[Devel] [PATCH RH9 02/28] ve/net/neighbour: per-ct limit for neighbour entries

[Andrey Zhadchenko]

From: Vasily Averin <vvs at virtuozzo.com>

Based on patch from Konstantin Khoerenko <khorenko at virtuozzo.com>

Currently neighours do not have per-namespace structures and use
common hash, locks, counters and global limits.

Total number of neigh entries on a Node can easily exceed the
global gc_thresh3 on the Node. (We've got an Node with 67 active
Containers which required ~6000 active neighbour entries).

Let's introduce additional counters for neigh entries per Container and
check max limit against them. Global counters for neigh entries are left
as is, they will be used for periodical and forced cleanups as it works
now.

https://jira.sw.ru/browse/PSBM-87155

Signed-off-by: Vasily Averin <vvs at virtuozzo.com>
Acked-by: Konstantin Khorenko <khorenko at virtuozzo.com>

(cherry picked from vz7 commit c75e42c15e7d ("ve/net/neighbour: per-ct limit for
neighbour entries"))

VZ 8 rebase part https://jira.sw.ru/browse/PSBM-127837

Signed-off-by: Alexander Mikhalitsyn <alexander.mikhalitsyn at virtuozzo.com>

(cherry picked from vz8 commit 690ab3c461b97672274252713a7f5684f3f86bfd)
Signed-off-by: Andrey Zhadchenko <andrey.zhadchenko at virtuozzo.com>
---
 include/linux/ve.h   |  2 ++
 kernel/ve/ve.c       |  4 ++++
 net/core/neighbour.c | 46 +++++++++++++++++++++++++++++++++++++++-------
 3 files changed, 45 insertions(+), 7 deletions(-)

diff --git a/include/linux/ve.h b/include/linux/ve.h
index d349985..60947c2 100644
--- a/include/linux/ve.h
+++ b/include/linux/ve.h
@@ -76,6 +76,8 @@ struct ve_struct {
 
 	int			_randomize_va_space;
 
+	atomic_t		arp_neigh_nr;
+	atomic_t		nd_neigh_nr;
 	unsigned long		meminfo_val;
 
 	atomic_t		mnt_nr; /* number of present VE mounts */
diff --git a/kernel/ve/ve.c b/kernel/ve/ve.c
index 8471790..2f4b746 100644
--- a/kernel/ve/ve.c
+++ b/kernel/ve/ve.c
@@ -68,6 +68,8 @@ struct ve_struct ve0 = {
 					2,
 #endif
 
+	.arp_neigh_nr		= ATOMIC_INIT(0),
+	.nd_neigh_nr		= ATOMIC_INIT(0),
 	.mnt_nr			= ATOMIC_INIT(0),
 	.meminfo_val		= VE_MEMINFO_SYSTEM,
 	.vdso_64		= (struct vdso_image*)&vdso_image_64,
@@ -720,6 +722,8 @@ static struct cgroup_subsys_state *ve_create(struct cgroup_subsys_state *parent_
 	INIT_LIST_HEAD(&ve->ve_list);
 	kmapset_init_key(&ve->sysfs_perms_key);
 
+	atomic_set(&ve->arp_neigh_nr, 0);
+	atomic_set(&ve->nd_neigh_nr, 0);
 	atomic_set(&ve->mnt_nr, 0);
 
 #ifdef CONFIG_COREDUMP
diff --git a/net/core/neighbour.c b/net/core/neighbour.c
index 53e85c7..76c749f 100644
--- a/net/core/neighbour.c
+++ b/net/core/neighbour.c
@@ -38,6 +38,7 @@
 #include <linux/log2.h>
 #include <linux/inetdevice.h>
 #include <net/addrconf.h>
+#include <linux/ve.h>
 
 #include <trace/events/neigh.h>
 
@@ -221,7 +222,7 @@ bool neigh_remove_one(struct neighbour *ndel, struct neigh_table *tbl)
 	return false;
 }
 
-static int neigh_forced_gc(struct neigh_table *tbl)
+static int neigh_forced_gc(struct neigh_table *tbl, struct ve_struct *ve)
 {
 	int max_clean = atomic_read(&tbl->gc_entries) - tbl->gc_thresh2;
 	unsigned long tref = jiffies - 5 * HZ;
@@ -235,6 +236,7 @@ static int neigh_forced_gc(struct neigh_table *tbl)
 	list_for_each_entry_safe(n, tmp, &tbl->gc_list, gc_list) {
 		if (refcount_read(&n->refcnt) == 1) {
 			bool remove = false;
+			bool same_ve = (dev_net(n->dev)->owner_ve == ve);
 
 			write_lock(&n->lock);
 			if ((n->nud_state == NUD_FAILED) ||
@@ -246,7 +248,8 @@ static int neigh_forced_gc(struct neigh_table *tbl)
 			write_unlock(&n->lock);
 
 			if (remove && neigh_remove_one(n, tbl))
-				shrunk++;
+				if (same_ve)
+					shrunk++;
 			if (shrunk >= max_clean)
 				break;
 		}
@@ -377,22 +380,44 @@ int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
 }
 EXPORT_SYMBOL(neigh_ifdown);
 
+static inline atomic_t *get_perve_tbl_entries_counter(struct neigh_table *tbl,
+							struct ve_struct *ve)
+{
+	switch (tbl->family) {
+	case AF_INET:
+		return &ve->arp_neigh_nr;
+	case AF_INET6:
+		return &ve->nd_neigh_nr;
+	}
+	return NULL;
+}
+
 static struct neighbour *neigh_alloc(struct neigh_table *tbl,
 				     struct net_device *dev,
 				     bool exempt_from_gc)
 {
 	struct neighbour *n = NULL;
 	unsigned long now = jiffies;
-	int entries;
+	int entries, glob_entries;
+	atomic_t *cnt;
+	struct ve_struct *ve = dev_net(dev)->owner_ve;
 
 	if (exempt_from_gc)
 		goto do_alloc;
 
-	entries = atomic_inc_return(&tbl->gc_entries) - 1;
+	/* If per-VE counter of neighbour entries exist
+	 * it will be limited by tbl->gc_thresh3
+	 * and according global counter (tbl->entries) become unlimited.
+	 */
+
+	glob_entries = atomic_inc_return(&tbl->entries) - 1;
+	cnt = get_perve_tbl_entries_counter(tbl, ve);
+	entries = cnt ? atomic_inc_return(cnt) - 1 : glob_entries;
+
 	if (entries >= tbl->gc_thresh3 ||
-	    (entries >= tbl->gc_thresh2 &&
+	    (glob_entries >= tbl->gc_thresh2 &&
 	     time_after(now, tbl->last_flush + 5 * HZ))) {
-		if (!neigh_forced_gc(tbl) &&
+		if (!neigh_forced_gc(tbl, ve) &&
 		    entries >= tbl->gc_thresh3) {
 			net_info_ratelimited("%s: neighbor table overflow!\n",
 					     tbl->id);
@@ -427,8 +452,11 @@ static struct neighbour *neigh_alloc(struct neigh_table *tbl,
 	return n;
 
 out_entries:
-	if (!exempt_from_gc)
+	if (!exempt_from_gc) {
 		atomic_dec(&tbl->gc_entries);
+		if (cnt)
+			atomic_dec(cnt);
+	}
 	goto out;
 }
 
@@ -834,6 +862,8 @@ static inline void neigh_parms_put(struct neigh_parms *parms)
 void neigh_destroy(struct neighbour *neigh)
 {
 	struct net_device *dev = neigh->dev;
+	struct ve_struct *ve = dev_net(dev)->owner_ve;
+	atomic_t *cnt = get_perve_tbl_entries_counter(neigh->tbl, ve);
 
 	NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
 
@@ -859,6 +889,8 @@ void neigh_destroy(struct neighbour *neigh)
 
 	neigh_dbg(2, "neigh %p is destroyed\n", neigh);
 
+	if (cnt)
+		 atomic_dec(cnt);
 	atomic_dec(&neigh->tbl->entries);
 	kfree_rcu(neigh, rcu);
 }
-- 
1.8.3.1