support for per/namespace routing cache cleanup

/proc/sys/net/route/flush should be accessible inside the net namespace. Though, the complete opening of this file will result in a DoS or significant entire host slowdown if a namespace process will continually flush routes.

This patch introduces per/namespace route flush facility.

Each namespace wanted to flush a cache copies global generation count to itself and starts the timer. The cache is dropped for a specific namespace iff the namespace counter is greater or equal global ones.

So, in general, unwanted namespaces do nothing. They hold very old low counter and they are unaffected by the requested cleanup.

Signed-of-by Denis V. Lunev <[email protected]>

include/net/net_namespace.h +2 -0

index 85abf14..b492ce8 100644
@@ -143,6 +143,8 @@ struct net {
 
 	/* iptable_filter.c */
 	struct xt_table		*ip_packet_filter;
+
+	unsigned long		rt_flush_required;
 };
 
 extern struct net init_net;

net/ipv4/route.c +78 -2

index f9a59ff..413587c 100644
@@ -91,6 +91,7 @@
 #include <linux/jhash.h>
 #include <linux/rcupdate.h>
 #include <linux/times.h>
+#include <linux/nsproxy.h>
 #include <net/net_namespace.h>
 #include <net/protocol.h>
 #include <net/ip.h>

@@ -642,6 +643,51 @@ static void rt_check_expire(unsigned long dummy)
 	mod_timer(&rt_periodic_timer, jiffies + ip_rt_gc_interval);
 }
 
+
+static DEFINE_SPINLOCK(rt_flush_lock);
+
+#ifdef CONFIG_NET_NS
+static unsigned long rt_flush_gen;
+
+/* called under rt_flush_lock */
+static void rt_flush_required_set(struct net *net)
+{
+	/*
+	 * If the global generation rt_flush_gen is equal to G, then
+	 * the pass considering entries labelled by G is yet to come.
+	 */
+	net->rt_flush_required = rt_flush_gen;
+}
+
+static unsigned long rt_flush_required_reset(void)
+{
+	unsigned long g;
+
+	spin_lock_bh(&rt_flush_lock);
+	g = rt_flush_gen++;
+	spin_unlock_bh(&rt_flush_lock);
+	return g;
+}
+
+static int rt_flush_required_check(struct net *net, unsigned long gen)
+{
+	/* can be checked without the lock */
+	return net->rt_flush_required >= gen;
+}
+
+#else
+
+static void rt_flush_required_reset(struct net *net)
+{
+}
+
+static unsigned long rt_flush_required_reset(void)
+{
+	return 0;
+}
+#endif
+
+
 /* This can run from both BH and non-BH contexts, the latter
  * in the case of a forced flush event.
  */

@@ -649,16 +695,46 @@ static void rt_run_flush(unsigned long dummy)
 {
 	int i;
 	struct rtable *rth, *next;
+	unsigned long gen;
 
 	rt_deadline = 0;
 
 	get_random_bytes(&rt_hash_rnd, 4);
+	gen = rt_flush_required_reset();
 
 	for (i = rt_hash_mask; i >= 0; i--) {
+#ifdef CONFIG_NET_NS
+		struct rtable **prev, *p, *tail;
+
+		spin_lock_bh(rt_hash_lock_addr(i));
+		rth = rt_hash_table[i].chain;
+
+		/* defer releasing the head of the list after spin_unlock */
+		for (tail = rth; tail; tail = tail->u.dst.rt_next)
+			if (!rt_flush_required_check(tail->fl.fl_net, gen))
+				break;
+		if (rth != tail)
+			rt_hash_table[i].chain = tail;
+
+		/* call rt_free on entries after the tail requiring flush */
+		prev = &rt_hash_table[i].chain;
+		for (p = *prev; p; p = next) {
+			next = p->u.dst.rt_next;
+			if (!rt_flush_required_check(p->fl.fl_net, gen)) {
+				prev = &p->u.dst.rt_next;
+			} else {
+				*prev = next;
+				rt_free(p);
+			}
+		}
+#else
 		spin_lock_bh(rt_hash_lock_addr(i));
 		rth = rt_hash_table[i].chain;
 		if (rth)
 			rt_hash_table[i].chain = NULL;
+		tail = NULL;
+
+#endif
 		spin_unlock_bh(rt_hash_lock_addr(i));
 
