mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
* [PATCH v2] exfat: hold s_lock for cluster-chain updates in __exfat_cont_expand()
@ 2026-11-18  1:58 Youling Tang
  2026-11-18  3:22 ` Chi Zhiling
  0 siblings, 1 reply; 2+ messages in thread
From: Youling Tang @ 2026-11-18  1:58 UTC (permalink / raw)
  To: Namjae Jeon, Sungjong Seo
  Cc: Yuezhang Mo, Chi Zhiling, exfat, linux-kernel, youling.tang,
	Youling Tang

From: Youling Tang <tangyouling@kylinos.cn>

__exfat_cont_expand() allocates clusters, appends them to the FAT chain
and updates ei->start_clu, ei->flags and i_blocks without holding
sbi->s_lock.  Readers such as writeback call exfat_map_cluster() with
s_lock held but without inode_lock, so they are not serialized against
the expansion and can observe the chain metadata mid-update.

Hold s_lock around the cluster allocation, chain append and i_blocks
update.  This nests inside the invalidate_lock held by
exfat_cont_expand(), matching the order used by exfat_truncate().  i_size
is published via exfat_pagecache_isize_extended(), which takes the
straddling folio lock; since writeback takes folio lock -> s_lock, this
must happen after s_lock is released to avoid inverting the lock order.
i_blocks is therefore updated first, under s_lock, which also ensures a
reader never sees i_size beyond the allocated chain.

Signed-off-by: Youling Tang <tangyouling@kylinos.cn>
---
v2: rebase onto exfat/dev and adapt to __exfat_cont_expand().

 fs/exfat/file.c | 18 +++++++++++++-----
 1 file changed, 13 insertions(+), 5 deletions(-)

diff --git a/fs/exfat/file.c b/fs/exfat/file.c
index 9dfd99ad9265..c7991e98a07a 100644
--- a/fs/exfat/file.c
+++ b/fs/exfat/file.c
@@ -105,6 +105,7 @@ static int __exfat_cont_expand(struct inode *inode, loff_t size)
 	if (ret)
 		return ret;
 
+	mutex_lock(&sbi->s_lock);
 	num_clusters = exfat_bytes_to_cluster(sbi, exfat_ondisk_size(inode));
 	/* integer overflow is already checked in inode_newsize_ok(). */
 	new_num_clusters = exfat_bytes_to_cluster_round_up(sbi, size);
@@ -116,7 +117,7 @@ static int __exfat_cont_expand(struct inode *inode, loff_t size)
 		exfat_chain_set(&clu, ei->start_clu, num_clusters, ei->flags);
 		ret = exfat_find_last_cluster(sb, &clu, &last_clu);
 		if (ret)
-			return ret;
+			goto unlock;
 
 		clu.dir = last_clu + 1;
 	} else {
@@ -130,7 +131,7 @@ static int __exfat_cont_expand(struct inode *inode, loff_t size)
 	ret = exfat_alloc_cluster(inode, new_num_clusters - num_clusters,
 			&clu, inode_needs_sync(inode), false);
 	if (ret)
-		return ret;
+		goto unlock;
 
 	/* Append new clusters to chain */
 	if (num_clusters) {
@@ -147,15 +148,19 @@ static int __exfat_cont_expand(struct inode *inode, loff_t size)
 	ei->flags = clu.flags;
 
 out:
+	inode->i_blocks = round_up(size, sbi->cluster_size) >> 9;
 	inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
+	mutex_unlock(&sbi->s_lock);
+
 	/*
 	 * When extending file size, call exfat_pagecache_isize_extended()
 	 * to update i_size and ensure the straddling folio is properly
 	 * marked RO so page_mkwrite() is called and post-EOF area is zeroed.
+	 *
+	 * It takes the folio lock; call it without s_lock (writeback uses
+	 * folio lock -> s_lock).
 	 */
 	exfat_pagecache_isize_extended(inode, oldsize, size);
-
-	inode->i_blocks = round_up(size, sbi->cluster_size) >> 9;
 	mark_inode_dirty(inode);
 
 	if (IS_SYNC(inode))
@@ -165,7 +170,10 @@ static int __exfat_cont_expand(struct inode *inode, loff_t size)
 
 free_clu:
 	exfat_free_cluster(inode, &clu);
-	return -EIO;
+	ret = -EIO;
+unlock:
+	mutex_unlock(&sbi->s_lock);
+	return ret;
 }
 
 static int exfat_cont_expand(struct inode *inode, loff_t size)
-- 
2.43.0


^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: [PATCH v2] exfat: hold s_lock for cluster-chain updates in __exfat_cont_expand()
  2026-11-18  1:58 [PATCH v2] exfat: hold s_lock for cluster-chain updates in __exfat_cont_expand() Youling Tang
@ 2026-11-18  3:22 ` Chi Zhiling
  0 siblings, 0 replies; 2+ messages in thread
From: Chi Zhiling @ 2026-11-18  3:22 UTC (permalink / raw)
  To: Youling Tang, Namjae Jeon, Sungjong Seo
  Cc: Yuezhang Mo, exfat, linux-kernel, Youling Tang

