fs/btrfs/scrub.c

   1 /*
   2  * Copyright (C) 2011 STRATO.  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/blkdev.h>
  20 #include <linux/ratelimit.h>
  21 #include "ctree.h"
  22 #include "volumes.h"
  23 #include "disk-io.h"
  24 #include "ordered-data.h"
  25 #include "backref.h"
  26
  27 /*
  28  * This is only the first step towards a full-features scrub. It reads all
  29  * extent and super block and verifies the checksums. In case a bad checksum
  30  * is found or the extent cannot be read, good data will be written back if
  31  * any can be found.
  32  *
  33  * Future enhancements:
  34  *  - To enhance the performance, better read-ahead strategies for the
  35  *    extent-tree can be employed.
  36  *  - In case an unrepairable extent is encountered, track which files are
  37  *    affected and report them
  38  *  - In case of a read error on files with nodatasum, map the file and read
  39  *    the extent to trigger a writeback of the good copy
  40  *  - track and record media errors, throw out bad devices
  41  *  - add a mode to also read unallocated space
  42  *  - make the prefetch cancellable
  43  */
  44
  45 struct scrub_bio;
  46 struct scrub_page;
  47 struct scrub_dev;
  48 static void scrub_bio_end_io(struct bio *bio, int err);
  49 static void scrub_checksum(struct btrfs_work *work);
  50 static int scrub_checksum_data(struct scrub_dev *sdev,
  51                                struct scrub_page *spag, void *buffer);
  52 static int scrub_checksum_tree_block(struct scrub_dev *sdev,
  53                                      struct scrub_page *spag, u64 logical,
  54                                      void *buffer);
  55 static int scrub_checksum_super(struct scrub_bio *sbio, void *buffer);
  56 static int scrub_fixup_check(struct scrub_bio *sbio, int ix);
  57 static void scrub_fixup_end_io(struct bio *bio, int err);
  58 static int scrub_fixup_io(int rw, struct block_device *bdev, sector_t sector,
  59                           struct page *page);
  60 static void scrub_fixup(struct scrub_bio *sbio, int ix);
  61
  62 #define SCRUB_PAGES_PER_BIO     16      /* 64k per bio */
  63 #define SCRUB_BIOS_PER_DEV      16      /* 1 MB per device in flight */
  64
  65 struct scrub_page {
  66         u64                     flags;  /* extent flags */
  67         u64                     generation;
  68         u64                     mirror_num;
  69         int                     have_csum;
  70         u8                      csum[BTRFS_CSUM_SIZE];
  71 };
  72
  73 struct scrub_bio {
  74         int                     index;
  75         struct scrub_dev        *sdev;
  76         struct bio              *bio;
  77         int                     err;
  78         u64                     logical;
  79         u64                     physical;
  80         struct scrub_page       spag[SCRUB_PAGES_PER_BIO];
  81         u64                     count;
  82         int                     next_free;
  83         struct btrfs_work       work;
  84 };
  85
  86 struct scrub_dev {
  87         struct scrub_bio        *bios[SCRUB_BIOS_PER_DEV];
  88         struct btrfs_device     *dev;
  89         int                     first_free;
  90         int                     curr;
  91         atomic_t                in_flight;
  92         spinlock_t              list_lock;
  93         wait_queue_head_t       list_wait;
  94         u16                     csum_size;
  95         struct list_head        csum_list;
  96         atomic_t                cancel_req;
  97         int                     readonly;
  98         /*
  99          * statistics
 100          */
 101         struct btrfs_scrub_progress stat;
 102         spinlock_t              stat_lock;
 103 };
 104
 105 struct scrub_warning {
 106         struct btrfs_path       *path;
 107         u64                     extent_item_size;
 108         char                    *scratch_buf;
 109         char                    *msg_buf;
 110         const char              *errstr;
 111         sector_t                sector;
 112         u64                     logical;
 113         struct btrfs_device     *dev;
 114         int                     msg_bufsize;
 115         int                     scratch_bufsize;
 116 };
 117
 118 static void scrub_free_csums(struct scrub_dev *sdev)
 119 {
 120         while (!list_empty(&sdev->csum_list)) {
 121                 struct btrfs_ordered_sum *sum;
 122                 sum = list_first_entry(&sdev->csum_list,
 123                                        struct btrfs_ordered_sum, list);
 124                 list_del(&sum->list);
 125                 kfree(sum);
 126         }
 127 }
 128
 129 static void scrub_free_bio(struct bio *bio)
 130 {
 131         int i;
 132         struct page *last_page = NULL;
 133
 134         if (!bio)
 135                 return;
 136
 137         for (i = 0; i < bio->bi_vcnt; ++i) {
 138                 if (bio->bi_io_vec[i].bv_page == last_page)
 139                         continue;
 140                 last_page = bio->bi_io_vec[i].bv_page;
 141                 __free_page(last_page);
 142         }
 143         bio_put(bio);
 144 }
 145
 146 static noinline_for_stack void scrub_free_dev(struct scrub_dev *sdev)
 147 {
 148         int i;
 149
 150         if (!sdev)
 151                 return;
 152
 153         for (i = 0; i < SCRUB_BIOS_PER_DEV; ++i) {
 154                 struct scrub_bio *sbio = sdev->bios[i];
 155
 156                 if (!sbio)
 157                         break;
 158
 159                 scrub_free_bio(sbio->bio);
 160                 kfree(sbio);
 161         }
 162
 163         scrub_free_csums(sdev);
 164         kfree(sdev);
 165 }
 166
 167 static noinline_for_stack
 168 struct scrub_dev *scrub_setup_dev(struct btrfs_device *dev)
 169 {
 170         struct scrub_dev *sdev;
 171         int             i;
 172         struct btrfs_fs_info *fs_info = dev->dev_root->fs_info;
 173
 174         sdev = kzalloc(sizeof(*sdev), GFP_NOFS);
 175         if (!sdev)
 176                 goto nomem;
 177         sdev->dev = dev;
 178         for (i = 0; i < SCRUB_BIOS_PER_DEV; ++i) {
 179                 struct scrub_bio *sbio;
 180
 181                 sbio = kzalloc(sizeof(*sbio), GFP_NOFS);
 182                 if (!sbio)
 183                         goto nomem;
 184                 sdev->bios[i] = sbio;
 185
 186                 sbio->index = i;
 187                 sbio->sdev = sdev;
 188                 sbio->count = 0;
 189                 sbio->work.func = scrub_checksum;
 190
 191                 if (i != SCRUB_BIOS_PER_DEV-1)
 192                         sdev->bios[i]->next_free = i + 1;
 193                  else
 194                         sdev->bios[i]->next_free = -1;
 195         }
 196         sdev->first_free = 0;
 197         sdev->curr = -1;