 		for (; rth; rth = next) {

@@ -668,8 +744,6 @@ static void rt_run_flush(unsigned long dummy)
 	}
 }
 
-static DEFINE_SPINLOCK(rt_flush_lock);
-
 void rt_cache_flush(int delay)
 {
 	unsigned long now = jiffies;

@@ -697,6 +771,8 @@ void rt_cache_flush(int delay)
 			delay = tmo;
 	}
 
+	rt_flush_required_set(current->nsproxy->net_ns);
+
 	if (delay <= 0) {
 		spin_unlock_bh(&rt_flush_lock);
 		rt_run_flush(0);

Quoted 11 lines · Denis V. Lunev

/proc/sys/net/route/flush should be accessible inside the net namespace. Though, the complete opening of this file will result in a DoS or significant entire host slowdown if a namespace process will continually flush routes.

This patch introduces per/namespace route flush facility.

Each namespace wanted to flush a cache copies global generation count to itself and starts the timer. The cache is dropped for a specific namespace iff the namespace counter is greater or equal global ones.

Quoted 4 lines

So, in general, unwanted namespaces do nothing. They hold very old low counter and they are unaffected by the requested cleanup.

Signed-of-by Denis V. Lunev <[email protected]>

That's right and that will happen when manipulating ip addresses of the network devices too. But I am not confortable with your patchset. It touches the routing flush function too hardly and it uses current->nsproxy->net_ns.

IMHO we should have two flush functions. One taking a network namespace parameter and one without the network namespace parameter. The first one is called when a write to /proc/sys/net/ipv4/route/flush is done (we must use the network namespace of the writer) or when a interface address is changed or shutdown|up. The last one is called by the timer, so we have a global timer flushing the routing cache for all the namespaces.

Quoted 12 lines · Daniel Lezcano

Quoted 11 lines · Denis V. Lunev

/proc/sys/net/route/flush should be accessible inside the net namespace. Though, the complete opening of this file will result in a DoS or significant entire host slowdown if a namespace process will continually flush routes.

This patch introduces per/namespace route flush facility.

Each namespace wanted to flush a cache copies global generation count to itself and starts the timer. The cache is dropped for a specific namespace iff the namespace counter is greater or equal global ones.

Quoted 8 lines

Quoted 4 lines

So, in general, unwanted namespaces do nothing. They hold very old low counter and they are unaffected by the requested cleanup.

Signed-of-by Denis V. Lunev <[email protected]>

That's right and that will happen when manipulating ip addresses of the network devices too. But I am not confortable with your patchset. It touches the routing flush function too hardly and it uses current->nsproxy->net_ns.

IMHO we should have two flush functions. One taking a network namespace parameter and one without the network namespace parameter. The first one is called when a write to /proc/sys/net/ipv4/route/flush is done (we must use the network namespace of the writer) or when a interface address is changed or shutdown|up. The last one is called by the timer, so we have a global timer flushing the routing cache for all the namespaces.

we can't :( The unfortunate thing is that the actual cleanup is called indirectly and asynchronously. The user _schedule_ the garbage collector to run NOW and we are moving over a large routing cache. Really large.

The idea to iterate over the list of each namespace to flush is bad. We are in atomic context. The list is protected by the mutex.

The idea of several timers (per namespace) is also bad. You will iterate over large cache several times.

No other acceptable way here for me :(.

As for "the trigger" - rt_cache_flush, looks like you are right. We should pass namespace as a parameter. This should be done as a separate patch.

