2005-04-17 02:20:36 +04:00
|
|
|
/*
|
|
|
|
* INETPEER - A storage for permanent information about peers
|
|
|
|
*
|
|
|
|
* This source is covered by the GNU GPL, the same as all kernel sources.
|
|
|
|
*
|
|
|
|
* Authors: Andrey V. Savochkin <saw@msu.ru>
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <linux/module.h>
|
|
|
|
#include <linux/types.h>
|
|
|
|
#include <linux/slab.h>
|
|
|
|
#include <linux/interrupt.h>
|
|
|
|
#include <linux/spinlock.h>
|
|
|
|
#include <linux/random.h>
|
|
|
|
#include <linux/timer.h>
|
|
|
|
#include <linux/time.h>
|
|
|
|
#include <linux/kernel.h>
|
|
|
|
#include <linux/mm.h>
|
|
|
|
#include <linux/net.h>
|
2012-03-07 01:20:26 +04:00
|
|
|
#include <linux/workqueue.h>
|
2005-08-16 09:18:02 +04:00
|
|
|
#include <net/ip.h>
|
2005-04-17 02:20:36 +04:00
|
|
|
#include <net/inetpeer.h>
|
2011-08-04 07:50:44 +04:00
|
|
|
#include <net/secure_seq.h>
|
2005-04-17 02:20:36 +04:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Theory of operations.
|
|
|
|
* We keep one entry for each peer IP address. The nodes contains long-living
|
|
|
|
* information about the peer which doesn't depend on routes.
|
|
|
|
*
|
|
|
|
* Nodes are removed only when reference counter goes to 0.
|
|
|
|
* When it's happened the node may be removed when a sufficient amount of
|
|
|
|
* time has been passed since its last use. The less-recently-used entry can
|
|
|
|
* also be removed if the pool is overloaded i.e. if the total amount of
|
|
|
|
* entries is greater-or-equal than the threshold.
|
|
|
|
*
|
|
|
|
* Node pool is organised as an AVL tree.
|
|
|
|
* Such an implementation has been chosen not just for fun. It's a way to
|
|
|
|
* prevent easy and efficient DoS attacks by creating hash collisions. A huge
|
|
|
|
* amount of long living nodes in a single hash slot would significantly delay
|
|
|
|
* lookups performed with disabled BHs.
|
|
|
|
*
|
|
|
|
* Serialisation issues.
|
inetpeer: RCU conversion
inetpeer currently uses an AVL tree protected by an rwlock.
It's possible to make most lookups use RCU
1) Add a struct rcu_head to struct inet_peer
2) add a lookup_rcu_bh() helper to perform lockless and opportunistic
lookup. This is a normal function, not a macro like lookup().
3) Add a limit to number of links followed by lookup_rcu_bh(). This is
needed in case we fall in a loop.
4) add an smp_wmb() in link_to_pool() right before node insert.
5) make unlink_from_pool() use atomic_cmpxchg() to make sure it can take
last reference to an inet_peer, since lockless readers could increase
refcount, even while we hold peers.lock.
6) Delay struct inet_peer freeing after rcu grace period so that
lookup_rcu_bh() cannot crash.
7) inet_getpeer() first attempts lockless lookup.
Note this lookup can fail even if target is in AVL tree, but a
concurrent writer can let tree in a non correct form.
If this attemps fails, lock is taken a regular lookup is performed
again.
8) convert peers.lock from rwlock to a spinlock
9) Remove SLAB_HWCACHE_ALIGN when peer_cachep is created, because
rcu_head adds 16 bytes on 64bit arches, doubling effective size (64 ->
128 bytes)
In a future patch, this is probably possible to revert this part, if rcu
field is put in an union to share space with rid, ip_id_count, tcp_ts &
tcp_ts_stamp. These fields being manipulated only with refcnt > 0.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-15 12:23:14 +04:00
|
|
|
* 1. Nodes may appear in the tree only with the pool lock held.
|
|
|
|
* 2. Nodes may disappear from the tree only with the pool lock held
|
2005-04-17 02:20:36 +04:00
|
|
|
* AND reference count being 0.
|
2011-06-08 17:35:34 +04:00
|
|
|
* 3. Global variable peer_total is modified under the pool lock.
|
|
|
|
* 4. struct inet_peer fields modification:
|
2005-04-17 02:20:36 +04:00
|
|
|
* avl_left, avl_right, avl_parent, avl_height: pool lock
|
|
|
|
* refcnt: atomically against modifications on other CPU;
|
|
|
|
* usually under some other lock to prevent node disappearing
|
2010-11-30 22:53:55 +03:00
|
|
|
* daddr: unchangeable
|
2005-04-17 02:20:36 +04:00
|
|
|
*/
|
|
|
|
|
2006-12-07 07:33:20 +03:00
|
|
|
static struct kmem_cache *peer_cachep __read_mostly;
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2012-03-07 01:20:26 +04:00
|
|
|
static LIST_HEAD(gc_list);
|
|
|
|
static const int gc_delay = 60 * HZ;
|
|
|
|
static struct delayed_work gc_work;
|
|
|
|
static DEFINE_SPINLOCK(gc_lock);
|
|
|
|
|
2005-04-17 02:20:36 +04:00
|
|
|
#define node_height(x) x->avl_height
|
2010-06-14 23:35:21 +04:00
|
|
|
|
|
|
|
#define peer_avl_empty ((struct inet_peer *)&peer_fake_node)
|
2010-10-26 03:55:38 +04:00
|
|
|
#define peer_avl_empty_rcu ((struct inet_peer __rcu __force *)&peer_fake_node)
|
2010-06-14 23:35:21 +04:00
|
|
|
static const struct inet_peer peer_fake_node = {
|
2010-10-26 03:55:38 +04:00
|
|
|
.avl_left = peer_avl_empty_rcu,
|
|
|
|
.avl_right = peer_avl_empty_rcu,
|
2005-04-17 02:20:36 +04:00
|
|
|
.avl_height = 0
|
|
|
|
};
|
2010-06-14 23:35:21 +04:00
|
|
|
|
2012-06-10 03:27:05 +04:00
|
|
|
void inet_peer_base_init(struct inet_peer_base *bp)
|
|
|
|
{
|
|
|
|
bp->root = peer_avl_empty_rcu;
|
|
|
|
seqlock_init(&bp->lock);
|
|
|
|
bp->total = 0;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(inet_peer_base_init);
|
2010-11-30 23:12:23 +03:00
|
|
|
|
2005-04-17 02:20:36 +04:00
|
|
|
#define PEER_MAXDEPTH 40 /* sufficient for about 2^27 nodes */
|
|
|
|
|
|
|
|
/* Exported for sysctl_net_ipv4. */
|
2007-03-07 07:23:10 +03:00
|
|
|
int inet_peer_threshold __read_mostly = 65536 + 128; /* start to throw entries more
|
2005-04-17 02:20:36 +04:00
|
|
|
* aggressively at this stage */
|
2007-03-07 07:23:10 +03:00
|
|
|
int inet_peer_minttl __read_mostly = 120 * HZ; /* TTL under high load: 120 sec */
|
|
|
|
int inet_peer_maxttl __read_mostly = 10 * 60 * HZ; /* usual time to live: 10 min */
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2012-03-07 01:20:26 +04:00
|
|
|
static void inetpeer_gc_worker(struct work_struct *work)
|
|
|
|
{
|
2012-06-20 08:02:10 +04:00
|
|
|
struct inet_peer *p, *n, *c;
|
2014-04-25 04:53:29 +04:00
|
|
|
struct list_head list;
|
2012-03-07 01:20:26 +04:00
|
|
|
|
|
|
|
spin_lock_bh(&gc_lock);
|
|
|
|
list_replace_init(&gc_list, &list);
|
|
|
|
spin_unlock_bh(&gc_lock);
|
|
|
|
|
|
|
|
if (list_empty(&list))
|
|
|
|
return;
|
|
|
|
|
|
|
|
list_for_each_entry_safe(p, n, &list, gc_list) {
|
|
|
|
|
2012-06-20 08:02:10 +04:00
|
|
|
if (need_resched())
|
