit_transaction(trans, root);
1738         /* run commit again to  drop the original snapshot */
1739         trans = btrfs_start_transaction(root, 1);
1740         btrfs_commit_transaction(trans, root);
1741         ret = btrfs_write_and_wait_transaction(NULL, root);
1742         BUG_ON(ret);
1743
1744         write_ctree_super(NULL, root);
1745
1746         if (fs_info->delalloc_bytes) {
1747                 printk("btrfs: at unmount delalloc count %Lu\n",
1748                        fs_info->delalloc_bytes);
1749         }
1750         if (fs_info->total_ref_cache_size) {
1751                 printk("btrfs: at umount reference cache size %Lu\n",
1752                         fs_info->total_ref_cache_size);
1753         }
1754
1755         if (fs_info->extent_root->node)
1756                 free_extent_buffer(fs_info->extent_root->node);
1757
1758         if (fs_info->tree_root->node)
1759                 free_extent_buffer(fs_info->tree_root->node);
1760
1761         if (root->fs_info->chunk_root->node);
1762                 free_extent_buffer(root->fs_info->chunk_root->node);
1763
1764         if (root->fs_info->dev_root->node);
1765                 free_extent_buffer(root->fs_info->dev_root->node);
1766
1767         btrfs_free_block_groups(root->fs_info);
1768         fs_info->closing = 2;
1769         del_fs_roots(fs_info);
1770
1771         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1772
1773         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1774
1775         btrfs_stop_workers(&fs_info->fixup_workers);
1776         btrfs_stop_workers(&fs_info->workers);
1777         btrfs_stop_workers(&fs_info->endio_workers);
1778         btrfs_stop_workers(&fs_info->endio_write_workers);
1779         btrfs_stop_workers(&fs_info->submit_workers);
1780
1781         iput(fs_info->btree_inode);
1782 #if 0
1783         while(!list_empty(&fs_info->hashers)) {
1784                 struct btrfs_hasher *hasher;
1785                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1786                                     hashers);
1787                 list_del(&hasher->hashers);
1788                 crypto_free_hash(&fs_info->hash_tfm);
1789                 kfree(hasher);
1790         }
1791 #endif
1792         btrfs_close_devices(fs_info->fs_devices);
1793         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1794
1795 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1796         bdi_destroy(&fs_info->bdi);
1797 #endif
1798
1799         kfree(fs_info->extent_root);
1800         kfree(fs_info->tree_root);
1801         kfree(fs_info->chunk_root);
1802         kfree(fs_info->dev_root);
1803         return 0;
1804 }
1805
1806 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1807 {
1808         int ret;
1809         struct inode *btree_inode = buf->first_page->mapping->host;
1810
1811         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1812         if (!ret)
1813                 return ret;
1814
1815         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
1816                                     parent_transid);
1817         return !ret;
1818 }
1819
1820 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1821 {
1822         struct inode *btree_inode = buf->first_page->mapping->host;
1823         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1824                                           buf);
1825 }
1826
1827 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1828 {
1829         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1830         u64 transid = btrfs_header_generation(buf);
1831         struct inode *btree_inode = root->fs_info->btree_inode;
1832
1833         WARN_ON(!btrfs_tree_locked(buf));
1834         if (transid != root->fs_info->generation) {
1835                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1836                         (unsigned long long)buf->start,
1837                         transid, root->fs_info->generation);
1838                 WARN_ON(1);
1839         }
1840         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1841 }
1842
1843 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1844 {
1845         /*
1846          * looks as though older kernels can get into trouble with
1847          * this code, they end up stuck in balance_dirty_pages forever
1848          */
1849         struct extent_io_tree *tree;
1850         u64 num_dirty;
1851         u64 start = 0;
1852         unsigned long thresh = 2 * 1024 * 1024;
1853         tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
1854
1855         if (current_is_pdflush())
1856                 return;
1857
1858         num_dirty = count_range_bits(tree, &start, (u64)-1,
1859                                      thresh, EXTENT_DIRTY);
1860         if (num_dirty > thresh) {
1861                 balance_dirty_pages_ratelimited_nr(
1862                                    root->fs_info->btree_inode->i_mapping, 1);
1863         }
1864         return;
1865 }
1866
1867 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
1868 {
1869         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1870         int ret;
1871         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1872         if (ret == 0) {
1873                 buf->flags |= EXTENT_UPTODATE;
1874         }
1875         return ret;
1876 }
1877
1878 static struct extent_io_ops btree_extent_io_ops = {
1879         .writepage_io_hook = btree_writepage_io_hook,
1880         .readpage_end_io_hook = btree_readpage_end_io_hook,
1881         .submit_bio_hook = btree_submit_bio_hook,
1882         /* note we're sharing with inode.c for the merge bio hook */
1883         .merge_bio_hook = btrfs_merge_bio_hook,
1884 };