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 #include <linux/freezer.h>
  30 #include "compat.h"
  31 #include "crc32c.h"
  32 #include "ctree.h"
  33 #include "disk-io.h"
  34 #include "transaction.h"
  35 #include "btrfs_inode.h"
  36 #include "volumes.h"
  37 #include "print-tree.h"
  38 #include "async-thread.h"
  39 #include "locking.h"
  40 #include "ref-cache.h"
  41 #include "tree-log.h"
  42
  43 #if 0
  44 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
  45 {
  46         if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
  47                 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
  48                        (unsigned long long)extent_buffer_blocknr(buf),
  49                        (unsigned long long)btrfs_header_blocknr(buf));
  50                 return 1;
  51         }
  52         return 0;
  53 }
  54 #endif
  55
  56 static struct extent_io_ops btree_extent_io_ops;
  57 static void end_workqueue_fn(struct btrfs_work *work);
  58
  59 /*
  60  * end_io_wq structs are used to do processing in task context when an IO is
  61  * complete.  This is used during reads to verify checksums, and it is used
  62  * by writes to insert metadata for new file extents after IO is complete.
  63  */
  64 struct end_io_wq {
  65         struct bio *bio;
  66         bio_end_io_t *end_io;
  67         void *private;
  68         struct btrfs_fs_info *info;
  69         int error;
  70         int metadata;
  71         struct list_head list;
  72         struct btrfs_work work;
  73 };
  74
  75 /*
  76  * async submit bios are used to offload expensive checksumming
  77  * onto the worker threads.  They checksum file and metadata bios
  78  * just before they are sent down the IO stack.
  79  */
  80 struct async_submit_bio {
  81         struct inode *inode;
  82         struct bio *bio;
  83         struct list_head list;
  84         extent_submit_bio_hook_t *submit_bio_start;
  85         extent_submit_bio_hook_t *submit_bio_done;
  86         int rw;
  87         int mirror_num;
  88         unsigned long bio_flags;
  89         struct btrfs_work work;
  90 };
  91
  92 /*
  93  * extents on the btree inode are pretty simple, there's one extent
  94  * that covers the entire device
  95  */
  96 static struct extent_map *btree_get_extent(struct inode *inode,
  97                 struct page *page, size_t page_offset, u64 start, u64 len,
  98                 int create)
  99 {
 100         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 101         struct extent_map *em;
 102         int ret;
 103
 104         spin_lock(&em_tree->lock);
 105         em = lookup_extent_mapping(em_tree, start, len);
 106         if (em) {
 107                 em->bdev =
 108                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 109                 spin_unlock(&em_tree->lock);
 110                 goto out;
 111         }
 112         spin_unlock(&em_tree->lock);
 113
 114         em = alloc_extent_map(GFP_NOFS);
 115         if (!em) {
 116                 em = ERR_PTR(-ENOMEM);
 117                 goto out;
 118         }
 119         em->start = 0;
 120         em->len = (u64)-1;
 121         em->block_len = (u64)-1;
 122         em->block_start = 0;
 123         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 124
 125         spin_lock(&em_tree->lock);
 126         ret = add_extent_mapping(em_tree, em);
 127         if (ret == -EEXIST) {
 128                 u64 failed_start = em->start;
 129                 u64 failed_len = em->len;
 130
 131                 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
 132                        em->start, em->len, em->block_start);
 133                 free_extent_map(em);
 134                 em = lookup_extent_mapping(em_tree, start, len);
 135                 if (em) {
 136                         printk("after failing, found %Lu %Lu %Lu\n",
 137                                em->start, em->len, em->block_start);
 138                         ret = 0;
 139                 } else {
 140                         em = lookup_extent_mapping(em_tree, failed_start,
 141                                                    failed_len);
 142                         if (em) {
 143                                 printk("double failure lookup gives us "
 144                                        "%Lu %Lu -> %Lu\n", em->start,
 145                                        em->len, em->block_start);
 146                                 free_extent_map(em);
 147                         }
 148                         ret = -EIO;
 149                 }
 150         } else if (ret) {
 151                 free_extent_map(em);
 152                 em = NULL;
 153         }
 154         spin_unlock(&em_tree->lock);
 155
 156         if (ret)
 157                 em = ERR_PTR(ret);
 158 out:
 159         return em;
 160 }
 161
 162 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 163 {
 164         return btrfs_crc32c(seed, data, len);
 165 }
 166
 167 void btrfs_csum_final(u32 crc, char *result)
 168 {
 169         *(__le32 *)result = ~cpu_to_le32(crc);
 170 }
 171
 172 /*
 173  * compute the csum for a btree block, and either verify it or write it
 174  * into the csum field of the block.
 175  */
 176 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 177                            int verify)
 178 {
 179         u16 csum_size =
 180                 btrfs_super_csum_size(&root->fs_info->super_copy);
 181         char *result = NULL;
 182         unsigned long len;
 183         unsigned long cur_len;
 184         unsigned long offset = BTRFS_CSUM_SIZE;
 185         char *map_token = NULL;
 186         char *kaddr;
 187         unsigned long map_start;
 188         unsigned long map_len;
 189         int err;
 190         u32 crc = ~(u32)0;
 191         unsigned long inline_result;
 192
 193         len = buf->len - offset;
 194         while(len > 0) {
 195                 err = map_private_extent_buffer(buf, offset, 32,
 196                                         &map_token, &kaddr,
 197                                         &map_start, &map_len, KM_USER0);
 198                 if (err) {
 199                         printk("failed to map extent buffer! %lu\n",
 200                                offset);
 201                         return 1;
 202                 }
 203                 cur_len = min(len, map_len - (offset - map_start));
 204                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 205                                       crc, cur_len);
 206                 len -= cur_len;
 207                 offset += cur_len;
 208                 unmap_extent_buffer(buf, map_token, KM_USER0);
 209         }
 210         if (csum_size > sizeof(inline_result)) {
 211                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
 212                 if (!result)
 213                         return 1;
 214         } else {
 215                 result = (char *)&inline_result;
 216         }
 217
 218         btrfs_csum_final(crc, result);
 219
 220         if (verify) {
 221                 /* FIXME, this is not good */
 222                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 223                         u32 val;
 224                         u32 found = 0;
 225                         memcpy(&found, result, csum_size);
 226
 227                         read_extent_buffer(buf, &val, 0, csum_size);
 228                         printk("btrfs: %s checksum verify failed on %llu "
 229                                "wanted %X found %X level %d\n",
 230                                root->fs_info->sb->s_id,
 231                                buf->start, val, found, btrfs_header_level(buf));
 232                         if (result != (char *)&inline_result)
 233                                 kfree(result);
 234                         return 1;
 235                 }
 236         } else {
 237                 write_extent_buffer(buf, result, 0, csum_size);
 238         }
 239         if (result != (char *)&inline_result)
 240                 kfree(result);
 241         return 0;
 242 }
 243
 244 /*
 245  * we can't consider a given block up to date unless the transid of the
 246  * block matches the transid in the parent node's pointer.  This is how we
 247  * detect blocks that either didn't get written at all or got written
 248  * in the wrong place.
 249  */
 250 static int verify_parent_transid(struct extent_io_tree *io_tree,
 251                                  struct extent_buffer *eb, u64 parent_transid)
 252 {
 253         int ret;
 254
 255         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 256                 return 0;
 257
 258         lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
 259         if (extent_buffer_uptodate(io_tree, eb) &&
 260             btrfs_header_generation(eb) == parent_transid) {
 261                 ret = 0;
 262                 goto out;
 263         }
 264         printk("parent transid verify failed on %llu wanted %llu found %llu\n",
 265                (unsigned long long)eb->start,
 266                (unsigned long long)parent_transid,
 267                (unsigned long long)btrfs_header_generation(eb));
 268         ret = 1;
 269         clear_extent_buffer_uptodate(io_tree, eb);
 270 out:
 271         unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
 272                       GFP_NOFS);
 273         return ret;
 274 }
 275
 276 /*
 277  * helper to read a given tree block, doing retries as required when
 278  * the checksums don't match and we have alternate mirrors to try.