2012-03-07 01:20:26 +04:00
|
|
|
cond_resched();
|
|
|
|
|
2012-06-20 08:02:10 +04:00
|
|
|
c = rcu_dereference_protected(p->avl_left, 1);
|
|
|
|
if (c != peer_avl_empty) {
|
|
|
|
list_add_tail(&c->gc_list, &list);
|
|
|
|
p->avl_left = peer_avl_empty_rcu;
|
2012-03-07 01:20:26 +04:00
|
|
|
}
|
|
|
|
|
2012-06-20 08:02:10 +04:00
|
|
|
c = rcu_dereference_protected(p->avl_right, 1);
|
|
|
|
if (c != peer_avl_empty) {
|
|
|
|
list_add_tail(&c->gc_list, &list);
|
|
|
|
p->avl_right = peer_avl_empty_rcu;
|
2012-03-07 01:20:26 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
n = list_entry(p->gc_list.next, struct inet_peer, gc_list);
|
|
|
|
|
|
|
|
if (!atomic_read(&p->refcnt)) {
|
|
|
|
list_del(&p->gc_list);
|
|
|
|
kmem_cache_free(peer_cachep, p);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (list_empty(&list))
|
|
|
|
return;
|
|
|
|
|
|
|
|
spin_lock_bh(&gc_lock);
|
|
|
|
list_splice(&list, &gc_list);
|
|
|
|
spin_unlock_bh(&gc_lock);
|
|
|
|
|
|
|
|
schedule_delayed_work(&gc_work, gc_delay);
|
|
|
|
}
|
2005-04-17 02:20:36 +04:00
|
|
|
|
|
|
|
/* Called from ip_output.c:ip_init */
|
|
|
|
void __init inet_initpeers(void)
|
|
|
|
{
|
|
|
|
struct sysinfo si;
|
|
|
|
|
|
|
|
/* Use the straight interface to information about memory. */
|
|
|
|
si_meminfo(&si);
|
|
|
|
/* The values below were suggested by Alexey Kuznetsov
|
|
|
|
* <kuznet@ms2.inr.ac.ru>. I don't have any opinion about the values
|
|
|
|
* myself. --SAW
|
|
|
|
*/
|
|
|
|
if (si.totalram <= (32768*1024)/PAGE_SIZE)
|
|
|
|
inet_peer_threshold >>= 1; /* max pool size about 1MB on IA32 */
|
|
|
|
if (si.totalram <= (16384*1024)/PAGE_SIZE)
|
|
|
|
inet_peer_threshold >>= 1; /* about 512KB */
|
|
|
|
if (si.totalram <= (8192*1024)/PAGE_SIZE)
|
|
|
|
inet_peer_threshold >>= 2; /* about 128KB */
|
|
|
|
|
|
|
|
peer_cachep = kmem_cache_create("inet_peer_cache",
|
|
|
|
sizeof(struct inet_peer),
|
2010-06-16 08:52:13 +04:00
|
|
|
0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
|
2007-07-20 05:11:58 +04:00
|
|
|
NULL);
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2012-08-22 00:18:23 +04:00
|
|
|
INIT_DEFERRABLE_WORK(&gc_work, inetpeer_gc_worker);
|
2005-04-17 02:20:36 +04:00
|
|
|
}
|
|
|
|
|
2011-03-05 01:33:59 +03:00
|
|
|
#define rcu_deref_locked(X, BASE) \
|
|
|
|
rcu_dereference_protected(X, lockdep_is_held(&(BASE)->lock.lock))
|
|
|
|
|
2007-03-07 07:23:10 +03:00
|
|
|
/*
|
|
|
|
* Called with local BH disabled and the pool lock held.
|
|
|
|
*/
|
2010-11-30 22:41:59 +03:00
|
|
|
#define lookup(_daddr, _stack, _base) \
|
2005-04-17 02:20:36 +04:00
|
|
|
({ \
|
2010-10-26 03:55:38 +04:00
|
|
|
struct inet_peer *u; \
|
|
|
|
struct inet_peer __rcu **v; \
|
inetpeer: RCU conversion
inetpeer currently uses an AVL tree protected by an rwlock.
It's possible to make most lookups use RCU
1) Add a struct rcu_head to struct inet_peer
2) add a lookup_rcu_bh() helper to perform lockless and opportunistic
lookup. This is a normal function, not a macro like lookup().
3) Add a limit to number of links followed by lookup_rcu_bh(). This is
needed in case we fall in a loop.
4) add an smp_wmb() in link_to_pool() right before node insert.
5) make unlink_from_pool() use atomic_cmpxchg() to make sure it can take
last reference to an inet_peer, since lockless readers could increase
refcount, even while we hold peers.lock.
6) Delay struct inet_peer freeing after rcu grace period so that
lookup_rcu_bh() cannot crash.
7) inet_getpeer() first attempts lockless lookup.
Note this lookup can fail even if target is in AVL tree, but a
concurrent writer can let tree in a non correct form.
If this attemps fails, lock is taken a regular lookup is performed
again.
8) convert peers.lock from rwlock to a spinlock
9) Remove SLAB_HWCACHE_ALIGN when peer_cachep is created, because
rcu_head adds 16 bytes on 64bit arches, doubling effective size (64 ->
128 bytes)
In a future patch, this is probably possible to revert this part, if rcu
field is put in an union to share space with rid, ip_id_count, tcp_ts &
tcp_ts_stamp. These fields being manipulated only with refcnt > 0.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-15 12:23:14 +04:00
|
|
|
\
|
|
|
|
stackptr = _stack; \
|
2010-11-30 22:41:59 +03:00
|
|
|
*stackptr++ = &_base->root; \
|
2011-03-05 01:33:59 +03:00
|
|
|
for (u = rcu_deref_locked(_base->root, _base); \
|
2013-12-23 10:37:27 +04:00
|
|
|
u != peer_avl_empty;) { \
|
2015-08-28 02:07:01 +03:00
|
|
|
int cmp = inetpeer_addr_cmp(_daddr, &u->daddr); \
|
2010-11-30 23:08:53 +03:00
|
|
|
if (cmp == 0) \
|
2005-04-17 02:20:36 +04:00
|
|
|
break; \
|
2010-11-30 23:08:53 +03:00
|
|
|
if (cmp == -1) \
|
2005-04-17 02:20:36 +04:00
|
|
|
v = &u->avl_left; \
|
|
|
|
else \
|
|
|
|
v = &u->avl_right; \
|
inetpeer: RCU conversion
inetpeer currently uses an AVL tree protected by an rwlock.
It's possible to make most lookups use RCU
1) Add a struct rcu_head to struct inet_peer
2) add a lookup_rcu_bh() helper to perform lockless and opportunistic
lookup. This is a normal function, not a macro like lookup().
3) Add a limit to number of links followed by lookup_rcu_bh(). This is
needed in case we fall in a loop.
4) add an smp_wmb() in link_to_pool() right before node insert.
5) make unlink_from_pool() use atomic_cmpxchg() to make sure it can take
last reference to an inet_peer, since lockless readers could increase
refcount, even while we hold peers.lock.
6) Delay struct inet_peer freeing after rcu grace period so that
lookup_rcu_bh() cannot crash.
7) inet_getpeer() first attempts lockless lookup.
Note this lookup can fail even if target is in AVL tree, but a
concurrent writer can let tree in a non correct form.
If this attemps fails, lock is taken a regular lookup is performed
again.
8) convert peers.lock from rwlock to a spinlock
9) Remove SLAB_HWCACHE_ALIGN when peer_cachep is created, because
rcu_head adds 16 bytes on 64bit arches, doubling effective size (64 ->
128 bytes)
In a future patch, this is probably possible to revert this part, if rcu
field is put in an union to share space with rid, ip_id_count, tcp_ts &
tcp_ts_stamp. These fields being manipulated only with refcnt > 0.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-15 12:23:14 +04:00
|
|
|
*stackptr++ = v; \
|
2011-03-05 01:33:59 +03:00
|
|
|
u = rcu_deref_locked(*v, _base); \
|
2005-04-17 02:20:36 +04:00
|
|
|
} \
|
|
|
|
u; \
|
|
|
|
})
|
|
|
|
|
inetpeer: RCU conversion
inetpeer currently uses an AVL tree protected by an rwlock.
