This patch makes VLAN use RCU for the group lookup and removes
the group_lock. Since all operations that are done under group_lock
already have the rtnetlink semaphore (RTNL), the group_lock is unnecessary.
The result is the vlan code becomes basically lock free in the send/receive
path.
Signed-off-by: Stephen Hemminger <shemminger@xxxxxxxx>
diff -Nru a/include/linux/if_vlan.h b/include/linux/if_vlan.h
--- a/include/linux/if_vlan.h 2004-08-03 13:38:50 -07:00
+++ b/include/linux/if_vlan.h 2004-08-03 13:38:50 -07:00
@@ -22,6 +22,7 @@
struct packet_type;
struct vlan_collection;
struct vlan_dev_info;
+struct hlist_node;
#include <linux/proc_fs.h> /* for proc_dir_entry */
#include <linux/netdevice.h>
@@ -67,9 +68,9 @@
struct vlan_group {
int real_dev_ifindex; /* The ifindex of the ethernet(like) device the
vlan is attached to. */
+ struct hlist_node hlist; /* linked list */
struct net_device *vlan_devices[VLAN_GROUP_ARRAY_LEN];
-
- struct vlan_group *next; /* the next in the list */
+ struct rcu_head rcu;
};
struct vlan_priority_tci_mapping {
diff -Nru a/net/8021q/vlan.c b/net/8021q/vlan.c
--- a/net/8021q/vlan.c 2004-08-03 13:38:50 -07:00
+++ b/net/8021q/vlan.c 2004-08-03 13:38:50 -07:00
@@ -38,8 +38,7 @@
/* Global VLAN variables */
/* Our listing of VLAN group(s) */
-struct vlan_group *vlan_group_hash[VLAN_GRP_HASH_SIZE];
-spinlock_t vlan_group_lock = SPIN_LOCK_UNLOCKED;
+struct hlist_head vlan_group_hash[VLAN_GRP_HASH_SIZE];
#define vlan_grp_hashfn(IDX) ((((IDX) >> VLAN_GRP_HASH_SHIFT) ^ (IDX)) &
VLAN_GRP_HASH_MASK)
static char vlan_fullname[] = "802.1Q VLAN Support";
@@ -150,8 +149,7 @@
* references left.
*/
for (i = 0; i < VLAN_GRP_HASH_SIZE; i++) {
- if (vlan_group_hash[i] != NULL)
- BUG();
+ BUG_ON(!hlist_empty(&vlan_group_hash[i]));
}
vlan_proc_cleanup();
@@ -161,48 +159,24 @@
module_init(vlan_proto_init);
module_exit(vlan_cleanup_module);
-/* Must be invoked with vlan_group_lock held. */
+/* Must be invoked with RCU read lock (no preempt) */
static struct vlan_group *__vlan_find_group(int real_dev_ifindex)
{
struct vlan_group *grp;
+ struct hlist_node *n;
+ int hash = vlan_grp_hashfn(real_dev_ifindex);
- for (grp = vlan_group_hash[vlan_grp_hashfn(real_dev_ifindex)];
- grp != NULL;
- grp = grp->next) {
+ hlist_for_each_entry_rcu(grp, n, &vlan_group_hash[hash], hlist) {
if (grp->real_dev_ifindex == real_dev_ifindex)
- break;
+ return grp;
}
- return grp;
-}
-
-/* Must hold vlan_group_lock. */
-static void __grp_hash(struct vlan_group *grp)
-{
- struct vlan_group **head;
-
- head = &vlan_group_hash[vlan_grp_hashfn(grp->real_dev_ifindex)];
- grp->next = *head;
- *head = grp;
-}
-
-/* Must hold vlan_group_lock. */
-static void __grp_unhash(struct vlan_group *grp)
-{
- struct vlan_group *next, **pprev;
-
- pprev = &vlan_group_hash[vlan_grp_hashfn(grp->real_dev_ifindex)];
- next = *pprev;
- while (next != grp) {
- pprev = &next->next;
- next = *pprev;
- }
- *pprev = grp->next;
+ return NULL;
}
/* Find the protocol handler. Assumes VID < VLAN_VID_MASK.
