Re: binfmt_flat.c && bprm->cred (Was: [PATCH 0/1] exec: do not sleep in TASK_TRACED under ->cred_guard_mutex)

[Oleg Nesterov <oleg@redhat.com>]

On 09/03, Cyrill Gorcunov wrote:
>
> [Oleg Nesterov - Thu, Sep 03, 2009 at 06:29:39PM +0200]
> | On 09/03, Oleg Nesterov wrote:
> | >
> | > load_flat_shared_library() does something strange (but hopefully this
> | > patch doesn't break it). I do not understand why does it create the
> | > new bprm. Afaics, it could reuse bprm pointer which comes as an argument
> | > of ->load_binary(), all we need is to temporary change/restore bprm->file
> | > for load_flat_file().
> |
> | IOW, afaics the patch below makes sense. Imho it is a bit ugly binfmt_flat.c
> | plays with prepare_exec_creds().
> |
> | But again, I don't understand this code, and I didn't even try to compile
> | this patch.
> |
> | Oleg.
> |
> ...
> | -static int load_flat_shared_library(int id, struct lib_info *libs)
> | +static int load_flat_shared_library(struct linux_binprm *bprm, int id,
> | +						struct lib_info *libs)
> |  {
> ...
> | +	sprintf(buf, "/lib/lib%d.so", id);
>
> Hi Oleg, perhaps it is a good moment to switch sprintf to snprintf
> as well? buf is only 16 bytes long so we have 4 byte room for number.

Agreed. As you pointed out privately we have MAX_SHARED_LIBS=4, but
still snprintf() is safer.

Oleg.