On 11/18/26 09:58, Youling Tang wrote:
> From: Youling Tang <tangyouling@kylinos.cn>
> 
> __exfat_cont_expand() allocates clusters, appends them to the FAT chain
> and updates ei->start_clu, ei->flags and i_blocks without holding
> sbi->s_lock.  Readers such as writeback call exfat_map_cluster() with
> s_lock held but without inode_lock, so they are not serialized against
> the expansion and can observe the chain metadata mid-update.
> 
> Hold s_lock around the cluster allocation, chain append and i_blocks
> update.  This nests inside the invalidate_lock held by
> exfat_cont_expand(), matching the order used by exfat_truncate().  i_size
> is published via exfat_pagecache_isize_extended(), which takes the
> straddling folio lock; since writeback takes folio lock -> s_lock, this
> must happen after s_lock is released to avoid inverting the lock order.
> i_blocks is therefore updated first, under s_lock, which also ensures a
> reader never sees i_size beyond the allocated chain.
> 
> Signed-off-by: Youling Tang <tangyouling@kylinos.cn>

Looks good to me:

Reviewed-by: Chi Zhiling <chizhiling@kylinos.cn>

> ---
> v2: rebase onto exfat/dev and adapt to __exfat_cont_expand().
> 
>   fs/exfat/file.c | 18 +++++++++++++-----
>   1 file changed, 13 insertions(+), 5 deletions(-)
> 
> diff --git a/fs/exfat/file.c b/fs/exfat/file.c
> index 9dfd99ad9265..c7991e98a07a 100644
> --- a/fs/exfat/file.c
> +++ b/fs/exfat/file.c
> @@ -105,6 +105,7 @@ static int __exfat_cont_expand(struct inode *inode, loff_t size)
>   	if (ret)
>   		return ret;
>   
> +	mutex_lock(&sbi->s_lock);
>   	num_clusters = exfat_bytes_to_cluster(sbi, exfat_ondisk_size(inode));
>   	/* integer overflow is already checked in inode_newsize_ok(). */
>   	new_num_clusters = exfat_bytes_to_cluster_round_up(sbi, size);
> @@ -116,7 +117,7 @@ static int __exfat_cont_expand(struct inode *inode, loff_t size)
>   		exfat_chain_set(&clu, ei->start_clu, num_clusters, ei->flags);
>   		ret = exfat_find_last_cluster(sb, &clu, &last_clu);
>   		if (ret)
> -			return ret;
> +			goto unlock;
>   
>   		clu.dir = last_clu + 1;
>   	} else {
> @@ -130,7 +131,7 @@ static int __exfat_cont_expand(struct inode *inode, loff_t size)
>   	ret = exfat_alloc_cluster(inode, new_num_clusters - num_clusters,
>   			&clu, inode_needs_sync(inode), false);
>   	if (ret)
> -		return ret;
> +		goto unlock;
>   
>   	/* Append new clusters to chain */
>   	if (num_clusters) {
> @@ -147,15 +148,19 @@ static int __exfat_cont_expand(struct inode *inode, loff_t size)
>   	ei->flags = clu.flags;
>   
>   out:
> +	inode->i_blocks = round_up(size, sbi->cluster_size) >> 9;
>   	inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
> +	mutex_unlock(&sbi->s_lock);
> +
>   	/*
>   	 * When extending file size, call exfat_pagecache_isize_extended()
>   	 * to update i_size and ensure the straddling folio is properly
>   	 * marked RO so page_mkwrite() is called and post-EOF area is zeroed.
> +	 *
> +	 * It takes the folio lock; call it without s_lock (writeback uses
> +	 * folio lock -> s_lock).
>   	 */
>   	exfat_pagecache_isize_extended(inode, oldsize, size);
> -
> -	inode->i_blocks = round_up(size, sbi->cluster_size) >> 9;
>   	mark_inode_dirty(inode);
>   
>   	if (IS_SYNC(inode))
> @@ -165,7 +170,10 @@ static int __exfat_cont_expand(struct inode *inode, loff_t size)
>   
>   free_clu:
>   	exfat_free_cluster(inode, &clu);
> -	return -EIO;
> +	ret = -EIO;
> +unlock:
> +	mutex_unlock(&sbi->s_lock);
> +	return ret;
>   }
>   
>   static int exfat_cont_expand(struct inode *inode, loff_t size)


^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2026-11-18  3:22 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-11-18  1:58 [PATCH v2] exfat: hold s_lock for cluster-chain updates in __exfat_cont_expand() Youling Tang
2026-11-18  3:22 ` Chi Zhiling

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®