From 7ffe3de53a885dbb5836541c2178bd07d1bad7df Mon Sep 17 00:00:00 2001 From: Davidlohr Bueso Date: Thu, 17 Apr 2025 18:59:15 -0700 Subject: fs/buffer: split locking for pagecache lookups Callers of __find_get_block() may or may not allow for blocking semantics, and is currently assumed that it will not. Layout two paths based on this. The the private_lock scheme will continued to be used for atomic contexts. Otherwise take the folio lock instead, which protects the buffers, such as vs migration and try_to_free_buffers(). Per the "hack idea", the latter can alleviate contention on the private_lock for bdev mappings. For reasons of determinism and avoid making bugs hard to reproduce, the trylocking is not attempted. No change in semantics. All lookup users still take the spinlock. Reviewed-by: Jan Kara Signed-off-by: Davidlohr Bueso Link: https://kdevops.org/ext4/v6.15-rc2.html # [0] Link: https://lore.kernel.org/all/aAAEvcrmREWa1SKF@bombadil.infradead.org/ # [1] Link: https://lore.kernel.org/20250418015921.132400-2-dave@stgolabs.net Tested-by: kdevops@lists.linux.dev Reviewed-by: Luis Chamberlain Signed-off-by: Christian Brauner --- fs/buffer.c | 41 +++++++++++++++++++++++++---------------- 1 file changed, 25 insertions(+), 16 deletions(-) (limited to 'fs/buffer.c') diff --git a/fs/buffer.c b/fs/buffer.c index c7abb4a029dc..f8fcffdbe5d9 100644 --- a/fs/buffer.c +++ b/fs/buffer.c @@ -176,18 +176,8 @@ void end_buffer_write_sync(struct buffer_head *bh, int uptodate) } EXPORT_SYMBOL(end_buffer_write_sync); -/* - * Various filesystems appear to want __find_get_block to be non-blocking. - * But it's the page lock which protects the buffers. To get around this, - * we get exclusion from try_to_free_buffers with the blockdev mapping's - * i_private_lock. - * - * Hack idea: for the blockdev mapping, i_private_lock contention - * may be quite high. This code could TryLock the page, and if that - * succeeds, there is no need to take i_private_lock. - */ static struct buffer_head * -__find_get_block_slow(struct block_device *bdev, sector_t block) +__find_get_block_slow(struct block_device *bdev, sector_t block, bool atomic) { struct address_space *bd_mapping = bdev->bd_mapping; const int blkbits = bd_mapping->host->i_blkbits; @@ -204,7 +194,16 @@ __find_get_block_slow(struct block_device *bdev, sector_t block) if (IS_ERR(folio)) goto out; - spin_lock(&bd_mapping->i_private_lock); + /* + * Folio lock protects the buffers. Callers that cannot block + * will fallback to serializing vs try_to_free_buffers() via + * the i_private_lock. + */ + if (atomic) + spin_lock(&bd_mapping->i_private_lock); + else + folio_lock(folio); + head = folio_buffers(folio); if (!head) goto out_unlock; @@ -236,7 +235,10 @@ __find_get_block_slow(struct block_device *bdev, sector_t block) 1 << blkbits); } out_unlock: - spin_unlock(&bd_mapping->i_private_lock); + if (atomic) + spin_unlock(&bd_mapping->i_private_lock); + else + folio_unlock(folio); folio_put(folio); out: return ret; @@ -1388,14 +1390,15 @@ lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size) * it in the LRU and mark it as accessed. If it is not present then return * NULL */ -struct buffer_head * -__find_get_block(struct block_device *bdev, sector_t block, unsigned size) +static struct buffer_head * +find_get_block_common(struct block_device *bdev, sector_t block, + unsigned size, bool atomic) { struct buffer_head *bh = lookup_bh_lru(bdev, block, size); if (bh == NULL) { /* __find_get_block_slow will mark the page accessed */ - bh = __find_get_block_slow(bdev, block); + bh = __find_get_block_slow(bdev, block, atomic); if (bh) bh_lru_install(bh); } else @@ -1403,6 +1406,12 @@ __find_get_block(struct block_device *bdev, sector_t block, unsigned size) return bh; } + +struct buffer_head * +__find_get_block(struct block_device *bdev, sector_t block, unsigned size) +{ + return find_get_block_common(bdev, block, size, true); +} EXPORT_SYMBOL(__find_get_block); /** -- cgit v1.2.3 From 2814a7d3d2ff5d2cdd22936f641f758fdb971fa0 Mon Sep 17 00:00:00 2001 From: Davidlohr Bueso Date: Thu, 17 Apr 2025 18:59:16 -0700 Subject: fs/buffer: introduce sleeping flavors for pagecache lookups Add __find_get_block_nonatomic() and sb_find_get_block_nonatomic() calls for which users will be converted where safe. These versions will take the folio lock instead of the mapping's private_lock. Reviewed-by: Jan Kara Signed-off-by: Davidlohr Bueso Link: https://kdevops.org/ext4/v6.15-rc2.html # [0] Link: https://lore.kernel.org/all/aAAEvcrmREWa1SKF@bombadil.infradead.org/ # [1] Link: https://lore.kernel.org/20250418015921.132400-3-dave@stgolabs.net Tested-by: kdevops@lists.linux.dev Reviewed-by: Luis Chamberlain Signed-off-by: Christian Brauner --- fs/buffer.c | 9 +++++++++ include/linux/buffer_head.h | 8 ++++++++ 2 files changed, 17 insertions(+) (limited to 'fs/buffer.c') diff --git a/fs/buffer.c b/fs/buffer.c index f8fcffdbe5d9..5b1d74c818e9 100644 --- a/fs/buffer.c +++ b/fs/buffer.c @@ -1414,6 +1414,15 @@ __find_get_block(struct block_device *bdev, sector_t block, unsigned size) } EXPORT_SYMBOL(__find_get_block); +/* same as __find_get_block() but allows sleeping contexts */ +struct buffer_head * +__find_get_block_nonatomic(struct block_device *bdev, sector_t block, + unsigned size) +{ + return find_get_block_common(bdev, block, size, false); +} +EXPORT_SYMBOL(__find_get_block_nonatomic); + /** * bdev_getblk - Get a buffer_head in a block device's buffer cache. * @bdev: The block device. diff --git a/include/linux/buffer_head.h b/include/linux/buffer_head.h index f0a4ad7839b6..c791aa9a08da 100644 --- a/include/linux/buffer_head.h +++ b/include/linux/buffer_head.h @@ -222,6 +222,8 @@ void __wait_on_buffer(struct buffer_head *); wait_queue_head_t *bh_waitq_head(struct buffer_head *bh); struct buffer_head *__find_get_block(struct block_device *bdev, sector_t block, unsigned size); +struct buffer_head *__find_get_block_nonatomic(struct block_device *bdev, + sector_t block, unsigned size); struct buffer_head *bdev_getblk(struct block_device *bdev, sector_t block, unsigned size, gfp_t gfp); void __brelse(struct buffer_head *); @@ -397,6 +399,12 @@ sb_find_get_block(struct super_block *sb, sector_t block) return __find_get_block(sb->s_bdev, block, sb->s_blocksize); } +static inline struct buffer_head * +sb_find_get_block_nonatomic(struct super_block *sb, sector_t block) +{ + return __find_get_block_nonatomic(sb->s_bdev, block, sb->s_blocksize); +} + static inline void map_bh(struct buffer_head *bh, struct super_block *sb, sector_t block) { -- cgit v1.2.3 From 5b67d43976828dea2394eae2556b369bb7a61f64 Mon Sep 17 00:00:00 2001 From: Davidlohr Bueso Date: Thu, 17 Apr 2025 18:59:17 -0700 Subject: fs/buffer: use sleeping version of __find_get_block() Convert to the new nonatomic flavor to benefit from potential performance benefits and adapt in the future vs migration such that semantics are kept. Convert write_boundary_block() which already takes the buffer lock as well as bdev_getblk() depending on the respective gpf flags. There are no changes in semantics. Suggested-by: Jan Kara Reviewed-by: Jan Kara Signed-off-by: Davidlohr Bueso Link: https://kdevops.org/ext4/v6.15-rc2.html # [0] Link: https://lore.kernel.org/all/aAAEvcrmREWa1SKF@bombadil.infradead.org/ # [1] Link: https://lore.kernel.org/20250418015921.132400-4-dave@stgolabs.net Tested-by: kdevops@lists.linux.dev # [0] [1] Reviewed-by: Luis Chamberlain Signed-off-by: Christian Brauner --- fs/buffer.c | 11 +++++++++-- 1 file changed, 9 insertions(+), 2 deletions(-) (limited to 'fs/buffer.c') diff --git a/fs/buffer.c b/fs/buffer.c index 5b1d74c818e9..f8c9e5eb4685 100644 --- a/fs/buffer.c +++ b/fs/buffer.c @@ -658,7 +658,9 @@ EXPORT_SYMBOL(generic_buffers_fsync); void write_boundary_block(struct block_device *bdev, sector_t bblock, unsigned blocksize) { - struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize); + struct buffer_head *bh; + + bh = __find_get_block_nonatomic(bdev, bblock + 1, blocksize); if (bh) { if (buffer_dirty(bh)) write_dirty_buffer(bh, 0); @@ -1440,7 +1442,12 @@ EXPORT_SYMBOL(__find_get_block_nonatomic); struct buffer_head *bdev_getblk(struct block_device *bdev, sector_t block, unsigned size, gfp_t gfp) { - struct buffer_head *bh = __find_get_block(bdev, block, size); + struct buffer_head *bh; + + if (gfpflags_allow_blocking(gfp)) + bh = __find_get_block_nonatomic(bdev, block, size); + else + bh = __find_get_block(bdev, block, size); might_alloc(gfp); if (bh) -- cgit v1.2.3 From 2d900efff915fe24c3948d28eef9078953d87fec Mon Sep 17 00:00:00 2001 From: Davidlohr Bueso Date: Thu, 17 Apr 2025 18:59:21 -0700 Subject: mm/migrate: fix sleep in atomic for large folios and buffer heads The large folio + buffer head noref migration scenarios are being