 198         atomic_set(&sdev->in_flight, 0);
 199         atomic_set(&sdev->cancel_req, 0);
 200         sdev->csum_size = btrfs_super_csum_size(&fs_info->super_copy);
 201         INIT_LIST_HEAD(&sdev->csum_list);
 202
 203         spin_lock_init(&sdev->list_lock);
 204         spin_lock_init(&sdev->stat_lock);
 205         init_waitqueue_head(&sdev->list_wait);
 206         return sdev;
 207
 208 nomem:
 209         scrub_free_dev(sdev);
 210         return ERR_PTR(-ENOMEM);
 211 }
 212
 213 static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root, void *ctx)
 214 {
 215         u64 isize;
 216         u32 nlink;
 217         int ret;
 218         int i;
 219         struct extent_buffer *eb;
 220         struct btrfs_inode_item *inode_item;
 221         struct scrub_warning *swarn = ctx;
 222         struct btrfs_fs_info *fs_info = swarn->dev->dev_root->fs_info;
 223         struct inode_fs_paths *ipath = NULL;
 224         struct btrfs_root *local_root;
 225         struct btrfs_key root_key;
 226
 227         root_key.objectid = root;
 228         root_key.type = BTRFS_ROOT_ITEM_KEY;
 229         root_key.offset = (u64)-1;
 230         local_root = btrfs_read_fs_root_no_name(fs_info, &root_key);
 231         if (IS_ERR(local_root)) {
 232                 ret = PTR_ERR(local_root);
 233                 goto err;
 234         }
 235
 236         ret = inode_item_info(inum, 0, local_root, swarn->path);
 237         if (ret) {
 238                 btrfs_release_path(swarn->path);
 239                 goto err;
 240         }
 241
 242         eb = swarn->path->nodes[0];
 243         inode_item = btrfs_item_ptr(eb, swarn->path->slots[0],
 244                                         struct btrfs_inode_item);
 245         isize = btrfs_inode_size(eb, inode_item);
 246         nlink = btrfs_inode_nlink(eb, inode_item);
 247         btrfs_release_path(swarn->path);
 248
 249         ipath = init_ipath(4096, local_root, swarn->path);
 250         ret = paths_from_inode(inum, ipath);
 251
 252         if (ret < 0)
 253                 goto err;
 254
 255         /*
 256          * we deliberately ignore the bit ipath might have been too small to
 257          * hold all of the paths here
 258          */
 259         for (i = 0; i < ipath->fspath->elem_cnt; ++i)
 260                 printk(KERN_WARNING "btrfs: %s at logical %llu on dev "
 261                         "%s, sector %llu, root %llu, inode %llu, offset %llu, "
 262                         "length %llu, links %u (path: %s)\n", swarn->errstr,
 263                         swarn->logical, swarn->dev->name,
 264                         (unsigned long long)swarn->sector, root, inum, offset,
 265                         min(isize - offset, (u64)PAGE_SIZE), nlink,
 266                         ipath->fspath->str[i]);
 267
 268         free_ipath(ipath);
 269         return 0;
 270
 271 err:
 272         printk(KERN_WARNING "btrfs: %s at logical %llu on dev "
 273                 "%s, sector %llu, root %llu, inode %llu, offset %llu: path "
 274                 "resolving failed with ret=%d\n", swarn->errstr,
 275                 swarn->logical, swarn->dev->name,
 276                 (unsigned long long)swarn->sector, root, inum, offset, ret);
 277
 278         free_ipath(ipath);
 279         return 0;
 280 }
 281
 282 static void scrub_print_warning(const char *errstr, struct scrub_bio *sbio,
 283                                 int ix)
 284 {
 285         struct btrfs_device *dev = sbio->sdev->dev;
 286         struct btrfs_fs_info *fs_info = dev->dev_root->fs_info;
 287         struct btrfs_path *path;
 288         struct btrfs_key found_key;
 289         struct extent_buffer *eb;
 290         struct btrfs_extent_item *ei;
 291         struct scrub_warning swarn;
 292         u32 item_size;
 293         int ret;
 294         u64 ref_root;
 295         u8 ref_level;
 296         unsigned long ptr = 0;
 297         const int bufsize = 4096;
 298         u64 extent_offset;
 299
 300         path = btrfs_alloc_path();
 301
 302         swarn.scratch_buf = kmalloc(bufsize, GFP_NOFS);
 303         swarn.msg_buf = kmalloc(bufsize, GFP_NOFS);
 304         swarn.sector = (sbio->physical + ix * PAGE_SIZE) >> 9;
 305         swarn.logical = sbio->logical + ix * PAGE_SIZE;
 306         swarn.errstr = errstr;
 307         swarn.dev = dev;
 308         swarn.msg_bufsize = bufsize;
 309         swarn.scratch_bufsize = bufsize;
 310
 311         if (!path || !swarn.scratch_buf || !swarn.msg_buf)
 312                 goto out;
 313
 314         ret = extent_from_logical(fs_info, swarn.logical, path, &found_key);
 315         if (ret < 0)
 316                 goto out;
 317
 318         extent_offset = swarn.logical - found_key.objectid;
 319         swarn.extent_item_size = found_key.offset;
 320
 321         eb = path->nodes[0];
 322         ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
 323         item_size = btrfs_item_size_nr(eb, path->slots[0]);
 324
 325         if (ret & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
 326                 do {
 327                         ret = tree_backref_for_extent(&ptr, eb, ei, item_size,
 328                                                         &ref_root, &ref_level);
 329                         printk(KERN_WARNING "%s at logical %llu on dev %s, "
 330                                 "sector %llu: metadata %s (level %d) in tree "
 331                                 "%llu\n", errstr, swarn.logical, dev->name,
 332                                 (unsigned long long)swarn.sector,
 333                                 ref_level ? "node" : "leaf",
 334                                 ret < 0 ? -1 : ref_level,
 335                                 ret < 0 ? -1 : ref_root);
 336                 } while (ret != 1);
 337         } else {
 338                 swarn.path = path;
 339                 iterate_extent_inodes(fs_info, path, found_key.objectid,
 340                                         extent_offset,
 341                                         scrub_print_warning_inode, &swarn);
 342         }
 343
 344 out:
 345         btrfs_free_path(path);
 346         kfree(swarn.scratch_buf);
 347         kfree(swarn.msg_buf);
 348 }
 349
 350 /*
 351  * scrub_recheck_error gets called when either verification of the page
 352  * failed or the bio failed to read, e.g. with EIO. In the latter case,
 353  * recheck_error gets called for every page in the bio, even though only
 354  * one may be bad
 355  */
 356 static int scrub_recheck_error(struct scrub_bio *sbio, int ix)
 357 {
 358         struct scrub_dev *sdev = sbio->sdev;
 359         u64 sector = (sbio->physical + ix * PAGE_SIZE) >> 9;
 360         static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
 361                                         DEFAULT_RATELIMIT_BURST);
 362
