* [PATCH net-next 01/19] ipvs: change ip_vs_sched_lock to mutex
2013-03-22 9:46 [PATCH net-next 00/19] IPVS optimizations, part 2 Julian Anastasov
@ 2013-03-22 9:46 ` Julian Anastasov
2013-03-22 9:46 ` [PATCH net-next 02/19] ipvs: preparations for using rcu in schedulers Julian Anastasov
` (18 subsequent siblings)
19 siblings, 0 replies; 23+ messages in thread
From: Julian Anastasov @ 2013-03-22 9:46 UTC (permalink / raw)
To: Simon Horman; +Cc: lvs-devel, netdev
The global list with schedulers ip_vs_schedulers
is accessed only from user context - configuration and
scheduler module [un]registration. Use ip_vs_sched_mutex
instead.
Signed-off-by: Julian Anastasov <ja@ssi.bg>
---
net/netfilter/ipvs/ip_vs_sched.c | 24 ++++++++++++------------
1 files changed, 12 insertions(+), 12 deletions(-)
diff --git a/net/netfilter/ipvs/ip_vs_sched.c b/net/netfilter/ipvs/ip_vs_sched.c
index d6bf20d..7f11d3d 100644
--- a/net/netfilter/ipvs/ip_vs_sched.c
+++ b/net/netfilter/ipvs/ip_vs_sched.c
@@ -35,8 +35,8 @@ EXPORT_SYMBOL(ip_vs_scheduler_err);
*/
static LIST_HEAD(ip_vs_schedulers);
-/* lock for service table */
-static DEFINE_SPINLOCK(ip_vs_sched_lock);
+/* semaphore for schedulers */
+static DEFINE_MUTEX(ip_vs_sched_mutex);
/*
@@ -92,7 +92,7 @@ static struct ip_vs_scheduler *ip_vs_sched_getbyname(const char *sched_name)
IP_VS_DBG(2, "%s(): sched_name \"%s\"\n", __func__, sched_name);
- spin_lock_bh(&ip_vs_sched_lock);
+ mutex_lock(&ip_vs_sched_mutex);
list_for_each_entry(sched, &ip_vs_schedulers, n_list) {
/*
@@ -106,14 +106,14 @@ static struct ip_vs_scheduler *ip_vs_sched_getbyname(const char *sched_name)
}
if (strcmp(sched_name, sched->name)==0) {
/* HIT */
- spin_unlock_bh(&ip_vs_sched_lock);
+ mutex_unlock(&ip_vs_sched_mutex);
return sched;
}
if (sched->module)
module_put(sched->module);
}
- spin_unlock_bh(&ip_vs_sched_lock);
+ mutex_unlock(&ip_vs_sched_mutex);
return NULL;
}
@@ -192,10 +192,10 @@ int register_ip_vs_scheduler(struct ip_vs_scheduler *scheduler)
/* increase the module use count */
ip_vs_use_count_inc();
- spin_lock_bh(&ip_vs_sched_lock);
+ mutex_lock(&ip_vs_sched_mutex);
if (!list_empty(&scheduler->n_list)) {
- spin_unlock_bh(&ip_vs_sched_lock);
+ mutex_unlock(&ip_vs_sched_mutex);
ip_vs_use_count_dec();
pr_err("%s(): [%s] scheduler already linked\n",
__func__, scheduler->name);
@@ -208,7 +208,7 @@ int register_ip_vs_scheduler(struct ip_vs_scheduler *scheduler)
*/
list_for_each_entry(sched, &ip_vs_schedulers, n_list) {
if (strcmp(scheduler->name, sched->name) == 0) {
- spin_unlock_bh(&ip_vs_sched_lock);
+ mutex_unlock(&ip_vs_sched_mutex);
ip_vs_use_count_dec();
pr_err("%s(): [%s] scheduler already existed "
"in the system\n", __func__, scheduler->name);
@@ -219,7 +219,7 @@ int register_ip_vs_scheduler(struct ip_vs_scheduler *scheduler)
* Add it into the d-linked scheduler list
*/
list_add(&scheduler->n_list, &ip_vs_schedulers);
- spin_unlock_bh(&ip_vs_sched_lock);
+ mutex_unlock(&ip_vs_sched_mutex);
pr_info("[%s] scheduler registered.\n", scheduler->name);
@@ -237,9 +237,9 @@ int unregister_ip_vs_scheduler(struct ip_vs_scheduler *scheduler)
return -EINVAL;
}
- spin_lock_bh(&ip_vs_sched_lock);
+ mutex_lock(&ip_vs_sched_mutex);
if (list_empty(&scheduler->n_list)) {
- spin_unlock_bh(&ip_vs_sched_lock);
+ mutex_unlock(&ip_vs_sched_mutex);
pr_err("%s(): [%s] scheduler is not in the list. failed\n",
__func__, scheduler->name);
return -EINVAL;
@@ -249,7 +249,7 @@ int unregister_ip_vs_scheduler(struct ip_vs_scheduler *scheduler)
* Remove it from the d-linked scheduler list
*/
list_del(&scheduler->n_list);
- spin_unlock_bh(&ip_vs_sched_lock);
+ mutex_unlock(&ip_vs_sched_mutex);
/* decrease the module use count */
ip_vs_use_count_dec();
--
1.7.3.4
^ permalink raw reply related [flat|nested] 23+ messages in thread* [PATCH net-next 02/19] ipvs: preparations for using rcu in schedulers
2013-03-22 9:46 [PATCH net-next 00/19] IPVS optimizations, part 2 Julian Anastasov
2013-03-22 9:46 ` [PATCH net-next 01/19] ipvs: change ip_vs_sched_lock to mutex Julian Anastasov
@ 2013-03-22 9:46 ` Julian Anastasov
2013-03-22 9:46 ` [PATCH net-next 03/19] ipvs: add ip_vs_dest_hold and ip_vs_dest_put Julian Anastasov
` (17 subsequent siblings)
19 siblings, 0 replies; 23+ messages in thread
From: Julian Anastasov @ 2013-03-22 9:46 UTC (permalink / raw)
To: Simon Horman; +Cc: lvs-devel, netdev
Allow schedulers to use rcu_dereference when
returning destination on lookup. The RCU read-side critical
section will allow ip_vs_bind_dest to get dest refcnt as
preparation for the step where destinations will be
deleted without an IP_VS_WAIT_WHILE guard that holds the
packet processing during update.
Add new optional scheduler methods add_dest,
del_dest and upd_dest. For now the methods are called
together with update_service but update_service will be
removed in a following change.
Signed-off-by: Julian Anastasov <ja@ssi.bg>
---
include/net/ip_vs.h | 6 ++++++
net/netfilter/ipvs/ip_vs_core.c | 6 ++++++
net/netfilter/ipvs/ip_vs_ctl.c | 8 ++++++++
3 files changed, 20 insertions(+), 0 deletions(-)
diff --git a/include/net/ip_vs.h b/include/net/ip_vs.h
index 5ed8939..5af338b 100644
--- a/include/net/ip_vs.h
+++ b/include/net/ip_vs.h
@@ -805,6 +805,12 @@ struct ip_vs_scheduler {
int (*done_service)(struct ip_vs_service *svc);
/* scheduler updating service */
int (*update_service)(struct ip_vs_service *svc);
+ /* dest is linked */
+ int (*add_dest)(struct ip_vs_service *svc, struct ip_vs_dest *dest);
+ /* dest is unlinked */
+ int (*del_dest)(struct ip_vs_service *svc, struct ip_vs_dest *dest);
+ /* dest is updated */
+ int (*upd_dest)(struct ip_vs_service *svc, struct ip_vs_dest *dest);
/* selecting a server from the given service */
struct ip_vs_dest* (*schedule)(struct ip_vs_service *svc,
diff --git a/net/netfilter/ipvs/ip_vs_core.c b/net/netfilter/ipvs/ip_vs_core.c
index 207c3b9..7bf62d0 100644
--- a/net/netfilter/ipvs/ip_vs_core.c
+++ b/net/netfilter/ipvs/ip_vs_core.c
@@ -304,8 +304,10 @@ ip_vs_sched_persist(struct ip_vs_service *svc,
* template is not available.
* return *ignored=0 i.e. ICMP and NF_DROP
*/
+ rcu_read_lock();
dest = svc->scheduler->schedule(svc, skb);
if (!dest) {
+ rcu_read_unlock();
IP_VS_DBG(1, "p-schedule: no dest found.\n");
kfree(param.pe_data);
*ignored = 0;
@@ -321,6 +323,7 @@ ip_vs_sched_persist(struct ip_vs_service *svc,
* when the template expires */
ct = ip_vs_conn_new(¶m, &dest->addr, dport,
IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
+ rcu_read_unlock();
if (ct == NULL) {
kfree(param.pe_data);
*ignored = -1;
@@ -449,8 +452,10 @@ ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
return NULL;
}
+ rcu_read_lock();
dest = svc->scheduler->schedule(svc, skb);
if (dest == NULL) {
+ rcu_read_unlock();
IP_VS_DBG(1, "Schedule: no dest found.\n");
return NULL;
}
@@ -471,6 +476,7 @@ ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
cp = ip_vs_conn_new(&p, &dest->addr,
dest->port ? dest->port : pptr[1],
flags, dest, skb->mark);
+ rcu_read_unlock();
if (!cp) {
*ignored = -1;
return NULL;
diff --git a/net/netfilter/ipvs/ip_vs_ctl.c b/net/netfilter/ipvs/ip_vs_ctl.c
index a66b1c7..ae765b8 100644
--- a/net/netfilter/ipvs/ip_vs_ctl.c
+++ b/net/netfilter/ipvs/ip_vs_ctl.c
@@ -823,6 +823,11 @@ __ip_vs_update_dest(struct ip_vs_service *svc, struct ip_vs_dest *dest,
if (add) {
list_add(&dest->n_list, &svc->destinations);
svc->num_dests++;
+ if (svc->scheduler->add_dest)
+ svc->scheduler->add_dest(svc, dest);
+ } else {
+ if (svc->scheduler->upd_dest)
+ svc->scheduler->upd_dest(svc, dest);
}
/* call the update_service, because server weight may be changed */
@@ -1069,6 +1074,9 @@ static void __ip_vs_unlink_dest(struct ip_vs_service *svc,
list_del(&dest->n_list);
svc->num_dests--;
+ if (svcupd && svc->scheduler->del_dest)
+ svc->scheduler->del_dest(svc, dest);
+
/*
* Call the update_service function of its scheduler
*/
--
1.7.3.4
^ permalink raw reply related [flat|nested] 23+ messages in thread* [PATCH net-next 03/19] ipvs: add ip_vs_dest_hold and ip_vs_dest_put
2013-03-22 9:46 [PATCH net-next 00/19] IPVS optimizations, part 2 Julian Anastasov
2013-03-22 9:46 ` [PATCH net-next 01/19] ipvs: change ip_vs_sched_lock to mutex Julian Anastasov
2013-03-22 9:46 ` [PATCH net-next 02/19] ipvs: preparations for using rcu in schedulers Julian Anastasov
@ 2013-03-22 9:46 ` Julian Anastasov
2013-03-22 9:46 ` [PATCH net-next 04/19] ipvs: convert dh scheduler to rcu Julian Anastasov
` (16 subsequent siblings)
19 siblings, 0 replies; 23+ messages in thread
From: Julian Anastasov @ 2013-03-22 9:46 UTC (permalink / raw)
To: Simon Horman; +Cc: lvs-devel, netdev
ip_vs_dest_hold will be used under RCU lock
while ip_vs_dest_put can be called even after dest
is removed from service, as it happens for conns and
some schedulers.
Signed-off-by: Julian Anastasov <ja@ssi.bg>
---
include/net/ip_vs.h | 10 ++++++++++
net/netfilter/ipvs/ip_vs_conn.c | 9 ++-------
net/netfilter/ipvs/ip_vs_ctl.c | 4 ++--
net/netfilter/ipvs/ip_vs_sync.c | 4 ++--
4 files changed, 16 insertions(+), 11 deletions(-)
diff --git a/include/net/ip_vs.h b/include/net/ip_vs.h
index 5af338b..2752c86 100644
--- a/include/net/ip_vs.h
+++ b/include/net/ip_vs.h
@@ -1427,6 +1427,16 @@ ip_vs_find_dest(struct net *net, int af, const union nf_inet_addr *daddr,
__u16 protocol, __u32 fwmark, __u32 flags);
extern struct ip_vs_dest *ip_vs_try_bind_dest(struct ip_vs_conn *cp);
+static inline void ip_vs_dest_hold(struct ip_vs_dest *dest)
+{
+ atomic_inc(&dest->refcnt);
+}
+
+static inline void ip_vs_dest_put(struct ip_vs_dest *dest)
+{
+ smp_mb__before_atomic_dec();
+ atomic_dec(&dest->refcnt);
+}
/*
* IPVS sync daemon data and function prototypes
diff --git a/net/netfilter/ipvs/ip_vs_conn.c b/net/netfilter/ipvs/ip_vs_conn.c
index e3e2b4d..1b29e4a 100644
--- a/net/netfilter/ipvs/ip_vs_conn.c
+++ b/net/netfilter/ipvs/ip_vs_conn.c
@@ -554,7 +554,7 @@ ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest)
return;
/* Increase the refcnt counter of the dest */
- atomic_inc(&dest->refcnt);
+ ip_vs_dest_hold(dest);
conn_flags = atomic_read(&dest->conn_flags);
if (cp->protocol != IPPROTO_UDP)
@@ -700,12 +700,7 @@ static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp)
dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
}
- /*
- * Simply decrease the refcnt of the dest, because the
- * dest will be either in service's destination list
- * or in the trash.
- */
- atomic_dec(&dest->refcnt);
+ ip_vs_dest_put(dest);
}
static int expire_quiescent_template(struct netns_ipvs *ipvs,
diff --git a/net/netfilter/ipvs/ip_vs_ctl.c b/net/netfilter/ipvs/ip_vs_ctl.c
index ae765b8..36d9cd8 100644
--- a/net/netfilter/ipvs/ip_vs_ctl.c
+++ b/net/netfilter/ipvs/ip_vs_ctl.c
@@ -614,7 +614,7 @@ struct ip_vs_dest *ip_vs_find_dest(struct net *net, int af,
if (!dest)
dest = ip_vs_lookup_dest(svc, daddr, port ^ dport);
if (dest)
- atomic_inc(&dest->refcnt);
+ ip_vs_dest_hold(dest);
ip_vs_service_put(svc);
return dest;
}
@@ -1054,7 +1054,7 @@ static void __ip_vs_del_dest(struct net *net, struct ip_vs_dest *dest)
ntohs(dest->port),
atomic_read(&dest->refcnt));
list_add(&dest->n_list, &ipvs->dest_trash);
- atomic_inc(&dest->refcnt);
+ ip_vs_dest_hold(dest);
}
}
diff --git a/net/netfilter/ipvs/ip_vs_sync.c b/net/netfilter/ipvs/ip_vs_sync.c
index 44fd10c..6cc3e52 100644
--- a/net/netfilter/ipvs/ip_vs_sync.c
+++ b/net/netfilter/ipvs/ip_vs_sync.c
@@ -861,7 +861,7 @@ static void ip_vs_proc_conn(struct net *net, struct ip_vs_conn_param *param,
if (!dest) {
dest = ip_vs_try_bind_dest(cp);
if (dest)
- atomic_dec(&dest->refcnt);
+ ip_vs_dest_put(dest);
}
} else {
/*
@@ -874,7 +874,7 @@ static void ip_vs_proc_conn(struct net *net, struct ip_vs_conn_param *param,
cp = ip_vs_conn_new(param, daddr, dport, flags, dest, fwmark);
if (dest)
- atomic_dec(&dest->refcnt);
+ ip_vs_dest_put(dest);
if (!cp) {
if (param->pe_data)
kfree(param->pe_data);
--
1.7.3.4
^ permalink raw reply related [flat|nested] 23+ messages in thread* [PATCH net-next 04/19] ipvs: convert dh scheduler to rcu
2013-03-22 9:46 [PATCH net-next 00/19] IPVS optimizations, part 2 Julian Anastasov
` (2 preceding siblings ...)
2013-03-22 9:46 ` [PATCH net-next 03/19] ipvs: add ip_vs_dest_hold and ip_vs_dest_put Julian Anastasov
@ 2013-03-22 9:46 ` Julian Anastasov
2013-03-22 9:46 ` [PATCH net-next 05/19] ipvs: convert lblc " Julian Anastasov
` (15 subsequent siblings)
19 siblings, 0 replies; 23+ messages in thread
From: Julian Anastasov @ 2013-03-22 9:46 UTC (permalink / raw)
To: Simon Horman; +Cc: lvs-devel, netdev
Use the new add_dest and del_dest methods
to reassign dests.
Signed-off-by: Julian Anastasov <ja@ssi.bg>
---
net/netfilter/ipvs/ip_vs_dh.c | 81 +++++++++++++++++++++++------------------
1 files changed, 45 insertions(+), 36 deletions(-)
diff --git a/net/netfilter/ipvs/ip_vs_dh.c b/net/netfilter/ipvs/ip_vs_dh.c
index 7f3b0cc..ebe80f4 100644
--- a/net/netfilter/ipvs/ip_vs_dh.c
+++ b/net/netfilter/ipvs/ip_vs_dh.c
@@ -51,7 +51,7 @@
* IPVS DH bucket
*/
struct ip_vs_dh_bucket {
- struct ip_vs_dest *dest; /* real server (cache) */
+ struct ip_vs_dest __rcu *dest; /* real server (cache) */
};
/*
@@ -64,6 +64,10 @@ struct ip_vs_dh_bucket {
#define IP_VS_DH_TAB_SIZE (1 << IP_VS_DH_TAB_BITS)
#define IP_VS_DH_TAB_MASK (IP_VS_DH_TAB_SIZE - 1)
+struct ip_vs_dh_state {
+ struct ip_vs_dh_bucket buckets[IP_VS_DH_TAB_SIZE];
+ struct rcu_head rcu_head;
+};
/*
* Returns hash value for IPVS DH entry
@@ -85,10 +89,9 @@ static inline unsigned int ip_vs_dh_hashkey(int af, const union nf_inet_addr *ad
* Get ip_vs_dest associated with supplied parameters.
*/
static inline struct ip_vs_dest *
-ip_vs_dh_get(int af, struct ip_vs_dh_bucket *tbl,
- const union nf_inet_addr *addr)
+ip_vs_dh_get(int af, struct ip_vs_dh_state *s, const union nf_inet_addr *addr)
{
- return (tbl[ip_vs_dh_hashkey(af, addr)]).dest;
+ return rcu_dereference(s->buckets[ip_vs_dh_hashkey(af, addr)].dest);
}
@@ -96,25 +99,30 @@ ip_vs_dh_get(int af, struct ip_vs_dh_bucket *tbl,
* Assign all the hash buckets of the specified table with the service.
*/
static int
-ip_vs_dh_assign(struct ip_vs_dh_bucket *tbl, struct ip_vs_service *svc)
+ip_vs_dh_reassign(struct ip_vs_dh_state *s, struct ip_vs_service *svc)
{
int i;
struct ip_vs_dh_bucket *b;
struct list_head *p;
struct ip_vs_dest *dest;
+ bool empty;
- b = tbl;
+ b = &s->buckets[0];
p = &svc->destinations;
+ empty = list_empty(p);
for (i=0; i<IP_VS_DH_TAB_SIZE; i++) {
- if (list_empty(p)) {
- b->dest = NULL;
- } else {
+ dest = rcu_dereference_protected(b->dest, 1);
+ if (dest)
+ ip_vs_dest_put(dest);
+ if (empty)
+ RCU_INIT_POINTER(b->dest, NULL);
+ else {
if (p == &svc->destinations)
p = p->next;
dest = list_entry(p, struct ip_vs_dest, n_list);
- atomic_inc(&dest->refcnt);
- b->dest = dest;
+ ip_vs_dest_hold(dest);
+ RCU_INIT_POINTER(b->dest, dest);
p = p->next;
}
@@ -127,16 +135,18 @@ ip_vs_dh_assign(struct ip_vs_dh_bucket *tbl, struct ip_vs_service *svc)
/*
* Flush all the hash buckets of the specified table.
*/
-static void ip_vs_dh_flush(struct ip_vs_dh_bucket *tbl)
+static void ip_vs_dh_flush(struct ip_vs_dh_state *s)
{
int i;
struct ip_vs_dh_bucket *b;
+ struct ip_vs_dest *dest;
- b = tbl;
+ b = &s->buckets[0];
for (i=0; i<IP_VS_DH_TAB_SIZE; i++) {
- if (b->dest) {
- atomic_dec(&b->dest->refcnt);
- b->dest = NULL;
+ dest = rcu_dereference_protected(b->dest, 1);
+ if (dest) {
+ ip_vs_dest_put(dest);
+ RCU_INIT_POINTER(b->dest, NULL);
}
b++;
}
@@ -145,21 +155,20 @@ static void ip_vs_dh_flush(struct ip_vs_dh_bucket *tbl)
static int ip_vs_dh_init_svc(struct ip_vs_service *svc)
{
- struct ip_vs_dh_bucket *tbl;
+ struct ip_vs_dh_state *s;
/* allocate the DH table for this service */
- tbl = kmalloc(sizeof(struct ip_vs_dh_bucket)*IP_VS_DH_TAB_SIZE,
- GFP_KERNEL);
- if (tbl == NULL)
+ s = kzalloc(sizeof(struct ip_vs_dh_state), GFP_KERNEL);
+ if (s == NULL)
return -ENOMEM;
- svc->sched_data = tbl;
+ svc->sched_data = s;
IP_VS_DBG(6, "DH hash table (memory=%Zdbytes) allocated for "
"current service\n",
sizeof(struct ip_vs_dh_bucket)*IP_VS_DH_TAB_SIZE);
- /* assign the hash buckets with the updated service */
- ip_vs_dh_assign(tbl, svc);
+ /* assign the hash buckets with current dests */
+ ip_vs_dh_reassign(s, svc);
return 0;
}
@@ -167,13 +176,13 @@ static int ip_vs_dh_init_svc(struct ip_vs_service *svc)
static int ip_vs_dh_done_svc(struct ip_vs_service *svc)
{
- struct ip_vs_dh_bucket *tbl = svc->sched_data;
+ struct ip_vs_dh_state *s = svc->sched_data;
/* got to clean up hash buckets here */
- ip_vs_dh_flush(tbl);
+ ip_vs_dh_flush(s);
/* release the table itself */
- kfree(svc->sched_data);
+ kfree_rcu(s, rcu_head);
IP_VS_DBG(6, "DH hash table (memory=%Zdbytes) released\n",
sizeof(struct ip_vs_dh_bucket)*IP_VS_DH_TAB_SIZE);
@@ -181,15 +190,13 @@ static int ip_vs_dh_done_svc(struct ip_vs_service *svc)
}
-static int ip_vs_dh_update_svc(struct ip_vs_service *svc)
+static int ip_vs_dh_dest_changed(struct ip_vs_service *svc,
+ struct ip_vs_dest *dest)
{
- struct ip_vs_dh_bucket *tbl = svc->sched_data;
-
- /* got to clean up hash buckets here */
- ip_vs_dh_flush(tbl);
+ struct ip_vs_dh_state *s = svc->sched_data;
/* assign the hash buckets with the updated service */
- ip_vs_dh_assign(tbl, svc);
+ ip_vs_dh_reassign(s, svc);
return 0;
}
@@ -212,19 +219,20 @@ static struct ip_vs_dest *
ip_vs_dh_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
{
struct ip_vs_dest *dest;
- struct ip_vs_dh_bucket *tbl;
+ struct ip_vs_dh_state *s;
struct ip_vs_iphdr iph;
ip_vs_fill_iph_addr_only(svc->af, skb, &iph);
IP_VS_DBG(6, "%s(): Scheduling...\n", __func__);
- tbl = (struct ip_vs_dh_bucket *)svc->sched_data;
- dest = ip_vs_dh_get(svc->af, tbl, &iph.daddr);
+ s = (struct ip_vs_dh_state *) svc->sched_data;
+ dest = ip_vs_dh_get(svc->af, s, &iph.daddr);
if (!dest
|| !(dest->flags & IP_VS_DEST_F_AVAILABLE)
|| atomic_read(&dest->weight) <= 0
|| is_overloaded(dest)) {
+ ip_vs_scheduler_err(svc, "no destination available");
return NULL;
}
@@ -248,7 +256,8 @@ static struct ip_vs_scheduler ip_vs_dh_scheduler =
.n_list = LIST_HEAD_INIT(ip_vs_dh_scheduler.n_list),
.init_service = ip_vs_dh_init_svc,
.done_service = ip_vs_dh_done_svc,
- .update_service = ip_vs_dh_update_svc,
+ .add_dest = ip_vs_dh_dest_changed,
+ .del_dest = ip_vs_dh_dest_changed,
.schedule = ip_vs_dh_schedule,
};
--
1.7.3.4
^ permalink raw reply related [flat|nested] 23+ messages in thread* [PATCH net-next 05/19] ipvs: convert lblc scheduler to rcu
2013-03-22 9:46 [PATCH net-next 00/19] IPVS optimizations, part 2 Julian Anastasov
` (3 preceding siblings ...)
2013-03-22 9:46 ` [PATCH net-next 04/19] ipvs: convert dh scheduler to rcu Julian Anastasov
@ 2013-03-22 9:46 ` Julian Anastasov
2013-03-22 9:46 ` [PATCH net-next 06/19] ipvs: convert lblcr " Julian Anastasov
` (14 subsequent siblings)
19 siblings, 0 replies; 23+ messages in thread
From: Julian Anastasov @ 2013-03-22 9:46 UTC (permalink / raw)
To: Simon Horman; +Cc: lvs-devel, netdev
The schedule method now needs _rcu list-traversal
primitive for svc->destinations. The read_lock for sched_lock is
removed. Use a dead flag to prevent new entries to be created
while scheduler is reclaimed. Use hlist for the hash table.
Signed-off-by: Julian Anastasov <ja@ssi.bg>
---
net/netfilter/ipvs/ip_vs_lblc.c | 96 ++++++++++++++++++++++-----------------
1 files changed, 55 insertions(+), 41 deletions(-)
diff --git a/net/netfilter/ipvs/ip_vs_lblc.c b/net/netfilter/ipvs/ip_vs_lblc.c
index fdd89b9..b873e17 100644
--- a/net/netfilter/ipvs/ip_vs_lblc.c
+++ b/net/netfilter/ipvs/ip_vs_lblc.c
@@ -90,11 +90,12 @@
* IP address and its destination server
*/
struct ip_vs_lblc_entry {
- struct list_head list;
+ struct hlist_node list;
int af; /* address family */
union nf_inet_addr addr; /* destination IP address */
- struct ip_vs_dest *dest; /* real server (cache) */
+ struct ip_vs_dest __rcu *dest; /* real server (cache) */
unsigned long lastuse; /* last used time */
+ struct rcu_head rcu_head;
};
@@ -102,12 +103,14 @@ struct ip_vs_lblc_entry {
* IPVS lblc hash table
*/
struct ip_vs_lblc_table {
- struct list_head bucket[IP_VS_LBLC_TAB_SIZE]; /* hash bucket */
+ struct rcu_head rcu_head;
+ struct hlist_head __rcu bucket[IP_VS_LBLC_TAB_SIZE]; /* hash bucket */
+ struct timer_list periodic_timer; /* collect stale entries */
atomic_t entries; /* number of entries */
int max_size; /* maximum size of entries */
- struct timer_list periodic_timer; /* collect stale entries */
int rover; /* rover for expire check */
int counter; /* counter for no expire */
+ bool dead;
};
@@ -129,13 +132,16 @@ static ctl_table vs_vars_table[] = {
static inline void ip_vs_lblc_free(struct ip_vs_lblc_entry *en)
{
- list_del(&en->list);
+ struct ip_vs_dest *dest;
+
+ hlist_del_rcu(&en->list);
/*
* We don't kfree dest because it is referred either by its service
* or the trash dest list.
