: bug fix in multipath drr code.

PATCH Applied pravin 3 participants 6 messages May 23, 2005 – May 26, 2005

Subsystems (3)

Networking Network drivers IPv4/IPv6

lore

Share & tools

lore

pravin

author

2005-05-23 12:26 UTC #0 lore raw

PATCH 1/1 : bug fix in multipath drr code.

hi
AFAIU, there is a race condition in multipath_drr code,
these code paths try to access & change last_selection variable without any synchronization.

Please correct me if I am wrong.

Code Path - 1

__ip_route_output_key(...)
{
...
calls drr_select_route

      if(FLOWI_FLAG_MULTIPATHOLDROUTE set){
          check last_selection.
          return  last_selection //so it can return NULL  pointer
      }

....
}

Code Path - 2

rt_secret_rebuild()
{
.....

rt_run_flush(..);
     ->rt_free(rth);
             -> multipath_remove(rt)
                     ->drr_remove()
                             reset last_selection.

....
}

Attached patch also fixes bug in function drr_init ::
multipath_alg_register(..) is called with wrong algorithm ID.

Regards
Pravin.

Signed-off by: Pravin B. Shelar <[email protected]>

net/ipv4/multipath_drr.c +8 -6

Index: linux-2.6.12-rc4/net/ipv4/multipath_drr.c
===================================================================
@@ -145,11 +145,13 @@
 	int cur_min_devidx = -1;
 
        	/* if necessary and possible utilize the old alternative */
-	if ((flp->flags & FLOWI_FLAG_MULTIPATHOLDROUTE) != 0 &&
-	    last_selection != NULL) {
-		result = last_selection;
-		*rp = result;
-		return;
+	if ((flp->flags & FLOWI_FLAG_MULTIPATHOLDROUTE) != 0 ) {
+		struct rtable *last_result = last_selection;
+		if(last_result != NULL &&
+		   multipath_comparekeys(&last_result->fl, flp)) {
+			*rp = last_result;
+			return;
+		}
 	}
 
 	/* 1. make sure all alt. nexthops have the same GC related data */

@@ -244,7 +246,7 @@
 	if (err)
 		return err;
 
-	err = multipath_alg_register(&drr_ops, IP_MP_ALG_RR);
+	err = multipath_alg_register(&drr_ops, IP_MP_ALG_DRR);
 	if (err)
 		goto fail;

net/ipv4/multipath_rr.c +7 -4

Index: linux-2.6.12-rc4/net/ipv4/multipath_rr.c
===================================================================
@@ -64,10 +64,13 @@
 	int min_use = -1;
 
 	/* if necessary and possible utilize the old alternative */
-	if ((flp->flags & FLOWI_FLAG_MULTIPATHOLDROUTE) != 0 &&
-	    last_used != NULL) {
-		result = last_used;
-		goto out;
+	if ((flp->flags & FLOWI_FLAG_MULTIPATHOLDROUTE) != 0 ) {
+		struct rtable *last_result = last_used;
+		if(last_result != NULL &&
+		   multipath_comparekeys(&last_result->fl, flp)) {
+			result = last_result;
+			goto out;
+		}
 	}
 
 	/* 1. make sure all alt. nexthops have the same GC related data

Herbert Xu

2005-05-23 23:22 UTC #1 lore raw

Quoted 3 lines · pravin

AFAIU, there is a race condition in multipath_drr code,
these code paths try to access & change last_selection variable without any synchronization.

Yes you're right.

Quoted 11 lines

Index: linux-2.6.12-rc4/net/ipv4/multipath_drr.c
===================================================================
net/ipv4/multipath_drr.c +0 -3
@@ -145,11 +145,13 @@
 	int cur_min_devidx = -1;
 
        	/* if necessary and possible utilize the old alternative */
-	if ((flp->flags & FLOWI_FLAG_MULTIPATHOLDROUTE) != 0 &&
-	    last_selection != NULL) {
-		result = last_selection;

net/ipv4/multipath_drr.c line 155

Quoted 10 lines

net/ipv4/multipath_drr.c +7 -2

-		*rp = result;
-		return;
+	if ((flp->flags & FLOWI_FLAG_MULTIPATHOLDROUTE) != 0 ) {
+		struct rtable *last_result = last_selection;
+		if(last_result != NULL &&
+		   multipath_comparekeys(&last_result->fl, flp)) {
+			*rp = last_result;
+			return;
+		}
 	}

You don't need all this code. All you need to do is do result = last_selection before the if condition and then check result in the if condition instead of last_selection.

The multipath_comparekeys isn't necessary either.

Cheers,

Signature · 4 lines

Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

pravin b shelar

author

2005-05-24 06:46 UTC #2 lore raw

Quoted 24 lines · Herbert Xu

Quoted 13 lines · pravin

       	/* if necessary and possible utilize the old alternative */

net/ipv4/multipath_drr.c +7 -5

-	if ((flp->flags & FLOWI_FLAG_MULTIPATHOLDROUTE) != 0 &&
-	    last_selection != NULL) {
-		result = last_selection;
-		*rp = result;
-		return;
+	if ((flp->flags & FLOWI_FLAG_MULTIPATHOLDROUTE) != 0 ) {
+		struct rtable *last_result = last_selection;
+		if(last_result != NULL &&
+		   multipath_comparekeys(&last_result->fl, flp)) {
+			*rp = last_result;
+			return;
+		}

Quoted 3 lines

You don't need all this code. All you need to do is do result = last_selection before the if condition and then check result in the if condition instead of last_selection.

Quoted 4 lines

The multipath_comparekeys isn't necessary either.

In concurrent invocations of drr_select_route() last_selection will change to different route.
So in that case we can not use last_selection route on basis of FLOWI_FLAG_MULTIPATHOLDROUTE flag only, since old
route might be totally different due another invocation of same function. So, I think multipath_comparekeys is necessary.

Please correct me if I am wrong.

Regards,
Pravin.

Herbert Xu

2005-05-24 07:11 UTC #3 lore raw

Quoted 9 lines · pravin b shelar

Quoted 2 lines
The multipath_comparekeys isn't necessary either.
In concurrent invocations of drr_select_route() last_selection will change to different route.
So in that case we can not use last_selection route on basis of FLOWI_FLAG_MULTIPATHOLDROUTE flag only, since old
route might be totally different due another invocation of same function. So, I think multipath_comparekeys is necessary.

Indeed. In fact this whole MULTIPATHOLDROUTE thing can't possibly work at all. I'd suggest that it be removed.

Cheers,

Signature · 4 lines

Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

pravin b shelar

author

2005-05-24 09:57 UTC #4 lore raw

Quoted 7 lines · Herbert Xu

On Tue, May 24, 2005 at 12:16:38PM +0530, pravin b shelar wrote:
In fact this whole MULTIPATHOLDROUTE thing can't possibly work at all. I'd suggest that it be removed.

So I think we should ignore MULTIPATHOLDROUTE flag
in drr and rr multipath algorithms as done in random multipath algorithm.

Attached patch does same thing.

Regards,
Pravin.

Signed-off by: Pravin B. Shelar <[email protected]>

include/net/route.h +0 -3

Index: linux-2.6.12-rc4/include/net/route.h
===================================================================
@@ -181,9 +181,6 @@
 		memcpy(&fl, &(*rp)->fl, sizeof(fl));
 		fl.fl_ip_sport = sport;
 		fl.fl_ip_dport = dport;
-#if defined(CONFIG_IP_ROUTE_MULTIPATH_CACHED)
-		fl.flags |= FLOWI_FLAG_MULTIPATHOLDROUTE;
-#endif
 		ip_rt_put(*rp);
 		*rp = NULL;
 		return ip_route_output_flow(rp, &fl, sk, 0);

net/ipv4/multipath_drr.c +1 ~~-17~~

Index: linux-2.6.12-rc4/net/ipv4/multipath_drr.c
===================================================================
@@ -57,7 +57,6 @@
 
 static struct multipath_device state[MULTIPATH_MAX_DEVICECANDIDATES];
 static DEFINE_SPINLOCK(state_lock);
-static struct rtable *last_selection = NULL;
 
 static int inline __multipath_findslot(void)
 {

@@ -111,11 +110,6 @@
 	.notifier_call	= drr_dev_event,
 };
 
-static void drr_remove(struct rtable *rt)
-{
-	if (last_selection == rt)
-		last_selection = NULL;
-}
 
 static void drr_safe_inc(atomic_t *usecount)
 {

@@ -144,14 +138,6 @@
 	int devidx = -1;
 	int cur_min_devidx = -1;
 
-       	/* if necessary and possible utilize the old alternative */
-	if ((flp->flags & FLOWI_FLAG_MULTIPATHOLDROUTE) != 0 &&
-	    last_selection != NULL) {
-		result = last_selection;
-		*rp = result;
-		return;
-	}
-
 	/* 1. make sure all alt. nexthops have the same GC related data */
 	/* 2. determine the new candidate to be returned */
 	result = NULL;

@@ -229,12 +215,10 @@
 	}
 
 	*rp = result;
-	last_selection = result;
 }
 
 static struct ip_mp_alg_ops drr_ops = {
 	.mp_alg_select_route	=	drr_select_route,
-	.mp_alg_remove		=	drr_remove,
 };
 
 static int __init drr_init(void)

@@ -244,7 +228,7 @@
 	if (err)
 		return err;
 
-	err = multipath_alg_register(&drr_ops, IP_MP_ALG_RR);
+	err = multipath_alg_register(&drr_ops, IP_MP_ALG_DRR);
 	if (err)
 		goto fail;

net/ipv4/multipath_rr.c +0 ~~-20~~

Index: linux-2.6.12-rc4/net/ipv4/multipath_rr.c
===================================================================
@@ -47,29 +47,12 @@
 #include <net/checksum.h>
 #include <net/ip_mp_alg.h>
 
-#define MULTIPATH_MAX_CANDIDATES 40
-
-static struct rtable* last_used = NULL;
-
-static void rr_remove(struct rtable *rt)
-{
-	if (last_used == rt)
-		last_used = NULL;
-}
-
 static void rr_select_route(const struct flowi *flp,
 			    struct rtable *first, struct rtable **rp)
 {
 	struct rtable *nh, *result, *min_use_cand = NULL;
 	int min_use = -1;
 
-	/* if necessary and possible utilize the old alternative */
-	if ((flp->flags & FLOWI_FLAG_MULTIPATHOLDROUTE) != 0 &&
-	    last_used != NULL) {
-		result = last_used;
-		goto out;
-	}
-
 	/* 1. make sure all alt. nexthops have the same GC related data
 	 * 2. determine the new candidate to be returned
 	 */

@@ -90,15 +73,12 @@
 	if (!result)
 		result = first;
 
-out:
-	last_used = result;
 	result->u.dst.__use++;
 	*rp = result;
 }
 
 static struct ip_mp_alg_ops rr_ops = {
 	.mp_alg_select_route	=	rr_select_route,
-	.mp_alg_remove		=	rr_remove,
 };
 
 static int __init rr_init(void)

David S. Miller

maintainer Applied

2005-05-26 22:13 UTC #5 lore raw

From: pravin b shelar <[email protected]>
Date: Tue, 24 May 2005 15:27:16 +0530

Quoted 4 lines

So I think we should ignore MULTIPATHOLDROUTE flag
in drr and rr multipath algorithms as done in random multipath algorithm.
Attached patch does same thing.

I've applied this patch, thanks.