Re: [PATCH] autofs4: Fix endless loop in autofs4_write

[Ian Kent]

On Fri, 2016-06-10 at 19:07 +0200, Laurent Dufour wrote:
> The 'commit e9a7c2f1a548 ("autofs4: coding style fixes")' removed the
> check done on the __vfs_write()'s returned value in autofs4_write().
> This may lead to a spinning process which can't catch any signal.

Yeah, sorry my bad.

> 
> Call stack showed in xmon could be :
> [c0000003a76c7500] c00000000030df74 __vfs_write+0x134/0x1c0
> (unreliable)
> [c0000003a76c75a0] d0000000052a35d4 autofs4_notify_daemon+0x174/0x3f0
> [autofs4]
> [c0000003a76c7780] d0000000052a3fa0 autofs4_wait+0x750/0xa10 [autofs4]
> [c0000003a76c78b0] d0000000052a24d8 autofs4_mount_wait+0x78/0x140
> [autofs4]
> [c0000003a76c7930] d0000000052a2f48 autofs4_d_automount+0x1d8/0x370
> [autofs4]
> [c0000003a76c79c0] c0000000003221e4 follow_managed+0x204/0x3a0
> [c0000003a76c7a20] c000000000322c10 lookup_fast+0x220/0x420
> [c0000003a76c7a90] c00000000032324c walk_component+0x5c/0x3e0
> [c0000003a76c7b00] c000000000323794 link_path_walk+0x1c4/0x5f0
> [c0000003a76c7b90] c000000000324b00 path_openat+0xf0/0x1620
> [c0000003a76c7c90] c000000000327f6c do_filp_open+0xfc/0x170
> [c0000003a76c7dc0] c00000000030d06c do_sys_open+0x1bc/0x2e0
> [c0000003a76c7e30] c000000000009260 system_call+0x38/0x108
> --- Exception: c01 (System Call) at 00003fffa38a0988
> 
> Cc: Ian Kent <raven@xxxxxxxxxx>
> Cc: autofs@xxxxxxxxxxxxxxx
> Cc: linux-kernel@xxxxxxxxxxxxxxx
> Cc: stable@xxxxxxxxxxxxxxx
> Reviewed-by: Greg Kurz <groug@xxxxxxxx>
> Signed-off-by: Laurent Dufour <ldufour@xxxxxxxxxxxxxxxxxx>
> Fixes: e9a7c2f1a548 ("autofs4: coding style fixes")
> ---
>  fs/autofs4/waitq.c | 7 ++++---
>  1 file changed, 4 insertions(+), 3 deletions(-)
> 
> diff --git a/fs/autofs4/waitq.c b/fs/autofs4/waitq.c
> index 0146d911f468..106d94139281 100644
> --- a/fs/autofs4/waitq.c
> +++ b/fs/autofs4/waitq.c
> @@ -66,11 +66,12 @@ static int autofs4_write(struct autofs_sb_info *sbi,
>  	set_fs(KERNEL_DS);
>  
>  	mutex_lock(&sbi->pipe_mutex);
> -	wr = __vfs_write(file, data, bytes, &file->f_pos);
> -	while (bytes && wr) {

Right but why not just wr >= 0 here.

I guess this patch probably saves a few bytes.

I'll add it to the series.
 
> +	while (bytes) {
> +		wr = __vfs_write(file, data, bytes, &file->f_pos);
> +		if (wr < 0)
> +			break;
>  		data += wr;
>  		bytes -= wr;
> -		wr = __vfs_write(file, data, bytes, &file->f_pos);
>  	}
>  	mutex_unlock(&sbi->pipe_mutex);
>