fs/btrfs/disk-io.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/version.h>
  20 #include <linux/fs.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/scatterlist.h>
  23 #include <linux/swap.h>
  24 #include <linux/radix-tree.h>
  25 #include <linux/writeback.h>
  26 #include <linux/buffer_head.h> // for block_sync_page
  27 #include <linux/workqueue.h>
  28 #include <linux/kthread.h>
  29 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,20)
  30 # include <linux/freezer.h>
  31 #else
  32 # include <linux/sched.h>
  33 #endif
  34 #include "crc32c.h"
  35 #include "ctree.h"
  36 #include "disk-io.h"
  37 #include "transaction.h"
  38 #include "btrfs_inode.h"
  39 #include "volumes.h"
  40 #include "print-tree.h"
  41 #include "async-thread.h"
  42 #include "locking.h"
  43
  44 #if 0
  45 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
  46 {
  47         if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
  48                 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
  49                        (unsigned long long)extent_buffer_blocknr(buf),
  50                        (unsigned long long)btrfs_header_blocknr(buf));
  51                 return 1;
  52         }
  53         return 0;
  54 }
  55 #endif
  56
  57 static struct extent_io_ops btree_extent_io_ops;
  58 static void end_workqueue_fn(struct btrfs_work *work);
  59
  60 struct end_io_wq {
  61         struct bio *bio;
  62         bio_end_io_t *end_io;
  63         void *private;
  64         struct btrfs_fs_info *info;
  65         int error;
  66         int metadata;
  67         struct list_head list;
  68         struct btrfs_work work;
  69 };
  70
  71 struct async_submit_bio {
  72         struct inode *inode;
  73         struct bio *bio;
  74         struct list_head list;
  75         extent_submit_bio_hook_t *submit_bio_hook;
  76         int rw;
  77         int mirror_num;
  78         struct btrfs_work work;
  79 };
  80
  81 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
  82                                     size_t page_offset, u64 start, u64 len,
  83                                     int create)
  84 {
  85         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
  86         struct extent_map *em;
  87         int ret;
  88
  89         spin_lock(&em_tree->lock);
  90         em = lookup_extent_mapping(em_tree, start, len);
  91         if (em) {
  92                 em->bdev =
  93                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
  94                 spin_unlock(&em_tree->lock);
  95                 goto out;
  96         }
  97         spin_unlock(&em_tree->lock);
  98
  99         em = alloc_extent_map(GFP_NOFS);
 100         if (!em) {
 101                 em = ERR_PTR(-ENOMEM);
 102                 goto out;
 103         }
 104         em->start = 0;
 105         em->len = (u64)-1;
 106         em->block_start = 0;
 107         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 108
 109         spin_lock(&em_tree->lock);
 110         ret = add_extent_mapping(em_tree, em);
 111         if (ret == -EEXIST) {
 112                 u64 failed_start = em->start;
 113                 u64 failed_len = em->len;
 114
 115                 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
 116                        em->start, em->len, em->block_start);
 117                 free_extent_map(em);
 118                 em = lookup_extent_mapping(em_tree, start, len);
 119                 if (em) {
 120                         printk("after failing, found %Lu %Lu %Lu\n",
 121                                em->start, em->len, em->block_start);
 122                         ret = 0;
 123                 } else {
 124                         em = lookup_extent_mapping(em_tree, failed_start,
 125                                                    failed_len);
 126                         if (em) {
 127                                 printk("double failure lookup gives us "
 128                                        "%Lu %Lu -> %Lu\n", em->start,
 129                                        em->len, em->block_start);
 130                                 free_extent_map(em);
 131                         }
 132                         ret = -EIO;
 133                 }
 134         } else if (ret) {
 135                 free_extent_map(em);
 136                 em = NULL;
 137         }
 138         spin_unlock(&em_tree->lock);
 139
 140         if (ret)
 141                 em = ERR_PTR(ret);
 142 out:
 143         return em;
 144 }
 145
 146 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 147 {
 148         return btrfs_crc32c(seed, data, len);
 149 }
 150
 151 void btrfs_csum_final(u32 crc, char *result)
 152 {
 153         *(__le32 *)result = ~cpu_to_le32(crc);
 154 }
 155
 156 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 157                            int verify)
 158 {
 159         char result[BTRFS_CRC32_SIZE];
 160         unsigned long len;
 161         unsigned long cur_len;
 162         unsigned long offset = BTRFS_CSUM_SIZE;
 163         char *map_token = NULL;
 164         char *kaddr;
 165         unsigned long map_start;
 166         unsigned long map_len;
 167         int err;
 168         u32 crc = ~(u32)0;
 169
 170         len = buf->len - offset;
 171         while(len > 0) {
 172                 err = map_private_extent_buffer(buf, offset, 32,
 173                                         &map_token, &kaddr,
 174                                         &map_start, &map_len, KM_USER0);
 175                 if (err) {
 176                         printk("failed to map extent buffer! %lu\n",
 177                                offset);
 178                         return 1;
 179                 }
 180                 cur_len = min(len, map_len - (offset - map_start));
 181                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 182                                       crc, cur_len);
 183                 len -= cur_len;
 184                 offset += cur_len;
 185                 unmap_extent_buffer(buf, map_token, KM_USER0);
 186         }
 187         btrfs_csum_final(crc, result);
 188
 189         if (verify) {
 190                 int from_this_trans = 0;
 191
 192                 if (root->fs_info->running_transaction &&
 193                     btrfs_header_generation(buf) ==
 194                     root->fs_info->running_transaction->transid)
 195                         from_this_trans = 1;
 196
 197                 /* FIXME, this is not good */
 198                 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
 199                         u32 val;
 200                         u32 found = 0;
 201                         memcpy(&found, result, BTRFS_CRC32_SIZE);
 202
 203                         read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
 204                         printk("btrfs: %s checksum verify failed on %llu "
 205                                "wanted %X found %X from_this_trans %d "
 206                                "level %d\n",
 207                                root->fs_info->sb->s_id,
 208                                buf->start, val, found, from_this_trans,
 209                                btrfs_header_level(buf));
 210                         return 1;
 211                 }
 212         } else {
 213                 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
 214         }
 215         return 0;
 216 }
 217
 218 static int verify_parent_transid(struct extent_io_tree *io_tree,
 219                                  struct extent_buffer *eb, u64 parent_transid)
 220 {
 221         int ret;
 222
 223         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 224                 return 0;
 225
 226         lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
 227         if (extent_buffer_uptodate(io_tree, eb) &&
 228             btrfs_header_generation(eb) == parent_transid) {
 229                 ret = 0;
 230                 goto out;
 231         }
 232         printk("parent transid verify failed on %llu wanted %llu found %llu\n",
 233                (unsigned long long)eb->start,
 234                (unsigned long long)parent_transid,
 235                (unsigned long long)btrfs_header_generation(eb));
 236         ret = 1;
 237 out:
 238         clear_extent_buffer_uptodate(io_tree, eb);
 239         unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
 240                       GFP_NOFS);
 241         return ret;
 242
 243 }
 244
 245 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 246                                           struct extent_buffer *eb,
 247                                           u64 start, u64 parent_transid)
 248 {
 249         struct extent_io_tree *io_tree;
 250         int ret;