It's possible to make most lookups use RCU
1) Add a struct rcu_head to struct inet_peer
2) add a lookup_rcu_bh() helper to perform lockless and opportunistic
lookup. This is a normal function, not a macro like lookup().
3) Add a limit to number of links followed by lookup_rcu_bh(). This is
needed in case we fall in a loop.
4) add an smp_wmb() in link_to_pool() right before node insert.
5) make unlink_from_pool() use atomic_cmpxchg() to make sure it can take
last reference to an inet_peer, since lockless readers could increase
refcount, even while we hold peers.lock.
6) Delay struct inet_peer freeing after rcu grace period so that
lookup_rcu_bh() cannot crash.
7) inet_getpeer() first attempts lockless lookup.
Note this lookup can fail even if target is in AVL tree, but a
concurrent writer can let tree in a non correct form.
If this attemps fails, lock is taken a regular lookup is performed
again.
8) convert peers.lock from rwlock to a spinlock
9) Remove SLAB_HWCACHE_ALIGN when peer_cachep is created, because
rcu_head adds 16 bytes on 64bit arches, doubling effective size (64 ->
128 bytes)
In a future patch, this is probably possible to revert this part, if rcu
field is put in an union to share space with rid, ip_id_count, tcp_ts &
tcp_ts_stamp. These fields being manipulated only with refcnt > 0.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-15 12:23:14 +04:00
|
|
|
/*
|
2011-03-09 01:59:28 +03:00
|
|
|
* Called with rcu_read_lock()
|
inetpeer: RCU conversion
inetpeer currently uses an AVL tree protected by an rwlock.
It's possible to make most lookups use RCU
1) Add a struct rcu_head to struct inet_peer
2) add a lookup_rcu_bh() helper to perform lockless and opportunistic
lookup. This is a normal function, not a macro like lookup().
3) Add a limit to number of links followed by lookup_rcu_bh(). This is
needed in case we fall in a loop.
4) add an smp_wmb() in link_to_pool() right before node insert.
5) make unlink_from_pool() use atomic_cmpxchg() to make sure it can take
last reference to an inet_peer, since lockless readers could increase
refcount, even while we hold peers.lock.
6) Delay struct inet_peer freeing after rcu grace period so that
lookup_rcu_bh() cannot crash.
7) inet_getpeer() first attempts lockless lookup.
Note this lookup can fail even if target is in AVL tree, but a
concurrent writer can let tree in a non correct form.
If this attemps fails, lock is taken a regular lookup is performed
again.
8) convert peers.lock from rwlock to a spinlock
9) Remove SLAB_HWCACHE_ALIGN when peer_cachep is created, because
rcu_head adds 16 bytes on 64bit arches, doubling effective size (64 ->
128 bytes)
In a future patch, this is probably possible to revert this part, if rcu
field is put in an union to share space with rid, ip_id_count, tcp_ts &
tcp_ts_stamp. These fields being manipulated only with refcnt > 0.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-15 12:23:14 +04:00
|
|
|
* Because we hold no lock against a writer, its quite possible we fall
|
|
|
|
* in an endless loop.
|
|
|
|
* But every pointer we follow is guaranteed to be valid thanks to RCU.
|
|
|
|
* We exit from this function if number of links exceeds PEER_MAXDEPTH
|
|
|
|
*/
|
2011-03-09 01:59:28 +03:00
|
|
|
static struct inet_peer *lookup_rcu(const struct inetpeer_addr *daddr,
|
2011-06-08 17:35:34 +04:00
|
|
|
struct inet_peer_base *base)
|
inetpeer: RCU conversion
inetpeer currently uses an AVL tree protected by an rwlock.
It's possible to make most lookups use RCU
1) Add a struct rcu_head to struct inet_peer
2) add a lookup_rcu_bh() helper to perform lockless and opportunistic
lookup. This is a normal function, not a macro like lookup().
3) Add a limit to number of links followed by lookup_rcu_bh(). This is
needed in case we fall in a loop.
4) add an smp_wmb() in link_to_pool() right before node insert.
5) make unlink_from_pool() use atomic_cmpxchg() to make sure it can take
last reference to an inet_peer, since lockless readers could increase
refcount, even while we hold peers.lock.
6) Delay struct inet_peer freeing after rcu grace period so that
lookup_rcu_bh() cannot crash.
7) inet_getpeer() first attempts lockless lookup.
Note this lookup can fail even if target is in AVL tree, but a
concurrent writer can let tree in a non correct form.
If this attemps fails, lock is taken a regular lookup is performed
again.
8) convert peers.lock from rwlock to a spinlock
9) Remove SLAB_HWCACHE_ALIGN when peer_cachep is created, because
rcu_head adds 16 bytes on 64bit arches, doubling effective size (64 ->
128 bytes)
In a future patch, this is probably possible to revert this part, if rcu
field is put in an union to share space with rid, ip_id_count, tcp_ts &
tcp_ts_stamp. These fields being manipulated only with refcnt > 0.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-15 12:23:14 +04:00
|
|
|
{
|
2011-03-09 01:59:28 +03:00
|
|
|
struct inet_peer *u = rcu_dereference(base->root);
|
inetpeer: RCU conversion
inetpeer currently uses an AVL tree protected by an rwlock.
It's possible to make most lookups use RCU
1) Add a struct rcu_head to struct inet_peer
2) add a lookup_rcu_bh() helper to perform lockless and opportunistic
lookup. This is a normal function, not a macro like lookup().
3) Add a limit to number of links followed by lookup_rcu_bh(). This is
needed in case we fall in a loop.
4) add an smp_wmb() in link_to_pool() right before node insert.
5) make unlink_from_pool() use atomic_cmpxchg() to make sure it can take
last reference to an inet_peer, since lockless readers could increase
refcount, even while we hold peers.lock.
6) Delay struct inet_peer freeing after rcu grace period so that
lookup_rcu_bh() cannot crash.
7) inet_getpeer() first attempts lockless lookup.
Note this lookup can fail even if target is in AVL tree, but a
concurrent writer can let tree in a non correct form.
If this attemps fails, lock is taken a regular lookup is performed
again.
8) convert peers.lock from rwlock to a spinlock
9) Remove SLAB_HWCACHE_ALIGN when peer_cachep is created, because
rcu_head adds 16 bytes on 64bit arches, doubling effective size (64 ->
128 bytes)
In a future patch, this is probably possible to revert this part, if rcu
field is put in an union to share space with rid, ip_id_count, tcp_ts &
tcp_ts_stamp. These fields being manipulated only with refcnt > 0.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-15 12:23:14 +04:00
|
|
|
int count = 0;
|
|
|
|
|
|
|
|
while (u != peer_avl_empty) {
|
2015-08-28 02:07:01 +03:00
|
|
|
int cmp = inetpeer_addr_cmp(daddr, &u->daddr);
|
2010-11-30 23:08:53 +03:00
|
|
|
if (cmp == 0) {
|
2010-06-16 08:47:39 +04:00
|
|
|
/* Before taking a reference, check if this entry was
|
2011-06-08 17:35:34 +04:00
|
|
|
* deleted (refcnt=-1)
|
2010-06-16 08:47:39 +04:00
|
|
|
*/
|
2011-06-08 17:35:34 +04:00
|
|
|
if (!atomic_add_unless(&u->refcnt, 1, -1))
|
inetpeer: RCU conversion
inetpeer currently uses an AVL tree protected by an rwlock.
It's possible to make most lookups use RCU
1) Add a struct rcu_head to struct inet_peer
2) add a lookup_rcu_bh() helper to perform lockless and opportunistic
lookup. This is a normal function, not a macro like lookup().
3) Add a limit to number of links followed by lookup_rcu_bh(). This is
needed in case we fall in a loop.