*
- * Must be invoked with vlan_group_lock held.
+ * Must be invoked with RCU read lock (no preempt)
*/
struct net_device *__find_vlan_dev(struct net_device *real_dev,
unsigned short VID)
@@ -215,6 +189,12 @@
return NULL;
}
+static void vlan_rcu_free(struct rcu_head *rcu)
+{
+ kfree(container_of(rcu, struct vlan_group, rcu));
+}
+
+
/* This returns 0 if everything went fine.
* It will return 1 if the group was killed as a result.
* A negative return indicates failure.
@@ -237,9 +217,8 @@
if (vlan_id >= VLAN_VID_MASK)
return -EINVAL;
- spin_lock_bh(&vlan_group_lock);
+ ASSERT_RTNL();
grp = __vlan_find_group(real_dev_ifindex);
- spin_unlock_bh(&vlan_group_lock);
ret = 0;
@@ -279,16 +258,12 @@
if (real_dev->features & NETIF_F_HW_VLAN_RX)
real_dev->vlan_rx_register(real_dev,
NULL);
- spin_lock_bh(&vlan_group_lock);
- __grp_unhash(grp);
- spin_unlock_bh(&vlan_group_lock);
-
- /* Free the group, after we have removed it
- * from the hash.
- */
- kfree(grp);
- grp = NULL;
+ hlist_del_rcu(&grp->hlist);
+ /* Free the group, after all cpu's are done. */
+ call_rcu(&grp->rcu, vlan_rcu_free);
+
+ grp = NULL;
ret = 1;
}
}
@@ -375,7 +350,6 @@
struct vlan_group *grp;
struct net_device *new_dev;
struct net_device *real_dev; /* the ethernet device */
- int r;
char name[IFNAMSIZ];
#ifdef VLAN_DEBUG
@@ -424,11 +398,7 @@
if (!(real_dev->flags & IFF_UP))
goto out_unlock;
- spin_lock_bh(&vlan_group_lock);
- r = (__find_vlan_dev(real_dev, VLAN_ID) != NULL);
- spin_unlock_bh(&vlan_group_lock);
-
- if (r) {
+ if (__find_vlan_dev(real_dev, VLAN_ID) != NULL) {
/* was already registered. */
printk(VLAN_DBG "%s: ALREADY had VLAN registered\n",
__FUNCTION__);
goto out_unlock;
@@ -527,9 +497,7 @@
/* So, got the sucker initialized, now lets place
* it into our local structure.
*/
- spin_lock_bh(&vlan_group_lock);
grp = __vlan_find_group(real_dev->ifindex);
- spin_unlock_bh(&vlan_group_lock);
/* Note, we are running under the RTNL semaphore
* so it cannot "appear" on us.
