Skip to contentSkip to conversation

get_proximity() crashes on interfaces with no addresses

PATCH Doug Nazar 3 participants 6 messages May 12, 2013 – May 16, 2013 Kernel automounter

lore
Share & tools
lore

Doug Nazar

author
#0 lore raw

Since commit aa6f7793 [autofs-5.0.7 - fix ipv6 proximity calculation] uses getifaddrs however it crashes on interfaces with no addresses. Fix the NULL check to ignore interfaces with no addresses.

Should it also check for the IFF_UP flag?

Doug

modules/replicated.c +2 -2
diff -ur autofs-5.0.7.orig/modules/replicated.c autofs-5.0.7/modules/replicated.c
@@ -166,7 +166,7 @@
 	this = ifa;
 	while (this) {
 		if (this->ifa_flags & IFF_POINTOPOINT ||
-		    this->ifa_addr->sa_data == NULL) {
+		    this->ifa_addr == NULL) {
 			this = this->ifa_next;
 			continue;
 		}
@@ -203,7 +203,7 @@
 	this = ifa;
 	while (this) {
 		if (this->ifa_flags & IFF_POINTOPOINT ||
-		    this->ifa_addr->sa_data == NULL) {
+		    this->ifa_addr == NULL) {
 			this = this->ifa_next;
 			continue;
 		}
#1 lore raw
Quoted 3 lines · Doug Nazar

Since commit aa6f7793 [autofs-5.0.7 - fix ipv6 proximity calculation] uses getifaddrs however it crashes on interfaces with no addresses. Fix the NULL check to ignore interfaces with no addresses.

Hi,

Your patch removes the check on ifa_addr->sa_data. I'm wondering if it's possible to have a valid ifa_addr and a NULL sa_data. Do you know? Maybe it's safer to just test both:

if (this->ifa_flags & IFF_POINTOPOINT ||

		    this->ifa_addr == NULL ||
		    this->ifa_addr->sa_data == NULL) {

this = this->ifa_next;
continue;
}

Quoted 1 line

Should it also check for the IFF_UP flag?

I think it makes sense to test it.

Thanks,
Leonardo

author
#2 lore raw
Quoted 7 lines · Leonardo Chiquitto
Quoted 3 lines · Doug Nazar

Since commit aa6f7793 [autofs-5.0.7 - fix ipv6 proximity calculation] uses getifaddrs however it crashes on interfaces with no addresses. Fix the NULL check to ignore interfaces with no addresses.

Your patch removes the check on ifa_addr->sa_data. I'm wondering if it's possible to have a valid ifa_addr and a NULL sa_data. Do you know? Maybe it's safer to just test both:

I don't see how. ifa_addr is of type struct sockaddr. The sa_data field is a char array and only has meaning after interpreting the sa_family field.

Quoted 3 lines
Quoted 1 line

Should it also check for the IFF_UP flag?

I think it makes sense to test it.

Ok, here you go.

Doug

commit d0b5d4961004a41f6881b2ac2bb32ba3002654e5

Author: Doug Nazar <[email protected]>
Date:   Sun May 12 16:22:04 2013 -0400

Since commit aa6f7793 [autofs-5.0.7 - fix ipv6 proximity calculation] get_proximity() uses getifaddrs however it crashes on interfaces with no addresses. Fix the NULL check to ignore interfaces with no addresses.

Also skip interfaces which are not currently running.

modules/replicated.c +6 -4
index 26f64b8..6dbdade 100644
@@ -165,8 +165,9 @@ static unsigned int get_proximity(struct sockaddr *host_addr)
 
 	this = ifa;
 	while (this) {
-		if (this->ifa_flags & IFF_POINTOPOINT ||
-		    this->ifa_addr->sa_data == NULL) {
+		if (!(this->ifa_flags & IFF_UP) ||
+		    this->ifa_flags & IFF_POINTOPOINT ||
+		    this->ifa_addr == NULL) {
 			this = this->ifa_next;
 			continue;
 		}
@@ -202,8 +203,9 @@ static unsigned int get_proximity(struct sockaddr *host_addr)
 