4) add an smp_wmb() in link_to_pool() right before node insert.
5) make unlink_from_pool() use atomic_cmpxchg() to make sure it can take
last reference to an inet_peer, since lockless readers could increase
refcount, even while we hold peers.lock.
6) Delay struct inet_peer freeing after rcu grace period so that
lookup_rcu_bh() cannot crash.
7) inet_getpeer() first attempts lockless lookup.
Note this lookup can fail even if target is in AVL tree, but a
concurrent writer can let tree in a non correct form.
If this attemps fails, lock is taken a regular lookup is performed
again.
8) convert peers.lock from rwlock to a spinlock
9) Remove SLAB_HWCACHE_ALIGN when peer_cachep is created, because
rcu_head adds 16 bytes on 64bit arches, doubling effective size (64 ->
128 bytes)
In a future patch, this is probably possible to revert this part, if rcu
field is put in an union to share space with rid, ip_id_count, tcp_ts &
tcp_ts_stamp. These fields being manipulated only with refcnt > 0.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-15 12:23:14 +04:00
|
|
|
u = NULL;
|
|
|
|
return u;
|
|
|
|
}
|
2010-11-30 23:08:53 +03:00
|
|
|
if (cmp == -1)
|
2011-03-09 01:59:28 +03:00
|
|
|
u = rcu_dereference(u->avl_left);
|
inetpeer: RCU conversion
inetpeer currently uses an AVL tree protected by an rwlock.
It's possible to make most lookups use RCU
1) Add a struct rcu_head to struct inet_peer
2) add a lookup_rcu_bh() helper to perform lockless and opportunistic
lookup. This is a normal function, not a macro like lookup().
3) Add a limit to number of links followed by lookup_rcu_bh(). This is
needed in case we fall in a loop.
4) add an smp_wmb() in link_to_pool() right before node insert.
5) make unlink_from_pool() use atomic_cmpxchg() to make sure it can take
last reference to an inet_peer, since lockless readers could increase
refcount, even while we hold peers.lock.
6) Delay struct inet_peer freeing after rcu grace period so that
lookup_rcu_bh() cannot crash.
7) inet_getpeer() first attempts lockless lookup.
Note this lookup can fail even if target is in AVL tree, but a
concurrent writer can let tree in a non correct form.
If this attemps fails, lock is taken a regular lookup is performed
again.
8) convert peers.lock from rwlock to a spinlock
9) Remove SLAB_HWCACHE_ALIGN when peer_cachep is created, because
rcu_head adds 16 bytes on 64bit arches, doubling effective size (64 ->
128 bytes)
In a future patch, this is probably possible to revert this part, if rcu
field is put in an union to share space with rid, ip_id_count, tcp_ts &
tcp_ts_stamp. These fields being manipulated only with refcnt > 0.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-15 12:23:14 +04:00
|
|
|
else
|
2011-03-09 01:59:28 +03:00
|
|
|
u = rcu_dereference(u->avl_right);
|
inetpeer: RCU conversion
inetpeer currently uses an AVL tree protected by an rwlock.
It's possible to make most lookups use RCU
1) Add a struct rcu_head to struct inet_peer
2) add a lookup_rcu_bh() helper to perform lockless and opportunistic
lookup. This is a normal function, not a macro like lookup().
3) Add a limit to number of links followed by lookup_rcu_bh(). This is
needed in case we fall in a loop.
4) add an smp_wmb() in link_to_pool() right before node insert.
5) make unlink_from_pool() use atomic_cmpxchg() to make sure it can take
last reference to an inet_peer, since lockless readers could increase
refcount, even while we hold peers.lock.
6) Delay struct inet_peer freeing after rcu grace period so that
lookup_rcu_bh() cannot crash.
7) inet_getpeer() first attempts lockless lookup.
Note this lookup can fail even if target is in AVL tree, but a
concurrent writer can let tree in a non correct form.
If this attemps fails, lock is taken a regular lookup is performed
again.
8) convert peers.lock from rwlock to a spinlock
9) Remove SLAB_HWCACHE_ALIGN when peer_cachep is created, because
rcu_head adds 16 bytes on 64bit arches, doubling effective size (64 ->
128 bytes)
In a future patch, this is probably possible to revert this part, if rcu
field is put in an union to share space with rid, ip_id_count, tcp_ts &
tcp_ts_stamp. These fields being manipulated only with refcnt > 0.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-15 12:23:14 +04:00
|
|
|
if (unlikely(++count == PEER_MAXDEPTH))
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Called with local BH disabled and the pool lock held. */
|
2010-11-30 22:41:59 +03:00
|
|
|
#define lookup_rightempty(start, base) \
|
2005-04-17 02:20:36 +04:00
|
|
|
({ \
|
2010-10-26 03:55:38 +04:00
|
|
|
struct inet_peer *u; \
|
|
|
|
struct inet_peer __rcu **v; \
|
2005-04-17 02:20:36 +04:00
|
|
|
*stackptr++ = &start->avl_left; \
|
|
|
|
v = &start->avl_left; \
|
2011-03-05 01:33:59 +03:00
|
|
|
for (u = rcu_deref_locked(*v, base); \
|
2013-12-23 10:37:27 +04:00
|
|
|
u->avl_right != peer_avl_empty_rcu;) { \
|
2005-04-17 02:20:36 +04:00
|
|
|
v = &u->avl_right; \
|
|
|
|
*stackptr++ = v; \
|
2011-03-05 01:33:59 +03:00
|
|
|
u = rcu_deref_locked(*v, base); \
|
2005-04-17 02:20:36 +04:00
|
|
|
} \
|
|
|
|
u; \
|
|
|
|
})
|
|
|
|
|
inetpeer: RCU conversion
inetpeer currently uses an AVL tree protected by an rwlock.
It's possible to make most lookups use RCU
1) Add a struct rcu_head to struct inet_peer
2) add a lookup_rcu_bh() helper to perform lockless and opportunistic
lookup. This is a normal function, not a macro like lookup().
3) Add a limit to number of links followed by lookup_rcu_bh(). This is
needed in case we fall in a loop.
4) add an smp_wmb() in link_to_pool() right before node insert.
5) make unlink_from_pool() use atomic_cmpxchg() to make sure it can take
last reference to an inet_peer, since lockless readers could increase
refcount, even while we hold peers.lock.
6) Delay struct inet_peer freeing after rcu grace period so that
lookup_rcu_bh() cannot crash.
7) inet_getpeer() first attempts lockless lookup.
Note this lookup can fail even if target is in AVL tree, but a
concurrent writer can let tree in a non correct form.
If this attemps fails, lock is taken a regular lookup is performed
again.
8) convert peers.lock from rwlock to a spinlock
9) Remove SLAB_HWCACHE_ALIGN when peer_cachep is created, because
rcu_head adds 16 bytes on 64bit arches, doubling effective size (64 ->
128 bytes)
In a future patch, this is probably possible to revert this part, if rcu
field is put in an union to share space with rid, ip_id_count, tcp_ts &
tcp_ts_stamp. These fields being manipulated only with refcnt > 0.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-15 12:23:14 +04:00
|
|
|
/* Called with local BH disabled and the pool lock held.
|
2005-04-17 02:20:36 +04:00
|
|
|
* Variable names are the proof of operation correctness.
|
inetpeer: RCU conversion
inetpeer currently uses an AVL tree protected by an rwlock.
It's possible to make most lookups use RCU
1) Add a struct rcu_head to struct inet_peer
2) add a lookup_rcu_bh() helper to perform lockless and opportunistic
lookup. This is a normal function, not a macro like lookup().
3) Add a limit to number of links followed by lookup_rcu_bh(). This is
needed in case we fall in a loop.
4) add an smp_wmb() in link_to_pool() right before node insert.
5) make unlink_from_pool() use atomic_cmpxchg() to make sure it can take
last reference to an inet_peer, since lockless readers could increase
refcount, even while we hold peers.lock.
6) Delay struct inet_peer freeing after rcu grace period so that
lookup_rcu_bh() cannot crash.
7) inet_getpeer() first attempts lockless lookup.