Quoted 24 lines · Denis V. Lunev

Quoted 12 lines · Daniel Lezcano

Quoted 11 lines · Denis V. Lunev

/proc/sys/net/route/flush should be accessible inside the net namespace. Though, the complete opening of this file will result in a DoS or significant entire host slowdown if a namespace process will continually flush routes.

This patch introduces per/namespace route flush facility.

Each namespace wanted to flush a cache copies global generation count to itself and starts the timer. The cache is dropped for a specific namespace iff the namespace counter is greater or equal global ones.

Quoted 8 lines

Quoted 4 lines

So, in general, unwanted namespaces do nothing. They hold very old low counter and they are unaffected by the requested cleanup.

Signed-of-by Denis V. Lunev <[email protected]>

That's right and that will happen when manipulating ip addresses of the network devices too. But I am not confortable with your patchset. It touches the routing flush function too hardly and it uses current->nsproxy->net_ns.

IMHO we should have two flush functions. One taking a network namespace parameter and one without the network namespace parameter. The first one is called when a write to /proc/sys/net/ipv4/route/flush is done (we must use the network namespace of the writer) or when a interface address is changed or shutdown|up. The last one is called by the timer, so we have a global timer flushing the routing cache for all the namespaces.

we can't :( The unfortunate thing is that the actual cleanup is called indirectly and asynchronously. The user _schedule_ the garbage collector to run NOW and we are moving over a large routing cache. Really large.

Quoted 7 lines

The idea to iterate over the list of each namespace to flush is bad. We are in atomic context. The list is protected by the mutex.

The idea of several timers (per namespace) is also bad. You will iterate over large cache several times.

No other acceptable way here for me :(.

As for "the trigger" - rt_cache_flush, looks like you are right. We should pass namespace as a parameter. This should be done as a separate patch.

If we change:

    rt_cache_flush(struct net *net, int delay);

and inside we call rt_cache_flush((unsigned long)net);

And then we check in the rt_run_flush function,

    struct net *net = (struct net *)dummy;

    ...
    for (i = rt_hash_mask; i >= 0; i--) {
        ...
        if (dummy && rth->fl.fl_net != net)
            continue
        ...
    ...

So when rt_run_flush is called synchronously, the netns is specified in dummy and only the routes belonging to netns are flushed. Otherwise when it is called by the timer, netns is not set so all routes are flushed.

Quoted 51 lines · Daniel Lezcano

Quoted 24 lines · Denis V. Lunev

Quoted 12 lines · Daniel Lezcano

Quoted 11 lines · Denis V. Lunev

/proc/sys/net/route/flush should be accessible inside the net namespace. Though, the complete opening of this file will result in a DoS or significant entire host slowdown if a namespace process will continually flush routes.

This patch introduces per/namespace route flush facility.

Each namespace wanted to flush a cache copies global generation count to itself and starts the timer. The cache is dropped for a specific namespace iff the namespace counter is greater or equal global ones.

Quoted 8 lines

Quoted 4 lines

So, in general, unwanted namespaces do nothing. They hold very old low counter and they are unaffected by the requested cleanup.

Signed-of-by Denis V. Lunev <[email protected]>

That's right and that will happen when manipulating ip addresses of the network devices too. But I am not confortable with your patchset. It touches the routing flush function too hardly and it uses current->nsproxy->net_ns.

IMHO we should have two flush functions. One taking a network namespace parameter and one without the network namespace parameter. The first one is called when a write to /proc/sys/net/ipv4/route/flush is done (we must use the network namespace of the writer) or when a interface address is changed or shutdown|up. The last one is called by the timer, so we have a global timer flushing the routing cache for all the namespaces.

we can't :( The unfortunate thing is that the actual cleanup is called indirectly and asynchronously. The user _schedule_ the garbage collector to run NOW and we are moving over a large routing cache. Really large.

Quoted 7 lines

The idea to iterate over the list of each namespace to flush is bad. We are in atomic context. The list is protected by the mutex.

The idea of several timers (per namespace) is also bad. You will iterate over large cache several times.

No other acceptable way here for me :(.

