Re: [PATCH] kernel: module: strncpy issue, using strlcpy instead of strncpy

[Rusty Russell <rusty@rustcorp.com.au>]

Chen Gang <gang.chen@asianux.com> writes:
>   ownername and namebuf are all NUL terminated string.
>
>   need always let them ended by '\0'.
>
> Signed-off-by: Chen Gang <gang.chen@asianux.com>
> ---
>  kernel/module.c |    4 ++--
>  1 files changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/kernel/module.c b/kernel/module.c
> index 3c2c72d..597efd8 100644
> --- a/kernel/module.c
> +++ b/kernel/module.c
> @@ -1283,7 +1283,7 @@ static const struct kernel_symbol *resolve_symbol(struct module *mod,
>  
>  getname:
>  	/* We must make copy under the lock if we failed to get ref. */
> -	strncpy(ownername, module_name(owner), MODULE_NAME_LEN);
> +	strlcpy(ownername, module_name(owner), MODULE_NAME_LEN);

This should just be strcpy().

>  unlock:
>  	mutex_unlock(&module_mutex);
>  	return sym;
> @@ -3464,7 +3464,7 @@ const char *module_address_lookup(unsigned long addr,
>  	}
>  	/* Make a copy in here where it's safe */
>  	if (ret) {
> -		strncpy(namebuf, ret, KSYM_NAME_LEN - 1);
> +		strlcpy(namebuf, ret, KSYM_NAME_LEN);

This isn't a bug, because the caller (kallsyms_lookup) puts a NUL in
ret[KSYM_NAME_LEN].

However, kallsyms_lookup also calls kallsyms_expand_symbol, which
doesn't stop at KSYM_NAME_LEN, so if a name was longer than that we'd
have a real bug.

Would you like to take a look at that, too?

Thanks,
Rusty.