 251         int num_copies = 0;
 252         int mirror_num = 0;
 253
 254         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 255         while (1) {
 256                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 257                                                btree_get_extent, mirror_num);
 258                 if (!ret &&
 259                     !verify_parent_transid(io_tree, eb, parent_transid))
 260                         return ret;
 261
 262                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 263                                               eb->start, eb->len);
 264                 if (num_copies == 1)
 265                         return ret;
 266
 267                 mirror_num++;
 268                 if (mirror_num > num_copies)
 269                         return ret;
 270         }
 271         return -EIO;
 272 }
 273
 274 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 275 {
 276         struct extent_io_tree *tree;
 277         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 278         u64 found_start;
 279         int found_level;
 280         unsigned long len;
 281         struct extent_buffer *eb;
 282         int ret;
 283
 284         tree = &BTRFS_I(page->mapping->host)->io_tree;
 285
 286         if (page->private == EXTENT_PAGE_PRIVATE)
 287                 goto out;
 288         if (!page->private)
 289                 goto out;
 290         len = page->private >> 2;
 291         if (len == 0) {
 292                 WARN_ON(1);
 293         }
 294         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 295         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 296                                              btrfs_header_generation(eb));
 297         BUG_ON(ret);
 298         found_start = btrfs_header_bytenr(eb);
 299         if (found_start != start) {
 300                 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
 301                        start, found_start, len);
 302                 WARN_ON(1);
 303                 goto err;
 304         }
 305         if (eb->first_page != page) {
 306                 printk("bad first page %lu %lu\n", eb->first_page->index,
 307                        page->index);
 308                 WARN_ON(1);
 309                 goto err;
 310         }
 311         if (!PageUptodate(page)) {
 312                 printk("csum not up to date page %lu\n", page->index);
 313                 WARN_ON(1);
 314                 goto err;
 315         }
 316         found_level = btrfs_header_level(eb);
 317         spin_lock(&root->fs_info->hash_lock);
 318         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
 319         spin_unlock(&root->fs_info->hash_lock);
 320         csum_tree_block(root, eb, 0);
 321 err:
 322         free_extent_buffer(eb);
 323 out:
 324         return 0;
 325 }
 326
 327 static int btree_writepage_io_hook(struct page *page, u64 start, u64 end)
 328 {
 329         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 330
 331         csum_dirty_buffer(root, page);
 332         return 0;
 333 }
 334
 335 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 336                                struct extent_state *state)
 337 {
 338         struct extent_io_tree *tree;
 339         u64 found_start;
 340         int found_level;
 341         unsigned long len;
 342         struct extent_buffer *eb;
 343         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 344         int ret = 0;
 345
 346         tree = &BTRFS_I(page->mapping->host)->io_tree;
 347         if (page->private == EXTENT_PAGE_PRIVATE)
 348                 goto out;
 349         if (!page->private)
 350                 goto out;
 351         len = page->private >> 2;
 352         if (len == 0) {
 353                 WARN_ON(1);
 354         }
 355         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 356
 357         found_start = btrfs_header_bytenr(eb);
 358         if (found_start != start) {
 359                 ret = -EIO;
 360                 goto err;
 361         }
 362         if (eb->first_page != page) {
 363                 printk("bad first page %lu %lu\n", eb->first_page->index,
 364                        page->index);
 365                 WARN_ON(1);
 366                 ret = -EIO;
 367                 goto err;
 368         }
 369         if (memcmp_extent_buffer(eb, root->fs_info->fsid,
 370                                  (unsigned long)btrfs_header_fsid(eb),
 371                                  BTRFS_FSID_SIZE)) {
 372                 printk("bad fsid on block %Lu\n", eb->start);
 373                 ret = -EIO;
 374                 goto err;
 375         }
 376         found_level = btrfs_header_level(eb);
 377
 378         ret = csum_tree_block(root, eb, 1);
 379         if (ret)
 380                 ret = -EIO;
 381
 382         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 383         end = eb->start + end - 1;
 384         release_extent_buffer_tail_pages(eb);
 385 err:
 386         free_extent_buffer(eb);
 387 out:
 388         return ret;
 389 }
 390
 391 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
 392 static void end_workqueue_bio(struct bio *bio, int err)
 393 #else
 394 static int end_workqueue_bio(struct bio *bio,
 395                                    unsigned int bytes_done, int err)
 396 #endif
 397 {
 398         struct end_io_wq *end_io_wq = bio->bi_private;
 399         struct btrfs_fs_info *fs_info;
 400
 401 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 402         if (bio->bi_size)
 403                 return 1;
 404 #endif
 405
 406         fs_info = end_io_wq->info;
 407         end_io_wq->error = err;
 408         end_io_wq->work.func = end_workqueue_fn;
 409         end_io_wq->work.flags = 0;
 410         if (bio->bi_rw & (1 << BIO_RW))
 411                 btrfs_queue_worker(&fs_info->endio_write_workers,
 412                                    &end_io_wq->work);
 413         else
 414                 btrfs_queue_worker(&fs_info->endio_workers, &end_io_wq->work);
 415
 416 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 417         return 0;
 418 #endif
 419 }
 420
 421 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 422                         int metadata)
 423 {
 424         struct end_io_wq *end_io_wq;
 425         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 426         if (!end_io_wq)
 427                 return -ENOMEM;
 428
 429         end_io_wq->private = bio->bi_private;
 430         end_io_wq->end_io = bio->bi_end_io;
 431         end_io_wq->info = info;
 432         end_io_wq->error = 0;
 433         end_io_wq->bio = bio;
 434         end_io_wq->metadata = metadata;
 435
 436         bio->bi_private = end_io_wq;
 437         bio->bi_end_io = end_workqueue_bio;
 438         return 0;
 439 }
 440
 441 static void run_one_async_submit(struct btrfs_work *work)
 442 {
 443         struct btrfs_fs_info *fs_info;
 444         struct async_submit_bio *async;
 445
 446         async = container_of(work, struct  async_submit_bio, work);
 447         fs_info = BTRFS_I(async->inode)->root->fs_info;
 448         atomic_dec(&fs_info->nr_async_submits);
 449         async->submit_bio_hook(async->inode, async->rw, async->bio,
 450                                async->mirror_num);
 451         kfree(async);
 452 }
 453
 454 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 455                         int rw, struct bio *bio, int mirror_num,
 456                         extent_submit_bio_hook_t *submit_bio_hook)
 457 {
 458         struct async_submit_bio *async;
 459
 460         async = kmalloc(sizeof(*async), GFP_NOFS);
 461         if (!async)
 462                 return -ENOMEM;
 463
 464         async->inode = inode;
 465         async->rw = rw;
 466         async->bio = bio;
 467         async->mirror_num = mirror_num;
 468         async->submit_bio_hook = submit_bio_hook;
 469         async->work.func = run_one_async_submit;
 470         async->work.flags = 0;
 471         atomic_inc(&fs_info->nr_async_submits);
 472         btrfs_queue_worker(&fs_info->workers, &async->work);
 473         return 0;
 474 }
 475
 476 static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 477                                  int mirror_num)
 478 {
 479         struct btrfs_root *root = BTRFS_I(inode)->root;
 480         u64 offset;
 481         int ret;
 482
 483         offset = bio->bi_sector << 9;
 484
 485         /*
 486          * when we're called for a write, we're already in the async
 487          * submission context.  Just jump ingo btrfs_map_bio
 488          */
 489         if (rw & (1 << BIO_RW)) {
 490                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 491                                      mirror_num, 0);
 492         }
 493
