diff options
author | Linus Torvalds <torvalds@linux-foundation.org> | 2012-12-18 17:42:05 (GMT) |
---|---|---|
committer | Linus Torvalds <torvalds@linux-foundation.org> | 2012-12-18 17:42:05 (GMT) |
commit | a22180d2666c018f4fef6818074d78bb76ff2bda (patch) | |
tree | a633aaf423ff39f94d00502d03dbbd99dab4b2ee /fs/btrfs/scrub.c | |
parent | 2d4dce0070448bcb5ccd04553a4be4635417f565 (diff) | |
parent | 213490b301773ea9c6fb89a86424a6901fcdd069 (diff) | |
download | linux-fsl-qoriq-a22180d2666c018f4fef6818074d78bb76ff2bda.tar.xz |
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/linux-btrfs
Pull btrfs update from Chris Mason:
"A big set of fixes and features.
In terms of line count, most of the code comes from Stefan, who added
the ability to replace a single drive in place. This is different
from how btrfs normally replaces drives, and is much much much faster.
Josef is plowing through our synchronous write performance. This pull
request does not include the DIO_OWN_WAITING patch that was discussed
on the list, but it has a number of other improvements to cut down our
latencies and CPU time during fsync/O_DIRECT writes.
Miao Xie has a big series of fixes and is spreading out ordered
operations over more CPUs. This improves performance and reduces
contention.
I've put in fixes for error handling around hash collisions. These
are going back to individual stable kernels as I test against them.
Otherwise we have a lot of fixes and cleanups, thanks everyone!
raid5/6 is being rebased against the device replacement code. I'll
have it posted this Friday along with a nice series of benchmarks."
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/linux-btrfs: (115 commits)
Btrfs: fix a bug of per-file nocow
Btrfs: fix hash overflow handling
Btrfs: don't take inode delalloc mutex if we're a free space inode
Btrfs: fix autodefrag and umount lockup
Btrfs: fix permissions of empty files not affected by umask
Btrfs: put raid properties into global table
Btrfs: fix BUG() in scrub when first superblock reading gives EIO
Btrfs: do not call file_update_time in aio_write
Btrfs: only unlock and relock if we have to
Btrfs: use tokens where we can in the tree log
Btrfs: optimize leaf_space_used
Btrfs: don't memset new tokens
Btrfs: only clear dirty on the buffer if it is marked as dirty
Btrfs: move checks in set_page_dirty under DEBUG
Btrfs: log changed inodes based on the extent map tree
Btrfs: add path->really_keep_locks
Btrfs: do not mark ems as prealloc if we are writing to them
Btrfs: keep track of the extents original block length
Btrfs: inline csums if we're fsyncing
Btrfs: don't bother copying if we're only logging the inode
...
Diffstat (limited to 'fs/btrfs/scrub.c')
-rw-r--r-- | fs/btrfs/scrub.c | 1836 |
1 files changed, 1333 insertions, 503 deletions
diff --git a/fs/btrfs/scrub.c b/fs/btrfs/scrub.c index 27892f6..bdbb94f 100644 --- a/fs/btrfs/scrub.c +++ b/fs/btrfs/scrub.c @@ -1,5 +1,5 @@ /* - * Copyright (C) 2011 STRATO. All rights reserved. + * Copyright (C) 2011, 2012 STRATO. All rights reserved. * * This program is free software; you can redistribute it and/or * modify it under the terms of the GNU General Public @@ -25,6 +25,7 @@ #include "transaction.h" #include "backref.h" #include "extent_io.h" +#include "dev-replace.h" #include "check-integrity.h" #include "rcu-string.h" @@ -42,10 +43,23 @@ */ struct scrub_block; -struct scrub_dev; +struct scrub_ctx; -#define SCRUB_PAGES_PER_BIO 16 /* 64k per bio */ -#define SCRUB_BIOS_PER_DEV 16 /* 1 MB per device in flight */ +/* + * the following three values only influence the performance. + * The last one configures the number of parallel and outstanding I/O + * operations. The first two values configure an upper limit for the number + * of (dynamically allocated) pages that are added to a bio. + */ +#define SCRUB_PAGES_PER_RD_BIO 32 /* 128k per bio */ +#define SCRUB_PAGES_PER_WR_BIO 32 /* 128k per bio */ +#define SCRUB_BIOS_PER_SCTX 64 /* 8MB per device in flight */ + +/* + * the following value times PAGE_SIZE needs to be large enough to match the + * largest node/leaf/sector size that shall be supported. + * Values larger than BTRFS_STRIPE_LEN are not supported. + */ #define SCRUB_MAX_PAGES_PER_BLOCK 16 /* 64k per node/leaf/sector */ struct scrub_page { @@ -56,6 +70,8 @@ struct scrub_page { u64 generation; u64 logical; u64 physical; + u64 physical_for_dev_replace; + atomic_t ref_count; struct { unsigned int mirror_num:8; unsigned int have_csum:1; @@ -66,23 +82,28 @@ struct scrub_page { struct scrub_bio { int index; - struct scrub_dev *sdev; + struct scrub_ctx *sctx; + struct btrfs_device *dev; struct bio *bio; int err; u64 logical; u64 physical; - struct scrub_page *pagev[SCRUB_PAGES_PER_BIO]; +#if SCRUB_PAGES_PER_WR_BIO >= SCRUB_PAGES_PER_RD_BIO + struct scrub_page *pagev[SCRUB_PAGES_PER_WR_BIO]; +#else + struct scrub_page *pagev[SCRUB_PAGES_PER_RD_BIO]; +#endif int page_count; int next_free; struct btrfs_work work; }; struct scrub_block { - struct scrub_page pagev[SCRUB_MAX_PAGES_PER_BLOCK]; + struct scrub_page *pagev[SCRUB_MAX_PAGES_PER_BLOCK]; int page_count; atomic_t outstanding_pages; atomic_t ref_count; /* free mem on transition to zero */ - struct scrub_dev *sdev; + struct scrub_ctx *sctx; struct { unsigned int header_error:1; unsigned int checksum_error:1; @@ -91,23 +112,35 @@ struct scrub_block { }; }; -struct scrub_dev { - struct scrub_bio *bios[SCRUB_BIOS_PER_DEV]; - struct btrfs_device *dev; +struct scrub_wr_ctx { + struct scrub_bio *wr_curr_bio; + struct btrfs_device *tgtdev; + int pages_per_wr_bio; /* <= SCRUB_PAGES_PER_WR_BIO */ + atomic_t flush_all_writes; + struct mutex wr_lock; +}; + +struct scrub_ctx { + struct scrub_bio *bios[SCRUB_BIOS_PER_SCTX]; + struct btrfs_root *dev_root; int first_free; int curr; - atomic_t in_flight; - atomic_t fixup_cnt; + atomic_t bios_in_flight; + atomic_t workers_pending; spinlock_t list_lock; wait_queue_head_t list_wait; u16 csum_size; struct list_head csum_list; atomic_t cancel_req; int readonly; - int pages_per_bio; /* <= SCRUB_PAGES_PER_BIO */ + int pages_per_rd_bio; u32 sectorsize; u32 nodesize; u32 leafsize; + + int is_dev_replace; + struct scrub_wr_ctx wr_ctx; + /* * statistics */ @@ -116,13 +149,23 @@ struct scrub_dev { }; struct scrub_fixup_nodatasum { - struct scrub_dev *sdev; + struct scrub_ctx *sctx; + struct btrfs_device *dev; u64 logical; struct btrfs_root *root; struct btrfs_work work; int mirror_num; }; +struct scrub_copy_nocow_ctx { + struct scrub_ctx *sctx; + u64 logical; + u64 len; + int mirror_num; + u64 physical_for_dev_replace; + struct btrfs_work work; +}; + struct scrub_warning { struct btrfs_path *path; u64 extent_item_size; @@ -137,15 +180,20 @@ struct scrub_warning { }; +static void scrub_pending_bio_inc(struct scrub_ctx *sctx); +static void scrub_pending_bio_dec(struct scrub_ctx *sctx); +static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx); +static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx); static int scrub_handle_errored_block(struct scrub_block *sblock_to_check); -static int scrub_setup_recheck_block(struct scrub_dev *sdev, - struct btrfs_mapping_tree *map_tree, +static int scrub_setup_recheck_block(struct scrub_ctx *sctx, + struct btrfs_fs_info *fs_info, + struct scrub_block *original_sblock, u64 length, u64 logical, - struct scrub_block *sblock); -static int scrub_recheck_block(struct btrfs_fs_info *fs_info, - struct scrub_block *sblock, int is_metadata, - int have_csum, u8 *csum, u64 generation, - u16 csum_size); + struct scrub_block *sblocks_for_recheck); +static void scrub_recheck_block(struct btrfs_fs_info *fs_info, + struct scrub_block *sblock, int is_metadata, + int have_csum, u8 *csum, u64 generation, + u16 csum_size); static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info, struct scrub_block *sblock, int is_metadata, int have_csum, @@ -158,118 +206,221 @@ static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad, static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad, struct scrub_block *sblock_good, int page_num, int force_write); +static void scrub_write_block_to_dev_replace(struct scrub_block *sblock); +static int scrub_write_page_to_dev_replace(struct scrub_block *sblock, + int page_num); static int scrub_checksum_data(struct scrub_block *sblock); static int scrub_checksum_tree_block(struct scrub_block *sblock); static int scrub_checksum_super(struct scrub_block *sblock); static void scrub_block_get(struct scrub_block *sblock); static void scrub_block_put(struct scrub_block *sblock); -static int scrub_add_page_to_bio(struct scrub_dev *sdev, - struct scrub_page *spage); -static int scrub_pages(struct scrub_dev *sdev, u64 logical, u64 len, - u64 physical, u64 flags, u64 gen, int mirror_num, - u8 *csum, int force); +static void scrub_page_get(struct scrub_page *spage); +static void scrub_page_put(struct scrub_page *spage); +static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx, + struct scrub_page *spage); +static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len, + u64 physical, struct btrfs_device *dev, u64 flags, + u64 gen, int mirror_num, u8 *csum, int force, + u64 physical_for_dev_replace); static void scrub_bio_end_io(struct bio *bio, int err); static void scrub_bio_end_io_worker(struct btrfs_work *work); static void scrub_block_complete(struct scrub_block *sblock); +static void scrub_remap_extent(struct btrfs_fs_info *fs_info, + u64 extent_logical, u64 extent_len, + u64 *extent_physical, + struct btrfs_device **extent_dev, + int *extent_mirror_num); +static int scrub_setup_wr_ctx(struct scrub_ctx *sctx, + struct scrub_wr_ctx *wr_ctx, + struct btrfs_fs_info *fs_info, + struct btrfs_device *dev, + int is_dev_replace); +static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx); +static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx, + struct scrub_page *spage); +static void scrub_wr_submit(struct scrub_ctx *sctx); +static void scrub_wr_bio_end_io(struct bio *bio, int err); +static void scrub_wr_bio_end_io_worker(struct btrfs_work *work); +static int write_page_nocow(struct scrub_ctx *sctx, + u64 physical_for_dev_replace, struct page *page); +static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root, + void *ctx); +static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len, + int mirror_num, u64 physical_for_dev_replace); +static void copy_nocow_pages_worker(struct btrfs_work *work); + + +static void scrub_pending_bio_inc(struct scrub_ctx *sctx) +{ + atomic_inc(&sctx->bios_in_flight); +} + +static void scrub_pending_bio_dec(struct scrub_ctx *sctx) +{ + atomic_dec(&sctx->bios_in_flight); + wake_up(&sctx->list_wait); +} + +/* + * used for workers that require transaction commits (i.e., for the + * NOCOW case) + */ +static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx) +{ + struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; + + /* + * increment scrubs_running to prevent cancel requests from + * completing as long as a worker is running. we must also + * increment scrubs_paused to prevent deadlocking on pause + * requests used for transactions commits (as the worker uses a + * transaction context). it is safe to regard the worker + * as paused for all matters practical. effectively, we only + * avoid cancellation requests from completing. + */ + mutex_lock(&fs_info->scrub_lock); + atomic_inc(&fs_info->scrubs_running); + atomic_inc(&fs_info->scrubs_paused); + mutex_unlock(&fs_info->scrub_lock); + atomic_inc(&sctx->workers_pending); +} +/* used