Skip to contentSkip to conversation

xtables: use exponential delay when waiting for xtables lock

PATCH In review Subash Abhinov Kasiviswanathan 3 participants 4 messages Apr 8, 2016 – Apr 28, 2016 netfilter

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 xtables: use exponential delay when waiting for xtables lock

ip[6]tables currently waits for 1 second for the xtables lock to be freed if the -w option is used. We have seen that the lock is held much less than that resulting in unnecessary delay when trying to acquire the lock. This problem is even severe in case of latency sensitive applications.

Introduce an exponential delay starting from 10ms (experimentally determined) up to a second.

---
 iptables/xshared.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)
iptables/xshared.c +8 -2
index 81c2581..d90ac13 100644
@@ -247,6 +247,7 @@ void xs_init_match(struct xtables_match *match)
 bool xtables_lock(int wait)
 {
 	int fd, waited = 0, i = 0;
+	useconds_t base_delay = 10000;
 
 	fd = open(XT_LOCK_NAME, O_CREAT, 0600);
 	if (fd < 0)
@@ -257,10 +258,15 @@ bool xtables_lock(int wait)
 			return true;
 		else if (wait >= 0 && waited >= wait)
 			return false;
-		if (++i % 2 == 0)
+		if ((++i % 2 == 0) && (base_delay >= 200000))
 			fprintf(stderr, "Another app is currently holding the xtables lock; "
 				"waiting (%ds) for it to exit...\n", waited);
 		waited++;
-		sleep(1);
+		if (base_delay > 1000000) {
+			sleep(1);
+		} else {
+			usleep(base_delay);
+			base_delay *= 2;
+		}
 	}
 }
Signature · 1 line
1.8.2.1
#1 lore raw

2016-04-08 11:07 GMT+08:00 Subash Abhinov Kasiviswanathan <[email protected]>:

Quoted 27 lines

ip[6]tables currently waits for 1 second for the xtables lock to be freed if the -w option is used. We have seen that the lock is held much less than that resulting in unnecessary delay when trying to acquire the lock. This problem is even severe in case of latency sensitive applications.

Introduce an exponential delay starting from 10ms (experimentally determined) up to a second.

---
 iptables/xshared.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)
iptables/xshared.c +1 -0
index 81c2581..d90ac13 100644
@@ -247,6 +247,7 @@ void xs_init_match(struct xtables_match *match)
 bool xtables_lock(int wait)
 {
        int fd, waited = 0, i = 0;
+       useconds_t base_delay = 10000;
 
        fd = open(XT_LOCK_NAME, O_CREAT, 0600);
        if (fd < 0)

iptables/xshared.c bool xtables_lock(int wait) · line 264

Quoted 9 lines
iptables/xshared.c +1 -2
@@ -257,10 +258,15 @@ bool xtables_lock(int wait)
                        return true;
                else if (wait >= 0 && waited >= wait)
                        return false;
-               if (++i % 2 == 0)
+               if ((++i % 2 == 0) && (base_delay >= 200000))
                        fprintf(stderr, "Another app is currently holding the xtables lock; "
                                "waiting (%ds) for it to exit...\n", waited);
                waited++;
-               sleep(1);

This break the "-w" option's semantic, i.e. if the user input "iptables -w 1", and concurrency happen,
we will just only wait 10ms and return an error.

Quoted 15 lines
iptables/xshared.c +6 -1
+               if (base_delay > 1000000) {
+                       sleep(1);
+               } else {
+                       usleep(base_delay);
+                       base_delay *= 2;
+               }
        }
 }
--

1.8.2.1

--
To unsubscribe from this list: send the line "unsubscribe netfilter-devel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#2 lore raw
Quoted 15 lines · Liping Zhang

2016-04-08 11:07 GMT+08:00 Subash Abhinov Kasiviswanathan <[email protected]>:

iptables/xshared.c bool xtables_lock(int wait) · line 264

Quoted 10 lines
iptables/xshared.c +1 -2
@@ -257,10 +258,15 @@ bool xtables_lock(int wait)
                        return true;
                else if (wait >= 0 && waited >= wait)
                        return false;
-               if (++i % 2 == 0)
+               if ((++i % 2 == 0) && (base_delay >= 200000))
                        fprintf(stderr, "Another app is currently holding the xtables lock; "
                                "waiting (%ds) for it to exit...\n", waited);
                waited++;
-               sleep(1);

This break the "-w" option's semantic, i.e. if the user input "iptables -w 1", and concurrency happen,
we will just only wait 10ms and return an error.

If there's any chance this patch can break existing setups then we can't take this.

I'd suggest you add support to express millisecond precision using a dot notation, ie.

        -w .000001

that means 10 ms.

#3 lore raw
Quoted 8 lines
Quoted 4 lines

This break the "-w" option's semantic, i.e. if the user input "iptables -w 1", and concurrency happen,
we will just only wait 10ms and return an error.

If there's any chance this patch can break existing setups then we can't take this.

Quoted 6 lines

I'd suggest you add support to express millisecond precision using a dot notation, ie.

        -w .000001

that means 10 ms.

Thanks Liping / Pablo for your comments.

I'll raise a new patchset for this.

--
Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project.