As for "the trigger" - rt_cache_flush, looks like you are right. We should pass namespace as a parameter. This should be done as a separate patch.

If we change:

    rt_cache_flush(struct net *net, int delay);

and inside we call rt_cache_flush((unsigned long)net);

And then we check in the rt_run_flush function,

    struct net *net = (struct net *)dummy;

    ...
    for (i = rt_hash_mask; i >= 0; i--) {
        ...
        if (dummy && rth->fl.fl_net != net)
            continue
        ...
    ...

Quoted 2 lines

So when rt_run_flush is called synchronously, the netns is specified in dummy and only the routes belonging to netns are flushed. Otherwise when it is called by the timer, netns is not set so all routes are flushed.

this does not look good for me. The size of this cache for 4GB host is 2*10^6 entries for IPv4 with a 131072 chains. The conventional mainstream kernel wants to merge the routing cache cleanup requests from the different sources if they are delayed (default).

The main idea for this patch is to protect all other namespaces from the current one. This cache is an important resource. You proposal will work for the forced synchronous cleanups. Though, there are some requests with results in the delayed rt_run_flush via [mod/add]_timer

How should we handle them?

Quoted 19 lines · Daniel Lezcano

Quoted 18 lines · Denis V. Lunev

Quoted 17 lines · Daniel Lezcano

Quoted 15 lines · Denis V. Lunev

/proc/sys/net/route/flush should be accessible inside the net namespace. Though, the complete opening of this file will result in a DoS or significant entire host slowdown if a namespace process will continually flush routes.

This patch introduces per/namespace route flush facility.

Each namespace wanted to flush a cache copies global generation count to itself and starts the timer. The cache is dropped for a specific namespace iff the namespace counter is greater or equal global ones.

So, in general, unwanted namespaces do nothing. They hold very old low counter and they are unaffected by the requested cleanup.

Signed-of-by Denis V. Lunev <[email protected]>

Quoted 7 lines

Quoted 7 lines

Quoted 3 lines

That's right and that will happen when manipulating ip addresses of the network devices too. But I am not confortable with your patchset. It touches the routing flush function too hardly and it uses current->nsproxy->net_ns.

IMHO we should have two flush functions. One taking a network namespace parameter and one without the network namespace parameter. The first one is called when a write to /proc/sys/net/ipv4/route/flush is done (we must use the network namespace of the writer) or when a interface address is changed or shutdown|up. The last one is called by the timer, so we have a global timer flushing the routing cache for all the namespaces.

we can't :( The unfortunate thing is that the actual cleanup is called indirectly and asynchronously. The user _schedule_ the garbage collector to run NOW and we are moving over a large routing cache. Really large.

The idea to iterate over the list of each namespace to flush is bad. We are in atomic context. The list is protected by the mutex.

Oh, by the way, I forgot something important you spotted with the list protected by the mutex.

When looking at ipv6/fib_hash.c with Benjamin, we need to browse the network namespaces list for the garbage collecting, but we are in an interrupt handler, so I can not use rtnl_lock.

Why is not possible to protect the list with a simple spinlock ? so we can call spin_lock_bh when we are in interrupt handler.

Quoted 6 lines · Daniel Lezcano

Oh, by the way, I forgot something important you spotted with the list protected by the mutex.

When looking at ipv6/fib_hash.c with Benjamin, we need to browse the network namespaces list for the garbage collecting, but we are in an interrupt handler, so I can not use rtnl_lock.

where exactly....

all interesting places are called under rtnl already...

Quoted 2 lines

Why is not possible to protect the list with a simple spinlock ? so we can call spin_lock_bh when we are in interrupt handler.

see my old patch with a locking rules :)

By the way, I have forgotten to mention, that the original patch works from 2004 for OpenVz :)

Quoted 6 lines · Daniel Lezcano

Oh, by the way, I forgot something important you spotted with the list protected by the mutex.

When looking at ipv6/fib_hash.c with Benjamin, we need to browse the network namespaces list for the garbage collecting, but we are in an interrupt handler, so I can not use rtnl_lock.

where exactly....

all interesting places are called under rtnl already...

Quoted 2 lines

Why is not possible to protect the list with a simple spinlock ? so we can call spin_lock_bh when we are in interrupt handler.

see my old patch with a locking rules :)

By the way, I have forgotten to mention, that the original patch works from 2004 for OpenVz :)

Regards,
Den

Quoted 7 lines · Denis V. Lunev

Quoted 6 lines · Daniel Lezcano

Oh, by the way, I forgot something important you spotted with the list protected by the mutex.

When looking at ipv6/fib_hash.c with Benjamin, we need to browse the network namespaces list for the garbage collecting, but we are in an interrupt handler, so I can not use rtnl_lock.

Quoted 1 line

where exactly....

Actually, it is in net/ipv6/ip6_fib.c, in fib6_clean_all().

fib6_clean_all() is called by fib6_run_gc() handler of the ip6_fib_timer. If we don't want to have one such timer per net namespace, in fib6_clean_all() we have to go through all net to clean their own fib_table_hash (using for_each_net() protected by rtnl_lock).

Quoted 7 lines · Denis V. Lunev

Quoted 6 lines · Daniel Lezcano

Oh, by the way, I forgot something important you spotted with the list protected by the mutex.

When looking at ipv6/fib_hash.c with Benjamin, we need to browse the network namespaces list for the garbage collecting, but we are in an interrupt handler, so I can not use rtnl_lock.

Quoted 1 line

where exactly....

Actually, it is in net/ipv6/ip6_fib.c, in fib6_clean_all().

fib6_clean_all() is called by fib6_run_gc() handler of the ip6_fib_timer. If we don't want to have one such timer per net namespace, in fib6_clean_all() we have to go through all net to clean their own fib_table_hash (using for_each_net() protected by rtnl_lock).

Quoted 12 lines · Benjamin Thery

Quoted 8 lines · Denis V. Lunev

Quoted 6 lines · Daniel Lezcano

Oh, by the way, I forgot something important you spotted with the list protected by the mutex.

When looking at ipv6/fib_hash.c with Benjamin, we need to browse the network namespaces list for the garbage collecting, but we are in an interrupt handler, so I can not use rtnl_lock.

where exactly....

Actually, it is in net/ipv6/ip6_fib.c, in fib6_clean_all().

Quoted 5 lines

fib6_clean_all() is called by fib6_run_gc() handler of the ip6_fib_timer. If we don't want to have one such timer per net namespace, in fib6_clean_all() we have to go through all net to clean their own
fib_table_hash (using for_each_net() protected by rtnl_lock).

after careful thinking, one timer per/namespace looks better for me :)

Quoted 12 lines · Denis V. Lunev

Quoted 11 lines · Benjamin Thery

Quoted 8 lines · Denis V. Lunev

Quoted 6 lines · Daniel Lezcano

Oh, by the way, I forgot something important you spotted with the list protected by the mutex.

When looking at ipv6/fib_hash.c with Benjamin, we need to browse the network namespaces list for the garbage collecting, but we are in an interrupt handler, so I can not use rtnl_lock.

where exactly....

Actually, it is in net/ipv6/ip6_fib.c, in fib6_clean_all().

Quoted 6 lines

Quoted 5 lines

fib6_clean_all() is called by fib6_run_gc() handler of the ip6_fib_timer. If we don't want to have one such timer per net namespace, in fib6_clean_all() we have to go through all net to clean their own
fib_table_hash (using for_each_net() protected by rtnl_lock).

after careful thinking, one timer per/namespace looks better for me :)

Why? :)

Then you'll have to find a way pass the target net to fib6_run_gc() (the timer handler). current->nsproxy->net_ns won't work :) One timer for all looked simpler to me.

Can there be an impact on performance if we have several GC timers for the several netns running?

