Re: [PATCH v2 1/2] can: dev: can_put_echo_skb(): use dev_kfree_skb_any()

[sashiko-bot@kernel.org]

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

Pre-existing issues:
- [High] The patch fails to make can_put_echo_skb() completely safe for hardware interrupt context because it overlooks the inline helper can_create_echo_skb().
- [High] The out-of-bounds error path in can_put_echo_skb() leaks the SKB.
- [High] Unsafe kfree_skb() usage remains in CAN SKB allocation helpers which are explicitly called from hardware interrupt handlers.
- [High] can_dev_dropped_skb() and can_dropped_invalid_skb() use kfree_skb() instead of dev_kfree_skb_any(), risking execution in IRQ-disabled contexts.
--

--- Patch [1]: [PATCH v2 1/2] can: dev: can_put_echo_skb(): use dev_kfree_skb_any() ---
commit af16d148470d669cda96f722b1dffb79b817c3f9
Author: Cunhao Lu <1579567540@qq.com>

can: dev: can_put_echo_skb(): use dev_kfree_skb_any()

can_put_echo_skb() is a public helper used by CAN drivers, so its callers
do not have a guaranteed IRQ context. kfree_skb() cannot be called from
hardware interrupt context or with hardware interrupts disabled.

Use dev_kfree_skb_any() for the existing skb drop paths so that they are
safe regardless of the caller's IRQ context.

> diff --git a/drivers/net/can/dev/skb.c b/drivers/net/can/dev/skb.c
> index 95fcdc1026f80..d7b5a5d17ff22 100644
> --- a/drivers/net/can/dev/skb.c
> +++ b/drivers/net/can/dev/skb.c

[Severity: High]
This isn't a bug introduced by this patch, but does the out-of-bounds error
path in can_put_echo_skb() leak the SKB?

If the bounds check fails, the function logs a BUG and returns without 
freeing the skb:

drivers/net/can/dev/skb.c:can_put_echo_skb() {
    if (idx >= priv->echo_skb_max) {
        netdev_err(dev, "%s: BUG! ...\n", __func__, idx, priv->echo_skb_max);
        return -EINVAL;
    }
    ...
}

Many CAN driver callers assume can_put_echo_skb() takes ownership of the skb
and frees it on error. Does this lead to a memory leak when the bounds check
fails?

> @@ -62,7 +62,7 @@ int can_put_echo_skb(struct sk_buff *skb, struct net_device *dev,
>  	    (skb->protocol != htons(ETH_P_CAN) &&
>  	     skb->protocol != htons(ETH_P_CANFD) &&
>  	     skb->protocol != htons(ETH_P_CANXL))) {
> -		kfree_skb(skb);
> +		dev_kfree_skb_any(skb);
>  		return 0;
>  	}
>  

[Severity: High]
This is a pre-existing issue, but does this patch fail to make 
can_put_echo_skb() completely safe for hardware interrupt context?

While dev_kfree_skb_any() is now used in the direct drop paths, the main 
execution path calls the inline helper can_create_echo_skb():

drivers/net/can/dev/skb.c:can_put_echo_skb() {
    ...
    if (!priv->echo_skb[idx]) {
        skb = can_create_echo_skb(skb);
    ...
}

can_create_echo_skb() continues to invoke kfree_skb() on allocation failure
and consume_skb() on success:

include/linux/can/skb.h:can_create_echo_skb() {
    ...
    kfree_skb(skb);
    ...
    consume_skb(skb);
    ...
}

Since both are unsafe to call from hardirq context or with interrupts
disabled, does this defeat the purpose of the patch?

> @@ -90,7 +90,7 @@ int can_put_echo_skb(struct sk_buff *skb, struct net_device *dev,
>  	} else {
>  		/* locking problem with netif_stop_queue() ?? */
>  		netdev_err(dev, "%s: BUG! echo_skb %d is occupied!\n", __func__, idx);
> -		kfree_skb(skb);
> +		dev_kfree_skb_any(skb);
>  		return -EBUSY;
>  	}
>  

[Severity: High]
This is a pre-existing issue, but do can_dev_dropped_skb() and 
can_dropped_invalid_skb() still risk execution in IRQ-disabled contexts?

These functions are called in the exact same ndo_start_xmit paths, right
before can_put_echo_skb(), and they still invoke kfree_skb() on their drop
paths:

include/linux/can/dev.h:can_dev_dropped_skb() {
    ...
invalid_skb:
    kfree_skb(skb);
    dev->stats.tx_dropped++;
    return true;
}

drivers/net/can/dev/skb.c:can_dropped_invalid_skb() {
    ...
inval_skb:
    kfree_skb(skb);
    dev->stats.tx_dropped++;
    return true;
}

This leaves the system vulnerable to the same crashes this patch attempts to
resolve.

[Severity: High]
This isn't a bug introduced by this patch, but do unsafe kfree_skb() calls
remain in CAN SKB allocation helpers that run in hardware interrupt handlers?

CAN drivers use alloc_can_skb(), alloc_canfd_skb(), and alloc_canxl_skb() in 
their RX paths, which often run in hardirq context. If memory allocation for
the CAN SKB extension fails in can_skb_ext_add(), these helpers drop the
allocated SKB using kfree_skb():

drivers/net/can/dev/skb.c:alloc_can_skb() {
    ...
    if (!csx) {
        kfree_skb(skb);
        goto out_error_cc;
    }
    ...
}

Calling kfree_skb() in hardirq context is strictly forbidden and can trigger
warnings or panics. Should these also be updated to use dev_kfree_skb_any()?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260731-master-v2-0-204d772f2e36@qq.com?part=1