[PATCH 5.15 39/72] i2c: mpc: Avoid out of bounds memory access

From: Greg Kroah-Hartman
Date: Mon Jan 10 2022 - 02:47:46 EST


From: Chris Packham <chris.packham@xxxxxxxxxxxxxxxxxxx>

commit 72a4a87da8f7bcf868b338615a814b6542f277f3 upstream.

When performing an I2C transfer where the last message was a write KASAN
would complain:

BUG: KASAN: slab-out-of-bounds in mpc_i2c_do_action+0x154/0x630
Read of size 2 at addr c814e310 by task swapper/2/0

CPU: 2 PID: 0 Comm: swapper/2 Tainted: G B 5.16.0-rc8 #1
Call Trace:
[e5ee9d50] [c08418e8] dump_stack_lvl+0x4c/0x6c (unreliable)
[e5ee9d70] [c02f8a14] print_address_description.constprop.13+0x64/0x3b0
[e5ee9da0] [c02f9030] kasan_report+0x1f0/0x204
[e5ee9de0] [c0c76ee4] mpc_i2c_do_action+0x154/0x630
[e5ee9e30] [c0c782c4] mpc_i2c_isr+0x164/0x240
[e5ee9e60] [c00f3a04] __handle_irq_event_percpu+0xf4/0x3b0
[e5ee9ec0] [c00f3d40] handle_irq_event_percpu+0x80/0x110
[e5ee9f40] [c00f3e48] handle_irq_event+0x78/0xd0
[e5ee9f60] [c00fcfec] handle_fasteoi_irq+0x19c/0x370
[e5ee9fa0] [c00f1d84] generic_handle_irq+0x54/0x80
[e5ee9fc0] [c0006b54] __do_irq+0x64/0x200
[e5ee9ff0] [c0007958] __do_IRQ+0xe8/0x1c0
[c812dd50] [e3eaab20] 0xe3eaab20
[c812dd90] [c0007a4c] do_IRQ+0x1c/0x30
[c812dda0] [c0000c04] ExternalInput+0x144/0x160
--- interrupt: 500 at arch_cpu_idle+0x34/0x60
NIP: c000b684 LR: c000b684 CTR: c0019688
REGS: c812ddb0 TRAP: 0500 Tainted: G B (5.16.0-rc8)
MSR: 00029002 <CE,EE,ME> CR: 22000488 XER: 20000000

GPR00: c10ef7fc c812de90 c80ff200 c2394718 00000001 00000001 c10e3f90 00000003
GPR08: 00000000 c0019688 c2394718 fc7d625b 22000484 00000000 21e17000 c208228c
GPR16: e3e99284 00000000 ffffffff c2390000 c001bac0 c2082288 c812df60 c001ba60
GPR24: c23949c0 00000018 00080000 00000004 c80ff200 00000002 c2348ee4 c2394718
NIP [c000b684] arch_cpu_idle+0x34/0x60
LR [c000b684] arch_cpu_idle+0x34/0x60
--- interrupt: 500
[c812de90] [c10e3f90] rcu_eqs_enter.isra.60+0xc0/0x110 (unreliable)
[c812deb0] [c10ef7fc] default_idle_call+0xbc/0x230
[c812dee0] [c00af0e8] do_idle+0x1c8/0x200
[c812df10] [c00af3c0] cpu_startup_entry+0x20/0x30
[c812df20] [c001e010] start_secondary+0x5d0/0xba0
[c812dff0] [c00028a0] __secondary_start+0x90/0xdc

This happened because we would overrun the i2c->msgs array on the final
interrupt for the I2C STOP. This didn't happen if the last message was a
read because there is no interrupt in that case. Ensure that we only
access the current message if we are not processing a I2C STOP
condition.

Fixes: 1538d82f4647 ("i2c: mpc: Interrupt driven transfer")
Reported-by: Maxime Bizon <mbizon@xxxxxxxxxx>
Signed-off-by: Chris Packham <chris.packham@xxxxxxxxxxxxxxxxxxx>
Signed-off-by: Wolfram Sang <wsa@xxxxxxxxxx>
Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
---
drivers/i2c/busses/i2c-mpc.c | 15 +++++++++------
1 file changed, 9 insertions(+), 6 deletions(-)

--- a/drivers/i2c/busses/i2c-mpc.c
+++ b/drivers/i2c/busses/i2c-mpc.c
@@ -492,7 +492,7 @@ static void mpc_i2c_finish(struct mpc_i2

static void mpc_i2c_do_action(struct mpc_i2c *i2c)
{
- struct i2c_msg *msg = &i2c->msgs[i2c->curr_msg];
+ struct i2c_msg *msg = NULL;
int dir = 0;
int recv_len = 0;
u8 byte;
@@ -501,10 +501,13 @@ static void mpc_i2c_do_action(struct mpc

i2c->cntl_bits &= ~(CCR_RSTA | CCR_MTX | CCR_TXAK);

- if (msg->flags & I2C_M_RD)
- dir = 1;
- if (msg->flags & I2C_M_RECV_LEN)
- recv_len = 1;
+ if (i2c->action != MPC_I2C_ACTION_STOP) {
+ msg = &i2c->msgs[i2c->curr_msg];
+ if (msg->flags & I2C_M_RD)
+ dir = 1;
+ if (msg->flags & I2C_M_RECV_LEN)
+ recv_len = 1;
+ }

switch (i2c->action) {
case MPC_I2C_ACTION_RESTART:
@@ -581,7 +584,7 @@ static void mpc_i2c_do_action(struct mpc
break;
}

- if (msg->len == i2c->byte_posn) {
+ if (msg && msg->len == i2c->byte_posn) {
i2c->curr_msg++;
i2c->byte_posn = 0;