Broadcom: Multiple memory corruptions in "dhd_pno_process_anqpo_result" (CVE-2017-0572)

ID SSV:92897
Type seebug
Reporter Root
Modified 2017-04-05T00:00:00


Detailed analysis of reference : the

  • the first part
  • Part II

Broadcom produces the Wi-Fi HardMAC SoCs which are used to handle the PHY and MAC layer processing. These chips are present in both mobile devices and Wi-Fi routers, and are capable of handling many Wi-Fi related events without delegating to the host OS. On Android devices, the "bcmdhd" driver is used in order to communicate with the Wi-Fi SoC (also referred to as "dongle").

When the dongle wishes to notify the host OS of an event, it does so by encoding a special "packet" and transmitting it to the host. These packets have an ether type of 0x886C (referred to as ETHER_TYPE_BRCM), and do not contain actual packet data, but rather encapsulate information about events which must be handled by the driver.

After reading packets from the SDIO interface, the "bcmdhd" driver calls the function "dhd_rx_frame" to handle each of the received frames. If a frame has the special Broadcom ether type, it is passed on to an internal handling function, "dhd_wl_host_event". This function inspects the event code, and passes it onto the registered handlers for the given event type.

The function "wl_notify_gscan_event" is the registered handler for events of the following types: -WLC_E_PFN_BEST_BATCHING -WLC_E_PFN_SCAN_COMPLETE -WLC_E_PFN_GSCAN_FULL_RESULT -WLC_E_PFN_SWC -WLC_E_PFN_BSSID_NET_FOUND -WLC_E_PFN_BSSID_NET_LOST -WLC_E_PFN_SSID_EXT -WLC_E_GAS_FRAGMENT_RX (for reference, see "wl_init_event_handler")

Specifically, when the event code "WLC_E_PFN_BSSID_NET_FOUND" is received, the gscan handler function calls "dhd_pno_process_anqpo_result" in order to process the event's data, like so:

1. dhd_pno_process_anqpo_result(dhd_pub_t *dhd, const void *data, uint32 event, int *size) 2. { 3. wl_bss_info_t *bi = (wl_bss_info_t *)data; 4. wifi_gscan_full_result_t *result = NULL; 5. wl_event_gas_t *gas_data = (wl_event_gas_t *)((uint8 *)data + 6. bi->ie_offset + 7. bi->ie_length); 8. ... 9. uint32 mem_needed = OFFSETOF(wifi_gscan_full_result_t, ie_data) + 10. bi->ie_length + 11. OFFSETOF(wl_event_gas_t, data) + gas_data->data_len + 12. sizeof(int); 13. result = (wifi_gscan_full_result_t *) kmalloc(mem_needed, GFP_KERNEL); 14. if (result == NULL) { 15. DHD_ERROR(("%s Cannot Malloc %d bytes!!\ n", __FUNCTION__, mem_needed)); 16. return NULL; 17. } 18. memcpy(result->fixed. ssid, bi->SSID, bi->SSID_len); 19. result->fixed. ssid[bi->SSID_len] = '\0'; 20. ... 21. memcpy(result->ie_data, ((uint8 *)bi + bi->ie_offset), bi->ie_length); 22. return result; 23. }

(where "data" is the arbitrary data encapsulated in the event passed in from the dongle)

First, in line 18, the function uses the "bi->SSID_len" argument in order to copy the SSID from the event to the destination buffer ("result"), but fails to verify that it does not exceed the maximal SSID length (32). Assuming an attacker has arbitrary control over the dongle, they would be able to overflow the allocated buffer by setting: - bi->ie_length = 0 - bi->SSID_len = 255

Since OFFSETOF(wifi_gscan_full_result_t, ie_data) == 86 and OFFSETOF(wl_event_gas_t, data) + gas_data->data_len == 8, the size of the allocated buffer will only be 86 + 8 + 4 = 98, causing the memcpy call above to overflow it.

Alternately, an attacker controlling the dongle could supply an abnormally large value in the "bi->ie_length" (which is a a uint32_t that). This would cause the calculation of "mem_needed" (lines 9-12) to overflow, resulting in a small value. However, later in the function an additional memcpy operation is performed using the value of "bi->ie_length", thus on paying the allocated buffer.

I've been able to statically verify these issues on the "bcmdhd-3.10" driver, and in the corresponding "bcmdhd" driver on the Nexus 6P's kernel (the angler).