Note this lookup can fail even if target is in AVL tree, but a
concurrent writer can let tree in a non correct form.
If this attemps fails, lock is taken a regular lookup is performed
again.
8) convert peers.lock from rwlock to a spinlock
9) Remove SLAB_HWCACHE_ALIGN when peer_cachep is created, because
rcu_head adds 16 bytes on 64bit arches, doubling effective size (64 ->
128 bytes)
In a future patch, this is probably possible to revert this part, if rcu
field is put in an union to share space with rid, ip_id_count, tcp_ts &
tcp_ts_stamp. These fields being manipulated only with refcnt > 0.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-15 12:23:14 +04:00
|
|
|
* Look into mm/map_avl.c for more detail description of the ideas.
|
|
|
|
*/
|
2010-10-26 03:55:38 +04:00
|
|
|
static void peer_avl_rebalance(struct inet_peer __rcu **stack[],
|
2010-11-30 22:41:59 +03:00
|
|
|
struct inet_peer __rcu ***stackend,
|
|
|
|
struct inet_peer_base *base)
|
2005-04-17 02:20:36 +04:00
|
|
|
{
|
2010-10-26 03:55:38 +04:00
|
|
|
struct inet_peer __rcu **nodep;
|
|
|
|
struct inet_peer *node, *l, *r;
|
2005-04-17 02:20:36 +04:00
|
|
|
int lh, rh;
|
|
|
|
|
|
|
|
while (stackend > stack) {
|
|
|
|
nodep = *--stackend;
|
2011-03-05 01:33:59 +03:00
|
|
|
node = rcu_deref_locked(*nodep, base);
|
|
|
|
l = rcu_deref_locked(node->avl_left, base);
|
|
|
|
r = rcu_deref_locked(node->avl_right, base);
|
2005-04-17 02:20:36 +04:00
|
|
|
lh = node_height(l);
|
|
|
|
rh = node_height(r);
|
|
|
|
if (lh > rh + 1) { /* l: RH+2 */
|
|
|
|
struct inet_peer *ll, *lr, *lrl, *lrr;
|
|
|
|
int lrh;
|
2011-03-05 01:33:59 +03:00
|
|
|
ll = rcu_deref_locked(l->avl_left, base);
|
|
|
|
lr = rcu_deref_locked(l->avl_right, base);
|
2005-04-17 02:20:36 +04:00
|
|
|
lrh = node_height(lr);
|
|
|
|
if (lrh <= node_height(ll)) { /* ll: RH+1 */
|
2010-10-26 03:55:38 +04:00
|
|
|
RCU_INIT_POINTER(node->avl_left, lr); /* lr: RH or RH+1 */
|
|
|
|
RCU_INIT_POINTER(node->avl_right, r); /* r: RH */
|
2005-04-17 02:20:36 +04:00
|
|
|
node->avl_height = lrh + 1; /* RH+1 or RH+2 */
|
2010-10-26 03:55:38 +04:00
|
|
|
RCU_INIT_POINTER(l->avl_left, ll); /* ll: RH+1 */
|
|
|
|
RCU_INIT_POINTER(l->avl_right, node); /* node: RH+1 or RH+2 */
|
2005-04-17 02:20:36 +04:00
|
|
|
l->avl_height = node->avl_height + 1;
|
2010-10-26 03:55:38 +04:00
|
|
|
RCU_INIT_POINTER(*nodep, l);
|
2005-04-17 02:20:36 +04:00
|
|
|
} else { /* ll: RH, lr: RH+1 */
|
2011-03-05 01:33:59 +03:00
|
|
|
lrl = rcu_deref_locked(lr->avl_left, base);/* lrl: RH or RH-1 */
|
|
|
|
lrr = rcu_deref_locked(lr->avl_right, base);/* lrr: RH or RH-1 */
|
2010-10-26 03:55:38 +04:00
|
|
|
RCU_INIT_POINTER(node->avl_left, lrr); /* lrr: RH or RH-1 */
|
|
|
|
RCU_INIT_POINTER(node->avl_right, r); /* r: RH */
|
2005-04-17 02:20:36 +04:00
|
|
|
node->avl_height = rh + 1; /* node: RH+1 */
|
2010-10-26 03:55:38 +04:00
|
|
|
RCU_INIT_POINTER(l->avl_left, ll); /* ll: RH */
|
|
|
|
RCU_INIT_POINTER(l->avl_right, lrl); /* lrl: RH or RH-1 */
|
2005-04-17 02:20:36 +04:00
|
|
|
l->avl_height = rh + 1; /* l: RH+1 */
|
2010-10-26 03:55:38 +04:00
|
|
|
RCU_INIT_POINTER(lr->avl_left, l); /* l: RH+1 */
|
|
|
|
RCU_INIT_POINTER(lr->avl_right, node); /* node: RH+1 */
|
2005-04-17 02:20:36 +04:00
|
|
|
lr->avl_height = rh + 2;
|
2010-10-26 03:55:38 +04:00
|
|
|
RCU_INIT_POINTER(*nodep, lr);
|
2005-04-17 02:20:36 +04:00
|
|
|
}
|
|
|
|
} else if (rh > lh + 1) { /* r: LH+2 */
|
|
|
|
struct inet_peer *rr, *rl, *rlr, *rll;
|
|
|
|
int rlh;
|
2011-03-05 01:33:59 +03:00
|
|
|
rr = rcu_deref_locked(r->avl_right, base);
|
|
|
|
rl = rcu_deref_locked(r->avl_left, base);
|
2005-04-17 02:20:36 +04:00
|
|
|
rlh = node_height(rl);
|
|
|
|
if (rlh <= node_height(rr)) { /* rr: LH+1 */
|
2010-10-26 03:55:38 +04:00
|
|
|
RCU_INIT_POINTER(node->avl_right, rl); /* rl: LH or LH+1 */
|
|
|
|
RCU_INIT_POINTER(node->avl_left, l); /* l: LH */
|
2005-04-17 02:20:36 +04:00
|
|
|
node->avl_height = rlh + 1; /* LH+1 or LH+2 */
|
2010-10-26 03:55:38 +04:00
|
|
|
RCU_INIT_POINTER(r->avl_right, rr); /* rr: LH+1 */
|
|
|
|
RCU_INIT_POINTER(r->avl_left, node); /* node: LH+1 or LH+2 */
|
2005-04-17 02:20:36 +04:00
|
|
|
r->avl_height = node->avl_height + 1;
|
2010-10-26 03:55:38 +04:00
|
|
|
RCU_INIT_POINTER(*nodep, r);
|
2005-04-17 02:20:36 +04:00
|
|
|
} else { /* rr: RH, rl: RH+1 */
|
2011-03-05 01:33:59 +03:00
|
|
|
rlr = rcu_deref_locked(rl->avl_right, base);/* rlr: LH or LH-1 */
|
|
|
|
rll = rcu_deref_locked(rl->avl_left, base);/* rll: LH or LH-1 */
|
2010-10-26 03:55:38 +04:00
|
|
|
RCU_INIT_POINTER(node->avl_right, rll); /* rll: LH or LH-1 */
|
|
|
|
RCU_INIT_POINTER(node->avl_left, l); /* l: LH */
|
2005-04-17 02:20:36 +04:00
|
|
|
node->avl_height = lh + 1; /* node: LH+1 */
|
2010-10-26 03:55:38 +04:00
|
|
|
RCU_INIT_POINTER(r->avl_right, rr); /* rr: LH */
|
|
|
|
RCU_INIT_POINTER(r->avl_left, rlr); /* rlr: LH or LH-1 */
|
2005-04-17 02:20:36 +04:00
|
|
|
r->avl_height = lh + 1; /* r: LH+1 */
|
2010-10-26 03:55:38 +04:00
|
|
|
RCU_INIT_POINTER(rl->avl_right, r); /* r: LH+1 */
|
|
|
|
RCU_INIT_POINTER(rl->avl_left, node); /* node: LH+1 */
|
2005-04-17 02:20:36 +04:00
|
|
|
rl->avl_height = lh + 2;
|
2010-10-26 03:55:38 +04:00
|
|
|
RCU_INIT_POINTER(*nodep, rl);
|
2005-04-17 02:20:36 +04:00
|
|
|
}
|
|
|
|
} else {
|
|
|
|
node->avl_height = (lh > rh ? lh : rh) + 1;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
inetpeer: RCU conversion
inetpeer currently uses an AVL tree protected by an rwlock.
