rtw88: Fix array overrun in rtw_get_tx_power_params()
authorPing-Ke Shih <pkshih@realtek.com>
Thu, 1 Apr 2021 19:27:17 +0000 (14:27 -0500)
committerKalle Valo <kvalo@codeaurora.org>
Sun, 18 Apr 2021 06:34:28 +0000 (09:34 +0300)
Using a kernel with the Undefined Behaviour Sanity Checker (UBSAN) enabled, the
following array overrun is logged:

================================================================================
UBSAN: array-index-out-of-bounds in /home/finger/wireless-drivers-next/drivers/net/wireless/realtek/rtw88/phy.c:1789:34
index 5 is out of range for type 'u8 [5]'
CPU: 2 PID: 84 Comm: kworker/u16:3 Tainted: G           O      5.12.0-rc5-00086-gd88bba47038e-dirty #651
Hardware name: TOSHIBA TECRA A50-A/TECRA A50-A, BIOS Version 4.50   09/29/2014
Workqueue: phy0 ieee80211_scan_work [mac80211]
Call Trace:
 dump_stack+0x64/0x7c
 ubsan_epilogue+0x5/0x40
 __ubsan_handle_out_of_bounds.cold+0x43/0x48
 rtw_get_tx_power_params+0x83a/drivers/net/wireless/realtek/rtw88/0xad0 [rtw_core]
 ? rtw_pci_read16+0x20/0x20 [rtw_pci]
 ? check_hw_ready+0x50/0x90 [rtw_core]
 rtw_phy_get_tx_power_index+0x4d/0xd0 [rtw_core]
 rtw_phy_set_tx_power_level+0xee/0x1b0 [rtw_core]
 rtw_set_channel+0xab/0x110 [rtw_core]
 rtw_ops_config+0x87/0xc0 [rtw_core]
 ieee80211_hw_config+0x9d/0x130 [mac80211]
 ieee80211_scan_state_set_channel+0x81/0x170 [mac80211]
 ieee80211_scan_work+0x19f/0x2a0 [mac80211]
 process_one_work+0x1dd/0x3a0
 worker_thread+0x49/0x330
 ? rescuer_thread+0x3a0/0x3a0
 kthread+0x134/0x150
 ? kthread_create_worker_on_cpu+0x70/0x70
 ret_from_fork+0x22/0x30
================================================================================

The statement where an array is being overrun is shown in the following snippet:

if (rate <= DESC_RATE11M)
tx_power = pwr_idx_2g->cck_base[group];
else
====> tx_power = pwr_idx_2g->bw40_base[group];

The associated arrays are defined in main.h as follows:

struct rtw_2g_txpwr_idx {
u8 cck_base[6];
u8 bw40_base[5];
struct rtw_2g_1s_pwr_idx_diff ht_1s_diff;
struct rtw_2g_ns_pwr_idx_diff ht_2s_diff;
struct rtw_2g_ns_pwr_idx_diff ht_3s_diff;
struct rtw_2g_ns_pwr_idx_diff ht_4s_diff;
};

The problem arises because the value of group is 5 for channel 14. The trivial
increase in the dimension of bw40_base fails as this struct must match the layout of
efuse. The fix is to add the rate as an argument to rtw_get_channel_group() and set
the group for channel 14 to 4 if rate <= DESC_RATE11M.

This patch fixes commit fa6dfe6bff24 ("rtw88: resolve order of tx power setting routines")

Fixes: fa6dfe6bff24 ("rtw88: resolve order of tx power setting routines")
Reported-by: Богдан Пилипенко <bogdan.pylypenko107@gmail.com>
Signed-off-by: Larry Finger <Larry.Finger@lwfinger.net>
Signed-off-by: Ping-Ke Shih <pkshih@realtek.com>
Cc: Stable <stable@vger.kernel.org>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
Link: https://lore.kernel.org/r/20210401192717.28927-1-Larry.Finger@lwfinger.net
drivers/net/wireless/realtek/rtw88/phy.c

index ae401a7..fdd5da0 100644 (file)
@@ -1586,7 +1586,7 @@ void rtw_phy_load_tables(struct rtw_dev *rtwdev)
 }
 EXPORT_SYMBOL(rtw_phy_load_tables);
 
-static u8 rtw_get_channel_group(u8 channel)
+static u8 rtw_get_channel_group(u8 channel, u8 rate)
 {
        switch (channel) {
        default:
@@ -1630,6 +1630,7 @@ static u8 rtw_get_channel_group(u8 channel)
        case 106:
                return 4;
        case 14:
+               return rate <= DESC_RATE11M ? 5 : 4;
        case 108:
        case 110:
        case 112:
@@ -1881,7 +1882,7 @@ void rtw_get_tx_power_params(struct rtw_dev *rtwdev, u8 path, u8 rate, u8 bw,
        s8 *remnant = &pwr_param->pwr_remnant;
 
        pwr_idx = &rtwdev->efuse.txpwr_idx_table[path];
-       group = rtw_get_channel_group(ch);
+       group = rtw_get_channel_group(ch, rate);
 
        /* base power index for 2.4G/5G */
        if (IS_CH_2G_BAND(ch)) {