*/
- atomic_dec(&en->dest->refcnt);
- kfree(en);
+ dest = rcu_dereference_protected(en->dest, 1);
+ ip_vs_dest_put(dest);
+ kfree_rcu(en, rcu_head);
}
@@ -165,15 +171,12 @@ ip_vs_lblc_hash(struct ip_vs_lblc_table *tbl, struct ip_vs_lblc_entry *en)
{
unsigned int hash = ip_vs_lblc_hashkey(en->af, &en->addr);
- list_add(&en->list, &tbl->bucket[hash]);
+ hlist_add_head_rcu(&en->list, &tbl->bucket[hash]);
atomic_inc(&tbl->entries);
}
-/*
- * Get ip_vs_lblc_entry associated with supplied parameters. Called under read
- * lock
- */
+/* Get ip_vs_lblc_entry associated with supplied parameters. */
static inline struct ip_vs_lblc_entry *
ip_vs_lblc_get(int af, struct ip_vs_lblc_table *tbl,
const union nf_inet_addr *addr)
@@ -181,7 +184,7 @@ ip_vs_lblc_get(int af, struct ip_vs_lblc_table *tbl,
unsigned int hash = ip_vs_lblc_hashkey(af, addr);
struct ip_vs_lblc_entry *en;
- list_for_each_entry(en, &tbl->bucket[hash], list)
+ hlist_for_each_entry_rcu(en, &tbl->bucket[hash], list)
if (ip_vs_addr_equal(af, &en->addr, addr))
return en;
@@ -209,14 +212,20 @@ ip_vs_lblc_new(struct ip_vs_lblc_table *tbl, const union nf_inet_addr *daddr,
ip_vs_addr_copy(dest->af, &en->addr, daddr);
en->lastuse = jiffies;
- atomic_inc(&dest->refcnt);
- en->dest = dest;
+ ip_vs_dest_hold(dest);
+ RCU_INIT_POINTER(en->dest, dest);
ip_vs_lblc_hash(tbl, en);
- } else if (en->dest != dest) {
- atomic_dec(&en->dest->refcnt);
- atomic_inc(&dest->refcnt);
- en->dest = dest;
+ } else {
+ struct ip_vs_dest *old_dest;
+
+ old_dest = rcu_dereference_protected(en->dest, 1);
+ if (old_dest != dest) {
+ ip_vs_dest_put(old_dest);
+ ip_vs_dest_hold(dest);
+ /* No ordering constraints for refcnt */
+ RCU_INIT_POINTER(en->dest, dest);
+ }
}
return en;
@@ -226,17 +235,22 @@ ip_vs_lblc_new(struct ip_vs_lblc_table *tbl, const union nf_inet_addr *daddr,
/*
* Flush all the entries of the specified table.
*/
-static void ip_vs_lblc_flush(struct ip_vs_lblc_table *tbl)
+static void ip_vs_lblc_flush(struct ip_vs_service *svc)
{
- struct ip_vs_lblc_entry *en, *nxt;
+ struct ip_vs_lblc_table *tbl = svc->sched_data;
+ struct ip_vs_lblc_entry *en;
+ struct hlist_node *next;
int i;
+ write_lock_bh(&svc->sched_lock);
+ tbl->dead = 1;
for (i=0; i<IP_VS_LBLC_TAB_SIZE; i++) {
- list_for_each_entry_safe(en, nxt, &tbl->bucket[i], list) {
+ hlist_for_each_entry_safe(en, next, &tbl->bucket[i], list) {
ip_vs_lblc_free(en);
atomic_dec(&tbl->entries);
}
}
+ write_unlock_bh(&svc->sched_lock);
}
static int sysctl_lblc_expiration(struct ip_vs_service *svc)
@@ -252,7 +266,8 @@ static int sysctl_lblc_expiration(struct ip_vs_service *svc)
static inline void ip_vs_lblc_full_check(struct ip_vs_service *svc)
{
struct ip_vs_lblc_table *tbl = svc->sched_data;
- struct ip_vs_lblc_entry *en, *nxt;
+ struct ip_vs_lblc_entry *en;
+ struct hlist_node *next;
unsigned long now = jiffies;
int i, j;
@@ -260,7 +275,7 @@ static inline void ip_vs_lblc_full_check(struct ip_vs_service *svc)
j = (j + 1) & IP_VS_LBLC_TAB_MASK;
write_lock(&svc->sched_lock);
- list_for_each_entry_safe(en, nxt, &tbl->bucket[j], list) {
+ hlist_for_each_entry_safe(en, next, &tbl->bucket[j], list) {
if (time_before(now,
en->lastuse +
sysctl_lblc_expiration(svc)))
@@ -293,7 +308,8 @@ static void ip_vs_lblc_check_expire(unsigned long data)
unsigned long now = jiffies;
int goal;
int i, j;
- struct ip_vs_lblc_entry *en, *nxt;
+ struct ip_vs_lblc_entry *en;
+ struct hlist_node *next;
if ((tbl->counter % COUNT_FOR_FULL_EXPIRATION) == 0) {
/* do full expiration check */
@@ -315,7 +331,7 @@ static void ip_vs_lblc_check_expire(unsigned long data)
j = (j + 1) & IP_VS_LBLC_TAB_MASK;
write_lock(&svc->sched_lock);
- list_for_each_entry_safe(en, nxt, &tbl->bucket[j], list) {
+ hlist_for_each_entry_safe(en, next, &tbl->bucket[j], list) {
if (time_before(now, en->lastuse + ENTRY_TIMEOUT))
continue;
@@ -354,11 +370,12 @@ static int ip_vs_lblc_init_svc(struct ip_vs_service *svc)
* Initialize the hash buckets
*/
for (i=0; i<IP_VS_LBLC_TAB_SIZE; i++) {
- INIT_LIST_HEAD(&tbl->bucket[i]);
+ INIT_HLIST_HEAD(&tbl->bucket[i]);
}
tbl->max_size = IP_VS_LBLC_TAB_SIZE*16;
tbl->rover = 0;
tbl->counter = 1;
+ tbl->dead = 0;
/*
* Hook periodic timer for garbage collection
@@ -379,10 +396,10 @@ static int ip_vs_lblc_done_svc(struct ip_vs_service *svc)
del_timer_sync(&tbl->periodic_timer);
/* got to clean up table entries here */
- ip_vs_lblc_flush(tbl);
+ ip_vs_lblc_flush(svc);
/* release the table itself */
- kfree(tbl);
+ kfree_rcu(tbl, rcu_head);
IP_VS_DBG(6, "LBLC hash table (memory=%Zdbytes) released\n",
sizeof(*tbl));
@@ -408,7 +425,7 @@ __ip_vs_lblc_schedule(struct ip_vs_service *svc)
* The server with weight=0 is quiesced and will not receive any
* new connection.
*/
- list_for_each_entry(dest, &svc->destinations, n_list) {
+ list_for_each_entry_rcu(dest, &svc->destinations, n_list) {
if (dest->flags & IP_VS_DEST_F_OVERLOAD)
continue;
if (atomic_read(&dest->weight) > 0) {
@@ -423,7 +440,7 @@ __ip_vs_lblc_schedule(struct ip_vs_service *svc)
* Find the destination with the least load.
*/
nextstage:
- list_for_each_entry_continue(dest, &svc->destinations, n_list) {
+ list_for_each_entry_continue_rcu(dest, &svc->destinations, n_list) {
if (dest->flags & IP_VS_DEST_F_OVERLOAD)
continue;
@@ -457,7 +474,7 @@ is_overloaded(struct ip_vs_dest *dest, struct ip_vs_service *svc)
if (atomic_read(&dest->activeconns) > atomic_read(&dest->weight)) {
struct ip_vs_dest *d;
- list_for_each_entry(d, &svc->destinations, n_list) {
+ list_for_each_entry_rcu(d, &svc->destinations, n_list) {
if (atomic_read(&d->activeconns)*2
< atomic_read(&d->weight)) {
return 1;
@@ -484,7 +501,6 @@ ip_vs_lblc_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
IP_VS_DBG(6, "%s(): Scheduling...\n", __func__);
/* First look in our cache */
- read_lock(&svc->sched_lock);
en = ip_vs_lblc_get(svc->af, tbl, &iph.daddr);
if (en) {
/* We only hold a read lock, but this is atomic */
@@ -499,14 +515,11 @@ ip_vs_lblc_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
* free up entries from the trash at any time.
*/
- if (en->dest->flags & IP_VS_DEST_F_AVAILABLE)
- dest = en->dest;
+ dest = rcu_dereference(en->dest);
+ if ((dest->flags & IP_VS_DEST_F_AVAILABLE) &&
+ atomic_read(&dest->weight) > 0 && !is_overloaded(dest, svc))
+ goto out;
}
- read_unlock(&svc->sched_lock);
-
- /* If the destination has a weight and is not overloaded, use it */
- if (dest && atomic_read(&dest->weight) > 0 && !is_overloaded(dest, svc))
- goto out;
/* No cache entry or it is invalid, time to schedule */
dest = __ip_vs_lblc_schedule(svc);
@@ -517,7 +530,8 @@ ip_vs_lblc_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
/* If we fail to create a cache entry, we'll just use the valid dest */
write_lock(&svc->sched_lock);
- ip_vs_lblc_new(tbl, &iph.daddr, dest);
+ if (!tbl->dead)
+ ip_vs_lblc_new(tbl, &iph.daddr, dest);
write_unlock(&svc->sched_lock);
out:
--
1.7.3.4
^ permalink raw reply related [flat|nested] 23+ messages in thread* [PATCH net-next 06/19] ipvs: convert lblcr scheduler to rcu
2013-03-22 9:46 [PATCH net-next 00/19] IPVS optimizations, part 2 Julian Anastasov
` (4 preceding siblings ...)
2013-03-22 9:46 ` [PATCH net-next 05/19] ipvs: convert lblc " Julian Anastasov
@ 2013-03-22 9:46 ` Julian Anastasov
2013-03-22 9:46 ` [PATCH net-next 07/19] ipvs: convert lc " Julian Anastasov
` (13 subsequent siblings)
19 siblings, 0 replies; 23+ messages in thread
From: Julian Anastasov @ 2013-03-22 9:46 UTC (permalink / raw)
To: Simon Horman; +Cc: lvs-devel, netdev
The schedule method now needs _rcu list-traversal
primitive for svc->destinations. The read_lock for sched_lock is
removed. The set.lock is removed because now it is used in
rare cases, mostly under sched_lock.
Signed-off-by: Julian Anastasov <ja@ssi.bg>
---
net/netfilter/ipvs/ip_vs_lblcr.c | 171 ++++++++++++++++++++------------------
1 files changed, 90 insertions(+), 81 deletions(-)
diff --git a/net/netfilter/ipvs/ip_vs_lblcr.c b/net/netfilter/ipvs/ip_vs_lblcr.c
index c03b6a3..c22f173 100644
--- a/net/netfilter/ipvs/ip_vs_lblcr.c
+++ b/net/netfilter/ipvs/ip_vs_lblcr.c
@@ -89,40 +89,44 @@
*/
struct ip_vs_dest_set_elem {
struct list_head list; /* list link */
- struct ip_vs_dest *dest; /* destination server */
+ struct ip_vs_dest __rcu *dest; /* destination server */
+ struct rcu_head rcu_head;
};
struct ip_vs_dest_set {
atomic_t size; /* set size */
unsigned long lastmod; /* last modified time */
struct list_head list; /* destination list */
- rwlock_t lock; /* lock for this list */
};
-static struct ip_vs_dest_set_elem *
-ip_vs_dest_set_insert(struct ip_vs_dest_set *set, struct ip_vs_dest *dest)
+static void ip_vs_dest_set_insert(struct ip_vs_dest_set *set,
+ struct ip_vs_dest *dest, bool check)
{
struct ip_vs_dest_set_elem *e;
- list_for_each_entry(e, &set->list, list) {
- if (e->dest == dest)
- /* already existed */
- return NULL;
+ if (check) {
+ list_for_each_entry(e, &set->list, list) {
+ struct ip_vs_dest *d;
+
+ d = rcu_dereference_protected(e->dest, 1);
+ if (d == dest)
+ /* already existed */
+ return;
+ }
}
e = kmalloc(sizeof(*e), GFP_ATOMIC);
if (e == NULL)
- return NULL;
+ return;
- atomic_inc(&dest->refcnt);
- e->dest = dest;
+ ip_vs_dest_hold(dest);
+ RCU_INIT_POINTER(e->dest, dest);
- list_add(&e->list, &set->list);
+ list_add_rcu(&e->list, &set->list);
atomic_inc(&set->size);
set->lastmod = jiffies;
- return e;
}
static void
@@ -131,13 +135,16 @@ ip_vs_dest_set_erase(struct ip_vs_dest_set *set, struct ip_vs_dest *dest)
struct ip_vs_dest_set_elem *e;
list_for_each_entry(e, &set->list, list) {
- if (e->dest == dest) {
+ struct ip_vs_dest *d;
+
+ d = rcu_dereference_protected(e->dest, 1);
+ if (d == dest) {
/* HIT */
atomic_dec(&set->size);
set->lastmod = jiffies;
- atomic_dec(&e->dest->refcnt);
- list_del(&e->list);
- kfree(e);
+ ip_vs_dest_put(dest);
+ list_del_rcu(&e->list);
+ kfree_rcu(e, rcu_head);
break;
}
}
@@ -147,17 +154,18 @@ static void ip_vs_dest_set_eraseall(struct ip_vs_dest_set *set)
{
struct ip_vs_dest_set_elem *e, *ep;
- write_lock(&set->lock);
list_for_each_entry_safe(e, ep, &set->list, list) {
+ struct ip_vs_dest *d;
+
+ d = rcu_dereference_protected(e->dest, 1);
/*
* We don't kfree dest because it is referred either
* by its service or by the trash dest list.
*/
- atomic_dec(&e->dest->refcnt);
- list_del(&e->list);
- kfree(e);
+ ip_vs_dest_put(d);
+ list_del_rcu(&e->list);
+ kfree_rcu(e, rcu_head);
}
- write_unlock(&set->lock);
}
/* get weighted least-connection node in the destination set */
@@ -171,8 +179,8 @@ static inline struct ip_vs_dest *ip_vs_dest_set_min(struct ip_vs_dest_set *set)
return NULL;
/* select the first destination server, whose weight > 0 */
- list_for_each_entry(e, &set->list, list) {
- least = e->dest;
+ list_for_each_entry_rcu(e, &set->list, list) {
+ least = rcu_dereference(e->dest);
if (least->flags & IP_VS_DEST_F_OVERLOAD)
continue;
@@ -186,8 +194,8 @@ static inline struct ip_vs_dest *ip_vs_dest_set_min(struct ip_vs_dest_set *set)
/* find the destination with the weighted least load */
nextstage:
- list_for_each_entry(e, &set->list, list) {
- dest = e->dest;
+ list_for_each_entry_continue_rcu(e, &set->list, list) {
+ dest = rcu_dereference(e->dest);
if (dest->flags & IP_VS_DEST_F_OVERLOAD)
continue;
@@ -224,7 +232,7 @@ static inline struct ip_vs_dest *ip_vs_dest_set_max(struct ip_vs_dest_set *set)
/* select the first destination server, whose weight > 0 */
list_for_each_entry(e, &set->list, list) {
- most = e->dest;
+ most = rcu_dereference_protected(e->dest, 1);
if (atomic_read(&most->weight) > 0) {
moh = ip_vs_dest_conn_overhead(most);
goto nextstage;
@@ -234,8 +242,8 @@ static inline struct ip_vs_dest *ip_vs_dest_set_max(struct ip_vs_dest_set *set)
/* find the destination with the weighted most load */
nextstage:
- list_for_each_entry(e, &set->list, list) {
- dest = e->dest;
+ list_for_each_entry_continue(e, &set->list, list) {
+ dest = rcu_dereference_protected(e->dest, 1);
doh = ip_vs_dest_conn_overhead(dest);
/* moh/mw < doh/dw ==> moh*dw < doh*mw, where mw,dw>0 */
if ((moh * atomic_read(&dest->weight) <
@@ -262,11 +270,12 @@ static inline struct ip_vs_dest *ip_vs_dest_set_max(struct ip_vs_dest_set *set)
* IP address and its destination server set
*/
struct ip_vs_lblcr_entry {
- struct list_head list;
+ struct hlist_node list;
int af; /* address family */
union nf_inet_addr addr; /* destination IP address */
struct ip_vs_dest_set set; /* destination server set */
unsigned long lastuse; /* last used time */
+ struct rcu_head rcu_head;
};
@@ -274,12 +283,14 @@ struct ip_vs_lblcr_entry {
* IPVS lblcr hash table
*/
struct ip_vs_lblcr_table {
- struct list_head bucket[IP_VS_LBLCR_TAB_SIZE]; /* hash bucket */
+ struct rcu_head rcu_head;
+ struct hlist_head __rcu bucket[IP_VS_LBLCR_TAB_SIZE]; /* hash bucket */
atomic_t entries; /* number of entries */
int max_size; /* maximum size of entries */
struct timer_list periodic_timer; /* collect stale entries */
int rover; /* rover for expire check */
int counter; /* counter for no expire */
+ bool dead;
};
@@ -302,9 +313,9 @@ static ctl_table vs_vars_table[] = {
static inline void ip_vs_lblcr_free(struct ip_vs_lblcr_entry *en)
{
- list_del(&en->list);
+ hlist_del_rcu(&en->list);
ip_vs_dest_set_eraseall(&en->set);
- kfree(en);
+ kfree_rcu(en, rcu_head);
}
@@ -334,15 +345,12 @@ ip_vs_lblcr_hash(struct ip_vs_lblcr_table *tbl, struct ip_vs_lblcr_entry *en)
{
unsigned int hash = ip_vs_lblcr_hashkey(en->af, &en->addr);
- list_add(&en->list, &tbl->bucket[hash]);
+ hlist_add_head_rcu(&en->list, &tbl->bucket[hash]);
atomic_inc(&tbl->entries);
}
-/*
- * Get ip_vs_lblcr_entry associated with supplied parameters. Called under
- * read lock.
- */
+/* Get ip_vs_lblcr_entry associated with supplied parameters. */
static inline struct ip_vs_lblcr_entry *
ip_vs_lblcr_get(int af, struct ip_vs_lblcr_table *tbl,
const union nf_inet_addr *addr)
@@ -350,7 +358,7 @@ ip_vs_lblcr_get(int af, struct ip_vs_lblcr_table *tbl,
unsigned int hash = ip_vs_lblcr_hashkey(af, addr);
struct ip_vs_lblcr_entry *en;
- list_for_each_entry(en, &tbl->bucket[hash], list)
+ hlist_for_each_entry_rcu(en, &tbl->bucket[hash], list)
if (ip_vs_addr_equal(af, &en->addr, addr))
return en;
@@ -381,14 +389,14 @@ ip_vs_lblcr_new(struct ip_vs_lblcr_table *tbl, const union nf_inet_addr *daddr,
/* initialize its dest set */
atomic_set(&(en->set.size), 0);
INIT_LIST_HEAD(&en->set.list);
- rwlock_init(&en->set.lock);
+
+ ip_vs_dest_set_insert(&en->set, dest, false);
ip_vs_lblcr_hash(tbl, en);
+ return en;
}
- write_lock(&en->set.lock);
- ip_vs_dest_set_insert(&en->set, dest);
- write_unlock(&en->set.lock);
+ ip_vs_dest_set_insert(&en->set, dest, true);
return en;
}
@@ -397,17 +405,21 @@ ip_vs_lblcr_new(struct ip_vs_lblcr_table *tbl, const union nf_inet_addr *daddr,
/*
* Flush all the entries of the specified table.
*/
-static void ip_vs_lblcr_flush(struct ip_vs_lblcr_table *tbl)
+static void ip_vs_lblcr_flush(struct ip_vs_service *svc)
{
+ struct ip_vs_lblcr_table *tbl = svc->sched_data;
int i;
- struct ip_vs_lblcr_entry *en, *nxt;
+ struct ip_vs_lblcr_entry *en;
+ struct hlist_node *next;
- /* No locking required, only called during cleanup. */
+ write_lock_bh(&svc->sched_lock);
+ tbl->dead = 1;
for (i=0; i<IP_VS_LBLCR_TAB_SIZE; i++) {
- list_for_each_entry_safe(en, nxt, &tbl->bucket[i], list) {
+ hlist_for_each_entry_safe(en, next, &tbl->bucket[i], list) {
ip_vs_lblcr_free(en);
}
}
+ write_unlock_bh(&svc->sched_lock);
}
static int sysctl_lblcr_expiration(struct ip_vs_service *svc)
@@ -425,13 +437,14 @@ static inline void ip_vs_lblcr_full_check(struct ip_vs_service *svc)
struct ip_vs_lblcr_table *tbl = svc->sched_data;
unsigned long now = jiffies;
int i, j;
- struct ip_vs_lblcr_entry *en, *nxt;
+ struct ip_vs_lblcr_entry *en;
+ struct hlist_node *next;
for (i=0, j=tbl->rover; i<IP_VS_LBLCR_TAB_SIZE; i++) {
j = (j + 1) & IP_VS_LBLCR_TAB_MASK;
write_lock(&svc->sched_lock);
- list_for_each_entry_safe(en, nxt, &tbl->bucket[j], list) {
+ hlist_for_each_entry_safe(en, next, &tbl->bucket[j], list) {
if (time_after(en->lastuse +
sysctl_lblcr_expiration(svc), now))
continue;
@@ -463,7 +476,8 @@ static void ip_vs_lblcr_check_expire(unsigned long data)
unsigned long now = jiffies;
int goal;
int i, j;
- struct ip_vs_lblcr_entry *en, *nxt;
+ struct ip_vs_lblcr_entry *en;
+ struct hlist_node *next;
if ((tbl->counter % COUNT_FOR_FULL_EXPIRATION) == 0) {
/* do full expiration check */
@@ -485,7 +499,7 @@ static void ip_vs_lblcr_check_expire(unsigned long data)
j = (j + 1) & IP_VS_LBLCR_TAB_MASK;
write_lock(&svc->sched_lock);
- list_for_each_entry_safe(en, nxt, &tbl->bucket[j], list) {
+ hlist_for_each_entry_safe(en, next, &tbl->bucket[j], list) {
if (time_before(now, en->lastuse+ENTRY_TIMEOUT))
continue;
@@ -523,11 +537,12 @@ static int ip_vs_lblcr_init_svc(struct ip_vs_service *svc)
* Initialize the hash buckets
*/
for (i=0; i<IP_VS_LBLCR_TAB_SIZE; i++) {
- INIT_LIST_HEAD(&tbl->bucket[i]);
+ INIT_HLIST_HEAD(&tbl->bucket[i]);
}
tbl->max_size = IP_VS_LBLCR_TAB_SIZE*16;
tbl->rover = 0;
tbl->counter = 1;
+ tbl->dead = 0;
/*
* Hook periodic timer for garbage collection
@@ -548,10 +563,10 @@ static int ip_vs_lblcr_done_svc(struct ip_vs_service *svc)
del_timer_sync(&tbl->periodic_timer);
/* got to clean up table entries here */
- ip_vs_lblcr_flush(tbl);
+ ip_vs_lblcr_flush(svc);
/* release the table itself */
- kfree(tbl);
+ kfree_rcu(tbl, rcu_head);
IP_VS_DBG(6, "LBLCR hash table (memory=%Zdbytes) released\n",
sizeof(*tbl));
@@ -577,7 +592,7 @@ __ip_vs_lblcr_schedule(struct ip_vs_service *svc)
* The server with weight=0 is quiesced and will not receive any
* new connection.
*/
- list_for_each_entry(dest, &svc->destinations, n_list) {
+ list_for_each_entry_rcu(dest, &svc->destinations, n_list) {
if (dest->flags & IP_VS_DEST_F_OVERLOAD)
continue;
@@ -593,7 +608,7 @@ __ip_vs_lblcr_schedule(struct ip_vs_service *svc)
* Find the destination with the least load.
*/
nextstage:
- list_for_each_entry_continue(dest, &svc->destinations, n_list) {
+ list_for_each_entry_continue_rcu(dest, &svc->destinations, n_list) {
if (dest->flags & IP_VS_DEST_F_OVERLOAD)
continue;
@@ -627,7 +642,7 @@ is_overloaded(struct ip_vs_dest *dest, struct ip_vs_service *svc)
if (atomic_read(&dest->activeconns) > atomic_read(&dest->weight)) {
struct ip_vs_dest *d;
- list_for_each_entry(d, &svc->destinations, n_list) {
+ list_for_each_entry_rcu(d, &svc->destinations, n_list) {
if (atomic_read(&d->activeconns)*2
< atomic_read(&d->weight)) {
return 1;
@@ -646,7 +661,7 @@ ip_vs_lblcr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
{
struct ip_vs_lblcr_table *tbl = svc->sched_data;
struct ip_vs_iphdr iph;
- struct ip_vs_dest *dest = NULL;
+ struct ip_vs_dest *dest;
struct ip_vs_lblcr_entry *en;
ip_vs_fill_iph_addr_only(svc->af, skb, &iph);
@@ -654,53 +669,46 @@ ip_vs_lblcr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
IP_VS_DBG(6, "%s(): Scheduling...\n", __func__);
/* First look in our cache */
- read_lock(&svc->sched_lock);
en = ip_vs_lblcr_get(svc->af, tbl, &iph.daddr);
if (en) {
- /* We only hold a read lock, but this is atomic */
en->lastuse = jiffies;
/* Get the least loaded destination */
- read_lock(&en->set.lock);
dest = ip_vs_dest_set_min(&en->set);
- read_unlock(&en->set.lock);
/* More than one destination + enough time passed by, cleanup */
if (atomic_read(&en->set.size) > 1 &&
- time_after(jiffies, en->set.lastmod +
+ time_after(jiffies, en->set.lastmod +
sysctl_lblcr_expiration(svc))) {
- struct ip_vs_dest *m;
+ write_lock(&svc->sched_lock);
+ if (atomic_read(&en->set.size) > 1) {
+ struct ip_vs_dest *m;
- write_lock(&en->set.lock);
- m = ip_vs_dest_set_max(&en->set);
- if (m)
- ip_vs_dest_set_erase(&en->set, m);
- write_unlock(&en->set.lock);
+ m = ip_vs_dest_set_max(&en->set);
+ if (m)
+ ip_vs_dest_set_erase(&en->set, m);
+ }
+ write_unlock(&svc->sched_lock);
}
/* If the destination is not overloaded, use it */
- if (dest && !is_overloaded(dest, svc)) {
- read_unlock(&svc->sched_lock);
+ if (dest && !is_overloaded(dest, svc))
goto out;
- }
/* The cache entry is invalid, time to schedule */
dest = __ip_vs_lblcr_schedule(svc);
if (!dest) {
ip_vs_scheduler_err(svc, "no destination available");
- read_unlock(&svc->sched_lock);
return NULL;
}
/* Update our cache entry */
- write_lock(&en->set.lock);
- ip_vs_dest_set_insert(&en->set, dest);
- write_unlock(&en->set.lock);
- }
- read_unlock(&svc->sched_lock);
-
- if (dest)
+ write_lock(&svc->sched_lock);
+ if (!tbl->dead)
+ ip_vs_dest_set_insert(&en->set, dest, true);
+ write_unlock(&svc->sched_lock);
goto out;
+ }
/* No cache entry, time to schedule */
dest = __ip_vs_lblcr_schedule(svc);
@@ -711,7 +719,8 @@ ip_vs_lblcr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
/* If we fail to create a cache entry, we'll just use the valid dest */
write_lock(&svc->sched_lock);
- ip_vs_lblcr_new(tbl, &iph.daddr, dest);
+ if (!tbl->dead)
+ ip_vs_lblcr_new(tbl, &iph.daddr, dest);
write_unlock(&svc->sched_lock);
out:
--
1.7.3.4
^ permalink raw reply related [flat|nested] 23+ messages in thread* [PATCH net-next 07/19] ipvs: convert lc scheduler to rcu
2013-03-22 9:46 [PATCH net-next 00/19] IPVS optimizations, part 2 Julian Anastasov
` (5 preceding siblings ...)
2013-03-22 9:46 ` [PATCH net-next 06/19] ipvs: convert lblcr " Julian Anastasov
@ 2013-03-22 9:46 ` Julian Anastasov
2013-03-22 9:46 ` [PATCH net-next 08/19] ipvs: convert nq " Julian Anastasov
` (12 subsequent siblings)
19 siblings, 0 replies; 23+ messages in thread
From: Julian Anastasov @ 2013-03-22 9:46 UTC (permalink / raw)
To: Simon Horman; +Cc: lvs-devel, netdev
The schedule method now needs _rcu list-traversal
primitive for svc->destinations.
Signed-off-by: Julian Anastasov <ja@ssi.bg>
---
net/netfilter/ipvs/ip_vs_lc.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/net/netfilter/ipvs/ip_vs_lc.c b/net/netfilter/ipvs/ip_vs_lc.c
index f391819..0cabf78 100644
--- a/net/netfilter/ipvs/ip_vs_lc.c
+++ b/net/netfilter/ipvs/ip_vs_lc.c
@@ -42,7 +42,7 @@ ip_vs_lc_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
* served, but no new connection is assigned to the server.
*/
- list_for_each_entry(dest, &svc->destinations, n_list) {
+ list_for_each_entry_rcu(dest, &svc->destinations, n_list) {
if ((dest->flags & IP_VS_DEST_F_OVERLOAD) ||
atomic_read(&dest->weight) == 0)
continue;
--
1.7.3.4
^ permalink raw reply related [flat|nested] 23+ messages in thread* [PATCH net-next 08/19] ipvs: convert nq scheduler to rcu
2013-03-22 9:46 [PATCH net-next 00/19] IPVS optimizations, part 2 Julian Anastasov
` (6 preceding siblings ...)