Quoted 17 lines · Denis V. Lunev

Quoted 16 lines · Benjamin Thery

Quoted 8 lines · Denis V. Lunev

Quoted 6 lines · Daniel Lezcano

Oh, by the way, I forgot something important you spotted with the list protected by the mutex.

When looking at ipv6/fib_hash.c with Benjamin, we need to browse the network namespaces list for the garbage collecting, but we are in an interrupt handler, so I can not use rtnl_lock.

where exactly....

Actually, it is in net/ipv6/ip6_fib.c, in fib6_clean_all().

fib6_clean_all() is called by fib6_run_gc() handler of the ip6_fib_timer. If we don't want to have one such timer per net namespace, in fib6_clean_all() we have to go through all net to clean their own
fib_table_hash (using for_each_net() protected by rtnl_lock).

Quoted 1 line

after careful thinking, one timer per/namespace looks better for me :)

Why? :)
Then you'll have to find a way pass the target net to fib6_run_gc() (the timer handler). current->nsproxy->net_ns won't work :)

Quoted 18 lines · Benjamin Thery

Quoted 17 lines · Denis V. Lunev

Quoted 16 lines · Benjamin Thery

Quoted 8 lines · Denis V. Lunev

Quoted 6 lines · Daniel Lezcano

Oh, by the way, I forgot something important you spotted with the list protected by the mutex.

When looking at ipv6/fib_hash.c with Benjamin, we need to browse the network namespaces list for the garbage collecting, but we are in an interrupt handler, so I can not use rtnl_lock.

where exactly....

Actually, it is in net/ipv6/ip6_fib.c, in fib6_clean_all().

fib6_clean_all() is called by fib6_run_gc() handler of the ip6_fib_timer. If we don't want to have one such timer per net namespace, in fib6_clean_all() we have to go through all net to clean their own
fib_table_hash (using for_each_net() protected by rtnl_lock).

Quoted 4 lines

Quoted 1 line

after careful thinking, one timer per/namespace looks better for me :)

Why? :)
Then you'll have to find a way pass the target net to fib6_run_gc() (the timer handler). current->nsproxy->net_ns won't work :)

sorry for spam :) forget to mention that the namespace can be passed via timer data :) yes - additional structure will be required

By the way, why do you, guys, so dislike our approach :) In that case there will be no problem to get context from currrent. Yes. In soft interrupt also :))

Quoted 8 lines · Benjamin Thery

Quoted 7 lines · Denis V. Lunev

Quoted 6 lines · Daniel Lezcano

Oh, by the way, I forgot something important you spotted with the list protected by the mutex.

When looking at ipv6/fib_hash.c with Benjamin, we need to browse the network namespaces list for the garbage collecting, but we are in an interrupt handler, so I can not use rtnl_lock.

Quoted 6 lines

Quoted 1 line

where exactly....

Actually, it is in net/ipv6/ip6_fib.c, in fib6_clean_all().

fib6_clean_all() is called by fib6_run_gc() handler of the ip6_fib_timer. If we don't want to have one such timer per net namespace, in fib6_clean_all() we have to go through all net to clean their own fib_table_hash (using for_each_net() protected by rtnl_lock).

after careful thinking, one timer per/namespace looks better for me :)

So you prefer to have multiple timers flushing a portion of a shared routing cache ?

It is not impossible I already tryed that for ipv6. it will just touch much more files.

If we want, at all costs, to have the routing cache flushed per namespace, I am much more favorable to have a timer per namespace than using the algorithm you sent previously.

Quoted 9 lines · Daniel Lezcano

So you prefer to have multiple timers flushing a portion of a shared routing cache ?

It is not impossible I already tryed that for ipv6. it will just touch much more files.

If we want, at all costs, to have the routing cache flushed per namespace, I am much more favorable to have a timer per namespace than using the algorithm you sent previously.

as do I. But the IPv4 and IPv6 situation is much different. IPv6 cache is separated while IPv4 one is common.

So, the situation is not uniform, at least :(

By the way, we can move routing cache into the FIB tree like one for IPv4 :)