On Fri, Dec 27, 2024 at 07:04:59PM +0200, Eugen Hristev wrote:
> In psz_init_zones, if the requested area has a total_size less than
> record_size, kcalloc will be called with c == 0 and will return
> ZERO_SIZE_PTR.
> Further, this will lead to an oops.
>
> With this patch, in this scenario, it will look like this :
> [ 6.865545] pstore_zone: total size : 28672 Bytes
> [ 6.865547] pstore_zone: kmsg size : 65536 Bytes
> [ 6.865549] pstore_zone: pmsg size : 0 Bytes
> [ 6.865551] pstore_zone: console size : 0 Bytes
> [ 6.865553] pstore_zone: ftrace size : 0 Bytes
> [ 6.872095] pstore_zone: zone dmesg total_size too small
> [ 6.878234] pstore_zone: alloc zones failed
>
> Signed-off-by: Eugen Hristev <eugen.hris...@linaro.org>
> ---
> fs/pstore/zone.c | 4 ++++
> 1 file changed, 4 insertions(+)
>
> diff --git a/fs/pstore/zone.c b/fs/pstore/zone.c
> index 694db616663f..50a765999b54 100644
> --- a/fs/pstore/zone.c
> +++ b/fs/pstore/zone.c
> @@ -1217,6 +1217,10 @@ static struct pstore_zone **psz_init_zones(enum
> pstore_type_id type,
> pr_err("allocate for zones %s failed\n", name);
> return ERR_PTR(-ENOMEM);
> }
> + if (unlikely(zones == ZERO_SIZE_PTR)) {
> + pr_err("zone %s total_size too small\n", name);
> + return ERR_PTR(-EINVAL);
> + }
I'd rather catch this after the calculation of "c" rather than after a
zero-sized allocation. Can you rework the patch to test it then?
-Kees
>
> for (i = 0; i < c; i++) {
> zone = psz_init_zone(type, off, record_size);
> --
> 2.43.0
>
--
Kees Cook