 494         /*
 495          * called for a read, do the setup so that checksum validation
 496          * can happen in the async kernel threads
 497          */
 498         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
 499         BUG_ON(ret);
 500
 501         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 502 }
 503
 504 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 505                                  int mirror_num)
 506 {
 507         /*
 508          * kthread helpers are used to submit writes so that checksumming
 509          * can happen in parallel across all CPUs
 510          */
 511         if (!(rw & (1 << BIO_RW))) {
 512                 return __btree_submit_bio_hook(inode, rw, bio, mirror_num);
 513         }
 514         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 515                                    inode, rw, bio, mirror_num,
 516                                    __btree_submit_bio_hook);
 517 }
 518
 519 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 520 {
 521         struct extent_io_tree *tree;
 522         tree = &BTRFS_I(page->mapping->host)->io_tree;
 523         return extent_write_full_page(tree, page, btree_get_extent, wbc);
 524 }
 525
 526 static int btree_writepages(struct address_space *mapping,
 527                             struct writeback_control *wbc)
 528 {
 529         struct extent_io_tree *tree;
 530         tree = &BTRFS_I(mapping->host)->io_tree;
 531         if (wbc->sync_mode == WB_SYNC_NONE) {
 532                 u64 num_dirty;
 533                 u64 start = 0;
 534                 unsigned long thresh = 96 * 1024 * 1024;
 535
 536                 if (wbc->for_kupdate)
 537                         return 0;
 538
 539                 if (current_is_pdflush()) {
 540                         thresh = 96 * 1024 * 1024;
 541                 } else {
 542                         thresh = 8 * 1024 * 1024;
 543                 }
 544                 num_dirty = count_range_bits(tree, &start, (u64)-1,
 545                                              thresh, EXTENT_DIRTY);
 546                 if (num_dirty < thresh) {
 547                         return 0;
 548                 }
 549         }
 550         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 551 }
 552
 553 int btree_readpage(struct file *file, struct page *page)
 554 {
 555         struct extent_io_tree *tree;
 556         tree = &BTRFS_I(page->mapping->host)->io_tree;
 557         return extent_read_full_page(tree, page, btree_get_extent);
 558 }
 559
 560 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 561 {
 562         struct extent_io_tree *tree;
 563         struct extent_map_tree *map;
 564         int ret;
 565
 566         if (page_count(page) > 3) {
 567                 /* once for page->private, once for the caller, once
 568                  * once for the page cache
 569                  */
 570                 return 0;
 571         }
 572         tree = &BTRFS_I(page->mapping->host)->io_tree;
 573         map = &BTRFS_I(page->mapping->host)->extent_tree;
 574         ret = try_release_extent_state(map, tree, page, gfp_flags);
 575         if (ret == 1) {
 576                 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
 577                 ClearPagePrivate(page);
 578                 set_page_private(page, 0);
 579                 page_cache_release(page);
 580         }
 581         return ret;
 582 }
 583
 584 static void btree_invalidatepage(struct page *page, unsigned long offset)
 585 {
 586         struct extent_io_tree *tree;
 587         tree = &BTRFS_I(page->mapping->host)->io_tree;
 588         extent_invalidatepage(tree, page, offset);
 589         btree_releasepage(page, GFP_NOFS);
 590         if (PagePrivate(page)) {
 591                 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
 592                 ClearPagePrivate(page);
 593                 set_page_private(page, 0);
 594                 page_cache_release(page);
 595         }
 596 }
 597
 598 #if 0
 599 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 600 {
 601         struct buffer_head *bh;
 602         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 603         struct buffer_head *head;
 604         if (!page_has_buffers(page)) {
 605                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 606                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 607         }
 608         head = page_buffers(page);
 609         bh = head;
 610         do {
 611                 if (buffer_dirty(bh))
 612                         csum_tree_block(root, bh, 0);
 613                 bh = bh->b_this_page;
 614         } while (bh != head);
 615         return block_write_full_page(page, btree_get_block, wbc);
 616 }
 617 #endif
 618
 619 static struct address_space_operations btree_aops = {
 620         .readpage       = btree_readpage,
 621         .writepage      = btree_writepage,
 622         .writepages     = btree_writepages,
 623         .releasepage    = btree_releasepage,
 624         .invalidatepage = btree_invalidatepage,
 625         .sync_page      = block_sync_page,
 626 };
 627
 628 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 629                          u64 parent_transid)
 630 {
 631         struct extent_buffer *buf = NULL;
 632         struct inode *btree_inode = root->fs_info->btree_inode;
 633         int ret = 0;
 634
 635         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 636         if (!buf)
 637                 return 0;
 638         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 639                                  buf, 0, 0, btree_get_extent, 0);
 640         free_extent_buffer(buf);
 641         return ret;
 642 }
 643
 644 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 645                                             u64 bytenr, u32 blocksize)
 646 {
 647         struct inode *btree_inode = root->fs_info->btree_inode;
 648         struct extent_buffer *eb;
 649         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 650                                 bytenr, blocksize, GFP_NOFS);
 651         return eb;
 652 }
 653
 654 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 655                                                  u64 bytenr, u32 blocksize)
 656 {
 657         struct inode *btree_inode = root->fs_info->btree_inode;
 658         struct extent_buffer *eb;
 659
 660         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 661                                  bytenr, blocksize, NULL, GFP_NOFS);
 662         return eb;
 663 }
 664
 665
 666 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 667                                       u32 blocksize, u64 parent_transid)
 668 {
 669         struct extent_buffer *buf = NULL;
 670         struct inode *btree_inode = root->fs_info->btree_inode;
 671         struct extent_io_tree *io_tree;
 672         int ret;
 673
 674         io_tree = &BTRFS_I(btree_inode)->io_tree;
 675
 676         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 677         if (!buf)
 678                 return NULL;
 679
 680         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 681
 682         if (ret == 0) {
 683                 buf->flags |= EXTENT_UPTODATE;
 684         }
 685         return buf;
 686
 687 }
 688
 689 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 690                      struct extent_buffer *buf)
 691 {
 692         struct inode *btree_inode = root->fs_info->btree_inode;
 693         if (btrfs_header_generation(buf) ==
 694             root->fs_info->running_transaction->transid) {
 695                 WARN_ON(!btrfs_tree_locked(buf));
 696                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 697                                           buf);
 698         }
 699         return 0;
 700 }
 701
 702 int wait_on_tree_block_writeback(struct btrfs_root *root,
 703                                  struct extent_buffer *buf)
 704 {
 705         struct inode *btree_inode = root->fs_info->btree_inode;
 706         wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
 707                                         buf);
 708         return 0;
 709 }
 710
 711 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 712                         u32 stripesize, struct btrfs_root *root,
 713                         struct btrfs_fs_info *fs_info,
 714                         u64 objectid)
 715 {
 716         root->node = NULL;
 717         root->inode = NULL;
 718         root->commit_root = NULL;
 719         root->sectorsize = sectorsize;
 720         root->nodesize = nodesize;
 721         root->leafsize = leafsize;
 722         root->stripesize = stripesize;
 723         root->ref_cows = 0;
 724         root->track_dirty = 0;
 725
 726         root->fs_info = fs_info;
