Re: [PATCH] HID: mcp2221: reject short input reports

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 4 potential issue(s) to consider:

Pre-existing issues:
- [Critical] Use-After-Free and kernel stack corruption due to un-cleared mcp->rxbuf on timeout.
- [Critical] Stack buffer overflow due to using uninitialized memory as I2C read length.
- [High] Delayed USB reports indiscriminately complete pending waits, causing state desynchronization.
- [High] Static retries variable underflow causes infinite loop of workqueue rescheduling.
--

commit 421d556dea19ceb8f40a640c782b6801d60f43cd
Author: Yousef Alhouseen <alhouseenyousef@gmail.com>

HID: mcp2221: reject short input reports

The MCP2221 raw-event callback reads fixed offsets from the 64-byte input
report, including data[50] and a variable payload beginning at data[4].
This rejects reports whose size does not match the protocol and completes
the pending command with -EMSGSIZE.

Closes: https://syzkaller.appspot.com/bug?extid=1018672fe70298606e5f

> diff --git a/drivers/hid/hid-mcp2221.c b/drivers/hid/hid-mcp2221.c
> index e4ddd8e9293b6..311b51c17b4e4 100644
> --- a/drivers/hid/hid-mcp2221.c
> +++ b/drivers/hid/hid-mcp2221.c
> @@ -861,6 +861,12 @@ static int mcp2221_raw_event(struct hid_device *hdev,
>  	u8 *buf;
>  	struct mcp2221 *mcp = hid_get_drvdata(hdev);
>  
> +	if (size != sizeof(mcp->txbuf)) {
> +		mcp->status = -EMSGSIZE;
> +		complete(&mcp->wait_in_report);
> +		return 1;
> +	}
> +

[Severity: High]
This is a pre-existing issue, but does mcp2221_raw_event() indiscriminately
complete pending waits for any incoming report without matching it to the
pending request?

If a previous request times out, and a delayed USB response arrives later,
it appears it could prematurely wake up a new command's wait queue with
incorrect status and data:

drivers/hid/hid-mcp2221.c:mcp2221_raw_event() {
	...
	case MCP2221_I2C_WR_DATA:
	...
		complete(&mcp->wait_in_report);
		break;
	...
}

[Severity: Critical]
This isn't a bug introduced by this patch, but can this lead to stack
corruption if a request times out?

In mcp_i2c_smbus_read(), if mcp_send_data_req_status() fails or times out,
it returns immediately without clearing mcp->rxbuf:

drivers/hid/hid-mcp2221.c:mcp_i2c_smbus_read() {
	...
	ret = mcp_send_data_req_status(mcp, mcp->txbuf, 4);
	if (ret)
		return ret;
	...
}

Since mcp->rxbuf can point to a caller-provided stack buffer (like
union i2c_smbus_data), the stack frame is popped upon return. If a delayed
USB response arrives later, wouldn't mcp2221_raw_event() blindly write into
this dangling mcp->rxbuf pointer?

drivers/hid/hid-mcp2221.c:mcp2221_raw_event() {
	...
			if (data[2] == MCP2221_I2C_READ_COMPL ||
			    data[2] == MCP2221_I2C_READ_PARTIAL) {
				...
				buf = mcp->rxbuf;
				memcpy(&buf[mcp->rxbuf_idx], &data[4], data[3]);
	...
}

[Severity: Critical]
This is a pre-existing issue, but does mcp_smbus_xfer() use uninitialized
stack memory as the I2C read length during a block read?

drivers/hid/hid-mcp2221.c:mcp_smbus_xfer() {
	...
			ret = mcp_i2c_smbus_read(mcp, NULL,
						MCP2221_I2C_RD_RPT_START,
						addr, data->block[0] + 1,
						data->block);
	...
}

The union i2c_smbus_data provided by the caller is uninitialized stack memory
on read. This configures mcp->rxbuf_size to an arbitrarily large value. Could
this lead to a massive stack buffer overflow when mcp2221_raw_event() copies
up to this value into the statically sized 34-byte data->block stack buffer?

[Severity: High]
This is a pre-existing issue, but could the statically scoped retries variable
in mcp_init_work() cause an infinite reschedule loop?

drivers/hid/hid-mcp2221.c:mcp_init_work() {
	static int retries = 5;
	...
reschedule_task:
	...
	if (!retries--)
		return;

	/* Device is not ready to read SRAM or FLASH data, try again */
	schedule_delayed_work(&mcp->init_work, msecs_to_jiffies(100));
}

Since retries is static, when initialization fails 6 times, it reaches 0,
returns, and decrements to -1. On a subsequent failure, retries starts at -1.
Because !(-1) is false, it bypasses the return, decrements to -2, and
unconditionally schedules the delayed work every 100ms forever.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260628093658.43445-1-alhouseenyousef@gmail.com?part=1