@@ -543,9 +511,8 @@
memset(grp, 0, sizeof(struct vlan_group));
grp->real_dev_ifindex = real_dev->ifindex;
- spin_lock_bh(&vlan_group_lock);
- __grp_hash(grp);
- spin_unlock_bh(&vlan_group_lock);
+ hlist_add_head_rcu(&grp->hlist,
+
&vlan_group_hash[vlan_grp_hashfn(real_dev->ifindex)]);
if (real_dev->features & NETIF_F_HW_VLAN_RX)
real_dev->vlan_rx_register(real_dev, grp);
@@ -587,14 +554,10 @@
static int vlan_device_event(struct notifier_block *unused, unsigned long
event, void *ptr)
{
- struct net_device *dev = (struct net_device *)(ptr);
- struct vlan_group *grp = NULL;
+ struct net_device *dev = ptr;
+ struct vlan_group *grp = __vlan_find_group(dev->ifindex);
int i, flgs;
- struct net_device *vlandev = NULL;
-
- spin_lock_bh(&vlan_group_lock);
- grp = __vlan_find_group(dev->ifindex);
- spin_unlock_bh(&vlan_group_lock);
+ struct net_device *vlandev;
if (!grp)
goto out;
diff -Nru a/net/8021q/vlan.h b/net/8021q/vlan.h
--- a/net/8021q/vlan.h 2004-08-03 13:38:50 -07:00
+++ b/net/8021q/vlan.h 2004-08-03 13:38:50 -07:00
@@ -33,8 +33,7 @@
#define VLAN_GRP_HASH_SHIFT 5
#define VLAN_GRP_HASH_SIZE (1 << VLAN_GRP_HASH_SHIFT)
#define VLAN_GRP_HASH_MASK (VLAN_GRP_HASH_SIZE - 1)
-extern struct vlan_group *vlan_group_hash[VLAN_GRP_HASH_SIZE];
-extern spinlock_t vlan_group_lock;
+extern struct hlist_head vlan_group_hash[VLAN_GRP_HASH_SIZE];
/* Find a VLAN device by the MAC address of its Ethernet device, and
* it's VLAN ID. The default configuration is to have VLAN's scope
@@ -44,10 +43,8 @@
* NOT follow the spec for VLANs, but may be useful for doing very
* large quantities of VLAN MUX/DEMUX onto FrameRelay or ATM PVCs.
*
- * Must be invoked with vlan_group_lock held and that lock MUST NOT
- * be dropped until a reference is obtained on the returned device.
- * You may drop the lock earlier if you are running under the RTNL
- * semaphore, however.
+ * Must be invoked with rcu_read_lock (ie preempt disabled)
+ * or with RTNL.
*/
struct net_device *__find_vlan_dev(struct net_device* real_dev,
unsigned short VID); /* vlan.c */
diff -Nru a/net/8021q/vlan_dev.c b/net/8021q/vlan_dev.c
--- a/net/8021q/vlan_dev.c 2004-08-03 13:38:50 -07:00
+++ b/net/8021q/vlan_dev.c 2004-08-03 13:38:50 -07:00
@@ -138,15 +138,15 @@
/* We have 12 bits of vlan ID.
*
- * We must not drop the vlan_group_lock until we hold a
+ * We must not drop allow preempt until we hold a
* reference to the device (netif_rx does that) or we
* fail.
*/
- spin_lock_bh(&vlan_group_lock);
+ rcu_read_lock();
skb->dev = __find_vlan_dev(dev, vid);
if (!skb->dev) {
- spin_unlock_bh(&vlan_group_lock);
+ rcu_read_unlock();
#ifdef VLAN_DEBUG
printk(VLAN_DBG "%s: ERROR: No net_device for VID: %i on dev:
%s [%i]\n",
@@ -170,7 +170,7 @@
*/
if (dev != VLAN_DEV_INFO(skb->dev)->real_dev) {
- spin_unlock_bh(&vlan_group_lock);
+ rcu_read_unlock();
#ifdef VLAN_DEBUG
printk(VLAN_DBG "%s: dropping skb: %p because came in on wrong
device, dev: %s real_dev: %s, skb_dev: %s\n",
@@ -244,7 +244,7 @@
/* TODO: Add a more specific counter here. */
stats->rx_errors++;
}
- spin_unlock_bh(&vlan_group_lock);
+ rcu_read_lock();
return 0;
}
@@ -273,7 +273,7 @@
/* TODO: Add a more specific counter here. */
stats->rx_errors++;
}
- spin_unlock_bh(&vlan_group_lock);
+ rcu_read_unlock();
return 0;
}
@@ -296,7 +296,7 @@
/* TODO: Add a more specific counter here. */
stats->rx_errors++;
}
- spin_unlock_bh(&vlan_group_lock);
+ rcu_read_unlock();
return 0;
}
|