 279  */
 280 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 281                                           struct extent_buffer *eb,
 282                                           u64 start, u64 parent_transid)
 283 {
 284         struct extent_io_tree *io_tree;
 285         int ret;
 286         int num_copies = 0;
 287         int mirror_num = 0;
 288
 289         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 290         while (1) {
 291                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 292                                                btree_get_extent, mirror_num);
 293                 if (!ret &&
 294                     !verify_parent_transid(io_tree, eb, parent_transid))
 295                         return ret;
 296 printk("read extent buffer pages failed with ret %d mirror no %d\n", ret, mirror_num);
 297                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 298                                               eb->start, eb->len);
 299                 if (num_copies == 1)
 300                         return ret;
 301
 302                 mirror_num++;
 303                 if (mirror_num > num_copies)
 304                         return ret;
 305         }
 306         return -EIO;
 307 }
 308
 309 /*
 310  * checksum a dirty tree block before IO.  This has extra checks to make
 311  * sure we only fill in the checksum field in the first page of a multi-page block
 312  */
 313 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 314 {
 315         struct extent_io_tree *tree;
 316         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 317         u64 found_start;
 318         int found_level;
 319         unsigned long len;
 320         struct extent_buffer *eb;
 321         int ret;
 322
 323         tree = &BTRFS_I(page->mapping->host)->io_tree;
 324
 325         if (page->private == EXTENT_PAGE_PRIVATE)
 326                 goto out;
 327         if (!page->private)
 328                 goto out;
 329         len = page->private >> 2;
 330         if (len == 0) {
 331                 WARN_ON(1);
 332         }
 333         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 334         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 335                                              btrfs_header_generation(eb));
 336         BUG_ON(ret);
 337         found_start = btrfs_header_bytenr(eb);
 338         if (found_start != start) {
 339                 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
 340                        start, found_start, len);
 341                 WARN_ON(1);
 342                 goto err;
 343         }
 344         if (eb->first_page != page) {
 345                 printk("bad first page %lu %lu\n", eb->first_page->index,
 346                        page->index);
 347                 WARN_ON(1);
 348                 goto err;
 349         }
 350         if (!PageUptodate(page)) {
 351                 printk("csum not up to date page %lu\n", page->index);
 352                 WARN_ON(1);
 353                 goto err;
 354         }
 355         found_level = btrfs_header_level(eb);
 356
 357         csum_tree_block(root, eb, 0);
 358 err:
 359         free_extent_buffer(eb);
 360 out:
 361         return 0;
 362 }
 363
 364 static int check_tree_block_fsid(struct btrfs_root *root,
 365                                  struct extent_buffer *eb)
 366 {
 367         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 368         u8 fsid[BTRFS_UUID_SIZE];
 369         int ret = 1;
 370
 371         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
 372                            BTRFS_FSID_SIZE);
 373         while (fs_devices) {
 374                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 375                         ret = 0;
 376                         break;
 377                 }
 378                 fs_devices = fs_devices->seed;
 379         }
 380         return ret;
 381 }
 382
 383 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 384                                struct extent_state *state)
 385 {
 386         struct extent_io_tree *tree;
 387         u64 found_start;
 388         int found_level;
 389         unsigned long len;
 390         struct extent_buffer *eb;
 391         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 392         int ret = 0;
 393
 394         tree = &BTRFS_I(page->mapping->host)->io_tree;
 395         if (page->private == EXTENT_PAGE_PRIVATE)
 396                 goto out;
 397         if (!page->private)
 398                 goto out;
 399         len = page->private >> 2;
 400         if (len == 0) {
 401                 WARN_ON(1);
 402         }
 403         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 404
 405         found_start = btrfs_header_bytenr(eb);
 406         if (found_start != start) {
 407                 printk("bad tree block start %llu %llu\n",
 408                        (unsigned long long)found_start,
 409                        (unsigned long long)eb->start);
 410                 ret = -EIO;
 411                 goto err;
 412         }
 413         if (eb->first_page != page) {
 414                 printk("bad first page %lu %lu\n", eb->first_page->index,
 415                        page->index);
 416                 WARN_ON(1);
 417                 ret = -EIO;
 418                 goto err;
 419         }
 420         if (check_tree_block_fsid(root, eb)) {
 421                 printk("bad fsid on block %Lu\n", eb->start);
 422                 ret = -EIO;
 423                 goto err;
 424         }
 425         found_level = btrfs_header_level(eb);
 426
 427         ret = csum_tree_block(root, eb, 1);
 428         if (ret)
 429                 ret = -EIO;
 430
 431         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 432         end = eb->start + end - 1;
 433 err:
 434         free_extent_buffer(eb);
 435 out:
 436         return ret;
 437 }
 438
 439 static void end_workqueue_bio(struct bio *bio, int err)
 440 {
 441         struct end_io_wq *end_io_wq = bio->bi_private;
 442         struct btrfs_fs_info *fs_info;
 443
 444         fs_info = end_io_wq->info;
 445         end_io_wq->error = err;
 446         end_io_wq->work.func = end_workqueue_fn;
 447         end_io_wq->work.flags = 0;
 448
 449         if (bio->bi_rw & (1 << BIO_RW)) {
 450                 btrfs_queue_worker(&fs_info->endio_write_workers,
 451                                    &end_io_wq->work);
 452         } else {
 453                 if (end_io_wq->metadata)
 454                         btrfs_queue_worker(&fs_info->endio_meta_workers,
 455                                            &end_io_wq->work);
 456                 else
 457                         btrfs_queue_worker(&fs_info->endio_workers,
 458                                            &end_io_wq->work);
 459         }
 460 }
 461
 462 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 463                         int metadata)
 464 {
 465         struct end_io_wq *end_io_wq;
 466         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 467         if (!end_io_wq)
 468                 return -ENOMEM;
 469
 470         end_io_wq->private = bio->bi_private;
 471         end_io_wq->end_io = bio->bi_end_io;
 472         end_io_wq->info = info;
 473         end_io_wq->error = 0;
 474         end_io_wq->bio = bio;
 475         end_io_wq->metadata = metadata;
 476
 477         bio->bi_private = end_io_wq;
 478         bio->bi_end_io = end_workqueue_bio;
 479         return 0;
 480 }
 481
 482 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 483 {
 484         unsigned long limit = min_t(unsigned long,
 485                                     info->workers.max_workers,
 486                                     info->fs_devices->open_devices);
 487         return 256 * limit;
 488 }
 489
 490 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
 491 {
 492         return atomic_read(&info->nr_async_bios) >
 493                 btrfs_async_submit_limit(info);
 494 }
 495
 496 static void run_one_async_start(struct btrfs_work *work)
 497 {
 498         struct btrfs_fs_info *fs_info;
 499         struct async_submit_bio *async;
 500
 501         async = container_of(work, struct  async_submit_bio, work);
 502         fs_info = BTRFS_I(async->inode)->root->fs_info;
 503         async->submit_bio_start(async->inode, async->rw, async->bio,
 504                                async->mirror_num, async->bio_flags);
 505 }
 506
 507 static void run_one_async_done(struct btrfs_work *work)
 508 {
 509         struct btrfs_fs_info *fs_info;
 510         struct async_submit_bio *async;
 511         int limit;
 512
 513         async = container_of(work, struct  async_submit_bio, work);
 514         fs_info = BTRFS_I(async->inode)->root->fs_info;
 515
 516         limit = btrfs_async_submit_limit(fs_info);
 517         limit = limit * 2 / 3;
 518
 519         atomic_dec(&fs_info->nr_async_submits);
 520
 521         if (atomic_read(&fs_info->nr_async_submits) < limit &&
 522             waitqueue_active(&fs_info->async_submit_wait))
 523                 wake_up(&fs_info->async_submit_wait);
 524
 525         async->submit_bio_done(async->inode, async->rw, async->bio,
 526                                async->mirror_num, async->bio_flags);
 527 }
 528
 529 static void run_one_async_free(struct btrfs_work *work)
 530 {
 531         struct async_submit_bio *async;
 532
 533         async = container_of(work, struct  async_submit_bio, work);
 534         kfree(async);
 535 }
 536
 537 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 538                         int rw, struct bio *bio, int mirror_num,
 539                         unsigned long bio_flags,
 540                         extent_submit_bio_hook_t *submit_bio_start,
 541                         extent_submit_bio_hook_t *submit_bio_done)
 542 {
 543         struct async_submit_bio *async;
 544
 545         async = kmalloc(sizeof(*async), GFP_NOFS);
 546         if (!async)
 547                 return -ENOMEM;
 548
 549         async->inode = inode;
 550         async->rw = rw;
 551         async->bio = bio;
 552         async->mirror_num = mirror_num;
 553         async->submit_bio_start = submit_bio_start;
 554         async->submit_bio_done = submit_bio_done;
 555
 556         async->work.func = run_one_async_start;
 557         async->work.ordered_func = run_one_async_done;
 558         async->work.ordered_free = run_one_async_free;
 559
 560         async->work.flags = 0;
 561         async->bio_flags = bio_flags;
 562
 563         atomic_inc(&fs_info->nr_async_submits);
 564         btrfs_queue_worker(&fs_info->workers, &async->work);
 565 #if 0
 566         int limit = btrfs_async_submit_limit(fs_info);
 567         if (atomic_read(&fs_info->nr_async_submits) > limit) {
 568                 wait_event_timeout(fs_info->async_submit_wait,
 569                            (atomic_read(&fs_info->nr_async_submits) < limit),
 570                            HZ/10);
 571
 572                 wait_event_timeout(fs_info->async_submit_wait,
 573                            (atomic_read(&fs_info->nr_async_bios) < limit),
 574                            HZ/10);
 575         }
 576 #endif
 577         while(atomic_read(&fs_info->async_submit_draining) &&
 578               atomic_read(&fs_info->nr_async_submits)) {
 579                 wait_event(fs_info->async_submit_wait,
 580                            (atomic_read(&fs_info->nr_async_submits) == 0));
 581         }
 582
 583         return 0;
 584 }
 585
 586 static int btree_csum_one_bio(struct bio *bio)
 587 {
 588         struct bio_vec *bvec = bio->bi_io_vec;
 589         int bio_index = 0;
 590         struct btrfs_root *root;
 591
 592         WARN_ON(bio->bi_vcnt <= 0);
 593         while(bio_index < bio->bi_vcnt) {
 594                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 595                 csum_dirty_buffer(root, bvec->bv_page);
 596                 bio_index++;
 597                 bvec++;
 598         }
 599         return 0;
 600 }
 601