 727         root->objectid = objectid;
 728         root->last_trans = 0;
 729         root->highest_inode = 0;
 730         root->last_inode_alloc = 0;
 731         root->name = NULL;
 732         root->in_sysfs = 0;
 733
 734         INIT_LIST_HEAD(&root->dirty_list);
 735         spin_lock_init(&root->node_lock);
 736         mutex_init(&root->objectid_mutex);
 737         memset(&root->root_key, 0, sizeof(root->root_key));
 738         memset(&root->root_item, 0, sizeof(root->root_item));
 739         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 740         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 741         root->defrag_trans_start = fs_info->generation;
 742         init_completion(&root->kobj_unregister);
 743         root->defrag_running = 0;
 744         root->defrag_level = 0;
 745         root->root_key.objectid = objectid;
 746         return 0;
 747 }
 748
 749 static int find_and_setup_root(struct btrfs_root *tree_root,
 750                                struct btrfs_fs_info *fs_info,
 751                                u64 objectid,
 752                                struct btrfs_root *root)
 753 {
 754         int ret;
 755         u32 blocksize;
 756
 757         __setup_root(tree_root->nodesize, tree_root->leafsize,
 758                      tree_root->sectorsize, tree_root->stripesize,
 759                      root, fs_info, objectid);
 760         ret = btrfs_find_last_root(tree_root, objectid,
 761                                    &root->root_item, &root->root_key);
 762         BUG_ON(ret);
 763
 764         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 765         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 766                                      blocksize, 0);
 767         BUG_ON(!root->node);
 768         return 0;
 769 }
 770
 771 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
 772                                                struct btrfs_key *location)
 773 {
 774         struct btrfs_root *root;
 775         struct btrfs_root *tree_root = fs_info->tree_root;
 776         struct btrfs_path *path;
 777         struct extent_buffer *l;
 778         u64 highest_inode;
 779         u32 blocksize;
 780         int ret = 0;
 781
 782         root = kzalloc(sizeof(*root), GFP_NOFS);
 783         if (!root)
 784                 return ERR_PTR(-ENOMEM);
 785         if (location->offset == (u64)-1) {
 786                 ret = find_and_setup_root(tree_root, fs_info,
 787                                           location->objectid, root);
 788                 if (ret) {
 789                         kfree(root);
 790                         return ERR_PTR(ret);
 791                 }
 792                 goto insert;
 793         }
 794
 795         __setup_root(tree_root->nodesize, tree_root->leafsize,
 796                      tree_root->sectorsize, tree_root->stripesize,
 797                      root, fs_info, location->objectid);
 798
 799         path = btrfs_alloc_path();
 800         BUG_ON(!path);
 801         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 802         if (ret != 0) {
 803                 if (ret > 0)
 804                         ret = -ENOENT;
 805                 goto out;
 806         }
 807         l = path->nodes[0];
 808         read_extent_buffer(l, &root->root_item,
 809                btrfs_item_ptr_offset(l, path->slots[0]),
 810                sizeof(root->root_item));
 811         memcpy(&root->root_key, location, sizeof(*location));
 812         ret = 0;
 813 out:
 814         btrfs_release_path(root, path);
 815         btrfs_free_path(path);
 816         if (ret) {
 817                 kfree(root);
 818                 return ERR_PTR(ret);
 819         }
 820         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 821         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 822                                      blocksize, 0);
 823         BUG_ON(!root->node);
 824 insert:
 825         root->ref_cows = 1;
 826         ret = btrfs_find_highest_inode(root, &highest_inode);
 827         if (ret == 0) {
 828                 root->highest_inode = highest_inode;
 829                 root->last_inode_alloc = highest_inode;
 830         }
 831         return root;
 832 }
 833
 834 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
 835                                         u64 root_objectid)
 836 {
 837         struct btrfs_root *root;
 838
 839         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
 840                 return fs_info->tree_root;
 841         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
 842                 return fs_info->extent_root;
 843
 844         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 845                                  (unsigned long)root_objectid);
 846         return root;
 847 }
 848
 849 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
 850                                               struct btrfs_key *location)
 851 {
 852         struct btrfs_root *root;
 853         int ret;
 854
 855         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 856                 return fs_info->tree_root;
 857         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
 858                 return fs_info->extent_root;
 859         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
 860                 return fs_info->chunk_root;
 861         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
 862                 return fs_info->dev_root;
 863
 864         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 865                                  (unsigned long)location->objectid);
 866         if (root)
 867                 return root;
 868
 869         root = btrfs_read_fs_root_no_radix(fs_info, location);
 870         if (IS_ERR(root))
 871                 return root;
 872         ret = radix_tree_insert(&fs_info->fs_roots_radix,
 873                                 (unsigned long)root->root_key.objectid,
 874                                 root);
 875         if (ret) {
 876                 free_extent_buffer(root->node);
 877                 kfree(root);
 878                 return ERR_PTR(ret);
 879         }
 880         ret = btrfs_find_dead_roots(fs_info->tree_root,
 881                                     root->root_key.objectid, root);
 882         BUG_ON(ret);
 883
 884         return root;
 885 }
 886
 887 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 888                                       struct btrfs_key *location,
 889                                       const char *name, int namelen)
 890 {
 891         struct btrfs_root *root;
 892         int ret;
 893
 894         root = btrfs_read_fs_root_no_name(fs_info, location);
 895         if (!root)
 896                 return NULL;
 897
 898         if (root->in_sysfs)
 899                 return root;
 900
 901         ret = btrfs_set_root_name(root, name, namelen);
 902         if (ret) {
 903                 free_extent_buffer(root->node);
 904                 kfree(root);
 905                 return ERR_PTR(ret);
 906         }
 907
 908         ret = btrfs_sysfs_add_root(root);
 909         if (ret) {
 910                 free_extent_buffer(root->node);
 911                 kfree(root->name);
 912                 kfree(root);
 913                 return ERR_PTR(ret);
 914         }
 915         root->in_sysfs = 1;
 916         return root;
 917 }
 918 #if 0
 919 static int add_hasher(struct btrfs_fs_info *info, char *type) {
 920         struct btrfs_hasher *hasher;
 921
 922         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
 923         if (!hasher)
 924                 return -ENOMEM;
 925         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
 926         if (!hasher->hash_tfm) {
 927                 kfree(hasher);
 928                 return -EINVAL;
 929         }
 930         spin_lock(&info->hash_lock);
 931         list_add(&hasher->list, &info->hashers);
 932         spin_unlock(&info->hash_lock);
 933         return 0;
 934 }
 935 #endif
 936
 937 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
 938 {
 939         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
 940         int ret = 0;
 941         int limit = 256 * info->fs_devices->open_devices;
 942         struct list_head *cur;
 943         struct btrfs_device *device;
 944         struct backing_dev_info *bdi;
 945
 946         if ((bdi_bits & (1 << BDI_write_congested)) &&
 947             atomic_read(&info->nr_async_submits) > limit) {
 948                 return 1;
 949         }
 950
 951         list_for_each(cur, &info->fs_devices->devices) {
 952                 device = list_entry(cur, struct btrfs_device, dev_list);
 953                 if (!device->bdev)
 954                         continue;
