Re: [PATCH v3 2/2] wl3501_cs: Fix out-of-bounds warnings in wl3501_mgmt_join

[Kees Cook <keescook@chromium.org>]

On Wed, Apr 14, 2021 at 06:45:15PM -0500, Gustavo A. R. Silva wrote:
> Fix the following out-of-bounds warnings by adding a new structure
> wl3501_req instead of duplicating the same members in structure
> wl3501_join_req and wl3501_scan_confirm:
> 
> arch/x86/include/asm/string_32.h:182:25: warning: '__builtin_memcpy' offset [39, 108] from the object at 'sig' is out of the bounds of referenced subobject 'beacon_period' with type 'short unsigned int' at offset 36 [-Warray-bounds]
> arch/x86/include/asm/string_32.h:182:25: warning: '__builtin_memcpy' offset [25, 95] from the object at 'sig' is out of the bounds of referenced subobject 'beacon_period' with type 'short unsigned int' at offset 22 [-Warray-bounds]
> 
> Refactor the code, accordingly:
> 
> $ pahole -C wl3501_req drivers/net/wireless/wl3501_cs.o
> struct wl3501_req {
>         u16                        beacon_period;        /*     0     2 */
>         u16                        dtim_period;          /*     2     2 */
>         u16                        cap_info;             /*     4     2 */
>         u8                         bss_type;             /*     6     1 */
>         u8                         bssid[6];             /*     7     6 */
>         struct iw_mgmt_essid_pset  ssid;                 /*    13    34 */
>         struct iw_mgmt_ds_pset     ds_pset;              /*    47     3 */
>         struct iw_mgmt_cf_pset     cf_pset;              /*    50     8 */
>         struct iw_mgmt_ibss_pset   ibss_pset;            /*    58     4 */
>         struct iw_mgmt_data_rset   bss_basic_rset;       /*    62    10 */
> 
>         /* size: 72, cachelines: 2, members: 10 */
>         /* last cacheline: 8 bytes */
> };
> 
> $ pahole -C wl3501_join_req drivers/net/wireless/wl3501_cs.o
> struct wl3501_join_req {
>         u16                        next_blk;             /*     0     2 */
>         u8                         sig_id;               /*     2     1 */
>         u8                         reserved;             /*     3     1 */
>         struct iw_mgmt_data_rset   operational_rset;     /*     4    10 */
>         u16                        reserved2;            /*    14     2 */
>         u16                        timeout;              /*    16     2 */
>         u16                        probe_delay;          /*    18     2 */
>         u8                         timestamp[8];         /*    20     8 */
>         u8                         local_time[8];        /*    28     8 */
>         struct wl3501_req          req;                  /*    36    72 */
> 
>         /* size: 108, cachelines: 2, members: 10 */
>         /* last cacheline: 44 bytes */
> };
> 
> $ pahole -C wl3501_scan_confirm drivers/net/wireless/wl3501_cs.o
> struct wl3501_scan_confirm {
>         u16                        next_blk;             /*     0     2 */
>         u8                         sig_id;               /*     2     1 */
>         u8                         reserved;             /*     3     1 */
>         u16                        status;               /*     4     2 */
>         char                       timestamp[8];         /*     6     8 */
>         char                       localtime[8];         /*    14     8 */
>         struct wl3501_req          req;                  /*    22    72 */
>         /* --- cacheline 1 boundary (64 bytes) was 30 bytes ago --- */
>         u8                         rssi;                 /*    94     1 */
> 
>         /* size: 96, cachelines: 2, members: 8 */
>         /* padding: 1 */
>         /* last cacheline: 32 bytes */
> };
> 
> The problem is that the original code is trying to copy data into a
> bunch of struct members adjacent to each other in a single call to
> memcpy(). Now that a new struct wl3501_req enclosing all those adjacent
> members is introduced, memcpy() doesn't overrun the length of
> &sig.beacon_period and &this->bss_set[i].beacon_period, because the
> address of the new struct object _req_ is used as the destination,
> instead.
> 
> This helps with the ongoing efforts to globally enable -Warray-bounds
> and get us closer to being able to tighten the FORTIFY_SOURCE routines
> on memcpy().
> 
> Link: https://github.com/KSPP/linux/issues/109
> Reported-by: kernel test robot <lkp@intel.com>
> Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>

Awesome! Thank you for this solution.

Reviewed-by: Kees Cook <keescook@chromium.org>

-- 
Kees Cook