 602 static int __btree_submit_bio_start(struct inode *inode, int rw,
 603                                     struct bio *bio, int mirror_num,
 604                                     unsigned long bio_flags)
 605 {
 606         /*
 607          * when we're called for a write, we're already in the async
 608          * submission context.  Just jump into btrfs_map_bio
 609          */
 610         btree_csum_one_bio(bio);
 611         return 0;
 612 }
 613
 614 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 615                                  int mirror_num, unsigned long bio_flags)
 616 {
 617         /*
 618          * when we're called for a write, we're already in the async
 619          * submission context.  Just jump into btrfs_map_bio
 620          */
 621         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 622 }
 623
 624 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 625                                  int mirror_num, unsigned long bio_flags)
 626 {
 627         /*
 628          * kthread helpers are used to submit writes so that checksumming
 629          * can happen in parallel across all CPUs
 630          */
 631         if (!(rw & (1 << BIO_RW))) {
 632                 int ret;
 633                 /*
 634                  * called for a read, do the setup so that checksum validation
 635                  * can happen in the async kernel threads
 636                  */
 637                 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 638                                           bio, 1);
 639                 BUG_ON(ret);
 640
 641                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 642                                      mirror_num, 0);
 643         }
 644         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 645                                    inode, rw, bio, mirror_num, 0,
 646                                    __btree_submit_bio_start,
 647                                    __btree_submit_bio_done);
 648 }
 649
 650 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 651 {
 652         struct extent_io_tree *tree;
 653         tree = &BTRFS_I(page->mapping->host)->io_tree;
 654
 655         if (current->flags & PF_MEMALLOC) {
 656                 redirty_page_for_writepage(wbc, page);
 657                 unlock_page(page);
 658                 return 0;
 659         }
 660         return extent_write_full_page(tree, page, btree_get_extent, wbc);
 661 }
 662
 663 static int btree_writepages(struct address_space *mapping,
 664                             struct writeback_control *wbc)
 665 {
 666         struct extent_io_tree *tree;
 667         tree = &BTRFS_I(mapping->host)->io_tree;
 668         if (wbc->sync_mode == WB_SYNC_NONE) {
 669                 u64 num_dirty;
 670                 u64 start = 0;
 671                 unsigned long thresh = 32 * 1024 * 1024;
 672
 673                 if (wbc->for_kupdate)
 674                         return 0;
 675
 676                 num_dirty = count_range_bits(tree, &start, (u64)-1,
 677                                              thresh, EXTENT_DIRTY);
 678                 if (num_dirty < thresh) {
 679                         return 0;
 680                 }
 681         }
 682         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 683 }
 684
 685 static int btree_readpage(struct file *file, struct page *page)
 686 {
 687         struct extent_io_tree *tree;
 688         tree = &BTRFS_I(page->mapping->host)->io_tree;
 689         return extent_read_full_page(tree, page, btree_get_extent);
 690 }
 691
 692 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 693 {
 694         struct extent_io_tree *tree;
 695         struct extent_map_tree *map;
 696         int ret;
 697
 698         if (PageWriteback(page) || PageDirty(page))
 699             return 0;
 700
 701         tree = &BTRFS_I(page->mapping->host)->io_tree;
 702         map = &BTRFS_I(page->mapping->host)->extent_tree;
 703
 704         ret = try_release_extent_state(map, tree, page, gfp_flags);
 705         if (!ret) {
 706                 return 0;
 707         }
 708
 709         ret = try_release_extent_buffer(tree, page);
 710         if (ret == 1) {
 711                 ClearPagePrivate(page);
 712                 set_page_private(page, 0);
 713                 page_cache_release(page);
 714         }
 715
 716         return ret;
 717 }
 718
 719 static void btree_invalidatepage(struct page *page, unsigned long offset)
 720 {
 721         struct extent_io_tree *tree;
 722         tree = &BTRFS_I(page->mapping->host)->io_tree;
 723         extent_invalidatepage(tree, page, offset);
 724         btree_releasepage(page, GFP_NOFS);
 725         if (PagePrivate(page)) {
 726                 printk("warning page private not zero on page %Lu\n",
 727                        page_offset(page));
 728                 ClearPagePrivate(page);
 729                 set_page_private(page, 0);
 730                 page_cache_release(page);
 731         }
 732 }
 733
 734 #if 0
 735 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 736 {
 737         struct buffer_head *bh;
 738         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 739         struct buffer_head *head;
 740         if (!page_has_buffers(page)) {
 741                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 742                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 743         }
 744         head = page_buffers(page);
 745         bh = head;
 746         do {
 747                 if (buffer_dirty(bh))
 748                         csum_tree_block(root, bh, 0);
 749                 bh = bh->b_this_page;
 750         } while (bh != head);
 751         return block_write_full_page(page, btree_get_block, wbc);
 752 }
 753 #endif
 754
 755 static struct address_space_operations btree_aops = {
 756         .readpage       = btree_readpage,
 757         .writepage      = btree_writepage,
 758         .writepages     = btree_writepages,
 759         .releasepage    = btree_releasepage,
 760         .invalidatepage = btree_invalidatepage,
 761         .sync_page      = block_sync_page,
 762 };
 763
 764 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 765                          u64 parent_transid)
 766 {
 767         struct extent_buffer *buf = NULL;
 768         struct inode *btree_inode = root->fs_info->btree_inode;
 769         int ret = 0;
 770
 771         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 772         if (!buf)
 773                 return 0;
 774         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 775                                  buf, 0, 0, btree_get_extent, 0);
 776         free_extent_buffer(buf);
 777         return ret;
 778 }
 779
 780 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 781                                             u64 bytenr, u32 blocksize)
 782 {
 783         struct inode *btree_inode = root->fs_info->btree_inode;
 784         struct extent_buffer *eb;
 785         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 786                                 bytenr, blocksize, GFP_NOFS);
 787         return eb;
 788 }
 789
 790 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 791                                                  u64 bytenr, u32 blocksize)
 792 {
 793         struct inode *btree_inode = root->fs_info->btree_inode;
 794         struct extent_buffer *eb;
 795
 796         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 797                                  bytenr, blocksize, NULL, GFP_NOFS);
 798         return eb;
 799 }
 800
 801
 802 int btrfs_write_tree_block(struct extent_buffer *buf)
 803 {
 804         return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
 805                                       buf->start + buf->len - 1, WB_SYNC_ALL);
 806 }
 807
 808 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 809 {
 810         return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
 811                                   buf->start, buf->start + buf->len -1);
 812 }
 813
 814 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 815                                       u32 blocksize, u64 parent_transid)
 816 {
 817         struct extent_buffer *buf = NULL;
 818         struct inode *btree_inode = root->fs_info->btree_inode;
 819         struct extent_io_tree *io_tree;
 820         int ret;
 821
 822         io_tree = &BTRFS_I(btree_inode)->io_tree;
 823
 824         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 825         if (!buf)
 826                 return NULL;
 827
 828         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 829
 830         if (ret == 0) {
 831                 buf->flags |= EXTENT_UPTODATE;
 832         } else {
 833                 WARN_ON(1);
 834         }
 835         return buf;
 836
 837 }
 838
 839 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 840                      struct extent_buffer *buf)
 841 {
 842         struct inode *btree_inode = root->fs_info->btree_inode;
 843         if (btrfs_header_generation(buf) ==
 844             root->fs_info->running_transaction->transid) {
 845                 WARN_ON(!btrfs_tree_locked(buf));
 846                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 847                                           buf);
 848         }
 849         return 0;
 850 }
 851
 852 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 853                         u32 stripesize, struct btrfs_root *root,
 854                         struct btrfs_fs_info *fs_info,
 855                         u64 objectid)
 856 {
 857         root->node = NULL;
 858         root->commit_root = NULL;
 859         root->ref_tree = NULL;
 860         root->sectorsize = sectorsize;
 861         root->nodesize = nodesize;
 862         root->leafsize = leafsize;
 863         root->stripesize = stripesize;
 864         root->ref_cows = 0;
 865         root->track_dirty = 0;
 866
 867         root->fs_info = fs_info;
 868         root->objectid = objectid;
 869         root->last_trans = 0;
 870         root->highest_inode = 0;
 871         root->last_inode_alloc = 0;
 872         root->name = NULL;
 873         root->in_sysfs = 0;
 874
 875         INIT_LIST_HEAD(&root->dirty_list);
 876         INIT_LIST_HEAD(&root->orphan_list);
 877         INIT_LIST_HEAD(&root->dead_list);
 878         spin_lock_init(&root->node_lock);
 879         spin_lock_init(&root->list_lock);
 880         mutex_init(&root->objectid_mutex);
 881         mutex_init(&root->log_mutex);
 882         extent_io_tree_init(&root->dirty_log_pages,
 883                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 884
 885         btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
 886         root->ref_tree = &root->ref_tree_struct;
 887
 888         memset(&root->root_key, 0, sizeof(root->root_key));
 889         memset(&root->root_item, 0, sizeof(root->root_item));
 890         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 891         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 892         root->defrag_trans_start = fs_info->generation;
 893         init_completion(&root->kobj_unregister);
 894         root->defrag_running = 0;
 895         root->defrag_level = 0;
 896         root->root_key.objectid = objectid;
 897         root->anon_super.s_root = NULL;
 898         root->anon_super.s_dev = 0;
 899         INIT_LIST_HEAD(&root->anon_super.s_list);
 900         INIT_LIST_HEAD(&root->anon_super.s_instances);
 901         init_rwsem(&root->anon_super.s_umount);
 902
 903         return 0;
 904 }
 905
 906 static int find_and_setup_root(struct btrfs_root *tree_root,
 907                                struct btrfs_fs_info *fs_info,
 908                                u64 objectid,
 909                                struct btrfs_root *root)
 910 {
 911         int ret;
 912         u32 blocksize;
 913         u64 generation;
 914
 915         __setup_root(tree_root->nodesize, tree_root->leafsize,
 916                      tree_root->sectorsize, tree_root->stripesize,
 917                      root, fs_info, objectid);