 955                 bdi = blk_get_backing_dev_info(device->bdev);
 956                 if (bdi && bdi_congested(bdi, bdi_bits)) {
 957                         ret = 1;
 958                         break;
 959                 }
 960         }
 961         return ret;
 962 }
 963
 964 /*
 965  * this unplugs every device on the box, and it is only used when page
 966  * is null
 967  */
 968 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
 969 {
 970         struct list_head *cur;
 971         struct btrfs_device *device;
 972         struct btrfs_fs_info *info;
 973
 974         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
 975         list_for_each(cur, &info->fs_devices->devices) {
 976                 device = list_entry(cur, struct btrfs_device, dev_list);
 977                 bdi = blk_get_backing_dev_info(device->bdev);
 978                 if (bdi->unplug_io_fn) {
 979                         bdi->unplug_io_fn(bdi, page);
 980                 }
 981         }
 982 }
 983
 984 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
 985 {
 986         struct inode *inode;
 987         struct extent_map_tree *em_tree;
 988         struct extent_map *em;
 989         struct address_space *mapping;
 990         u64 offset;
 991
 992         /* the generic O_DIRECT read code does this */
 993         if (!page) {
 994                 __unplug_io_fn(bdi, page);
 995                 return;
 996         }
 997
 998         /*
 999          * page->mapping may change at any time.  Get a consistent copy
1000          * and use that for everything below
1001          */
1002         smp_mb();
1003         mapping = page->mapping;
1004         if (!mapping)
1005                 return;
1006
1007         inode = mapping->host;
1008         offset = page_offset(page);
1009
1010         em_tree = &BTRFS_I(inode)->extent_tree;
1011         spin_lock(&em_tree->lock);
1012         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1013         spin_unlock(&em_tree->lock);
1014         if (!em)
1015                 return;
1016
1017         offset = offset - em->start;
1018         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1019                           em->block_start + offset, page);
1020         free_extent_map(em);
1021 }
1022
1023 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1024 {
1025 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1026         bdi_init(bdi);
1027 #endif
1028         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1029         bdi->state              = 0;
1030         bdi->capabilities       = default_backing_dev_info.capabilities;
1031         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1032         bdi->unplug_io_data     = info;
1033         bdi->congested_fn       = btrfs_congested_fn;
1034         bdi->congested_data     = info;
1035         return 0;
1036 }
1037
1038 static int bio_ready_for_csum(struct bio *bio)
1039 {
1040         u64 length = 0;
1041         u64 buf_len = 0;
1042         u64 start = 0;
1043         struct page *page;
1044         struct extent_io_tree *io_tree = NULL;
1045         struct btrfs_fs_info *info = NULL;
1046         struct bio_vec *bvec;
1047         int i;
1048         int ret;
1049
1050         bio_for_each_segment(bvec, bio, i) {
1051                 page = bvec->bv_page;
1052                 if (page->private == EXTENT_PAGE_PRIVATE) {
1053                         length += bvec->bv_len;
1054                         continue;
1055                 }
1056                 if (!page->private) {
1057                         length += bvec->bv_len;
1058                         continue;
1059                 }
1060                 length = bvec->bv_len;
1061                 buf_len = page->private >> 2;
1062                 start = page_offset(page) + bvec->bv_offset;
1063                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1064                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1065         }
1066         /* are we fully contained in this bio? */
1067         if (buf_len <= length)
1068                 return 1;
1069
1070         ret = extent_range_uptodate(io_tree, start + length,
1071                                     start + buf_len - 1);
1072         if (ret == 1)
1073                 return ret;
1074         return ret;
1075 }
1076
1077 /*
1078  * called by the kthread helper functions to finally call the bio end_io
1079  * functions.  This is where read checksum verification actually happens
1080  */
1081 static void end_workqueue_fn(struct btrfs_work *work)
1082 {
1083         struct bio *bio;
1084         struct end_io_wq *end_io_wq;
1085         struct btrfs_fs_info *fs_info;
1086         int error;
1087
1088         end_io_wq = container_of(work, struct end_io_wq, work);
1089         bio = end_io_wq->bio;
1090         fs_info = end_io_wq->info;
1091
1092         /* metadata bios are special because the whole tree block must
1093          * be checksummed at once.  This makes sure the entire block is in
1094          * ram and up to date before trying to verify things.  For
1095          * blocksize <= pagesize, it is basically a noop
1096          */
1097         if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1098                 btrfs_queue_worker(&fs_info->endio_workers,
1099                                    &end_io_wq->work);
1100                 return;
1101         }
1102         error = end_io_wq->error;
1103         bio->bi_private = end_io_wq->private;
1104         bio->bi_end_io = end_io_wq->end_io;
1105         kfree(end_io_wq);
1106 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1107         bio_endio(bio, bio->bi_size, error);
1108 #else
1109         bio_endio(bio, error);
1110 #endif
1111 }
1112
1113 static int cleaner_kthread(void *arg)
1114 {
1115         struct btrfs_root *root = arg;
1116
1117         do {
1118                 smp_mb();
1119                 if (root->fs_info->closing)
1120                         break;
1121
1122                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1123                 mutex_lock(&root->fs_info->cleaner_mutex);
1124                 btrfs_clean_old_snapshots(root);
1125                 mutex_unlock(&root->fs_info->cleaner_mutex);
1126
1127                 if (freezing(current)) {
1128                         refrigerator();
1129                 } else {
1130                         smp_mb();
1131                         if (root->fs_info->closing)
1132                                 break;
1133                         set_current_state(TASK_INTERRUPTIBLE);
1134                         schedule();
1135                         __set_current_state(TASK_RUNNING);
1136                 }
1137         } while (!kthread_should_stop());
1138         return 0;
1139 }
1140
1141 static int transaction_kthread(void *arg)
1142 {
1143         struct btrfs_root *root = arg;
1144         struct btrfs_trans_handle *trans;
1145         struct btrfs_transaction *cur;
1146         unsigned long now;
1147         unsigned long delay;
1148         int ret;
1149
1150         do {
1151                 smp_mb();
1152                 if (root->fs_info->closing)
1153                         break;
1154
1155                 delay = HZ * 30;
1156                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1157                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1158
1159                 mutex_lock(&root->fs_info->trans_mutex);
1160                 cur = root->fs_info->running_transaction;
1161                 if (!cur) {
1162                         mutex_unlock(&root->fs_info->trans_mutex);
1163                         goto sleep;
1164                 }
1165                 now = get_seconds();
1166                 if (now < cur->start_time || now - cur->start_time < 30) {
1167                         mutex_unlock(&root->fs_info->trans_mutex);
1168                         delay = HZ * 5;
1169                         goto sleep;
1170                 }
1171                 mutex_unlock(&root->fs_info->trans_mutex);
1172                 trans = btrfs_start_transaction(root, 1);
1173                 ret = btrfs_commit_transaction(trans, root);
1174 sleep:
1175                 wake_up_process(root->fs_info->cleaner_kthread);
1176                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1177
1178                 if (freezing(current)) {
1179                         refrigerator();
1180                 } else {
1181                         if (root->fs_info->closing)
1182                                 break;
1183                         set_current_state(TASK_INTERRUPTIBLE);
1184                         schedule_timeout(delay);
1185                         __set_current_state(TASK_RUNNING);
1186                 }
1187         } while (!kthread_should_stop());
