Kurt Van Dijck wrote: > On Mon, Dec 14, 2009 at 08:57:37PM +0100, Oliver Hartkopp wrote: >> Kurt Van Dijck wrote: >>>>>> http://lxr.linux.no/#linux+v2.6.32/net/can/af_can.c#L218 >>>>> Wolfgang, >>>>> >>>>> To address your concern, would a construction like this suit, when >>>>> fitted in Oliver's proposal? >>>>> >>>>> inline int no_can_skb((struct sk_buff *skb) >>>>> { >>>>> struct can_frame *cf = (struct can_frame *)skb->data >>>>> >>>>> if ((skb->len != sizeof(*cf)) || ((cf->can_dlc > 8)) { >>>>> if (skb->sk && !sock_flag(skb->sk, SOCK_DEAD)) { >>>>> skb->sk->sk_err = EINVAL; >>>>> skb->sk->sk_error_report(skb->sk); /* can this block?*/ >>>>> } >>>>> >>>>> WARN_ONCE(1, "non conform skbuf: ..."); >>>>> "Dropped non conform skbuf: len %d, can_dlc %d\n", >>>>> skb->len, cf->can_dlc); >>>>> return 1; >>>>> } >>>>> return 0; >>>>> } >>>> My primary concern is about using *WARN_ONCE*. The BUG, WARN, functions >>>> and friends indicate to the user that there is a problem with the >>>> kernel, e.g. a bug and therefore I prefer a simple dev_err(). Also the >>> I see. >>> I agree that a dev_err seems more appropriate here. >>> Since this is TX path, kernel message flood is with respect to local >>> activity? >> Indeed i would like to have a PRINTK_ONCE() or DEV_ERR_ONCE() :-) >> >> >>>> word "skbuf" does say little to the normal Linux users. I find >>>> s/skbuf/packet/ more intuitive. Of course, if there is a better way to >>> ack. >>>> inform the user we should use it. Unfortunately, I can't tell if your >>>> approach will work. >>> I went into the code, in net/core/sock.c:sock_def_error_report. >>> that looks like atomic code to me, so the above sk_error_report() stuff >>> should work. >>> no_can_skb() would not even need a dev_err in that case. >> Using sk_error_report() and leaving out dev_err() is an interesting idea! >> >> Is that a usual way to provide this kind of error notification, e.g. for >> broken PF_PACKET packets? > I'm not familiar with PF_PACKET sockets. > > I encountered this sk_error_report() during my current development. > It appears to be _the_ method of saying > "something got in error", which must first be set by ->sk_err.
I wonder if socket layer error reporting is the right thing on netdriver level. ~/net-2.6/drivers/net$ find . -name \*.c | xargs grep sk_error returns nothing - and i'm sure accessing skb->sk->sk_error_report(skb->sk) on driver level will bounce quite hard on netdev-ML ;-) Looks like, we need to come back to the first approach. Regards, Oliver _______________________________________________ Socketcan-core mailing list [email protected] https://lists.berlios.de/mailman/listinfo/socketcan-core
