On Thu, 2020-12-03 at 12:08 +0300, Dan Carpenter wrote:
> Hello Luca Coelho,
>
> The patch fbb7957d28ac: "iwlwifi: support REDUCE_TX_POWER_CMD version
> 6" from Sep 28, 2020, leads to the following static checker warning:
>
> drivers/net/wireless/intel/iwlwifi/fw/acpi.c:462 iwl_sar_fill_table()
> error: buffer overflow 'prof->table' 10 <= 15
>
> drivers/net/wireless/intel/iwlwifi/fw/acpi.c
> 422 static int iwl_sar_fill_table(struct iwl_fw_runtime *fwrt,
> 423 __le16 *per_chain, u32 n_subbands,
> 424 int prof_a, int prof_b)
>
> Original n_subbands was ACPI_SAR_NUM_SUB_BANDS (5) but now it can be
> IWL_NUM_SUB_BANDS_V2 (11) as well.
>
> 425 {
> 426 int profs[ACPI_SAR_NUM_CHAIN_LIMITS] = { prof_a, prof_b };
> 427 int i, j, idx;
> 428
> 429 for (i = 0; i < ACPI_SAR_NUM_CHAIN_LIMITS; i++) {
> 430 struct iwl_sar_profile *prof;
> 431
> 432 /* don't allow SAR to be disabled (profile 0 means disable) */
> 433 if (profs[i] == 0)
> 434 return -EPERM;
> 435
> 436 /* we are off by one, so allow up to ACPI_SAR_PROFILE_NUM */
> 437 if (profs[i] > ACPI_SAR_PROFILE_NUM)
> 438 return -EINVAL;
> 439
> 440 /* profiles go from 1 to 4, so decrement to access the array */
> 441 prof = &fwrt->sar_profiles[profs[i] - 1];
> 442
> 443 /* if the profile is disabled, do nothing */
> 444 if (!prof->enabled) {
> 445 IWL_DEBUG_RADIO(fwrt, "SAR profile %d is disabled.\n",
> 446 profs[i]);
> 447 /*
> 448 * if one of the profiles is disabled, we
> 449 * ignore all of them and return 1 to
> 450 * differentiate disabled from other failures.
> 451 */
> 452 return 1;
> 453 }
> 454
> 455 IWL_DEBUG_INFO(fwrt,
> 456 "SAR EWRD: chain %d profile index %d\n",
> 457 i, profs[i]);
> 458 IWL_DEBUG_RADIO(fwrt, " Chain[%d]:\n", i);
> 459 for (j = 0; j < n_subbands; j++) {
> 460 idx = i * ACPI_SAR_NUM_SUB_BANDS + j;
> 461 per_chain[i * n_subbands + j] =
> 462 cpu_to_le16(prof->table[idx]);
> ^^^^^^^^^^^^^^^^
> But this table size wasn't increased so potentially we're reading beyond
> the end of the array?
This is a good catch! The intention of this patch is to support the
larger structures in the FW, but not from ACPI. So the solution for
this here, would be to check if j is larger than ACPI_SAR_NUM_SUB_BANDS
(which is the max ACPI table we support) and just fill in with zeros in
that case.
This code has already been refactored in our internal tree (and will go
upstream soon) to support the full size table from ACPI as well. But
since the changes are not only fixing this bug, I'll send a separate
bugfix for the mainline for now.
Thanks, Dan!
--
Cheers,
Luca.