1188         return 0;
1189 }
1190
1191 struct btrfs_root *open_ctree(struct super_block *sb,
1192                               struct btrfs_fs_devices *fs_devices,
1193                               char *options)
1194 {
1195         u32 sectorsize;
1196         u32 nodesize;
1197         u32 leafsize;
1198         u32 blocksize;
1199         u32 stripesize;
1200         struct buffer_head *bh;
1201         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
1202                                                  GFP_NOFS);
1203         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
1204                                                GFP_NOFS);
1205         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1206                                                 GFP_NOFS);
1207         struct btrfs_root *chunk_root = kmalloc(sizeof(struct btrfs_root),
1208                                                 GFP_NOFS);
1209         struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
1210                                               GFP_NOFS);
1211         int ret;
1212         int err = -EINVAL;
1213
1214         struct btrfs_super_block *disk_super;
1215
1216         if (!extent_root || !tree_root || !fs_info) {
1217                 err = -ENOMEM;
1218                 goto fail;
1219         }
1220         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1221         INIT_LIST_HEAD(&fs_info->trans_list);
1222         INIT_LIST_HEAD(&fs_info->dead_roots);
1223         INIT_LIST_HEAD(&fs_info->hashers);
1224         spin_lock_init(&fs_info->hash_lock);
1225         spin_lock_init(&fs_info->delalloc_lock);
1226         spin_lock_init(&fs_info->new_trans_lock);
1227
1228         init_completion(&fs_info->kobj_unregister);
1229         fs_info->tree_root = tree_root;
1230         fs_info->extent_root = extent_root;
1231         fs_info->chunk_root = chunk_root;
1232         fs_info->dev_root = dev_root;
1233         fs_info->fs_devices = fs_devices;
1234         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1235         INIT_LIST_HEAD(&fs_info->space_info);
1236         btrfs_mapping_init(&fs_info->mapping_tree);
1237         atomic_set(&fs_info->nr_async_submits, 0);
1238         atomic_set(&fs_info->throttles, 0);
1239         fs_info->sb = sb;
1240         fs_info->max_extent = (u64)-1;
1241         fs_info->max_inline = 8192 * 1024;
1242         setup_bdi(fs_info, &fs_info->bdi);
1243         fs_info->btree_inode = new_inode(sb);
1244         fs_info->btree_inode->i_ino = 1;
1245         fs_info->btree_inode->i_nlink = 1;
1246         fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1247
1248         sb->s_blocksize = 4096;
1249         sb->s_blocksize_bits = blksize_bits(4096);
1250
1251         /*
1252          * we set the i_size on the btree inode to the max possible int.
1253          * the real end of the address space is determined by all of
1254          * the devices in the system
1255          */
1256         fs_info->btree_inode->i_size = OFFSET_MAX;
1257         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1258         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1259
1260         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1261                              fs_info->btree_inode->i_mapping,
1262                              GFP_NOFS);
1263         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1264                              GFP_NOFS);
1265
1266         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1267
1268         extent_io_tree_init(&fs_info->free_space_cache,
1269                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1270         extent_io_tree_init(&fs_info->block_group_cache,
1271                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1272         extent_io_tree_init(&fs_info->pinned_extents,
1273                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1274         extent_io_tree_init(&fs_info->pending_del,
1275                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1276         extent_io_tree_init(&fs_info->extent_ins,
1277                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1278         fs_info->do_barriers = 1;
1279
1280         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1281         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1282                sizeof(struct btrfs_key));
1283         insert_inode_hash(fs_info->btree_inode);
1284         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1285
1286         mutex_init(&fs_info->trans_mutex);
1287         mutex_init(&fs_info->drop_mutex);
1288         mutex_init(&fs_info->alloc_mutex);
1289         mutex_init(&fs_info->chunk_mutex);
1290         mutex_init(&fs_info->transaction_kthread_mutex);
1291         mutex_init(&fs_info->cleaner_mutex);
1292         mutex_init(&fs_info->volume_mutex);
1293         init_waitqueue_head(&fs_info->transaction_throttle);
1294
1295 #if 0
1296         ret = add_hasher(fs_info, "crc32c");
1297         if (ret) {
1298                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1299                 err = -ENOMEM;
1300                 goto fail_iput;
1301         }
1302 #endif
1303         __setup_root(4096, 4096, 4096, 4096, tree_root,
1304                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1305
1306
1307         bh = __bread(fs_devices->latest_bdev,
1308                      BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1309         if (!bh)
1310                 goto fail_iput;
1311
1312         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1313         brelse(bh);
1314
1315         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1316
1317         disk_super = &fs_info->super_copy;
1318         if (!btrfs_super_root(disk_super))
1319                 goto fail_sb_buffer;
1320
1321         err = btrfs_parse_options(tree_root, options);
1322         if (err)
1323                 goto fail_sb_buffer;
1324
1325         /*
1326          * we need to start all the end_io workers up front because the
1327          * queue work function gets called at interrupt time, and so it
1328          * cannot dynamically grow.
1329          */
1330         btrfs_init_workers(&fs_info->workers, fs_info->thread_pool_size);
1331         btrfs_init_workers(&fs_info->submit_workers, fs_info->thread_pool_size);
1332         btrfs_init_workers(&fs_info->fixup_workers, 1);
1333         btrfs_init_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1334         btrfs_init_workers(&fs_info->endio_write_workers,
1335                            fs_info->thread_pool_size);
1336         btrfs_start_workers(&fs_info->workers, 1);
1337         btrfs_start_workers(&fs_info->submit_workers, 1);
1338         btrfs_start_workers(&fs_info->fixup_workers, 1);
1339         btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1340         btrfs_start_workers(&fs_info->endio_write_workers,
1341                             fs_info->thread_pool_size);
1342
1343         err = -EINVAL;
1344         if (btrfs_super_num_devices(disk_super) > fs_devices->open_devices) {
1345                 printk("Btrfs: wanted %llu devices, but found %llu\n",
1346                        (unsigned long long)btrfs_super_num_devices(disk_super),
1347                        (unsigned long long)fs_devices->open_devices);
1348                 if (btrfs_test_opt(tree_root, DEGRADED))
1349                         printk("continuing in degraded mode\n");
1350                 else {
1351                         goto fail_sb_buffer;
1352                 }
1353         }
1354
1355         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1356
1357         nodesize = btrfs_super_nodesize(disk_super);
1358         leafsize = btrfs_super_leafsize(disk_super);
1359         sectorsize = btrfs_super_sectorsize(disk_super);
1360         stripesize = btrfs_super_stripesize(disk_super);
1361         tree_root->nodesize = nodesize;
1362         tree_root->leafsize = leafsize;
1363         tree_root->sectorsize = sectorsize;
1364         tree_root->stripesize = stripesize;
1365
1366         sb->s_blocksize = sectorsize;
1367         sb->s_blocksize_bits = blksize_bits(sectorsize);
1368
1369         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1370                     sizeof(disk_super->magic))) {
1371                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1372                 goto fail_sb_buffer;
1373         }
1374
1375         mutex_lock(&fs_info->chunk_mutex);
1376         ret = btrfs_read_sys_array(tree_root);
1377         mutex_unlock(&fs_info->chunk_mutex);
1378         if (ret) {
1379                 printk("btrfs: failed to read the system array on %s\n",
1380                        sb->s_id);
1381                 goto fail_sys_array;
1382         }
1383
1384         blocksize = btrfs_level_size(tree_root,
1385                                      btrfs_super_chunk_root_level(disk_super));
1386
1387         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1388                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1389
1390         chunk_root->node = read_tree_block(chunk_root,
1391                                            btrfs_super_chunk_root(disk_super),
1392                                            blocksize, 0);
1393         BUG_ON(!chunk_root->node);
1394
1395         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1396                  (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1397                  BTRFS_UUID_SIZE);
1398
1399         mutex_lock(&fs_info->chunk_mutex);
1400         ret = btrfs_read_chunk_tree(chunk_root);
1401         mutex_unlock(&fs_info->chunk_mutex);
1402         BUG_ON(ret);
1403
1404         btrfs_close_extra_devices(fs_devices);
1405
1406         blocksize = btrfs_level_size(tree_root,
1407                                      btrfs_super_root_level(disk_super));
1408
1409
1410         tree_root->node = read_tree_block(tree_root,
1411                                           btrfs_super_root(disk_super),
1412                                           blocksize, 0);
1413         if (!tree_root->node)
1414                 goto fail_sb_buffer;
1415
1416
1417         ret = find_and_setup_root(tree_root, fs_info,
1418                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1419         if (ret)
1420                 goto fail_tree_root;
1421         extent_root->track_dirty = 1;
1422
1423         ret = find_and_setup_root(tree_root, fs_info,
1424                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1425         dev_root->track_dirty = 1;
1426
1427         if (ret)
1428                 goto fail_extent_root;
1429
1430         btrfs_read_block_groups(extent_root);
1431
1432         fs_info->generation = btrfs_super_generation(disk_super) + 1;
1433         fs_info->data_alloc_profile = (u64)-1;
1434         fs_info->metadata_alloc_profile = (u64)-1;
1435         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1436         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1437                                                "btrfs-cleaner");
1438         if (!fs_info->cleaner_kthread)
1439                 goto fail_extent_root;
1440
1441         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1442                                                    tree_root,
1443                                                    "btrfs-transaction");
1444         if (!fs_info->transaction_kthread)
1445                 goto fail_cleaner;
1446
1447
1448         return tree_root;
1449
1450 fail_cleaner:
1451         kthread_stop(fs_info->cleaner_kthread);
1452 fail_extent_root:
1453         free_extent_buffer(extent_root->node);
1454 fail_tree_root:
1455         free_extent_buffer(tree_root->node);
1456 fail_sys_array:
1457 fail_sb_buffer:
1458         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1459         btrfs_stop_workers(&fs_info->fixup_workers);
1460         btrfs_stop_workers(&fs_info->workers);
1461         btrfs_stop_workers(&fs_info->endio_workers);
1462         btrfs_stop_workers(&fs_info->endio_write_workers);
1463         btrfs_stop_workers(&fs_info->submit_workers);
1464 fail_iput:
1465         iput(fs_info->btree_inode);
1466 fail:
1467         btrfs_close_devices(fs_info->fs_devices);
1468         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1469
1470         kfree(extent_root);
1471         kfree(tree_root);
1472 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1473         bdi_destroy(&fs_info->bdi);
1474 #endif
1475         kfree(fs_info);
1476         return ERR_PTR(err);
1477 }
1478
1479 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1480 {
1481         char b[BDEVNAME_SIZE];
1482
1483         if (uptodate) {
1484                 set_buffer_uptodate(bh);
1485         } else {
1486                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1487                         printk(KERN_WARNING "lost page write due to "
1488                                         "I/O error on %s\n",
1489                                        bdevname(bh->b_bdev, b));
1490                 }
1491                 /* note, we dont' set_buffer_write_io_error because we have
1492                  * our own ways of dealing with the IO errors
1493                  */
1494                 clear_buffer_uptodate(bh);
1495         }
1496         unlock_buffer(bh);
1497         put_bh(bh);
1498 }
1499
1500 int write_all_supers(struct btrfs_root *root)
1501 {
1502         struct list_head *cur;
1503         struct list_head *head = &root->fs_info->fs_devices->devices;
1504         struct btrfs_device *dev;
1505         struct btrfs_super_block *sb;
1506         struct btrfs_dev_item *dev_item;
1507         struct buffer_head *bh;
1508         int ret;
1509         int do_barriers;
1510         int max_errors;
1511         int total_errors = 0;
1512         u32 crc;
1513         u64 flags;
1514
1515         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1516         do_barriers = !btrfs_test_opt(root, NOBARRIER);
1517
1518         sb = &root->fs_info->super_for_commit;
1519         dev_item = &sb->dev_item;
1520         list_for_each(cur, head) {
1521                 dev = list_entry(cur, struct btrfs_device, dev_list);
1522                 if (!dev->bdev) {
1523                         total_errors++;
1524                         continue;
1525                 }
1526                 if (!dev->in_fs_metadata)
1527                         continue;
1528
1529                 btrfs_set_stack_device_type(dev_item, dev->type);
1530                 btrfs_set_stack_device_id(dev_item, dev->devid);
1531                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1532                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1533                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1534                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1535                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1536                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1537                 flags = btrfs_super_flags(sb);
1538                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1539
1540
1541                 crc = ~(u32)0;
1542                 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1543                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1544                 btrfs_csum_final(crc, sb->csum);
1545
1546                 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1547                               BTRFS_SUPER_INFO_SIZE);
1548
1549                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1550                 dev->pending_io = bh;
1551
1552                 get_bh(bh);
1553                 set_buffer_uptodate(bh);
1554                 lock_buffer(bh);
1555                 bh->b_end_io = btrfs_end_buffer_write_sync;
1556
1557                 if (do_barriers && dev->barriers) {
1558                         ret = submit_bh(WRITE_BARRIER, bh);
1559                         if (ret == -EOPNOTSUPP) {
1560                                 printk("btrfs: disabling barriers on dev %s\n",
1561                                        dev->name);
1562                                 set_buffer_uptodate(bh);
1563                                 dev->barriers = 0;
1564                                 get_bh(bh);
1565                                 lock_buffer(bh);
1566                                 ret = submit_bh(WRITE, bh);
1567                         }
1568                 } else {
1569                         ret = submit_bh(WRITE, bh);
1570                 }
1571                 if (ret)
1572                         total_errors++;
1573         }
1574         if (total_errors > max_errors) {
1575                 printk("btrfs: %d errors while writing supers\n", total_errors);
1576                 BUG();
1577         }
1578         total_errors = 0;
1579
1580         list_for_each(cur, head) {
1581                 dev = list_entry(cur, struct btrfs_device, dev_list);
1582                 if (!dev->bdev)
1583                         continue;
1584                 if (!dev->in_fs_metadata)
1585                         continue;
1586
1587                 BUG_ON(!dev->pending_io);
1588                 bh = dev->pending_io;
1589                 wait_on_buffer(bh);
1590                 if (!buffer_uptodate(dev->pending_io)) {
1591                         if (do_barriers && dev->barriers) {
1592                                 printk("btrfs: disabling barriers on dev %s\n",
1593                                        dev->name);
