Skip to contentSkip to conversation

sch_dsmark bug

DISCUSSION Hugo Santos 3 participants 3 messages Apr 11, 2004

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

Hi there,

I tried to contact the person whose name is in the source file but with no success. While developing some code that handles with kernel's TC objects direcly (by netlink) i found a small problem with dsmark (maybe other scheds have it too, but i haven't checked). This is in linux 2.6.4 (there is no change in this context in 2.6.5); in dsmark_init, if opt is NULL, the module will crash and TC will get stuck. the module will have a ref count of 1 so i can't unload it, and i can't change any other TC stuff since the netlink socket gets stuck when dsmark crashes. So, a simple

if (!opt) return -EINVAL;

would prevent such problems.

If you prefer i may send a patch, but since it's only a one line patch i didn't send one.

Best regards,
Hugo Santos

#1 lore raw

Looks legit nit. Why are you sending empty opts anyways? ;-> patchlet attached - you should have sent one ;-> Maybe you can inspect other qdiscs.

cheers,
jamal

Quoted 13 lines · Hugo Santos

Hi there,

I tried to contact the person whose name is in the source file but with no success. While developing some code that handles with kernel's TC objects direcly (by netlink) i found a small problem with dsmark (maybe other scheds have it too, but i haven't checked). This is in linux 2.6.4 (there is no change in this context in 2.6.5); in dsmark_init, if opt is NULL, the module will crash and TC will get stuck. the module will have a ref count of 1 so i can't unload it, and i can't change any other TC stuff since the netlink socket gets stuck when dsmark crashes. So, a simple

if (!opt) return -EINVAL;

Quoted 12 lines

would prevent such problems.

If you prefer i may send a patch, but since it's only a one line patch i didn't send one.

Best regards,
Hugo Santos

sch_dsmark.c +1 -0
@@ -321,6 +321,7 @@
 	struct rtattr *tb[TCA_DSMARK_MAX];
 	__u16 tmp;
 
+	if (!opt) return -EINVAL;
 	DPRINTK("dsmark_init(sch %p,[qdisc %p],opt %p)\n",sch,p,opt);
 	if (rtattr_parse(tb,TCA_DSMARK_MAX,RTA_DATA(opt),RTA_PAYLOAD(opt)) < 0 ||
 	    !tb[TCA_DSMARK_INDICES-1] ||
#2 lore raw

On Sun, Apr 11, 2004 at 03:43:03PM -0400, jamal told us:

Quoted 8 lines

Looks legit nit. Why are you sending empty opts anyways? ;-> patchlet attached - you should have sent one ;-> Maybe you can inspect other qdiscs.

cheers,
jamal

I took a look at the others and found this ones...

Sven

Signature · 2 lines
Linux zion 2.6.5 #1 Sun Apr 4 19:56:55 CEST 2004 i686 athlon i386 GNU/Linux
 00:43:25  up 7 days,  1:27,  2 users,  load average: 0.02, 0.05, 0.01
sched/sch_cbq.c renamed from sched/sch_cbq.c.orig +2 -1
@@ -1399,7 +1399,8 @@ static int cbq_init(struct Qdisc *sch, s
 	struct rtattr *tb[TCA_CBQ_MAX];
 	struct tc_ratespec *r;
 
-	if (rtattr_parse(tb, TCA_CBQ_MAX, RTA_DATA(opt), RTA_PAYLOAD(opt)) < 0 ||
+	if (!opt || 
+	    rtattr_parse(tb, TCA_CBQ_MAX, RTA_DATA(opt), RTA_PAYLOAD(opt)) < 0 ||
 	    tb[TCA_CBQ_RTAB-1] == NULL || tb[TCA_CBQ_RATE-1] == NULL ||
 	    RTA_PAYLOAD(tb[TCA_CBQ_RATE-1]) < sizeof(struct tc_ratespec))
 		return -EINVAL;
sched/sch_csz.c renamed from sched/sch_csz.c.orig +2 -0
@@ -763,6 +763,8 @@ static int csz_init(struct Qdisc *sch, s
 	struct tc_csz_qopt *qopt;
 	int    i;
 
+	if (!opt)
+		return -EINVAL;
 	rtattr_parse(tb, TCA_CSZ_PTAB, RTA_DATA(opt), RTA_PAYLOAD(opt));
 	if (tb[TCA_CSZ_PARMS-1] == NULL ||
 	    RTA_PAYLOAD(tb[TCA_CSZ_PARMS-1]) < sizeof(*qopt))