 918         ret = btrfs_find_last_root(tree_root, objectid,
 919                                    &root->root_item, &root->root_key);
 920         BUG_ON(ret);
 921
 922         generation = btrfs_root_generation(&root->root_item);
 923         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 924         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 925                                      blocksize, generation);
 926         BUG_ON(!root->node);
 927         return 0;
 928 }
 929
 930 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
 931                              struct btrfs_fs_info *fs_info)
 932 {
 933         struct extent_buffer *eb;
 934         struct btrfs_root *log_root_tree = fs_info->log_root_tree;
 935         u64 start = 0;
 936         u64 end = 0;
 937         int ret;
 938
 939         if (!log_root_tree)
 940                 return 0;
 941
 942         while(1) {
 943                 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
 944                                     0, &start, &end, EXTENT_DIRTY);
 945                 if (ret)
 946                         break;
 947
 948                 clear_extent_dirty(&log_root_tree->dirty_log_pages,
 949                                    start, end, GFP_NOFS);
 950         }
 951         eb = fs_info->log_root_tree->node;
 952
 953         WARN_ON(btrfs_header_level(eb) != 0);
 954         WARN_ON(btrfs_header_nritems(eb) != 0);
 955
 956         ret = btrfs_free_reserved_extent(fs_info->tree_root,
 957                                 eb->start, eb->len);
 958         BUG_ON(ret);
 959
 960         free_extent_buffer(eb);
 961         kfree(fs_info->log_root_tree);
 962         fs_info->log_root_tree = NULL;
 963         return 0;
 964 }
 965
 966 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
 967                              struct btrfs_fs_info *fs_info)
 968 {
 969         struct btrfs_root *root;
 970         struct btrfs_root *tree_root = fs_info->tree_root;
 971
 972         root = kzalloc(sizeof(*root), GFP_NOFS);
 973         if (!root)
 974                 return -ENOMEM;
 975
 976         __setup_root(tree_root->nodesize, tree_root->leafsize,
 977                      tree_root->sectorsize, tree_root->stripesize,
 978                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
 979
 980         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
 981         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
 982         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
 983         root->ref_cows = 0;
 984
 985         root->node = btrfs_alloc_free_block(trans, root, root->leafsize,
 986                                             0, BTRFS_TREE_LOG_OBJECTID,
 987                                             trans->transid, 0, 0, 0);
 988
 989         btrfs_set_header_nritems(root->node, 0);
 990         btrfs_set_header_level(root->node, 0);
 991         btrfs_set_header_bytenr(root->node, root->node->start);
 992         btrfs_set_header_generation(root->node, trans->transid);
 993         btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
 994
 995         write_extent_buffer(root->node, root->fs_info->fsid,
 996                             (unsigned long)btrfs_header_fsid(root->node),
 997                             BTRFS_FSID_SIZE);
 998         btrfs_mark_buffer_dirty(root->node);
 999         btrfs_tree_unlock(root->node);
1000         fs_info->log_root_tree = root;
1001         return 0;
1002 }
1003
1004 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1005                                                struct btrfs_key *location)
1006 {
1007         struct btrfs_root *root;
1008         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1009         struct btrfs_path *path;
1010         struct extent_buffer *l;
1011         u64 highest_inode;
1012         u64 generation;
1013         u32 blocksize;
1014         int ret = 0;
1015
1016         root = kzalloc(sizeof(*root), GFP_NOFS);
1017         if (!root)
1018                 return ERR_PTR(-ENOMEM);
1019         if (location->offset == (u64)-1) {
1020                 ret = find_and_setup_root(tree_root, fs_info,
1021                                           location->objectid, root);
1022                 if (ret) {
1023                         kfree(root);
1024                         return ERR_PTR(ret);
1025                 }
1026                 goto insert;
1027         }
1028
1029         __setup_root(tree_root->nodesize, tree_root->leafsize,
1030                      tree_root->sectorsize, tree_root->stripesize,
1031                      root, fs_info, location->objectid);
1032
1033         path = btrfs_alloc_path();
1034         BUG_ON(!path);
1035         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1036         if (ret != 0) {
1037                 if (ret > 0)
1038                         ret = -ENOENT;
1039                 goto out;
1040         }
1041         l = path->nodes[0];
1042         read_extent_buffer(l, &root->root_item,
1043                btrfs_item_ptr_offset(l, path->slots[0]),
1044                sizeof(root->root_item));
1045         memcpy(&root->root_key, location, sizeof(*location));
1046         ret = 0;
1047 out:
1048         btrfs_release_path(root, path);
1049         btrfs_free_path(path);
1050         if (ret) {
1051                 kfree(root);
1052                 return ERR_PTR(ret);
1053         }
1054         generation = btrfs_root_generation(&root->root_item);
1055         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1056         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1057                                      blocksize, generation);
1058         BUG_ON(!root->node);
1059 insert:
1060         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1061                 root->ref_cows = 1;
1062                 ret = btrfs_find_highest_inode(root, &highest_inode);
1063                 if (ret == 0) {
1064                         root->highest_inode = highest_inode;
1065                         root->last_inode_alloc = highest_inode;
1066                 }
1067         }
1068         return root;
1069 }
1070
1071 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1072                                         u64 root_objectid)
1073 {
1074         struct btrfs_root *root;
1075
1076         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1077                 return fs_info->tree_root;
1078         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1079                 return fs_info->extent_root;
1080
1081         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1082                                  (unsigned long)root_objectid);
1083         return root;
1084 }
1085
1086 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1087                                               struct btrfs_key *location)
1088 {
1089         struct btrfs_root *root;
1090         int ret;
1091
1092         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1093                 return fs_info->tree_root;
1094         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1095                 return fs_info->extent_root;
1096         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1097                 return fs_info->chunk_root;
1098         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1099                 return fs_info->dev_root;
1100         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1101                 return fs_info->csum_root;
1102
1103         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1104                                  (unsigned long)location->objectid);
1105         if (root)
1106                 return root;
1107
1108         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1109         if (IS_ERR(root))
1110                 return root;
1111
1112         set_anon_super(&root->anon_super, NULL);
1113
1114         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1115                                 (unsigned long)root->root_key.objectid,
1116                                 root);
1117         if (ret) {
1118                 free_extent_buffer(root->node);
1119                 kfree(root);
1120                 return ERR_PTR(ret);
1121         }
1122         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
1123                 ret = btrfs_find_dead_roots(fs_info->tree_root,
1124                                             root->root_key.objectid, root);
1125                 BUG_ON(ret);
1126                 btrfs_orphan_cleanup(root);
1127         }
1128         return root;
1129 }
1130
1131 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1132                                       struct btrfs_key *location,
1133                                       const char *name, int namelen)
1134 {
1135         struct btrfs_root *root;
1136         int ret;
1137
1138         root = btrfs_read_fs_root_no_name(fs_info, location);
1139         if (!root)
1140                 return NULL;
1141
1142         if (root->in_sysfs)
1143                 return root;
1144
1145         ret = btrfs_set_root_name(root, name, namelen);
1146         if (ret) {
1147                 free_extent_buffer(root->node);
1148                 kfree(root);
1149                 return ERR_PTR(ret);
1150         }
1151 #if 0
1152         ret = btrfs_sysfs_add_root(root);
1153         if (ret) {
1154                 free_extent_buffer(root->node);
1155                 kfree(root->name);
1156                 kfree(root);
1157                 return ERR_PTR(ret);
1158         }
1159 #endif
1160         root->in_sysfs = 1;
1161         return root;
1162 }
1163 #if 0
1164 static int add_hasher(struct btrfs_fs_info *info, char *type) {
1165         struct btrfs_hasher *hasher;
1166
1167         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
1168         if (!hasher)
1169                 return -ENOMEM;
1170         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
1171         if (!hasher->hash_tfm) {
1172                 kfree(hasher);
1173                 return -EINVAL;
1174         }
1175         spin_lock(&info->hash_lock);
1176         list_add(&hasher->list, &info->hashers);
1177         spin_unlock(&info->hash_lock);
1178         return 0;
1179 }
1180 #endif
1181
1182 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1183 {
1184         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1185         int ret = 0;
1186         struct list_head *cur;
1187         struct btrfs_device *device;
1188         struct backing_dev_info *bdi;
1189 #if 0
1190         if ((bdi_bits & (1 << BDI_write_congested)) &&
1191             btrfs_congested_async(info, 0))
1192                 return 1;
1193 #endif
1194         list_for_each(cur, &info->fs_devices->devices) {
1195                 device = list_entry(cur, struct btrfs_device, dev_list);
1196                 if (!device->bdev)
1197                         continue;
1198                 bdi = blk_get_backing_dev_info(device->bdev);
1199                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1200                         ret = 1;
1201                         break;
1202                 }
1203         }
1204         return ret;
1205 }
1206
1207 /*
1208  * this unplugs every device on the box, and it is only used when page
1209  * is null
1210  */
1211 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1212 {
1213         struct list_head *cur;
1214         struct btrfs_device *device;
1215         struct btrfs_fs_info *info;
1216
1217         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1218         list_for_each(cur, &info->fs_devices->devices) {
1219                 device = list_entry(cur, struct btrfs_device, dev_list);
1220                 if (!device->bdev)
1221                         continue;
1222
1223                 bdi = blk_get_backing_dev_info(device->bdev);
1224                 if (bdi->unplug_io_fn) {
1225                         bdi->unplug_io_fn(bdi, page);
1226                 }
1227         }
1228 }
1229