It's possible to make most lookups use RCU
1) Add a struct rcu_head to struct inet_peer
2) add a lookup_rcu_bh() helper to perform lockless and opportunistic
lookup. This is a normal function, not a macro like lookup().
3) Add a limit to number of links followed by lookup_rcu_bh(). This is
needed in case we fall in a loop.
4) add an smp_wmb() in link_to_pool() right before node insert.
5) make unlink_from_pool() use atomic_cmpxchg() to make sure it can take
last reference to an inet_peer, since lockless readers could increase
refcount, even while we hold peers.lock.
6) Delay struct inet_peer freeing after rcu grace period so that
lookup_rcu_bh() cannot crash.
7) inet_getpeer() first attempts lockless lookup.
Note this lookup can fail even if target is in AVL tree, but a
concurrent writer can let tree in a non correct form.
If this attemps fails, lock is taken a regular lookup is performed
again.
8) convert peers.lock from rwlock to a spinlock
9) Remove SLAB_HWCACHE_ALIGN when peer_cachep is created, because
rcu_head adds 16 bytes on 64bit arches, doubling effective size (64 ->
128 bytes)
In a future patch, this is probably possible to revert this part, if rcu
field is put in an union to share space with rid, ip_id_count, tcp_ts &
tcp_ts_stamp. These fields being manipulated only with refcnt > 0.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-15 12:23:14 +04:00
|
|
|
/* Called with local BH disabled and the pool lock held. */
|
2010-11-30 22:41:59 +03:00
|
|
|
#define link_to_pool(n, base) \
|
2005-04-17 02:20:36 +04:00
|
|
|
do { \
|
|
|
|
n->avl_height = 1; \
|
2010-10-26 03:55:38 +04:00
|
|
|
n->avl_left = peer_avl_empty_rcu; \
|
|
|
|
n->avl_right = peer_avl_empty_rcu; \
|
|
|
|
/* lockless readers can catch us now */ \
|
|
|
|
rcu_assign_pointer(**--stackptr, n); \
|
2010-11-30 22:41:59 +03:00
|
|
|
peer_avl_rebalance(stack, stackptr, base); \
|
2010-06-14 23:35:21 +04:00
|
|
|
} while (0)
|
2005-04-17 02:20:36 +04:00
|
|
|
|
inetpeer: RCU conversion
inetpeer currently uses an AVL tree protected by an rwlock.
It's possible to make most lookups use RCU
1) Add a struct rcu_head to struct inet_peer
2) add a lookup_rcu_bh() helper to perform lockless and opportunistic
lookup. This is a normal function, not a macro like lookup().
3) Add a limit to number of links followed by lookup_rcu_bh(). This is
needed in case we fall in a loop.
4) add an smp_wmb() in link_to_pool() right before node insert.
5) make unlink_from_pool() use atomic_cmpxchg() to make sure it can take
last reference to an inet_peer, since lockless readers could increase
refcount, even while we hold peers.lock.
6) Delay struct inet_peer freeing after rcu grace period so that
lookup_rcu_bh() cannot crash.
7) inet_getpeer() first attempts lockless lookup.
Note this lookup can fail even if target is in AVL tree, but a
concurrent writer can let tree in a non correct form.
If this attemps fails, lock is taken a regular lookup is performed
again.
8) convert peers.lock from rwlock to a spinlock
9) Remove SLAB_HWCACHE_ALIGN when peer_cachep is created, because
rcu_head adds 16 bytes on 64bit arches, doubling effective size (64 ->
128 bytes)
In a future patch, this is probably possible to revert this part, if rcu
field is put in an union to share space with rid, ip_id_count, tcp_ts &
tcp_ts_stamp. These fields being manipulated only with refcnt > 0.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-15 12:23:14 +04:00
|
|
|
static void inetpeer_free_rcu(struct rcu_head *head)
|
|
|
|
{
|
|
|
|
kmem_cache_free(peer_cachep, container_of(head, struct inet_peer, rcu));
|
|
|
|
}
|
|
|
|
|
2011-04-12 02:39:40 +04:00
|
|
|
static void unlink_from_pool(struct inet_peer *p, struct inet_peer_base *base,
|
|
|
|
struct inet_peer __rcu **stack[PEER_MAXDEPTH])
|
2005-04-17 02:20:36 +04:00
|
|
|
{
|
2011-06-08 17:35:34 +04:00
|
|
|
struct inet_peer __rcu ***stackptr, ***delp;
|
|
|
|
|
|
|
|
if (lookup(&p->daddr, stack, base) != p)
|
|
|
|
BUG();
|
|
|
|
delp = stackptr - 1; /* *delp[0] == p */
|
|
|
|
if (p->avl_left == peer_avl_empty_rcu) {
|
|
|
|
*delp[0] = p->avl_right;
|
|
|
|
--stackptr;
|
|
|
|
} else {
|
|
|
|
/* look for a node to insert instead of p */
|
|
|
|
struct inet_peer *t;
|
|
|
|
t = lookup_rightempty(p, base);
|
|
|
|
BUG_ON(rcu_deref_locked(*stackptr[-1], base) != t);
|
|
|
|
**--stackptr = t->avl_left;
|
|
|
|
/* t is removed, t->daddr > x->daddr for any
|
|
|
|
* x in p->avl_left subtree.
|
|
|
|
* Put t in the old place of p. */
|
|
|
|
RCU_INIT_POINTER(*delp[0], t);
|
|
|
|
t->avl_left = p->avl_left;
|
|
|
|
t->avl_right = p->avl_right;
|
|
|
|
t->avl_height = p->avl_height;
|
|
|
|
BUG_ON(delp[1] != &p->avl_left);
|
|
|
|
delp[1] = &t->avl_left; /* was &p->avl_left */
|
2005-04-17 02:20:36 +04:00
|
|
|
}
|
2011-06-08 17:35:34 +04:00
|
|
|
peer_avl_rebalance(stack, stackptr, base);
|
|
|
|
base->total--;
|
|
|
|
call_rcu(&p->rcu, inetpeer_free_rcu);
|
2005-04-17 02:20:36 +04:00
|
|
|
}
|
|
|
|
|
2011-06-08 17:35:34 +04:00
|
|
|
/* perform garbage collect on all items stacked during a lookup */
|
|
|
|
static int inet_peer_gc(struct inet_peer_base *base,
|
|
|
|
struct inet_peer __rcu **stack[PEER_MAXDEPTH],
|
|
|
|
struct inet_peer __rcu ***stackptr)
|
2010-11-30 22:41:59 +03:00
|
|
|
{
|
2011-06-08 17:35:34 +04:00
|
|
|
struct inet_peer *p, *gchead = NULL;
|
|
|
|
__u32 delta, ttl;
|
|
|
|
int cnt = 0;
|
2007-11-13 08:27:28 +03:00
|
|
|
|
2011-06-08 17:35:34 +04:00
|
|
|
if (base->total >= inet_peer_threshold)
|
|
|
|
ttl = 0; /* be aggressive */
|
|
|
|
else
|
|
|
|
ttl = inet_peer_maxttl
|
|
|
|
- (inet_peer_maxttl - inet_peer_minttl) / HZ *
|
|
|
|
base->total / inet_peer_threshold * HZ;
|
|
|
|
stackptr--; /* last stack slot is peer_avl_empty */
|
|
|
|
while (stackptr > stack) {
|
|
|
|
stackptr--;
|
|
|
|
p = rcu_deref_locked(**stackptr, base);
|
2011-07-11 06:49:52 +04:00
|
|
|
if (atomic_read(&p->refcnt) == 0) {
|
|
|
|
smp_rmb();
|
|
|
|
delta = (__u32)jiffies - p->dtime;
|
|
|
|
if (delta >= ttl &&
|
|
|
|
atomic_cmpxchg(&p->refcnt, 0, -1) == 0) {
|
|
|
|
p->gc_next = gchead;
|
|
|
|
gchead = p;
|
|
|
|
}
|
2005-04-17 02:20:36 +04:00
|
|
|
}
|
|
|
|
}
|
2011-06-08 17:35:34 +04:00
|
|
|
while ((p = gchead) != NULL) {
|
|
|
|
gchead = p->gc_next;
|
|
|
|
cnt++;
|
|
|
|
unlink_from_pool(p, base, stack);
|
|
|
|
}
|
|
|
|
return cnt;
|
2005-04-17 02:20:36 +04:00
|
|
|
}
|
|
|
|
|
2012-06-10 06:12:36 +04:00
|
|
|
struct inet_peer *inet_getpeer(struct inet_peer_base *base,
|
2012-06-08 05:20:41 +04:00
|
|
|
const struct inetpeer_addr *daddr,
|
|
|
|
int create)
|
2005-04-17 02:20:36 +04:00
|
|
|
{
|
2010-10-26 03:55:38 +04:00
|
|
|
struct inet_peer __rcu **stack[PEER_MAXDEPTH], ***stackptr;
|
2010-11-30 22:41:59 +03:00
|
|
|
struct inet_peer *p;
|
2011-03-05 01:33:59 +03:00
|
|
|
unsigned int sequence;
|
2011-06-08 17:35:34 +04:00
|
|
|
int invalidated, gccnt = 0;
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2011-06-08 17:35:34 +04:00
|
|
|
/* Attempt a lockless lookup first.