 363         if (sbio->err) {
 364                 if (scrub_fixup_io(READ, sbio->sdev->dev->bdev, sector,
 365                                    sbio->bio->bi_io_vec[ix].bv_page) == 0) {
 366                         if (scrub_fixup_check(sbio, ix) == 0)
 367                                 return 0;
 368                 }
 369                 if (__ratelimit(&_rs))
 370                         scrub_print_warning("i/o error", sbio, ix);
 371         } else {
 372                 if (__ratelimit(&_rs))
 373                         scrub_print_warning("checksum error", sbio, ix);
 374         }
 375
 376         spin_lock(&sdev->stat_lock);
 377         ++sdev->stat.read_errors;
 378         spin_unlock(&sdev->stat_lock);
 379
 380         scrub_fixup(sbio, ix);
 381         return 1;
 382 }
 383
 384 static int scrub_fixup_check(struct scrub_bio *sbio, int ix)
 385 {
 386         int ret = 1;
 387         struct page *page;
 388         void *buffer;
 389         u64 flags = sbio->spag[ix].flags;
 390
 391         page = sbio->bio->bi_io_vec[ix].bv_page;
 392         buffer = kmap_atomic(page, KM_USER0);
 393         if (flags & BTRFS_EXTENT_FLAG_DATA) {
 394                 ret = scrub_checksum_data(sbio->sdev,
 395                                           sbio->spag + ix, buffer);
 396         } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
 397                 ret = scrub_checksum_tree_block(sbio->sdev,
 398                                                 sbio->spag + ix,
 399                                                 sbio->logical + ix * PAGE_SIZE,
 400                                                 buffer);
 401         } else {
 402                 WARN_ON(1);
 403         }
 404         kunmap_atomic(buffer, KM_USER0);
 405
 406         return ret;
 407 }
 408
 409 static void scrub_fixup_end_io(struct bio *bio, int err)
 410 {
 411         complete((struct completion *)bio->bi_private);
 412 }
 413
 414 static void scrub_fixup(struct scrub_bio *sbio, int ix)
 415 {
 416         struct scrub_dev *sdev = sbio->sdev;
 417         struct btrfs_fs_info *fs_info = sdev->dev->dev_root->fs_info;
 418         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
 419         struct btrfs_multi_bio *multi = NULL;
 420         u64 logical = sbio->logical + ix * PAGE_SIZE;
 421         u64 length;
 422         int i;
 423         int ret;
 424         DECLARE_COMPLETION_ONSTACK(complete);
 425
 426         if ((sbio->spag[ix].flags & BTRFS_EXTENT_FLAG_DATA) &&
 427             (sbio->spag[ix].have_csum == 0)) {
 428                 /*
 429                  * nodatasum, don't try to fix anything
 430                  * FIXME: we can do better, open the inode and trigger a
 431                  * writeback
 432                  */
 433                 goto uncorrectable;
 434         }
 435
 436         length = PAGE_SIZE;
 437         ret = btrfs_map_block(map_tree, REQ_WRITE, logical, &length,
 438                               &multi, 0);
 439         if (ret || !multi || length < PAGE_SIZE) {
 440                 printk(KERN_ERR
 441                        "scrub_fixup: btrfs_map_block failed us for %llu\n",
 442                        (unsigned long long)logical);
 443                 WARN_ON(1);
 444                 return;
 445         }
 446
 447         if (multi->num_stripes == 1)
 448                 /* there aren't any replicas */
 449                 goto uncorrectable;
 450
 451         /*
 452          * first find a good copy
 453          */
 454         for (i = 0; i < multi->num_stripes; ++i) {
 455                 if (i == sbio->spag[ix].mirror_num)
 456                         continue;
 457
 458                 if (scrub_fixup_io(READ, multi->stripes[i].dev->bdev,
 459                                    multi->stripes[i].physical >> 9,
 460                                    sbio->bio->bi_io_vec[ix].bv_page)) {
 461                         /* I/O-error, this is not a good copy */
 462                         continue;
 463                 }
 464
 465                 if (scrub_fixup_check(sbio, ix) == 0)
 466                         break;
 467         }
 468         if (i == multi->num_stripes)
 469                 goto uncorrectable;
 470
 471         if (!sdev->readonly) {
 472                 /*
 473                  * bi_io_vec[ix].bv_page now contains good data, write it back
 474                  */
 475                 if (scrub_fixup_io(WRITE, sdev->dev->bdev,
 476                                    (sbio->physical + ix * PAGE_SIZE) >> 9,
 477                                    sbio->bio->bi_io_vec[ix].bv_page)) {
 478                         /* I/O-error, writeback failed, give up */
 479                         goto uncorrectable;
 480                 }
 481         }
 482
 483         kfree(multi);
 484         spin_lock(&sdev->stat_lock);
 485         ++sdev->stat.corrected_errors;
 486         spin_unlock(&sdev->stat_lock);
 487
 488         printk_ratelimited(KERN_ERR "btrfs: fixed up error at logical %llu\n",
 489                                (unsigned long long)logical);
 490         return;
 491
 492 uncorrectable:
 493         kfree(multi);
 494         spin_lock(&sdev->stat_lock);
 495         ++sdev->stat.uncorrectable_errors;
 496         spin_unlock(&sdev->stat_lock);
 497
 498         printk_ratelimited(KERN_ERR "btrfs: unable to fixup (regular) error at "
 499                                 "logical %llu\n", (unsigned long long)logical);
 500 }
 501
 502 static int scrub_fixup_io(int rw, struct block_device *bdev, sector_t sector,
 503                          struct page *page)
 504 {
 505         struct bio *bio = NULL;
 506         int ret;
 507         DECLARE_COMPLETION_ONSTACK(complete);
 508
 509         bio = bio_alloc(GFP_NOFS, 1);
 510         bio->bi_bdev = bdev;
 511         bio->bi_sector = sector;
 512         bio_add_page(bio, page, PAGE_SIZE, 0);
 513         bio->bi_end_io = scrub_fixup_end_io;
 514         bio->bi_private = &complete;
 515         submit_bio(rw, bio);
 516
 517         /* this will also unplug the queue */
 518         wait_for_completion(&complete);
 519
 520         ret = !test_bit(BIO_UPTODATE, &bio->bi_flags);
 521         bio_put(bio);
 522         return ret;
 523 }
 524
 525 static void scrub_bio_end_io(struct bio *bio, int err)
 526 {
 527         struct scrub_bio *sbio = bio->bi_private;
 528         struct scrub_dev *sdev = sbio->sdev;
 529         struct btrfs_fs_info *fs_info = sdev->dev->dev_root->fs_info;
 530
 531         sbio->err = err;
 532         sbio->bio = bio;
 533
 534         btrfs_queue_worker(&fs_info->scrub_workers, &sbio->work);
 535 }
 536
 537 static void scrub_checksum(struct btrfs_work *work)
 538 {
 539         struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
 540         struct scrub_dev *sdev = sbio->sdev;
 541         struct page *page;