1230 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1231 {
1232         struct inode *inode;
1233         struct extent_map_tree *em_tree;
1234         struct extent_map *em;
1235         struct address_space *mapping;
1236         u64 offset;
1237
1238         /* the generic O_DIRECT read code does this */
1239         if (1 || !page) {
1240                 __unplug_io_fn(bdi, page);
1241                 return;
1242         }
1243
1244         /*
1245          * page->mapping may change at any time.  Get a consistent copy
1246          * and use that for everything below
1247          */
1248         smp_mb();
1249         mapping = page->mapping;
1250         if (!mapping)
1251                 return;
1252
1253         inode = mapping->host;
1254
1255         /*
1256          * don't do the expensive searching for a small number of
1257          * devices
1258          */
1259         if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1260                 __unplug_io_fn(bdi, page);
1261                 return;
1262         }
1263
1264         offset = page_offset(page);
1265
1266         em_tree = &BTRFS_I(inode)->extent_tree;
1267         spin_lock(&em_tree->lock);
1268         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1269         spin_unlock(&em_tree->lock);
1270         if (!em) {
1271                 __unplug_io_fn(bdi, page);
1272                 return;
1273         }
1274
1275         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1276                 free_extent_map(em);
1277                 __unplug_io_fn(bdi, page);
1278                 return;
1279         }
1280         offset = offset - em->start;
1281         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1282                           em->block_start + offset, page);
1283         free_extent_map(em);
1284 }
1285
1286 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1287 {
1288         bdi_init(bdi);
1289         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1290         bdi->state              = 0;
1291         bdi->capabilities       = default_backing_dev_info.capabilities;
1292         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1293         bdi->unplug_io_data     = info;
1294         bdi->congested_fn       = btrfs_congested_fn;
1295         bdi->congested_data     = info;
1296         return 0;
1297 }
1298
1299 static int bio_ready_for_csum(struct bio *bio)
1300 {
1301         u64 length = 0;
1302         u64 buf_len = 0;
1303         u64 start = 0;
1304         struct page *page;
1305         struct extent_io_tree *io_tree = NULL;
1306         struct btrfs_fs_info *info = NULL;
1307         struct bio_vec *bvec;
1308         int i;
1309         int ret;
1310
1311         bio_for_each_segment(bvec, bio, i) {
1312                 page = bvec->bv_page;
1313                 if (page->private == EXTENT_PAGE_PRIVATE) {
1314                         length += bvec->bv_len;
1315                         continue;
1316                 }
1317                 if (!page->private) {
1318                         length += bvec->bv_len;
1319                         continue;
1320                 }
1321                 length = bvec->bv_len;
1322                 buf_len = page->private >> 2;
1323                 start = page_offset(page) + bvec->bv_offset;
1324                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1325                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1326         }
1327         /* are we fully contained in this bio? */
1328         if (buf_len <= length)
1329                 return 1;
1330
1331         ret = extent_range_uptodate(io_tree, start + length,
1332                                     start + buf_len - 1);
1333         if (ret == 1)
1334                 return ret;
1335         return ret;
1336 }
1337
1338 /*
1339  * called by the kthread helper functions to finally call the bio end_io
1340  * functions.  This is where read checksum verification actually happens
1341  */
1342 static void end_workqueue_fn(struct btrfs_work *work)
1343 {
1344         struct bio *bio;
1345         struct end_io_wq *end_io_wq;
1346         struct btrfs_fs_info *fs_info;
1347         int error;
1348
1349         end_io_wq = container_of(work, struct end_io_wq, work);
1350         bio = end_io_wq->bio;
1351         fs_info = end_io_wq->info;
1352
1353         /* metadata bios are special because the whole tree block must
1354          * be checksummed at once.  This makes sure the entire block is in
1355          * ram and up to date before trying to verify things.  For
1356          * blocksize <= pagesize, it is basically a noop
1357          */
1358         if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1359                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1360                                    &end_io_wq->work);
1361                 return;
1362         }
1363         error = end_io_wq->error;
1364         bio->bi_private = end_io_wq->private;
1365         bio->bi_end_io = end_io_wq->end_io;
1366         kfree(end_io_wq);
1367         bio_endio(bio, error);
1368 }
1369
1370 static int cleaner_kthread(void *arg)
1371 {
1372         struct btrfs_root *root = arg;
1373
1374         do {
1375                 smp_mb();
1376                 if (root->fs_info->closing)
1377                         break;
1378
1379                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1380                 mutex_lock(&root->fs_info->cleaner_mutex);
1381                 btrfs_clean_old_snapshots(root);
1382                 mutex_unlock(&root->fs_info->cleaner_mutex);
1383
1384                 if (freezing(current)) {
1385                         refrigerator();
1386                 } else {
1387                         smp_mb();
1388                         if (root->fs_info->closing)
1389                                 break;
1390                         set_current_state(TASK_INTERRUPTIBLE);
1391                         schedule();
1392                         __set_current_state(TASK_RUNNING);
1393                 }
1394         } while (!kthread_should_stop());
1395         return 0;
1396 }
1397
1398 static int transaction_kthread(void *arg)
1399 {
1400         struct btrfs_root *root = arg;
1401         struct btrfs_trans_handle *trans;
1402         struct btrfs_transaction *cur;
1403         unsigned long now;
1404         unsigned long delay;
1405         int ret;
1406
1407         do {
1408                 smp_mb();
1409                 if (root->fs_info->closing)
1410                         break;
1411
1412                 delay = HZ * 30;
1413                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1414                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1415
1416                 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1417                         printk("btrfs: total reference cache size %Lu\n",
1418                                 root->fs_info->total_ref_cache_size);
1419                 }
1420
1421                 mutex_lock(&root->fs_info->trans_mutex);
1422                 cur = root->fs_info->running_transaction;
1423                 if (!cur) {
1424                         mutex_unlock(&root->fs_info->trans_mutex);
1425                         goto sleep;
1426                 }
1427
1428                 now = get_seconds();
1429                 if (now < cur->start_time || now - cur->start_time < 30) {
1430                         mutex_unlock(&root->fs_info->trans_mutex);
1431                         delay = HZ * 5;
1432                         goto sleep;
1433                 }
1434                 mutex_unlock(&root->fs_info->trans_mutex);
1435                 trans = btrfs_start_transaction(root, 1);
1436                 ret = btrfs_commit_transaction(trans, root);
1437 sleep:
1438                 wake_up_process(root->fs_info->cleaner_kthread);
1439                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1440
1441                 if (freezing(current)) {
1442                         refrigerator();
1443                 } else {
1444                         if (root->fs_info->closing)
1445                                 break;
1446                         set_current_state(TASK_INTERRUPTIBLE);
1447                         schedule_timeout(delay);
1448                         __set_current_state(TASK_RUNNING);
1449                 }
1450         } while (!kthread_should_stop());
1451         return 0;
1452 }
1453
1454 struct btrfs_root *open_ctree(struct super_block *sb,
1455                               struct btrfs_fs_devices *fs_devices,
1456                               char *options)
1457 {
1458         u32 sectorsize;
1459         u32 nodesize;
1460         u32 leafsize;
1461         u32 blocksize;
1462         u32 stripesize;
1463         u64 generation;
1464         u64 features;
1465         struct btrfs_key location;
1466         struct buffer_head *bh;
1467         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1468                                                  GFP_NOFS);
1469         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1470                                                  GFP_NOFS);
1471         struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1472                                                GFP_NOFS);
1473         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1474                                                 GFP_NOFS);
1475         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1476                                                 GFP_NOFS);
1477         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1478                                               GFP_NOFS);
1479         struct btrfs_root *log_tree_root;
1480
1481         int ret;
1482         int err = -EINVAL;
1483
1484         struct btrfs_super_block *disk_super;
1485
1486         if (!extent_root || !tree_root || !fs_info ||
1487             !chunk_root || !dev_root || !csum_root) {
1488                 err = -ENOMEM;
1489                 goto fail;
1490         }
1491         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1492         INIT_LIST_HEAD(&fs_info->trans_list);
1493         INIT_LIST_HEAD(&fs_info->dead_roots);
1494         INIT_LIST_HEAD(&fs_info->hashers);
1495         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1496         spin_lock_init(&fs_info->hash_lock);
1497         spin_lock_init(&fs_info->delalloc_lock);
1498         spin_lock_init(&fs_info->new_trans_lock);
1499         spin_lock_init(&fs_info->ref_cache_lock);
1500
1501         init_completion(&fs_info->kobj_unregister);
1502         fs_info->tree_root = tree_root;
1503         fs_info->extent_root = extent_root;
1504         fs_info->csum_root = csum_root;
1505         fs_info->chunk_root = chunk_root;
1506         fs_info->dev_root = dev_root;
1507         fs_info->fs_devices = fs_devices;
1508         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1509         INIT_LIST_HEAD(&fs_info->space_info);
1510         btrfs_mapping_init(&fs_info->mapping_tree);
1511         atomic_set(&fs_info->nr_async_submits, 0);
1512         atomic_set(&fs_info->async_delalloc_pages, 0);
1513         atomic_set(&fs_info->async_submit_draining, 0);
1514         atomic_set(&fs_info->nr_async_bios, 0);
1515         atomic_set(&fs_info->throttles, 0);
1516         atomic_set(&fs_info->throttle_gen, 0);
1517         fs_info->sb = sb;
1518         fs_info->max_extent = (u64)-1;
1519         fs_info->max_inline = 8192 * 1024;
1520         setup_bdi(fs_info, &fs_info->bdi);
1521         fs_info->btree_inode = new_inode(sb);
1522         fs_info->btree_inode->i_ino = 1;
1523         fs_info->btree_inode->i_nlink = 1;
1524
1525         fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1526
1527         INIT_LIST_HEAD(&fs_info->ordered_extents);
1528         spin_lock_init(&fs_info->ordered_extent_lock);
1529
1530         sb->s_blocksize = 4096;
1531         sb->s_blocksize_bits = blksize_bits(4096);
1532
1533         /*
1534          * we set the i_size on the btree inode to the max possible int.