for workers that require transaction commits */ +static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx) +{ + struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; -static void scrub_free_csums(struct scrub_dev *sdev) + /* + * see scrub_pending_trans_workers_inc() why we're pretending + * to be paused in the scrub counters + */ + mutex_lock(&fs_info->scrub_lock); + atomic_dec(&fs_info->scrubs_running); + atomic_dec(&fs_info->scrubs_paused); + mutex_unlock(&fs_info->scrub_lock); + atomic_dec(&sctx->workers_pending); + wake_up(&fs_info->scrub_pause_wait); + wake_up(&sctx->list_wait); +} + +static void scrub_free_csums(struct scrub_ctx *sctx) { - while (!list_empty(&sdev->csum_list)) { + while (!list_empty(&sctx->csum_list)) { struct btrfs_ordered_sum *sum; - sum = list_first_entry(&sdev->csum_list, + sum = list_first_entry(&sctx->csum_list, struct btrfs_ordered_sum, list); list_del(&sum->list); kfree(sum); } } -static noinline_for_stack void scrub_free_dev(struct scrub_dev *sdev) +static noinline_for_stack void scrub_free_ctx(struct scrub_ctx *sctx) { int i; - if (!sdev) + if (!sctx) return; + scrub_free_wr_ctx(&sctx->wr_ctx); + /* this can happen when scrub is cancelled */ - if (sdev->curr != -1) { - struct scrub_bio *sbio = sdev->bios[sdev->curr]; + if (sctx->curr != -1) { + struct scrub_bio *sbio = sctx->bios[sctx->curr]; for (i = 0; i < sbio->page_count; i++) { - BUG_ON(!sbio->pagev[i]); - BUG_ON(!sbio->pagev[i]->page); + WARN_ON(!sbio->pagev[i]->page); scrub_block_put(sbio->pagev[i]->sblock); } bio_put(sbio->bio); } - for (i = 0; i < SCRUB_BIOS_PER_DEV; ++i) { - struct scrub_bio *sbio = sdev->bios[i]; + for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) { + struct scrub_bio *sbio = sctx->bios[i]; if (!sbio) break; kfree(sbio); } - scrub_free_csums(sdev); - kfree(sdev); + scrub_free_csums(sctx); + kfree(sctx); } static noinline_for_stack -struct scrub_dev *scrub_setup_dev(struct btrfs_device *dev) +struct scrub_ctx *scrub_setup_ctx(struct btrfs_device *dev, int is_dev_replace) { - struct scrub_dev *sdev; + struct scrub_ctx *sctx; int i; struct btrfs_fs_info *fs_info = dev->dev_root->fs_info; - int pages_per_bio; + int pages_per_rd_bio; + int ret; - pages_per_bio = min_t(int, SCRUB_PAGES_PER_BIO, - bio_get_nr_vecs(dev->bdev)); - sdev = kzalloc(sizeof(*sdev), GFP_NOFS); - if (!sdev) + /* + * the setting of pages_per_rd_bio is correct for scrub but might + * be wrong for the dev_replace code where we might read from + * different devices in the initial huge bios. However, that + * code is able to correctly handle the case when adding a page + * to a bio fails. + */ + if (dev->bdev) + pages_per_rd_bio = min_t(int, SCRUB_PAGES_PER_RD_BIO, + bio_get_nr_vecs(dev->bdev)); + else + pages_per_rd_bio = SCRUB_PAGES_PER_RD_BIO; + sctx = kzalloc(sizeof(*sctx), GFP_NOFS); + if (!sctx) goto nomem; - sdev->dev = dev; - sdev->pages_per_bio = pages_per_bio; - sdev->curr = -1; - for (i = 0; i < SCRUB_BIOS_PER_DEV; ++i) { + sctx->is_dev_replace = is_dev_replace; + sctx->pages_per_rd_bio = pages_per_rd_bio; + sctx->curr = -1; + sctx->dev_root = dev->dev_root; + for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) { struct scrub_bio *sbio; sbio = kzalloc(sizeof(*sbio), GFP_NOFS); if (!sbio) goto nomem; - sdev->bios[i] = sbio; + sctx->bios[i] = sbio; sbio->index = i; - sbio->sdev = sdev; + sbio->sctx = sctx; sbio->page_count = 0; sbio->work.func = scrub_bio_end_io_worker; - if (i != SCRUB_BIOS_PER_DEV-1) - sdev->bios[i]->next_free = i + 1; + if (i != SCRUB_BIOS_PER_SCTX - 1) + sctx->bios[i]->next_free = i + 1; else - sdev->bios[i]->next_free = -1; - } - sdev->first_free = 0; - sdev->nodesize = dev->dev_root->nodesize; - sdev->leafsize = dev->dev_root->leafsize; - sdev->sectorsize = dev->dev_root->sectorsize; - atomic_set(&sdev->in_flight, 0); - atomic_set(&sdev->fixup_cnt, 0); - atomic_set(&sdev->cancel_req, 0); - sdev->csum_size = btrfs_super_csum_size(fs_info->super_copy); - INIT_LIST_HEAD(&sdev->csum_list); - - spin_lock_init(&sdev->list_lock); - spin_lock_init(&sdev->stat_lock); - init_waitqueue_head(&sdev->list_wait); - return sdev; + sctx->bios[i]->next_free = -1; + } + sctx->first_free = 0; + sctx->nodesize = dev->dev_root->nodesize; + sctx->leafsize = dev->dev_root->leafsize; + sctx->sectorsize = dev->dev_root->sectorsize; + atomic_set(&sctx->bios_in_flight, 0); + atomic_set(&sctx->workers_pending, 0); + atomic_set(&sctx->cancel_req, 0); + sctx->csum_size = btrfs_super_csum_size(fs_info->super_copy); + INIT_LIST_HEAD(&sctx->csum_list); + + spin_lock_init(&sctx->list_lock); + spin_lock_init(&sctx->stat_lock); + init_waitqueue_head(&sctx->list_wait); + + ret = scrub_setup_wr_ctx(sctx, &sctx->wr_ctx, fs_info, + fs_info->dev_replace.tgtdev, is_dev_replace); + if (ret) { + scrub_free_ctx(sctx); + return ERR_PTR(ret); + } + return sctx; nomem: - scrub_free_dev(sdev); + scrub_free_ctx(sctx); return ERR_PTR(-ENOMEM); } -static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root, void *ctx) +static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root, + void *warn_ctx) { u64 isize; u32 nlink; @@ -277,7 +428,7 @@ static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root, void *ctx) int i; struct extent_buffer *eb; struct btrfs_inode_item *inode_item; - struct scrub_warning *swarn = ctx; + struct scrub_warning *swarn = warn_ctx; struct btrfs_fs_info *fs_info = swarn->dev->dev_root->fs_info; struct inode_fs_paths *ipath = NULL; struct btrfs_root *local_root; @@ -345,8 +496,8 @@ err: static void scrub_print_warning(const char *errstr, struct scrub_block *sblock) { - struct btrfs_device *dev = sblock->sdev->dev; - struct btrfs_fs_info *fs_info = dev->dev_root->fs_info; + struct btrfs_device *dev; + struct btrfs_fs_info *fs_info; struct btrfs_path *path; struct btrfs_key found_key; struct extent_buffer *eb; @@ -361,15 +512,18 @@ static void scrub_print_warning(const char *errstr, struct scrub_block *sblock) const int bufsize = 4096; int ret; + WARN_ON(sblock->page_count < 1); + dev = sblock->pagev[0]->dev; + fs_info = sblock->sctx->dev_root->fs_info; + path = btrfs_alloc_path(); swarn.scratch_buf = kmalloc(bufsize, GFP_NOFS); swarn.msg_buf = kmalloc(bufsize, GFP_NOFS); - BUG_ON(sblock->page_count < 1); - swarn.sector = (sblock->pagev[0].physical) >> 9; - swarn.logical = sblock->pagev[0].logical; + swarn.sector = (sblock->pagev[0]->physical) >> 9; + swarn.logical = sblock->pagev[0]->logical; swarn.errstr = errstr; - swarn.dev = dev; + swarn.dev = NULL; swarn.msg_bufsize = bufsize; swarn.scratch_bufsize = bufsize; @@ -405,6 +559,7 @@ static void scrub_print_warning(const char *errstr, struct scrub_block *sblock) } while (ret != 1); } else { swarn.path = path; + swarn.dev = dev; iterate_extent_inodes(fs_info, found_key.objectid, extent_item_pos, 1, scrub_print_warning_inode, &swarn); @@ -416,11 +571,11 @@ out: kfree(swarn.msg_buf); } -static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *ctx) +static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *fixup_ctx) { struct page *page = NULL; unsigned long index; - struct scrub_fixup_nodatasum *fixup = ctx; + struct scrub_fixup_nodatasum *fixup = fixup_ctx; int ret; int corrected = 0; struct btrfs_key key; @@ -451,7 +606,7 @@ static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *ctx) } if (PageUptodate(page)) { - struct btrfs_mapping_tree *map_tree; + struct btrfs_fs_info *fs_info; if (PageDirty(page)) { /* * we need to write the data to the defect sector. the @@ -472,8 +627,8 @@ static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *ctx) ret = -EIO; goto out; } - map_tree = &BTRFS_I(inode)->root->fs_info->mapping_tree; - ret = repair_io_failure(map_tree, offset, PAGE_SIZE, + fs_info = BTRFS_I(inode)->root->fs_info; + ret = repair_io_failure(fs_info, offset, PAGE_SIZE, fixup->logical, page, fixup->mirror_num); unlock_page(page); @@ -530,21 +685,21 @@ static void scrub_fixup_nodatasum(struct btrfs_work *work) { int ret; struct scrub_fixup_nodatasum *fixup; - struct scrub_dev *sdev; + struct scrub_ctx *sctx; struct btrfs_trans_handle *trans = NULL; struct btrfs_fs_info *fs_info; struct btrfs_path *path; int uncorrectable = 0; fixup = container_of(work, struct scrub_fixup_nodatasum, work); - sdev = fixup->sdev; + sctx = fixup->sctx; fs_info = fixup->root->fs_info; path = btrfs_alloc_path(); if (!path) { - spin_lock(&sdev->stat_lock); - ++sdev->stat.malloc_errors; - spin_unlock(&sdev->stat_lock); + spin_lock(&sctx->stat_lock); + ++sctx->stat.malloc_errors; + spin_unlock(&sctx->stat_lock); uncorrectable = 1; goto out; } @@ -573,35 +728,30 @@ static void scrub_fixup_nodatasum(struct btrfs_work *work) } WARN_ON(ret != 1); - spin_lock(&sdev->stat_lock); - ++sdev->stat.corrected_errors; - spin_unlock(&sdev->stat_lock); + spin_lock(&sctx->stat_lock); + ++sctx->stat.corrected_errors; + spin_unlock(&sctx->stat_lock); out: if (trans && !IS_ERR(trans)) btrfs_end_transaction(trans, fixup->root); if (uncorrectable) { - spin_lock(&sdev->stat_lock); - ++sdev->stat.uncorrectable_errors; - spin_unlock(&sdev->stat_lock); - + spin_lock(&sctx->stat_lock); + ++sctx->stat.uncorrectable_errors; + spin_unlock(&sctx->stat_lock); + btrfs_dev_replace_stats_inc( + &sctx->dev_root->fs_info->dev_replace. + num_uncorrectable_read_errors); printk_ratelimited_in_rcu(KERN_ERR "btrfs: unable to fixup (nodatasum) error at logical %llu on dev %s\n", (unsigned long long)fixup->logical, - rcu_str_deref(sdev->dev->name)); + rcu_str_deref(fixup->dev->name)); } btrfs_free_path(path); kfree(fixup); - /* see caller why we're pretending to be paused in the scrub counters */ - mutex_lock(&fs_info->scrub_lock); - atomic_dec(&fs_info->scrubs_running); - atomic_dec(&fs_info->scrubs_paused); - mutex_unlock(&fs_info->scrub_lock); - atomic_dec(&sdev->fixup_cnt); - wake_up(&fs_info->scrub_pause_wait); - wake_up(&sdev->list_wait); + scrub_pending_trans_workers_dec(sctx); } /* @@ -614,7 +764,8 @@ out: */ static int scrub_handle_errored_block(struct scrub_block *sblock_to_check) { - struct scrub_dev *sdev = sblock_to_check->sdev; + struct scrub_ctx *sctx = sblock_to_check->sctx; + struct btrfs_device *dev; struct btrfs_fs_info *fs_info; u64 length; u64 logical; @@ -633,16 +784,33 @@ static int scrub_handle_errored_block(struct scrub_block *sblock_to_check) DEFAULT_RATELIMIT_BURST); BUG_ON(sblock_to_check->page_count < 1); - fs_info = sdev->dev->dev_root->fs_info; + fs_info = sctx->dev_root->fs_info; + if (sblock_to_check->pagev[0]->flags & BTRFS_EXTENT_FLAG_SUPER) { + /* + * if we find an error in a super block, we just report it. + * They will get written with the next transaction commit + * anyway + */ + spin_lock(&sctx->stat_lock); + ++sctx->stat.super_errors; + spin_unlock(&sctx->stat_lock); + return 0; + } length = sblock_to_check->page_count * PAGE_SIZE; - logical = sblock_to_check->pagev[0].logical; - generation = sblock_to_check->pagev[0].generation; - BUG_ON(sblock_to_check->pagev[0].mirror_num < 1); - failed_mirror_index = sblock_to_check->pagev[0].mirror_num - 1; - is_metadata = !(sblock_to_check->pagev[0].flags & + logical = sblock_to_check->pagev[0]->logical; + generation = sblock_to_check->pagev[0]->generation; + BUG_ON(sblock_to_check->pagev[0]->mirror_num < 1); + failed_mirror_index = sblock_to_check->pagev[0]->mirror_num - 1; + is_metadata = !