 542         void *buffer;
 543         int i;
 544         u64 flags;
 545         u64 logical;
 546         int ret;
 547
 548         if (sbio->err) {
 549                 ret = 0;
 550                 for (i = 0; i < sbio->count; ++i)
 551                         ret |= scrub_recheck_error(sbio, i);
 552                 if (!ret) {
 553                         spin_lock(&sdev->stat_lock);
 554                         ++sdev->stat.unverified_errors;
 555                         spin_unlock(&sdev->stat_lock);
 556                 }
 557
 558                 sbio->bio->bi_flags &= ~(BIO_POOL_MASK - 1);
 559                 sbio->bio->bi_flags |= 1 << BIO_UPTODATE;
 560                 sbio->bio->bi_phys_segments = 0;
 561                 sbio->bio->bi_idx = 0;
 562
 563                 for (i = 0; i < sbio->count; i++) {
 564                         struct bio_vec *bi;
 565                         bi = &sbio->bio->bi_io_vec[i];
 566                         bi->bv_offset = 0;
 567                         bi->bv_len = PAGE_SIZE;
 568                 }
 569                 goto out;
 570         }
 571         for (i = 0; i < sbio->count; ++i) {
 572                 page = sbio->bio->bi_io_vec[i].bv_page;
 573                 buffer = kmap_atomic(page, KM_USER0);
 574                 flags = sbio->spag[i].flags;
 575                 logical = sbio->logical + i * PAGE_SIZE;
 576                 ret = 0;
 577                 if (flags & BTRFS_EXTENT_FLAG_DATA) {
 578                         ret = scrub_checksum_data(sdev, sbio->spag + i, buffer);
 579                 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
 580                         ret = scrub_checksum_tree_block(sdev, sbio->spag + i,
 581                                                         logical, buffer);
 582                 } else if (flags & BTRFS_EXTENT_FLAG_SUPER) {
 583                         BUG_ON(i);
 584                         (void)scrub_checksum_super(sbio, buffer);
 585                 } else {
 586                         WARN_ON(1);
 587                 }
 588                 kunmap_atomic(buffer, KM_USER0);
 589                 if (ret) {
 590                         ret = scrub_recheck_error(sbio, i);
 591                         if (!ret) {
 592                                 spin_lock(&sdev->stat_lock);
 593                                 ++sdev->stat.unverified_errors;
 594                                 spin_unlock(&sdev->stat_lock);
 595                         }
 596                 }
 597         }
 598
 599 out:
 600         scrub_free_bio(sbio->bio);
 601         sbio->bio = NULL;
 602         spin_lock(&sdev->list_lock);
 603         sbio->next_free = sdev->first_free;
 604         sdev->first_free = sbio->index;
 605         spin_unlock(&sdev->list_lock);
 606         atomic_dec(&sdev->in_flight);
 607         wake_up(&sdev->list_wait);
 608 }
 609
 610 static int scrub_checksum_data(struct scrub_dev *sdev,
 611                                struct scrub_page *spag, void *buffer)
 612 {
 613         u8 csum[BTRFS_CSUM_SIZE];
 614         u32 crc = ~(u32)0;
 615         int fail = 0;
 616         struct btrfs_root *root = sdev->dev->dev_root;
 617
 618         if (!spag->have_csum)
 619                 return 0;
 620
 621         crc = btrfs_csum_data(root, buffer, crc, PAGE_SIZE);
 622         btrfs_csum_final(crc, csum);
 623         if (memcmp(csum, spag->csum, sdev->csum_size))
 624                 fail = 1;
 625
 626         spin_lock(&sdev->stat_lock);
 627         ++sdev->stat.data_extents_scrubbed;
 628         sdev->stat.data_bytes_scrubbed += PAGE_SIZE;
 629         if (fail)
 630                 ++sdev->stat.csum_errors;
 631         spin_unlock(&sdev->stat_lock);
 632
 633         return fail;
 634 }
 635
 636 static int scrub_checksum_tree_block(struct scrub_dev *sdev,
 637                                      struct scrub_page *spag, u64 logical,
 638                                      void *buffer)
 639 {
 640         struct btrfs_header *h;
 641         struct btrfs_root *root = sdev->dev->dev_root;
 642         struct btrfs_fs_info *fs_info = root->fs_info;
 643         u8 csum[BTRFS_CSUM_SIZE];
 644         u32 crc = ~(u32)0;
 645         int fail = 0;
 646         int crc_fail = 0;
 647
 648         /*
 649          * we don't use the getter functions here, as we
 650          * a) don't have an extent buffer and
 651          * b) the page is already kmapped
 652          */
 653         h = (struct btrfs_header *)buffer;
 654
 655         if (logical != le64_to_cpu(h->bytenr))
 656                 ++fail;
 657
 658         if (spag->generation != le64_to_cpu(h->generation))
 659                 ++fail;
 660
 661         if (memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE))
 662                 ++fail;
 663
 664         if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
 665                    BTRFS_UUID_SIZE))
 666                 ++fail;
 667
 668         crc = btrfs_csum_data(root, buffer + BTRFS_CSUM_SIZE, crc,
 669                               PAGE_SIZE - BTRFS_CSUM_SIZE);
 670         btrfs_csum_final(crc, csum);
 671         if (memcmp(csum, h->csum, sdev->csum_size))
 672                 ++crc_fail;
 673
 674         spin_lock(&sdev->stat_lock);
 675         ++sdev->stat.tree_extents_scrubbed;
 676         sdev->stat.tree_bytes_scrubbed += PAGE_SIZE;
 677         if (crc_fail)
 678                 ++sdev->stat.csum_errors;
 679         if (fail)
 680                 ++sdev->stat.verify_errors;
 681         spin_unlock(&sdev->stat_lock);
 682
 683         return fail || crc_fail;
 684 }
 685
 686 static int scrub_checksum_super(struct scrub_bio *sbio, void *buffer)
 687 {
 688         struct btrfs_super_block *s;
 689         u64 logical;
 690         struct scrub_dev *sdev = sbio->sdev;
 691         struct btrfs_root *root = sdev->dev->dev_root;
 692         struct btrfs_fs_info *fs_info = root->fs_info;
 693         u8 csum[BTRFS_CSUM_SIZE];
 694         u32 crc = ~(u32)0;
 695         int fail = 0;
 696
 697         s = (struct btrfs_super_block *)buffer;
 698         logical = sbio->logical;
 699
 700         if (logical != le64_to_cpu(s->bytenr))
 701                 ++fail;
 702
 703         if (sbio->spag[0].generation != le64_to_cpu(s->generation))
 704                 ++fail;
 705
 706         if (memcmp(s->fsid, fs_info->fsid, BTRFS_UUID_SIZE))
 707                 ++fail;
 708
 709         crc = btrfs_csum_data(root, buffer + BTRFS_CSUM_SIZE, crc,
 710                               PAGE_SIZE - BTRFS_CSUM_SIZE);
 711         btrfs_csum_final(crc, csum);
 712         if (memcmp(csum, s->csum, sbio->sdev->csum_size))
 713                 ++fail;
 714
 715         if (fail) {
 716                 /*
 717                  * if we find an error in a super block, we just report it.