1535          * the real end of the address space is determined by all of
1536          * the devices in the system
1537          */
1538         fs_info->btree_inode->i_size = OFFSET_MAX;
1539         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1540         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1541
1542         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1543                              fs_info->btree_inode->i_mapping,
1544                              GFP_NOFS);
1545         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1546                              GFP_NOFS);
1547
1548         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1549
1550         spin_lock_init(&fs_info->block_group_cache_lock);
1551         fs_info->block_group_cache_tree.rb_node = NULL;
1552
1553         extent_io_tree_init(&fs_info->pinned_extents,
1554                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1555         extent_io_tree_init(&fs_info->pending_del,
1556                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1557         extent_io_tree_init(&fs_info->extent_ins,
1558                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1559         fs_info->do_barriers = 1;
1560
1561         INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
1562         btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
1563         btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
1564
1565         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1566         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1567                sizeof(struct btrfs_key));
1568         insert_inode_hash(fs_info->btree_inode);
1569
1570         mutex_init(&fs_info->trans_mutex);
1571         mutex_init(&fs_info->tree_log_mutex);
1572         mutex_init(&fs_info->drop_mutex);
1573         mutex_init(&fs_info->extent_ins_mutex);
1574         mutex_init(&fs_info->pinned_mutex);
1575         mutex_init(&fs_info->chunk_mutex);
1576         mutex_init(&fs_info->transaction_kthread_mutex);
1577         mutex_init(&fs_info->cleaner_mutex);
1578         mutex_init(&fs_info->volume_mutex);
1579         mutex_init(&fs_info->tree_reloc_mutex);
1580         init_waitqueue_head(&fs_info->transaction_throttle);
1581         init_waitqueue_head(&fs_info->transaction_wait);
1582         init_waitqueue_head(&fs_info->async_submit_wait);
1583         init_waitqueue_head(&fs_info->tree_log_wait);
1584         atomic_set(&fs_info->tree_log_commit, 0);
1585         atomic_set(&fs_info->tree_log_writers, 0);
1586         fs_info->tree_log_transid = 0;
1587
1588 #if 0
1589         ret = add_hasher(fs_info, "crc32c");
1590         if (ret) {
1591                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1592                 err = -ENOMEM;
1593                 goto fail_iput;
1594         }
1595 #endif
1596         __setup_root(4096, 4096, 4096, 4096, tree_root,
1597                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1598
1599
1600         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1601         if (!bh)
1602                 goto fail_iput;
1603
1604         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1605         brelse(bh);
1606
1607         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1608
1609         disk_super = &fs_info->super_copy;
1610         if (!btrfs_super_root(disk_super))
1611                 goto fail_iput;
1612
1613         ret = btrfs_parse_options(tree_root, options);
1614         if (ret) {
1615                 err = ret;
1616                 goto fail_iput;
1617         }
1618
1619         features = btrfs_super_incompat_flags(disk_super) &
1620                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1621         if (features) {
1622                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1623                        "unsupported optional features (%Lx).\n",
1624                        features);
1625                 err = -EINVAL;
1626                 goto fail_iput;
1627         }
1628
1629         features = btrfs_super_compat_ro_flags(disk_super) &
1630                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1631         if (!(sb->s_flags & MS_RDONLY) && features) {
1632                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1633                        "unsupported option features (%Lx).\n",
1634                        features);
1635                 err = -EINVAL;
1636                 goto fail_iput;
1637         }
1638
1639         /*
1640          * we need to start all the end_io workers up front because the
1641          * queue work function gets called at interrupt time, and so it
1642          * cannot dynamically grow.
1643          */
1644         btrfs_init_workers(&fs_info->workers, "worker",
1645                            fs_info->thread_pool_size);
1646
1647         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1648                            fs_info->thread_pool_size);
1649
1650         btrfs_init_workers(&fs_info->submit_workers, "submit",
1651                            min_t(u64, fs_devices->num_devices,
1652                            fs_info->thread_pool_size));
1653
1654         /* a higher idle thresh on the submit workers makes it much more
1655          * likely that bios will be send down in a sane order to the
1656          * devices
1657          */
1658         fs_info->submit_workers.idle_thresh = 64;
1659
1660         fs_info->workers.idle_thresh = 16;
1661         fs_info->workers.ordered = 1;
1662
1663         fs_info->delalloc_workers.idle_thresh = 2;
1664         fs_info->delalloc_workers.ordered = 1;
1665
1666         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1667         btrfs_init_workers(&fs_info->endio_workers, "endio",
1668                            fs_info->thread_pool_size);
1669         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1670                            fs_info->thread_pool_size);
1671         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1672                            fs_info->thread_pool_size);
1673
1674         /*
1675          * endios are largely parallel and should have a very
1676          * low idle thresh
1677          */
1678         fs_info->endio_workers.idle_thresh = 4;
1679         fs_info->endio_write_workers.idle_thresh = 64;
1680
1681         btrfs_start_workers(&fs_info->workers, 1);
1682         btrfs_start_workers(&fs_info->submit_workers, 1);
1683         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1684         btrfs_start_workers(&fs_info->fixup_workers, 1);
1685         btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1686         btrfs_start_workers(&fs_info->endio_meta_workers,
1687                             fs_info->thread_pool_size);
1688         btrfs_start_workers(&fs_info->endio_write_workers,
1689                             fs_info->thread_pool_size);
1690
1691         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1692         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1693                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1694
1695         nodesize = btrfs_super_nodesize(disk_super);
1696         leafsize = btrfs_super_leafsize(disk_super);
1697         sectorsize = btrfs_super_sectorsize(disk_super);
1698         stripesize = btrfs_super_stripesize(disk_super);
1699         tree_root->nodesize = nodesize;
1700         tree_root->leafsize = leafsize;
1701         tree_root->sectorsize = sectorsize;
1702         tree_root->stripesize = stripesize;
1703
1704         sb->s_blocksize = sectorsize;
1705         sb->s_blocksize_bits = blksize_bits(sectorsize);
1706
1707         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1708                     sizeof(disk_super->magic))) {
1709                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1710                 goto fail_sb_buffer;
1711         }
1712
1713         mutex_lock(&fs_info->chunk_mutex);
1714         ret = btrfs_read_sys_array(tree_root);
1715         mutex_unlock(&fs_info->chunk_mutex);
1716         if (ret) {
1717                 printk("btrfs: failed to read the system array on %s\n",
1718                        sb->s_id);
1719                 goto fail_sys_array;
1720         }
1721
1722         blocksize = btrfs_level_size(tree_root,
1723                                      btrfs_super_chunk_root_level(disk_super));
1724         generation = btrfs_super_chunk_root_generation(disk_super);
1725
1726         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1727                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1728
1729         chunk_root->node = read_tree_block(chunk_root,
1730                                            btrfs_super_chunk_root(disk_super),
1731                                            blocksize, generation);
1732         BUG_ON(!chunk_root->node);
1733
1734         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1735                  (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1736                  BTRFS_UUID_SIZE);
1737
1738         mutex_lock(&fs_info->chunk_mutex);
1739         ret = btrfs_read_chunk_tree(chunk_root);
1740         mutex_unlock(&fs_info->chunk_mutex);
1741         if (ret) {
1742                 printk("btrfs: failed to read chunk tree on %s\n", sb->s_id);
1743                 goto fail_chunk_root;
1744         }
1745
1746         btrfs_close_extra_devices(fs_devices);
1747
1748         blocksize = btrfs_level_size(tree_root,
1749                                      btrfs_super_root_level(disk_super));
1750         generation = btrfs_super_generation(disk_super);
1751
1752         tree_root->node = read_tree_block(tree_root,
1753                                           btrfs_super_root(disk_super),
1754                                           blocksize, generation);
1755         if (!tree_root->node)
1756                 goto fail_chunk_root;
1757
1758
1759         ret = find_and_setup_root(tree_root, fs_info,
1760                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1761         if (ret)
1762                 goto fail_tree_root;
1763         extent_root->track_dirty = 1;
1764
1765         ret = find_and_setup_root(tree_root, fs_info,
1766                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1767         dev_root->track_dirty = 1;
1768
1769         if (ret)
1770                 goto fail_extent_root;
1771
1772         ret = find_and_setup_root(tree_root, fs_info,
1773                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
1774         if (ret)
1775                 goto fail_extent_root;
1776
1777         csum_root->track_dirty = 1;
1778
1779         btrfs_read_block_groups(extent_root);
1780
1781         fs_info->generation = generation + 1;
1782         fs_info->last_trans_committed = generation;
1783         fs_info->data_alloc_profile = (u64)-1;
1784         fs_info->metadata_alloc_profile = (u64)-1;
1785         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1786         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1787                                                "btrfs-cleaner");
1788         if (!fs_info->cleaner_kthread)
1789                 goto fail_csum_root;
1790
1791         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1792                                                    tree_root,
1793                                                    "btrfs-transaction");
1794         if (!fs_info->transaction_kthread)
1795                 goto fail_cleaner;
1796
1797         if (btrfs_super_log_root(disk_super) != 0) {