 	this = ifa;
 	while (this) {
-		if (this->ifa_flags & IFF_POINTOPOINT ||
-		    this->ifa_addr->sa_data == NULL) {
+		if (!(this->ifa_flags & IFF_UP) ||
+		    this->ifa_flags & IFF_POINTOPOINT ||
+		    this->ifa_addr == NULL) {
 			this = this->ifa_next;
 			continue;
 		}
maintainer
#3 lore raw
Quoted 11 lines · Doug Nazar
Quoted 7 lines · Leonardo Chiquitto
Quoted 3 lines · Doug Nazar

Since commit aa6f7793 [autofs-5.0.7 - fix ipv6 proximity calculation] uses getifaddrs however it crashes on interfaces with no addresses. Fix the NULL check to ignore interfaces with no addresses.

Your patch removes the check on ifa_addr->sa_data. I'm wondering if it's possible to have a valid ifa_addr and a NULL sa_data. Do you know? Maybe it's safer to just test both:

I don't see how. ifa_addr is of type struct sockaddr. The sa_data field is a char array and only has meaning after interpreting the sa_family field.

You would think so, I'll go with this.

Quoted 3 lines
Quoted 3 lines
Quoted 2 lines

Should it also check for the IFF_UP flag?

I think it makes sense to test it.

Indeed that's probably good to do too.

I'd be interested to hear if this patch functions OK with the IFF_UP check in place.

In the mean time I'll add the patch to my queue for the next commit.

Ian

author
#4 lore raw
Quoted 1 line · Ian Kent

Indeed that's probably good to do too. I'd be interested to hear if this patch functions OK with the IFF_UP check in place. In the mean time I'll add the patch to my queue for the next commit.

I've been running it for a day on a few boxes with no problems. The only additional line is a logging line I had when originally testing.

May 14 04:58:47 cockatrice automount[365]: get_proximity:168: getifaddrs: ifa_name lo, ifa_flags 0x00010049
May 14 04:58:47 cockatrice automount[365]: get_proximity:168: getifaddrs: ifa_name eql, ifa_flags 0x00000400
May 14 04:58:47 cockatrice automount[365]: get_proximity:168: getifaddrs: ifa_name eth0, ifa_flags 0x00011043
May 14 04:58:47 cockatrice automount[365]: get_proximity:168: getifaddrs: ifa_name wlan0, ifa_flags 0x00001002
May 14 04:58:47 cockatrice automount[365]: get_proximity:168: getifaddrs: ifa_name wmon0, ifa_flags 0x00011043

IFF_UP is 0x01 so that looks correct. eql was the one that it was crashing on with no ifa_addr.

Doug

maintainer
#5 lore raw
Quoted 17 lines · Doug Nazar
Quoted 3 lines · Ian Kent

Indeed that's probably good to do too. I'd be interested to hear if this patch functions OK with the IFF_UP check in place. In the mean time I'll add the patch to my queue for the next commit.

I've been running it for a day on a few boxes with no problems. The only additional line is a logging line I had when originally testing.

May 14 04:58:47 cockatrice automount[365]: get_proximity:168: 
getifaddrs: ifa_name lo, ifa_flags 0x00010049
May 14 04:58:47 cockatrice automount[365]: get_proximity:168: 
getifaddrs: ifa_name eql, ifa_flags 0x00000400
May 14 04:58:47 cockatrice automount[365]: get_proximity:168: 
getifaddrs: ifa_name eth0, ifa_flags 0x00011043
May 14 04:58:47 cockatrice automount[365]: get_proximity:168: 
getifaddrs: ifa_name wlan0, ifa_flags 0x00001002
May 14 04:58:47 cockatrice automount[365]: get_proximity:168: 
Quoted 4 lines

getifaddrs: ifa_name wmon0, ifa_flags 0x00011043

IFF_UP is 0x01 so that looks correct. eql was the one that it was crashing on with no ifa_addr.

OK, thanks for this.

Quoted 3 lines

Doug