 718                  * They will get written with the next transaction commit
 719                  * anyway
 720                  */
 721                 spin_lock(&sdev->stat_lock);
 722                 ++sdev->stat.super_errors;
 723                 spin_unlock(&sdev->stat_lock);
 724         }
 725
 726         return fail;
 727 }
 728
 729 static int scrub_submit(struct scrub_dev *sdev)
 730 {
 731         struct scrub_bio *sbio;
 732         struct bio *bio;
 733         int i;
 734
 735         if (sdev->curr == -1)
 736                 return 0;
 737
 738         sbio = sdev->bios[sdev->curr];
 739
 740         bio = bio_alloc(GFP_NOFS, sbio->count);
 741         if (!bio)
 742                 goto nomem;
 743
 744         bio->bi_private = sbio;
 745         bio->bi_end_io = scrub_bio_end_io;
 746         bio->bi_bdev = sdev->dev->bdev;
 747         bio->bi_sector = sbio->physical >> 9;
 748
 749         for (i = 0; i < sbio->count; ++i) {
 750                 struct page *page;
 751                 int ret;
 752
 753                 page = alloc_page(GFP_NOFS);
 754                 if (!page)
 755                         goto nomem;
 756
 757                 ret = bio_add_page(bio, page, PAGE_SIZE, 0);
 758                 if (!ret) {
 759                         __free_page(page);
 760                         goto nomem;
 761                 }
 762         }
 763
 764         sbio->err = 0;
 765         sdev->curr = -1;
 766         atomic_inc(&sdev->in_flight);
 767
 768         submit_bio(READ, bio);
 769
 770         return 0;
 771
 772 nomem:
 773         scrub_free_bio(bio);
 774
 775         return -ENOMEM;
 776 }
 777
 778 static int scrub_page(struct scrub_dev *sdev, u64 logical, u64 len,
 779                       u64 physical, u64 flags, u64 gen, u64 mirror_num,
 780                       u8 *csum, int force)
 781 {
 782         struct scrub_bio *sbio;
 783
 784 again:
 785         /*
 786          * grab a fresh bio or wait for one to become available
 787          */
 788         while (sdev->curr == -1) {
 789                 spin_lock(&sdev->list_lock);
 790                 sdev->curr = sdev->first_free;
 791                 if (sdev->curr != -1) {
 792                         sdev->first_free = sdev->bios[sdev->curr]->next_free;
 793                         sdev->bios[sdev->curr]->next_free = -1;
 794                         sdev->bios[sdev->curr]->count = 0;
 795                         spin_unlock(&sdev->list_lock);
 796                 } else {
 797                         spin_unlock(&sdev->list_lock);
 798                         wait_event(sdev->list_wait, sdev->first_free != -1);
 799                 }
 800         }
 801         sbio = sdev->bios[sdev->curr];
 802         if (sbio->count == 0) {
 803                 sbio->physical = physical;
 804                 sbio->logical = logical;
 805         } else if (sbio->physical + sbio->count * PAGE_SIZE != physical ||
 806                    sbio->logical + sbio->count * PAGE_SIZE != logical) {
 807                 int ret;
 808
 809                 ret = scrub_submit(sdev);
 810                 if (ret)
 811                         return ret;
 812                 goto again;
 813         }
 814         sbio->spag[sbio->count].flags = flags;
 815         sbio->spag[sbio->count].generation = gen;
 816         sbio->spag[sbio->count].have_csum = 0;
 817         sbio->spag[sbio->count].mirror_num = mirror_num;
 818         if (csum) {
 819                 sbio->spag[sbio->count].have_csum = 1;
 820                 memcpy(sbio->spag[sbio->count].csum, csum, sdev->csum_size);
 821         }
 822         ++sbio->count;
 823         if (sbio->count == SCRUB_PAGES_PER_BIO || force) {
 824                 int ret;
 825
 826                 ret = scrub_submit(sdev);
 827                 if (ret)
 828                         return ret;
 829         }
 830
 831         return 0;
 832 }
 833
 834 static int scrub_find_csum(struct scrub_dev *sdev, u64 logical, u64 len,
 835                            u8 *csum)
 836 {
 837         struct btrfs_ordered_sum *sum = NULL;
 838         int ret = 0;
 839         unsigned long i;
 840         unsigned long num_sectors;
 841         u32 sectorsize = sdev->dev->dev_root->sectorsize;
 842
 843         while (!list_empty(&sdev->csum_list)) {
 844                 sum = list_first_entry(&sdev->csum_list,
 845                                        struct btrfs_ordered_sum, list);
 846                 if (sum->bytenr > logical)
 847                         return 0;
 848                 if (sum->bytenr + sum->len > logical)
 849                         break;
 850
 851                 ++sdev->stat.csum_discards;
 852                 list_del(&sum->list);
 853                 kfree(sum);
 854                 sum = NULL;
 855         }
 856         if (!sum)
 857                 return 0;
 858
 859         num_sectors = sum->len / sectorsize;
 860         for (i = 0; i < num_sectors; ++i) {
 861                 if (sum->sums[i].bytenr == logical) {
 862                         memcpy(csum, &sum->sums[i].sum, sdev->csum_size);
 863                         ret = 1;
 864                         break;
 865                 }
 866         }
 867         if (ret && i == num_sectors - 1) {
 868                 list_del(&sum->list);
 869                 kfree(sum);
 870         }
 871         return ret;
 872 }
 873
 874 /* scrub extent tries to collect up to 64 kB for each bio */
 875 static int scrub_extent(struct scrub_dev *sdev, u64 logical, u64 len,
 876                         u64 physical, u64 flags, u64 gen, u64 mirror_num)
 877 {
 878         int ret;
 879         u8 csum[BTRFS_CSUM_SIZE];
 880
 881         while (len) {
 882                 u64 l = min_t(u64, len, PAGE_SIZE);
 883                 int have_csum = 0;
 884
 885                 if (flags & BTRFS_EXTENT_FLAG_DATA) {
 886                         /* push csums to sbio */
 887                         have_csum = scrub_find_csum(sdev, logical, l, csum);
 888                         if (have_csum == 0)
 889                                 ++sdev->stat.no_csum;
 890                 }
 891                 ret = scrub_page(sdev, logical, l, physical, flags, gen,
 892                                  mirror_num, have_csum ? csum : NULL, 0);
 893                 if (ret)
 894                         return ret;
 895                 len -= l;
 896                 logical += l;
 897                 physical += l;
 898         }
 899         return 0;
 900 }
 901
 902 static noinline_for_stack int scrub_stripe(struct scrub_dev *sdev,
 903         struct map_lookup *map, int num, u64 base, u64 length)
 904 {
 905         struct btrfs_path *path;
 906         struct btrfs_fs_info *fs_info = sdev->dev->dev_root->fs_info;
 907         struct btrfs_root *root = fs_info->extent_root;
 908         struct btrfs_root *csum_root = fs_info->csum_root;
 909         struct btrfs_extent_item *extent;
 910         struct blk_plug plug;
 911         u64 flags;
 912         int ret;
 913         int slot;
 914         int i;
 915         u64 nstripes;
 916         int start_stripe;
 917         struct extent_buffer *l;
 918         struct btrfs_key key;
 919         u64 physical;
 920         u64 logical;
 921         u64 generation;
 922         u64 mirror_num;
 923
 924         u64 increment = map->stripe_len;
 925         u64 offset;
 926
 927         nstripes = length;
 928         offset = 0;
 929         do_div(nstripes, map->stripe_len);
 930         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
 931                 offset = map->stripe_len * num;
 932                 increment = map->stripe_len * map->num_stripes;
 933                 mirror_num = 0;
 934         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
 935                 int factor = map->num_stripes / map->sub_stripes;
 936                 offset = map->stripe_len * (num / map->sub_stripes);
 937                 increment = map->stripe_len * factor;
 938                 mirror_num = num % map->sub_stripes;
 939         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
 940                 increment = map->stripe_len;
 941                 mirror_num = num % map->num_stripes;
