Re: [PATCH] fat: serialize primary and mirror updates
From: OGAWA Hirofumi
Date: Sat Oct 31 2026 - 00:21:06 EST
Matthias Goergens <matthias.goergens@xxxxxxxxx> writes:
> __fat_alloc_clusters() drops fat_lock before syncing and mirroring
> modified buffer heads. Another allocation can change the primary FAT
> while an earlier allocation is copying it to the mirror.
>
> fat_ent_write() also updates and mirrors without fat_lock. Take the lock
> in both paths, preserving the existing post-unlock error cleanup.
>
> The test detector reproduced a primary-FAT write during a mirror copy.
> Three exact-current trials completed with no FAT conflicts and matching
> mirrors.
What is the real issue?
> Link: https://syzkaller.appspot.com/bug?extid=018e6f695ad9bccb182a
> ---
> fs/fat/fatent.c | 7 +++++--
> 1 file changed, 5 insertions(+), 2 deletions(-)
>
> diff --git a/fs/fat/fatent.c b/fs/fat/fatent.c
> index 2d10a815a579..5f2779c7f947 100644
> --- a/fs/fat/fatent.c
> +++ b/fs/fat/fatent.c
> @@ -419,6 +419,7 @@ int fat_ent_write(struct inode *inode, struct fat_entry *fatent,
> const struct fatent_operations *ops = MSDOS_SB(sb)->fatent_ops;
> int err;
>
> + lock_fat(MSDOS_SB(sb));
> ops->ent_put(fatent, new);
> if (wait) {
> err = fat_sync_bhs(fatent->bhs, fatent->nr_bhs);
> @@ -429,7 +430,7 @@ int fat_ent_write(struct inode *inode, struct fat_entry *fatent,
> * would not have value to write again on I/O
> * error.
> */
> - return err;
> + goto out_unlock;
> }
> }
>
> @@ -440,6 +441,8 @@ int fat_ent_write(struct inode *inode, struct fat_entry *fatent,
> if (wait)
> fat_sync_bhs(fatent->bhs, fatent->nr_bhs);
> }
> +out_unlock:
> + unlock_fat(MSDOS_SB(sb));
> return err;
> }
>
> @@ -582,7 +585,6 @@ static int __fat_alloc_clusters(struct inode *inode, int *cluster, int nr_cluste
> err = -ENOSPC;
>
> out:
> - unlock_fat(sbi);
> mark_fsinfo_dirty(sb);
> fatent_brelse(&fatent);
> if (!err) {
> @@ -591,6 +593,7 @@ static int __fat_alloc_clusters(struct inode *inode, int *cluster, int nr_cluste
> if (!err)
> err = fat_mirror_bhs(sb, bhs, nr_bhs);
> }
> + unlock_fat(sbi);
> for (i = 0; i < nr_bhs; i++)
> brelse(bhs[i]);
>
>
> base-commit: d89e1d186ec9271bf78e62771779e9fc03bf0fc3
--
OGAWA Hirofumi <hirofumi@xxxxxxxxxxxxxxxxxx>