|
inetpeer: RCU conversion
inetpeer currently uses an AVL tree protected by an rwlock.
It's possible to make most lookups use RCU
1) Add a struct rcu_head to struct inet_peer
2) add a lookup_rcu_bh() helper to perform lockless and opportunistic
lookup. This is a normal function, not a macro like lookup().
3) Add a limit to number of links followed by lookup_rcu_bh(). This is
needed in case we fall in a loop.
4) add an smp_wmb() in link_to_pool() right before node insert.
5) make unlink_from_pool() use atomic_cmpxchg() to make sure it can take
last reference to an inet_peer, since lockless readers could increase
refcount, even while we hold peers.lock.
6) Delay struct inet_peer freeing after rcu grace period so that
lookup_rcu_bh() cannot crash.
7) inet_getpeer() first attempts lockless lookup.
Note this lookup can fail even if target is in AVL tree, but a
concurrent writer can let tree in a non correct form.
If this attemps fails, lock is taken a regular lookup is performed
again.
8) convert peers.lock from rwlock to a spinlock
9) Remove SLAB_HWCACHE_ALIGN when peer_cachep is created, because
rcu_head adds 16 bytes on 64bit arches, doubling effective size (64 ->
128 bytes)
In a future patch, this is probably possible to revert this part, if rcu
field is put in an union to share space with rid, ip_id_count, tcp_ts &
tcp_ts_stamp. These fields being manipulated only with refcnt > 0.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-15 12:23:14 +04:00
|
|
|
* Because of a concurrent writer, we might not find an existing entry.
|
|
|
|
*/
|
2011-03-09 01:59:28 +03:00
|
|
|
rcu_read_lock();
|
2011-03-05 01:33:59 +03:00
|
|
|
sequence = read_seqbegin(&base->lock);
|
2011-06-08 17:35:34 +04:00
|
|
|
p = lookup_rcu(daddr, base);
|
2011-03-05 01:33:59 +03:00
|
|
|
invalidated = read_seqretry(&base->lock, sequence);
|
2011-03-09 01:59:28 +03:00
|
|
|
rcu_read_unlock();
|
inetpeer: RCU conversion
inetpeer currently uses an AVL tree protected by an rwlock.
It's possible to make most lookups use RCU
1) Add a struct rcu_head to struct inet_peer
2) add a lookup_rcu_bh() helper to perform lockless and opportunistic
lookup. This is a normal function, not a macro like lookup().
3) Add a limit to number of links followed by lookup_rcu_bh(). This is
needed in case we fall in a loop.
4) add an smp_wmb() in link_to_pool() right before node insert.
5) make unlink_from_pool() use atomic_cmpxchg() to make sure it can take
last reference to an inet_peer, since lockless readers could increase
refcount, even while we hold peers.lock.
6) Delay struct inet_peer freeing after rcu grace period so that
lookup_rcu_bh() cannot crash.
7) inet_getpeer() first attempts lockless lookup.
Note this lookup can fail even if target is in AVL tree, but a
concurrent writer can let tree in a non correct form.
If this attemps fails, lock is taken a regular lookup is performed
again.
8) convert peers.lock from rwlock to a spinlock
9) Remove SLAB_HWCACHE_ALIGN when peer_cachep is created, because
rcu_head adds 16 bytes on 64bit arches, doubling effective size (64 ->
128 bytes)
In a future patch, this is probably possible to revert this part, if rcu
field is put in an union to share space with rid, ip_id_count, tcp_ts &
tcp_ts_stamp. These fields being manipulated only with refcnt > 0.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-15 12:23:14 +04:00
|
|
|
|
2011-06-08 17:35:34 +04:00
|
|
|
if (p)
|
inetpeer: RCU conversion
inetpeer currently uses an AVL tree protected by an rwlock.
It's possible to make most lookups use RCU
1) Add a struct rcu_head to struct inet_peer
2) add a lookup_rcu_bh() helper to perform lockless and opportunistic
lookup. This is a normal function, not a macro like lookup().
3) Add a limit to number of links followed by lookup_rcu_bh(). This is
needed in case we fall in a loop.
4) add an smp_wmb() in link_to_pool() right before node insert.
5) make unlink_from_pool() use atomic_cmpxchg() to make sure it can take
last reference to an inet_peer, since lockless readers could increase
refcount, even while we hold peers.lock.
6) Delay struct inet_peer freeing after rcu grace period so that
lookup_rcu_bh() cannot crash.
7) inet_getpeer() first attempts lockless lookup.
Note this lookup can fail even if target is in AVL tree, but a
concurrent writer can let tree in a non correct form.
If this attemps fails, lock is taken a regular lookup is performed
again.
8) convert peers.lock from rwlock to a spinlock
9) Remove SLAB_HWCACHE_ALIGN when peer_cachep is created, because
rcu_head adds 16 bytes on 64bit arches, doubling effective size (64 ->
128 bytes)
In a future patch, this is probably possible to revert this part, if rcu
field is put in an union to share space with rid, ip_id_count, tcp_ts &
tcp_ts_stamp. These fields being manipulated only with refcnt > 0.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-15 12:23:14 +04:00
|
|
|
return p;
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2011-03-05 01:33:59 +03:00
|
|
|
/* If no writer did a change during our lookup, we can return early. */
|
|
|
|
if (!create && !invalidated)
|
|
|
|
return NULL;
|
|
|
|
|
inetpeer: RCU conversion
inetpeer currently uses an AVL tree protected by an rwlock.
It's possible to make most lookups use RCU
1) Add a struct rcu_head to struct inet_peer
2) add a lookup_rcu_bh() helper to perform lockless and opportunistic
lookup. This is a normal function, not a macro like lookup().
3) Add a limit to number of links followed by lookup_rcu_bh(). This is
needed in case we fall in a loop.
4) add an smp_wmb() in link_to_pool() right before node insert.
5) make unlink_from_pool() use atomic_cmpxchg() to make sure it can take
last reference to an inet_peer, since lockless readers could increase
refcount, even while we hold peers.lock.
6) Delay struct inet_peer freeing after rcu grace period so that
lookup_rcu_bh() cannot crash.
7) inet_getpeer() first attempts lockless lookup.
Note this lookup can fail even if target is in AVL tree, but a
concurrent writer can let tree in a non correct form.
If this attemps fails, lock is taken a regular lookup is performed
again.
