Re: Oops on current Raspian when closing an SCTP connection

[Corey Minyard <minyard@acm.org>]

On Mon, Aug 17, 2020 at 10:44:57AM -0300, Marcelo Ricardo Leitner wrote:
> On Sun, Aug 16, 2020 at 06:06:24PM -0500, Corey Minyard wrote:
> > I'm seeing the following when an SCTP connection terminates.  This is on
> > Raspian on a Raspberry Pi, version is Linux version 5.4.51-v7+.  That's
> > 32-bit ARM.
> > 
> > I haven't looked into it yet, I thought I would report before trying to
> > chase anything down.  I'm not seeing it on 5.4 x86_64 systems.
> > 
> > Aug 16 17:59:01 access kernel: [510640.438008] Hardware name: BCM2835
> > Aug 16 17:59:01 access kernel: [510640.443823] PC is at sctp_ulpevent_free+0x38/0xa0 [sctp]
> > Aug 16 17:59:01 access kernel: [510640.451498] LR is at sctp_queue_purge_ulpevents+0x34/0x50 [sctp]
> 
> Not ringing a bell here. Can you pinpoint on which line this crash
> was? It seems, by the 0x8 offset and these function offsets, that this
> could be when it was trying to access event->rmem_len, but if event
> was NULL then it should have crashed earlier.
> 
>   Marcelo

I think so:

00015e38 <sctp_ulpevent_free>:
   15e38:       e1a0c00d        mov     ip, sp
   15e3c:       e92dd878        push    {r3, r4, r5, r6, fp, ip, lr, pc}
   15e40:       e24cb004        sub     fp, ip, #4
   15e44:       e52de004        push    {lr}            ; (str lr, [sp, #-4]!)
   15e48:       ebfffffe        bl      0 <__gnu_mcount_nc>

ulpevent.c:1102		if (sctp_ulpevent_is_notification(event))
   15e4c:       e1d032f0        ldrsh   r3, [r0, #32]
   15e50:       e1a05000        mov     r5, r0
   15e54:       e3530000        cmp     r3, #0
   15e58:       ba000011        blt     15ea4 <sctp_ulpevent_free+0x6c>

This is the false branch from the above (high bit isn't set).

ulpevent.c:1061		if (!skb->data_len) (just the fetch part)
   15e5c:       e5903040        ldr     r3, [r0, #64]   ; 0x40

ulpevent.c:1059		len = skb->len;
   15e60:       e590603c        ldr     r6, [r0, #60]   ; 0x3c

ulpevent.c:1061         if (!skb->data_len) (the compare part)
   15e64:       e3530000        cmp     r3, #0
   15e68:       0a000008        beq     15e90 <sctp_ulpevent_free+0x58>


ulpevent.c:1065		skb_walk_frags(skb, frag) {
skbuff.h:1395		return skb->end; (inside skb_shinfo())
   15e6c:       e5903084        ldr     r3, [r0, #132]  ; 0x84

skbuff.h:3461		iter = skb_shinfo(skb)->frag_list
   15e70:       e5934008        ldr     r4, [r3, #8] <-- Crash occurs here

   15e74:       e3540000        cmp     r4, #0
   15e78:       0a000004        beq     15e90 <sctp_ulpevent_free+0x58>
   15e7c:       e2840018        add     r0, r4, #24
   15e80:       ebfffa34        bl      14758 <sctp_ulpevent_release_frag_data>


So it appears that skb->end is NULL.

-corey