1594                                 set_buffer_uptodate(bh);
1595                                 get_bh(bh);
1596                                 lock_buffer(bh);
1597                                 dev->barriers = 0;
1598                                 ret = submit_bh(WRITE, bh);
1599                                 BUG_ON(ret);
1600                                 wait_on_buffer(bh);
1601                                 if (!buffer_uptodate(bh))
1602                                         total_errors++;
1603                         } else {
1604                                 total_errors++;
1605                         }
1606
1607                 }
1608                 dev->pending_io = NULL;
1609                 brelse(bh);
1610         }
1611         if (total_errors > max_errors) {
1612                 printk("btrfs: %d errors while writing supers\n", total_errors);
1613                 BUG();
1614         }
1615         return 0;
1616 }
1617
1618 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1619                       *root)
1620 {
1621         int ret;
1622
1623         ret = write_all_supers(root);
1624         return ret;
1625 }
1626
1627 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1628 {
1629         radix_tree_delete(&fs_info->fs_roots_radix,
1630                           (unsigned long)root->root_key.objectid);
1631         if (root->in_sysfs)
1632                 btrfs_sysfs_del_root(root);
1633         if (root->inode)
1634                 iput(root->inode);
1635         if (root->node)
1636                 free_extent_buffer(root->node);
1637         if (root->commit_root)
1638                 free_extent_buffer(root->commit_root);
1639         if (root->name)
1640                 kfree(root->name);
1641         kfree(root);
1642         return 0;
1643 }
1644
1645 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1646 {
1647         int ret;
1648         struct btrfs_root *gang[8];
1649         int i;
1650
1651         while(1) {
1652                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1653                                              (void **)gang, 0,
1654                                              ARRAY_SIZE(gang));
1655                 if (!ret)
1656                         break;
1657                 for (i = 0; i < ret; i++)
1658                         btrfs_free_fs_root(fs_info, gang[i]);
1659         }
1660         return 0;
1661 }
1662
1663 int close_ctree(struct btrfs_root *root)
1664 {
1665         int ret;
1666         struct btrfs_trans_handle *trans;
1667         struct btrfs_fs_info *fs_info = root->fs_info;
1668
1669         fs_info->closing = 1;
1670         smp_mb();
1671
1672         kthread_stop(root->fs_info->transaction_kthread);
1673         kthread_stop(root->fs_info->cleaner_kthread);
1674
1675         btrfs_clean_old_snapshots(root);
1676         trans = btrfs_start_transaction(root, 1);
1677         ret = btrfs_commit_transaction(trans, root);
1678         /* run commit again to  drop the original snapshot */
1679         trans = btrfs_start_transaction(root, 1);
1680         btrfs_commit_transaction(trans, root);
1681         ret = btrfs_write_and_wait_transaction(NULL, root);
1682         BUG_ON(ret);
1683
1684         write_ctree_super(NULL, root);
1685
1686         if (fs_info->delalloc_bytes) {
1687                 printk("btrfs: at unmount delalloc count %Lu\n",
1688                        fs_info->delalloc_bytes);
1689         }
1690         if (fs_info->extent_root->node)
1691                 free_extent_buffer(fs_info->extent_root->node);
1692
1693         if (fs_info->tree_root->node)
1694                 free_extent_buffer(fs_info->tree_root->node);
1695
1696         if (root->fs_info->chunk_root->node);
1697                 free_extent_buffer(root->fs_info->chunk_root->node);
1698
1699         if (root->fs_info->dev_root->node);
1700                 free_extent_buffer(root->fs_info->dev_root->node);
1701
1702         btrfs_free_block_groups(root->fs_info);
1703         del_fs_roots(fs_info);
1704
1705         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1706
1707         extent_io_tree_empty_lru(&fs_info->free_space_cache);
1708         extent_io_tree_empty_lru(&fs_info->block_group_cache);
1709         extent_io_tree_empty_lru(&fs_info->pinned_extents);
1710         extent_io_tree_empty_lru(&fs_info->pending_del);
1711         extent_io_tree_empty_lru(&fs_info->extent_ins);
1712         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1713
1714         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1715
1716         btrfs_stop_workers(&fs_info->fixup_workers);
1717         btrfs_stop_workers(&fs_info->workers);
1718         btrfs_stop_workers(&fs_info->endio_workers);
1719         btrfs_stop_workers(&fs_info->endio_write_workers);
1720         btrfs_stop_workers(&fs_info->submit_workers);
1721
1722         iput(fs_info->btree_inode);
1723 #if 0
1724         while(!list_empty(&fs_info->hashers)) {
1725                 struct btrfs_hasher *hasher;
1726                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1727                                     hashers);
1728                 list_del(&hasher->hashers);
1729                 crypto_free_hash(&fs_info->hash_tfm);
1730                 kfree(hasher);
1731         }
1732 #endif
1733         btrfs_close_devices(fs_info->fs_devices);
1734         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1735
1736 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1737         bdi_destroy(&fs_info->bdi);
1738 #endif
1739
1740         kfree(fs_info->extent_root);
1741         kfree(fs_info->tree_root);
1742         kfree(fs_info->chunk_root);
1743         kfree(fs_info->dev_root);
1744         return 0;
1745 }
1746
1747 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1748 {
1749         int ret;
1750         struct inode *btree_inode = buf->first_page->mapping->host;
1751
1752         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1753         if (!ret)
1754                 return ret;
1755
1756         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
1757                                     parent_transid);
1758         return !ret;
1759 }
1760
1761 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1762 {
1763         struct inode *btree_inode = buf->first_page->mapping->host;
1764         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1765                                           buf);
1766 }
1767
1768 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1769 {
1770         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1771         u64 transid = btrfs_header_generation(buf);
1772         struct inode *btree_inode = root->fs_info->btree_inode;
1773
1774         WARN_ON(!btrfs_tree_locked(buf));
1775         if (transid != root->fs_info->generation) {
1776                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1777                         (unsigned long long)buf->start,
1778                         transid, root->fs_info->generation);
1779                 WARN_ON(1);
1780         }
1781         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1782 }
1783
1784 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1785 {
1786         /*
1787          * looks as though older kernels can get into trouble with
1788          * this code, they end up stuck in balance_dirty_pages forever
1789          */
1790         struct extent_io_tree *tree;
1791         u64 num_dirty;
1792         u64 start = 0;
1793         unsigned long thresh = 16 * 1024 * 1024;
1794         tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
1795
1796         if (current_is_pdflush())
1797                 return;
1798
1799         num_dirty = count_range_bits(tree, &start, (u64)-1,
1800                                      thresh, EXTENT_DIRTY);
1801         if (num_dirty > thresh) {
1802                 balance_dirty_pages_ratelimited_nr(
1803                                    root->fs_info->btree_inode->i_mapping, 1);
1804         }
1805         return;
1806 }
1807
1808 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
1809 {
1810         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1811         int ret;
1812         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1813         if (ret == 0) {
1814                 buf->flags |= EXTENT_UPTODATE;
1815         }
1816         return ret;
1817 }
1818
1819 static struct extent_io_ops btree_extent_io_ops = {
1820         .writepage_io_hook = btree_writepage_io_hook,
1821         .readpage_end_io_hook = btree_readpage_end_io_hook,
1822         .submit_bio_hook = btree_submit_bio_hook,
1823         /* note we're sharing with inode.c for the merge bio hook */
1824         .merge_bio_hook = btrfs_merge_bio_hook,
1825 };