 942         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
 943                 increment = map->stripe_len;
 944                 mirror_num = num % map->num_stripes;
 945         } else {
 946                 increment = map->stripe_len;
 947                 mirror_num = 0;
 948         }
 949
 950         path = btrfs_alloc_path();
 951         if (!path)
 952                 return -ENOMEM;
 953
 954         path->reada = 2;
 955         path->search_commit_root = 1;
 956         path->skip_locking = 1;
 957
 958         /*
 959          * find all extents for each stripe and just read them to get
 960          * them into the page cache
 961          * FIXME: we can do better. build a more intelligent prefetching
 962          */
 963         logical = base + offset;
 964         physical = map->stripes[num].physical;
 965         ret = 0;
 966         for (i = 0; i < nstripes; ++i) {
 967                 key.objectid = logical;
 968                 key.type = BTRFS_EXTENT_ITEM_KEY;
 969                 key.offset = (u64)0;
 970
 971                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 972                 if (ret < 0)
 973                         goto out_noplug;
 974
 975                 /*
 976                  * we might miss half an extent here, but that doesn't matter,
 977                  * as it's only the prefetch
 978                  */
 979                 while (1) {
 980                         l = path->nodes[0];
 981                         slot = path->slots[0];
 982                         if (slot >= btrfs_header_nritems(l)) {
 983                                 ret = btrfs_next_leaf(root, path);
 984                                 if (ret == 0)
 985                                         continue;
 986                                 if (ret < 0)
 987                                         goto out_noplug;
 988
 989                                 break;
 990                         }
 991                         btrfs_item_key_to_cpu(l, &key, slot);
 992
 993                         if (key.objectid >= logical + map->stripe_len)
 994                                 break;
 995
 996                         path->slots[0]++;
 997                 }
 998                 btrfs_release_path(path);
 999                 logical += increment;
1000                 physical += map->stripe_len;
1001                 cond_resched();
1002         }
1003
1004         /*
1005          * collect all data csums for the stripe to avoid seeking during
1006          * the scrub. This might currently (crc32) end up to be about 1MB
1007          */
1008         start_stripe = 0;
1009         blk_start_plug(&plug);
1010 again:
1011         logical = base + offset + start_stripe * increment;
1012         for (i = start_stripe; i < nstripes; ++i) {
1013                 ret = btrfs_lookup_csums_range(csum_root, logical,
1014                                                logical + map->stripe_len - 1,
1015                                                &sdev->csum_list, 1);
1016                 if (ret)
1017                         goto out;
1018
1019                 logical += increment;
1020                 cond_resched();
1021         }
1022         /*
1023          * now find all extents for each stripe and scrub them
1024          */
1025         logical = base + offset + start_stripe * increment;
1026         physical = map->stripes[num].physical + start_stripe * map->stripe_len;
1027         ret = 0;
1028         for (i = start_stripe; i < nstripes; ++i) {
1029                 /*
1030                  * canceled?
1031                  */
1032                 if (atomic_read(&fs_info->scrub_cancel_req) ||
1033                     atomic_read(&sdev->cancel_req)) {
1034                         ret = -ECANCELED;
1035                         goto out;
1036                 }
1037                 /*
1038                  * check to see if we have to pause
1039                  */
1040                 if (atomic_read(&fs_info->scrub_pause_req)) {
1041                         /* push queued extents */
1042                         scrub_submit(sdev);
1043                         wait_event(sdev->list_wait,
1044                                    atomic_read(&sdev->in_flight) == 0);
1045                         atomic_inc(&fs_info->scrubs_paused);
1046                         wake_up(&fs_info->scrub_pause_wait);
1047                         mutex_lock(&fs_info->scrub_lock);
1048                         while (atomic_read(&fs_info->scrub_pause_req)) {
1049                                 mutex_unlock(&fs_info->scrub_lock);
1050                                 wait_event(fs_info->scrub_pause_wait,
1051                                    atomic_read(&fs_info->scrub_pause_req) == 0);
1052                                 mutex_lock(&fs_info->scrub_lock);
1053                         }
1054                         atomic_dec(&fs_info->scrubs_paused);
1055                         mutex_unlock(&fs_info->scrub_lock);
1056                         wake_up(&fs_info->scrub_pause_wait);
1057                         scrub_free_csums(sdev);
1058                         start_stripe = i;
1059                         goto again;
1060                 }
1061
1062                 key.objectid = logical;
1063                 key.type = BTRFS_EXTENT_ITEM_KEY;
1064                 key.offset = (u64)0;
1065
1066                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1067                 if (ret < 0)
1068                         goto out;
1069                 if (ret > 0) {
1070                         ret = btrfs_previous_item(root, path, 0,
1071                                                   BTRFS_EXTENT_ITEM_KEY);
1072                         if (ret < 0)
1073                                 goto out;
1074                         if (ret > 0) {
1075                                 /* there's no smaller item, so stick with the
1076                                  * larger one */
1077                                 btrfs_release_path(path);
1078                                 ret = btrfs_search_slot(NULL, root, &key,
1079                                                         path, 0, 0);
1080                                 if (ret < 0)
1081                                         goto out;
1082                         }
1083                 }
1084
1085                 while (1) {
1086                         l = path->nodes[0];
1087                         slot = path->slots[0];
1088                         if (slot >= btrfs_header_nritems(l)) {
1089                                 ret = btrfs_next_leaf(root, path);
1090                                 if (ret == 0)
1091                                         continue;
1092                                 if (ret < 0)
1093                                         goto out;
1094
1095                                 break;
1096                         }
1097                         btrfs_item_key_to_cpu(l, &key, slot);
1098
1099                         if (key.objectid + key.offset <= logical)
1100                                 goto next;
1101
1102                         if (key.objectid >= logical + map->stripe_len)
1103                                 break;
1104
1105                         if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY)
1106                                 goto next;
1107
1108                         extent = btrfs_item_ptr(l, slot,
1109                                                 struct btrfs_extent_item);
1110                         flags = btrfs_extent_flags(l, extent);
1111                         generation = btrfs_extent_generation(l, extent);
1112
1113                         if (key.objectid < logical &&
1114                             (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
1115                                 printk(KERN_ERR
1116                                        "btrfs scrub: tree block %llu spanning "
1117                                        "stripes, ignored. logical=%llu\n",
1118                                        (unsigned long long)key.objectid,
1119                                        (unsigned long long)logical);
1120                                 goto next;
1121                         }
