Re: OOPS: Multipath routing 2.4.17

Andi Kleen (ak@suse.de)
02 Mar 2002 00:00:56 +0100


I wrote:

>In theory yes, but the
>
>#if 1
> if (power <= 0) {
> printk(KERN_CRIT "impossible 777\n");
> return;
> }
>#endif
>
>should stop it; making it just not work, but not crash.
>If he still gets a division by zero then something else is fishy.

Ignore the objection. The division is using fi->fib_power not power,
so it is definitely racy and possible. Sorry for the brain fart.

Short term fix would be just to add a spinlock like this (untested).
I think using a new algorithm would be too risky at least for 2.4.

--- linux-work/net/ipv4/fib_semantics.c-FIBLOCK Tue Jan 15 11:05:17 2002
+++ linux-work/net/ipv4/fib_semantics.c Fri Mar 1 23:58:45 2002
@@ -35,6 +35,7 @@
#include <linux/skbuff.h>
#include <linux/netlink.h>
#include <linux/init.h>
+#include <linux/spinlock.h>

#include <net/ip.h>
#include <net/protocol.h>
@@ -45,6 +46,8 @@

#define FSprintk(a...)

+static spinlock_t fib_nh_lock = SPIN_LOCK_UNLOCKED;
+
static struct fib_info *fib_info_list;
static rwlock_t fib_info_lock = RW_LOCK_UNLOCKED;
int fib_info_cnt;
@@ -859,6 +862,8 @@
if (force)
scope = -1;

+ spin_lock_bh(&fib_nh_lock);
+
for_fib_info() {
if (local && fi->fib_prefsrc == local) {
fi->fib_flags |= RTNH_F_DEAD;
@@ -885,6 +890,7 @@
}
}
} endfor_fib_info();
+ spin_unlock_bh(&fib_nh_lock);
return ret;
}

@@ -902,6 +908,7 @@
if (!(dev->flags&IFF_UP))
return 0;

+ spin_lock_bh(&fib_nh_lock);
for_fib_info() {
int alive = 0;

@@ -924,6 +931,7 @@
ret++;
}
} endfor_fib_info();
+ spin_unlock_bh(&fib_nh_lock);
return ret;
}

@@ -937,6 +945,7 @@
struct fib_info *fi = res->fi;
int w;

+ spin_lock_bh(&fib_nh_lock);
if (fi->fib_power <= 0) {
int power = 0;
change_nexthops(fi) {
@@ -949,6 +958,7 @@
#if 1
if (power <= 0) {
printk(KERN_CRIT "impossible 777\n");
+ spin_unlock_bh(&fib_nh_lock);
return;
}
#endif
@@ -967,10 +977,13 @@
nh->nh_power--;
fi->fib_power--;
res->nh_sel = nhsel;
+ spin_unlock_bh(&fib_nh_lock);
return;
}
}
} endfor_nexthops(fi);
+
+ spin_unlock_bh(&fib_nh_lock);

#if 1
printk(KERN_CRIT "impossible 888\n");

-Andi
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/