[PATCH 6.14 030/783] fs/buffer: split locking for pagecache lookups

[PATCH 6.14 030/783] fs/buffer: split locking for pagecache lookups

[Greg Kroah-Hartman]

6.14-stable review patch.  If anyone has any objections, please let me know.

------------------

From: Davidlohr Bueso <dave@stgolabs.net>

[ Upstream commit 7ffe3de53a885dbb5836541c2178bd07d1bad7df ]

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 <jack@suse.cz>
Signed-off-by: Davidlohr Bueso <dave@stgolabs.net>
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 <mcgrof@kernel.org>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 fs/buffer.c | 41 +++++++++++++++++++++++++----------------
 1 file changed, 25 insertions(+), 16 deletions(-)

diff --git a/fs/buffer.c b/fs/buffer.c
index cc8452f602516..a03c245022dcf 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);
 
 /**
-- 
2.39.5

Re: [PATCH 6.14 030/783] fs/buffer: split locking for pagecache lookups

[Luis Chamberlain]

On Tue, May 27, 2025 at 06:17:07PM +0200, Greg Kroah-Hartman wrote:
> 6.14-stable review patch.  If anyone has any objections, please let me know.
> 
> ------------------
> 
> From: Davidlohr Bueso <dave@stgolabs.net>
> 
> [ Upstream commit 7ffe3de53a885dbb5836541c2178bd07d1bad7df ]
> 
> 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.

As noted to Sasha before, one of these changes apply to stable kernels,
since buffer-heads lacked folio min order support until v6.15. So these
patches from Davidlohr are not fixing anything on older kernels.

  Luis

[PATCH 6.14 031/783] fs/buffer: introduce sleeping flavors for pagecache lookups

[Greg Kroah-Hartman]

6.14-stable review patch.  If anyone has any objections, please let me know.

------------------

From: Davidlohr Bueso <dave@stgolabs.net>

[ Upstream commit 2814a7d3d2ff5d2cdd22936f641f758fdb971fa0 ]

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 <jack@suse.cz>
Signed-off-by: Davidlohr Bueso <dave@stgolabs.net>
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 <mcgrof@kernel.org>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 fs/buffer.c                 | 9 +++++++++
 include/linux/buffer_head.h | 8 ++++++++
 2 files changed, 17 insertions(+)

diff --git a/fs/buffer.c b/fs/buffer.c
index a03c245022dcf..7981097c846d4 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 932139c5d46f5..ffcd76d977703 100644
--- a/include/linux/buffer_head.h
+++ b/include/linux/buffer_head.h
@@ -223,6 +223,8 @@ 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 *);
@@ -398,6 +400,12 @@ 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)
 {
-- 
2.39.5

[PATCH 6.14 032/783] fs/buffer: use sleeping version of __find_get_block()

[Greg Kroah-Hartman]

6.14-stable review patch.  If anyone has any objections, please let me know.

------------------

From: Davidlohr Bueso <dave@stgolabs.net>

[ Upstream commit 5b67d43976828dea2394eae2556b369bb7a61f64 ]

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 <jack@suse.cz>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Davidlohr Bueso <dave@stgolabs.net>
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 <mcgrof@kernel.org>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 fs/buffer.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/fs/buffer.c b/fs/buffer.c
index 7981097c846d4..2494fe3a5e69e 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)
-- 
2.39.5

[PATCH 6.14 033/783] fs/ocfs2: use sleeping version of __find_get_block()

[Greg Kroah-Hartman]

6.14-stable review patch.  If anyone has any objections, please let me know.

------------------

From: Davidlohr Bueso <dave@stgolabs.net>

[ Upstream commit a0b5ff07491010789fcb012bc8f9dad9d26f9a8b ]

This is a path that allows for blocking as it does IO. Convert
to the new nonatomic flavor to benefit from potential performance
benefits and adapt in the future vs migration such that semantics
are kept.

Suggested-by: Jan Kara <jack@suse.cz>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Davidlohr Bueso <dave@stgolabs.net>
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/20250418015921.132400-5-dave@stgolabs.net
Tested-by: kdevops@lists.linux.dev
Reviewed-by: Luis Chamberlain <mcgrof@kernel.org>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 fs/ocfs2/journal.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/ocfs2/journal.c b/fs/ocfs2/journal.c
index f37831d5f95a1..e5f58ff2175f4 100644
--- a/fs/ocfs2/journal.c
+++ b/fs/ocfs2/journal.c
@@ -1271,7 +1271,7 @@ static int ocfs2_force_read_journal(struct inode *inode)
 		}
 
 		for (i = 0; i < p_blocks; i++, p_blkno++) {
-			bh = __find_get_block(osb->sb->s_bdev, p_blkno,
+			bh = __find_get_block_nonatomic(osb->sb->s_bdev, p_blkno,
 					osb->sb->s_blocksize);
 			/* block not cached. */
 			if (!bh)
-- 
2.39.5

[PATCH 6.14 034/783] fs/jbd2: use sleeping version of __find_get_block()

[Greg Kroah-Hartman]

6.14-stable review patch.  If anyone has any objections, please let me know.