1798                 u64 bytenr = btrfs_super_log_root(disk_super);
1799
1800                 if (fs_devices->rw_devices == 0) {
1801                         printk("Btrfs log replay required on RO media\n");
1802                         err = -EIO;
1803                         goto fail_trans_kthread;
1804                 }
1805                 blocksize =
1806                      btrfs_level_size(tree_root,
1807                                       btrfs_super_log_root_level(disk_super));
1808
1809                 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1810                                                       GFP_NOFS);
1811
1812                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1813                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1814
1815                 log_tree_root->node = read_tree_block(tree_root, bytenr,
1816                                                       blocksize,
1817                                                       generation + 1);
1818                 ret = btrfs_recover_log_trees(log_tree_root);
1819                 BUG_ON(ret);
1820
1821                 if (sb->s_flags & MS_RDONLY) {
1822                         ret =  btrfs_commit_super(tree_root);
1823                         BUG_ON(ret);
1824                 }
1825         }
1826
1827         if (!(sb->s_flags & MS_RDONLY)) {
1828                 ret = btrfs_cleanup_reloc_trees(tree_root);
1829                 BUG_ON(ret);
1830         }
1831
1832         location.objectid = BTRFS_FS_TREE_OBJECTID;
1833         location.type = BTRFS_ROOT_ITEM_KEY;
1834         location.offset = (u64)-1;
1835
1836         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1837         if (!fs_info->fs_root)
1838                 goto fail_trans_kthread;
1839         return tree_root;
1840
1841 fail_trans_kthread:
1842         kthread_stop(fs_info->transaction_kthread);
1843 fail_cleaner:
1844         kthread_stop(fs_info->cleaner_kthread);
1845
1846         /*
1847          * make sure we're done with the btree inode before we stop our
1848          * kthreads
1849          */
1850         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1851         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1852
1853 fail_csum_root:
1854         free_extent_buffer(csum_root->node);
1855 fail_extent_root:
1856         free_extent_buffer(extent_root->node);
1857 fail_tree_root:
1858         free_extent_buffer(tree_root->node);
1859 fail_chunk_root:
1860         free_extent_buffer(chunk_root->node);
1861 fail_sys_array:
1862         free_extent_buffer(dev_root->node);
1863 fail_sb_buffer:
1864         btrfs_stop_workers(&fs_info->fixup_workers);
1865         btrfs_stop_workers(&fs_info->delalloc_workers);
1866         btrfs_stop_workers(&fs_info->workers);
1867         btrfs_stop_workers(&fs_info->endio_workers);
1868         btrfs_stop_workers(&fs_info->endio_meta_workers);
1869         btrfs_stop_workers(&fs_info->endio_write_workers);
1870         btrfs_stop_workers(&fs_info->submit_workers);
1871 fail_iput:
1872         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1873         iput(fs_info->btree_inode);
1874 fail:
1875         btrfs_close_devices(fs_info->fs_devices);
1876         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1877
1878         kfree(extent_root);
1879         kfree(tree_root);
1880         bdi_destroy(&fs_info->bdi);
1881         kfree(fs_info);
1882         kfree(chunk_root);
1883         kfree(dev_root);
1884         kfree(csum_root);
1885         return ERR_PTR(err);
1886 }
1887
1888 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1889 {
1890         char b[BDEVNAME_SIZE];
1891
1892         if (uptodate) {
1893                 set_buffer_uptodate(bh);
1894         } else {
1895                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1896                         printk(KERN_WARNING "lost page write due to "
1897                                         "I/O error on %s\n",
1898                                        bdevname(bh->b_bdev, b));
1899                 }
1900                 /* note, we dont' set_buffer_write_io_error because we have
1901                  * our own ways of dealing with the IO errors
1902                  */
1903                 clear_buffer_uptodate(bh);
1904         }
1905         unlock_buffer(bh);
1906         put_bh(bh);
1907 }
1908
1909 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
1910 {
1911         struct buffer_head *bh;
1912         struct buffer_head *latest = NULL;
1913         struct btrfs_super_block *super;
1914         int i;
1915         u64 transid = 0;
1916         u64 bytenr;
1917
1918         /* we would like to check all the supers, but that would make
1919          * a btrfs mount succeed after a mkfs from a different FS.
1920          * So, we need to add a special mount option to scan for
1921          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1922          */
1923         for (i = 0; i < 1; i++) {
1924                 bytenr = btrfs_sb_offset(i);
1925                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
1926                         break;
1927                 bh = __bread(bdev, bytenr / 4096, 4096);
1928                 if (!bh)
1929                         continue;
1930
1931                 super = (struct btrfs_super_block *)bh->b_data;
1932                 if (btrfs_super_bytenr(super) != bytenr ||
1933                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
1934                             sizeof(super->magic))) {
1935                         brelse(bh);
1936                         continue;
1937                 }
1938
1939                 if (!latest || btrfs_super_generation(super) > transid) {
1940                         brelse(latest);
1941                         latest = bh;
1942                         transid = btrfs_super_generation(super);
1943                 } else {
1944                         brelse(bh);
1945                 }
1946         }
1947         return latest;
1948 }
1949
1950 static int write_dev_supers(struct btrfs_device *device,
1951                             struct btrfs_super_block *sb,
1952                             int do_barriers, int wait, int max_mirrors)
1953 {
1954         struct buffer_head *bh;
1955         int i;
1956         int ret;
1957         int errors = 0;
1958         u32 crc;
1959         u64 bytenr;
1960         int last_barrier = 0;
1961
1962         if (max_mirrors == 0)
1963                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
1964
1965         /* make sure only the last submit_bh does a barrier */
1966         if (do_barriers) {
1967                 for (i = 0; i < max_mirrors; i++) {
1968                         bytenr = btrfs_sb_offset(i);
1969                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1970                             device->total_bytes)
1971                                 break;
1972                         last_barrier = i;
1973                 }
1974         }
1975
1976         for (i = 0; i < max_mirrors; i++) {
1977                 bytenr = btrfs_sb_offset(i);
1978                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
1979                         break;
1980
1981                 if (wait) {
1982                         bh = __find_get_block(device->bdev, bytenr / 4096,
1983                                               BTRFS_SUPER_INFO_SIZE);
1984                         BUG_ON(!bh);
1985                         brelse(bh);
1986                         wait_on_buffer(bh);
1987                         if (buffer_uptodate(bh)) {
1988                                 brelse(bh);
1989                                 continue;
1990                         }
1991                 } else {
1992                         btrfs_set_super_bytenr(sb, bytenr);
1993
1994                         crc = ~(u32)0;
1995                         crc = btrfs_csum_data(NULL, (char *)sb +
1996                                               BTRFS_CSUM_SIZE, crc,
1997                                               BTRFS_SUPER_INFO_SIZE -
1998                                               BTRFS_CSUM_SIZE);
1999                         btrfs_csum_final(crc, sb->csum);
2000
2001                         bh = __getblk(device->bdev, bytenr / 4096,
2002                                       BTRFS_SUPER_INFO_SIZE);
2003                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2004
2005                         set_buffer_uptodate(bh);
2006                         get_bh(bh);
2007                         lock_buffer(bh);
2008                         bh->b_end_io = btrfs_end_buffer_write_sync;
2009                 }
2010
2011                 if (i == last_barrier && do_barriers && device->barriers) {
2012                         ret = submit_bh(WRITE_BARRIER, bh);
2013                         if (ret == -EOPNOTSUPP) {
2014                                 printk("btrfs: disabling barriers on dev %s\n",
2015                                        device->name);
2016                                 set_buffer_uptodate(bh);
2017                                 device->barriers = 0;
2018                                 get_bh(bh);
2019                                 lock_buffer(bh);
2020                                 ret = submit_bh(WRITE, bh);
2021                         }
2022                 } else {
2023                         ret = submit_bh(WRITE, bh);
2024                 }
2025
2026                 if (!ret && wait) {
2027                         wait_on_buffer(bh);
2028                         if (!buffer_uptodate(bh))
2029                                 errors++;
2030                 } else if (ret) {
2031                         errors++;
2032                 }
2033                 if (wait)
2034                         brelse(bh);
2035         }
2036         return errors < i ? 0 : -1;
2037 }
2038
2039 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2040 {
2041         struct list_head *cur;
2042         struct list_head *head = &root->fs_info->fs_devices->devices;
2043         struct btrfs_device *dev;
2044         struct btrfs_super_block *sb;
2045         struct btrfs_dev_item *dev_item;
2046         int ret;
2047         int do_barriers;
2048         int max_errors;
2049         int total_errors = 0;
2050         u64 flags;
2051
2052         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2053         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2054
2055         sb = &root->fs_info->super_for_commit;
2056         dev_item = &sb->dev_item;
2057         list_for_each(cur, head) {
2058                 dev = list_entry(cur, struct btrfs_device, dev_list);
2059                 if (!dev->bdev) {
2060                         total_errors++;
2061                         continue;
2062                 }
2063                 if (!dev->in_fs_metadata || !dev->writeable)
2064                         continue;
2065
2066                 btrfs_set_stack_device_generation(dev_item, 0);
2067                 btrfs_set_stack_device_type(dev_item, dev->type);
2068                 btrfs_set_stack_device_id(dev_item, dev->devid);
2069                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2070                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2071                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2072                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2073                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2074                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2075                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2076
