Re: [PATCH v8 4/4] can: isotp: fix state machine corruption on signal interruption

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:

Pre-existing issues:
- [High] The `so->tx_gen` mechanism fails to protect against delayed local echoes from aborted transfers if the CAN payload matches, leading to protocol pacing violations and timeouts.
- [High] Signal interruption during `wait_tx_done` causes racy state clobbering of concurrent transfers due to a stolen wakeup and lockless state update.
--

--- Patch [4]: [PATCH v8 4/4] can: isotp: fix state machine corruption on signal interruption ---
commit 7dee50f317c5e02a3ea6610e45f4dde065c0b96c
Author: Oliver Hartkopp <socketcan@hartkopp.net>

can: isotp: fix state machine corruption on signal interruption

This commit fixes a bug where a signal interrupting a thread waiting to start
a new transfer incorrectly aborts an ongoing transfer. By returning the error
directly, it avoids executing the cleanup path that forces the state back to
ISOTP_IDLE.

> diff --git a/net/can/isotp.c b/net/can/isotp.c
> index c5b44d298618b..ceb590ae058eb 100644
> --- a/net/can/isotp.c
> +++ b/net/can/isotp.c
> @@ -1047,7 +1047,7 @@ static int isotp_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
>  		/* wait for complete transmission of current pdu */
>  		err = wait_event_interruptible(so->wait, so->tx.state == ISOTP_IDLE);
>  		if (err)
> -			goto err_event_drop;
> +			return err;

[Severity: High]
This is a pre-existing issue, but does a similar race exist later in
isotp_sendmsg() when handling wait_tx_done?

If Thread A waits in wait_tx_done with a signal pending, and concurrently
the transfer finishes (setting state to ISOTP_IDLE and waking waiters),
Thread B could wake up and claim the socket locklessly:

Thread B:
    while (cmpxchg(&so->tx.state, ISOTP_IDLE, ISOTP_SENDING) != ISOTP_IDLE) {

Before Thread B increments so->tx_gen, if Thread A wakes up, evaluates its
wait condition, returns -ERESTARTSYS, and jumps to err_event_drop:

Thread A:
    if (wait_tx_done) {
        err = wait_event_interruptible(so->wait, so->tx.state == ISOTP_IDLE);
        if (err)
            goto err_event_drop;
    }

Would Thread A then incorrectly reset the state back to ISOTP_IDLE while
Thread B is actively using the socket?

Thread A in err_event_drop:
    if (READ_ONCE(so->tx_gen) == gen) {
        so->cfecho = 0;
        so->tx.state = ISOTP_IDLE;
    }

>  	}
>  
>  	/* new transfer: bump the generation (gen is used below and by

[Severity: High]
This isn't a bug introduced by this patch, but does the so->tx_gen mechanism
fully protect against delayed local echoes from aborted transfers?

If a transfer is aborted while its CAN frame is already queued, and a new
transfer begins with matching first 4 bytes of payload, could the delayed
local echo from the aborted frame arrive and be falsely validated?

In isotp_rcv_echo():

    if (so->cfecho != *(u32 *)cf->data)
        goto out_unlock;
    
    gen = READ_ONCE(so->tx_gen);
    ...
    if (so->tx.state != ISOTP_SENDING || READ_ONCE(so->tx_gen) != gen)
        goto out_unlock;

Since gen is read directly from the current socket state (so->tx_gen) rather
than being persistently attached to the SKB, does this allow a stale echo to
prematurely advance the new transfer's state machine, leading to pacing
(STmin) violations and timeouts?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260712-isotp-fixes-v8-0-c39220a4e442@hartkopp.net?part=4