------------------

From: Davidlohr Bueso <dave@stgolabs.net>

[ Upstream commit f76d4c28a46a9260d85e00dafc8f46d369365d33 ]

Convert to the new nonatomic flavor to benefit from potential
performance benefits and adapt in the future vs migration such
that semantics are kept.

- jbd2_journal_revoke(): can sleep (has might_sleep() in the beginning)

- jbd2_journal_cancel_revoke(): only used from do_get_write_access() and
    do_get_create_access() which do sleep. So can sleep.

- jbd2_clear_buffer_revoked_flags() - only called from journal commit code
    which sleeps. So can sleep.

Suggested-by: Jan Kara <jack@suse.cz>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Davidlohr Bueso <dave@stgolabs.net>
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/20250418015921.132400-6-dave@stgolabs.net
Tested-by: kdevops@lists.linux.dev
Reviewed-by: Luis Chamberlain <mcgrof@kernel.org>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 fs/jbd2/revoke.c | 15 +++++++++------
 1 file changed, 9 insertions(+), 6 deletions(-)

diff --git a/fs/jbd2/revoke.c b/fs/jbd2/revoke.c
index ce63d5fde9c3a..f68fc8c255f00 100644
--- a/fs/jbd2/revoke.c
+++ b/fs/jbd2/revoke.c
@@ -345,7 +345,8 @@ int jbd2_journal_revoke(handle_t *handle, unsigned long long blocknr,
 	bh = bh_in;
 
 	if (!bh) {
-		bh = __find_get_block(bdev, blocknr, journal->j_blocksize);
+		bh = __find_get_block_nonatomic(bdev, blocknr,
+						journal->j_blocksize);
 		if (bh)
 			BUFFER_TRACE(bh, "found on hash");
 	}
@@ -355,7 +356,8 @@ int jbd2_journal_revoke(handle_t *handle, unsigned long long blocknr,
 
 		/* If there is a different buffer_head lying around in
 		 * memory anywhere... */
-		bh2 = __find_get_block(bdev, blocknr, journal->j_blocksize);
+		bh2 = __find_get_block_nonatomic(bdev, blocknr,
+						 journal->j_blocksize);
 		if (bh2) {
 			/* ... and it has RevokeValid status... */
 			if (bh2 != bh && buffer_revokevalid(bh2))
@@ -466,7 +468,8 @@ int jbd2_journal_cancel_revoke(handle_t *handle, struct journal_head *jh)
 	 * state machine will get very upset later on. */
 	if (need_cancel) {
 		struct buffer_head *bh2;
-		bh2 = __find_get_block(bh->b_bdev, bh->b_blocknr, bh->b_size);
+		bh2 = __find_get_block_nonatomic(bh->b_bdev, bh->b_blocknr,
+						 bh->b_size);
 		if (bh2) {
 			if (bh2 != bh)
 				clear_buffer_revoked(bh2);
@@ -495,9 +498,9 @@ void jbd2_clear_buffer_revoked_flags(journal_t *journal)
 			struct jbd2_revoke_record_s *record;
 			struct buffer_head *bh;
 			record = (struct jbd2_revoke_record_s *)list_entry;
-			bh = __find_get_block(journal->j_fs_dev,
-					      record->blocknr,
-					      journal->j_blocksize);
+			bh = __find_get_block_nonatomic(journal->j_fs_dev,
+							record->blocknr,
+							journal->j_blocksize);
 			if (bh) {
 				clear_buffer_revoked(bh);
 				__brelse(bh);
-- 
2.39.5

[PATCH 6.14 035/783] fs/ext4: use sleeping version of sb_find_get_block()

[Greg Kroah-Hartman]

6.14-stable review patch.  If anyone has any objections, please let me know.

------------------

From: Davidlohr Bueso <dave@stgolabs.net>

[ Upstream commit 6e8f57fd09c9fb569d10b2ccc3878155b702591a ]

Enable ext4_free_blocks() to use it, which has a cond_resched to begin
with. Convert to the new nonatomic flavor to benefit from potential
performance benefits and adapt in the future vs migration such that
semantics are kept.

Suggested-by: Jan Kara <jack@suse.cz>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Davidlohr Bueso <dave@stgolabs.net>
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/20250418015921.132400-7-dave@stgolabs.net
Tested-by: kdevops@lists.linux.dev
Reviewed-by: Luis Chamberlain <mcgrof@kernel.org>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 fs/ext4/mballoc.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
index b25a27c866969..d6f1e61c6dc82 100644
--- a/fs/ext4/mballoc.c
+++ b/fs/ext4/mballoc.c
@@ -6644,7 +6644,8 @@ void ext4_free_blocks(handle_t *handle, struct inode *inode,
 		for (i = 0; i < count; i++) {
 			cond_resched();
 			if (is_metadata)
-				bh = sb_find_get_block(inode->i_sb, block + i);
+				bh = sb_find_get_block_nonatomic(inode->i_sb,
+								 block + i);
 			ext4_forget(handle, is_metadata, inode, bh, block + i);
 		}
 	}
-- 
2.39.5