1122
1123                         /*
1124                          * trim extent to this stripe
1125                          */
1126                         if (key.objectid < logical) {
1127                                 key.offset -= logical - key.objectid;
1128                                 key.objectid = logical;
1129                         }
1130                         if (key.objectid + key.offset >
1131                             logical + map->stripe_len) {
1132                                 key.offset = logical + map->stripe_len -
1133                                              key.objectid;
1134                         }
1135
1136                         ret = scrub_extent(sdev, key.objectid, key.offset,
1137                                            key.objectid - logical + physical,
1138                                            flags, generation, mirror_num);
1139                         if (ret)
1140                                 goto out;
1141
1142 next:
1143                         path->slots[0]++;
1144                 }
1145                 btrfs_release_path(path);
1146                 logical += increment;
1147                 physical += map->stripe_len;
1148                 spin_lock(&sdev->stat_lock);
1149                 sdev->stat.last_physical = physical;
1150                 spin_unlock(&sdev->stat_lock);
1151         }
1152         /* push queued extents */
1153         scrub_submit(sdev);
1154
1155 out:
1156         blk_finish_plug(&plug);
1157 out_noplug:
1158         btrfs_free_path(path);
1159         return ret < 0 ? ret : 0;
1160 }
1161
1162 static noinline_for_stack int scrub_chunk(struct scrub_dev *sdev,
1163         u64 chunk_tree, u64 chunk_objectid, u64 chunk_offset, u64 length)
1164 {
1165         struct btrfs_mapping_tree *map_tree =
1166                 &sdev->dev->dev_root->fs_info->mapping_tree;
1167         struct map_lookup *map;
1168         struct extent_map *em;
1169         int i;
1170         int ret = -EINVAL;
1171
1172         read_lock(&map_tree->map_tree.lock);
1173         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
1174         read_unlock(&map_tree->map_tree.lock);
1175
1176         if (!em)
1177                 return -EINVAL;
1178
1179         map = (struct map_lookup *)em->bdev;
1180         if (em->start != chunk_offset)
1181                 goto out;
1182
1183         if (em->len < length)
1184                 goto out;
1185
1186         for (i = 0; i < map->num_stripes; ++i) {
1187                 if (map->stripes[i].dev == sdev->dev) {
1188                         ret = scrub_stripe(sdev, map, i, chunk_offset, length);
1189                         if (ret)
1190                                 goto out;
1191                 }
1192         }
1193 out:
1194         free_extent_map(em);
1195
1196         return ret;
1197 }
1198
1199 static noinline_for_stack
1200 int scrub_enumerate_chunks(struct scrub_dev *sdev, u64 start, u64 end)
1201 {
1202         struct btrfs_dev_extent *dev_extent = NULL;
1203         struct btrfs_path *path;
1204         struct btrfs_root *root = sdev->dev->dev_root;
1205         struct btrfs_fs_info *fs_info = root->fs_info;
1206         u64 length;
1207         u64 chunk_tree;
1208         u64 chunk_objectid;
1209         u64 chunk_offset;
1210         int ret;
1211         int slot;
1212         struct extent_buffer *l;
1213         struct btrfs_key key;
1214         struct btrfs_key found_key;
1215         struct btrfs_block_group_cache *cache;
1216
1217         path = btrfs_alloc_path();
1218         if (!path)
1219                 return -ENOMEM;
1220
1221         path->reada = 2;
1222         path->search_commit_root = 1;
1223         path->skip_locking = 1;
1224
1225         key.objectid = sdev->dev->devid;
1226         key.offset = 0ull;
1227         key.type = BTRFS_DEV_EXTENT_KEY;
1228
1229
1230         while (1) {
1231                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1232                 if (ret < 0)
1233                         break;
1234                 if (ret > 0) {
1235                         if (path->slots[0] >=
1236                             btrfs_header_nritems(path->nodes[0])) {
1237                                 ret = btrfs_next_leaf(root, path);
1238                                 if (ret)
1239                                         break;
1240                         }
1241                 }
1242
1243                 l = path->nodes[0];
1244                 slot = path->slots[0];
1245
1246                 btrfs_item_key_to_cpu(l, &found_key, slot);
1247
1248                 if (found_key.objectid != sdev->dev->devid)
1249                         break;
1250
1251                 if (btrfs_key_type(&found_key) != BTRFS_DEV_EXTENT_KEY)
1252                         break;
1253
1254                 if (found_key.offset >= end)
1255                         break;
1256
1257                 if (found_key.offset < key.offset)
1258                         break;
1259
1260                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1261                 length = btrfs_dev_extent_length(l, dev_extent);
1262
1263                 if (found_key.offset + length <= start) {
1264                         key.offset = found_key.offset + length;
1265                         btrfs_release_path(path);
1266                         continue;
1267                 }
1268
1269                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1270                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1271                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1272
1273                 /*
1274                  * get a reference on the corresponding block group to prevent
1275                  * the chunk from going away while we scrub it
1276                  */
1277                 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
1278                 if (!cache) {
1279                         ret = -ENOENT;
1280                         break;
1281                 }
1282                 ret = scrub_chunk(sdev, chunk_tree, chunk_objectid,
1283                                   chunk_offset, length);
1284                 btrfs_put_block_group(cache);
1285                 if (ret)
1286                         break;
1287
1288                 key.offset = found_key.offset + length;
1289                 btrfs_release_path(path);
1290         }
1291
1292         btrfs_free_path(path);
1293
1294         /*
1295          * ret can still be 1 from search_slot or next_leaf,
1296          * that's not an error
1297          */
1298         return ret < 0 ? ret : 0;
1299 }
1300
1301 static noinline_for_stack int scrub_supers(struct scrub_dev *sdev)
1302 {
1303         int     i;
1304         u64     bytenr;
1305         u64     gen;
1306         int     ret;
1307         struct btrfs_device *device = sdev->dev;
1308         struct btrfs_root *root = device->dev_root;
1309
1310         gen = root->fs_info->last_trans_committed;
1311
1312         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1313                 bytenr = btrfs_sb_offset(i);
1314                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
1315                         break;
1316
1317                 ret = scrub_page(sdev, bytenr, PAGE_SIZE, bytenr,
1318                                  BTRFS_EXTENT_FLAG_SUPER, gen, i, NULL, 1);
1319                 if (ret)
1320                         return ret;
1321         }
1322         wait_event(sdev->list_wait, atomic_read(&sdev->in_flight) == 0);
1323
1324         return 0;
1325 }
1326
1327 /*
1328  * get a reference count on fs_info->scrub_workers. start worker if necessary
1329  */
1330 static noinline_for_stack int scrub_workers_get(struct btrfs_root *root)
1331 {
1332         struct btrfs_fs_info *fs_info = root->fs_info;
1333
1334         mutex_lock(&fs_info->scrub_lock);
1335         if (fs_info->scrub_workers_refcnt == 0) {
1336                 btrfs_init_workers(&fs_info->scrub_workers, "scrub",
1337                            fs_info->thread_pool_size, &fs_info->generic_worker);
1338                 fs_info->scrub_workers.idle_thresh = 4;
1339                 btrfs_start_workers(&fs_info->scrub_workers, 1);
1340         }
