Re: [PATCH net-next v2] af_unix: Refine UNIX pathname sockets autobind identifier length

[Kuniyuki Iwashima]

From: Liang Jie <buaajxlj@xxxxxxx>
Date: Thu,  6 Feb 2025 16:19:05 +0800
> Hi Kuniyuki,
> 
> The logs from 'netdev/build_allmodconfig_warn' is as follows:
>   ../net/unix/af_unix.c: In function ‘unix_autobind’:
>   ../net/unix/af_unix.c:1222:52: warning: ‘snprintf’ output truncated before the last format character [-Wformat-truncation=]
>    1222 |         snprintf(addr->name->sun_path + 1, 5, "%05x", ordernum);
>         |                                                    ^
>   ../net/unix/af_unix.c:1222:9: note: ‘snprintf’ output 6 bytes into a destination of size 5
>    1222 |         snprintf(addr->name->sun_path + 1, 5, "%05x", ordernum);
>         |         ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> 
> snprintf() also append a trailing '\0' at the end of the sun_path.

I didn't say snprintf() would work rather we need a variant of it that
does not terminate string with \0.


> 
> Now, I think of three options. Which one do you think we should choose?
> 
> 1. Allocate an additional byte during the kzalloc phase.
> 	addr = kzalloc(sizeof(*addr) + offsetof(struct sockaddr_un, sun_path) +
> 		       UNIX_AUTOBIND_LEN + 1, GFP_KERNEL);
> 
> 2. Use temp buffer and memcpy() for handling.
> 
> 3. Keep the current code as it is.
> 
> Do you have any other suggestions?

I'd choose 3. as said in v1 thread.  We can't avoid hard-coding and
adjustment like +1 and -1 here.