naughty and blocking while holding a spinlock. As a consequence of the pagecache lookup path taking the folio lock this serializes against migration paths, so they can wait for each other. For the private_lock atomic case, a new BH_Migrate flag is introduced which enables the lookup to bail. This allows the critical region of the private_lock on the migration path to be reduced to the way it was before ebdf4de5642fb6 ("mm: migrate: fix reference check race between __find_get_block() and migration"), that is covering the count checks. The scope is always noref migration. Reported-by: kernel test robot Reported-by: syzbot+f3c6fda1297c748a7076@syzkaller.appspotmail.com Closes: https://lore.kernel.org/oe-lkp/202503101536.27099c77-lkp@intel.com Fixes: 3c20917120ce61 ("block/bdev: enable large folio support for large logical block sizes") Reviewed-by: Jan Kara Co-developed-by: Luis Chamberlain Signed-off-by: Davidlohr Bueso Link: https://kdevops.org/ext4/v6.15-rc2.html # [0] Link: https://lore.kernel.org/all/aAAEvcrmREWa1SKF@bombadil.infradead.org/ # [1] Link: https://lore.kernel.org/20250418015921.132400-8-dave@stgolabs.net Tested-by: kdevops@lists.linux.dev # [0] [1] Reviewed-by: Luis Chamberlain Signed-off-by: Christian Brauner --- fs/buffer.c | 12 +++++++++++- fs/ext4/ialloc.c | 3 ++- include/linux/buffer_head.h | 1 + mm/migrate.c | 8 +++++--- 4 files changed, 19 insertions(+), 5 deletions(-) (limited to 'fs/buffer.c') diff --git a/fs/buffer.c b/fs/buffer.c index f8c9e5eb4685..7be23ff20b27 100644 --- a/fs/buffer.c +++ b/fs/buffer.c @@ -207,6 +207,15 @@ __find_get_block_slow(struct block_device *bdev, sector_t block, bool atomic) head = folio_buffers(folio); if (!head) goto out_unlock; + /* + * Upon a noref migration, the folio lock serializes here; + * otherwise bail. + */ + if (test_bit_acquire(BH_Migrate, &head->b_state)) { + WARN_ON(!atomic); + goto out_unlock; + } + bh = head; do { if (!buffer_mapped(bh)) @@ -1390,7 +1399,8 @@ lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size) /* * Perform a pagecache lookup for the matching buffer. If it's there, refresh * it in the LRU and mark it as accessed. If it is not present then return - * NULL + * NULL. Atomic context callers may also return NULL if the buffer is being + * migrated; similarly the page is not marked accessed either. */ static struct buffer_head * find_get_block_common(struct block_device *bdev, sector_t block, diff --git a/fs/ext4/ialloc.c b/fs/ext4/ialloc.c index 38bc8d74f4cc..e7ecc7c8a729 100644 --- a/fs/ext4/ialloc.c +++ b/fs/ext4/ialloc.c @@ -691,7 +691,8 @@ static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino) if (!bh || !buffer_uptodate(bh)) /* * If the block is not in the buffer cache, then it - * must have been written out. + * must have been written out, or, most unlikely, is + * being migrated - false failure should be OK here. */ goto out; diff --git a/include/linux/buffer_head.h b/include/linux/buffer_head.h index c791aa9a08da..0029ff880e27 100644 --- a/include/linux/buffer_head.h +++ b/include/linux/buffer_head.h @@ -34,6 +34,7 @@ enum bh_state_bits { BH_Meta, /* Buffer contains metadata */ BH_Prio, /* Buffer should be submitted with REQ_PRIO */ BH_Defer_Completion, /* Defer AIO completion to workqueue */ + BH_Migrate, /* Buffer is being migrated (norefs) */ BH_PrivateStart,/* not a state bit, but the first bit available * for private allocation by other entities diff --git a/mm/migrate.c b/mm/migrate.c index f3ee6d8d5e2e..676d9cfc7059 100644 --- a/mm/migrate.c +++ b/mm/migrate.c @@ -845,9 +845,11 @@ static int __buffer_migrate_folio(struct address_space *mapping, return -EAGAIN; if (check_refs) { - bool busy; + bool busy, migrating; bool invalidated = false; + migrating = test_and_set_bit_lock(BH_Migrate, &head->b_state); + VM_WARN_ON_ONCE(migrating); recheck_buffers: busy = false; spin_lock(&mapping->i_private_lock); @@ -859,12 +861,12 @@ recheck_buffers: } bh = bh->b_this_page; } while (bh != head); + spin_unlock(&mapping->i_private_lock); if (busy) { if (invalidated) { rc = -EAGAIN; goto unlock_buffers; } - spin_unlock(&mapping->i_private_lock); invalidate_bh_lrus(); invalidated = true; goto recheck_buffers; @@ -883,7 +885,7 @@ recheck_buffers: unlock_buffers: if (check_refs) - spin_unlock(&mapping->i_private_lock); + clear_bit_unlock(BH_Migrate, &head->b_state); bh = head; do { unlock_buffer(bh); -- cgit v1.2.3