1341         ++fs_info->scrub_workers_refcnt;
1342         mutex_unlock(&fs_info->scrub_lock);
1343
1344         return 0;
1345 }
1346
1347 static noinline_for_stack void scrub_workers_put(struct btrfs_root *root)
1348 {
1349         struct btrfs_fs_info *fs_info = root->fs_info;
1350
1351         mutex_lock(&fs_info->scrub_lock);
1352         if (--fs_info->scrub_workers_refcnt == 0)
1353                 btrfs_stop_workers(&fs_info->scrub_workers);
1354         WARN_ON(fs_info->scrub_workers_refcnt < 0);
1355         mutex_unlock(&fs_info->scrub_lock);
1356 }
1357
1358
1359 int btrfs_scrub_dev(struct btrfs_root *root, u64 devid, u64 start, u64 end,
1360                     struct btrfs_scrub_progress *progress, int readonly)
1361 {
1362         struct scrub_dev *sdev;
1363         struct btrfs_fs_info *fs_info = root->fs_info;
1364         int ret;
1365         struct btrfs_device *dev;
1366
1367         if (btrfs_fs_closing(root->fs_info))
1368                 return -EINVAL;
1369
1370         /*
1371          * check some assumptions
1372          */
1373         if (root->sectorsize != PAGE_SIZE ||
1374             root->sectorsize != root->leafsize ||
1375             root->sectorsize != root->nodesize) {
1376                 printk(KERN_ERR "btrfs_scrub: size assumptions fail\n");
1377                 return -EINVAL;
1378         }
1379
1380         ret = scrub_workers_get(root);
1381         if (ret)
1382                 return ret;
1383
1384         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1385         dev = btrfs_find_device(root, devid, NULL, NULL);
1386         if (!dev || dev->missing) {
1387                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1388                 scrub_workers_put(root);
1389                 return -ENODEV;
1390         }
1391         mutex_lock(&fs_info->scrub_lock);
1392
1393         if (!dev->in_fs_metadata) {
1394                 mutex_unlock(&fs_info->scrub_lock);
1395                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1396                 scrub_workers_put(root);
1397                 return -ENODEV;
1398         }
1399
1400         if (dev->scrub_device) {
1401                 mutex_unlock(&fs_info->scrub_lock);
1402                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1403                 scrub_workers_put(root);
1404                 return -EINPROGRESS;
1405         }
1406         sdev = scrub_setup_dev(dev);
1407         if (IS_ERR(sdev)) {
1408                 mutex_unlock(&fs_info->scrub_lock);
1409                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1410                 scrub_workers_put(root);
1411                 return PTR_ERR(sdev);
1412         }
1413         sdev->readonly = readonly;
1414         dev->scrub_device = sdev;
1415
1416         atomic_inc(&fs_info->scrubs_running);
1417         mutex_unlock(&fs_info->scrub_lock);
1418         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1419
1420         down_read(&fs_info->scrub_super_lock);
1421         ret = scrub_supers(sdev);
1422         up_read(&fs_info->scrub_super_lock);
1423
1424         if (!ret)
1425                 ret = scrub_enumerate_chunks(sdev, start, end);
1426
1427         wait_event(sdev->list_wait, atomic_read(&sdev->in_flight) == 0);
1428
1429         atomic_dec(&fs_info->scrubs_running);
1430         wake_up(&fs_info->scrub_pause_wait);
1431
1432         if (progress)
1433                 memcpy(progress, &sdev->stat, sizeof(*progress));
1434
1435         mutex_lock(&fs_info->scrub_lock);
1436         dev->scrub_device = NULL;
1437         mutex_unlock(&fs_info->scrub_lock);
1438
1439         scrub_free_dev(sdev);
1440         scrub_workers_put(root);
1441
1442         return ret;
1443 }
1444
1445 int btrfs_scrub_pause(struct btrfs_root *root)
1446 {
1447         struct btrfs_fs_info *fs_info = root->fs_info;
1448
1449         mutex_lock(&fs_info->scrub_lock);
1450         atomic_inc(&fs_info->scrub_pause_req);
1451         while (atomic_read(&fs_info->scrubs_paused) !=
1452                atomic_read(&fs_info->scrubs_running)) {
1453                 mutex_unlock(&fs_info->scrub_lock);
1454                 wait_event(fs_info->scrub_pause_wait,
1455                            atomic_read(&fs_info->scrubs_paused) ==
1456                            atomic_read(&fs_info->scrubs_running));
1457                 mutex_lock(&fs_info->scrub_lock);
1458         }
1459         mutex_unlock(&fs_info->scrub_lock);
1460
1461         return 0;
1462 }
1463
1464 int btrfs_scrub_continue(struct btrfs_root *root)
1465 {
1466         struct btrfs_fs_info *fs_info = root->fs_info;
1467
1468         atomic_dec(&fs_info->scrub_pause_req);
1469         wake_up(&fs_info->scrub_pause_wait);
1470         return 0;
1471 }
1472
1473 int btrfs_scrub_pause_super(struct btrfs_root *root)
1474 {
1475         down_write(&root->fs_info->scrub_super_lock);
1476         return 0;
1477 }
1478
1479 int btrfs_scrub_continue_super(struct btrfs_root *root)
1480 {
1481         up_write(&root->fs_info->scrub_super_lock);
1482         return 0;
1483 }
1484
1485 int btrfs_scrub_cancel(struct btrfs_root *root)
1486 {
1487         struct btrfs_fs_info *fs_info = root->fs_info;
1488
1489         mutex_lock(&fs_info->scrub_lock);
1490         if (!atomic_read(&fs_info->scrubs_running)) {
1491                 mutex_unlock(&fs_info->scrub_lock);
1492                 return -ENOTCONN;
1493         }
1494
1495         atomic_inc(&fs_info->scrub_cancel_req);
1496         while (atomic_read(&fs_info->scrubs_running)) {
1497                 mutex_unlock(&fs_info->scrub_lock);
1498                 wait_event(fs_info->scrub_pause_wait,
1499                            atomic_read(&fs_info->scrubs_running) == 0);
1500                 mutex_lock(&fs_info->scrub_lock);
1501         }
1502         atomic_dec(&fs_info->scrub_cancel_req);
1503         mutex_unlock(&fs_info->scrub_lock);
1504
1505         return 0;
1506 }
1507
1508 int btrfs_scrub_cancel_dev(struct btrfs_root *root, struct btrfs_device *dev)
1509 {
1510         struct btrfs_fs_info *fs_info = root->fs_info;
1511         struct scrub_dev *sdev;
1512
1513         mutex_lock(&fs_info->scrub_lock);
1514         sdev = dev->scrub_device;
1515         if (!sdev) {
1516                 mutex_unlock(&fs_info->scrub_lock);
1517                 return -ENOTCONN;
1518         }
1519         atomic_inc(&sdev->cancel_req);
1520         while (dev->scrub_device) {
1521                 mutex_unlock(&fs_info->scrub_lock);
1522                 wait_event(fs_info->scrub_pause_wait,
1523                            dev->scrub_device == NULL);
1524                 mutex_lock(&fs_info->scrub_lock);
1525         }
1526         mutex_unlock(&fs_info->scrub_lock);
1527
1528         return 0;
1529 }
1530 int btrfs_scrub_cancel_devid(struct btrfs_root *root, u64 devid)
1531 {
1532         struct btrfs_fs_info *fs_info = root->fs_info;
1533         struct btrfs_device *dev;
1534         int ret;
1535
1536         /*
1537          * we have to hold the device_list_mutex here so the device
1538          * does not go away in cancel_dev. FIXME: find a better solution
1539          */
1540         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1541         dev = btrfs_find_device(root, devid, NULL, NULL);
1542         if (!dev) {
1543                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1544                 return -ENODEV;
1545         }
1546         ret = btrfs_scrub_cancel_dev(root, dev);
1547         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1548
1549         return ret;
1550 }
1551
1552 int btrfs_scrub_progress(struct btrfs_root *root, u64 devid,
1553                          struct btrfs_scrub_progress *progress)
1554 {
1555         struct btrfs_device *dev;
1556         struct scrub_dev *sdev = NULL;
1557
1558         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1559         dev = btrfs_find_device(root, devid, NULL, NULL);
1560         if (dev)
1561                 sdev = dev->scrub_device;
1562         if (sdev)
1563                 memcpy(progress, &sdev->stat, sizeof(*progress));
1564         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1565
1566         return dev ? (sdev ? 0 : -ENOTCONN) : -ENODEV;
1567 }