(sblock_to_check->pagev[0]->flags & BTRFS_EXTENT_FLAG_DATA); - have_csum = sblock_to_check->pagev[0].have_csum; - csum = sblock_to_check->pagev[0].csum; + have_csum = sblock_to_check->pagev[0]->have_csum; + csum = sblock_to_check->pagev[0]->csum; + dev = sblock_to_check->pagev[0]->dev; + + if (sctx->is_dev_replace && !is_metadata && !have_csum) { + sblocks_for_recheck = NULL; + goto nodatasum_case; + } /* * read all mirrors one after the other. This includes to @@ -677,43 +845,32 @@ static int scrub_handle_errored_block(struct scrub_block *sblock_to_check) sizeof(*sblocks_for_recheck), GFP_NOFS); if (!sblocks_for_recheck) { - spin_lock(&sdev->stat_lock); - sdev->stat.malloc_errors++; - sdev->stat.read_errors++; - sdev->stat.uncorrectable_errors++; - spin_unlock(&sdev->stat_lock); - btrfs_dev_stat_inc_and_print(sdev->dev, - BTRFS_DEV_STAT_READ_ERRS); + spin_lock(&sctx->stat_lock); + sctx->stat.malloc_errors++; + sctx->stat.read_errors++; + sctx->stat.uncorrectable_errors++; + spin_unlock(&sctx->stat_lock); + btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); goto out; } /* setup the context, map the logical blocks and alloc the pages */ - ret = scrub_setup_recheck_block(sdev, &fs_info->mapping_tree, length, + ret = scrub_setup_recheck_block(sctx, fs_info, sblock_to_check, length, logical, sblocks_for_recheck); if (ret) { - spin_lock(&sdev->stat_lock); - sdev->stat.read_errors++; - sdev->stat.uncorrectable_errors++; - spin_unlock(&sdev->stat_lock); - btrfs_dev_stat_inc_and_print(sdev->dev, - BTRFS_DEV_STAT_READ_ERRS); + spin_lock(&sctx->stat_lock); + sctx->stat.read_errors++; + sctx->stat.uncorrectable_errors++; + spin_unlock(&sctx->stat_lock); + btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); goto out; } BUG_ON(failed_mirror_index >= BTRFS_MAX_MIRRORS); sblock_bad = sblocks_for_recheck + failed_mirror_index; /* build and submit the bios for the failed mirror, check checksums */ - ret = scrub_recheck_block(fs_info, sblock_bad, is_metadata, have_csum, - csum, generation, sdev->csum_size); - if (ret) { - spin_lock(&sdev->stat_lock); - sdev->stat.read_errors++; - sdev->stat.uncorrectable_errors++; - spin_unlock(&sdev->stat_lock); - btrfs_dev_stat_inc_and_print(sdev->dev, - BTRFS_DEV_STAT_READ_ERRS); - goto out; - } + scrub_recheck_block(fs_info, sblock_bad, is_metadata, have_csum, + csum, generation, sctx->csum_size); if (!sblock_bad->header_error && !sblock_bad->checksum_error && sblock_bad->no_io_error_seen) { @@ -725,50 +882,54 @@ static int scrub_handle_errored_block(struct scrub_block *sblock_to_check) * different bio (usually one of the two latter cases is * the cause) */ - spin_lock(&sdev->stat_lock); - sdev->stat.unverified_errors++; - spin_unlock(&sdev->stat_lock); + spin_lock(&sctx->stat_lock); + sctx->stat.unverified_errors++; + spin_unlock(&sctx->stat_lock); + if (sctx->is_dev_replace) + scrub_write_block_to_dev_replace(sblock_bad); goto out; } if (!sblock_bad->no_io_error_seen) { - spin_lock(&sdev->stat_lock); - sdev->stat.read_errors++; - spin_unlock(&sdev->stat_lock); + spin_lock(&sctx->stat_lock); + sctx->stat.read_errors++; + spin_unlock(&sctx->stat_lock); if (__ratelimit(&_rs)) scrub_print_warning("i/o error", sblock_to_check); - btrfs_dev_stat_inc_and_print(sdev->dev, - BTRFS_DEV_STAT_READ_ERRS); + btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); } else if (sblock_bad->checksum_error) { - spin_lock(&sdev->stat_lock); - sdev->stat.csum_errors++; - spin_unlock(&sdev->stat_lock); + spin_lock(&sctx->stat_lock); + sctx->stat.csum_errors++; + spin_unlock(&sctx->stat_lock); if (__ratelimit(&_rs)) scrub_print_warning("checksum error", sblock_to_check); - btrfs_dev_stat_inc_and_print(sdev->dev, + btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS); } else if (sblock_bad->header_error) { - spin_lock(&sdev->stat_lock); - sdev->stat.verify_errors++; - spin_unlock(&sdev->stat_lock); + spin_lock(&sctx->stat_lock); + sctx->stat.verify_errors++; + spin_unlock(&sctx->stat_lock); if (__ratelimit(&_rs)) scrub_print_warning("checksum/header error", sblock_to_check); if (sblock_bad->generation_error) - btrfs_dev_stat_inc_and_print(sdev->dev, + btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_GENERATION_ERRS); else - btrfs_dev_stat_inc_and_print(sdev->dev, + btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS); } - if (sdev->readonly) + if (sctx->readonly && !sctx->is_dev_replace) goto did_not_correct_error; if (!is_metadata && !have_csum) { struct scrub_fixup_nodatasum *fixup_nodatasum; +nodatasum_case: + WARN_ON(sctx->is_dev_replace); + /* * !is_metadata and !have_csum, this means that the data * might not be COW'ed, that it might be modified @@ -779,24 +940,12 @@ static int scrub_handle_errored_block(struct scrub_block *sblock_to_check) fixup_nodatasum = kzalloc(sizeof(*fixup_nodatasum), GFP_NOFS); if (!fixup_nodatasum) goto did_not_correct_error; - fixup_nodatasum->sdev = sdev; + fixup_nodatasum->sctx = sctx; + fixup_nodatasum->dev = dev; fixup_nodatasum->logical = logical; fixup_nodatasum->root = fs_info->extent_root; fixup_nodatasum->mirror_num = failed_mirror_index + 1; - /* - * increment scrubs_running to prevent cancel requests from - * completing as long as a fixup worker is running. we must also - * increment scrubs_paused to prevent deadlocking on pause - * requests used for transactions commits (as the worker uses a - * transaction context). it is safe to regard the fixup worker - * as paused for all matters practical. effectively, we only - * avoid cancellation requests from completing. - */ - mutex_lock(&fs_info->scrub_lock); - atomic_inc(&fs_info->scrubs_running); - atomic_inc(&fs_info->scrubs_paused); - mutex_unlock(&fs_info->scrub_lock); - atomic_inc(&sdev->fixup_cnt); + scrub_pending_trans_workers_inc(sctx); fixup_nodatasum->work.func = scrub_fixup_nodatasum; btrfs_queue_worker(&fs_info->scrub_workers, &fixup_nodatasum->work); @@ -805,26 +954,8 @@ static int scrub_handle_errored_block(struct scrub_block *sblock_to_check) /* * now build and submit the bios for the other mirrors, check - * checksums - */ - for (mirror_index = 0; - mirror_index < BTRFS_MAX_MIRRORS && - sblocks_for_recheck[mirror_index].page_count > 0; - mirror_index++) { - if (mirror_index == failed_mirror_index) - continue; - - /* build and submit the bios, check checksums */ - ret = scrub_recheck_block(fs_info, - sblocks_for_recheck + mirror_index, - is_metadata, have_csum, csum, - generation, sdev->csum_size); - if (ret) - goto did_not_correct_error; - } - - /* - * first try to pick the mirror which is completely without I/O + * checksums. + * First try to pick the mirror which is completely without I/O * errors and also does not have a checksum error. * If one is found, and if a checksum is present, the full block * that is known to contain an error is rewritten. Afterwards @@ -840,24 +971,93 @@ static int scrub_handle_errored_block(struct scrub_block *sblock_to_check) mirror_index < BTRFS_MAX_MIRRORS && sblocks_for_recheck[mirror_index].page_count > 0; mirror_index++) { - struct scrub_block *sblock_other = sblocks_for_recheck + - mirror_index; + struct scrub_block *sblock_other; + + if (mirror_index == failed_mirror_index) + continue; + sblock_other = sblocks_for_recheck + mirror_index; + + /* build and submit the bios, check checksums */ + scrub_recheck_block(fs_info, sblock_other, is_metadata, + have_csum, csum, generation, + sctx->csum_size); if (!sblock_other->header_error && !sblock_other->checksum_error && sblock_other->no_io_error_seen) { - int force_write = is_metadata || have_csum; - - ret = scrub_repair_block_from_good_copy(sblock_bad, - sblock_other, - force_write); + if (sctx->is_dev_replace) { + scrub_write_block_to_dev_replace(sblock_other); + } else { + int force_write = is_metadata || have_csum; + + ret = scrub_repair_block_from_good_copy( + sblock_bad, sblock_other, + force_write); + } if (0 == ret) goto corrected_error; } } /* - * in case of I/O errors in the area that is supposed to be + * for dev_replace, pick good pages and write to the target device. + */ + if (sctx->is_dev_replace) { + success = 1; + for (page_num = 0; page_num < sblock_bad->page_count; + page_num++) { + int sub_success; + + sub_success = 0; + for (mirror_index = 0; + mirror_index < BTRFS_MAX_MIRRORS && + sblocks_for_recheck[mirror_index].page_count > 0; + mirror_index++) { + struct scrub_block *sblock_other = + sblocks_for_recheck + mirror_index; + struct scrub_page *page_other = + sblock_other->pagev[page_num]; + + if (!page_other->io_error) { + ret = scrub_write_page_to_dev_replace( + sblock_other, page_num); + if (ret == 0) { + /* succeeded for this page */ + sub_success = 1; + break; + } else { + btrfs_dev_replace_stats_inc( + &sctx->dev_root-> + fs_info->dev_replace. + num_write_errors); + } + } + } + + if (!sub_success) { + /* + * did not find a mirror to fetch the page + * from. scrub_write_page_to_dev_replace() + * handles this case (page->io_error), by + * filling the block with zeros before + * submitting the write request + */ + success = 0; + ret = scrub_write_page_to_dev_replace( + sblock_bad, page_num); + if (ret) + btrfs_dev_replace_stats_inc( + &sctx->dev_root->fs_info-> + dev_replace.num_write_errors); + } + } + + goto out; + } + + /* + * for regular scrub, repair those pages that are errored. + * In case of I/O errors in the area that is supposed to be * repaired, continue by picking good copies of those pages. * Select the good pages from mirrors to rewrite bad pages from * the area to fix. Afterwards verify the checksum of the block @@ -887,7 +1087,7 @@ static int scrub_handle_errored_block(struct scrub_block *sblock_to_check) success = 1; for (page_num = 0; page_num < sblock_bad->page_count; page_num++) { - struct scrub_page *page_bad = sblock_bad->pagev + page_num; + struct scrub_page *page_bad = sblock_bad->pagev[page_num]; if (!page_bad->io_error) continue; @@ -898,8 +1098,8 @@ static int scrub_handle_errored_block(struct scrub_block *sblock_to_check) mirror_index++) { struct scrub_block *sblock_other = sblocks_for_recheck + mirror_index; - struct scrub_page *page_other = sblock_other->pagev + - page_num; + struct scrub_page *page_other = sblock_other->pagev[ + page_num]; if (!page_other->io_error) { ret = scrub_repair_page_from_good_copy( @@ -928,10 +1128,10 @@ static int scrub_handle_errored_block(struct scrub_block *sblock_to_check) * is verified, but most likely the data comes out * of the page cache. */ - ret = scrub_recheck_block(fs_info, sblock_bad, - is_metadata, have_csum, csum, - generation, sdev->csum_size); - if (!ret && !sblock_bad->header_error && + scrub_recheck_block(fs_info, sblock_bad, + is_metadata, have_csum, csum, + generation, sctx->csum_size); + if (!sblock_bad->header_error && !sblock_bad->checksum_error && sblock_bad->no_io_error_seen) goto corrected_error; @@ -939,23 +1139,23 @@ static int scrub_handle_errored_block(struct scrub_block *sblock_to_check) goto did_not_correct_error; } else { corrected_error: - spin_lock(&sdev->stat_lock); - sdev->stat.corrected_errors++; - spin_unlock(&sdev->stat_lock); + spin_lock(&sctx->stat_lock); + sctx->stat.corrected_errors++; + spin_unlock(&sctx->stat_lock); printk_ratelimited_in_rcu(KERN_ERR "btrfs: fixed up error at logical %llu on dev %s\n", (unsigned long long)logical, - rcu_str_deref(sdev->dev->name)); + rcu_str_deref(dev->name)); } } else { did_not_correct_error: - spin_lock(&sdev->stat_lock); - sdev->stat.uncorrectable_errors++; - spin_unlock(&sdev->stat_lock); + spin_lock(&sctx->stat_lock); + sctx->stat.uncorrectable_errors++; + spin_unlock(&sctx->stat_lock); printk_ratelimited_in_rcu(KERN_ERR "btrfs: unable to fixup (regular) error at logical %llu on dev %s\n", (unsigned long long)logical, - rcu_str_deref(sdev->dev->name)); + rcu_str_deref(dev->name)); } out: @@ -966,11 +1166,11 @@ out: mirror_index; int page_index; - for (page_index = 0; page_index < SCRUB_PAGES_PER_BIO; - page_index++) - if (sblock->pagev[page_index].page) - __free_page( - sblock->pagev[page_index].page); + for (page_index = 0; page_index < sblock->page_count; + page_index++) { + sblock->pagev[page_index]->sblock = NULL; + scrub_page_put(sblock->pagev[page_index]); + } } kfree(sblocks_for_recheck); } @@ -978,8 +1178,9 @@ out: return 0; } -static int scrub_setup_recheck_block(struct scrub_dev *sdev, - struct btrfs_mapping_tree *map_tree, +static int scrub_setup_recheck_block(struct scrub_ctx *sctx, + struct btrfs_fs_info *fs_info, + struct scrub_block *original_sblock, u64 length, u64 logical, struct scrub_block *sblocks_for_recheck) { @@ -988,7 +1189,7 @@ static int scrub_setup_recheck_block(struct scrub_dev *sdev, int ret; /* - * note: the three members sdev, ref_count and outstanding_pages + * note: the two members ref_count and outstanding_pages * are not used (and not set) in the blocks that are used for * the recheck procedure */ @@ -1003,14 +1204,14 @@ static int scrub_setup_recheck_block(struct scrub_dev *sdev, * with a length of PAGE_SIZE, each returned stripe * represents one mirror */ - ret = btrfs_map_block(map_tree, WRITE, logical, &mapped_length, - &bbio, 0); + ret = btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS, logical, + &mapped_length, &bbio, 0); if (ret || !bbio || mapped_length < sublen) { kfree(bbio); return -EIO; } - BUG_ON(page_index >= SCRUB_PAGES_PER_BIO); + BUG_ON(page_index >= SCRUB_PAGES_PER_RD_BIO); for (mirror_index = 0; mirror_index < (int)bbio->num_stripes; mirror_index++) { struct scrub_block *sblock; @@ -1020,21 +1221,31 @@ static int scrub_setup_recheck_block(struct scrub_dev *sdev, continue; sblock = sblocks_for_recheck + mirror_index; - page = sblock->pagev + page_index; + sblock->sctx = sctx; + page = kzalloc(sizeof(*page), GFP_NOFS); + if (!page) { +leave_nomem: + spin_lock(&sctx->stat_lock); + sctx->stat.malloc_errors++; + spin_unlock(&sctx->stat_lock); + kfree(bbio); + return -ENOMEM; + } + scrub_page_get(page); + sblock->pagev[page_index] = page; page->logical = logical; page->physical = bbio->stripes[mirror_index].physical; + BUG_ON(page_index >= original_sblock->page_count); + page->physical_for_dev_replace = + original_sblock->pagev[page_index]-> + physical_for_dev_replace; /* for missing devices, dev->bdev is NULL */ page->dev = bbio->stripes[mirror_index].dev; page->mirror_num = mirror_index + 1; - page->page = alloc_page(GFP_NOFS); - if (!page->page) { - spin_lock(&sdev->stat_lock); - sdev->stat.malloc_errors++; - spin_unlock(&sdev->stat_lock); - kfree(bbio); - return -ENOMEM; - } sblock->page_count++; + page->page = alloc_page(GFP_NOFS); + if (!page->page) + goto leave_nomem; } kfree(bbio); length -= sublen; @@ -1052,10 +1263,10 @@ static int scrub_setup_recheck_block(struct scrub_dev *sdev, * to take those pages that are not errored from all the mirrors so that * the pages that are errored in the just handled mirror can be repaired. */ -static int scrub_recheck_block(struct btrfs_fs_info *fs_info, - struct scrub_block *sblock, int is_metadata, - int have_csum, u8 *csum, u64 generation, - u16 csum_size) +static void scrub_recheck_block(struct btrfs_fs_info *fs_info, + struct scrub_block *sblock, int is_metadata, + int have_csum, u8 *csum, u64 generation, + u16 csum_size) { int page_num; @@ -1065,8 +1276,7 @@ static int scrub_recheck_block(struct btrfs_fs_info *fs_info, for (page_num = 0; page_num < sblock->page_count; page_num++) { struct bio *bio; - int ret; - struct scrub_page *page = sblock->pagev + page_num; + struct scrub_page *page = sblock->pagev[page_num]; DECLARE_COMPLETION_ONSTACK(complete); if (page->dev->bdev == NULL) { @@ -1075,20 +1285,19 @@ static int scrub_recheck_block(struct btrfs_fs_info *fs_info, continue; } - BUG_ON(!page->page); + WARN_ON(!page->page); bio = bio_alloc(GFP_NOFS, 1); - if (!bio) - return -EIO; + if (!bio) { + page->io_error = 1; + sblock->no_io_error_seen = 0; + continue; + } bio->bi_bdev = page->dev->bdev; bio->bi_sector = page->physical >> 9; bio->bi_end_io = scrub_complete_bio_end_io; bio->bi_private = &complete; - ret = bio_add_page(bio, page->page, PAGE_SIZE, 0); - if (PAGE_SIZE != ret) { - bio_put(bio); - return -EIO; - } + bio_add_page(bio, page->page, PAGE_SIZE, 0); btrfsic_submit_bio(READ, bio); /* this will also unplug the queue */ @@ -1105,7 +1314,7 @@ static int scrub_recheck_block(struct btrfs_fs_info *fs_info, have_csum, csum, generation, csum_size); - return 0; + return; } static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info, @@ -1120,14 +1329,14 @@ static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info, struct btrfs_root *root = fs_info->extent_root; void *mapped_buffer; - BUG_ON(!sblock->pagev[0].page); + WARN_ON(!sblock->pagev[0]->page); if (is_metadata) { struct btrfs_header *h; - mapped_buffer = kmap_atomic(sblock->pagev[0].page); + mapped_buffer = kmap_atomic(sblock->pagev[0]->page); h = (struct btrfs_header *)mapped_buffer; - if (sblock->pagev[0].logical != le64_to_cpu(h->bytenr) || + if (sblock->pagev[0]->logical != le64_to_cpu(h->bytenr) || memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE) || memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid, BTRFS_UUID_SIZE)) { @@ -1141,7 +1350,7 @@ static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info, if (!have_csum) return; - mapped_buffer = kmap_atomic(sblock->pagev[0].page); + mapped_buffer = kmap_atomic(sblock->pagev[0]->page); } for (page_num = 0;;) { @@ -1157,9 +1366,9 @@ static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info, page_num++; if (page_num >= sblock->page_count) break; - BUG_ON(!sblock->pagev[page_num].page); + WARN_ON(!sblock->pagev[page_num]->page); - mapped_buffer = kmap_atomic(sblock->pagev[page_num].page); + mapped_buffer = kmap_atomic(sblock->pagev[page_num]->page); } btrfs_csum_final(crc, calculated_csum); @@ -1197,17 +1406,23 @@ static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad, struct scrub_block *sblock_good, int page_num, int force_write) { - struct scrub_page *page_bad = sblock_bad->pagev + page_num; - struct scrub_page *page_good = sblock_good->pagev + page_num; + struct scrub_page *page_bad = sblock_bad->pagev[page_num]; + struct scrub_page *page_good = sblock_good->pagev[page_num]; - BUG_ON(sblock_bad->pagev[page_num].page == NULL); - BUG_ON(sblock_good->pagev[page_num].page == NULL); + BUG_ON(page_bad->page == NULL); + BUG_ON(page_good->page == NULL); if (force_write || sblock_bad->header_error || sblock_bad->checksum_error || page_bad->io_error) { struct bio *bio; int ret; DECLARE_COMPLETION_ONSTACK(complete); + if (!page_bad->dev->bdev) { + printk_ratelimited(KERN_WARNING + "btrfs: scrub_repair_page_from_good_copy(bdev == NULL) is unexpected!\n"); + return -EIO; + } + bio = bio_alloc(GFP_NOFS, 1); if (!bio) return -EIO; @@ -1228,6 +1443,9 @@ static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad, if (!bio_flagged(bio, BIO_UPTODATE)) { btrfs_dev_stat_inc_and_print(page_bad->dev, BTRFS_DEV_STAT_WRITE_ERRS); + btrfs_dev_replace_stats_inc( + &sblock_bad->sctx->dev_root->fs_info-> + dev_replace.num_write_errors); bio_put(bio); return -EIO; } @@ -1237,13 +1455,174 @@ static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad, return 0; } -static void scrub_checksum(struct scrub_block *sblock) +static void scrub_write_block_to_dev_replace(struct scrub_block *sblock) +{ + int page_num; + + for (page_num = 0; page_num < sblock->page_count; page_num++) { + int ret; + + ret = scrub_write_page_to_dev_replace(sblock, page_num); + if (ret) + btrfs_dev_replace_stats_inc( + &sblock->sctx->dev_root->fs_info->dev_replace. + num_write_errors); + } +} + +static int scrub_write_page_to_dev_replace(struct scrub_block *sblock, + int page_num) +{ + struct scrub_page *spage = sblock->pagev[page_num]; + + BUG_ON(spage->page == NULL); + if (spage->io_error) { + void *mapped_buffer = kmap_atomic(spage->page); + + memset(mapped_buffer, 0, PAGE_CACHE_SIZE); + flush_dcache_page(spage->page); + kunmap_atomic(mapped_buffer); + } + return scrub_add_page_to_wr_bio(sblock->sctx, spage); +} + +static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx, + struct scrub_page *spage) +{ + struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx; + struct scrub_bio *sbio; + int ret; + + mutex_lock(&wr_ctx->wr_lock); +again: + if (!wr_ctx->wr_curr_bio) { + wr_ctx->wr_curr_bio = kzalloc(sizeof(*wr_ctx->wr_curr_bio), + GFP_NOFS); + if (!wr_ctx->wr_curr_bio) { + mutex_unlock(&wr_ctx->wr_lock); + return -ENOMEM; + } + wr_ctx->wr_curr_bio->sctx = sctx; + wr_ctx->wr_curr_bio->page_count = 0; + } + sbio = wr_ctx->wr_curr_bio; + if (sbio->page_count == 0) { + struct bio *bio; + + sbio->physical = spage->physical_for_dev_replace; + sbio->logical = spage->logical; + sbio->dev = wr_ctx->tgtdev; + bio = sbio->bio; + if (!bio) { + bio = bio_alloc(GFP_NOFS, wr_ctx->pages_per_wr_bio); + if (!bio) { + mutex_unlock(&wr_ctx->wr_lock); + return -ENOMEM; + } + sbio->bio = bio; + } + + bio->bi_private = sbio; + bio->bi_end_io = scrub_wr_bio_end_io; + bio->bi_bdev = sbio->dev->bdev; + bio->bi_sector = sbio->physical >> 9; + sbio->err = 0; + } else if (sbio->physical + sbio->page_count * PAGE_SIZE != + spage->physical_for_dev_replace || + sbio->logical + sbio->page_count * PAGE_SIZE != + spage->logical) { + scrub_wr_submit(sctx); + goto again; + } + + ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0); + if (ret != PAGE_SIZE) { + if (sbio->page_count < 1) { + bio_put(sbio->bio); + sbio->bio = NULL; + mutex_unlock(&wr_ctx->wr_lock); + return -EIO; + } + scrub_wr_submit(sctx); + goto again; + } + + sbio->pagev[sbio->page_count] = spage; + scrub_page_get(spage); + sbio->page_count++; + if (sbio->page_count == wr_ctx->pages_per_wr_bio) + scrub_wr_submit(sctx); + mutex_unlock(&wr_ctx->wr_lock); + + return 0; +} + +static void scrub_wr_submit(struct scrub_ctx *sctx) +{ + struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx; + struct scrub_bio *sbio; + + if (!wr_ctx->wr_curr_bio) + return; + + sbio = wr_ctx->wr_curr_bio; + wr_ctx->wr_curr_bio = NULL; + WARN_ON(!sbio->bio->bi_bdev); + scrub_pending_bio_inc(sctx); + /* process all writes in a single worker thread. Then the block layer + * orders the requests before sending them to the driver which + * doubled the write performance on spinning disks when measured + * with Linux 3.5 */ + btrfsic_submit_bio(WRITE, sbio->bio); +} + +static void scrub_wr_bio_end_io(struct bio *bio, int err) +{ + struct scrub_bio *sbio = bio->bi_private; + struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info; + + sbio->err = err; + sbio->bio = bio; + + sbio->work.func = scrub_wr_bio_end_io_worker; + btrfs_queue_worker(&fs_info->scrub_wr_completion_workers, &sbio->work); +} + +static void scrub_wr_bio_end_io_worker(struct btrfs_work *work) +{ + struct scrub_bio *sbio = container_of(work, struct scrub_bio, work); + struct scrub_ctx *sctx = sbio->sctx; + int i; + + WARN_ON(sbio->page_count > SCRUB_PAGES_PER_WR_BIO); + if (sbio->err) { + struct btrfs_dev_replace *dev_replace = + &sbio->sctx->dev_root->fs_info->dev_replace; + + for (i = 0; i < sbio->page_count; i++) { + struct scrub_page *spage = sbio->pagev[i]; + + spage->io_error = 1; + btrfs_dev_replace_stats_inc(&dev_replace-> + num_write_errors); + } + } + + for (i = 0; i < sbio->page_count; i++) + scrub_page_put(sbio->pagev[i]); + + bio_put(sbio->bio); + kfree(sbio); + scrub_pending_bio_dec(sctx); +} + +static int scrub_checksum(struct scrub_block *sblock) { u64 flags; int ret; - BUG_ON(sblock->page_count < 1); - flags = sblock->pagev[0].flags; + WARN_ON(sblock->page_count < 1); + flags = sblock->pagev[0]->flags; ret = 0; if (flags & BTRFS_EXTENT_FLAG_DATA) ret = scrub_checksum_data(sblock); @@ -1255,30 +1634,32 @@ static void scrub_checksum(struct scrub_block *sblock) WARN_ON(1); if (ret) scrub_handle_errored_block(sblock); + + return ret; } static int scrub_checksum_data(struct scrub_block *sblock) { - struct scrub_dev *sdev = sblock->sdev; + struct scrub_ctx *sctx = sblock->sctx; u8 csum[BTRFS_CSUM_SIZE]; u8 *on_disk_csum; struct page *page; void *buffer; u32 crc = ~(u32)0; int fail = 0; - struct btrfs_root *root = sdev->dev->dev_root; + struct btrfs_root *root = sctx->dev_root; u64 len; int index; BUG_ON(sblock->page_count < 1); - if (!sblock->pagev[0].have_csum) + if (!sblock->pagev[0]->have_csum) return 0; - on_disk_csum = sblock->pagev[0].csum; - page = sblock->pagev[0].page; + on_disk_csum = sblock->pagev[0]->csum; + page = sblock->pagev[0]->page; buffer = kmap_atomic(page); - len = sdev->sectorsize; + len = sctx->sectorsize; index = 0; for (;;) { u64 l = min_t(u64, len, PAGE_SIZE); @@ -1290,13 +1671,13 @@ static int scrub_checksum_data(struct scrub_block *sblock) break; index++; BUG_ON(index >= sblock->page_count); - BUG_ON(!sblock->pagev[index].page); - page = sblock->pagev[index].page; + BUG_ON(!sblock->pagev[index]->page); + page = sblock->pagev[index]->page; buffer = kmap_atomic(page); } btrfs_csum_final(crc, csum); - if (memcmp(csum, on_disk_csum, sdev->csum_size)) + if (memcmp(csum, on_disk_csum, sctx->csum_size)) fail = 1; return fail; @@ -1304,9 +1685,9 @@ static int scrub_checksum_data(struct scrub_block *sblock) static int scrub_checksum_tree_block(struct scrub_block *sblock) { - struct scrub_dev *sdev = sblock->sdev; + struct scrub_ctx *sctx = sblock->sctx; struct btrfs_header *h; - struct btrfs_root *root = sdev->dev->dev_root; + struct btrfs_root *root = sctx->dev_root; struct btrfs_fs_info *fs_info = root->fs_info; u8 calculated_csum[BTRFS_CSUM_SIZE]; u8 on_disk_csum[BTRFS_CSUM_SIZE]; @@ -1321,10 +1702,10 @@ static int scrub_checksum_tree_block(struct scrub_block *sblock) int index; BUG_ON(sblock->page_count < 1); - page = sblock->pagev[0].page; + page = sblock->pagev[0]->page; mapped_buffer = kmap_atomic(page); h = (struct btrfs_header *)mapped_buffer; - memcpy(on_disk_csum, h->csum, sdev->csum_size); + memcpy(on_disk_csum, h->csum, sctx->csum_size); /* * we don't use the getter functions here, as we @@ -1332,10 +1713,10 @@ static int scrub_checksum_tree_block(struct scrub_block *sblock) * b) the page is already kmapped */ - if (sblock->pagev[0].logical != le64_to_cpu(h->bytenr)) + if (sblock->pagev[0]->logical != le64_to_cpu(h->bytenr)) ++fail; - if (sblock->pagev[0].generation != le64_to_cpu(h->generation)) + if (sblock->pagev[0]->generation != le64_to_cpu(h->generation)) ++fail; if (memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE)) @@ -1345,8 +1726,8 @@ static int scrub_checksum_tree_block(struct scrub_block *sblock) BTRFS_UUID_SIZE)) ++fail; - BUG_ON(sdev->nodesize != sdev->leafsize); - len = sdev->nodesize - BTRFS_CSUM_SIZE; + WARN_ON(sctx->nodesize != sctx->leafsize); + len = sctx->nodesize - BTRFS_CSUM_SIZE; mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE; p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE; index = 0; @@ -1360,15 +1741,15 @@ static int scrub_checksum_tree_block(struct scrub_block *sblock) break; index++; BUG_ON(index >= sblock->page_count); - BUG_ON(!sblock->pagev[index].page); - page = sblock->pagev[index].page; + BUG_ON(!sblock->pagev[index]->page); + page = sblock->pagev[index]->page; mapped_buffer = kmap_atomic(page); mapped_size = PAGE_SIZE; p = mapped_buffer; } btrfs_csum_final(crc, calculated_csum); - if (memcmp(calculated_csum, on_disk_csum, sdev->csum_size)) + if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size)) ++crc_fail; return fail || crc_fail; @@ -1377,8 +1758,8 @@ static int scrub_checksum_tree_block(struct scrub_block *sblock) static int scrub_checksum_super(struct scrub_block *sblock) { struct btrfs_super_block *s; - struct scrub_dev *sdev = sblock->sdev; - struct btrfs_root *root = sdev->dev->dev_root; + struct scrub_ctx *sctx = sblock->sctx; + struct btrfs_root *root = sctx->dev_root; struct btrfs_fs_info *fs_info = root->fs_info; u8 calculated_csum[BTRFS_CSUM_SIZE]; u8 on_disk_csum[BTRFS_CSUM_SIZE]; @@ -1393,15 +1774,15 @@ static int scrub_checksum_super(struct scrub_block *sblock) int index; BUG_ON(sblock->page_count < 1); - page = sblock->pagev[0].page; + page = sblock->pagev[0]->page; mapped_buffer = kmap_atomic(page); s = (struct btrfs_super_block *)mapped_buffer; - memcpy(on_disk_csum, s->csum, sdev->csum_size); + memcpy(on_disk_csum, s->csum, sctx->csum_size); - if (sblock->pagev[0].logical != le64_to_cpu(s->bytenr)) + if (sblock->pagev[0]->logical != le64_to_cpu(s->bytenr)) ++fail_cor; - if (sblock->pagev[0].generation != le64_to_cpu(s->generation)) + if (sblock->pagev[0]->generation != le64_to_cpu(s->generation)) ++fail_gen; if (memcmp(s->fsid, fs_info->fsid, BTRFS_UUID_SIZE)) @@ -1421,15 +1802,15 @@ static int scrub_checksum_super(struct scrub_block *sblock) break; index++; BUG_ON(index >= sblock->page_count); - BUG_ON(!sblock->pagev[index].page); - page = sblock->pagev[index].page; + BUG_ON(!sblock->pagev[index]->page); + page = sblock->pagev[index]->page; mapped_buffer = kmap_atomic(page); mapped_size = PAGE_SIZE; p = mapped_buffer; } btrfs_csum_final(crc, calculated_csum); - if (memcmp(calculated_csum, on_disk_csum, sdev->csum_size)) + if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size)) ++fail_cor; if (fail_cor + fail_gen) { @@ -1438,14 +1819,14 @@ static int scrub_checksum_super(struct scrub_block *sblock) * They will get written with the next transaction commit * anyway */ - spin_lock(&sdev->stat_lock); - ++sdev->stat.super_errors; - spin_unlock(&sdev->stat_lock); + spin_lock(&sctx->stat_lock); + ++sctx->stat.super_errors; + spin_unlock(&sctx->stat_lock); if (fail_cor) - btrfs_dev_stat_inc_and_print(sdev->dev, + btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev, BTRFS_DEV_STAT_CORRUPTION_ERRS); else - btrfs_dev_stat_inc_and_print(sdev->dev, + btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev, BTRFS_DEV_STAT_GENERATION_ERRS); } @@ -1463,28 +1844,54 @@ static void scrub_block_put(struct scrub_block *sblock) int i; for (i = 0; i < sblock->page_count; i++) - if (sblock->pagev[i].page) - __free_page(sblock->pagev[i].page); + scrub_page_put(sblock->pagev[i]); kfree(sblock); } } -static void scrub_submit(struct scrub_dev *sdev) +static void scrub_page_get(struct scrub_page *spage) +{ + atomic_inc(&spage->ref_count); +} + +static void scrub_page_put(struct scrub_page *spage) +{ + if (atomic_dec_and_test(&spage->ref_count)) { + if (spage->page) + __free_page(spage->page); + kfree(spage); + } +} + +static void scrub_submit(struct scrub_ctx *sctx) { struct scrub_bio *sbio; - if (sdev->curr == -1) + if (sctx->curr == -1) return; - sbio = sdev->bios[sdev->curr]; - sdev->curr = -1; - atomic_inc(&sdev->in_flight); + sbio = sctx->bios[sctx->curr]; + sctx->curr = -1; + scrub_pending_bio_inc(sctx); - btrfsic_submit_bio(READ, sbio->bio); + if (!sbio->bio->bi_bdev) { + /* + * this case should not happen. If btrfs_map_block() is + * wrong, it could happen for dev-replace operations on + * missing devices when no mirrors are available, but in + * this case it should already fail the mount. + * This case is handled correctly (but _very_ slowly). + */ + printk_ratelimited(KERN_WARNING + "btrfs: scrub_submit(bio bdev == NULL) is unexpected!\n"); + bio_endio(sbio->bio, -EIO); + } else { + btrfsic_submit_bio(READ, sbio->bio); + } } -static int scrub_add_page_to_bio(struct scrub_dev *sdev, - struct scrub_page *spage) +static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx, + struct scrub_page *spage) { struct scrub_block *sblock = spage->sblock; struct scrub_bio *sbio; @@ -1494,28 +1901,29 @@ again: /* * grab a fresh bio or wait for one to become available */ - while (sdev->curr == -1) { - spin_lock(&sdev->list_lock); - sdev->curr = sdev->first_free; - if (sdev->curr != -1) { - sdev->first_free = sdev->bios[sdev->curr]->next_free; - sdev->bios[sdev->curr]->next_free = -1; - sdev->bios[sdev->curr]->page_count = 0; - spin_unlock(&sdev->list_lock); + while (sctx->curr == -1) { + spin_lock(&sctx->list_lock); + sctx->curr = sctx->first_free; + if (sctx->curr != -1) { + sctx->first_free = sctx->bios[sctx->curr]->next_free; + sctx->bios[sctx->curr]->next_free = -1; + sctx->bios[sctx->curr]->page_count = 0; + spin_unlock(&sctx->list_lock); } else { - spin_unlock(&sdev->list_lock); - wait_event(sdev->list_wait, sdev->first_free != -1); + spin_unlock(&sctx->list_lock); + wait_event(sctx->list_wait, sctx->first_free != -1); } } - sbio = sdev->bios[sdev->curr]; + sbio = sctx->bios[sctx->curr]; if (sbio->page_count == 0) { struct bio *bio; sbio->physical = spage->physical; sbio->logical = spage->logical; + sbio->dev = spage->dev; bio = sbio->bio; if (!bio) { - bio = bio_alloc(GFP_NOFS, sdev->pages_per_bio); + bio = bio_alloc(GFP_NOFS, sctx->pages_per_rd_bio); if (!bio) return -ENOMEM; sbio->bio = bio; @@ -1523,14 +1931,15 @@ again: bio->bi_private = sbio; bio->bi_end_io = scrub_bio_end_io; - bio->bi_bdev = sdev->dev->bdev; - bio->bi_sector = spage->physical >> 9; + bio->bi_bdev = sbio->dev->bdev; + bio->bi_sector = sbio->physical >> 9; sbio->err = 0; } else if (sbio->physical + sbio->page_count * PAGE_SIZE != spage->physical || sbio->logical + sbio->page_count * PAGE_SIZE != - spage->logical) { - scrub_submit(sdev); + spage->logical || + sbio->dev != spage->dev) { + scrub_submit(sctx); goto again; } @@ -1542,81 +1951,87 @@ again: sbio->bio = NULL; return -EIO; } - scrub_submit(sdev); + scrub_submit(sctx); goto again; } - scrub_block_get(sblock); /* one for the added page */ + scrub_block_get(sblock); /* one for the page added to the bio */ atomic_inc(&sblock->outstanding_pages); sbio->page_count++; - if (sbio->page_count == sdev->pages_per_bio) - scrub_submit(sdev); + if (sbio->page_count == sctx->pages_per_rd_bio) + scrub_submit(sctx); return 0; } -static int scrub_pages(struct scrub_dev *sdev, u64 logical, u64 len, - u64 physical, u64 flags, u64 gen, int mirror_num, - u8 *csum, int force) +static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len, + u64 physical, struct btrfs_device *dev, u64 flags, + u64 gen, int mirror_num, u8 *csum, int force, + u64 physical_for_dev_replace) { struct scrub_block *sblock; int index; sblock = kzalloc(sizeof(*sblock), GFP_NOFS); if (!sblock) { - spin_lock(&sdev->stat_lock); - sdev->stat.malloc_errors++; - spin_unlock(&sdev->stat_lock); + spin_lock(&sctx->stat_lock); + sctx->stat.malloc_errors++; + spin_unlock(&sctx->stat_lock); return -ENOMEM; } - /* one ref inside this function, plus one for each page later on */ + /* one ref inside this function, plus one for each page added to + * a bio later on */ atomic_set(&sblock->ref_count, 1); - sblock->sdev = sdev; + sblock->sctx = sctx; sblock->no_io_error_seen = 1; for (index = 0; len > 0; index++) { - struct scrub_page *spage = sblock->pagev + index; + struct scrub_page *spage; u64 l = min_t(u64, len, PAGE_SIZE); - BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK); - spage->page = alloc_page(GFP_NOFS); - if (!spage->page) { - spin_lock(&sdev->stat_lock); - sdev->stat.malloc_errors++; - spin_unlock(&sdev->stat_lock); - while (index > 0) { - index--; - __free_page(sblock->pagev[index].page); - } - kfree(sblock); + spage = kzalloc(sizeof(*spage), GFP_NOFS); + if (!spage) { +leave_nomem: + spin_lock(&sctx->stat_lock); + sctx->stat.malloc_errors++; + spin_unlock(&sctx->stat_lock); + scrub_block_put(sblock); return -ENOMEM; } + BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK); + scrub_page_get(spage); + sblock->pagev[index] = spage; spage->sblock = sblock; - spage->dev = sdev->dev; + spage->dev = dev; spage->flags = flags; spage->generation = gen; spage->logical = logical; spage->physical = physical; + spage->physical_for_dev_replace = physical_for_dev_replace; spage->mirror_num = mirror_num; if (csum) { spage->have_csum = 1; - memcpy(spage->csum, csum, sdev->csum_size); + memcpy(spage->csum, csum, sctx->csum_size); } else { spage->have_csum = 0; } sblock->page_count++; + spage->page = alloc_page(GFP_NOFS); + if (!spage->page) + goto leave_nomem; len -= l; logical += l; physical += l; + physical_for_dev_replace += l; } - BUG_ON(sblock->page_count == 0); + WARN_ON(sblock->page_count == 0); for (index = 0; index < sblock->page_count; index++) { - struct scrub_page *spage = sblock->pagev + index; + struct scrub_page *spage = sblock->pagev[index]; int ret; - ret = scrub_add_page_to_bio(sdev, spage); + ret = scrub_add_page_to_rd_bio(sctx, spage); if (ret) { scrub_block_put(sblock); return ret; @@ -1624,7 +2039,7 @@ static int scrub_pages(struct scrub_dev *sdev, u64 logical, u64 len, } if (force) - scrub_submit(sdev); + scrub_submit(sctx); /* last one frees, either here or in bio completion for last page */ scrub_block_put(sblock); @@ -1634,8 +2049,7 @@ static int scrub_pages(struct scrub_dev *sdev, u64 logical, u64 len, static void scrub_bio_end_io(struct bio *bio, int err) { struct scrub_bio *sbio = bio->bi_private; - struct scrub_dev *sdev = sbio->sdev; - struct btrfs_fs_info *fs_info = sdev->dev->dev_root->fs_info; + struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info; sbio->err = err; sbio->bio = bio; @@ -1646,10 +2060,10 @@ static void scrub_bio_end_io(struct bio *bio, int err) static void scrub_bio_end_io_worker(struct btrfs_work *work) { struct scrub_bio *sbio = container_of(work, struct scrub_bio, work); - struct scrub_dev *sdev = sbio->sdev; + struct scrub_ctx *sctx = sbio->sctx; int i; - BUG_ON(sbio->page_count > SCRUB_PAGES_PER_BIO); + BUG_ON(sbio->page_count > SCRUB_PAGES_PER_RD_BIO); if (sbio->err) { for (i = 0; i < sbio->page_count; i++) { struct scrub_page *spage = sbio->pagev[i]; @@ -1671,23 +2085,37 @@ static void scrub_bio_end_io_worker(struct btrfs_work *work) bio_put(sbio->bio); sbio->bio = NULL; - spin_lock(&sdev->list_lock); - sbio->next_free = sdev->first_free; - sdev->first_free = sbio->index; - spin_unlock(&sdev->list_lock); - atomic_dec(&sdev->in_flight); - wake_up(&sdev->list_wait); + spin_lock(&sctx->list_lock); + sbio->next_free = sctx->first_free; + sctx->first_free = sbio->index; + spin_unlock(&sctx->list_lock); + + if (sctx->is_dev_replace && + atomic_read(&sctx->wr_ctx.flush_all_writes)) { + mutex_lock(&sctx->wr_ctx.wr_lock); + scrub_wr_submit(sctx); + mutex_unlock(&sctx->wr_ctx.wr_lock); + } + + scrub_pending_bio_dec(sctx); } static void scrub_block_complete(struct scrub_block *sblock) { - if (!sblock->no_io_error_seen) + if (!sblock->no_io_error_seen) { scrub_handle_errored_block(sblock); - else - scrub_checksum(sblock); + } else { + /* + * if has checksum error, write via repair mechanism in + * dev replace case, otherwise write here in dev replace + * case. + */ + if (!scrub_checksum(sblock) && sblock->sctx->is_dev_replace) + scrub_write_block_to_dev_replace(sblock); + } } -static int scrub_find_csum(struct scrub_dev *sdev, u64 logical, u64 len, +static int scrub_find_csum(struct scrub_ctx *sctx, u64 logical, u64 len, u8 *csum) { struct btrfs_ordered_sum *sum = NULL; @@ -1695,15 +2123,15 @@ static int scrub_find_csum(struct scrub_dev *sdev, u64 logical, u64 len, unsigned long i; unsigned long num_sectors; - while (!list_empty(&sdev->csum_list)) { - sum = list_first_entry(&sdev->csum_list, + while (!list_empty(&sctx->csum_list)) { + sum = list_first_entry(&sctx->csum_list, struct btrfs_ordered_sum, list); if (sum->bytenr > logical) return 0; if (sum->bytenr + sum->len > logical) break; - ++sdev->stat.csum_discards; + ++sctx->stat.csum_discards; list_del(&sum->list); kfree(sum); sum = NULL; @@ -1711,10 +2139,10 @@ static int scrub_find_csum(struct scrub_dev *sdev, u64 logical, u64 len, if (!sum) return 0; - num_sectors = sum->len / sdev->sectorsize; + num_sectors = sum->len / sctx->sectorsize; for (i = 0; i < num_sectors; ++i) { if (sum->sums[i].bytenr == logical) { - memcpy(csum, &sum->sums[i].sum, sdev->csum_size); + memcpy(csum, &sum->sums[i].sum, sctx->csum_size); ret = 1; break; } @@ -1727,29 +2155,30 @@ static int scrub_find_csum(struct scrub_dev *sdev, u64 logical, u64 len, } /* scrub extent tries to collect up to 64 kB for each bio */ -static int scrub_extent(struct scrub_dev *sdev, u64 logical, u64 len, - u64 physical, u64 flags, u64 gen, int mirror_num) +static int scrub_extent(struct scrub_ctx *sctx, u64 logical, u64 len, + u64 physical, struct btrfs_device *dev, u64 flags, + u64 gen, int mirror_num, u64 physical_for_dev_replace) { int ret; u8 csum[BTRFS_CSUM_SIZE]; u32 blocksize; if (flags & BTRFS_EXTENT_FLAG_DATA) { - blocksize = sdev->sectorsize; - spin_lock(&sdev->stat_lock); - sdev->stat.data_extents_scrubbed++; - sdev->stat.data_bytes_scrubbed += len; - spin_unlock(&sdev->stat_lock); + blocksize = sctx->sectorsize; + spin_lock(&sctx->stat_lock); + sctx->stat.data_extents_scrubbed++; + sctx->stat.data_bytes_scrubbed += len; + spin_unlock(&sctx->stat_lock); } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { - BUG_ON(sdev->nodesize != sdev->leafsize); - blocksize = sdev->nodesize; - spin_lock(&sdev->stat_lock); - sdev->stat.tree_extents_scrubbed++; - sdev->stat.tree_bytes_scrubbed += len; - spin_unlock(&sdev->stat_lock); + WARN_ON(sctx->nodesize != sctx->leafsize); + blocksize = sctx->nodesize; + spin_lock(&sctx->stat_lock); + sctx->stat.tree_extents_scrubbed++; + sctx->stat.tree_bytes_scrubbed += len; + spin_unlock(&sctx->stat_lock); } else { - blocksize = sdev->sectorsize; - BUG_ON(1); + blocksize = sctx->sectorsize; + WARN_ON(1); } while (len) { @@ -1758,26 +2187,38 @@ static int scrub_extent(struct scrub_dev *sdev, u64 logical, u64 len, if (flags & BTRFS_EXTENT_FLAG_DATA) { /* push csums to sbio */ - have_csum = scrub_find_csum(sdev, logical, l, csum); + have_csum = scrub_find_csum(sctx, logical, l, csum); if (have_csum == 0) - ++sdev->stat.no_csum; + ++sctx->stat.no_csum; + if (sctx->is_dev_replace && !have_csum) { + ret = copy_nocow_pages(sctx, logical, l, + mirror_num, + physical_for_dev_replace); + goto behind_scrub_pages; + } } - ret = scrub_pages(sdev, logical, l, physical, flags, gen, - mirror_num, have_csum ? csum : NULL, 0); + ret = scrub_pages(sctx, logical, l, physical, dev, flags, gen, + mirror_num, have_csum ? csum : NULL, 0, + physical_for_dev_replace); +behind_scrub_pages: if (ret) return ret; len -= l; logical += l; physical += l; + physical_for_dev_replace += l; } return 0; } -static noinline_for_stack int scrub_stripe(struct scrub_dev *sdev, - struct map_lookup *map, int num, u64 base, u64 length) +static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx, + struct map_lookup *map, + struct btrfs_device *scrub_dev, + int num, u64 base, u64 length, + int is_dev_replace) { struct btrfs_path *path; - struct btrfs_fs_info *fs_info = sdev->dev->dev_root->fs_info; + struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; struct btrfs_root *root = fs_info->extent_root; struct btrfs_root *csum_root = fs_info->csum_root; struct btrfs_extent_item *extent; @@ -1797,9 +2238,13 @@ static noinline_for_stack int scrub_stripe(struct scrub_dev *sdev, struct reada_control *reada2; struct btrfs_key key_start; struct btrfs_key key_end; - u64 increment = map->stripe_len; u64 offset; + u64 extent_logical; + u64 extent_physical; + u64 extent_len; + struct btrfs_device *extent_dev; + int extent_mirror_num; nstripes = length; offset = 0; @@ -1843,8 +2288,8 @@ static noinline_for_stack int scrub_stripe(struct scrub_dev *sdev, */ logical = base + offset; - wait_event(sdev->list_wait, - atomic_read(&sdev->in_flight) == 0); + wait_event(sctx->list_wait, + atomic_read(&sctx->bios_in_flight) == 0); atomic_inc(&fs_info->scrubs_paused); wake_up(&fs_info->scrub_pause_wait); @@ -1898,7 +2343,7 @@ static noinline_for_stack int scrub_stripe(struct scrub_dev *sdev, * canceled? */ if (atomic_read(&fs_info->scrub_cancel_req) || - atomic_read(&sdev->cancel_req)) { + atomic_read(&sctx->cancel_req)) { ret = -ECANCELED; goto out; } @@ -1907,9 +2352,14 @@ static noinline_for_stack int scrub_stripe(struct scrub_dev *sdev, */ if (atomic_read(&fs_info->scrub_pause_req)) { /* push queued extents */ - scrub_submit(sdev); - wait_event(sdev->list_wait, - atomic_read(&sdev->in_flight) == 0); + atomic_set(&sctx->wr_ctx.flush_all_writes, 1); + scrub_submit(sctx); + mutex_lock(&sctx->wr_ctx.wr_lock); + scrub_wr_submit(sctx); + mutex_unlock(&sctx->wr_ctx.wr_lock); + wait_event(sctx->list_wait, + atomic_read(&sctx->bios_in_flight) == 0); + atomic_set(&sctx->wr_ctx.flush_all_writes, 0); atomic_inc(&fs_info->scrubs_paused); wake_up(&fs_info->scrub_pause_wait); mutex_lock(&fs_info->scrub_lock); @@ -1926,7 +2376,7 @@ static noinline_for_stack int scrub_stripe(struct scrub_dev *sdev, ret = btrfs_lookup_csums_range(csum_root, logical, logical + map->stripe_len - 1, - &sdev->csum_list, 1); + &sctx->csum_list, 1); if (ret) goto out; @@ -2004,9 +2454,20 @@ static noinline_for_stack int scrub_stripe(struct scrub_dev *sdev, key.objectid; } - ret = scrub_extent(sdev, key.objectid, key.offset, - key.objectid - logical + physical, - flags, generation, mirror_num); + extent_logical = key.objectid; + extent_physical = key.objectid - logical + physical; + extent_len = key.offset; + extent_dev = scrub_dev; + extent_mirror_num = mirror_num; + if (is_dev_replace) + scrub_remap_extent(fs_info, extent_logical, + extent_len, &extent_physical, + &extent_dev, + &extent_mirror_num); + ret = scrub_extent(sctx, extent_logical, extent_len, + extent_physical, extent_dev, flags, + generation, extent_mirror_num, + key.objectid - logical + physical); if (ret) goto out; @@ -2016,29 +2477,34 @@ next: btrfs_release_path(path); logical += increment; physical += map->stripe_len; - spin_lock(&sdev->stat_lock); - sdev->stat.last_physical = physical; - spin_unlock(&sdev->stat_lock); + spin_lock(&sctx->stat_lock); + sctx->stat.last_physical = physical; + spin_unlock(&sctx->stat_lock); } +out: /* push queued extents */ - scrub_submit(sdev); + scrub_submit(sctx); + mutex_lock(&sctx->wr_ctx.wr_lock); + scrub_wr_submit(sctx); + mutex_unlock(&sctx->wr_ctx.wr_lock); -out: blk_finish_plug(&plug); btrfs_free_path(path); return ret < 0 ? ret : 0; } -static noinline_for_stack int scrub_chunk(struct scrub_dev *sdev, - u64 chunk_tree, u64 chunk_objectid, u64 chunk_offset, u64 length, - u64 dev_offset) +static noinline_for_stack int scrub_chunk(struct scrub_ctx *sctx, + struct btrfs_device *scrub_dev, + u64 chunk_tree, u64 chunk_objectid, + u64 chunk_offset, u64 length, + u64 dev_offset, int is_dev_replace) { struct btrfs_mapping_tree *map_tree = - &sdev->dev->dev_root->fs_info->mapping_tree; + &sctx->dev_root->fs_info->mapping_tree; struct map_lookup *map; struct extent_map *em; int i; - int ret = -EINVAL; + int ret = 0; read_lock(&map_tree->map_tree.lock); em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); @@ -2055,9 +2521,11 @@ static noinline_for_stack int scrub_chunk(struct scrub_dev *sdev, goto out; for (i = 0; i < map->num_stripes; ++i) { - if (map->stripes[i].dev == sdev->dev && + if (map->stripes[i].dev->bdev == scrub_dev->bdev && map->stripes[i].physical == dev_offset) { - ret = scrub_stripe(sdev, map, i, chunk_offset, length); + ret = scrub_stripe(sctx, map, scrub_dev, i, + chunk_offset, length, + is_dev_replace); if (ret) goto out; } @@ -2069,11 +2537,13 @@ out: } static noinline_for_stack -int scrub_enumerate_chunks(struct scrub_dev *sdev, u64 start, u64 end) +int scrub_enumerate_chunks(struct scrub_ctx *sctx, + struct btrfs_device *scrub_dev, u64 start, u64 end, + int is_dev_replace) { struct btrfs_dev_extent *dev_extent = NULL; struct btrfs_path *path; - struct btrfs_root *root = sdev->dev->dev_root; + struct btrfs_root *root = sctx->dev_root; struct btrfs_fs_info *fs_info = root->fs_info; u64 length; u64 chunk_tree; @@ -2085,6 +2555,7 @@ int scrub_enumerate_chunks(struct scrub_dev *sdev, u64 start, u64 end) struct btrfs_key key; struct btrfs_key found_key; struct btrfs_block_group_cache *cache; + struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; path = btrfs_alloc_path(); if (!path) @@ -2094,11 +2565,10 @@ int scrub_enumerate_chunks(struct scrub_dev *sdev, u64 start, u64 end) path->search_commit_root = 1; path->skip_locking = 1; - key.objectid = sdev->dev->devid; + key.objectid = scrub_dev->devid; key.offset = 0ull; key.type = BTRFS_DEV_EXTENT_KEY; - while (1) { ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); if (ret < 0) @@ -2117,7 +2587,7 @@ int scrub_enumerate_chunks(struct scrub_dev *sdev, u64 start, u64 end) btrfs_item_key_to_cpu(l, &found_key, slot); - if (found_key.objectid != sdev->dev->devid) + if (found_key.objectid != scrub_dev->devid) break; if (btrfs_key_type(&found_key) != BTRFS_DEV_EXTENT_KEY) @@ -2151,11 +2621,62 @@ int scrub_enumerate_chunks(struct scrub_dev *sdev, u64 start, u64 end) ret = -ENOENT; break; } - ret = scrub_chunk(sdev, chunk_tree, chunk_objectid, - chunk_offset, length, found_key.offset); + dev_replace->cursor_right = found_key.offset + length; + dev_replace->cursor_left = found_key.offset; + dev_replace->item_needs_writeback = 1; + ret = scrub_chunk(sctx, scrub_dev, chunk_tree, chunk_objectid, + chunk_offset, length, found_key.offset, + is_dev_replace); + + /* + * flush, submit all pending read and write bios, afterwards + * wait for them. + * Note that in the dev replace case, a read request causes + * write requests that are submitted in the read completion + * worker. Therefore in the current situation, it is required + * that all write requests are flushed, so that all read and + * write requests are really completed when bios_in_flight + * changes to 0. + */ + atomic_set(&sctx->wr_ctx.flush_all_writes, 1); + scrub_submit(sctx); + mutex_lock(&sctx->wr_ctx.wr_lock); + scrub_wr_submit(sctx); + mutex_unlock(&sctx->wr_ctx.wr_lock); + + wait_event(sctx->list_wait, + atomic_read(&sctx->bios_in_flight) == 0); + atomic_set(&sctx->wr_ctx.flush_all_writes, 0); + atomic_inc(&fs_info->scrubs_paused); + wake_up(&fs_info->scrub_pause_wait); + wait_event(sctx->list_wait, + atomic_read(&sctx->workers_pending) == 0); + + mutex_lock(&fs_info->scrub_lock); + while (atomic_read(&fs_info->scrub_pause_req)) { + mutex_unlock(&fs_info->scrub_lock); + wait_event(fs_info->scrub_pause_wait, + atomic_read(&fs_info->scrub_pause_req) == 0); + mutex_lock(&fs_info->scrub_lock); + } + atomic_dec(&fs_info->scrubs_paused); + mutex_unlock(&fs_info->scrub_lock); + wake_up(&fs_info->scrub_pause_wait); + + dev_replace->cursor_left = dev_replace->cursor_right; + dev_replace->item_needs_writeback = 1; btrfs_put_block_group(cache); if (ret) break; + if (is_dev_replace && + atomic64_read(&dev_replace->num_write_errors) > 0) { + ret = -EIO; + break; + } + if (sctx->stat.malloc_errors > 0) { + ret = -ENOMEM; + break; + } key.offset = found_key.offset + length; btrfs_release_path(path); @@ -2170,14 +2691,14 @@ int scrub_enumerate_chunks(struct scrub_dev *sdev, u64 start, u64 end) return ret < 0 ? ret : 0; } -static noinline_for_stack int scrub_supers(struct scrub_dev *sdev) +static noinline_for_stack int scrub_supers(struct scrub_ctx *sctx, + struct btrfs_device *scrub_dev) { int i; u64 bytenr; u64 gen; int ret; - struct btrfs_device *device = sdev->dev; - struct btrfs_root *root = device->dev_root; + struct btrfs_root *root = sctx->dev_root; if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) return -EIO; @@ -2186,15 +2707,16 @@ static noinline_for_stack int scrub_supers(struct scrub_dev *sdev) for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { bytenr = btrfs_sb_offset(i); - if (bytenr + BTRFS_SUPER_INFO_SIZE > device->total_bytes) + if (bytenr + BTRFS_SUPER_INFO_SIZE > scrub_dev->total_bytes) break; - ret = scrub_pages(sdev, bytenr, BTRFS_SUPER_INFO_SIZE, bytenr, - BTRFS_EXTENT_FLAG_SUPER, gen, i, NULL, 1); + ret = scrub_pages(sctx, bytenr, BTRFS_SUPER_INFO_SIZE, bytenr, + scrub_dev, BTRFS_EXTENT_FLAG_SUPER, gen, i, + NULL, 1, bytenr); if (ret) return ret; } - wait_event(sdev->list_wait, atomic_read(&sdev->in_flight) == 0); + wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0); return 0; } @@ -2202,19 +2724,38 @@ static noinline_for_stack int scrub_supers(struct scrub_dev *sdev) /* * get a reference count on fs_info->scrub_workers. start worker if necessary */ -static noinline_for_stack int scrub_workers_get(struct btrfs_root *root) +static noinline_for_stack int scrub_workers_get(struct btrfs_fs_info *fs_info, + int is_dev_replace) { - struct btrfs_fs_info *fs_info = root->fs_info; int ret = 0; mutex_lock(&fs_info->scrub_lock); if (fs_info->scrub_workers_refcnt == 0) { - btrfs_init_workers(&fs_info->scrub_workers, "scrub", - fs_info->thread_pool_size, &fs_info->generic_worker); + if (is_dev_replace) + btrfs_init_workers(&fs_info->scrub_workers, "scrub", 1, + &fs_info->generic_worker); + else + btrfs_init_workers(&fs_info->scrub_workers, "scrub", + fs_info->thread_pool_size, + &fs_info->generic_worker); fs_info->scrub_workers.idle_thresh = 4; ret = btrfs_start_workers(&fs_info->scrub_workers); if (ret) goto out; + btrfs_init_workers(&fs_info->scrub_wr_completion_workers, + "scrubwrc", + fs_info->thread_pool_size, + &fs_info->generic_worker); + fs_info->scrub_wr_completion_workers.idle_thresh = 2; + ret = btrfs_start_workers( + &fs_info->scrub_wr_completion_workers); + if (ret) + goto out; + btrfs_init_workers(&fs_info->scrub_nocow_workers, "scrubnc", 1, + &fs_info->generic_worker); + ret = btrfs_start_workers(&fs_info->scrub_nocow_workers); + if (ret) + goto out; } ++fs_info->scrub_workers_refcnt; out: @@ -2223,40 +2764,41 @@ out: return ret; } -static noinline_for_stack void scrub_workers_put(struct btrfs_root *root) +static noinline_for_stack void scrub_workers_put(struct btrfs_fs_info *fs_info) { - struct btrfs_fs_info *fs_info = root->fs_info; - mutex_lock(&fs_info->scrub_lock); - if (--fs_info->scrub_workers_refcnt == 0) + if (--fs_info->scrub_workers_refcnt == 0) { btrfs_stop_workers(&fs_info->scrub_workers); + btrfs_stop_workers(&fs_info->scrub_wr_completion_workers); + btrfs_stop_workers(&fs_info->scrub_nocow_workers); + } WARN_ON(fs_info->scrub_workers_refcnt < 0); mutex_unlock(&fs_info->scrub_lock); } - -int btrfs_scrub_dev(struct btrfs_root *root, u64 devid, u64 start, u64 end, - struct btrfs_scrub_progress *progress, int readonly) +int btrfs_scrub_dev(struct btrfs_fs_info *fs_info, u64 devid, u64 start, + u64 end, struct btrfs_scrub_progress *progress, + int readonly, int is_dev_replace) { - struct scrub_dev *sdev; - struct btrfs_fs_info *fs_info = root->fs_info; + struct scrub_ctx *sctx; int ret; struct btrfs_device *dev; - if (btrfs_fs_closing(root->fs_info)) + if (btrfs_fs_closing(fs_info)) return -EINVAL; /* * check some assumptions */ - if (root->nodesize != root->leafsize) { + if (fs_info->chunk_root->nodesize != fs_info->chunk_root->leafsize) { printk(KERN_ERR "btrfs_scrub: size assumption nodesize == leafsize (%d == %d) fails\n", - root->nodesize, root->leafsize); + fs_info->chunk_root->nodesize, + fs_info->chunk_root->leafsize); return -EINVAL; } - if (root->nodesize > BTRFS_STRIPE_LEN) { + if (fs_info->chunk_root->nodesize > BTRFS_STRIPE_LEN) { /* * in this case scrub is unable to calculate the checksum * the way scrub is implemented. Do not handle this @@ -2264,80 +2806,105 @@ int btrfs_scrub_dev(struct btrfs_root *root, u64 devid, u64 start, u64 end, */ printk(KERN_ERR "btrfs_scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails\n", - root->nodesize, BTRFS_STRIPE_LEN); + fs_info->chunk_root->nodesize, BTRFS_STRIPE_LEN); return -EINVAL; } - if (root->sectorsize != PAGE_SIZE) { + if (fs_info->chunk_root->sectorsize != PAGE_SIZE) { /* not supported for data w/o checksums */ printk(KERN_ERR "btrfs_scrub: size assumption sectorsize != PAGE_SIZE (%d != %lld) fails\n", - root->sectorsize, (unsigned long long)PAGE_SIZE); + fs_info->chunk_root->sectorsize, + (unsigned long long)PAGE_SIZE); return -EINVAL; } - ret = scrub_workers_get(root); + if (fs_info->chunk_root->nodesize > + PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK || + fs_info->chunk_root->sectorsize > + PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK) { + /* + * would exhaust the array bounds of pagev member in + * struct scrub_block + */ + pr_err("btrfs_scrub: size assumption nodesize and sectorsize <= SCRUB_MAX_PAGES_PER_BLOCK (%d <= %d && %d <= %d) fails\n", + fs_info->chunk_root->nodesize, + SCRUB_MAX_PAGES_PER_BLOCK, + fs_info->chunk_root->sectorsize, + SCRUB_MAX_PAGES_PER_BLOCK); + return -EINVAL; + } + + ret = scrub_workers_get(fs_info, is_dev_replace); if (ret) return ret; - mutex_lock(&root->fs_info->fs_devices->device_list_mutex); - dev = btrfs_find_device(root, devid, NULL, NULL); - if (!dev || dev->missing) { - mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); - scrub_workers_put(root); + mutex_lock(&fs_info->fs_devices->device_list_mutex); + dev = btrfs_find_device(fs_info, devid, NULL, NULL); + if (!dev || (dev->missing && !is_dev_replace)) { + mutex_unlock(&fs_info->fs_devices->device_list_mutex); + scrub_workers_put(fs_info); return -ENODEV; } mutex_lock(&fs_info->scrub_lock); - if (!dev->in_fs_metadata) { + if (!dev->in_fs_metadata || dev->is_tgtdev_for_dev_replace) { mutex_unlock(&fs_info->scrub_lock); - mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); - scrub_workers_put(root); - return -ENODEV; + mutex_unlock(&fs_info->fs_devices->device_list_mutex); + scrub_workers_put(fs_info); + return -EIO; } - if (dev->scrub_device) { + btrfs_dev_replace_lock(&fs_info->dev_replace); + if (dev->scrub_device || + (!is_dev_replace && + btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))) { + btrfs_dev_replace_unlock(&fs_info->dev_replace); mutex_unlock(&fs_info->scrub_lock); - mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); - scrub_workers_put(root); + mutex_unlock(&fs_info->fs_devices->device_list_mutex); + scrub_workers_put(fs_info); return -EINPROGRESS; } - sdev = scrub_setup_dev(dev); - if (IS_ERR(sdev)) { + btrfs_dev_replace_unlock(&fs_info->dev_replace); + sctx = scrub_setup_ctx(dev, is_dev_replace); + if (IS_ERR(sctx)) { mutex_unlock(&fs_info->scrub_lock); - mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); - scrub_workers_put(root); - return PTR_ERR(sdev); + mutex_unlock(&fs_info->fs_devices->device_list_mutex); + scrub_workers_put(fs_info); + return PTR_ERR(sctx); } - sdev->readonly = readonly; - dev->scrub_device = sdev; + sctx->readonly = readonly; + dev->scrub_device = sctx; atomic_inc(&fs_info->scrubs_running); mutex_unlock(&fs_info->scrub_lock); - mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); + mutex_unlock(&fs_info->fs_devices->device_list_mutex); - down_read(&fs_info->scrub_super_lock); - ret = scrub_supers(sdev); - up_read(&fs_info->scrub_super_lock); + if (!is_dev_replace) { + down_read(&fs_info->scrub_super_lock); + ret = scrub_supers(sctx, dev); + up_read(&fs_info->scrub_super_lock); + } if (!ret) - ret = scrub_enumerate_chunks(sdev, start, end); + ret = scrub_enumerate_chunks(sctx, dev, start, end, + is_dev_replace); - wait_event(sdev->list_wait, atomic_read(&sdev->in_flight) == 0); + wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0); atomic_dec(&fs_info->scrubs_running); wake_up(&fs_info->scrub_pause_wait); - wait_event(sdev->list_wait, atomic_read(&sdev->fixup_cnt) == 0); + wait_event(sctx->list_wait, atomic_read(&sctx->workers_pending) == 0); if (progress) - memcpy(progress, &sdev->stat, sizeof(*progress)); + memcpy(progress, &sctx->stat, sizeof(*progress)); mutex_lock(&fs_info->scrub_lock); dev->scrub_device = NULL; mutex_unlock(&fs_info->scrub_lock); - scrub_free_dev(sdev); - scrub_workers_put(root); + scrub_free_ctx(sctx); + scrub_workers_put(fs_info); return ret; } @@ -2377,9 +2944,8 @@ void btrfs_scrub_continue_super(struct btrfs_root *root) up_write(&root->fs_info->scrub_super_lock); } -int __btrfs_scrub_cancel(struct btrfs_fs_info *fs_info) +int btrfs_scrub_cancel(struct btrfs_fs_info *fs_info) { - mutex_lock(&fs_info->scrub_lock); if (!atomic_read(&fs_info->scrubs_running)) { mutex_unlock(&fs_info->scrub_lock); @@ -2399,23 +2965,18 @@ int __btrfs_scrub_cancel(struct btrfs_fs_info *fs_info) return 0; } -int btrfs_scrub_cancel(struct btrfs_root *root) +int btrfs_scrub_cancel_dev(struct btrfs_fs_info *fs_info, + struct btrfs_device *dev) { - return __btrfs_scrub_cancel(root->fs_info); -} - -int btrfs_scrub_cancel_dev(struct btrfs_root *root, struct btrfs_device *dev) -{ - struct btrfs_fs_info *fs_info = root->fs_info; - struct scrub_dev *sdev; + struct scrub_ctx *sctx; mutex_lock(&fs_info->scrub_lock); - sdev = dev->scrub_device; - if (!sdev) { + sctx = dev->scrub_device; + if (!sctx) { mutex_unlock(&fs_info->scrub_lock); return -ENOTCONN; } - atomic_inc(&sdev->cancel_req); + atomic_inc(&sctx->cancel_req); while (dev->scrub_device) { mutex_unlock(&fs_info->scrub_lock); wait_event(fs_info->scrub_pause_wait, @@ -2438,12 +2999,12 @@ int btrfs_scrub_cancel_devid(struct btrfs_root *root, u64 devid) * does not go away in cancel_dev. FIXME: find a better solution */ mutex_lock(&fs_info->fs_devices->device_list_mutex); - dev = btrfs_find_device(root, devid, NULL, NULL); + dev = btrfs_find_device(fs_info, devid, NULL, NULL); if (!dev) { mutex_unlock(&fs_info->fs_devices->device_list_mutex); return -ENODEV; } - ret = btrfs_scrub_cancel_dev(root, dev); + ret = btrfs_scrub_cancel_dev(fs_info, dev); mutex_unlock(&fs_info->fs_devices->device_list_mutex); return ret; @@ -2453,15 +3014,284 @@ int btrfs_scrub_progress(struct btrfs_root *root, u64 devid, struct btrfs_scrub_progress *progress) { struct btrfs_device *dev; - struct scrub_dev *sdev = NULL; + struct scrub_ctx *sctx = NULL; mutex_lock(&root->fs_info->fs_devices->device_list_mutex); - dev = btrfs_find_device(root, devid, NULL, NULL); + dev = btrfs_find_device(root->fs_info, devid, NULL, NULL); if (dev) - sdev = dev->scrub_device; - if (sdev) - memcpy(progress, &sdev->stat, sizeof(*progress)); + sctx = dev->scrub_device; + if (sctx) + memcpy(progress, &sctx->stat, sizeof(*progress)); mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); - return dev ? (sdev ? 0 : -ENOTCONN) : -ENODEV; + return dev ? (sctx ? 0 : -ENOTCONN) : -ENODEV; +} + +static void scrub_remap_extent(struct btrfs_fs_info *fs_info, + u64 extent_logical, u64 extent_len, + u64 *extent_physical, + struct btrfs_device **extent_dev, + int *extent_mirror_num) +{ + u64 mapped_length; + struct btrfs_bio *bbio = NULL; + int ret; + + mapped_length = extent_len; + ret = btrfs_map_block(fs_info, READ, extent_logical, + &mapped_length, &bbio, 0); + if (ret || !bbio || mapped_length < extent_len || + !bbio->stripes[0].dev->bdev) { + kfree(bbio); + return; + } + + *extent_physical = bbio->stripes[0].physical; + *extent_mirror_num = bbio->mirror_num; + *extent_dev = bbio->stripes[0].dev; + kfree(bbio); +} + +static int scrub_setup_wr_ctx(struct scrub_ctx *sctx, + struct scrub_wr_ctx *wr_ctx, + struct btrfs_fs_info *fs_info, + struct btrfs_device *dev, + int is_dev_replace) +{ + WARN_ON(wr_ctx->wr_curr_bio != NULL); + + mutex_init(&wr_ctx->wr_lock); + wr_ctx->wr_curr_bio = NULL; + if (!is_dev_replace) + return 0; + + WARN_ON(!dev->bdev); + wr_ctx->pages_per_wr_bio = min_t(int, SCRUB_PAGES_PER_WR_BIO, + bio_get_nr_vecs(dev->bdev)); + wr_ctx->tgtdev = dev; + atomic_set(&wr_ctx->flush_all_writes, 0); + return 0; +} + +static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx) +{ + mutex_lock(&wr_ctx->wr_lock); + kfree(wr_ctx->wr_curr_bio); + wr_ctx->wr_curr_bio = NULL; + mutex_unlock(&wr_ctx->wr_lock); +} + +static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len, + int mirror_num, u64 physical_for_dev_replace) +{ + struct scrub_copy_nocow_ctx *nocow_ctx; + struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; + + nocow_ctx = kzalloc(sizeof(*nocow_ctx), GFP_NOFS); + if (!nocow_ctx) { + spin_lock(&sctx->stat_lock); + sctx->stat.malloc_errors++; + spin_unlock(&sctx->stat_lock); + return -ENOMEM; + } + + scrub_pending_trans_workers_inc(sctx); + + nocow_ctx->sctx = sctx; + nocow_ctx->logical = logical; + nocow_ctx->len = len; + nocow_ctx->mirror_num = mirror_num; + nocow_ctx->physical_for_dev_replace = physical_for_dev_replace; + nocow_ctx->work.func = copy_nocow_pages_worker; + btrfs_queue_worker(&fs_info->scrub_nocow_workers, + &nocow_ctx->work); + + return 0; +} + +static void copy_nocow_pages_worker(struct btrfs_work *work) +{ + struct scrub_copy_nocow_ctx *nocow_ctx = + container_of(work, struct scrub_copy_nocow_ctx, work); + struct scrub_ctx *sctx = nocow_ctx->sctx; + u64 logical = nocow_ctx->logical; + u64 len = nocow_ctx->len; + int mirror_num = nocow_ctx->mirror_num; + u64 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace; + int ret; + struct btrfs_trans_handle *trans = NULL; + struct btrfs_fs_info *fs_info; + struct btrfs_path *path; + struct btrfs_root *root; + int not_written = 0; + + fs_info = sctx->dev_root->fs_info; + root = fs_info->extent_root; + + path = btrfs_alloc_path(); + if (!path) { + spin_lock(&sctx->stat_lock); + sctx->stat.malloc_errors++; + spin_unlock(&sctx->stat_lock); + not_written = 1; + goto out; + } + + trans = btrfs_join_transaction(root); + if (IS_ERR(trans)) { + not_written = 1; + goto out; + } + + ret = iterate_inodes_from_logical(logical, fs_info, path, + copy_nocow_pages_for_inode, + nocow_ctx); + if (ret != 0 && ret != -ENOENT) { + pr_warn("iterate_inodes_from_logical() failed: log %llu, phys %llu, len %llu, mir %llu, ret %d\n", + (unsigned long long)logical, + (unsigned long long)physical_for_dev_replace, + (unsigned long long)len, + (unsigned long long)mirror_num, ret); + not_written = 1; + goto out; + } + +out: + if (trans && !IS_ERR(trans)) + btrfs_end_transaction(trans, root); + if (not_written) + btrfs_dev_replace_stats_inc(&fs_info->dev_replace. + num_uncorrectable_read_errors); + + btrfs_free_path(path); + kfree(nocow_ctx); + + scrub_pending_trans_workers_dec(sctx); +} + +static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root, void *ctx) +{ + unsigned long index; + struct scrub_copy_nocow_ctx *nocow_ctx = ctx; + int ret = 0; + struct btrfs_key key; + struct inode *inode = NULL; + struct btrfs_root *local_root; + u64 physical_for_dev_replace; + u64 len; + struct btrfs_fs_info *fs_info = nocow_ctx->sctx->dev_root->fs_info; + + key.objectid = root; + key.type = BTRFS_ROOT_ITEM_KEY; + key.offset = (u64)-1; + local_root = btrfs_read_fs_root_no_name(fs_info, &key); + if (IS_ERR(local_root)) + return PTR_ERR(local_root); + + key.type = BTRFS_INODE_ITEM_KEY; + key.objectid = inum; + key.offset = 0; + inode = btrfs_iget(fs_info->sb, &key, local_root, NULL); + if (IS_ERR(inode)) + return PTR_ERR(inode); + + physical_for_dev_replace = nocow_ctx->physical_for_dev_replace; + len = nocow_ctx->len; + while (len >= PAGE_CACHE_SIZE) { + struct page *page = NULL; + int ret_sub; + + index = offset >> PAGE_CACHE_SHIFT; + + page = find_or_create_page(inode->i_mapping, index, GFP_NOFS); + if (!page) { + pr_err("find_or_create_page() failed\n"); + ret = -ENOMEM; + goto next_page; + } + + if (PageUptodate(page)) { + if (PageDirty(page)) + goto next_page; + } else { + ClearPageError(page); + ret_sub = extent_read_full_page(&BTRFS_I(inode)-> + io_tree, + page, btrfs_get_extent, + nocow_ctx->mirror_num); + if (ret_sub) { + ret = ret_sub; + goto next_page; + } + wait_on_page_locked(page); + if (!PageUptodate(page)) { + ret = -EIO; + goto next_page; + } + } + ret_sub = write_page_nocow(nocow_ctx->sctx, + physical_for_dev_replace, page); + if (ret_sub) { + ret = ret_sub; + goto next_page; + } + +next_page: + if (page) { + unlock_page(page); + put_page(page); + } + offset += PAGE_CACHE_SIZE; + physical_for_dev_replace += PAGE_CACHE_SIZE; + len -= PAGE_CACHE_SIZE; + } + + if (inode) + iput(inode); + return ret; +} + +static int write_page_nocow(struct scrub_ctx *sctx, + u64 physical_for_dev_replace, struct page *page) +{ + struct bio *bio; + struct btrfs_device *dev; + int ret; + DECLARE_COMPLETION_ONSTACK(compl); + + dev = sctx->wr_ctx.tgtdev; + if (!dev) + return -EIO; + if (!dev->bdev) { + printk_ratelimited(KERN_WARNING + "btrfs: scrub write_page_nocow(bdev == NULL) is unexpected!\n"); + return -EIO; + } + bio = bio_alloc(GFP_NOFS, 1); + if (!bio) { + spin_lock(&sctx->stat_lock); + sctx->stat.malloc_errors++; + spin_unlock(&sctx->stat_lock); + return -ENOMEM; + } + bio->bi_private = &compl; + bio->bi_end_io = scrub_complete_bio_end_io; + bio->bi_size = 0; + bio->bi_sector = physical_for_dev_replace >> 9; + bio->bi_bdev = dev->bdev; + ret = bio_add_page(bio, page, PAGE_CACHE_SIZE, 0); + if (ret != PAGE_CACHE_SIZE) { +leave_with_eio: + bio_put(bio); + btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); + return -EIO; + } + btrfsic_submit_bio(WRITE_SYNC, bio); + wait_for_completion(&compl); + + if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) + goto leave_with_eio; + + bio_put(bio); + return 0; } |