Re: [f2fs-dev] [PATCH v2 1/1] f2fs: fix lockdep WARN of sbi->cp_global_sem and q->q_usage_counter

[Jaegeuk Kim via Linux-f2fs-devel <linux-f2fs-devel@lists.sourceforge.net>]

On 02/18, Shin'ichiro Kawasaki wrote:
> From: Shin'ichiro Kawasaki via Linux-f2fs-devel <linux-f2fs-devel@lists.sourceforge.net>
> 
> A lockdep WARN is observed recently under the following steps:
> 
> 1) Create a zoned TCMU device
> 2) Create a f2fs filesystem on the zoned TCMU device and mount it
> 3) Fill the filesystem with files and trigger GC
> 4) Unmout the filesystem
> 5) Remove the zoned TCMU device
> 
> The lockdep WARN indicates that a circular lock depedency formed by four
> contexts, as described below.
> 
> a) TCMU device removal context:
>  - call del_gendisk() to get q->q_usage_counter
>  - call start_flush_work() to get work_completion of wb->dwork
> b) f2fs writeback context:
>  - in wb_workfn(), which holds work_completion of wb->dwork
>  - call f2fs_balance_fs() to get sbi->gc_lock
> c) f2fs vfs_write context:
>  - call f2fs_gc() to get sbi->gc_lock
>  - call f2fs_write_checkpoint() to get sbi->cp_global_sem
> d) f2fs mount context:
>  - call recover_fsync_data() to get sbi->cp_global_sem
>  - call f2fs_check_and_fix_write_pointer() to call blkdev_report_zones()
>    that goes down to blk_mq_alloc_request and get q->q_usage_counter
> 
> To suppress the WARN, cut the dependency d) between sbi->cp_global_sem
> and q->q_usage_counter. For that purpose, move the
> f2fs_check_and_fix_write_pointer() call outside of the critical section
> of sbi->cp_global_sem in f2fs_recovery_fsync_data(). This change is fine
> because the write pointer fix operation only affects the main segments
> and does not interact with the check point metadata. Furthermore,
> conflicts between the write pointer fix operation and data/node flush
> operations remain protected by SBI_POR_DOING.
> 
> Fixes: c426d99127b1 ("f2fs: Check write pointer consistency of open zones")
> Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
> Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
> ---
>  fs/f2fs/recovery.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/fs/f2fs/recovery.c b/fs/f2fs/recovery.c
> index a26071f2b0bc..87fd6cd436fe 100644
> --- a/fs/f2fs/recovery.c
> +++ b/fs/f2fs/recovery.c
> @@ -922,6 +922,8 @@ int f2fs_recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only)
>  		truncate_inode_pages_final(META_MAPPING(sbi));
>  	}
>  
> +	f2fs_up_write_trace(&sbi->cp_global_sem, &lc);
> +
>  	/*
>  	 * If fsync data succeeds or there is no fsync data to recover,
>  	 * and the f2fs is not read only, check and fix zoned block devices'
> @@ -933,8 +935,6 @@ int f2fs_recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only)
>  	if (!err)
>  		clear_sbi_flag(sbi, SBI_POR_DOING);
>  
> -	f2fs_up_write_trace(&sbi->cp_global_sem, &lc);
> -

This was a guard to prevent checkpoint during f2fs_check_and_fix_write_pointer()
where it changes the checkpoint as well?

>  	/* let's drop all the directory inodes for clean checkpoint */
>  	destroy_fsync_dnodes(&dir_list, err);
>  
> -- 
> 2.49.0


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