2013-03-22 9:46 ` [PATCH net-next 07/19] ipvs: convert lc " Julian Anastasov
@ 2013-03-22 9:46 ` Julian Anastasov
2013-03-22 9:46 ` [PATCH net-next 09/19] ipvs: convert rr " Julian Anastasov
` (11 subsequent siblings)
19 siblings, 0 replies; 23+ messages in thread
From: Julian Anastasov @ 2013-03-22 9:46 UTC (permalink / raw)
To: Simon Horman; +Cc: lvs-devel, netdev
The schedule method now needs _rcu list-traversal
primitive for svc->destinations.
Signed-off-by: Julian Anastasov <ja@ssi.bg>
---
net/netfilter/ipvs/ip_vs_nq.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/net/netfilter/ipvs/ip_vs_nq.c b/net/netfilter/ipvs/ip_vs_nq.c
index 984d9c1..51dc0cf 100644
--- a/net/netfilter/ipvs/ip_vs_nq.c
+++ b/net/netfilter/ipvs/ip_vs_nq.c
@@ -75,7 +75,7 @@ ip_vs_nq_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
* new connections.
*/
- list_for_each_entry(dest, &svc->destinations, n_list) {
+ list_for_each_entry_rcu(dest, &svc->destinations, n_list) {
if (dest->flags & IP_VS_DEST_F_OVERLOAD ||
!atomic_read(&dest->weight))
--
1.7.3.4
^ permalink raw reply related [flat|nested] 23+ messages in thread* [PATCH net-next 09/19] ipvs: convert rr scheduler to rcu
2013-03-22 9:46 [PATCH net-next 00/19] IPVS optimizations, part 2 Julian Anastasov
` (7 preceding siblings ...)
2013-03-22 9:46 ` [PATCH net-next 08/19] ipvs: convert nq " Julian Anastasov
@ 2013-03-22 9:46 ` Julian Anastasov
2013-03-22 9:46 ` [PATCH net-next 10/19] ipvs: convert sed " Julian Anastasov
` (10 subsequent siblings)
19 siblings, 0 replies; 23+ messages in thread
From: Julian Anastasov @ 2013-03-22 9:46 UTC (permalink / raw)
To: Simon Horman; +Cc: lvs-devel, netdev
The schedule method now needs _rcu list-traversal
primitive for svc->destinations. As the previous entry
could be unlinked, limit the list traversals to 2 when
lookup started from previous entry.
Signed-off-by: Julian Anastasov <ja@ssi.bg>
---
net/netfilter/ipvs/ip_vs_rr.c | 55 ++++++++++++++++++++++++++---------------
1 files changed, 35 insertions(+), 20 deletions(-)
diff --git a/net/netfilter/ipvs/ip_vs_rr.c b/net/netfilter/ipvs/ip_vs_rr.c
index c49b388..3942890 100644
--- a/net/netfilter/ipvs/ip_vs_rr.c
+++ b/net/netfilter/ipvs/ip_vs_rr.c
@@ -35,9 +35,18 @@ static int ip_vs_rr_init_svc(struct ip_vs_service *svc)
}
-static int ip_vs_rr_update_svc(struct ip_vs_service *svc)
+static int ip_vs_rr_del_dest(struct ip_vs_service *svc, struct ip_vs_dest *dest)
{
- svc->sched_data = &svc->destinations;
+ struct list_head *p;
+
+ write_lock_bh(&svc->sched_lock);
+ p = (struct list_head *) svc->sched_data;
+ /* dest is already unlinked, so p->prev is not valid but
+ * p->next is valid, use it to reach previous entry.
+ */
+ if (p == &dest->n_list)
+ svc->sched_data = p->next->prev;
+ write_unlock_bh(&svc->sched_lock);
return 0;
}
@@ -48,35 +57,40 @@ static int ip_vs_rr_update_svc(struct ip_vs_service *svc)
static struct ip_vs_dest *
ip_vs_rr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
{
- struct list_head *p, *q;
- struct ip_vs_dest *dest;
+ struct list_head *p;
+ struct ip_vs_dest *dest, *last;
+ int pass = 0;
IP_VS_DBG(6, "%s(): Scheduling...\n", __func__);
write_lock(&svc->sched_lock);
- p = (struct list_head *)svc->sched_data;
- p = p->next;
- q = p;
+ p = (struct list_head *) svc->sched_data;
+ last = dest = list_entry(p, struct ip_vs_dest, n_list);
+
do {
- /* skip list head */
- if (q == &svc->destinations) {
- q = q->next;
- continue;
+ list_for_each_entry_continue_rcu(dest,
+ &svc->destinations,
+ n_list) {
+ if (!(dest->flags & IP_VS_DEST_F_OVERLOAD) &&
+ atomic_read(&dest->weight) > 0)
+ /* HIT */
+ goto out;
+ if (dest == last)
+ goto stop;
}
+ pass++;
+ /* Previous dest could be unlinked, do not loop forever.
+ * If we stay at head there is no need for 2nd pass.
+ */
+ } while (pass < 2 && p != &svc->destinations);
- dest = list_entry(q, struct ip_vs_dest, n_list);
- if (!(dest->flags & IP_VS_DEST_F_OVERLOAD) &&
- atomic_read(&dest->weight) > 0)
- /* HIT */
- goto out;
- q = q->next;
- } while (q != p);
+stop:
write_unlock(&svc->sched_lock);
ip_vs_scheduler_err(svc, "no destination available");
return NULL;
out:
- svc->sched_data = q;
+ svc->sched_data = &dest->n_list;
write_unlock(&svc->sched_lock);
IP_VS_DBG_BUF(6, "RR: server %s:%u "
"activeconns %d refcnt %d weight %d\n",
@@ -94,7 +108,8 @@ static struct ip_vs_scheduler ip_vs_rr_scheduler = {
.module = THIS_MODULE,
.n_list = LIST_HEAD_INIT(ip_vs_rr_scheduler.n_list),
.init_service = ip_vs_rr_init_svc,
- .update_service = ip_vs_rr_update_svc,
+ .add_dest = NULL,
+ .del_dest = ip_vs_rr_del_dest,
.schedule = ip_vs_rr_schedule,
};
--
1.7.3.4
^ permalink raw reply related [flat|nested] 23+ messages in thread* [PATCH net-next 10/19] ipvs: convert sed scheduler to rcu
2013-03-22 9:46 [PATCH net-next 00/19] IPVS optimizations, part 2 Julian Anastasov
` (8 preceding siblings ...)
2013-03-22 9:46 ` [PATCH net-next 09/19] ipvs: convert rr " Julian Anastasov
@ 2013-03-22 9:46 ` Julian Anastasov
2013-03-22 9:46 ` [PATCH net-next 11/19] ipvs: convert sh " Julian Anastasov
` (9 subsequent siblings)
19 siblings, 0 replies; 23+ messages in thread
From: Julian Anastasov @ 2013-03-22 9:46 UTC (permalink / raw)
To: Simon Horman; +Cc: lvs-devel, netdev
The schedule method now needs _rcu list-traversal
primitive for svc->destinations.
Signed-off-by: Julian Anastasov <ja@ssi.bg>
---
net/netfilter/ipvs/ip_vs_sed.c | 4 ++--
1 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/net/netfilter/ipvs/ip_vs_sed.c b/net/netfilter/ipvs/ip_vs_sed.c
index 89ead246..d011870 100644
--- a/net/netfilter/ipvs/ip_vs_sed.c
+++ b/net/netfilter/ipvs/ip_vs_sed.c
@@ -79,7 +79,7 @@ ip_vs_sed_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
* new connections.
*/
- list_for_each_entry(dest, &svc->destinations, n_list) {
+ list_for_each_entry_rcu(dest, &svc->destinations, n_list) {
if (!(dest->flags & IP_VS_DEST_F_OVERLOAD) &&
atomic_read(&dest->weight) > 0) {
least = dest;
@@ -94,7 +94,7 @@ ip_vs_sed_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
* Find the destination with the least load.
*/
nextstage:
- list_for_each_entry_continue(dest, &svc->destinations, n_list) {
+ list_for_each_entry_continue_rcu(dest, &svc->destinations, n_list) {
if (dest->flags & IP_VS_DEST_F_OVERLOAD)
continue;
doh = ip_vs_sed_dest_overhead(dest);
--
1.7.3.4
^ permalink raw reply related [flat|nested] 23+ messages in thread* [PATCH net-next 11/19] ipvs: convert sh scheduler to rcu
2013-03-22 9:46 [PATCH net-next 00/19] IPVS optimizations, part 2 Julian Anastasov
` (9 preceding siblings ...)
2013-03-22 9:46 ` [PATCH net-next 10/19] ipvs: convert sed " Julian Anastasov
@ 2013-03-22 9:46 ` Julian Anastasov
2013-03-22 9:46 ` [PATCH net-next 12/19] ipvs: convert wlc " Julian Anastasov
` (8 subsequent siblings)
19 siblings, 0 replies; 23+ messages in thread
From: Julian Anastasov @ 2013-03-22 9:46 UTC (permalink / raw)
To: Simon Horman; +Cc: lvs-devel, netdev
Use the 3 new methods to reassign dests.
Signed-off-by: Julian Anastasov <ja@ssi.bg>
---
net/netfilter/ipvs/ip_vs_sh.c | 81 +++++++++++++++++++++++------------------
1 files changed, 45 insertions(+), 36 deletions(-)
diff --git a/net/netfilter/ipvs/ip_vs_sh.c b/net/netfilter/ipvs/ip_vs_sh.c
index e331269..55e76d8 100644
--- a/net/netfilter/ipvs/ip_vs_sh.c
+++ b/net/netfilter/ipvs/ip_vs_sh.c
@@ -53,7 +53,7 @@
* IPVS SH bucket
*/
struct ip_vs_sh_bucket {
- struct ip_vs_dest *dest; /* real server (cache) */
+ struct ip_vs_dest __rcu *dest; /* real server (cache) */
};
/*
@@ -66,6 +66,10 @@ struct ip_vs_sh_bucket {
#define IP_VS_SH_TAB_SIZE (1 << IP_VS_SH_TAB_BITS)
#define IP_VS_SH_TAB_MASK (IP_VS_SH_TAB_SIZE - 1)
+struct ip_vs_sh_state {
+ struct ip_vs_sh_bucket buckets[IP_VS_SH_TAB_SIZE];
+ struct rcu_head rcu_head;
+};
/*
* Returns hash value for IPVS SH entry
@@ -87,10 +91,9 @@ static inline unsigned int ip_vs_sh_hashkey(int af, const union nf_inet_addr *ad
* Get ip_vs_dest associated with supplied parameters.
*/
static inline struct ip_vs_dest *
-ip_vs_sh_get(int af, struct ip_vs_sh_bucket *tbl,
- const union nf_inet_addr *addr)
+ip_vs_sh_get(int af, struct ip_vs_sh_state *s, const union nf_inet_addr *addr)
{
- return (tbl[ip_vs_sh_hashkey(af, addr)]).dest;
+ return rcu_dereference(s->buckets[ip_vs_sh_hashkey(af, addr)].dest);
}
@@ -98,27 +101,32 @@ ip_vs_sh_get(int af, struct ip_vs_sh_bucket *tbl,
* Assign all the hash buckets of the specified table with the service.
*/
static int
-ip_vs_sh_assign(struct ip_vs_sh_bucket *tbl, struct ip_vs_service *svc)
+ip_vs_sh_reassign(struct ip_vs_sh_state *s, struct ip_vs_service *svc)
{
int i;
struct ip_vs_sh_bucket *b;
struct list_head *p;
struct ip_vs_dest *dest;
int d_count;
+ bool empty;
- b = tbl;
+ b = &s->buckets[0];
p = &svc->destinations;
+ empty = list_empty(p);
d_count = 0;
for (i=0; i<IP_VS_SH_TAB_SIZE; i++) {
- if (list_empty(p)) {
- b->dest = NULL;
- } else {
+ dest = rcu_dereference_protected(b->dest, 1);
+ if (dest)
+ ip_vs_dest_put(dest);
+ if (empty)
+ RCU_INIT_POINTER(b->dest, NULL);
+ else {
if (p == &svc->destinations)
p = p->next;
dest = list_entry(p, struct ip_vs_dest, n_list);
- atomic_inc(&dest->refcnt);
- b->dest = dest;
+ ip_vs_dest_hold(dest);
+ RCU_INIT_POINTER(b->dest, dest);
IP_VS_DBG_BUF(6, "assigned i: %d dest: %s weight: %d\n",
i, IP_VS_DBG_ADDR(svc->af, &dest->addr),
@@ -140,16 +148,18 @@ ip_vs_sh_assign(struct ip_vs_sh_bucket *tbl, struct ip_vs_service *svc)
/*
* Flush all the hash buckets of the specified table.
*/
-static void ip_vs_sh_flush(struct ip_vs_sh_bucket *tbl)
+static void ip_vs_sh_flush(struct ip_vs_sh_state *s)
{
int i;
struct ip_vs_sh_bucket *b;
+ struct ip_vs_dest *dest;
- b = tbl;
+ b = &s->buckets[0];
for (i=0; i<IP_VS_SH_TAB_SIZE; i++) {
- if (b->dest) {
- atomic_dec(&b->dest->refcnt);
- b->dest = NULL;
+ dest = rcu_dereference_protected(b->dest, 1);
+ if (dest) {
+ ip_vs_dest_put(dest);
+ RCU_INIT_POINTER(b->dest, NULL);
}
b++;
}
@@ -158,21 +168,20 @@ static void ip_vs_sh_flush(struct ip_vs_sh_bucket *tbl)
static int ip_vs_sh_init_svc(struct ip_vs_service *svc)
{
- struct ip_vs_sh_bucket *tbl;
+ struct ip_vs_sh_state *s;
/* allocate the SH table for this service */
- tbl = kmalloc(sizeof(struct ip_vs_sh_bucket)*IP_VS_SH_TAB_SIZE,
- GFP_KERNEL);
- if (tbl == NULL)
+ s = kzalloc(sizeof(struct ip_vs_sh_state), GFP_KERNEL);
+ if (s == NULL)
return -ENOMEM;
- svc->sched_data = tbl;
+ svc->sched_data = s;
IP_VS_DBG(6, "SH hash table (memory=%Zdbytes) allocated for "
"current service\n",
sizeof(struct ip_vs_sh_bucket)*IP_VS_SH_TAB_SIZE);
- /* assign the hash buckets with the updated service */
- ip_vs_sh_assign(tbl, svc);
+ /* assign the hash buckets with current dests */
+ ip_vs_sh_reassign(s, svc);
return 0;
}
@@ -180,13 +189,13 @@ static int ip_vs_sh_init_svc(struct ip_vs_service *svc)
static int ip_vs_sh_done_svc(struct ip_vs_service *svc)
{
- struct ip_vs_sh_bucket *tbl = svc->sched_data;
+ struct ip_vs_sh_state *s = svc->sched_data;
/* got to clean up hash buckets here */
- ip_vs_sh_flush(tbl);
+ ip_vs_sh_flush(s);
/* release the table itself */
- kfree(svc->sched_data);
+ kfree_rcu(s, rcu_head);
IP_VS_DBG(6, "SH hash table (memory=%Zdbytes) released\n",
sizeof(struct ip_vs_sh_bucket)*IP_VS_SH_TAB_SIZE);
@@ -194,15 +203,13 @@ static int ip_vs_sh_done_svc(struct ip_vs_service *svc)
}
-static int ip_vs_sh_update_svc(struct ip_vs_service *svc)
+static int ip_vs_sh_dest_changed(struct ip_vs_service *svc,
+ struct ip_vs_dest *dest)
{
- struct ip_vs_sh_bucket *tbl = svc->sched_data;
-
- /* got to clean up hash buckets here */
- ip_vs_sh_flush(tbl);
+ struct ip_vs_sh_state *s = svc->sched_data;
/* assign the hash buckets with the updated service */
- ip_vs_sh_assign(tbl, svc);
+ ip_vs_sh_reassign(s, svc);
return 0;
}
@@ -225,15 +232,15 @@ static struct ip_vs_dest *
ip_vs_sh_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
{
struct ip_vs_dest *dest;
- struct ip_vs_sh_bucket *tbl;
+ struct ip_vs_sh_state *s;
struct ip_vs_iphdr iph;
ip_vs_fill_iph_addr_only(svc->af, skb, &iph);
IP_VS_DBG(6, "ip_vs_sh_schedule(): Scheduling...\n");
- tbl = (struct ip_vs_sh_bucket *)svc->sched_data;
- dest = ip_vs_sh_get(svc->af, tbl, &iph.saddr);
+ s = (struct ip_vs_sh_state *) svc->sched_data;
+ dest = ip_vs_sh_get(svc->af, s, &iph.saddr);
if (!dest
|| !(dest->flags & IP_VS_DEST_F_AVAILABLE)
|| atomic_read(&dest->weight) <= 0
@@ -262,7 +269,9 @@ static struct ip_vs_scheduler ip_vs_sh_scheduler =
.n_list = LIST_HEAD_INIT(ip_vs_sh_scheduler.n_list),
.init_service = ip_vs_sh_init_svc,
.done_service = ip_vs_sh_done_svc,
- .update_service = ip_vs_sh_update_svc,
+ .add_dest = ip_vs_sh_dest_changed,
+ .del_dest = ip_vs_sh_dest_changed,
+ .upd_dest = ip_vs_sh_dest_changed,
.schedule = ip_vs_sh_schedule,
};
--
1.7.3.4
^ permalink raw reply related [flat|nested] 23+ messages in thread* [PATCH net-next 12/19] ipvs: convert wlc scheduler to rcu
2013-03-22 9:46 [PATCH net-next 00/19] IPVS optimizations, part 2 Julian Anastasov
` (10 preceding siblings ...)
2013-03-22 9:46 ` [PATCH net-next 11/19] ipvs: convert sh " Julian Anastasov
@ 2013-03-22 9:46 ` Julian Anastasov
2013-03-22 9:46 ` [PATCH net-next 13/19] ipvs: convert wrr " Julian Anastasov
` (7 subsequent siblings)
19 siblings, 0 replies; 23+ messages in thread
From: Julian Anastasov @ 2013-03-22 9:46 UTC (permalink / raw)
To: Simon Horman; +Cc: lvs-devel, netdev
The schedule method now needs _rcu list-traversal
primitive for svc->destinations.
Signed-off-by: Julian Anastasov <ja@ssi.bg>
---
net/netfilter/ipvs/ip_vs_wlc.c | 4 ++--
1 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/net/netfilter/ipvs/ip_vs_wlc.c b/net/netfilter/ipvs/ip_vs_wlc.c
index bc1bfc4..dafae88 100644
--- a/net/netfilter/ipvs/ip_vs_wlc.c
+++ b/net/netfilter/ipvs/ip_vs_wlc.c
@@ -51,7 +51,7 @@ ip_vs_wlc_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
* new connections.
*/
- list_for_each_entry(dest, &svc->destinations, n_list) {
+ list_for_each_entry_rcu(dest, &svc->destinations, n_list) {
if (!(dest->flags & IP_VS_DEST_F_OVERLOAD) &&
atomic_read(&dest->weight) > 0) {
least = dest;
@@ -66,7 +66,7 @@ ip_vs_wlc_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
* Find the destination with the least load.
*/
nextstage:
- list_for_each_entry_continue(dest, &svc->destinations, n_list) {
+ list_for_each_entry_continue_rcu(dest, &svc->destinations, n_list) {
if (dest->flags & IP_VS_DEST_F_OVERLOAD)
continue;
doh = ip_vs_dest_conn_overhead(dest);
--
1.7.3.4
^ permalink raw reply related [flat|nested] 23+ messages in thread* [PATCH net-next 13/19] ipvs: convert wrr scheduler to rcu
2013-03-22 9:46 [PATCH net-next 00/19] IPVS optimizations, part 2 Julian Anastasov
` (11 preceding siblings ...)
2013-03-22 9:46 ` [PATCH net-next 12/19] ipvs: convert wlc " Julian Anastasov
@ 2013-03-22 9:46 ` Julian Anastasov
2013-03-22 9:46 ` [PATCH net-next 14/19] ipvs: reorganize dest trash Julian Anastasov
` (6 subsequent siblings)
19 siblings, 0 replies; 23+ messages in thread
From: Julian Anastasov @ 2013-03-22 9:46 UTC (permalink / raw)
To: Simon Horman; +Cc: lvs-devel, netdev
The schedule method now needs _rcu list-traversal
primitive for svc->destinations. As the weight for some
dest can be reduced during dest selection, change the
algorithm to check weights by using minimum weights in the
1 .. max_weight-(di-1) range, with the same step (di). By this
way we ensure that there will be always a weight >= 1 check
before claiming that all destinations are overloaded.
Signed-off-by: Julian Anastasov <ja@ssi.bg>
---
net/netfilter/ipvs/ip_vs_wrr.c | 167 ++++++++++++++++++++++++---------------
1 files changed, 103 insertions(+), 64 deletions(-)
diff --git a/net/netfilter/ipvs/ip_vs_wrr.c b/net/netfilter/ipvs/ip_vs_wrr.c
index 231be7d..98cb05e 100644
--- a/net/netfilter/ipvs/ip_vs_wrr.c
+++ b/net/netfilter/ipvs/ip_vs_wrr.c
@@ -29,14 +29,45 @@
#include <net/ip_vs.h>
+/* The WRR algorithm depends on some caclulations:
+ * - mw: maximum weight
+ * - di: weight step, greatest common divisor from all weights
+ * - cw: current required weight
+ * As result, all weights are in the [di..mw] range with a step=di.
+ *
+ * First, we start with cw = mw and select dests with weight >= cw.
+ * Then cw is reduced with di and all dests are checked again.
+ * Last pass should be with cw = di. We have mw/di passes in total:
+ *
+ * pass 1: cw = max weight
+ * pass 2: cw = max weight - di
+ * pass 3: cw = max weight - 2 * di
+ * ...
+ * last pass: cw = di
+ *
+ * Weights are supposed to be >= di but we run in parallel with
+ * weight changes, it is possible some dest weight to be reduced
+ * below di, bad if it is the only available dest.
+ *
+ * So, we modify how mw is calculated, now it is reduced with (di - 1),
+ * so that last cw is 1 to catch such dests with weight below di:
+ * pass 1: cw = max weight - (di - 1)
+ * pass 2: cw = max weight - di - (di - 1)
+ * pass 3: cw = max weight - 2 * di - (di - 1)
+ * ...
+ * last pass: cw = 1
+ *
+ */
+
/*
* current destination pointer for weighted round-robin scheduling
*/
struct ip_vs_wrr_mark {
- struct list_head *cl; /* current list head */
+ struct ip_vs_dest *cl; /* current dest or head */
int cw; /* current weight */
int mw; /* maximum weight */
int di; /* decreasing interval */
+ struct rcu_head rcu_head;
};
@@ -88,10 +119,10 @@ static int ip_vs_wrr_init_svc(struct ip_vs_service *svc)
if (mark == NULL)
return -ENOMEM;
- mark->cl = &svc->destinations;
- mark->cw = 0;
- mark->mw = ip_vs_wrr_max_weight(svc);
+ mark->cl = list_entry(&svc->destinations, struct ip_vs_dest, n_list);
mark->di = ip_vs_wrr_gcd_weight(svc);
+ mark->mw = ip_vs_wrr_max_weight(svc) - (mark->di - 1);
+ mark->cw = mark->mw;
svc->sched_data = mark;
return 0;
@@ -100,24 +131,31 @@ static int ip_vs_wrr_init_svc(struct ip_vs_service *svc)
static int ip_vs_wrr_done_svc(struct ip_vs_service *svc)
{
+ struct ip_vs_wrr_mark *mark = svc->sched_data;
+
/*
* Release the mark variable
*/
- kfree(svc->sched_data);
+ kfree_rcu(mark, rcu_head);
return 0;
}
-static int ip_vs_wrr_update_svc(struct ip_vs_service *svc)
+static int ip_vs_wrr_dest_changed(struct ip_vs_service *svc,
+ struct ip_vs_dest *dest)
{
struct ip_vs_wrr_mark *mark = svc->sched_data;
- mark->cl = &svc->destinations;
- mark->mw = ip_vs_wrr_max_weight(svc);
+ write_lock_bh(&svc->sched_lock);
+ mark->cl = list_entry(&svc->destinations, struct ip_vs_dest, n_list);
mark->di = ip_vs_wrr_gcd_weight(svc);
- if (mark->cw > mark->mw)
- mark->cw = 0;
+ mark->mw = ip_vs_wrr_max_weight(svc) - (mark->di - 1);
+ if (mark->cw > mark->mw || !mark->cw)
+ mark->cw = mark->mw;
+ else if (mark->di > 1)
+ mark->cw = (mark->cw / mark->di) * mark->di + 1;
+ write_unlock_bh(&svc->sched_lock);
return 0;
}
@@ -128,80 +166,79 @@ static int ip_vs_wrr_update_svc(struct ip_vs_service *svc)
static struct ip_vs_dest *
ip_vs_wrr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
{
- struct ip_vs_dest *dest;
+ struct ip_vs_dest *dest, *last, *stop = NULL;
struct ip_vs_wrr_mark *mark = svc->sched_data;
- struct list_head *p;
+ bool last_pass = false, restarted = false;
IP_VS_DBG(6, "%s(): Scheduling...\n", __func__);
- /*
- * This loop will always terminate, because mark->cw in (0, max_weight]
- * and at least one server has its weight equal to max_weight.
- */
write_lock(&svc->sched_lock);
- p = mark->cl;
+ dest = mark->cl;
+ /* No available dests? */
+ if (mark->mw == 0)
+ goto err_noavail;
+ last = dest;
+ /* Stop only after all dests were checked for weight >= 1 (last pass) */
while (1) {
- if (mark->cl == &svc->destinations) {
- /* it is at the head of the destination list */
-
- if (mark->cl == mark->cl->next) {
- /* no dest entry */
- ip_vs_scheduler_err(svc,
- "no destination available: "
- "no destinations present");
- dest = NULL;
- goto out;
- }
-
- mark->cl = svc->destinations.next;
- mark->cw -= mark->di;
- if (mark->cw <= 0) {
- mark->cw = mark->mw;
- /*
- * Still zero, which means no available servers.
- */
- if (mark->cw == 0) {
- mark->cl = &svc->destinations;
- ip_vs_scheduler_err(svc,
- "no destination available");
- dest = NULL;
- goto out;
- }
- }
- } else
- mark->cl = mark->cl->next;
-
- if (mark->cl != &svc->destinations) {
- /* not at the head of the list */
- dest = list_entry(mark->cl, struct ip_vs_dest, n_list);
+ list_for_each_entry_continue_rcu(dest,
+ &svc->destinations,
+ n_list) {
if (!(dest->flags & IP_VS_DEST_F_OVERLOAD) &&
- atomic_read(&dest->weight) >= mark->cw) {
- /* got it */
- break;
- }
+ atomic_read(&dest->weight) >= mark->cw)
+ goto found;
+ if (dest == stop)
+ goto err_over;
}
-
- if (mark->cl == p && mark->cw == mark->di) {
- /* back to the start, and no dest is found.
- It is only possible when all dests are OVERLOADED */
- dest = NULL;
- ip_vs_scheduler_err(svc,
- "no destination available: "
- "all destinations are overloaded");
- goto out;
+ mark->cw -= mark->di;
+ if (mark->cw <= 0) {
+ mark->cw = mark->mw;
+ /* Stop if we tried last pass from first dest:
+ * 1. last_pass: we started checks when cw > di but
+ * then all dests were checked for w >= 1
+ * 2. last was head: the first and only traversal
+ * was for weight >= 1, for all dests.
+ */
+ if (last_pass ||
+ &last->n_list == &svc->destinations)
+ goto err_over;
+ restarted = true;
+ }
+ last_pass = mark->cw <= mark->di;
+ if (last_pass && restarted &&
+ &last->n_list != &svc->destinations) {
+ /* First traversal was for w >= 1 but only
+ * for dests after 'last', now do the same
+ * for all dests up to 'last'.
+ */
+ stop = last;
}
}
+found:
IP_VS_DBG_BUF(6, "WRR: server %s:%u "
"activeconns %d refcnt %d weight %d\n",
IP_VS_DBG_ADDR(svc->af, &dest->addr), ntohs(dest->port),
atomic_read(&dest->activeconns),
atomic_read(&dest->refcnt),
atomic_read(&dest->weight));
+ mark->cl = dest;
out:
write_unlock(&svc->sched_lock);
return dest;
+
+err_noavail:
+ mark->cl = dest;
+ dest = NULL;
+ ip_vs_scheduler_err(svc, "no destination available");
+ goto out;
+
+err_over:
+ mark->cl = dest;
+ dest = NULL;
+ ip_vs_scheduler_err(svc, "no destination available: "
+ "all destinations are overloaded");
+ goto out;
}
@@ -212,7 +249,9 @@ static struct ip_vs_scheduler ip_vs_wrr_scheduler = {
.n_list = LIST_HEAD_INIT(ip_vs_wrr_scheduler.n_list),
.init_service = ip_vs_wrr_init_svc,
.done_service = ip_vs_wrr_done_svc,
- .update_service = ip_vs_wrr_update_svc,
+ .add_dest = ip_vs_wrr_dest_changed,
+ .del_dest = ip_vs_wrr_dest_changed,
+ .upd_dest = ip_vs_wrr_dest_changed,
.schedule = ip_vs_wrr_schedule,
};
--
1.7.3.4
^ permalink raw reply related [flat|nested] 23+ messages in thread* [PATCH net-next 14/19] ipvs: reorganize dest trash
2013-03-22 9:46 [PATCH net-next 00/19] IPVS optimizations, part 2 Julian Anastasov
` (12 preceding siblings ...)
