Re: [f2fs-dev] [PATCH] f2fs: avoid reading already updated pages during GC

[Chao Yu via Linux-f2fs-devel <linux-f2fs-devel@lists.sourceforge.net>]

On 2026/2/9 14:31, Jianan Huang wrote:
> We found the following issue during fuzz testing:
> 
> page: refcount:3 mapcount:0 mapping:00000000b6e89c65 index:0x18b2dc pfn:0x161ba9
> memcg:f8ffff800e269c00
> aops:f2fs_meta_aops ino:2
> flags: 0x52880000000080a9(locked|waiters|uptodate|lru|private|zone=1|kasantag=0x4a)
> raw: 52880000000080a9 fffffffec6e17588 fffffffec0ccc088 a7ffff8067063618
> raw: 000000000018b2dc 0000000000000009 00000003ffffffff f8ffff800e269c00
> page dumped because: VM_BUG_ON_FOLIO(folio_test_uptodate(folio))
> page_owner tracks the page as allocated
>   post_alloc_hook+0x58c/0x5ec
>   prep_new_page+0x34/0x284
>   get_page_from_freelist+0x2dcc/0x2e8c
>   __alloc_pages_noprof+0x280/0x76c
>   __folio_alloc_noprof+0x18/0xac
>   __filemap_get_folio+0x6bc/0xdc4
>   pagecache_get_page+0x3c/0x104
>   do_garbage_collect+0x5c78/0x77a4
>   f2fs_gc+0xd74/0x25f0
>   gc_thread_func+0xb28/0x2930
>   kthread+0x464/0x5d8
>   ret_from_fork+0x10/0x20
> ------------[ cut here ]------------
> kernel BUG at mm/filemap.c:1563!
>   folio_end_read+0x140/0x168
>   f2fs_finish_read_bio+0x5c4/0xb80
>   f2fs_read_end_io+0x64c/0x708
>   bio_endio+0x85c/0x8c0
>   blk_update_request+0x690/0x127c
>   scsi_end_request+0x9c/0xb8c
>   scsi_io_completion+0xf0/0x250
>   scsi_finish_command+0x430/0x45c
>   scsi_complete+0x178/0x6d4
>   blk_mq_complete_request+0xcc/0x104
>   scsi_done_internal+0x214/0x454
>   scsi_done+0x24/0x34
> 
> which is similar to the problem reported by syzbot:
> https://syzkaller.appspot.com/bug?extid=3686758660f980b402dc
> 
> This case is consistent with the description in commit 9bf1a3f
> ("f2fs: avoid GC causing encrypted file corrupted"):
> Page 1 is moved from blkaddr A to blkaddr B by move_data_block, and after
> being written it is marked as uptodate. Then, Page 1 is moved from blkaddr
> B to blkaddr C, VM_BUG_ON_FOLIO was triggered in the endio initiated by
> ra_data_block.
> 
> There is no need to read Page 1 again from blkaddr B, since it has already
> been updated. Therefore, avoid initiating I/O in this case.
> 
> Fixes: 6aa58d8ad20a ("f2fs: readahead encrypted block during GC")
> Signed-off-by: Jianan Huang <huangjianan@xiaomi.com>
> Signed-off-by: Sheng Yong <shengyong1@xiaomi.com>
> ---
>   fs/f2fs/gc.c | 5 ++++-
>   1 file changed, 4 insertions(+), 1 deletion(-)
> 
> diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c
> index 60378614bc54..c184ae6a616e 100644
> --- a/fs/f2fs/gc.c
> +++ b/fs/f2fs/gc.c
> @@ -1228,7 +1228,7 @@ static int ra_data_block(struct inode *inode, pgoff_t index)
>   		.encrypted_page = NULL,
>   		.in_list = 0,
>   	};
> -	int err;
> +	int err = 0;
>   
>   	folio = f2fs_grab_cache_folio(mapping, index, true);
>   	if (IS_ERR(folio))
> @@ -1279,6 +1279,9 @@ static int ra_data_block(struct inode *inode, pgoff_t index)
>   		goto put_folio;
>   	}
>   
> +	if (folio_test_uptodate(efolio))
> +		goto put_encrypted_page;
> +
>   	fio.encrypted_page = &efolio->page;

Should relocate changes here, otherwise, fio.encrypted_page will be used before
initialization.

1299         return 0;
1300 put_encrypted_page:
1301         f2fs_put_page(fio.encrypted_page, true);
1302 put_folio:
1303         f2fs_folio_put(folio, true);
1304         return err;

Thanks,

>   
>   	err = f2fs_submit_page_bio(&fio);



_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel