Skip to contentSkip to conversation

bonding: Fix jiffies overflow problems (again)

PATCH Superseded Jean Delvare 3 participants 4 messages Sep 2, 2010 – Sep 6, 2010

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 bonding: Fix jiffies overflow problems (again)

From: Jiri Bohac <[email protected]>

The time_before_eq()/time_after_eq() functions operate on unsigned long and only work if the difference between the two compared values is smaller than half the range of unsigned long (31 bits on i386).

Some of the variables (slave->jiffies, dev->trans_start, dev->last_rx) used by bonding store a copy of jiffies and may not be updated for a long time. With HZ=1000, time_before_eq()/time_after_eq() will start giving bad results after ~25 days.

jiffies will never be before slave->jiffies, dev->trans_start, dev->last_rx by more than possibly a couple ticks caused by preemption of this code. This allows us to detect/prevent these overflows by replacing time_before_eq()/time_after_eq() with time_in_range().

---
 drivers/net/bonding/bond_main.c |   48 +++++++++++++++++++++++++++-------------
 1 file changed, 33 insertions(+), 15 deletions(-)
drivers/net/bonding/bond_main.c +33 -15
@@ -2798,8 +2798,12 @@ void bond_loadbalance_arp_mon(struct wor
 	 */
 	bond_for_each_slave(bond, slave, i) {
 		if (slave->link != BOND_LINK_UP) {
-			if (time_before_eq(jiffies, dev_trans_start(slave->dev) + delta_in_ticks) &&
-			    time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
+			if (time_in_range(jiffies,
+				dev_trans_start(slave->dev) - delta_in_ticks,
+				dev_trans_start(slave->dev) + delta_in_ticks) &&
+			    time_in_range(jiffies,
+				slave->dev->last_rx - delta_in_ticks,
+				slave->dev->last_rx + delta_in_ticks)) {
 
 				slave->link  = BOND_LINK_UP;
 				slave->state = BOND_STATE_ACTIVE;
@@ -2827,8 +2831,12 @@ void bond_loadbalance_arp_mon(struct wor
 			 * when the source ip is 0, so don't take the link down
 			 * if we don't know our ip yet
 			 */
-			if (time_after_eq(jiffies, dev_trans_start(slave->dev) + 2*delta_in_ticks) ||
-			    (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
+			if (!time_in_range(jiffies,
+				dev_trans_start(slave->dev) - delta_in_ticks,
+				dev_trans_start(slave->dev) + 2*delta_in_ticks) ||
+			    (!time_in_range(jiffies,
+				slave->dev->last_rx - delta_in_ticks,
+				slave->dev->last_rx + 2*delta_in_ticks))) {
 
 				slave->link  = BOND_LINK_DOWN;
 				slave->state = BOND_STATE_BACKUP;
@@ -2888,8 +2896,10 @@ static int bond_ab_arp_inspect(struct bo
 		slave->new_link = BOND_LINK_NOCHANGE;
 
 		if (slave->link != BOND_LINK_UP) {
-			if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
-					   delta_in_ticks)) {
+			if (time_in_range(jiffies,
+				slave_last_rx(bond, slave) - delta_in_ticks,
+				slave_last_rx(bond, slave) + delta_in_ticks)) {
+
 				slave->new_link = BOND_LINK_UP;
 				commit++;
 			}
@@ -2902,8 +2912,9 @@ static int bond_ab_arp_inspect(struct bo
 		 * active.  This avoids bouncing, as the last receive
 		 * times need a full ARP monitor cycle to be updated.
 		 */
-		if (!time_after_eq(jiffies, slave->jiffies +
-				   2 * delta_in_ticks))
+		if (time_in_range(jiffies,
+				  slave->jiffies - delta_in_ticks,
+				  slave->jiffies + 2 * delta_in_ticks))
 			continue;
 
 		/*
@@ -2921,8 +2932,10 @@ static int bond_ab_arp_inspect(struct bo
 		 */
 		if (slave->state == BOND_STATE_BACKUP &&
 		    !bond->current_arp_slave &&
-		    time_after(jiffies, slave_last_rx(bond, slave) +
-			       3 * delta_in_ticks)) {
+		    !time_in_range(jiffies,
+			slave_last_rx(bond, slave) - delta_in_ticks,
+			slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
+
 			slave->new_link = BOND_LINK_DOWN;
 			commit++;
 		}
@@ -2934,10 +2947,13 @@ static int bond_ab_arp_inspect(struct bo
 		 *    the bond has an IP address)
 		 */
 		if ((slave->state == BOND_STATE_ACTIVE) &&
-		    (time_after_eq(jiffies, dev_trans_start(slave->dev) +
-				    2 * delta_in_ticks) ||
-		      (time_after_eq(jiffies, slave_last_rx(bond, slave)
-				     + 2 * delta_in_ticks)))) {
+		    (!time_in_range(jiffies,
+			dev_trans_start(slave->dev) - delta_in_ticks,
+			dev_trans_start(slave->dev) + 2 * delta_in_ticks) ||
+		     (!time_in_range(jiffies,
+			slave_last_rx(bond, slave) - delta_in_ticks,
+			slave_last_rx(bond, slave) + 2 * delta_in_ticks)))) {
+
 			slave->new_link = BOND_LINK_DOWN;
 			commit++;
 		}
@@ -2964,7 +2980,9 @@ static void bond_ab_arp_commit(struct bo
 
 		case BOND_LINK_UP:
 			if ((!bond->curr_active_slave &&
-			     time_before_eq(jiffies,
+			     time_in_range(jiffies,
+					    dev_trans_start(slave->dev) -
+					    delta_in_ticks,
 					    dev_trans_start(slave->dev) +
 					    delta_in_ticks)) ||
 			    bond->curr_active_slave != slave) {
Signature · 2 lines
Jean Delvare
Suse L3
#1 lore raw
Quoted 25 lines · Jean Delvare

From: Jiri Bohac <[email protected]>

The time_before_eq()/time_after_eq() functions operate on unsigned long and only work if the difference between the two compared values is smaller than half the range of unsigned long (31 bits on i386).

Some of the variables (slave->jiffies, dev->trans_start, dev->last_rx) used by bonding store a copy of jiffies and may not be updated for a long time. With HZ=1000, time_before_eq()/time_after_eq() will start giving bad results after ~25 days.

jiffies will never be before slave->jiffies, dev->trans_start, dev->last_rx by more than possibly a couple ticks caused by preemption of this code. This allows us to detect/prevent these overflows by replacing time_before_eq()/time_after_eq() with time_in_range().

---
 drivers/net/bonding/bond_main.c |   48 +++++++++++++++++++++++++++-------------
 1 file changed, 33 insertions(+), 15 deletions(-)

drivers/net/bonding/bond_main.c void bond_loadbalance_arp_mon(struct wor · line 2803

Quoted 8 lines
drivers/net/bonding/bond_main.c +3 -2
@@ -2798,8 +2798,12 @@ void bond_loadbalance_arp_mon(struct wor
 	 */
 	bond_for_each_slave(bond, slave, i) {
 		if (slave->link != BOND_LINK_UP) {
-			if (time_before_eq(jiffies, dev_trans_start(slave->dev) + delta_in_ticks) &&
-			    time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
+			if (time_in_range(jiffies,
+				dev_trans_start(slave->dev) - delta_in_ticks,
+				dev_trans_start(slave->dev) + delta_in_ticks) &&

since dev_trans_start() might be expensive, you probably should cache its result.

Quoted 89 lines
drivers/net/bonding/bond_main.c +30 -13
+			    time_in_range(jiffies,
+				slave->dev->last_rx - delta_in_ticks,
+				slave->dev->last_rx + delta_in_ticks)) {
 
 				slave->link  = BOND_LINK_UP;
 				slave->state = BOND_STATE_ACTIVE;
@@ -2827,8 +2831,12 @@ void bond_loadbalance_arp_mon(struct wor
 			 * when the source ip is 0, so don't take the link down
 			 * if we don't know our ip yet
 			 */
-			if (time_after_eq(jiffies, dev_trans_start(slave->dev) + 2*delta_in_ticks) ||
-			    (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
+			if (!time_in_range(jiffies,
+				dev_trans_start(slave->dev) - delta_in_ticks,
+				dev_trans_start(slave->dev) + 2*delta_in_ticks) ||
+			    (!time_in_range(jiffies,
+				slave->dev->last_rx - delta_in_ticks,
+				slave->dev->last_rx + 2*delta_in_ticks))) {
 
 				slave->link  = BOND_LINK_DOWN;
 				slave->state = BOND_STATE_BACKUP;
@@ -2888,8 +2896,10 @@ static int bond_ab_arp_inspect(struct bo
 		slave->new_link = BOND_LINK_NOCHANGE;
 
 		if (slave->link != BOND_LINK_UP) {
-			if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
-					   delta_in_ticks)) {
+			if (time_in_range(jiffies,
+				slave_last_rx(bond, slave) - delta_in_ticks,
+				slave_last_rx(bond, slave) + delta_in_ticks)) {
+
 				slave->new_link = BOND_LINK_UP;
 				commit++;
 			}
@@ -2902,8 +2912,9 @@ static int bond_ab_arp_inspect(struct bo
 		 * active.  This avoids bouncing, as the last receive
 		 * times need a full ARP monitor cycle to be updated.
 		 */
-		if (!time_after_eq(jiffies, slave->jiffies +
-				   2 * delta_in_ticks))
+		if (time_in_range(jiffies,
+				  slave->jiffies - delta_in_ticks,
+				  slave->jiffies + 2 * delta_in_ticks))
 			continue;
 
 		/*
@@ -2921,8 +2932,10 @@ static int bond_ab_arp_inspect(struct bo
 		 */
 		if (slave->state == BOND_STATE_BACKUP &&
 		    !bond->current_arp_slave &&
-		    time_after(jiffies, slave_last_rx(bond, slave) +
-			       3 * delta_in_ticks)) {
+		    !time_in_range(jiffies,
+			slave_last_rx(bond, slave) - delta_in_ticks,
+			slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
+
 			slave->new_link = BOND_LINK_DOWN;
 			commit++;
 		}
@@ -2934,10 +2947,13 @@ static int bond_ab_arp_inspect(struct bo
 		 *    the bond has an IP address)
 		 */
 		if ((slave->state == BOND_STATE_ACTIVE) &&
-		    (time_after_eq(jiffies, dev_trans_start(slave->dev) +
-				    2 * delta_in_ticks) ||
-		      (time_after_eq(jiffies, slave_last_rx(bond, slave)
-				     + 2 * delta_in_ticks)))) {
+		    (!time_in_range(jiffies,
+			dev_trans_start(slave->dev) - delta_in_ticks,
+			dev_trans_start(slave->dev) + 2 * delta_in_ticks) ||
+		     (!time_in_range(jiffies,
+			slave_last_rx(bond, slave) - delta_in_ticks,
+			slave_last_rx(bond, slave) + 2 * delta_in_ticks)))) {
+
 			slave->new_link = BOND_LINK_DOWN;
 			commit++;
 		}
@@ -2964,7 +2980,9 @@ static void bond_ab_arp_commit(struct bo
 
 		case BOND_LINK_UP:
 			if ((!bond->curr_active_slave &&
-			     time_before_eq(jiffies,
+			     time_in_range(jiffies,
+					    dev_trans_start(slave->dev) -
+					    delta_in_ticks,
 					    dev_trans_start(slave->dev) +
 					    delta_in_ticks)) ||
 			    bond->curr_active_slave != slave) {
#2 lore raw

PATCH v2 1/1 bonding: Fix jiffies overflow problems (again)

From: Jiri Bohac <[email protected]>

The time_before_eq()/time_after_eq() functions operate on unsigned long and only work if the difference between the two compared values is smaller than half the range of unsigned long (31 bits on i386).

Some of the variables (slave->jiffies, dev->trans_start, dev->last_rx) used by bonding store a copy of jiffies and may not be updated for a long time. With HZ=1000, time_before_eq()/time_after_eq() will start giving bad results after ~25 days.

jiffies will never be before slave->jiffies, dev->trans_start, dev->last_rx by more than possibly a couple ticks caused by preemption of this code. This allows us to detect/prevent these overflows by replacing time_before_eq()/time_after_eq() with time_in_range().

---
 drivers/net/bonding/bond_main.c |   56 +++++++++++++++++++++++++++------------
 1 file changed, 39 insertions(+), 17 deletions(-)

Changes in V2:
* Cache dev_trans_start() result as this function can be slow (as suggested by Eric Dumazet.)

drivers/net/bonding/bond_main.c +39 -17
@@ -2797,9 +2797,15 @@ void bond_loadbalance_arp_mon(struct wor
 	 *       so it can wait
 	 */
 	bond_for_each_slave(bond, slave, i) {
+		unsigned long trans_start = dev_trans_start(slave->dev);
+
 		if (slave->link != BOND_LINK_UP) {
-			if (time_before_eq(jiffies, dev_trans_start(slave->dev) + delta_in_ticks) &&
-			    time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
+			if (time_in_range(jiffies,
+				trans_start - delta_in_ticks,
+				trans_start + delta_in_ticks) &&
+			    time_in_range(jiffies,
+				slave->dev->last_rx - delta_in_ticks,
+				slave->dev->last_rx + delta_in_ticks)) {
 
 				slave->link  = BOND_LINK_UP;
 				slave->state = BOND_STATE_ACTIVE;
@@ -2827,8 +2833,12 @@ void bond_loadbalance_arp_mon(struct wor
 			 * when the source ip is 0, so don't take the link down
 			 * if we don't know our ip yet
 			 */
-			if (time_after_eq(jiffies, dev_trans_start(slave->dev) + 2*delta_in_ticks) ||
-			    (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
+			if (!time_in_range(jiffies,
+				trans_start - delta_in_ticks,
+				trans_start + 2 * delta_in_ticks) ||
+			    !time_in_range(jiffies,
+				slave->dev->last_rx - delta_in_ticks,
+				slave->dev->last_rx + 2 * delta_in_ticks)) {
 
 				slave->link  = BOND_LINK_DOWN;
 				slave->state = BOND_STATE_BACKUP;
@@ -2883,13 +2893,16 @@ static int bond_ab_arp_inspect(struct bo
 {
 	struct slave *slave;
 	int i, commit = 0;
+	unsigned long trans_start;
 
 	bond_for_each_slave(bond, slave, i) {
 		slave->new_link = BOND_LINK_NOCHANGE;
 
 		if (slave->link != BOND_LINK_UP) {
-			if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
-					   delta_in_ticks)) {
+			if (time_in_range(jiffies,
+				slave_last_rx(bond, slave) - delta_in_ticks,
+				slave_last_rx(bond, slave) + delta_in_ticks)) {
+
 				slave->new_link = BOND_LINK_UP;
 				commit++;
 			}
@@ -2902,8 +2915,9 @@ static int bond_ab_arp_inspect(struct bo
 		 * active.  This avoids bouncing, as the last receive
 		 * times need a full ARP monitor cycle to be updated.
 		 */
-		if (!time_after_eq(jiffies, slave->jiffies +
-				   2 * delta_in_ticks))
+		if (time_in_range(jiffies,
+				  slave->jiffies - delta_in_ticks,
+				  slave->jiffies + 2 * delta_in_ticks))
 			continue;
 
 		/*
@@ -2921,8 +2935,10 @@ static int bond_ab_arp_inspect(struct bo
 		 */
 		if (slave->state == BOND_STATE_BACKUP &&
 		    !bond->current_arp_slave &&
-		    time_after(jiffies, slave_last_rx(bond, slave) +
-			       3 * delta_in_ticks)) {
+		    !time_in_range(jiffies,
+			slave_last_rx(bond, slave) - delta_in_ticks,
+			slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
+
 			slave->new_link = BOND_LINK_DOWN;
 			commit++;
 		}
@@ -2933,11 +2949,15 @@ static int bond_ab_arp_inspect(struct bo
 		 * - (more than 2*delta since receive AND
 		 *    the bond has an IP address)
 		 */
+		trans_start = dev_trans_start(slave->dev);
 		if ((slave->state == BOND_STATE_ACTIVE) &&
-		    (time_after_eq(jiffies, dev_trans_start(slave->dev) +
-				    2 * delta_in_ticks) ||
-		      (time_after_eq(jiffies, slave_last_rx(bond, slave)
-				     + 2 * delta_in_ticks)))) {
+		    (!time_in_range(jiffies,
+			trans_start - delta_in_ticks,
+			trans_start + 2 * delta_in_ticks) ||
+		     !time_in_range(jiffies,
+			slave_last_rx(bond, slave) - delta_in_ticks,
+			slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
+
 			slave->new_link = BOND_LINK_DOWN;
 			commit++;
 		}
@@ -2956,6 +2976,7 @@ static void bond_ab_arp_commit(struct bo
 {
 	struct slave *slave;
 	int i;
+	unsigned long trans_start;
 
 	bond_for_each_slave(bond, slave, i) {
 		switch (slave->new_link) {
@@ -2963,10 +2984,11 @@ static void bond_ab_arp_commit(struct bo
 			continue;
 
 		case BOND_LINK_UP:
+			trans_start = dev_trans_start(slave->dev);
 			if ((!bond->curr_active_slave &&
-			     time_before_eq(jiffies,
-					    dev_trans_start(slave->dev) +
-					    delta_in_ticks)) ||
+			     time_in_range(jiffies,
+					   trans_start - delta_in_ticks,
+					   trans_start + delta_in_ticks)) ||
 			    bond->curr_active_slave != slave) {
 				slave->link = BOND_LINK_UP;
 				bond->current_arp_slave = NULL;
Signature · 2 lines
Jean Delvare
Suse L3
maintainer Applied
#3 lore raw

From: Jean Delvare <[email protected]>
Date: Thu, 2 Sep 2010 17:45:54 +0200

Quoted 11 lines

From: Jiri Bohac <[email protected]>

The time_before_eq()/time_after_eq() functions operate on unsigned long and only work if the difference between the two compared values is smaller than half the range of unsigned long (31 bits on i386).

Some of the variables (slave->jiffies, dev->trans_start, dev->last_rx) used by bonding store a copy of jiffies and may not be updated for a long time. With HZ=1000, time_before_eq()/time_after_eq() will start giving bad results after ~25 days.

Quoted 7 lines

jiffies will never be before slave->jiffies, dev->trans_start, dev->last_rx by more than possibly a couple ticks caused by preemption of this code. This allows us to detect/prevent these overflows by replacing time_before_eq()/time_after_eq() with time_in_range().

Applied, thanks.