2013-03-22 9:46 ` [PATCH net-next 13/19] ipvs: convert wrr " Julian Anastasov
@ 2013-03-22 9:46 ` Julian Anastasov
2013-03-22 9:46 ` [PATCH net-next 15/19] ipvs: do not expect result from done_service Julian Anastasov
` (5 subsequent siblings)
19 siblings, 0 replies; 23+ messages in thread
From: Julian Anastasov @ 2013-03-22 9:46 UTC (permalink / raw)
To: Simon Horman; +Cc: lvs-devel, netdev
All dests will go to trash, no exceptions.
But we have to use new list node t_list for this, due
to RCU changes in following patches. Dests will wait there
initial grace period and later all conns and schedulers to
put their reference. The dests don't get reference for
staying in dest trash as before.
As result, we do not load ip_vs_dest_put with
extra checks for last refcnt and the schedulers do not
need to play games with atomic_inc_not_zero while
selecting best destination.
Signed-off-by: Julian Anastasov <ja@ssi.bg>
---
include/net/ip_vs.h | 9 ++
net/netfilter/ipvs/ip_vs_ctl.c | 178 +++++++++++++++++++++++-----------------
2 files changed, 113 insertions(+), 74 deletions(-)
diff --git a/include/net/ip_vs.h b/include/net/ip_vs.h
index 2752c86..76747d9 100644
--- a/include/net/ip_vs.h
+++ b/include/net/ip_vs.h
@@ -749,6 +749,8 @@ struct ip_vs_dest_dst {
struct rcu_head rcu_head;
};
+/* In grace period after removing */
+#define IP_VS_DEST_STATE_REMOVING 0x01
/*
* The real server destination forwarding entry
* with ip address, port number, and so on.
@@ -766,6 +768,7 @@ struct ip_vs_dest {
atomic_t refcnt; /* reference counter */
struct ip_vs_stats stats; /* statistics */
+ unsigned long state; /* state flags */
/* connection counters and thresholds */
atomic_t activeconns; /* active connections */
@@ -785,6 +788,7 @@ struct ip_vs_dest {
union nf_inet_addr vaddr; /* virtual IP address */
__u32 vfwmark; /* firewall mark of service */
+ struct list_head t_list; /* in dest_trash */
struct rcu_head rcu_head;
unsigned int in_rs_table:1; /* we are in rs_table */
};
@@ -912,6 +916,9 @@ struct ipvs_master_sync_state {
struct netns_ipvs *ipvs;
};
+/* How much time to keep dests in trash */
+#define IP_VS_DEST_TRASH_PERIOD (120 * HZ)
+
/* IPVS in network namespace */
struct netns_ipvs {
int gen; /* Generation */
@@ -961,6 +968,8 @@ struct netns_ipvs {
/* Trash for destinations */
struct list_head dest_trash;
+ spinlock_t dest_trash_lock;
+ struct timer_list dest_trash_timer; /* expiration timer */
/* Service counters */
atomic_t ftpsvc_counter;
atomic_t nullsvc_counter;
diff --git a/net/netfilter/ipvs/ip_vs_ctl.c b/net/netfilter/ipvs/ip_vs_ctl.c
index 36d9cd8..a6c206f 100644
--- a/net/netfilter/ipvs/ip_vs_ctl.c
+++ b/net/netfilter/ipvs/ip_vs_ctl.c
@@ -71,7 +71,7 @@ int ip_vs_get_debug_level(void)
/* Protos */
-static void __ip_vs_del_service(struct ip_vs_service *svc);
+static void __ip_vs_del_service(struct ip_vs_service *svc, bool cleanup);
#ifdef CONFIG_IP_VS_IPV6
@@ -655,19 +655,25 @@ static struct ip_vs_dest *
ip_vs_trash_get_dest(struct ip_vs_service *svc, const union nf_inet_addr *daddr,
__be16 dport)
{
- struct ip_vs_dest *dest, *nxt;
+ struct ip_vs_dest *dest;
struct netns_ipvs *ipvs = net_ipvs(svc->net);
/*
* Find the destination in trash
*/
- list_for_each_entry_safe(dest, nxt, &ipvs->dest_trash, n_list) {
+ spin_lock_bh(&ipvs->dest_trash_lock);
+ list_for_each_entry(dest, &ipvs->dest_trash, t_list) {
IP_VS_DBG_BUF(3, "Destination %u/%s:%u still in trash, "
"dest->refcnt=%d\n",
dest->vfwmark,
IP_VS_DBG_ADDR(svc->af, &dest->addr),
ntohs(dest->port),
atomic_read(&dest->refcnt));
+ /* We can not reuse dest while in grace period
+ * because conns still can use dest->svc
+ */
+ if (test_bit(IP_VS_DEST_STATE_REMOVING, &dest->state))
+ continue;
if (dest->af == svc->af &&
ip_vs_addr_equal(svc->af, &dest->addr, daddr) &&
dest->port == dport &&
@@ -677,29 +683,27 @@ ip_vs_trash_get_dest(struct ip_vs_service *svc, const union nf_inet_addr *daddr,
(ip_vs_addr_equal(svc->af, &dest->vaddr, &svc->addr) &&
dest->vport == svc->port))) {
/* HIT */
- return dest;
- }
-
- /*
- * Try to purge the destination from trash if not referenced
- */
- if (atomic_read(&dest->refcnt) == 1) {
- IP_VS_DBG_BUF(3, "Removing destination %u/%s:%u "
- "from trash\n",
- dest->vfwmark,
- IP_VS_DBG_ADDR(svc->af, &dest->addr),
- ntohs(dest->port));
- list_del(&dest->n_list);
- __ip_vs_dst_cache_reset(dest);
- __ip_vs_unbind_svc(dest);
- free_percpu(dest->stats.cpustats);
- kfree_rcu(dest, rcu_head);
+ list_del(&dest->t_list);
+ ip_vs_dest_hold(dest);
+ goto out;
}
}
- return NULL;
+ dest = NULL;
+
+out:
+ spin_unlock_bh(&ipvs->dest_trash_lock);
+
+ return dest;
}
+static void ip_vs_dest_free(struct ip_vs_dest *dest)
+{
+ __ip_vs_dst_cache_reset(dest);
+ __ip_vs_unbind_svc(dest);
+ free_percpu(dest->stats.cpustats);
+ kfree(dest);
+}
/*
* Clean up all the destinations in the trash
@@ -708,19 +712,18 @@ ip_vs_trash_get_dest(struct ip_vs_service *svc, const union nf_inet_addr *daddr,
* When the ip_vs_control_clearup is activated by ipvs module exit,
* the service tables must have been flushed and all the connections
* are expired, and the refcnt of each destination in the trash must
- * be 1, so we simply release them here.
+ * be 0, so we simply release them here.
*/
static void ip_vs_trash_cleanup(struct net *net)
{
struct ip_vs_dest *dest, *nxt;
struct netns_ipvs *ipvs = net_ipvs(net);
- list_for_each_entry_safe(dest, nxt, &ipvs->dest_trash, n_list) {
- list_del(&dest->n_list);
- __ip_vs_dst_cache_reset(dest);
- __ip_vs_unbind_svc(dest);
- free_percpu(dest->stats.cpustats);
- kfree_rcu(dest, rcu_head);
+ del_timer_sync(&ipvs->dest_trash_timer);
+ /* No need to use dest_trash_lock */
+ list_for_each_entry_safe(dest, nxt, &ipvs->dest_trash, t_list) {
+ list_del(&dest->t_list);
+ ip_vs_dest_free(dest);
}
}
@@ -953,11 +956,6 @@ ip_vs_add_dest(struct ip_vs_service *svc, struct ip_vs_dest_user_kern *udest)
IP_VS_DBG_ADDR(svc->af, &dest->vaddr),
ntohs(dest->vport));
- /*
- * Get the destination from the trash
- */
- list_del(&dest->n_list);
-
__ip_vs_update_dest(svc, dest, udest, 1);
ret = 0;
} else {
@@ -1013,11 +1011,21 @@ ip_vs_edit_dest(struct ip_vs_service *svc, struct ip_vs_dest_user_kern *udest)
return 0;
}
+static void ip_vs_dest_wait_readers(struct rcu_head *head)
+{
+ struct ip_vs_dest *dest = container_of(head, struct ip_vs_dest,
+ rcu_head);
+
+ /* End of grace period after unlinking */
+ clear_bit(IP_VS_DEST_STATE_REMOVING, &dest->state);
+}
+
/*
* Delete a destination (must be already unlinked from the service)
*/
-static void __ip_vs_del_dest(struct net *net, struct ip_vs_dest *dest)
+static void __ip_vs_del_dest(struct net *net, struct ip_vs_dest *dest,
+ bool cleanup)
{
struct netns_ipvs *ipvs = net_ipvs(net);
@@ -1028,34 +1036,22 @@ static void __ip_vs_del_dest(struct net *net, struct ip_vs_dest *dest)
*/
ip_vs_rs_unhash(dest);
- /*
- * Decrease the refcnt of the dest, and free the dest
- * if nobody refers to it (refcnt=0). Otherwise, throw
- * the destination into the trash.
- */
- if (atomic_dec_and_test(&dest->refcnt)) {
- IP_VS_DBG_BUF(3, "Removing destination %u/%s:%u\n",
- dest->vfwmark,
- IP_VS_DBG_ADDR(dest->af, &dest->addr),
- ntohs(dest->port));
- __ip_vs_dst_cache_reset(dest);
- /* simply decrease svc->refcnt here, let the caller check
- and release the service if nobody refers to it.
- Only user context can release destination and service,
- and only one user context can update virtual service at a
- time, so the operation here is OK */
- atomic_dec(&dest->svc->refcnt);
- free_percpu(dest->stats.cpustats);
- kfree_rcu(dest, rcu_head);
- } else {
- IP_VS_DBG_BUF(3, "Moving dest %s:%u into trash, "
- "dest->refcnt=%d\n",
- IP_VS_DBG_ADDR(dest->af, &dest->addr),
- ntohs(dest->port),
- atomic_read(&dest->refcnt));
- list_add(&dest->n_list, &ipvs->dest_trash);
- ip_vs_dest_hold(dest);
+ if (!cleanup) {
+ set_bit(IP_VS_DEST_STATE_REMOVING, &dest->state);
+ call_rcu(&dest->rcu_head, ip_vs_dest_wait_readers);
}
+
+ spin_lock_bh(&ipvs->dest_trash_lock);
+ IP_VS_DBG_BUF(3, "Moving dest %s:%u into trash, dest->refcnt=%d\n",
+ IP_VS_DBG_ADDR(dest->af, &dest->addr), ntohs(dest->port),
+ atomic_read(&dest->refcnt));
+ if (list_empty(&ipvs->dest_trash) && !cleanup)
+ mod_timer(&ipvs->dest_trash_timer,
+ jiffies + IP_VS_DEST_TRASH_PERIOD);
+ /* dest lives in trash without reference */
+ list_add(&dest->t_list, &ipvs->dest_trash);
+ spin_unlock_bh(&ipvs->dest_trash_lock);
+ ip_vs_dest_put(dest);
}
@@ -1120,13 +1116,38 @@ ip_vs_del_dest(struct ip_vs_service *svc, struct ip_vs_dest_user_kern *udest)
/*
* Delete the destination
*/
- __ip_vs_del_dest(svc->net, dest);
+ __ip_vs_del_dest(svc->net, dest, false);
LeaveFunction(2);
return 0;
}
+static void ip_vs_dest_trash_expire(unsigned long data)
+{
+ struct net *net = (struct net *) data;
+ struct netns_ipvs *ipvs = net_ipvs(net);
+ struct ip_vs_dest *dest, *next;
+
+ spin_lock(&ipvs->dest_trash_lock);
+ list_for_each_entry_safe(dest, next, &ipvs->dest_trash, t_list) {
+ /* Skip if dest is in grace period */
+ if (test_bit(IP_VS_DEST_STATE_REMOVING, &dest->state))
+ continue;
+ if (atomic_read(&dest->refcnt) > 0)
+ continue;
+ IP_VS_DBG_BUF(3, "Removing destination %u/%s:%u from trash\n",
+ dest->vfwmark,
+ IP_VS_DBG_ADDR(dest->svc->af, &dest->addr),
+ ntohs(dest->port));
+ list_del(&dest->t_list);
+ ip_vs_dest_free(dest);
+ }
+ if (!list_empty(&ipvs->dest_trash))
+ mod_timer(&ipvs->dest_trash_timer,
+ jiffies + IP_VS_DEST_TRASH_PERIOD);
+ spin_unlock(&ipvs->dest_trash_lock);
+}
/*
* Add a service into the service hash table
@@ -1356,7 +1377,7 @@ out:
* - The service must be unlinked, unlocked and not referenced!
* - We are called under _bh lock
*/
-static void __ip_vs_del_service(struct ip_vs_service *svc)
+static void __ip_vs_del_service(struct ip_vs_service *svc, bool cleanup)
{
struct ip_vs_dest *dest, *nxt;
struct ip_vs_scheduler *old_sched;
@@ -1392,7 +1413,7 @@ static void __ip_vs_del_service(struct ip_vs_service *svc)
*/
list_for_each_entry_safe(dest, nxt, &svc->destinations, n_list) {
__ip_vs_unlink_dest(svc, dest, 0);
- __ip_vs_del_dest(svc->net, dest);
+ __ip_vs_del_dest(svc->net, dest, cleanup);
}
/*
@@ -1422,7 +1443,7 @@ static void __ip_vs_del_service(struct ip_vs_service *svc)
/*
* Unlink a service from list and try to delete it if its refcnt reached 0
*/
-static void ip_vs_unlink_service(struct ip_vs_service *svc)
+static void ip_vs_unlink_service(struct ip_vs_service *svc, bool cleanup)
{
/*
* Unhash it from the service table
@@ -1436,7 +1457,7 @@ static void ip_vs_unlink_service(struct ip_vs_service *svc)
*/
IP_VS_WAIT_WHILE(atomic_read(&svc->usecnt) > 0);
- __ip_vs_del_service(svc);
+ __ip_vs_del_service(svc, cleanup);
write_unlock_bh(&__ip_vs_svc_lock);
}
@@ -1448,7 +1469,7 @@ static int ip_vs_del_service(struct ip_vs_service *svc)
{
if (svc == NULL)
return -EEXIST;
- ip_vs_unlink_service(svc);
+ ip_vs_unlink_service(svc, false);
return 0;
}
@@ -1457,7 +1478,7 @@ static int ip_vs_del_service(struct ip_vs_service *svc)
/*
* Flush all the virtual services
*/
-static int ip_vs_flush(struct net *net)
+static int ip_vs_flush(struct net *net, bool cleanup)
{
int idx;
struct ip_vs_service *svc, *nxt;
@@ -1469,7 +1490,7 @@ static int ip_vs_flush(struct net *net)
list_for_each_entry_safe(svc, nxt, &ip_vs_svc_table[idx],
s_list) {
if (net_eq(svc->net, net))
- ip_vs_unlink_service(svc);
+ ip_vs_unlink_service(svc, cleanup);
}
}
@@ -1480,7 +1501,7 @@ static int ip_vs_flush(struct net *net)
list_for_each_entry_safe(svc, nxt,
&ip_vs_svc_fwm_table[idx], f_list) {
if (net_eq(svc->net, net))
- ip_vs_unlink_service(svc);
+ ip_vs_unlink_service(svc, cleanup);
}
}
@@ -1496,7 +1517,7 @@ void ip_vs_service_net_cleanup(struct net *net)
EnterFunction(2);
/* Check for "full" addressed entries */
mutex_lock(&__ip_vs_mutex);
- ip_vs_flush(net);
+ ip_vs_flush(net, true);
mutex_unlock(&__ip_vs_mutex);
LeaveFunction(2);
}
@@ -1556,9 +1577,11 @@ static int ip_vs_dst_event(struct notifier_block *this, unsigned long event,
}
}
- list_for_each_entry(dest, &ipvs->dest_trash, n_list) {
+ spin_lock_bh(&ipvs->dest_trash_lock);
+ list_for_each_entry(dest, &ipvs->dest_trash, t_list) {
ip_vs_forget_dev(dest, dev);
}
+ spin_unlock_bh(&ipvs->dest_trash_lock);
mutex_unlock(&__ip_vs_mutex);
LeaveFunction(2);
return NOTIFY_DONE;
@@ -2392,7 +2415,7 @@ do_ip_vs_set_ctl(struct sock *sk, int cmd, void __user *user, unsigned int len)
if (cmd == IP_VS_SO_SET_FLUSH) {
/* Flush the virtual service */
- ret = ip_vs_flush(net);
+ ret = ip_vs_flush(net, false);
goto out_unlock;
} else if (cmd == IP_VS_SO_SET_TIMEOUT) {
/* Set timeout values for (tcp tcpfin udp) */
@@ -3401,7 +3424,7 @@ static int ip_vs_genl_set_cmd(struct sk_buff *skb, struct genl_info *info)
mutex_lock(&__ip_vs_mutex);
if (cmd == IPVS_CMD_FLUSH) {
- ret = ip_vs_flush(net);
+ ret = ip_vs_flush(net, false);
goto out;
} else if (cmd == IPVS_CMD_SET_CONFIG) {
ret = ip_vs_genl_set_config(net, info->attrs);
@@ -3798,6 +3821,9 @@ int __net_init ip_vs_control_net_init(struct net *net)
INIT_HLIST_HEAD(&ipvs->rs_table[idx]);
INIT_LIST_HEAD(&ipvs->dest_trash);
+ spin_lock_init(&ipvs->dest_trash_lock);
+ setup_timer(&ipvs->dest_trash_timer, ip_vs_dest_trash_expire,
+ (unsigned long) net);
atomic_set(&ipvs->ftpsvc_counter, 0);
atomic_set(&ipvs->nullsvc_counter, 0);
@@ -3827,6 +3853,10 @@ void __net_exit ip_vs_control_net_cleanup(struct net *net)
{
struct netns_ipvs *ipvs = net_ipvs(net);
+ /* Some dest can be in grace period even before cleanup, we have to
+ * defer ip_vs_trash_cleanup until ip_vs_dest_wait_readers is called.
+ */
+ rcu_barrier();
ip_vs_trash_cleanup(net);
ip_vs_stop_estimator(net, &ipvs->tot_stats);
ip_vs_control_net_cleanup_sysctl(net);
--
1.7.3.4
^ permalink raw reply related [flat|nested] 23+ messages in thread* [PATCH net-next 15/19] ipvs: do not expect result from done_service
2013-03-22 9:46 [PATCH net-next 00/19] IPVS optimizations, part 2 Julian Anastasov
` (13 preceding siblings ...)
2013-03-22 9:46 ` [PATCH net-next 14/19] ipvs: reorganize dest trash Julian Anastasov
@ 2013-03-22 9:46 ` Julian Anastasov
2013-03-22 9:46 ` [PATCH net-next 16/19] ipvs: convert sched_lock to spin lock Julian Anastasov
` (4 subsequent siblings)
19 siblings, 0 replies; 23+ messages in thread
From: Julian Anastasov @ 2013-03-22 9:46 UTC (permalink / raw)
To: Simon Horman; +Cc: lvs-devel, netdev
This method releases the scheduler state,
it can not fail. Such change will help to properly
replace the scheduler in following patch.
Signed-off-by: Julian Anastasov <ja@ssi.bg>
---
include/net/ip_vs.h | 4 ++--
net/netfilter/ipvs/ip_vs_ctl.c | 5 +----
net/netfilter/ipvs/ip_vs_dh.c | 4 +---
net/netfilter/ipvs/ip_vs_lblc.c | 4 +---
net/netfilter/ipvs/ip_vs_lblcr.c | 4 +---
net/netfilter/ipvs/ip_vs_sched.c | 13 ++++---------
net/netfilter/ipvs/ip_vs_sh.c | 4 +---
net/netfilter/ipvs/ip_vs_wrr.c | 4 +---
8 files changed, 12 insertions(+), 30 deletions(-)
diff --git a/include/net/ip_vs.h b/include/net/ip_vs.h
index 76747d9..8597a71 100644
--- a/include/net/ip_vs.h
+++ b/include/net/ip_vs.h
@@ -806,7 +806,7 @@ struct ip_vs_scheduler {
/* scheduler initializing service */
int (*init_service)(struct ip_vs_service *svc);
/* scheduling service finish */
- int (*done_service)(struct ip_vs_service *svc);
+ void (*done_service)(struct ip_vs_service *svc);
/* scheduler updating service */
int (*update_service)(struct ip_vs_service *svc);
/* dest is linked */
@@ -1392,7 +1392,7 @@ extern int register_ip_vs_scheduler(struct ip_vs_scheduler *scheduler);
extern int unregister_ip_vs_scheduler(struct ip_vs_scheduler *scheduler);
extern int ip_vs_bind_scheduler(struct ip_vs_service *svc,
struct ip_vs_scheduler *scheduler);
-extern int ip_vs_unbind_scheduler(struct ip_vs_service *svc);
+extern void ip_vs_unbind_scheduler(struct ip_vs_service *svc);
extern struct ip_vs_scheduler *ip_vs_scheduler_get(const char *sched_name);
extern void ip_vs_scheduler_put(struct ip_vs_scheduler *scheduler);
extern struct ip_vs_conn *
diff --git a/net/netfilter/ipvs/ip_vs_ctl.c b/net/netfilter/ipvs/ip_vs_ctl.c
index a6c206f..e010deb 100644
--- a/net/netfilter/ipvs/ip_vs_ctl.c
+++ b/net/netfilter/ipvs/ip_vs_ctl.c
@@ -1332,10 +1332,7 @@ ip_vs_edit_service(struct ip_vs_service *svc, struct ip_vs_service_user_kern *u)
/*
* Unbind the old scheduler
*/
- if ((ret = ip_vs_unbind_scheduler(svc))) {
- old_sched = sched;
- goto out_unlock;
- }
+ ip_vs_unbind_scheduler(svc);
/*
* Bind the new scheduler
diff --git a/net/netfilter/ipvs/ip_vs_dh.c b/net/netfilter/ipvs/ip_vs_dh.c
index ebe80f4..89c2723 100644
--- a/net/netfilter/ipvs/ip_vs_dh.c
+++ b/net/netfilter/ipvs/ip_vs_dh.c
@@ -174,7 +174,7 @@ static int ip_vs_dh_init_svc(struct ip_vs_service *svc)
}
-static int ip_vs_dh_done_svc(struct ip_vs_service *svc)
+static void ip_vs_dh_done_svc(struct ip_vs_service *svc)
{
struct ip_vs_dh_state *s = svc->sched_data;
@@ -185,8 +185,6 @@ static int ip_vs_dh_done_svc(struct ip_vs_service *svc)
kfree_rcu(s, rcu_head);
IP_VS_DBG(6, "DH hash table (memory=%Zdbytes) released\n",
sizeof(struct ip_vs_dh_bucket)*IP_VS_DH_TAB_SIZE);
-
- return 0;
}
diff --git a/net/netfilter/ipvs/ip_vs_lblc.c b/net/netfilter/ipvs/ip_vs_lblc.c
index b873e17..c7ff978 100644
--- a/net/netfilter/ipvs/ip_vs_lblc.c
+++ b/net/netfilter/ipvs/ip_vs_lblc.c
@@ -388,7 +388,7 @@ static int ip_vs_lblc_init_svc(struct ip_vs_service *svc)
}
-static int ip_vs_lblc_done_svc(struct ip_vs_service *svc)
+static void ip_vs_lblc_done_svc(struct ip_vs_service *svc)
{
struct ip_vs_lblc_table *tbl = svc->sched_data;
@@ -402,8 +402,6 @@ static int ip_vs_lblc_done_svc(struct ip_vs_service *svc)
kfree_rcu(tbl, rcu_head);
IP_VS_DBG(6, "LBLC hash table (memory=%Zdbytes) released\n",
sizeof(*tbl));
-
- return 0;
}
diff --git a/net/netfilter/ipvs/ip_vs_lblcr.c b/net/netfilter/ipvs/ip_vs_lblcr.c
index c22f173..6049b85 100644
--- a/net/netfilter/ipvs/ip_vs_lblcr.c
+++ b/net/netfilter/ipvs/ip_vs_lblcr.c
@@ -555,7 +555,7 @@ static int ip_vs_lblcr_init_svc(struct ip_vs_service *svc)
}
-static int ip_vs_lblcr_done_svc(struct ip_vs_service *svc)
+static void ip_vs_lblcr_done_svc(struct ip_vs_service *svc)
{
struct ip_vs_lblcr_table *tbl = svc->sched_data;
@@ -569,8 +569,6 @@ static int ip_vs_lblcr_done_svc(struct ip_vs_service *svc)
kfree_rcu(tbl, rcu_head);
IP_VS_DBG(6, "LBLCR hash table (memory=%Zdbytes) released\n",
sizeof(*tbl));
-
- return 0;
}
diff --git a/net/netfilter/ipvs/ip_vs_sched.c b/net/netfilter/ipvs/ip_vs_sched.c
index 7f11d3d..1b715d0 100644
--- a/net/netfilter/ipvs/ip_vs_sched.c
+++ b/net/netfilter/ipvs/ip_vs_sched.c
@@ -64,22 +64,17 @@ int ip_vs_bind_scheduler(struct ip_vs_service *svc,
/*
* Unbind a service with its scheduler
*/
-int ip_vs_unbind_scheduler(struct ip_vs_service *svc)
+void ip_vs_unbind_scheduler(struct ip_vs_service *svc)
{
struct ip_vs_scheduler *sched = svc->scheduler;
if (!sched)
- return 0;
+ return;
- if (sched->done_service) {
- if (sched->done_service(svc) != 0) {
- pr_err("%s(): done error\n", __func__);
- return -EINVAL;
- }
- }
+ if (sched->done_service)
+ sched->done_service(svc);
svc->scheduler = NULL;
- return 0;
}
diff --git a/net/netfilter/ipvs/ip_vs_sh.c b/net/netfilter/ipvs/ip_vs_sh.c
index 55e76d8..81c1a10 100644
--- a/net/netfilter/ipvs/ip_vs_sh.c
+++ b/net/netfilter/ipvs/ip_vs_sh.c
@@ -187,7 +187,7 @@ static int ip_vs_sh_init_svc(struct ip_vs_service *svc)
}
-static int ip_vs_sh_done_svc(struct ip_vs_service *svc)
+static void ip_vs_sh_done_svc(struct ip_vs_service *svc)
{
struct ip_vs_sh_state *s = svc->sched_data;
@@ -198,8 +198,6 @@ static int ip_vs_sh_done_svc(struct ip_vs_service *svc)
kfree_rcu(s, rcu_head);
IP_VS_DBG(6, "SH hash table (memory=%Zdbytes) released\n",
sizeof(struct ip_vs_sh_bucket)*IP_VS_SH_TAB_SIZE);
-
- return 0;
}
diff --git a/net/netfilter/ipvs/ip_vs_wrr.c b/net/netfilter/ipvs/ip_vs_wrr.c
index 98cb05e..a74fd9b 100644
--- a/net/netfilter/ipvs/ip_vs_wrr.c
+++ b/net/netfilter/ipvs/ip_vs_wrr.c
@@ -129,7 +129,7 @@ static int ip_vs_wrr_init_svc(struct ip_vs_service *svc)
}
-static int ip_vs_wrr_done_svc(struct ip_vs_service *svc)
+static void ip_vs_wrr_done_svc(struct ip_vs_service *svc)
{
struct ip_vs_wrr_mark *mark = svc->sched_data;
@@ -137,8 +137,6 @@ static int ip_vs_wrr_done_svc(struct ip_vs_service *svc)
* Release the mark variable
*/
kfree_rcu(mark, rcu_head);
-
- return 0;
}
--
1.7.3.4
^ permalink raw reply related [flat|nested] 23+ messages in thread* [PATCH net-next 16/19] ipvs: convert sched_lock to spin lock
2013-03-22 9:46 [PATCH net-next 00/19] IPVS optimizations, part 2 Julian Anastasov
` (14 preceding siblings ...)
2013-03-22 9:46 ` [PATCH net-next 15/19] ipvs: do not expect result from done_service Julian Anastasov
@ 2013-03-22 9:46 ` Julian Anastasov
2013-03-22 9:46 ` [PATCH net-next 17/19] ipvs: convert dests to rcu Julian Anastasov
` (3 subsequent siblings)
19 siblings, 0 replies; 23+ messages in thread
From: Julian Anastasov @ 2013-03-22 9:46 UTC (permalink / raw)
To: Simon Horman; +Cc: lvs-devel, netdev
As all read_locks are gone spin lock is preferred.
Signed-off-by: Julian Anastasov <ja@ssi.bg>
---
include/net/ip_vs.h | 2 +-
net/netfilter/ipvs/ip_vs_ctl.c | 2 +-
net/netfilter/ipvs/ip_vs_lblc.c | 18 +++++++++---------
net/netfilter/ipvs/ip_vs_lblcr.c | 26 +++++++++++++-------------
net/netfilter/ipvs/ip_vs_rr.c | 10 +++++-----
net/netfilter/ipvs/ip_vs_wrr.c | 8 ++++----
6 files changed, 33 insertions(+), 33 deletions(-)
diff --git a/include/net/ip_vs.h b/include/net/ip_vs.h
index 8597a71..034ce38 100644
--- a/include/net/ip_vs.h
+++ b/include/net/ip_vs.h
@@ -734,7 +734,7 @@ struct ip_vs_service {
/* for scheduling */
struct ip_vs_scheduler *scheduler; /* bound scheduler object */
- rwlock_t sched_lock; /* lock sched_data */
+ spinlock_t sched_lock; /* lock sched_data */
void *sched_data; /* scheduler application data */
/* alternate persistence engine */
diff --git a/net/netfilter/ipvs/ip_vs_ctl.c b/net/netfilter/ipvs/ip_vs_ctl.c
index e010deb..ada61a8 100644
--- a/net/netfilter/ipvs/ip_vs_ctl.c
+++ b/net/netfilter/ipvs/ip_vs_ctl.c
@@ -1217,7 +1217,7 @@ ip_vs_add_service(struct net *net, struct ip_vs_service_user_kern *u,
svc->net = net;
INIT_LIST_HEAD(&svc->destinations);
- rwlock_init(&svc->sched_lock);
+ spin_lock_init(&svc->sched_lock);
spin_lock_init(&svc->stats.lock);
/* Bind the scheduler */
diff --git a/net/netfilter/ipvs/ip_vs_lblc.c b/net/netfilter/ipvs/ip_vs_lblc.c
index c7ff978..ffef8a1 100644
--- a/net/netfilter/ipvs/ip_vs_lblc.c
+++ b/net/netfilter/ipvs/ip_vs_lblc.c
@@ -194,7 +194,7 @@ ip_vs_lblc_get(int af, struct ip_vs_lblc_table *tbl,
/*
* Create or update an ip_vs_lblc_entry, which is a mapping of a destination IP
- * address to a server. Called under write lock.
+ * address to a server. Called under spin lock.
*/
static inline struct ip_vs_lblc_entry *
ip_vs_lblc_new(struct ip_vs_lblc_table *tbl, const union nf_inet_addr *daddr,
@@ -242,7 +242,7 @@ static void ip_vs_lblc_flush(struct ip_vs_service *svc)
struct hlist_node *next;
int i;
- write_lock_bh(&svc->sched_lock);
+ spin_lock_bh(&svc->sched_lock);
tbl->dead = 1;
for (i=0; i<IP_VS_LBLC_TAB_SIZE; i++) {
hlist_for_each_entry_safe(en, next, &tbl->bucket[i], list) {
@@ -250,7 +250,7 @@ static void ip_vs_lblc_flush(struct ip_vs_service *svc)
atomic_dec(&tbl->entries);
}
}
- write_unlock_bh(&svc->sched_lock);
+ spin_unlock_bh(&svc->sched_lock);
}
static int sysctl_lblc_expiration(struct ip_vs_service *svc)
@@ -274,7 +274,7 @@ static inline void ip_vs_lblc_full_check(struct ip_vs_service *svc)
for (i=0, j=tbl->rover; i<IP_VS_LBLC_TAB_SIZE; i++) {
j = (j + 1) & IP_VS_LBLC_TAB_MASK;
- write_lock(&svc->sched_lock);
+ spin_lock(&svc->sched_lock);
hlist_for_each_entry_safe(en, next, &tbl->bucket[j], list) {
if (time_before(now,
en->lastuse +
@@ -284,7 +284,7 @@ static inline void ip_vs_lblc_full_check(struct ip_vs_service *svc)
ip_vs_lblc_free(en);
atomic_dec(&tbl->entries);
}
- write_unlock(&svc->sched_lock);
+ spin_unlock(&svc->sched_lock);
}
tbl->rover = j;
}
@@ -330,7 +330,7 @@ static void ip_vs_lblc_check_expire(unsigned long data)
for (i=0, j=tbl->rover; i<IP_VS_LBLC_TAB_SIZE; i++) {
j = (j + 1) & IP_VS_LBLC_TAB_MASK;
- write_lock(&svc->sched_lock);
+ spin_lock(&svc->sched_lock);
hlist_for_each_entry_safe(en, next, &tbl->bucket[j], list) {
if (time_before(now, en->lastuse + ENTRY_TIMEOUT))
continue;
@@ -339,7 +339,7 @@ static void ip_vs_lblc_check_expire(unsigned long data)
atomic_dec(&tbl->entries);
goal--;
}
- write_unlock(&svc->sched_lock);
+ spin_unlock(&svc->sched_lock);
if (goal <= 0)
break;
}
@@ -527,10 +527,10 @@ ip_vs_lblc_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
}
/* If we fail to create a cache entry, we'll just use the valid dest */
- write_lock(&svc->sched_lock);
+ spin_lock(&svc->sched_lock);
if (!tbl->dead)
ip_vs_lblc_new(tbl, &iph.daddr, dest);
- write_unlock(&svc->sched_lock);
+ spin_unlock(&svc->sched_lock);
out:
IP_VS_DBG_BUF(6, "LBLC: destination IP address %s --> server %s:%d\n",
diff --git a/net/netfilter/ipvs/ip_vs_lblcr.c b/net/netfilter/ipvs/ip_vs_lblcr.c
index 6049b85..cdfe6a9 100644
--- a/net/netfilter/ipvs/ip_vs_lblcr.c
+++ b/net/netfilter/ipvs/ip_vs_lblcr.c
@@ -368,7 +368,7 @@ ip_vs_lblcr_get(int af, struct ip_vs_lblcr_table *tbl,
/*
* Create or update an ip_vs_lblcr_entry, which is a mapping of a destination
- * IP address to a server. Called under write lock.
+ * IP address to a server. Called under spin lock.
*/
static inline struct ip_vs_lblcr_entry *
ip_vs_lblcr_new(struct ip_vs_lblcr_table *tbl, const union nf_inet_addr *daddr,
@@ -412,14 +412,14 @@ static void ip_vs_lblcr_flush(struct ip_vs_service *svc)
struct ip_vs_lblcr_entry *en;
struct hlist_node *next;
- write_lock_bh(&svc->sched_lock);
+ spin_lock_bh(&svc->sched_lock);
tbl->dead = 1;
for (i=0; i<IP_VS_LBLCR_TAB_SIZE; i++) {
hlist_for_each_entry_safe(en, next, &tbl->bucket[i], list) {
ip_vs_lblcr_free(en);
}
}
- write_unlock_bh(&svc->sched_lock);
+ spin_unlock_bh(&svc->sched_lock);
}
static int sysctl_lblcr_expiration(struct ip_vs_service *svc)
@@ -443,7 +443,7 @@ static inline void ip_vs_lblcr_full_check(struct ip_vs_service *svc)
for (i=0, j=tbl->rover; i<IP_VS_LBLCR_TAB_SIZE; i++) {
j = (j + 1) & IP_VS_LBLCR_TAB_MASK;
- write_lock(&svc->sched_lock);
+ spin_lock(&svc->sched_lock);
hlist_for_each_entry_safe(en, next, &tbl->bucket[j], list) {
if (time_after(en->lastuse +
sysctl_lblcr_expiration(svc), now))
@@ -452,7 +452,7 @@ static inline void ip_vs_lblcr_full_check(struct ip_vs_service *svc)
ip_vs_lblcr_free(en);
atomic_dec(&tbl->entries);
}
- write_unlock(&svc->sched_lock);
+ spin_unlock(&svc->sched_lock);
}
tbl->rover = j;
}
@@ -498,7 +498,7 @@ static void ip_vs_lblcr_check_expire(unsigned long data)
for (i=0, j=tbl->rover; i<IP_VS_LBLCR_TAB_SIZE; i++) {
j = (j + 1) & IP_VS_LBLCR_TAB_MASK;
- write_lock(&svc->sched_lock);
+ spin_lock(&svc->sched_lock);
hlist_for_each_entry_safe(en, next, &tbl->bucket[j], list) {
if (time_before(now, en->lastuse+ENTRY_TIMEOUT))
continue;
@@ -507,7 +507,7 @@ static void ip_vs_lblcr_check_expire(unsigned long data)
atomic_dec(&tbl->entries);
goal--;
}
- write_unlock(&svc->sched_lock);
+ spin_unlock(&svc->sched_lock);
if (goal <= 0)
break;
}
@@ -678,7 +678,7 @@ ip_vs_lblcr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
if (atomic_read(&en->set.size) > 1 &&
time_after(jiffies, en->set.lastmod +
sysctl_lblcr_expiration(svc))) {
- write_lock(&svc->sched_lock);
+ spin_lock(&svc->sched_lock);
if (atomic_read(&en->set.size) > 1) {
struct ip_vs_dest *m;
@@ -686,7 +686,7 @@ ip_vs_lblcr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
if (m)
ip_vs_dest_set_erase(&en->set, m);
}
- write_unlock(&svc->sched_lock);
+ spin_unlock(&svc->sched_lock);
}
/* If the destination is not overloaded, use it */
@@ -701,10 +701,10 @@ ip_vs_lblcr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
}
/* Update our cache entry */
- write_lock(&svc->sched_lock);
+ spin_lock(&svc->sched_lock);
if (!tbl->dead)
ip_vs_dest_set_insert(&en->set, dest, true);
- write_unlock(&svc->sched_lock);
+ spin_unlock(&svc->sched_lock);
goto out;
}
@@ -716,10 +716,10 @@ ip_vs_lblcr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
}
/* If we fail to create a cache entry, we'll just use the valid dest */
- write_lock(&svc->sched_lock);
+ spin_lock(&svc->sched_lock);
if (!tbl->dead)
ip_vs_lblcr_new(tbl, &iph.daddr, dest);
- write_unlock(&svc->sched_lock);
+ spin_unlock(&svc->sched_lock);
out:
IP_VS_DBG_BUF(6, "LBLCR: destination IP address %s --> server %s:%d\n",
diff --git a/net/netfilter/ipvs/ip_vs_rr.c b/net/netfilter/ipvs/ip_vs_rr.c
index 3942890..aa4601f 100644
--- a/net/netfilter/ipvs/ip_vs_rr.c
+++ b/net/netfilter/ipvs/ip_vs_rr.c
@@ -39,14 +39,14 @@ static int ip_vs_rr_del_dest(struct ip_vs_service *svc, struct ip_vs_dest *dest)
{
struct list_head *p;
- write_lock_bh(&svc->sched_lock);
+ spin_lock_bh(&svc->sched_lock);
p = (struct list_head *) svc->sched_data;
/* dest is already unlinked, so p->prev is not valid but
* p->next is valid, use it to reach previous entry.
*/
if (p == &dest->n_list)
svc->sched_data = p->next->prev;
- write_unlock_bh(&svc->sched_lock);
+ spin_unlock_bh(&svc->sched_lock);
return 0;
}
@@ -63,7 +63,7 @@ ip_vs_rr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
IP_VS_DBG(6, "%s(): Scheduling...\n", __func__);
- write_lock(&svc->sched_lock);
+ spin_lock(&svc->sched_lock);
p = (struct list_head *) svc->sched_data;
last = dest = list_entry(p, struct ip_vs_dest, n_list);
@@ -85,13 +85,13 @@ ip_vs_rr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
} while (pass < 2 && p != &svc->destinations);
stop:
- write_unlock(&svc->sched_lock);
+ spin_unlock(&svc->sched_lock);
ip_vs_scheduler_err(svc, "no destination available");
return NULL;
out:
svc->sched_data = &dest->n_list;
- write_unlock(&svc->sched_lock);
+ spin_unlock(&svc->sched_lock);
IP_VS_DBG_BUF(6, "RR: server %s:%u "
"activeconns %d refcnt %d weight %d\n",
IP_VS_DBG_ADDR(svc->af, &dest->addr), ntohs(dest->port),
diff --git a/net/netfilter/ipvs/ip_vs_wrr.c b/net/netfilter/ipvs/ip_vs_wrr.c
index a74fd9b..b173ef9 100644
--- a/net/netfilter/ipvs/ip_vs_wrr.c
+++ b/net/netfilter/ipvs/ip_vs_wrr.c
@@ -145,7 +145,7 @@ static int ip_vs_wrr_dest_changed(struct ip_vs_service *svc,
{
struct ip_vs_wrr_mark *mark = svc->sched_data;
- write_lock_bh(&svc->sched_lock);
+ spin_lock_bh(&svc->sched_lock);
mark->cl = list_entry(&svc->destinations, struct ip_vs_dest, n_list);
mark->di = ip_vs_wrr_gcd_weight(svc);
mark->mw = ip_vs_wrr_max_weight(svc) - (mark->di - 1);
@@ -153,7 +153,7 @@ static int ip_vs_wrr_dest_changed(struct ip_vs_service *svc,
mark->cw = mark->mw;
else if (mark->di > 1)
mark->cw = (mark->cw / mark->di) * mark->di + 1;
- write_unlock_bh(&svc->sched_lock);
+ spin_unlock_bh(&svc->sched_lock);
return 0;
}
@@ -170,7 +170,7 @@ ip_vs_wrr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
IP_VS_DBG(6, "%s(): Scheduling...\n", __func__);
- write_lock(&svc->sched_lock);
+ spin_lock(&svc->sched_lock);
dest = mark->cl;
/* No available dests? */
if (mark->mw == 0)
@@ -222,7 +222,7 @@ found:
mark->cl = dest;
out:
- write_unlock(&svc->sched_lock);
+ spin_unlock(&svc->sched_lock);
return dest;
err_noavail:
--
1.7.3.4
^ permalink raw reply related [flat|nested] 23+ messages in thread* [PATCH net-next 17/19] ipvs: convert dests to rcu
2013-03-22 9:46 [PATCH net-next 00/19] IPVS optimizations, part 2 Julian Anastasov
` (15 preceding siblings ...)
2013-03-22 9:46 ` [PATCH net-next 16/19] ipvs: convert sched_lock to spin lock Julian Anastasov
@ 2013-03-22 9:46 ` Julian Anastasov
2013-03-22 9:46 ` [PATCH net-next 18/19] ipvs: convert services " Julian Anastasov
` (2 subsequent siblings)
19 siblings, 0 replies; 23+ messages in thread
From: Julian Anastasov @ 2013-03-22 9:46 UTC (permalink / raw)
To: Simon Horman; +Cc: lvs-devel, netdev
In previous commits the schedulers started to access
svc->destinations with _rcu list traversal primitives
because the IP_VS_WAIT_WHILE macro still plays the role of
grace period. Now it is time to finish the updating part,
i.e. adding and deleting of dests with _rcu suffix before
removing the IP_VS_WAIT_WHILE in next commit.
We use the same rule for conns as for the
schedulers: dests can be searched in RCU read-side critical
section where ip_vs_dest_hold can be called by ip_vs_bind_dest.
Some things are not perfect, for example, calling
functions like ip_vs_lookup_dest from updating code under
RCU, just because we use some function both from reader
and from updater.
Signed-off-by: Julian Anastasov <ja@ssi.bg>
---
include/net/ip_vs.h | 2 +-
net/netfilter/ipvs/ip_vs_conn.c | 8 +++++---
net/netfilter/ipvs/ip_vs_ctl.c | 32 +++++++++++++++++---------------
net/netfilter/ipvs/ip_vs_sync.c | 11 ++++-------
4 files changed, 27 insertions(+), 26 deletions(-)
diff --git a/include/net/ip_vs.h b/include/net/ip_vs.h
index 034ce38..40bb2a6 100644
--- a/include/net/ip_vs.h
+++ b/include/net/ip_vs.h
@@ -1434,7 +1434,7 @@ extern struct ip_vs_dest *
ip_vs_find_dest(struct net *net, int af, const union nf_inet_addr *daddr,
__be16 dport, const union nf_inet_addr *vaddr, __be16 vport,
__u16 protocol, __u32 fwmark, __u32 flags);
-extern struct ip_vs_dest *ip_vs_try_bind_dest(struct ip_vs_conn *cp);
+extern void ip_vs_try_bind_dest(struct ip_vs_conn *cp);
static inline void ip_vs_dest_hold(struct ip_vs_dest *dest)
{
diff --git a/net/netfilter/ipvs/ip_vs_conn.c b/net/netfilter/ipvs/ip_vs_conn.c
index 1b29e4a..54de340 100644
--- a/net/netfilter/ipvs/ip_vs_conn.c
+++ b/net/netfilter/ipvs/ip_vs_conn.c
@@ -611,10 +611,11 @@ ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest)
* Check if there is a destination for the connection, if so
* bind the connection to the destination.
*/
-struct ip_vs_dest *ip_vs_try_bind_dest(struct ip_vs_conn *cp)
+void ip_vs_try_bind_dest(struct ip_vs_conn *cp)
{
struct ip_vs_dest *dest;
+ rcu_read_lock();
dest = ip_vs_find_dest(ip_vs_conn_net(cp), cp->af, &cp->daddr,
cp->dport, &cp->vaddr, cp->vport,
cp->protocol, cp->fwmark, cp->flags);
@@ -624,7 +625,8 @@ struct ip_vs_dest *ip_vs_try_bind_dest(struct ip_vs_conn *cp)
spin_lock(&cp->lock);
if (cp->dest) {
spin_unlock(&cp->lock);
- return dest;
+ rcu_read_unlock();
+ return;
}
/* Applications work depending on the forwarding method
@@ -648,7 +650,7 @@ struct ip_vs_dest *ip_vs_try_bind_dest(struct ip_vs_conn *cp)
if (pd && atomic_read(&pd->appcnt))
ip_vs_bind_app(cp, pd->pp);
}
- return dest;
+ rcu_read_unlock();
}
diff --git a/net/netfilter/ipvs/ip_vs_ctl.c b/net/netfilter/ipvs/ip_vs_ctl.c
index ada61a8..8778927 100644
--- a/net/netfilter/ipvs/ip_vs_ctl.c
+++ b/net/netfilter/ipvs/ip_vs_ctl.c
@@ -563,8 +563,8 @@ bool ip_vs_has_real_service(struct net *net, int af, __u16 protocol,
return false;
}
-/*
- * Lookup destination by {addr,port} in the given service
+/* Lookup destination by {addr,port} in the given service
+ * Called under RCU lock.
*/
static struct ip_vs_dest *
ip_vs_lookup_dest(struct ip_vs_service *svc, const union nf_inet_addr *daddr,
@@ -575,7 +575,7 @@ ip_vs_lookup_dest(struct ip_vs_service *svc, const union nf_inet_addr *daddr,
/*
* Find the destination for the given service
*/
- list_for_each_entry(dest, &svc->destinations, n_list) {
+ list_for_each_entry_rcu(dest, &svc->destinations, n_list) {
if ((dest->af == svc->af)
&& ip_vs_addr_equal(svc->af, &dest->addr, daddr)
&& (dest->port == dport)) {
@@ -589,10 +589,11 @@ ip_vs_lookup_dest(struct ip_vs_service *svc, const union nf_inet_addr *daddr,
/*
* Find destination by {daddr,dport,vaddr,protocol}
- * Cretaed to be used in ip_vs_process_message() in
+ * Created to be used in ip_vs_process_message() in
* the backup synchronization daemon. It finds the
* destination to be bound to the received connection
* on the backup.
+ * Called under RCU lock, no refcnt is returned.
*/
struct ip_vs_dest *ip_vs_find_dest(struct net *net, int af,
const union nf_inet_addr *daddr,
@@ -613,8 +614,6 @@ struct ip_vs_dest *ip_vs_find_dest(struct net *net, int af,
dest = ip_vs_lookup_dest(svc, daddr, port);
if (!dest)
dest = ip_vs_lookup_dest(svc, daddr, port ^ dport);
- if (dest)
- ip_vs_dest_hold(dest);
ip_vs_service_put(svc);
return dest;
}
@@ -824,7 +823,7 @@ __ip_vs_update_dest(struct ip_vs_service *svc, struct ip_vs_dest *dest,
IP_VS_WAIT_WHILE(atomic_read(&svc->usecnt) > 0);
if (add) {
- list_add(&dest->n_list, &svc->destinations);
+ list_add_rcu(&dest->n_list, &svc->destinations);
svc->num_dests++;
if (svc->scheduler->add_dest)
svc->scheduler->add_dest(svc, dest);
@@ -931,10 +930,10 @@ ip_vs_add_dest(struct ip_vs_service *svc, struct ip_vs_dest_user_kern *udest)
ip_vs_addr_copy(svc->af, &daddr, &udest->addr);
- /*
- * Check if the dest already exists in the list
- */
+ /* We use function that requires RCU lock */
+ rcu_read_lock();
dest = ip_vs_lookup_dest(svc, &daddr, dport);
+ rcu_read_unlock();
if (dest != NULL) {
IP_VS_DBG(1, "%s(): dest already exists\n", __func__);
@@ -995,10 +994,10 @@ ip_vs_edit_dest(struct ip_vs_service *svc, struct ip_vs_dest_user_kern *udest)
ip_vs_addr_copy(svc->af, &daddr, &udest->addr);
- /*
- * Lookup the destination list
- */
+ /* We use function that requires RCU lock */
+ rcu_read_lock();
dest = ip_vs_lookup_dest(svc, &daddr, dport);
+ rcu_read_unlock();
if (dest == NULL) {
IP_VS_DBG(1, "%s(): dest doesn't exist\n", __func__);
@@ -1067,7 +1066,7 @@ static void __ip_vs_unlink_dest(struct ip_vs_service *svc,
/*
* Remove it from the d-linked destination list.
*/
- list_del(&dest->n_list);
+ list_del_rcu(&dest->n_list);
svc->num_dests--;
if (svcupd && svc->scheduler->del_dest)
@@ -1092,7 +1091,10 @@ ip_vs_del_dest(struct ip_vs_service *svc, struct ip_vs_dest_user_kern *udest)
EnterFunction(2);
+ /* We use function that requires RCU lock */
+ rcu_read_lock();
dest = ip_vs_lookup_dest(svc, &udest->addr, dport);
+ rcu_read_unlock();
if (dest == NULL) {
IP_VS_DBG(1, "%s(): destination not found!\n", __func__);
@@ -2102,7 +2104,7 @@ static int ip_vs_info_seq_show(struct seq_file *seq, void *v)
else
seq_putc(seq, '\n');
- list_for_each_entry(dest, &svc->destinations, n_list) {
+ list_for_each_entry_rcu(dest, &svc->destinations, n_list) {
#ifdef CONFIG_IP_VS_IPV6
if (dest->af == AF_INET6)
seq_printf(seq,
diff --git a/net/netfilter/ipvs/ip_vs_sync.c b/net/netfilter/ipvs/ip_vs_sync.c
index 6cc3e52..9724174 100644
--- a/net/netfilter/ipvs/ip_vs_sync.c
+++ b/net/netfilter/ipvs/ip_vs_sync.c
@@ -858,23 +858,20 @@ static void ip_vs_proc_conn(struct net *net, struct ip_vs_conn_param *param,
flags |= cp->flags & ~IP_VS_CONN_F_BACKUP_UPD_MASK;
cp->flags = flags;
spin_unlock(&cp->lock);
- if (!dest) {
- dest = ip_vs_try_bind_dest(cp);
- if (dest)
- ip_vs_dest_put(dest);
- }
+ if (!dest)
+ ip_vs_try_bind_dest(cp);
} else {
/*
* Find the appropriate destination for the connection.
* If it is not found the connection will remain unbound
* but still handled.
*/
+ rcu_read_lock();
dest = ip_vs_find_dest(net, type, daddr, dport, param->vaddr,
param->vport, protocol, fwmark, flags);
cp = ip_vs_conn_new(param, daddr, dport, flags, dest, fwmark);
- if (dest)
- ip_vs_dest_put(dest);
+ rcu_read_unlock();
if (!cp) {
if (param->pe_data)
kfree(param->pe_data);
--
1.7.3.4
^ permalink raw reply related [flat|nested] 23+ messages in thread* [PATCH net-next 18/19] ipvs: convert services to rcu
2013-03-22 9:46 [PATCH net-next 00/19] IPVS optimizations, part 2 Julian Anastasov
` (16 preceding siblings ...)
2013-03-22 9:46 ` [PATCH net-next 17/19] ipvs: convert dests to rcu Julian Anastasov
@ 2013-03-22 9:46 ` Julian Anastasov
2013-03-22 9:46 ` [PATCH net-next 19/19] ipvs: do not disable bh for long time Julian Anastasov
2013-03-28 5:41 ` [PATCH net-next 00/19] IPVS optimizations, part 2 Simon Horman
19 siblings, 0 replies; 23+ messages in thread
From: Julian Anastasov @ 2013-03-22 9:46 UTC (permalink / raw)
To: Simon Horman; +Cc: lvs-devel, netdev
This is the final step in RCU conversion.
Things that are removed:
- svc->usecnt: now svc is accessed under RCU read lock
- svc->inc: and some unused code
- ip_vs_bind_pe and ip_vs_unbind_pe: no ability to replace PE
- __ip_vs_svc_lock: replaced with RCU
- IP_VS_WAIT_WHILE: now readers lookup svcs and dests under
RCU and work in parallel with configuration
Other changes:
- before now, a RCU read-side critical section included the
calling of the schedule method, now it is extended to include
service lookup
- ip_vs_svc_table and ip_vs_svc_fwm_table are now using hlist
- svc->pe and svc->scheduler remain to the end (of grace period),
the schedulers are prepared for such RCU readers
even after done_service is called but they need
to use synchronize_rcu because last ip_vs_scheduler_put
can happen while RCU read-side critical sections
use an outdated svc->scheduler pointer
- as planned, update_service is removed
- empty services can be freed immediately after grace period.
If dests were present, the services are freed from
the dest trash code
Signed-off-by: Julian Anastasov <ja@ssi.bg>
---
include/net/ip_vs.h | 28 +--
net/netfilter/ipvs/ip_vs_core.c | 26 +--
net/netfilter/ipvs/ip_vs_ctl.c | 299 ++++++++++++++-------------------
net/netfilter/ipvs/ip_vs_dh.c | 1 +
net/netfilter/ipvs/ip_vs_lblc.c | 1 +
net/netfilter/ipvs/ip_vs_lblcr.c | 1 +
net/netfilter/ipvs/ip_vs_lc.c | 1 +
net/netfilter/ipvs/ip_vs_nq.c | 1 +
net/netfilter/ipvs/ip_vs_pe.c | 12 --
net/netfilter/ipvs/ip_vs_proto_sctp.c | 14 +-
net/netfilter/ipvs/ip_vs_proto_tcp.c | 14 +-
net/netfilter/ipvs/ip_vs_proto_udp.c | 14 +-
net/netfilter/ipvs/ip_vs_rr.c | 1 +
net/netfilter/ipvs/ip_vs_sched.c | 28 ++--
net/netfilter/ipvs/ip_vs_sed.c | 1 +
net/netfilter/ipvs/ip_vs_sh.c | 1 +
net/netfilter/ipvs/ip_vs_wlc.c | 1 +
net/netfilter/ipvs/ip_vs_wrr.c | 1 +
18 files changed, 186 insertions(+), 259 deletions(-)
diff --git a/include/net/ip_vs.h b/include/net/ip_vs.h
index 40bb2a6..3ee1f9e 100644
--- a/include/net/ip_vs.h
+++ b/include/net/ip_vs.h
@@ -359,8 +359,6 @@ static inline const char *ip_vs_dbg_addr(int af, char *buf, size_t buf_len,
#define LeaveFunction(level) do {} while (0)
#endif
-#define IP_VS_WAIT_WHILE(expr) while (expr) { cpu_relax(); }
-
/*
* The port number of FTP service (in network order).
@@ -712,10 +710,9 @@ struct ip_vs_dest_user_kern {
* and the forwarding entries
*/
struct ip_vs_service {
- struct list_head s_list; /* for normal service table */
- struct list_head f_list; /* for fwmark-based service table */
+ struct hlist_node s_list; /* for normal service table */
+ struct hlist_node f_list; /* for fwmark-based service table */
atomic_t refcnt; /* reference counter */
- atomic_t usecnt; /* use counter */
u16 af; /* address family */
__u16 protocol; /* which protocol (TCP/UDP) */
@@ -730,15 +727,16 @@ struct ip_vs_service {
struct list_head destinations; /* real server d-linked list */
__u32 num_dests; /* number of servers */
struct ip_vs_stats stats; /* statistics for the service */
- struct ip_vs_app *inc; /* bind conns to this app inc */
/* for scheduling */
- struct ip_vs_scheduler *scheduler; /* bound scheduler object */
+ struct ip_vs_scheduler __rcu *scheduler; /* bound scheduler object */
spinlock_t sched_lock; /* lock sched_data */
void *sched_data; /* scheduler application data */
/* alternate persistence engine */
- struct ip_vs_pe *pe;
+ struct ip_vs_pe __rcu *pe;
+
+ struct rcu_head rcu_head;
};
/* Information for cached dst */
@@ -807,8 +805,6 @@ struct ip_vs_scheduler {
int (*init_service)(struct ip_vs_service *svc);
/* scheduling service finish */
void (*done_service)(struct ip_vs_service *svc);
- /* scheduler updating service */
- int (*update_service)(struct ip_vs_service *svc);
/* dest is linked */
int (*add_dest)(struct ip_vs_service *svc, struct ip_vs_dest *dest);
/* dest is unlinked */
@@ -1344,8 +1340,6 @@ extern void ip_vs_app_inc_put(struct ip_vs_app *inc);
extern int ip_vs_app_pkt_out(struct ip_vs_conn *, struct sk_buff *skb);
extern int ip_vs_app_pkt_in(struct ip_vs_conn *, struct sk_buff *skb);
-void ip_vs_bind_pe(struct ip_vs_service *svc, struct ip_vs_pe *pe);
-void ip_vs_unbind_pe(struct ip_vs_service *svc);
int register_ip_vs_pe(struct ip_vs_pe *pe);
int unregister_ip_vs_pe(struct ip_vs_pe *pe);
struct ip_vs_pe *ip_vs_pe_getbyname(const char *name);
@@ -1392,7 +1386,8 @@ extern int register_ip_vs_scheduler(struct ip_vs_scheduler *scheduler);
extern int unregister_ip_vs_scheduler(struct ip_vs_scheduler *scheduler);
extern int ip_vs_bind_scheduler(struct ip_vs_service *svc,
struct ip_vs_scheduler *scheduler);
-extern void ip_vs_unbind_scheduler(struct ip_vs_service *svc);
+extern void ip_vs_unbind_scheduler(struct ip_vs_service *svc,
+ struct ip_vs_scheduler *sched);
extern struct ip_vs_scheduler *ip_vs_scheduler_get(const char *sched_name);
extern void ip_vs_scheduler_put(struct ip_vs_scheduler *scheduler);
extern struct ip_vs_conn *
@@ -1412,14 +1407,9 @@ extern struct ip_vs_stats ip_vs_stats;
extern int sysctl_ip_vs_sync_ver;
extern struct ip_vs_service *
-ip_vs_service_get(struct net *net, int af, __u32 fwmark, __u16 protocol,
+ip_vs_service_find(struct net *net, int af, __u32 fwmark, __u16 protocol,
const union nf_inet_addr *vaddr, __be16 vport);
-static inline void ip_vs_service_put(struct ip_vs_service *svc)
-{
- atomic_dec(&svc->usecnt);
-}
-
extern bool
ip_vs_has_real_service(struct net *net, int af, __u16 protocol,
const union nf_inet_addr *daddr, __be16 dport);
diff --git a/net/netfilter/ipvs/ip_vs_core.c b/net/netfilter/ipvs/ip_vs_core.c
index 7bf62d0..8814f02 100644
--- a/net/netfilter/ipvs/ip_vs_core.c
+++ b/net/netfilter/ipvs/ip_vs_core.c
@@ -206,7 +206,7 @@ ip_vs_conn_fill_param_persist(const struct ip_vs_service *svc,
{
ip_vs_conn_fill_param(svc->net, svc->af, protocol, caddr, cport, vaddr,
vport, p);
- p->pe = svc->pe;
+ p->pe = rcu_dereference(svc->pe);
if (p->pe && p->pe->fill_param)
return p->pe->fill_param(p, skb);
@@ -299,15 +299,16 @@ ip_vs_sched_persist(struct ip_vs_service *svc,
/* Check if a template already exists */
ct = ip_vs_ct_in_get(¶m);
if (!ct || !ip_vs_check_template(ct)) {
+ struct ip_vs_scheduler *sched;
+
/*
* No template found or the dest of the connection
* template is not available.
* return *ignored=0 i.e. ICMP and NF_DROP
*/
- rcu_read_lock();
- dest = svc->scheduler->schedule(svc, skb);
+ sched = rcu_dereference(svc->scheduler);
+ dest = sched->schedule(svc, skb);
if (!dest) {
- rcu_read_unlock();
IP_VS_DBG(1, "p-schedule: no dest found.\n");
kfree(param.pe_data);
*ignored = 0;
@@ -323,7 +324,6 @@ ip_vs_sched_persist(struct ip_vs_service *svc,
* when the template expires */
ct = ip_vs_conn_new(¶m, &dest->addr, dport,
IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
- rcu_read_unlock();
if (ct == NULL) {
kfree(param.pe_data);
*ignored = -1;
@@ -397,6 +397,7 @@ ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
{
struct ip_vs_protocol *pp = pd->pp;
struct ip_vs_conn *cp = NULL;
+ struct ip_vs_scheduler *sched;
struct ip_vs_dest *dest;
__be16 _ports[2], *pptr;
unsigned int flags;
@@ -452,10 +453,9 @@ ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
return NULL;
}
- rcu_read_lock();
- dest = svc->scheduler->schedule(svc, skb);
+ sched = rcu_dereference(svc->scheduler);
+ dest = sched->schedule(svc, skb);
if (dest == NULL) {
- rcu_read_unlock();
IP_VS_DBG(1, "Schedule: no dest found.\n");
return NULL;
}
@@ -476,7 +476,6 @@ ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
cp = ip_vs_conn_new(&p, &dest->addr,
dest->port ? dest->port : pptr[1],
flags, dest, skb->mark);
- rcu_read_unlock();
if (!cp) {
*ignored = -1;
return NULL;
@@ -513,7 +512,6 @@ int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
if (pptr == NULL) {
- ip_vs_service_put(svc);
return NF_DROP;
}
@@ -539,8 +537,6 @@ int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
IP_VS_CONN_F_ONE_PACKET : 0;
union nf_inet_addr daddr = { .all = { 0, 0, 0, 0 } };
- ip_vs_service_put(svc);
-
/* create a new connection entry */
IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
{
@@ -577,12 +573,8 @@ int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
* listed in the ipvs table), pass the packets, because it is
* not ipvs job to decide to drop the packets.
*/
- if ((svc->port == FTPPORT) && (pptr[1] != FTPPORT)) {
- ip_vs_service_put(svc);
+ if ((svc->port == FTPPORT) && (pptr[1] != FTPPORT))
return NF_ACCEPT;
- }
-
- ip_vs_service_put(svc);
/*
* Notify the client that the destination is unreachable, and
diff --git a/net/netfilter/ipvs/ip_vs_ctl.c b/net/netfilter/ipvs/ip_vs_ctl.c
index 8778927..26c87a5 100644
--- a/net/netfilter/ipvs/ip_vs_ctl.c
+++ b/net/netfilter/ipvs/ip_vs_ctl.c
@@ -55,9 +55,6 @@
/* semaphore for IPVS sockopts. And, [gs]etsockopt may sleep. */
static DEFINE_MUTEX(__ip_vs_mutex);
-/* lock for service table */
-static DEFINE_RWLOCK(__ip_vs_svc_lock);
-
/* sysctl variables */
#ifdef CONFIG_IP_VS_DEBUG
@@ -257,9 +254,9 @@ ip_vs_use_count_dec(void)
#define IP_VS_SVC_TAB_MASK (IP_VS_SVC_TAB_SIZE - 1)
/* the service table hashed by <protocol, addr, port> */
-static struct list_head ip_vs_svc_table[IP_VS_SVC_TAB_SIZE];
+static struct hlist_head ip_vs_svc_table[IP_VS_SVC_TAB_SIZE];
/* the service table hashed by fwmark */
-static struct list_head ip_vs_svc_fwm_table[IP_VS_SVC_TAB_SIZE];
+static struct hlist_head ip_vs_svc_fwm_table[IP_VS_SVC_TAB_SIZE];
/*
@@ -312,13 +309,13 @@ static int ip_vs_svc_hash(struct ip_vs_service *svc)
*/
hash = ip_vs_svc_hashkey(svc->net, svc->af, svc->protocol,
&svc->addr, svc->port);
- list_add(&svc->s_list, &ip_vs_svc_table[hash]);
+ hlist_add_head_rcu(&svc->s_list, &ip_vs_svc_table[hash]);
} else {
/*
* Hash it by fwmark in svc_fwm_table
*/
hash = ip_vs_svc_fwm_hashkey(svc->net, svc->fwmark);
- list_add(&svc->f_list, &ip_vs_svc_fwm_table[hash]);
+ hlist_add_head_rcu(&svc->f_list, &ip_vs_svc_fwm_table[hash]);
}
svc->flags |= IP_VS_SVC_F_HASHED;
@@ -342,10 +339,10 @@ static int ip_vs_svc_unhash(struct ip_vs_service *svc)
if (svc->fwmark == 0) {
/* Remove it from the svc_table table */
- list_del(&svc->s_list);
+ hlist_del_rcu(&svc->s_list);
} else {
/* Remove it from the svc_fwm_table table */
- list_del(&svc->f_list);
+ hlist_del_rcu(&svc->f_list);
}
svc->flags &= ~IP_VS_SVC_F_HASHED;
@@ -367,7 +364,7 @@ __ip_vs_service_find(struct net *net, int af, __u16 protocol,
/* Check for "full" addressed entries */
hash = ip_vs_svc_hashkey(net, af, protocol, vaddr, vport);
- list_for_each_entry(svc, &ip_vs_svc_table[hash], s_list){
+ hlist_for_each_entry_rcu(svc, &ip_vs_svc_table[hash], s_list) {
if ((svc->af == af)
&& ip_vs_addr_equal(af, &svc->addr, vaddr)
&& (svc->port == vport)
@@ -394,7 +391,7 @@ __ip_vs_svc_fwm_find(struct net *net, int af, __u32 fwmark)
/* Check for fwmark addressed entries */
hash = ip_vs_svc_fwm_hashkey(net, fwmark);
- list_for_each_entry(svc, &ip_vs_svc_fwm_table[hash], f_list) {
+ hlist_for_each_entry_rcu(svc, &ip_vs_svc_fwm_table[hash], f_list) {
if (svc->fwmark == fwmark && svc->af == af
&& net_eq(svc->net, net)) {
/* HIT */
@@ -405,15 +402,14 @@ __ip_vs_svc_fwm_find(struct net *net, int af, __u32 fwmark)
return NULL;
}
+/* Find service, called under RCU lock */
struct ip_vs_service *
-ip_vs_service_get(struct net *net, int af, __u32 fwmark, __u16 protocol,
- const union nf_inet_addr *vaddr, __be16 vport)
+ip_vs_service_find(struct net *net, int af, __u32 fwmark, __u16 protocol,
+ const union nf_inet_addr *vaddr, __be16 vport)
{
struct ip_vs_service *svc;
struct netns_ipvs *ipvs = net_ipvs(net);
- read_lock(&__ip_vs_svc_lock);
-
/*
* Check the table hashed by fwmark first
*/
@@ -449,10 +445,6 @@ ip_vs_service_get(struct net *net, int af, __u32 fwmark, __u16 protocol,
}
out:
- if (svc)
- atomic_inc(&svc->usecnt);
- read_unlock(&__ip_vs_svc_lock);
-
IP_VS_DBG_BUF(9, "lookup service: fwm %u %s %s:%u %s\n",
fwmark, ip_vs_proto_name(protocol),
IP_VS_DBG_ADDR(af, vaddr), ntohs(vport),
@@ -469,6 +461,13 @@ __ip_vs_bind_svc(struct ip_vs_dest *dest, struct ip_vs_service *svc)
dest->svc = svc;
}
+static void ip_vs_service_free(struct ip_vs_service *svc)
+{
+ if (svc->stats.cpustats)
+ free_percpu(svc->stats.cpustats);
+ kfree(svc);
+}
+
static void
__ip_vs_unbind_svc(struct ip_vs_dest *dest)
{
@@ -476,12 +475,11 @@ __ip_vs_unbind_svc(struct ip_vs_dest *dest)
dest->svc = NULL;
if (atomic_dec_and_test(&svc->refcnt)) {
- IP_VS_DBG_BUF(3, "Removing service %u/%s:%u usecnt=%d\n",
+ IP_VS_DBG_BUF(3, "Removing service %u/%s:%u\n",
svc->fwmark,
IP_VS_DBG_ADDR(svc->af, &svc->addr),
- ntohs(svc->port), atomic_read(&svc->usecnt));
- free_percpu(svc->stats.cpustats);
- kfree(svc);
+ ntohs(svc->port));
+ ip_vs_service_free(svc);
}
}
@@ -606,7 +604,7 @@ struct ip_vs_dest *ip_vs_find_dest(struct net *net, int af,
struct ip_vs_service *svc;
__be16 port = dport;
- svc = ip_vs_service_get(net, af, fwmark, protocol, vaddr, vport);
+ svc = ip_vs_service_find(net, af, fwmark, protocol, vaddr, vport);
if (!svc)
return NULL;
if (fwmark && (flags & IP_VS_CONN_F_FWD_MASK) != IP_VS_CONN_F_MASQ)
@@ -614,7 +612,6 @@ struct ip_vs_dest *ip_vs_find_dest(struct net *net, int af,
dest = ip_vs_lookup_dest(svc, daddr, port);
if (!dest)
dest = ip_vs_lookup_dest(svc, daddr, port ^ dport);
- ip_vs_service_put(svc);
return dest;
}
@@ -772,6 +769,7 @@ __ip_vs_update_dest(struct ip_vs_service *svc, struct ip_vs_dest *dest,
struct ip_vs_dest_user_kern *udest, int add)
{
struct netns_ipvs *ipvs = net_ipvs(svc->net);
+ struct ip_vs_scheduler *sched;
int conn_flags;
/* set the weight and the flags */
@@ -814,29 +812,17 @@ __ip_vs_update_dest(struct ip_vs_service *svc, struct ip_vs_dest *dest,
__ip_vs_dst_cache_reset(dest);
spin_unlock_bh(&dest->dst_lock);
- if (add)
- ip_vs_start_estimator(svc->net, &dest->stats);
-
- write_lock_bh(&__ip_vs_svc_lock);
-
- /* Wait until all other svc users go away */
- IP_VS_WAIT_WHILE(atomic_read(&svc->usecnt) > 0);
-
+ sched = rcu_dereference_protected(svc->scheduler, 1);
if (add) {
+ ip_vs_start_estimator(svc->net, &dest->stats);
list_add_rcu(&dest->n_list, &svc->destinations);
svc->num_dests++;
- if (svc->scheduler->add_dest)
- svc->scheduler->add_dest(svc, dest);
+ if (sched->add_dest)
+ sched->add_dest(svc, dest);
} else {
- if (svc->scheduler->upd_dest)
- svc->scheduler->upd_dest(svc, dest);
+ if (sched->upd_dest)
+ sched->upd_dest(svc, dest);
}
-
- /* call the update_service, because server weight may be changed */
- if (svc->scheduler->update_service)
- svc->scheduler->update_service(svc);
-
- write_unlock_bh(&__ip_vs_svc_lock);
}
@@ -1069,14 +1055,13 @@ static void __ip_vs_unlink_dest(struct ip_vs_service *svc,
list_del_rcu(&dest->n_list);
svc->num_dests--;
- if (svcupd && svc->scheduler->del_dest)
- svc->scheduler->del_dest(svc, dest);
+ if (svcupd) {
+ struct ip_vs_scheduler *sched;
- /*
- * Call the update_service function of its scheduler
- */
- if (svcupd && svc->scheduler->update_service)
- svc->scheduler->update_service(svc);
+ sched = rcu_dereference_protected(svc->scheduler, 1);
+ if (sched->del_dest)
+ sched->del_dest(svc, dest);
+ }
}
@@ -1101,20 +1086,11 @@ ip_vs_del_dest(struct ip_vs_service *svc, struct ip_vs_dest_user_kern *udest)
return -ENOENT;
}
- write_lock_bh(&__ip_vs_svc_lock);
-
- /*
- * Wait until all other svc users go away.
- */
- IP_VS_WAIT_WHILE(atomic_read(&svc->usecnt) > 0);
-
/*
* Unlink dest from the service
*/
__ip_vs_unlink_dest(svc, dest, 1);
- write_unlock_bh(&__ip_vs_svc_lock);
-
/*
* Delete the destination
*/
@@ -1205,7 +1181,6 @@ ip_vs_add_service(struct net *net, struct ip_vs_service_user_kern *u,
}
/* I'm the first user of the service */
- atomic_set(&svc->usecnt, 0);
atomic_set(&svc->refcnt, 0);
svc->af = u->af;
@@ -1229,7 +1204,7 @@ ip_vs_add_service(struct net *net, struct ip_vs_service_user_kern *u,
sched = NULL;
/* Bind the ct retriever */
- ip_vs_bind_pe(svc, pe);
+ RCU_INIT_POINTER(svc->pe, pe);
pe = NULL;
/* Update the virtual service counters */
@@ -1245,9 +1220,7 @@ ip_vs_add_service(struct net *net, struct ip_vs_service_user_kern *u,
ipvs->num_services++;
/* Hash the service into the service table */
- write_lock_bh(&__ip_vs_svc_lock);
ip_vs_svc_hash(svc);
- write_unlock_bh(&__ip_vs_svc_lock);
*svc_p = svc;
/* Now there is a service - full throttle */
@@ -1257,15 +1230,8 @@ ip_vs_add_service(struct net *net, struct ip_vs_service_user_kern *u,
out_err:
if (svc != NULL) {
- ip_vs_unbind_scheduler(svc);
- if (svc->inc) {
- local_bh_disable();
- ip_vs_app_inc_put(svc->inc);
- local_bh_enable();
- }
- if (svc->stats.cpustats)
- free_percpu(svc->stats.cpustats);
- kfree(svc);
+ ip_vs_unbind_scheduler(svc, sched);
+ ip_vs_service_free(svc);
}
ip_vs_scheduler_put(sched);
ip_vs_pe_put(pe);
@@ -1315,12 +1281,17 @@ ip_vs_edit_service(struct ip_vs_service *svc, struct ip_vs_service_user_kern *u)
}
#endif
- write_lock_bh(&__ip_vs_svc_lock);
-
- /*
- * Wait until all other svc users go away.
- */
- IP_VS_WAIT_WHILE(atomic_read(&svc->usecnt) > 0);
+ old_sched = rcu_dereference_protected(svc->scheduler, 1);
+ if (sched != old_sched) {
+ /* Bind the new scheduler */
+ ret = ip_vs_bind_scheduler(svc, sched);
+ if (ret) {
+ old_sched = sched;
+ goto out;
+ }
+ /* Unbind the old scheduler on success */
+ ip_vs_unbind_scheduler(svc, old_sched);
+ }
/*
* Set the flags and timeout value
@@ -1329,47 +1300,23 @@ ip_vs_edit_service(struct ip_vs_service *svc, struct ip_vs_service_user_kern *u)
svc->timeout = u->timeout * HZ;
svc->netmask = u->netmask;
- old_sched = svc->scheduler;
- if (sched != old_sched) {
- /*
- * Unbind the old scheduler
- */
- ip_vs_unbind_scheduler(svc);
+ old_pe = rcu_dereference_protected(svc->pe, 1);
+ if (pe != old_pe)
+ rcu_assign_pointer(svc->pe, pe);
- /*
- * Bind the new scheduler
- */
- if ((ret = ip_vs_bind_scheduler(svc, sched))) {
- /*
- * If ip_vs_bind_scheduler fails, restore the old
- * scheduler.
- * The main reason of failure is out of memory.
- *
- * The question is if the old scheduler can be
- * restored all the time. TODO: if it cannot be
- * restored some time, we must delete the service,
- * otherwise the system may crash.
- */
- ip_vs_bind_scheduler(svc, old_sched);
- old_sched = sched;
- goto out_unlock;
- }
- }
-
- old_pe = svc->pe;
- if (pe != old_pe) {
- ip_vs_unbind_pe(svc);
- ip_vs_bind_pe(svc, pe);
- }
-
-out_unlock:
- write_unlock_bh(&__ip_vs_svc_lock);
out:
ip_vs_scheduler_put(old_sched);
ip_vs_pe_put(old_pe);
return ret;
}
+static void ip_vs_service_rcu_free(struct rcu_head *head)
+{
+ struct ip_vs_service *svc;
+
+ svc = container_of(head, struct ip_vs_service, rcu_head);
+ ip_vs_service_free(svc);
+}
/*
* Delete a service from the service list
@@ -1392,21 +1339,14 @@ static void __ip_vs_del_service(struct ip_vs_service *svc, bool cleanup)
ip_vs_stop_estimator(svc->net, &svc->stats);
/* Unbind scheduler */
- old_sched = svc->scheduler;
- ip_vs_unbind_scheduler(svc);
+ old_sched = rcu_dereference_protected(svc->scheduler, 1);
+ ip_vs_unbind_scheduler(svc, old_sched);
ip_vs_scheduler_put(old_sched);
- /* Unbind persistence engine */
- old_pe = svc->pe;
- ip_vs_unbind_pe(svc);
+ /* Unbind persistence engine, keep svc->pe */
+ old_pe = rcu_dereference_protected(svc->pe, 1);
ip_vs_pe_put(old_pe);
- /* Unbind app inc */
- if (svc->inc) {
- ip_vs_app_inc_put(svc->inc);
- svc->inc = NULL;
- }
-
/*
* Unlink the whole destination list
*/
@@ -1426,13 +1366,12 @@ static void __ip_vs_del_service(struct ip_vs_service *svc, bool cleanup)
/*
* Free the service if nobody refers to it
*/
- if (atomic_read(&svc->refcnt) == 0) {
- IP_VS_DBG_BUF(3, "Removing service %u/%s:%u usecnt=%d\n",
+ if (atomic_dec_and_test(&svc->refcnt)) {
+ IP_VS_DBG_BUF(3, "Removing service %u/%s:%u\n",
svc->fwmark,
IP_VS_DBG_ADDR(svc->af, &svc->addr),
- ntohs(svc->port), atomic_read(&svc->usecnt));
- free_percpu(svc->stats.cpustats);
- kfree(svc);
+ ntohs(svc->port));
+ call_rcu(&svc->rcu_head, ip_vs_service_rcu_free);
}
/* decrease the module use count */
@@ -1444,21 +1383,14 @@ static void __ip_vs_del_service(struct ip_vs_service *svc, bool cleanup)
*/
static void ip_vs_unlink_service(struct ip_vs_service *svc, bool cleanup)
{
+ /* Hold svc to avoid double release from dest_trash */
+ atomic_inc(&svc->refcnt);
/*
* Unhash it from the service table
*/
- write_lock_bh(&__ip_vs_svc_lock);
-
ip_vs_svc_unhash(svc);
- /*
- * Wait until all the svc users go away.
- */
- IP_VS_WAIT_WHILE(atomic_read(&svc->usecnt) > 0);
-
__ip_vs_del_service(svc, cleanup);
-
- write_unlock_bh(&__ip_vs_svc_lock);
}
/*
@@ -1480,14 +1412,15 @@ static int ip_vs_del_service(struct ip_vs_service *svc)
static int ip_vs_flush(struct net *net, bool cleanup)
{
int idx;
- struct ip_vs_service *svc, *nxt;
+ struct ip_vs_service *svc;
+ struct hlist_node *n;
/*
* Flush the service table hashed by <netns,protocol,addr,port>
*/
for(idx = 0; idx < IP_VS_SVC_TAB_SIZE; idx++) {
- list_for_each_entry_safe(svc, nxt, &ip_vs_svc_table[idx],
- s_list) {
+ hlist_for_each_entry_safe(svc, n, &ip_vs_svc_table[idx],
+ s_list) {
if (net_eq(svc->net, net))
ip_vs_unlink_service(svc, cleanup);
}
@@ -1497,8 +1430,8 @@ static int ip_vs_flush(struct net *net, bool cleanup)
* Flush the service table hashed by fwmark
*/
for(idx = 0; idx < IP_VS_SVC_TAB_SIZE; idx++) {
- list_for_each_entry_safe(svc, nxt,
- &ip_vs_svc_fwm_table[idx], f_list) {
+ hlist_for_each_entry_safe(svc, n, &ip_vs_svc_fwm_table[idx],
+ f_list) {
if (net_eq(svc->net, net))
ip_vs_unlink_service(svc, cleanup);
}
@@ -1556,7 +1489,7 @@ static int ip_vs_dst_event(struct notifier_block *this, unsigned long event,
EnterFunction(2);
mutex_lock(&__ip_vs_mutex);
for (idx = 0; idx < IP_VS_SVC_TAB_SIZE; idx++) {
- list_for_each_entry(svc, &ip_vs_svc_table[idx], s_list) {
+ hlist_for_each_entry(svc, &ip_vs_svc_table[idx], s_list) {
if (net_eq(svc->net, net)) {
list_for_each_entry(dest, &svc->destinations,
n_list) {
@@ -1565,7 +1498,7 @@ static int ip_vs_dst_event(struct notifier_block *this, unsigned long event,
}
}
- list_for_each_entry(svc, &ip_vs_svc_fwm_table[idx], f_list) {
+ hlist_for_each_entry(svc, &ip_vs_svc_fwm_table[idx], f_list) {
if (net_eq(svc->net, net)) {
list_for_each_entry(dest, &svc->destinations,
n_list) {
@@ -1593,12 +1526,10 @@ static int ip_vs_zero_service(struct ip_vs_service *svc)
{
struct ip_vs_dest *dest;
- write_lock_bh(&__ip_vs_svc_lock);
list_for_each_entry(dest, &svc->destinations, n_list) {
ip_vs_zero_stats(&dest->stats);
}
ip_vs_zero_stats(&svc->stats);
- write_unlock_bh(&__ip_vs_svc_lock);
return 0;
}
@@ -1608,14 +1539,14 @@ static int ip_vs_zero_all(struct net *net)
struct ip_vs_service *svc;
for(idx = 0; idx < IP_VS_SVC_TAB_SIZE; idx++) {
- list_for_each_entry(svc, &ip_vs_svc_table[idx], s_list) {
+ hlist_for_each_entry(svc, &ip_vs_svc_table[idx], s_list) {
if (net_eq(svc->net, net))
ip_vs_zero_service(svc);
}
}
for(idx = 0; idx < IP_VS_SVC_TAB_SIZE; idx++) {
- list_for_each_entry(svc, &ip_vs_svc_fwm_table[idx], f_list) {
+ hlist_for_each_entry(svc, &ip_vs_svc_fwm_table[idx], f_list) {
if (net_eq(svc->net, net))
ip_vs_zero_service(svc);
}
@@ -1943,7 +1874,7 @@ static struct ctl_table vs_vars[] = {
struct ip_vs_iter {
struct seq_net_private p; /* Do not move this, netns depends upon it*/
- struct list_head *table;
+ struct hlist_head *table;
int bucket;
};
@@ -1976,7 +1907,7 @@ static struct ip_vs_service *ip_vs_info_array(struct seq_file *seq, loff_t pos)
/* look in hash by protocol */
for (idx = 0; idx < IP_VS_SVC_TAB_SIZE; idx++) {
- list_for_each_entry(svc, &ip_vs_svc_table[idx], s_list) {
+ hlist_for_each_entry_rcu(svc, &ip_vs_svc_table[idx], s_list) {
if (net_eq(svc->net, net) && pos-- == 0) {
iter->table = ip_vs_svc_table;
iter->bucket = idx;
@@ -1987,7 +1918,8 @@ static struct ip_vs_service *ip_vs_info_array(struct seq_file *seq, loff_t pos)
/* keep looking in fwmark */
for (idx = 0; idx < IP_VS_SVC_TAB_SIZE; idx++) {
- list_for_each_entry(svc, &ip_vs_svc_fwm_table[idx], f_list) {
+ hlist_for_each_entry_rcu(svc, &ip_vs_svc_fwm_table[idx],
+ f_list) {
if (net_eq(svc->net, net) && pos-- == 0) {
iter->table = ip_vs_svc_fwm_table;
iter->bucket = idx;
@@ -2000,17 +1932,16 @@ static struct ip_vs_service *ip_vs_info_array(struct seq_file *seq, loff_t pos)
}
static void *ip_vs_info_seq_start(struct seq_file *seq, loff_t *pos)
-__acquires(__ip_vs_svc_lock)
{
- read_lock_bh(&__ip_vs_svc_lock);
+ rcu_read_lock();
return *pos ? ip_vs_info_array(seq, *pos - 1) : SEQ_START_TOKEN;
}
static void *ip_vs_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
{
- struct list_head *e;
+ struct hlist_node *e;
struct ip_vs_iter *iter;
struct ip_vs_service *svc;
@@ -2023,13 +1954,14 @@ static void *ip_vs_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
if (iter->table == ip_vs_svc_table) {
/* next service in table hashed by protocol */
- if ((e = svc->s_list.next) != &ip_vs_svc_table[iter->bucket])
- return list_entry(e, struct ip_vs_service, s_list);
-
+ e = rcu_dereference(hlist_next_rcu(&svc->s_list));
+ if (e)
+ return hlist_entry(e, struct ip_vs_service, s_list);
while (++iter->bucket < IP_VS_SVC_TAB_SIZE) {
- list_for_each_entry(svc,&ip_vs_svc_table[iter->bucket],
- s_list) {
+ hlist_for_each_entry_rcu(svc,
+ &ip_vs_svc_table[iter->bucket],
+ s_list) {
return svc;
}
}
@@ -2040,13 +1972,15 @@ static void *ip_vs_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
}
/* next service in hashed by fwmark */
- if ((e = svc->f_list.next) != &ip_vs_svc_fwm_table[iter->bucket])
- return list_entry(e, struct ip_vs_service, f_list);
+ e = rcu_dereference(hlist_next_rcu(&svc->f_list));
+ if (e)
+ return hlist_entry(e, struct ip_vs_service, f_list);
scan_fwmark:
while (++iter->bucket < IP_VS_SVC_TAB_SIZE) {
- list_for_each_entry(svc, &ip_vs_svc_fwm_table[iter->bucket],
- f_list)
+ hlist_for_each_entry_rcu(svc,
+ &ip_vs_svc_fwm_table[iter->bucket],
+ f_list)
return svc;
}
@@ -2054,9 +1988,8 @@ static void *ip_vs_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
}
static void ip_vs_info_seq_stop(struct seq_file *seq, void *v)
-__releases(__ip_vs_svc_lock)
{
- read_unlock_bh(&__ip_vs_svc_lock);
+ rcu_read_unlock();
}
@@ -2074,6 +2007,7 @@ static int ip_vs_info_seq_show(struct seq_file *seq, void *v)
const struct ip_vs_service *svc = v;
const struct ip_vs_iter *iter = seq->private;
const struct ip_vs_dest *dest;
+ struct ip_vs_scheduler *sched = rcu_dereference(svc->scheduler);
if (iter->table == ip_vs_svc_table) {
#ifdef CONFIG_IP_VS_IPV6
@@ -2082,18 +2016,18 @@ static int ip_vs_info_seq_show(struct seq_file *seq, void *v)
ip_vs_proto_name(svc->protocol),
&svc->addr.in6,
ntohs(svc->port),
- svc->scheduler->name);
+ sched->name);
else
#endif
seq_printf(seq, "%s %08X:%04X %s %s ",
ip_vs_proto_name(svc->protocol),
ntohl(svc->addr.ip),
ntohs(svc->port),
- svc->scheduler->name,
+ sched->name,
(svc->flags & IP_VS_SVC_F_ONEPACKET)?"ops ":"");
} else {
seq_printf(seq, "FWM %08X %s %s",
- svc->fwmark, svc->scheduler->name,
+ svc->fwmark, sched->name,
(svc->flags & IP_VS_SVC_F_ONEPACKET)?"ops ":"");
}
@@ -2449,11 +2383,13 @@ do_ip_vs_set_ctl(struct sock *sk, int cmd, void __user *user, unsigned int len)
}
/* Lookup the exact service by <protocol, addr, port> or fwmark */
+ rcu_read_lock();
if (usvc.fwmark == 0)
svc = __ip_vs_service_find(net, usvc.af, usvc.protocol,
&usvc.addr, usvc.port);
else
svc = __ip_vs_svc_fwm_find(net, usvc.af, usvc.fwmark);
+ rcu_read_unlock();
if (cmd != IP_VS_SO_SET_ADD
&& (svc == NULL || svc->protocol != usvc.protocol)) {
@@ -2505,11 +2441,14 @@ do_ip_vs_set_ctl(struct sock *sk, int cmd, void __user *user, unsigned int len)
static void
ip_vs_copy_service(struct ip_vs_service_entry *dst, struct ip_vs_service *src)
{
+ struct ip_vs_scheduler *sched;
+
+ sched = rcu_dereference_protected(src->scheduler, 1);
dst->protocol = src->protocol;
dst->addr = src->addr.ip;
dst->port = src->port;
dst->fwmark = src->fwmark;
- strlcpy(dst->sched_name, src->scheduler->name, sizeof(dst->sched_name));
+ strlcpy(dst->sched_name, sched->name, sizeof(dst->sched_name));
dst->flags = src->flags;
dst->timeout = src->timeout / HZ;
dst->netmask = src->netmask;
@@ -2528,7 +2467,7 @@ __ip_vs_get_service_entries(struct net *net,
int ret = 0;
for (idx = 0; idx < IP_VS_SVC_TAB_SIZE; idx++) {
- list_for_each_entry(svc, &ip_vs_svc_table[idx], s_list) {
+ hlist_for_each_entry(svc, &ip_vs_svc_table[idx], s_list) {
/* Only expose IPv4 entries to old interface */
if (svc->af != AF_INET || !net_eq(svc->net, net))
continue;
@@ -2547,7 +2486,7 @@ __ip_vs_get_service_entries(struct net *net,
}
for (idx = 0; idx < IP_VS_SVC_TAB_SIZE; idx++) {
- list_for_each_entry(svc, &ip_vs_svc_fwm_table[idx], f_list) {
+ hlist_for_each_entry(svc, &ip_vs_svc_fwm_table[idx], f_list) {
/* Only expose IPv4 entries to old interface */
if (svc->af != AF_INET || !net_eq(svc->net, net))
continue;
@@ -2576,11 +2515,13 @@ __ip_vs_get_dest_entries(struct net *net, const struct ip_vs_get_dests *get,
union nf_inet_addr addr = { .ip = get->addr };
int ret = 0;
+ rcu_read_lock();
if (get->fwmark)
svc = __ip_vs_svc_fwm_find(net, AF_INET, get->fwmark);
else
svc = __ip_vs_service_find(net, AF_INET, get->protocol, &addr,
get->port);
+ rcu_read_unlock();
if (svc) {
int count = 0;
@@ -2763,12 +2704,14 @@ do_ip_vs_get_ctl(struct sock *sk, int cmd, void __user *user, int *len)
entry = (struct ip_vs_service_entry *)arg;
addr.ip = entry->addr;
+ rcu_read_lock();
if (entry->fwmark)
svc = __ip_vs_svc_fwm_find(net, AF_INET, entry->fwmark);
else
svc = __ip_vs_service_find(net, AF_INET,
entry->protocol, &addr,
entry->port);
+ rcu_read_unlock();
if (svc) {
ip_vs_copy_service(entry, svc);
if (copy_to_user(user, entry, sizeof(*entry)) != 0)
@@ -2925,6 +2868,7 @@ nla_put_failure:
static int ip_vs_genl_fill_service(struct sk_buff *skb,
struct ip_vs_service *svc)
{
+ struct ip_vs_scheduler *sched;
struct nlattr *nl_service;
struct ip_vs_flags flags = { .flags = svc->flags,
.mask = ~0 };
@@ -2945,7 +2889,8 @@ static int ip_vs_genl_fill_service(struct sk_buff *skb,
goto nla_put_failure;
}
- if (nla_put_string(skb, IPVS_SVC_ATTR_SCHED_NAME, svc->scheduler->name) ||
+ sched = rcu_dereference_protected(svc->scheduler, 1);
+ if (nla_put_string(skb, IPVS_SVC_ATTR_SCHED_NAME, sched->name) ||
(svc->pe &&
nla_put_string(skb, IPVS_SVC_ATTR_PE_NAME, svc->pe->name)) ||
nla_put(skb, IPVS_SVC_ATTR_FLAGS, sizeof(flags), &flags) ||
@@ -2996,7 +2941,7 @@ static int ip_vs_genl_dump_services(struct sk_buff *skb,
mutex_lock(&__ip_vs_mutex);
for (i = 0; i < IP_VS_SVC_TAB_SIZE; i++) {
- list_for_each_entry(svc, &ip_vs_svc_table[i], s_list) {
+ hlist_for_each_entry(svc, &ip_vs_svc_table[i], s_list) {
if (++idx <= start || !net_eq(svc->net, net))
continue;
if (ip_vs_genl_dump_service(skb, svc, cb) < 0) {
@@ -3007,7 +2952,7 @@ static int ip_vs_genl_dump_services(struct sk_buff *skb,
}
for (i = 0; i < IP_VS_SVC_TAB_SIZE; i++) {
- list_for_each_entry(svc, &ip_vs_svc_fwm_table[i], f_list) {
+ hlist_for_each_entry(svc, &ip_vs_svc_fwm_table[i], f_list) {
if (++idx <= start || !net_eq(svc->net, net))
continue;
if (ip_vs_genl_dump_service(skb, svc, cb) < 0) {
@@ -3067,11 +3012,13 @@ static int ip_vs_genl_parse_service(struct net *net,
usvc->fwmark = 0;
}
+ rcu_read_lock();
if (usvc->fwmark)
svc = __ip_vs_svc_fwm_find(net, usvc->af, usvc->fwmark);
else
svc = __ip_vs_service_find(net, usvc->af, usvc->protocol,
&usvc->addr, usvc->port);
+ rcu_read_unlock();
*ret_svc = svc;
/* If a full entry was requested, check for the additional fields */
@@ -3903,8 +3850,8 @@ int __init ip_vs_control_init(void)
/* Initialize svc_table, ip_vs_svc_fwm_table */
for (idx = 0; idx < IP_VS_SVC_TAB_SIZE; idx++) {
- INIT_LIST_HEAD(&ip_vs_svc_table[idx]);
- INIT_LIST_HEAD(&ip_vs_svc_fwm_table[idx]);
+ INIT_HLIST_HEAD(&ip_vs_svc_table[idx]);
+ INIT_HLIST_HEAD(&ip_vs_svc_fwm_table[idx]);
}
smp_wmb(); /* Do we really need it now ? */
diff --git a/net/netfilter/ipvs/ip_vs_dh.c b/net/netfilter/ipvs/ip_vs_dh.c
index 89c2723..ccab120 100644
--- a/net/netfilter/ipvs/ip_vs_dh.c
+++ b/net/netfilter/ipvs/ip_vs_dh.c
@@ -269,6 +269,7 @@ static int __init ip_vs_dh_init(void)
static void __exit ip_vs_dh_cleanup(void)
{
unregister_ip_vs_scheduler(&ip_vs_dh_scheduler);
+ synchronize_rcu();
}
diff --git a/net/netfilter/ipvs/ip_vs_lblc.c b/net/netfilter/ipvs/ip_vs_lblc.c
index ffef8a1..d8e5238 100644
--- a/net/netfilter/ipvs/ip_vs_lblc.c
+++ b/net/netfilter/ipvs/ip_vs_lblc.c
@@ -633,6 +633,7 @@ static void __exit ip_vs_lblc_cleanup(void)
{
unregister_ip_vs_scheduler(&ip_vs_lblc_scheduler);
unregister_pernet_subsys(&ip_vs_lblc_ops);
+ synchronize_rcu();
}
diff --git a/net/netfilter/ipvs/ip_vs_lblcr.c b/net/netfilter/ipvs/ip_vs_lblcr.c
index cdfe6a9..041b7cc 100644
--- a/net/netfilter/ipvs/ip_vs_lblcr.c
+++ b/net/netfilter/ipvs/ip_vs_lblcr.c
@@ -821,6 +821,7 @@ static void __exit ip_vs_lblcr_cleanup(void)
{
unregister_ip_vs_scheduler(&ip_vs_lblcr_scheduler);
unregister_pernet_subsys(&ip_vs_lblcr_ops);
+ synchronize_rcu();
}
diff --git a/net/netfilter/ipvs/ip_vs_lc.c b/net/netfilter/ipvs/ip_vs_lc.c
index 0cabf78..5128e33 100644
--- a/net/netfilter/ipvs/ip_vs_lc.c
+++ b/net/netfilter/ipvs/ip_vs_lc.c
@@ -84,6 +84,7 @@ static int __init ip_vs_lc_init(void)
static void __exit ip_vs_lc_cleanup(void)
{
unregister_ip_vs_scheduler(&ip_vs_lc_scheduler);
+ synchronize_rcu();
}
module_init(ip_vs_lc_init);
diff --git a/net/netfilter/ipvs/ip_vs_nq.c b/net/netfilter/ipvs/ip_vs_nq.c
index 51dc0cf..646cfd4 100644
--- a/net/netfilter/ipvs/ip_vs_nq.c
+++ b/net/netfilter/ipvs/ip_vs_nq.c
@@ -133,6 +133,7 @@ static int __init ip_vs_nq_init(void)
static void __exit ip_vs_nq_cleanup(void)
{
unregister_ip_vs_scheduler(&ip_vs_nq_scheduler);
+ synchronize_rcu();
}
module_init(ip_vs_nq_init);
diff --git a/net/netfilter/ipvs/ip_vs_pe.c b/net/netfilter/ipvs/ip_vs_pe.c
index 5d9774c..1a82b29 100644
--- a/net/netfilter/ipvs/ip_vs_pe.c
+++ b/net/netfilter/ipvs/ip_vs_pe.c
@@ -16,18 +16,6 @@ static LIST_HEAD(ip_vs_pe);
/* semaphore for IPVS PEs. */
static DEFINE_MUTEX(ip_vs_pe_mutex);
-/* Bind a service with a pe */
-void ip_vs_bind_pe(struct ip_vs_service *svc, struct ip_vs_pe *pe)
-{
- svc->pe = pe;
-}
-
-/* Unbind a service from its pe */
-void ip_vs_unbind_pe(struct ip_vs_service *svc)
-{
- svc->pe = NULL;
-}
-
/* Get pe in the pe list by name */
struct ip_vs_pe *__ip_vs_pe_getbyname(const char *pe_name)
{
diff --git a/net/netfilter/ipvs/ip_vs_proto_sctp.c b/net/netfilter/ipvs/ip_vs_proto_sctp.c
index f7190cd..4de5176 100644
--- a/net/netfilter/ipvs/ip_vs_proto_sctp.c
+++ b/net/netfilter/ipvs/ip_vs_proto_sctp.c
@@ -27,9 +27,10 @@ sctp_conn_schedule(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
if (sch == NULL)
return 0;
net = skb_net(skb);
+ rcu_read_lock();
if ((sch->type == SCTP_CID_INIT) &&
- (svc = ip_vs_service_get(net, af, skb->mark, iph->protocol,
- &iph->daddr, sh->dest))) {
+ (svc = ip_vs_service_find(net, af, skb->mark, iph->protocol,
+ &iph->daddr, sh->dest))) {
int ignored;
if (ip_vs_todrop(net_ipvs(net))) {
@@ -37,7 +38,7 @@ sctp_conn_schedule(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
* It seems that we are very loaded.
* We have to drop this packet :(
*/
- ip_vs_service_put(svc);
+ rcu_read_unlock();
*verdict = NF_DROP;
return 0;
}
@@ -49,14 +50,13 @@ sctp_conn_schedule(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
if (!*cpp && ignored <= 0) {
if (!ignored)
*verdict = ip_vs_leave(svc, skb, pd, iph);
- else {
- ip_vs_service_put(svc);
+ else
*verdict = NF_DROP;
- }
+ rcu_read_unlock();
return 0;
}
- ip_vs_service_put(svc);
}
+ rcu_read_unlock();
/* NF_ACCEPT */
return 1;
}
diff --git a/net/netfilter/ipvs/ip_vs_proto_tcp.c b/net/netfilter/ipvs/ip_vs_proto_tcp.c
index 0bbc3fe..7de3342 100644
--- a/net/netfilter/ipvs/ip_vs_proto_tcp.c
+++ b/net/netfilter/ipvs/ip_vs_proto_tcp.c
@@ -47,9 +47,10 @@ tcp_conn_schedule(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
}
net = skb_net(skb);
/* No !th->ack check to allow scheduling on SYN+ACK for Active FTP */
+ rcu_read_lock();
if (th->syn &&
- (svc = ip_vs_service_get(net, af, skb->mark, iph->protocol,
- &iph->daddr, th->dest))) {
+ (svc = ip_vs_service_find(net, af, skb->mark, iph->protocol,
+ &iph->daddr, th->dest))) {
int ignored;
if (ip_vs_todrop(net_ipvs(net))) {
@@ -57,7 +58,7 @@ tcp_conn_schedule(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
* It seems that we are very loaded.
* We have to drop this packet :(
*/
- ip_vs_service_put(svc);
+ rcu_read_unlock();
*verdict = NF_DROP;
return 0;
}
@@ -70,14 +71,13 @@ tcp_conn_schedule(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
if (!*cpp && ignored <= 0) {
if (!ignored)
*verdict = ip_vs_leave(svc, skb, pd, iph);
- else {
- ip_vs_service_put(svc);
+ else
*verdict = NF_DROP;
- }
+ rcu_read_unlock();
return 0;
}
- ip_vs_service_put(svc);
}
+ rcu_read_unlock();
/* NF_ACCEPT */
return 1;
}
diff --git a/net/netfilter/ipvs/ip_vs_proto_udp.c b/net/netfilter/ipvs/ip_vs_proto_udp.c
index 1a03e2d..b62a3c0 100644
--- a/net/netfilter/ipvs/ip_vs_proto_udp.c
+++ b/net/netfilter/ipvs/ip_vs_proto_udp.c
@@ -44,8 +44,9 @@ udp_conn_schedule(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
return 0;
}
net = skb_net(skb);
- svc = ip_vs_service_get(net, af, skb->mark, iph->protocol,
- &iph->daddr, uh->dest);
+ rcu_read_lock();
+ svc = ip_vs_service_find(net, af, skb->mark, iph->protocol,
+ &iph->daddr, uh->dest);
if (svc) {
int ignored;
@@ -54,7 +55,7 @@ udp_conn_schedule(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
* It seems that we are very loaded.
* We have to drop this packet :(
*/
- ip_vs_service_put(svc);
+ rcu_read_unlock();
*verdict = NF_DROP;
return 0;
}
@@ -67,14 +68,13 @@ udp_conn_schedule(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
if (!*cpp && ignored <= 0) {
if (!ignored)
*verdict = ip_vs_leave(svc, skb, pd, iph);
- else {
- ip_vs_service_put(svc);
+ else
*verdict = NF_DROP;
- }
+ rcu_read_unlock();
return 0;
}
- ip_vs_service_put(svc);
}
+ rcu_read_unlock();
/* NF_ACCEPT */
return 1;
}
diff --git a/net/netfilter/ipvs/ip_vs_rr.c b/net/netfilter/ipvs/ip_vs_rr.c
index aa4601f..749c98a 100644
--- a/net/netfilter/ipvs/ip_vs_rr.c
+++ b/net/netfilter/ipvs/ip_vs_rr.c
@@ -121,6 +121,7 @@ static int __init ip_vs_rr_init(void)
static void __exit ip_vs_rr_cleanup(void)
{
unregister_ip_vs_scheduler(&ip_vs_rr_scheduler);
+ synchronize_rcu();
}
module_init(ip_vs_rr_init);
diff --git a/net/netfilter/ipvs/ip_vs_sched.c b/net/netfilter/ipvs/ip_vs_sched.c
index 1b715d0..4dbcda6 100644
--- a/net/netfilter/ipvs/ip_vs_sched.c
+++ b/net/netfilter/ipvs/ip_vs_sched.c
@@ -47,8 +47,6 @@ int ip_vs_bind_scheduler(struct ip_vs_service *svc,
{
int ret;
- svc->scheduler = scheduler;
-
if (scheduler->init_service) {
ret = scheduler->init_service(svc);
if (ret) {
@@ -56,7 +54,7 @@ int ip_vs_bind_scheduler(struct ip_vs_service *svc,
return ret;
}
}
-
+ rcu_assign_pointer(svc->scheduler, scheduler);
return 0;
}
@@ -64,17 +62,19 @@ int ip_vs_bind_scheduler(struct ip_vs_service *svc,
/*
* Unbind a service with its scheduler
*/
-void ip_vs_unbind_scheduler(struct ip_vs_service *svc)
+void ip_vs_unbind_scheduler(struct ip_vs_service *svc,
+ struct ip_vs_scheduler *sched)
{
- struct ip_vs_scheduler *sched = svc->scheduler;
+ struct ip_vs_scheduler *cur_sched;
- if (!sched)
+ cur_sched = rcu_dereference_protected(svc->scheduler, 1);
+ /* This check proves that old 'sched' was installed */
+ if (!cur_sched)
return;
if (sched->done_service)
sched->done_service(svc);
-
- svc->scheduler = NULL;
+ /* svc->scheduler can not be set to NULL */
}
@@ -148,21 +148,21 @@ void ip_vs_scheduler_put(struct ip_vs_scheduler *scheduler)
void ip_vs_scheduler_err(struct ip_vs_service *svc, const char *msg)
{
+ struct ip_vs_scheduler *sched;
+
+ sched = rcu_dereference(svc->scheduler);
if (svc->fwmark) {
IP_VS_ERR_RL("%s: FWM %u 0x%08X - %s\n",
- svc->scheduler->name, svc->fwmark,
- svc->fwmark, msg);
+ sched->name, svc->fwmark, svc->fwmark, msg);
#ifdef CONFIG_IP_VS_IPV6
} else if (svc->af == AF_INET6) {
IP_VS_ERR_RL("%s: %s [%pI6c]:%d - %s\n",
- svc->scheduler->name,
- ip_vs_proto_name(svc->protocol),
+ sched->name, ip_vs_proto_name(svc->protocol),
&svc->addr.in6, ntohs(svc->port), msg);
#endif
} else {
IP_VS_ERR_RL("%s: %s %pI4:%d - %s\n",
- svc->scheduler->name,
- ip_vs_proto_name(svc->protocol),
+ sched->name, ip_vs_proto_name(svc->protocol),
&svc->addr.ip, ntohs(svc->port), msg);
}
}
diff --git a/net/netfilter/ipvs/ip_vs_sed.c b/net/netfilter/ipvs/ip_vs_sed.c
index d011870..f320592 100644
--- a/net/netfilter/ipvs/ip_vs_sed.c
+++ b/net/netfilter/ipvs/ip_vs_sed.c
@@ -134,6 +134,7 @@ static int __init ip_vs_sed_init(void)
static void __exit ip_vs_sed_cleanup(void)
{
unregister_ip_vs_scheduler(&ip_vs_sed_scheduler);
+ synchronize_rcu();
}
module_init(ip_vs_sed_init);
diff --git a/net/netfilter/ipvs/ip_vs_sh.c b/net/netfilter/ipvs/ip_vs_sh.c
index 81c1a10..0df269d 100644
--- a/net/netfilter/ipvs/ip_vs_sh.c
+++ b/net/netfilter/ipvs/ip_vs_sh.c
@@ -283,6 +283,7 @@ static int __init ip_vs_sh_init(void)
static void __exit ip_vs_sh_cleanup(void)
{
unregister_ip_vs_scheduler(&ip_vs_sh_scheduler);
+ synchronize_rcu();
}
diff --git a/net/netfilter/ipvs/ip_vs_wlc.c b/net/netfilter/ipvs/ip_vs_wlc.c
index dafae88..c60a81c 100644
--- a/net/netfilter/ipvs/ip_vs_wlc.c
+++ b/net/netfilter/ipvs/ip_vs_wlc.c
@@ -106,6 +106,7 @@ static int __init ip_vs_wlc_init(void)
static void __exit ip_vs_wlc_cleanup(void)
{
unregister_ip_vs_scheduler(&ip_vs_wlc_scheduler);
+ synchronize_rcu();
}
module_init(ip_vs_wlc_init);
diff --git a/net/netfilter/ipvs/ip_vs_wrr.c b/net/netfilter/ipvs/ip_vs_wrr.c
index b173ef9..32c646e 100644
--- a/net/netfilter/ipvs/ip_vs_wrr.c
+++ b/net/netfilter/ipvs/ip_vs_wrr.c
@@ -261,6 +261,7 @@ static int __init ip_vs_wrr_init(void)
static void __exit ip_vs_wrr_cleanup(void)
{
unregister_ip_vs_scheduler(&ip_vs_wrr_scheduler);
+ synchronize_rcu();
}
module_init(ip_vs_wrr_init);
--
1.7.3.4
^ permalink raw reply related [flat|nested] 23+ messages in thread* [PATCH net-next 19/19] ipvs: do not disable bh for long time
2013-03-22 9:46 [PATCH net-next 00/19] IPVS optimizations, part 2 Julian Anastasov
` (17 preceding siblings ...)
2013-03-22 9:46 ` [PATCH net-next 18/19] ipvs: convert services " Julian Anastasov
@ 2013-03-22 9:46 ` Julian Anastasov
2013-03-28 5:41 ` [PATCH net-next 00/19] IPVS optimizations, part 2 Simon Horman
19 siblings, 0 replies; 23+ messages in thread
From: Julian Anastasov @ 2013-03-22 9:46 UTC (permalink / raw)
To: Simon Horman; +Cc: lvs-devel, netdev
We used a global BH disable in LOCAL_OUT hook.
Add _bh suffix to all places that need it and remove
the disabling from LOCAL_OUT and sync code.
Functions like ip_defrag need protection from
BH, so add it. As for nf_nat_mangle_tcp_packet, it needs
RCU lock.
Signed-off-by: Julian Anastasov <ja@ssi.bg>
---
net/netfilter/ipvs/ip_vs_app.c | 4 +-
net/netfilter/ipvs/ip_vs_conn.c | 30 +++++++++++++-------------
net/netfilter/ipvs/ip_vs_core.c | 37 +++++++-------------------------
net/netfilter/ipvs/ip_vs_ftp.c | 2 +
net/netfilter/ipvs/ip_vs_lblc.c | 4 +-
net/netfilter/ipvs/ip_vs_lblcr.c | 12 +++++-----
net/netfilter/ipvs/ip_vs_proto_sctp.c | 4 +-
net/netfilter/ipvs/ip_vs_proto_tcp.c | 8 +++---
net/netfilter/ipvs/ip_vs_rr.c | 6 ++--
net/netfilter/ipvs/ip_vs_sync.c | 24 +++++++++------------
net/netfilter/ipvs/ip_vs_wrr.c | 4 +-
net/netfilter/ipvs/ip_vs_xmit.c | 16 +++++++-------
12 files changed, 64 insertions(+), 87 deletions(-)
diff --git a/net/netfilter/ipvs/ip_vs_app.c b/net/netfilter/ipvs/ip_vs_app.c
index a956030..dfd7b65 100644
--- a/net/netfilter/ipvs/ip_vs_app.c
+++ b/net/netfilter/ipvs/ip_vs_app.c
@@ -352,14 +352,14 @@ static inline void vs_seq_update(struct ip_vs_conn *cp, struct ip_vs_seq *vseq,
unsigned int flag, __u32 seq, int diff)
{
/* spinlock is to keep updating cp->flags atomic */
- spin_lock(&cp->lock);
+ spin_lock_bh(&cp->lock);
if (!(cp->flags & flag) || after(seq, vseq->init_seq)) {
vseq->previous_delta = vseq->delta;
vseq->delta += diff;
vseq->init_seq = seq;
cp->flags |= flag;
}
- spin_unlock(&cp->lock);
+ spin_unlock_bh(&cp->lock);
}
static inline int app_tcp_pkt_out(struct ip_vs_conn *cp, struct sk_buff *skb,
diff --git a/net/netfilter/ipvs/ip_vs_conn.c b/net/netfilter/ipvs/ip_vs_conn.c
index 54de340..de64758 100644
--- a/net/netfilter/ipvs/ip_vs_conn.c
+++ b/net/netfilter/ipvs/ip_vs_conn.c
@@ -86,14 +86,14 @@ struct ip_vs_aligned_lock
static struct ip_vs_aligned_lock
__ip_vs_conntbl_lock_array[CT_LOCKARRAY_SIZE] __cacheline_aligned;
-static inline void ct_write_lock(unsigned int key)
+static inline void ct_write_lock_bh(unsigned int key)
{
- spin_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
+ spin_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
}
-static inline void ct_write_unlock(unsigned int key)
+static inline void ct_write_unlock_bh(unsigned int key)
{
- spin_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
+ spin_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
}
@@ -167,7 +167,7 @@ static inline int ip_vs_conn_hash(struct ip_vs_conn *cp)
/* Hash by protocol, client address and port */
hash = ip_vs_conn_hashkey_conn(cp);
- ct_write_lock(hash);
+ ct_write_lock_bh(hash);
spin_lock(&cp->lock);
if (!(cp->flags & IP_VS_CONN_F_HASHED)) {
@@ -182,7 +182,7 @@ static inline int ip_vs_conn_hash(struct ip_vs_conn *cp)
}
spin_unlock(&cp->lock);
- ct_write_unlock(hash);
+ ct_write_unlock_bh(hash);
return ret;
}
@@ -200,7 +200,7 @@ static inline int ip_vs_conn_unhash(struct ip_vs_conn *cp)
/* unhash it and decrease its reference counter */
hash = ip_vs_conn_hashkey_conn(cp);
- ct_write_lock(hash);
+ ct_write_lock_bh(hash);
spin_lock(&cp->lock);
if (cp->flags & IP_VS_CONN_F_HASHED) {
@@ -212,7 +212,7 @@ static inline int ip_vs_conn_unhash(struct ip_vs_conn *cp)
ret = 0;
spin_unlock(&cp->lock);
- ct_write_unlock(hash);
+ ct_write_unlock_bh(hash);
return ret;
}
@@ -227,7 +227,7 @@ static inline bool ip_vs_conn_unlink(struct ip_vs_conn *cp)
hash = ip_vs_conn_hashkey_conn(cp);
- ct_write_lock(hash);
+ ct_write_lock_bh(hash);
spin_lock(&cp->lock);
if (cp->flags & IP_VS_CONN_F_HASHED) {
@@ -242,7 +242,7 @@ static inline bool ip_vs_conn_unlink(struct ip_vs_conn *cp)
ret = atomic_read(&cp->refcnt) ? false : true;
spin_unlock(&cp->lock);
- ct_write_unlock(hash);
+ ct_write_unlock_bh(hash);
return ret;
}
@@ -462,13 +462,13 @@ void ip_vs_conn_put(struct ip_vs_conn *cp)
void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport)
{
if (ip_vs_conn_unhash(cp)) {
- spin_lock(&cp->lock);
+ spin_lock_bh(&cp->lock);
if (cp->flags & IP_VS_CONN_F_NO_CPORT) {
atomic_dec(&ip_vs_conn_no_cport_cnt);
cp->flags &= ~IP_VS_CONN_F_NO_CPORT;
cp->cport = cport;
}
- spin_unlock(&cp->lock);
+ spin_unlock_bh(&cp->lock);
/* hash on new dport */
ip_vs_conn_hash(cp);
@@ -622,9 +622,9 @@ void ip_vs_try_bind_dest(struct ip_vs_conn *cp)
if (dest) {
struct ip_vs_proto_data *pd;
- spin_lock(&cp->lock);
+ spin_lock_bh(&cp->lock);
if (cp->dest) {
- spin_unlock(&cp->lock);
+ spin_unlock_bh(&cp->lock);
rcu_read_unlock();
return;
}
@@ -635,7 +635,7 @@ void ip_vs_try_bind_dest(struct ip_vs_conn *cp)
ip_vs_unbind_app(cp);
ip_vs_bind_dest(cp, dest);
- spin_unlock(&cp->lock);
+ spin_unlock_bh(&cp->lock);
/* Update its packet transmitter */
cp->packet_xmit = NULL;
diff --git a/net/netfilter/ipvs/ip_vs_core.c b/net/netfilter/ipvs/ip_vs_core.c
index 8814f02..1ac56c7 100644
--- a/net/netfilter/ipvs/ip_vs_core.c
+++ b/net/netfilter/ipvs/ip_vs_core.c
@@ -641,8 +641,11 @@ static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
static inline int ip_vs_gather_frags(struct sk_buff *skb, u_int32_t user)
{
- int err = ip_defrag(skb, user);
+ int err;
+ local_bh_disable();
+ err = ip_defrag(skb, user);
+ local_bh_enable();
if (!err)
ip_send_check(ip_hdr(skb));
@@ -1223,13 +1226,7 @@ ip_vs_local_reply4(unsigned int hooknum, struct sk_buff *skb,
const struct net_device *in, const struct net_device *out,
int (*okfn)(struct sk_buff *))
{
- unsigned int verdict;
-
- /* Disable BH in LOCAL_OUT until all places are fixed */
- local_bh_disable();
- verdict = ip_vs_out(hooknum, skb, AF_INET);
- local_bh_enable();
- return verdict;
+ return ip_vs_out(hooknum, skb, AF_INET);
}
#ifdef CONFIG_IP_VS_IPV6
@@ -1256,13 +1253,7 @@ ip_vs_local_reply6(unsigned int hooknum, struct sk_buff *skb,
const struct net_device *in, const struct net_device *out,
int (*okfn)(struct sk_buff *))
{
- unsigned int verdict;
-
- /* Disable BH in LOCAL_OUT until all places are fixed */
- local_bh_disable();
- verdict = ip_vs_out(hooknum, skb, AF_INET6);
- local_bh_enable();
- return verdict;
+ return ip_vs_out(hooknum, skb, AF_INET6);
}
#endif
@@ -1720,13 +1711,7 @@ ip_vs_local_request4(unsigned int hooknum, struct sk_buff *skb,
const struct net_device *in, const struct net_device *out,
int (*okfn)(struct sk_buff *))
{
- unsigned int verdict;
-
- /* Disable BH in LOCAL_OUT until all places are fixed */
- local_bh_disable();
- verdict = ip_vs_in(hooknum, skb, AF_INET);
- local_bh_enable();
- return verdict;
+ return ip_vs_in(hooknum, skb, AF_INET);
}
#ifdef CONFIG_IP_VS_IPV6
@@ -1785,13 +1770,7 @@ ip_vs_local_request6(unsigned int hooknum, struct sk_buff *skb,
const struct net_device *in, const struct net_device *out,
int (*okfn)(struct sk_buff *))
{
- unsigned int verdict;
-
- /* Disable BH in LOCAL_OUT until all places are fixed */
- local_bh_disable();
- verdict = ip_vs_in(hooknum, skb, AF_INET6);
- local_bh_enable();
- return verdict;
+ return ip_vs_in(hooknum, skb, AF_INET6);
}
#endif
diff --git a/net/netfilter/ipvs/ip_vs_ftp.c b/net/netfilter/ipvs/ip_vs_ftp.c
index 7f90825..77c1732 100644
--- a/net/netfilter/ipvs/ip_vs_ftp.c
+++ b/net/netfilter/ipvs/ip_vs_ftp.c
@@ -267,10 +267,12 @@ static int ip_vs_ftp_out(struct ip_vs_app *app, struct ip_vs_conn *cp,
* hopefully it will succeed on the retransmitted
* packet.
*/
+ rcu_read_lock();
ret = nf_nat_mangle_tcp_packet(skb, ct, ctinfo,
iph->ihl * 4,
start-data, end-start,
buf, buf_len);
+ rcu_read_unlock();
if (ret) {
ip_vs_nfct_expect_related(skb, ct, n_cp,
IPPROTO_TCP, 0, 0);
diff --git a/net/netfilter/ipvs/ip_vs_lblc.c b/net/netfilter/ipvs/ip_vs_lblc.c
index d8e5238..b2cc252 100644
--- a/net/netfilter/ipvs/ip_vs_lblc.c
+++ b/net/netfilter/ipvs/ip_vs_lblc.c
@@ -527,10 +527,10 @@ ip_vs_lblc_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
}
/* If we fail to create a cache entry, we'll just use the valid dest */
- spin_lock(&svc->sched_lock);
+ spin_lock_bh(&svc->sched_lock);
if (!tbl->dead)
ip_vs_lblc_new(tbl, &iph.daddr, dest);
- spin_unlock(&svc->sched_lock);
+ spin_unlock_bh(&svc->sched_lock);
out:
IP_VS_DBG_BUF(6, "LBLC: destination IP address %s --> server %s:%d\n",
diff --git a/net/netfilter/ipvs/ip_vs_lblcr.c b/net/netfilter/ipvs/ip_vs_lblcr.c
index 041b7cc..feb9656 100644
--- a/net/netfilter/ipvs/ip_vs_lblcr.c
+++ b/net/netfilter/ipvs/ip_vs_lblcr.c
@@ -678,7 +678,7 @@ ip_vs_lblcr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
if (atomic_read(&en->set.size) > 1 &&
time_after(jiffies, en->set.lastmod +
sysctl_lblcr_expiration(svc))) {
- spin_lock(&svc->sched_lock);
+ spin_lock_bh(&svc->sched_lock);
if (atomic_read(&en->set.size) > 1) {
struct ip_vs_dest *m;
@@ -686,7 +686,7 @@ ip_vs_lblcr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
if (m)
ip_vs_dest_set_erase(&en->set, m);
}
- spin_unlock(&svc->sched_lock);
+ spin_unlock_bh(&svc->sched_lock);
}
/* If the destination is not overloaded, use it */
@@ -701,10 +701,10 @@ ip_vs_lblcr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
}
/* Update our cache entry */
- spin_lock(&svc->sched_lock);
+ spin_lock_bh(&svc->sched_lock);
if (!tbl->dead)
ip_vs_dest_set_insert(&en->set, dest, true);
- spin_unlock(&svc->sched_lock);
+ spin_unlock_bh(&svc->sched_lock);
goto out;
}
@@ -716,10 +716,10 @@ ip_vs_lblcr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
}
/* If we fail to create a cache entry, we'll just use the valid dest */
- spin_lock(&svc->sched_lock);
+ spin_lock_bh(&svc->sched_lock);
if (!tbl->dead)
ip_vs_lblcr_new(tbl, &iph.daddr, dest);
- spin_unlock(&svc->sched_lock);
+ spin_unlock_bh(&svc->sched_lock);
out:
IP_VS_DBG_BUF(6, "LBLCR: destination IP address %s --> server %s:%d\n",
diff --git a/net/netfilter/ipvs/ip_vs_proto_sctp.c b/net/netfilter/ipvs/ip_vs_proto_sctp.c
index 4de5176..6e14a7b 100644
--- a/net/netfilter/ipvs/ip_vs_proto_sctp.c
+++ b/net/netfilter/ipvs/ip_vs_proto_sctp.c
@@ -994,9 +994,9 @@ static void
sctp_state_transition(struct ip_vs_conn *cp, int direction,
const struct sk_buff *skb, struct ip_vs_proto_data *pd)
{
- spin_lock(&cp->lock);
+ spin_lock_bh(&cp->lock);
set_sctp_state(pd, cp, direction, skb);
- spin_unlock(&cp->lock);
+ spin_unlock_bh(&cp->lock);
}
static inline __u16 sctp_app_hashkey(__be16 port)
diff --git a/net/netfilter/ipvs/ip_vs_proto_tcp.c b/net/netfilter/ipvs/ip_vs_proto_tcp.c
index 7de3342..50a1594 100644
--- a/net/netfilter/ipvs/ip_vs_proto_tcp.c
+++ b/net/netfilter/ipvs/ip_vs_proto_tcp.c
@@ -557,9 +557,9 @@ tcp_state_transition(struct ip_vs_conn *cp, int direction,
if (th == NULL)
return;
- spin_lock(&cp->lock);
+ spin_lock_bh(&cp->lock);
set_tcp_state(pd, cp, direction, th);
- spin_unlock(&cp->lock);
+ spin_unlock_bh(&cp->lock);
}
static inline __u16 tcp_app_hashkey(__be16 port)
@@ -655,11 +655,11 @@ void ip_vs_tcp_conn_listen(struct net *net, struct ip_vs_conn *cp)
{
struct ip_vs_proto_data *pd = ip_vs_proto_data_get(net, IPPROTO_TCP);
- spin_lock(&cp->lock);
+ spin_lock_bh(&cp->lock);
cp->state = IP_VS_TCP_S_LISTEN;
cp->timeout = (pd ? pd->timeout_table[IP_VS_TCP_S_LISTEN]
: tcp_timeouts[IP_VS_TCP_S_LISTEN]);
- spin_unlock(&cp->lock);
+ spin_unlock_bh(&cp->lock);
}
/* ---------------------------------------------
diff --git a/net/netfilter/ipvs/ip_vs_rr.c b/net/netfilter/ipvs/ip_vs_rr.c
index 749c98a..c35986c 100644
--- a/net/netfilter/ipvs/ip_vs_rr.c
+++ b/net/netfilter/ipvs/ip_vs_rr.c
@@ -63,7 +63,7 @@ ip_vs_rr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
IP_VS_DBG(6, "%s(): Scheduling...\n", __func__);
- spin_lock(&svc->sched_lock);
+ spin_lock_bh(&svc->sched_lock);
p = (struct list_head *) svc->sched_data;
last = dest = list_entry(p, struct ip_vs_dest, n_list);
@@ -85,13 +85,13 @@ ip_vs_rr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
} while (pass < 2 && p != &svc->destinations);
stop:
- spin_unlock(&svc->sched_lock);
+ spin_unlock_bh(&svc->sched_lock);
ip_vs_scheduler_err(svc, "no destination available");
return NULL;
out:
svc->sched_data = &dest->n_list;
- spin_unlock(&svc->sched_lock);
+ spin_unlock_bh(&svc->sched_lock);
IP_VS_DBG_BUF(6, "RR: server %s:%u "
"activeconns %d refcnt %d weight %d\n",
IP_VS_DBG_ADDR(svc->af, &dest->addr), ntohs(dest->port),
diff --git a/net/netfilter/ipvs/ip_vs_sync.c b/net/netfilter/ipvs/ip_vs_sync.c
index 9724174..8e57077 100644
--- a/net/netfilter/ipvs/ip_vs_sync.c
+++ b/net/netfilter/ipvs/ip_vs_sync.c
@@ -531,9 +531,9 @@ static void ip_vs_sync_conn_v0(struct net *net, struct ip_vs_conn *cp,
if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
return;
- spin_lock(&ipvs->sync_buff_lock);
+ spin_lock_bh(&ipvs->sync_buff_lock);
if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
- spin_unlock(&ipvs->sync_buff_lock);
+ spin_unlock_bh(&ipvs->sync_buff_lock);
return;
}
@@ -552,7 +552,7 @@ static void ip_vs_sync_conn_v0(struct net *net, struct ip_vs_conn *cp,
if (!buff) {
buff = ip_vs_sync_buff_create_v0(ipvs);
if (!buff) {
- spin_unlock(&ipvs->sync_buff_lock);
+ spin_unlock_bh(&ipvs->sync_buff_lock);
pr_err("ip_vs_sync_buff_create failed.\n");
return;
}
@@ -590,7 +590,7 @@ static void ip_vs_sync_conn_v0(struct net *net, struct ip_vs_conn *cp,
sb_queue_tail(ipvs, ms);
ms->sync_buff = NULL;
}
- spin_unlock(&ipvs->sync_buff_lock);
+ spin_unlock_bh(&ipvs->sync_buff_lock);
/* synchronize its controller if it has */
cp = cp->control;
@@ -641,9 +641,9 @@ sloop:
pe_name_len = strnlen(cp->pe->name, IP_VS_PENAME_MAXLEN);
}
- spin_lock(&ipvs->sync_buff_lock);
+ spin_lock_bh(&ipvs->sync_buff_lock);
if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
- spin_unlock(&ipvs->sync_buff_lock);
+ spin_unlock_bh(&ipvs->sync_buff_lock);
return;
}
@@ -683,7 +683,7 @@ sloop:
if (!buff) {
buff = ip_vs_sync_buff_create(ipvs);
if (!buff) {
- spin_unlock(&ipvs->sync_buff_lock);
+ spin_unlock_bh(&ipvs->sync_buff_lock);
pr_err("ip_vs_sync_buff_create failed.\n");
return;
}
@@ -750,7 +750,7 @@ sloop:
}
}
- spin_unlock(&ipvs->sync_buff_lock);
+ spin_unlock_bh(&ipvs->sync_buff_lock);
control:
/* synchronize its controller if it has */
@@ -843,7 +843,7 @@ static void ip_vs_proc_conn(struct net *net, struct ip_vs_conn_param *param,
kfree(param->pe_data);
dest = cp->dest;
- spin_lock(&cp->lock);
+ spin_lock_bh(&cp->lock);
if ((cp->flags ^ flags) & IP_VS_CONN_F_INACTIVE &&
!(flags & IP_VS_CONN_F_TEMPLATE) && dest) {
if (flags & IP_VS_CONN_F_INACTIVE) {
@@ -857,7 +857,7 @@ static void ip_vs_proc_conn(struct net *net, struct ip_vs_conn_param *param,
flags &= IP_VS_CONN_F_BACKUP_UPD_MASK;
flags |= cp->flags & ~IP_VS_CONN_F_BACKUP_UPD_MASK;
cp->flags = flags;
- spin_unlock(&cp->lock);
+ spin_unlock_bh(&cp->lock);
if (!dest)
ip_vs_try_bind_dest(cp);
} else {
@@ -1689,11 +1689,7 @@ static int sync_thread_backup(void *data)
break;
}
- /* disable bottom half, because it accesses the data
- shared by softirq while getting/creating conns */
- local_bh_disable();
ip_vs_process_message(tinfo->net, tinfo->buf, len);
- local_bh_enable();
}
}
diff --git a/net/netfilter/ipvs/ip_vs_wrr.c b/net/netfilter/ipvs/ip_vs_wrr.c
index 32c646e..0e68555 100644
--- a/net/netfilter/ipvs/ip_vs_wrr.c
+++ b/net/netfilter/ipvs/ip_vs_wrr.c
@@ -170,7 +170,7 @@ ip_vs_wrr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
IP_VS_DBG(6, "%s(): Scheduling...\n", __func__);
- spin_lock(&svc->sched_lock);
+ spin_lock_bh(&svc->sched_lock);
dest = mark->cl;
/* No available dests? */
if (mark->mw == 0)
@@ -222,7 +222,7 @@ found:
mark->cl = dest;
out:
- spin_unlock(&svc->sched_lock);
+ spin_unlock_bh(&svc->sched_lock);
return dest;
err_noavail:
diff --git a/net/netfilter/ipvs/ip_vs_xmit.c b/net/netfilter/ipvs/ip_vs_xmit.c
index 3db7889..b75ff64 100644
--- a/net/netfilter/ipvs/ip_vs_xmit.c
+++ b/net/netfilter/ipvs/ip_vs_xmit.c
@@ -177,22 +177,22 @@ __ip_vs_get_out_rt(struct sk_buff *skb, struct ip_vs_dest *dest,
rt = (struct rtable *) dest_dst->dst_cache;
else {
dest_dst = ip_vs_dest_dst_alloc();
- spin_lock(&dest->dst_lock);
+ spin_lock_bh(&dest->dst_lock);
if (!dest_dst) {
__ip_vs_dst_set(dest, NULL, NULL, 0);
- spin_unlock(&dest->dst_lock);
+ spin_unlock_bh(&dest->dst_lock);
goto err_unreach;
}
rt = do_output_route4(net, dest->addr.ip, rt_mode,
&dest_dst->dst_saddr.ip);
if (!rt) {
__ip_vs_dst_set(dest, NULL, NULL, 0);
- spin_unlock(&dest->dst_lock);
+ spin_unlock_bh(&dest->dst_lock);
ip_vs_dest_dst_free(dest_dst);
goto err_unreach;
}
__ip_vs_dst_set(dest, dest_dst, &rt->dst, 0);
- spin_unlock(&dest->dst_lock);
+ spin_unlock_bh(&dest->dst_lock);
IP_VS_DBG(10, "new dst %pI4, src %pI4, refcnt=%d\n",
&dest->addr.ip, &dest_dst->dst_saddr.ip,
atomic_read(&rt->dst.__refcnt));
@@ -358,10 +358,10 @@ __ip_vs_get_out_rt_v6(struct sk_buff *skb, struct ip_vs_dest *dest,
u32 cookie;
dest_dst = ip_vs_dest_dst_alloc();
- spin_lock(&dest->dst_lock);
+ spin_lock_bh(&dest->dst_lock);
if (!dest_dst) {
__ip_vs_dst_set(dest, NULL, NULL, 0);
- spin_unlock(&dest->dst_lock);
+ spin_unlock_bh(&dest->dst_lock);
goto err_unreach;
}
dst = __ip_vs_route_output_v6(net, &dest->addr.in6,
@@ -369,14 +369,14 @@ __ip_vs_get_out_rt_v6(struct sk_buff *skb, struct ip_vs_dest *dest,
do_xfrm);
if (!dst) {
__ip_vs_dst_set(dest, NULL, NULL, 0);
- spin_unlock(&dest->dst_lock);
+ spin_unlock_bh(&dest->dst_lock);
ip_vs_dest_dst_free(dest_dst);
goto err_unreach;
}
rt = (struct rt6_info *) dst;
cookie = rt->rt6i_node ? rt->rt6i_node->fn_sernum : 0;
__ip_vs_dst_set(dest, dest_dst, &rt->dst, cookie);
- spin_unlock(&dest->dst_lock);
+ spin_unlock_bh(&dest->dst_lock);
IP_VS_DBG(10, "new dst %pI6, src %pI6, refcnt=%d\n",
&dest->addr.in6, &dest_dst->dst_saddr.in6,
atomic_read(&rt->dst.__refcnt));
--
1.7.3.4
^ permalink raw reply related [flat|nested] 23+ messages in thread* Re: [PATCH net-next 00/19] IPVS optimizations, part 2
2013-03-22 9:46 [PATCH net-next 00/19] IPVS optimizations, part 2 Julian Anastasov
` (18 preceding siblings ...)
2013-03-22 9:46 ` [PATCH net-next 19/19] ipvs: do not disable bh for long time Julian Anastasov
@ 2013-03-28 5:41 ` Simon Horman
2013-03-28 8:57 ` Julian Anastasov
19 siblings, 1 reply; 23+ messages in thread
From: Simon Horman @ 2013-03-28 5:41 UTC (permalink / raw)
To: Julian Anastasov; +Cc: lvs-devel, netdev
On Fri, Mar 22, 2013 at 11:46:35AM +0200, Julian Anastasov wrote:
> This is the second patchset with IPVS optimizations.
> Now we convert the schedulers, dests and services to RCU.
>
> All patches are for net-next based on the first
> patchset v3. The idea is after discussion and review Simon to
> apply the patchset after a week or so to ipvs-next tree.
>
> The changes in this patchset eliminate global locks
> from packet processing by using RCU. There are more details
> in the patches.
>
> After this patchset the situation is as follows:
>
> - dests:
> - lookups under RCU lock allow ip_vs_dest_hold, used
> for binding dest to conn or to select dest by scheduler
> - dests are freed by dest_trash code long after grace period
>
> - services:
> - no global read_lock
> - lookups under RCU lock allow scheduler to select
> dests under RCU lock
> - grace period implemented with IP_VS_WAIT_WHILE is
> gone allowing scheduler's dest selection and scheduler
> reconfiguration to run in parallel
>
> - schedulers:
> - schedule method runs under RCU lock, needs _rcu
> if using svc->destinations, needs _bh suffix to locks
> because it can be called in LOCAL_OUT hook
> - when dest is added, the add_dest method is called
> instead of update_service
> - when dest is deleted, the del_dest method is called
> instead of update_service
> - when dest is updated, the upd_dest method is called
> instead of update_service
> - scheduler can hold dests in its state long after they are
> unlinked from svc, even without providing del_dest handler.
> But such dests must not be returned by the
> schedule method (needs IP_VS_DEST_F_AVAILABLE check)
> - sched_data must be freed after grace period and
> module exit should be delayed with synchronize_rcu
> to wait all RCU read-side critical sections to
> complete
>
> - BH:
> - we do not disable BHs in LOCAL_OUT and sync code anymore
> - _bh suffixes are added to all places that need them,
> except timer handlers
Hi Julian,
what is the status of this series?
N.B: THis series is different to "[PATCH 00/15 v3] IPVS optimizations (repost)"
^ permalink raw reply [flat|nested] 23+ messages in thread* Re: [PATCH net-next 00/19] IPVS optimizations, part 2
2013-03-28 5:41 ` [PATCH net-next 00/19] IPVS optimizations, part 2 Simon Horman
@ 2013-03-28 8:57 ` Julian Anastasov
2013-03-28 9:05 ` Simon Horman
0 siblings, 1 reply; 23+ messages in thread
From: Julian Anastasov @ 2013-03-28 8:57 UTC (permalink / raw)
To: Simon Horman; +Cc: lvs-devel, netdev
Hello,
On Thu, 28 Mar 2013, Simon Horman wrote:
> On Fri, Mar 22, 2013 at 11:46:35AM +0200, Julian Anastasov wrote:
> > This is the second patchset with IPVS optimizations.
> > Now we convert the schedulers, dests and services to RCU.
> >
> > All patches are for net-next based on the first
> > patchset v3. The idea is after discussion and review Simon to
> > apply the patchset after a week or so to ipvs-next tree.
> >
> > The changes in this patchset eliminate global locks
> > from packet processing by using RCU. There are more details
> > in the patches.
> >
> > After this patchset the situation is as follows:
> >
> > - dests:
> > - lookups under RCU lock allow ip_vs_dest_hold, used
> > for binding dest to conn or to select dest by scheduler
> > - dests are freed by dest_trash code long after grace period
> >
> > - services:
> > - no global read_lock
> > - lookups under RCU lock allow scheduler to select
> > dests under RCU lock
> > - grace period implemented with IP_VS_WAIT_WHILE is
> > gone allowing scheduler's dest selection and scheduler
> > reconfiguration to run in parallel
> >
> > - schedulers:
> > - schedule method runs under RCU lock, needs _rcu
> > if using svc->destinations, needs _bh suffix to locks
> > because it can be called in LOCAL_OUT hook
> > - when dest is added, the add_dest method is called
> > instead of update_service
> > - when dest is deleted, the del_dest method is called
> > instead of update_service
> > - when dest is updated, the upd_dest method is called
> > instead of update_service
> > - scheduler can hold dests in its state long after they are
> > unlinked from svc, even without providing del_dest handler.
> > But such dests must not be returned by the
> > schedule method (needs IP_VS_DEST_F_AVAILABLE check)
> > - sched_data must be freed after grace period and
> > module exit should be delayed with synchronize_rcu
> > to wait all RCU read-side critical sections to
> > complete
> >
> > - BH:
> > - we do not disable BHs in LOCAL_OUT and sync code anymore
> > - _bh suffixes are added to all places that need them,
> > except timer handlers
>
> Hi Julian,
>
> what is the status of this series?
>
> N.B: THis series is different to "[PATCH 00/15 v3] IPVS optimizations (repost)"
Yes, I posted 2 parts with optimizations. Part 1 is
at v3 while part 2 is at its first version. Both series
are ready for applying. BTW, Hans plans more tests in
the next days, may be we will have some numbers for
comparison.
Regards
--
Julian Anastasov <ja@ssi.bg>
^ permalink raw reply [flat|nested] 23+ messages in thread
* Re: [PATCH net-next 00/19] IPVS optimizations, part 2
2013-03-28 8:57 ` Julian Anastasov
@ 2013-03-28 9:05 ` Simon Horman
0 siblings, 0 replies; 23+ messages in thread
From: Simon Horman @ 2013-03-28 9:05 UTC (permalink / raw)
To: Julian Anastasov; +Cc: lvs-devel, netdev
On Thu, Mar 28, 2013 at 10:57:12AM +0200, Julian Anastasov wrote:
>
> Hello,
>
> On Thu, 28 Mar 2013, Simon Horman wrote:
>
> > On Fri, Mar 22, 2013 at 11:46:35AM +0200, Julian Anastasov wrote:
> > > This is the second patchset with IPVS optimizations.
> > > Now we convert the schedulers, dests and services to RCU.
> > >
> > > All patches are for net-next based on the first
> > > patchset v3. The idea is after discussion and review Simon to
> > > apply the patchset after a week or so to ipvs-next tree.
> > >
> > > The changes in this patchset eliminate global locks
> > > from packet processing by using RCU. There are more details
> > > in the patches.
> > >
> > > After this patchset the situation is as follows:
> > >
> > > - dests:
> > > - lookups under RCU lock allow ip_vs_dest_hold, used
> > > for binding dest to conn or to select dest by scheduler
> > > - dests are freed by dest_trash code long after grace period
> > >
> > > - services:
> > > - no global read_lock
> > > - lookups under RCU lock allow scheduler to select
> > > dests under RCU lock
> > > - grace period implemented with IP_VS_WAIT_WHILE is
> > > gone allowing scheduler's dest selection and scheduler
> > > reconfiguration to run in parallel
> > >
> > > - schedulers:
> > > - schedule method runs under RCU lock, needs _rcu
> > > if using svc->destinations, needs _bh suffix to locks
> > > because it can be called in LOCAL_OUT hook
> > > - when dest is added, the add_dest method is called
> > > instead of update_service
> > > - when dest is deleted, the del_dest method is called
> > > instead of update_service
> > > - when dest is updated, the upd_dest method is called
> > > instead of update_service
> > > - scheduler can hold dests in its state long after they are
> > > unlinked from svc, even without providing del_dest handler.
> > > But such dests must not be returned by the
> > > schedule method (needs IP_VS_DEST_F_AVAILABLE check)
> > > - sched_data must be freed after grace period and
> > > module exit should be delayed with synchronize_rcu
> > > to wait all RCU read-side critical sections to
> > > complete
> > >
> > > - BH:
> > > - we do not disable BHs in LOCAL_OUT and sync code anymore
> > > - _bh suffixes are added to all places that need them,
> > > except timer handlers
> >
> > Hi Julian,
> >
> > what is the status of this series?
> >
> > N.B: THis series is different to "[PATCH 00/15 v3] IPVS optimizations (repost)"
>
> Yes, I posted 2 parts with optimizations. Part 1 is
> at v3 while part 2 is at its first version. Both series
> are ready for applying. BTW, Hans plans more tests in
> the next days, may be we will have some numbers for
> comparison.
Thanks, I'll apply these changes.
^ permalink raw reply [flat|nested] 23+ messages in thread