2077                 flags = btrfs_super_flags(sb);
2078                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2079
2080                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2081                 if (ret)
2082                         total_errors++;
2083         }
2084         if (total_errors > max_errors) {
2085                 printk("btrfs: %d errors while writing supers\n", total_errors);
2086                 BUG();
2087         }
2088
2089         total_errors = 0;
2090         list_for_each(cur, head) {
2091                 dev = list_entry(cur, struct btrfs_device, dev_list);
2092                 if (!dev->bdev)
2093                         continue;
2094                 if (!dev->in_fs_metadata || !dev->writeable)
2095                         continue;
2096
2097                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2098                 if (ret)
2099                         total_errors++;
2100         }
2101         if (total_errors > max_errors) {
2102                 printk("btrfs: %d errors while writing supers\n", total_errors);
2103                 BUG();
2104         }
2105         return 0;
2106 }
2107
2108 int write_ctree_super(struct btrfs_trans_handle *trans,
2109                       struct btrfs_root *root, int max_mirrors)
2110 {
2111         int ret;
2112
2113         ret = write_all_supers(root, max_mirrors);
2114         return ret;
2115 }
2116
2117 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2118 {
2119         radix_tree_delete(&fs_info->fs_roots_radix,
2120                           (unsigned long)root->root_key.objectid);
2121         if (root->anon_super.s_dev) {
2122                 down_write(&root->anon_super.s_umount);
2123                 kill_anon_super(&root->anon_super);
2124         }
2125 #if 0
2126         if (root->in_sysfs)
2127                 btrfs_sysfs_del_root(root);
2128 #endif
2129         if (root->node)
2130                 free_extent_buffer(root->node);
2131         if (root->commit_root)
2132                 free_extent_buffer(root->commit_root);
2133         if (root->name)
2134                 kfree(root->name);
2135         kfree(root);
2136         return 0;
2137 }
2138
2139 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2140 {
2141         int ret;
2142         struct btrfs_root *gang[8];
2143         int i;
2144
2145         while(1) {
2146                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2147                                              (void **)gang, 0,
2148                                              ARRAY_SIZE(gang));
2149                 if (!ret)
2150                         break;
2151                 for (i = 0; i < ret; i++)
2152                         btrfs_free_fs_root(fs_info, gang[i]);
2153         }
2154         return 0;
2155 }
2156
2157 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2158 {
2159         u64 root_objectid = 0;
2160         struct btrfs_root *gang[8];
2161         int i;
2162         int ret;
2163
2164         while (1) {
2165                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2166                                              (void **)gang, root_objectid,
2167                                              ARRAY_SIZE(gang));
2168                 if (!ret)
2169                         break;
2170                 for (i = 0; i < ret; i++) {
2171                         root_objectid = gang[i]->root_key.objectid;
2172                         ret = btrfs_find_dead_roots(fs_info->tree_root,
2173                                                     root_objectid, gang[i]);
2174                         BUG_ON(ret);
2175                         btrfs_orphan_cleanup(gang[i]);
2176                 }
2177                 root_objectid++;
2178         }
2179         return 0;
2180 }
2181
2182 int btrfs_commit_super(struct btrfs_root *root)
2183 {
2184         struct btrfs_trans_handle *trans;
2185         int ret;
2186
2187         mutex_lock(&root->fs_info->cleaner_mutex);
2188         btrfs_clean_old_snapshots(root);
2189         mutex_unlock(&root->fs_info->cleaner_mutex);
2190         trans = btrfs_start_transaction(root, 1);
2191         ret = btrfs_commit_transaction(trans, root);
2192         BUG_ON(ret);
2193         /* run commit again to drop the original snapshot */
2194         trans = btrfs_start_transaction(root, 1);
2195         btrfs_commit_transaction(trans, root);
2196         ret = btrfs_write_and_wait_transaction(NULL, root);
2197         BUG_ON(ret);
2198
2199         ret = write_ctree_super(NULL, root, 0);
2200         return ret;
2201 }
2202
2203 int close_ctree(struct btrfs_root *root)
2204 {
2205         struct btrfs_fs_info *fs_info = root->fs_info;
2206         int ret;
2207
2208         fs_info->closing = 1;
2209         smp_mb();
2210
2211         kthread_stop(root->fs_info->transaction_kthread);
2212         kthread_stop(root->fs_info->cleaner_kthread);
2213
2214         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2215                 ret =  btrfs_commit_super(root);
2216                 if (ret) {
2217                         printk("btrfs: commit super returns %d\n", ret);
2218                 }
2219         }
2220
2221         if (fs_info->delalloc_bytes) {
2222                 printk("btrfs: at unmount delalloc count %Lu\n",
2223                        fs_info->delalloc_bytes);
2224         }
2225         if (fs_info->total_ref_cache_size) {
2226                 printk("btrfs: at umount reference cache size %Lu\n",
2227                         fs_info->total_ref_cache_size);
2228         }
2229
2230         if (fs_info->extent_root->node)
2231                 free_extent_buffer(fs_info->extent_root->node);
2232
2233         if (fs_info->tree_root->node)
2234                 free_extent_buffer(fs_info->tree_root->node);
2235
2236         if (root->fs_info->chunk_root->node);
2237                 free_extent_buffer(root->fs_info->chunk_root->node);
2238
2239         if (root->fs_info->dev_root->node);
2240                 free_extent_buffer(root->fs_info->dev_root->node);
2241
2242         if (root->fs_info->csum_root->node);
2243                 free_extent_buffer(root->fs_info->csum_root->node);
2244
2245         btrfs_free_block_groups(root->fs_info);
2246
2247         del_fs_roots(fs_info);
2248
2249         iput(fs_info->btree_inode);
2250
2251         btrfs_stop_workers(&fs_info->fixup_workers);
2252         btrfs_stop_workers(&fs_info->delalloc_workers);
2253         btrfs_stop_workers(&fs_info->workers);
2254         btrfs_stop_workers(&fs_info->endio_workers);
2255         btrfs_stop_workers(&fs_info->endio_meta_workers);
2256         btrfs_stop_workers(&fs_info->endio_write_workers);
2257         btrfs_stop_workers(&fs_info->submit_workers);
2258
2259 #if 0
2260         while(!list_empty(&fs_info->hashers)) {
2261                 struct btrfs_hasher *hasher;
2262                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
2263                                     hashers);
2264                 list_del(&hasher->hashers);
2265                 crypto_free_hash(&fs_info->hash_tfm);
2266                 kfree(hasher);
2267         }
2268 #endif
2269         btrfs_close_devices(fs_info->fs_devices);
2270         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2271
2272         bdi_destroy(&fs_info->bdi);
2273
2274         kfree(fs_info->extent_root);
2275         kfree(fs_info->tree_root);
2276         kfree(fs_info->chunk_root);
2277         kfree(fs_info->dev_root);
2278         kfree(fs_info->csum_root);
2279         return 0;
2280 }
2281
2282 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2283 {
2284         int ret;
2285         struct inode *btree_inode = buf->first_page->mapping->host;
2286
2287         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2288         if (!ret)
2289                 return ret;
2290
2291         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2292                                     parent_transid);
2293         return !ret;
2294 }
2295
2296 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2297 {
2298         struct inode *btree_inode = buf->first_page->mapping->host;
2299         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2300                                           buf);
2301 }
2302
2303 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2304 {
2305         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2306         u64 transid = btrfs_header_generation(buf);
2307         struct inode *btree_inode = root->fs_info->btree_inode;
2308
2309         WARN_ON(!btrfs_tree_locked(buf));
2310         if (transid != root->fs_info->generation) {
2311                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
2312                         (unsigned long long)buf->start,
2313                         transid, root->fs_info->generation);
2314                 WARN_ON(1);
2315         }
2316         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
2317 }
2318
2319 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2320 {
2321         /*
2322          * looks as though older kernels can get into trouble with
2323          * this code, they end up stuck in balance_dirty_pages forever
2324          */
2325         struct extent_io_tree *tree;
2326         u64 num_dirty;
2327         u64 start = 0;
2328         unsigned long thresh = 32 * 1024 * 1024;
2329         tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
2330
2331         if (current_is_pdflush() || current->flags & PF_MEMALLOC)
2332                 return;
2333
2334         num_dirty = count_range_bits(tree, &start, (u64)-1,
2335                                      thresh, EXTENT_DIRTY);
2336         if (num_dirty > thresh) {
2337                 balance_dirty_pages_ratelimited_nr(
2338                                    root->fs_info->btree_inode->i_mapping, 1);
2339         }
2340         return;
2341 }
2342
2343 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2344 {
2345         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2346         int ret;
2347         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2348         if (ret == 0) {
2349                 buf->flags |= EXTENT_UPTODATE;
2350         }
2351         return ret;
2352 }
2353
2354 int btree_lock_page_hook(struct page *page)
2355 {
2356         struct inode *inode = page->mapping->host;
2357         struct btrfs_root *root = BTRFS_I(inode)->root;
2358         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2359         struct extent_buffer *eb;
2360         unsigned long len;
2361         u64 bytenr = page_offset(page);
2362
2363         if (page->private == EXTENT_PAGE_PRIVATE)
2364                 goto out;
2365
2366         len = page->private >> 2;
2367         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2368         if (!eb)
2369                 goto out;
2370
2371         btrfs_tree_lock(eb);
2372         spin_lock(&root->fs_info->hash_lock);
2373         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2374         spin_unlock(&root->fs_info->hash_lock);
2375         btrfs_tree_unlock(eb);
2376         free_extent_buffer(eb);
2377 out:
2378         lock_page(page);
2379         return 0;
2380 }
2381
2382 static struct extent_io_ops btree_extent_io_ops = {
2383         .write_cache_pages_lock_hook = btree_lock_page_hook,
2384         .readpage_end_io_hook = btree_readpage_end_io_hook,
2385         .submit_bio_hook = btree_submit_bio_hook,
2386         /* note we're sharing with inode.c for the merge bio hook */
2387         .merge_bio_hook = btrfs_merge_bio_hook,
2388 };