Re: [RFC PATCH 4/8] fat: clean stale bdev buffers of newly allocated clusters
From: Matheus Alves de Almeida
Date: Wed Nov 11 2026 - 23:22:27 EST
On Sun, Nov 08, 2026 at 11:57:29PM +0900, OGAWA Hirofumi wrote:
>Matheus Alves de Almeida <matheus.aalmeida@xxxxxxxxxxxx> writes:
>
>> Directory I/O uses buffer_heads on the block device, which may leave
>> stale buffers behind when clusters are freed and reused. The buffer_head
>> I/O paths currently call clean_bdev_aliases() for newly allocated blocks,
>> preventing stale buffers from overwriting new data.
>>
>> Buffered iomap I/O does not provide this protection. In preparation for
>> the buffered iomap conversion, clean bdev aliases on newly allocated
>> clusters, like exfat patch "exfat: clean stale bdev buffers of newly
>> allocated clusters".
>>
>> Signed-off-by: Matheus Alves de Almeida <matheus.aalmeida@xxxxxxxxxxxx>
>> ---
>> fs/fat/fatent.c | 34 ++++++++++++++++++++++++++++++++--
>> 1 file changed, 32 insertions(+), 2 deletions(-)
>>
>> diff --git a/fs/fat/fatent.c b/fs/fat/fatent.c
>> index 4fa1fae5abf9..46899efa3cf8 100644
>> --- a/fs/fat/fatent.c
>> +++ b/fs/fat/fatent.c
>> @@ -523,6 +523,21 @@ static int fat_flush_bhs(struct super_block *sb, struct buffer_head **bhs,
>> return 0;
>> }
>>
>> +/*
>> + * Newly allocated clusters may still have dirty buffer_heads in the block
>> + * device mapping, e.g. the dentries of a directory that was just removed.
>> + * File data does not go through those buffers, so drop them; otherwise a
>> + * later flush of the block device writes them over the new data.
>> + */
>> +static void fat_clean_bdev_aliases(struct super_block *sb,
>> + unsigned int clu, unsigned int count)
>> +{
>> + struct msdos_sb_info *sbi = MSDOS_SB(sb);
>> +
>> + clean_bdev_aliases(sb->s_bdev, fat_clus_to_blknr(sbi, clu),
>> + (sector_t)count * sbi->sec_per_clus);
>> +}
>> +
>> static int __fat_alloc_clusters(struct inode *inode, int *cluster, int nr_cluster,
>> bool return_all)
>> {
>> @@ -531,7 +546,7 @@ static int __fat_alloc_clusters(struct inode *inode, int *cluster, int nr_cluste
>> const struct fatent_operations *ops = sbi->fatent_ops;
>> struct fat_entry fatent, prev_ent;
>> struct buffer_head *bhs[MAX_BUF_PER_PAGE];
>> - int i, count, err, nr_bhs, idx_clus;
>> + int i, count, err, nr_bhs, idx_clus, cont_start, cont_len;
>>
>> lock_fat(sbi);
>> if (sbi->free_clusters != -1 && sbi->free_clus_valid &&
>> @@ -540,14 +555,20 @@ static int __fat_alloc_clusters(struct inode *inode, int *cluster, int nr_cluste
>> return -ENOSPC;
>> }
>>
>> + cont_len = 0;
>> err = nr_bhs = idx_clus = 0;
>> count = FAT_START_ENT;
>> fatent_init(&prev_ent);
>> fatent_init(&fatent);
>> fatent_set_entry(&fatent, sbi->prev_free + 1);
>> while (count < sbi->max_cluster) {
>> - if (fatent.entry >= sbi->max_cluster)
>> + if (fatent.entry >= sbi->max_cluster) {
>> + if (cont_len) {
>> + fat_clean_bdev_aliases(sb, cont_start, cont_len);
>> + cont_len = 0;
>> + }
>> fatent.entry = FAT_START_ENT;
>> + }
>> fatent_set_entry(&fatent, fatent.entry);
>> err = fat_ent_read_block(sb, &fatent);
>> if (err)
>> @@ -573,6 +594,10 @@ static int __fat_alloc_clusters(struct inode *inode, int *cluster, int nr_cluste
>> *cluster = entry;
>> idx_clus++;
>>
>> + if (cont_len == 0)
>> + cont_start = entry;
>> + cont_len++;
>> +
>> if (nr_bhs + fatent.nr_bhs > MAX_BUF_PER_PAGE) {
>> err = fat_flush_bhs(sb, bhs, &nr_bhs,
>> inode_needs_sync(inode));
>> @@ -589,6 +614,9 @@ static int __fat_alloc_clusters(struct inode *inode, int *cluster, int nr_cluste
>> * so we can still use the prev_ent.
>> */
>> prev_ent = fatent;
>> + } else if (cont_len) {
>> + fat_clean_bdev_aliases(sb, cont_start, cont_len);
>> + cont_len = 0;
>> }
>> count++;
>> if (count == sbi->max_cluster)
>> @@ -602,6 +630,8 @@ static int __fat_alloc_clusters(struct inode *inode, int *cluster, int nr_cluste
>> err = -ENOSPC;
>>
>> out:
>> + if (cont_len)
>> + fat_clean_bdev_aliases(sb, cont_start, cont_len);
>
>We would want to make as possible as shorter lock_fat. At least, this
>part can move outside of unlock_fat(sbi)?
>
Yes, this can be moved. Good catch.
>> unlock_fat(sbi);
>> mark_fsinfo_dirty(sb);
>> fatent_brelse(&fatent);
Thanks,
Matheus