8) convert peers.lock from rwlock to a spinlock
9) Remove SLAB_HWCACHE_ALIGN when peer_cachep is created, because
rcu_head adds 16 bytes on 64bit arches, doubling effective size (64 ->
128 bytes)
In a future patch, this is probably possible to revert this part, if rcu
field is put in an union to share space with rid, ip_id_count, tcp_ts &
tcp_ts_stamp. These fields being manipulated only with refcnt > 0.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-15 12:23:14 +04:00
|
|
|
/* retry an exact lookup, taking the lock before.
|
|
|
|
* At least, nodes should be hot in our cache.
|
|
|
|
*/
|
2011-03-05 01:33:59 +03:00
|
|
|
write_seqlock_bh(&base->lock);
|
2011-06-08 17:35:34 +04:00
|
|
|
relookup:
|
2010-11-30 23:08:53 +03:00
|
|
|
p = lookup(daddr, stack, base);
|
2005-04-17 02:20:36 +04:00
|
|
|
if (p != peer_avl_empty) {
|
2011-06-08 17:35:34 +04:00
|
|
|
atomic_inc(&p->refcnt);
|
2011-03-05 01:33:59 +03:00
|
|
|
write_sequnlock_bh(&base->lock);
|
2011-06-08 17:35:34 +04:00
|
|
|
return p;
|
|
|
|
}
|
|
|
|
if (!gccnt) {
|
|
|
|
gccnt = inet_peer_gc(base, stack, stackptr);
|
|
|
|
if (gccnt && create)
|
|
|
|
goto relookup;
|
2005-04-17 02:20:36 +04:00
|
|
|
}
|
inetpeer: RCU conversion
inetpeer currently uses an AVL tree protected by an rwlock.
It's possible to make most lookups use RCU
1) Add a struct rcu_head to struct inet_peer
2) add a lookup_rcu_bh() helper to perform lockless and opportunistic
lookup. This is a normal function, not a macro like lookup().
3) Add a limit to number of links followed by lookup_rcu_bh(). This is
needed in case we fall in a loop.
4) add an smp_wmb() in link_to_pool() right before node insert.
5) make unlink_from_pool() use atomic_cmpxchg() to make sure it can take
last reference to an inet_peer, since lockless readers could increase
refcount, even while we hold peers.lock.
6) Delay struct inet_peer freeing after rcu grace period so that
lookup_rcu_bh() cannot crash.
7) inet_getpeer() first attempts lockless lookup.
Note this lookup can fail even if target is in AVL tree, but a
concurrent writer can let tree in a non correct form.
If this attemps fails, lock is taken a regular lookup is performed
again.
8) convert peers.lock from rwlock to a spinlock
9) Remove SLAB_HWCACHE_ALIGN when peer_cachep is created, because
rcu_head adds 16 bytes on 64bit arches, doubling effective size (64 ->
128 bytes)
In a future patch, this is probably possible to revert this part, if rcu
field is put in an union to share space with rid, ip_id_count, tcp_ts &
tcp_ts_stamp. These fields being manipulated only with refcnt > 0.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-15 12:23:14 +04:00
|
|
|
p = create ? kmem_cache_alloc(peer_cachep, GFP_ATOMIC) : NULL;
|
|
|
|
if (p) {
|
2010-11-30 22:54:19 +03:00
|
|
|
p->daddr = *daddr;
|
inetpeer: RCU conversion
inetpeer currently uses an AVL tree protected by an rwlock.
It's possible to make most lookups use RCU
1) Add a struct rcu_head to struct inet_peer
2) add a lookup_rcu_bh() helper to perform lockless and opportunistic
lookup. This is a normal function, not a macro like lookup().
3) Add a limit to number of links followed by lookup_rcu_bh(). This is
needed in case we fall in a loop.
4) add an smp_wmb() in link_to_pool() right before node insert.
5) make unlink_from_pool() use atomic_cmpxchg() to make sure it can take
last reference to an inet_peer, since lockless readers could increase
refcount, even while we hold peers.lock.
6) Delay struct inet_peer freeing after rcu grace period so that
lookup_rcu_bh() cannot crash.
7) inet_getpeer() first attempts lockless lookup.
Note this lookup can fail even if target is in AVL tree, but a
concurrent writer can let tree in a non correct form.
If this attemps fails, lock is taken a regular lookup is performed
again.
8) convert peers.lock from rwlock to a spinlock
9) Remove SLAB_HWCACHE_ALIGN when peer_cachep is created, because
rcu_head adds 16 bytes on 64bit arches, doubling effective size (64 ->
128 bytes)
In a future patch, this is probably possible to revert this part, if rcu
field is put in an union to share space with rid, ip_id_count, tcp_ts &
tcp_ts_stamp. These fields being manipulated only with refcnt > 0.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-15 12:23:14 +04:00
|
|
|
atomic_set(&p->refcnt, 1);
|
|
|
|
atomic_set(&p->rid, 0);
|
2011-01-28 00:52:16 +03:00
|
|
|
p->metrics[RTAX_LOCK-1] = INETPEER_METRICS_NEW;
|
2011-02-05 02:55:25 +03:00
|
|
|
p->rate_tokens = 0;
|
2012-09-27 08:11:00 +04:00
|
|
|
/* 60*HZ is arbitrary, but chosen enough high so that the first
|
|
|
|
* calculation of tokens is at its maximum.
|
|
|
|
*/
|
|
|
|
p->rate_last = jiffies - 60*HZ;
|
2012-03-07 01:20:26 +04:00
|
|
|
INIT_LIST_HEAD(&p->gc_list);
|
inetpeer: RCU conversion
inetpeer currently uses an AVL tree protected by an rwlock.
It's possible to make most lookups use RCU
1) Add a struct rcu_head to struct inet_peer
2) add a lookup_rcu_bh() helper to perform lockless and opportunistic
lookup. This is a normal function, not a macro like lookup().
3) Add a limit to number of links followed by lookup_rcu_bh(). This is
needed in case we fall in a loop.
4) add an smp_wmb() in link_to_pool() right before node insert.
5) make unlink_from_pool() use atomic_cmpxchg() to make sure it can take
last reference to an inet_peer, since lockless readers could increase
refcount, even while we hold peers.lock.
6) Delay struct inet_peer freeing after rcu grace period so that
lookup_rcu_bh() cannot crash.
7) inet_getpeer() first attempts lockless lookup.
Note this lookup can fail even if target is in AVL tree, but a
concurrent writer can let tree in a non correct form.
If this attemps fails, lock is taken a regular lookup is performed
again.
8) convert peers.lock from rwlock to a spinlock
9) Remove SLAB_HWCACHE_ALIGN when peer_cachep is created, because
rcu_head adds 16 bytes on 64bit arches, doubling effective size (64 ->
128 bytes)
In a future patch, this is probably possible to revert this part, if rcu
field is put in an union to share space with rid, ip_id_count, tcp_ts &
tcp_ts_stamp. These fields being manipulated only with refcnt > 0.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-15 12:23:14 +04:00
|
|
|
|
|
|
|
/* Link the node. */
|
2010-11-30 22:41:59 +03:00
|
|
|
link_to_pool(p, base);
|
|
|
|
base->total++;
|
inetpeer: RCU conversion
inetpeer currently uses an AVL tree protected by an rwlock.
It's possible to make most lookups use RCU
1) Add a struct rcu_head to struct inet_peer
2) add a lookup_rcu_bh() helper to perform lockless and opportunistic
lookup. This is a normal function, not a macro like lookup().
3) Add a limit to number of links followed by lookup_rcu_bh(). This is
needed in case we fall in a loop.
4) add an smp_wmb() in link_to_pool() right before node insert.
5) make unlink_from_pool() use atomic_cmpxchg() to make sure it can take
last reference to an inet_peer, since lockless readers could increase
refcount, even while we hold peers.lock.
6) Delay struct inet_peer freeing after rcu grace period so that
lookup_rcu_bh() cannot crash.
7) inet_getpeer() first attempts lockless lookup.
Note this lookup can fail even if target is in AVL tree, but a
concurrent writer can let tree in a non correct form.
If this attemps fails, lock is taken a regular lookup is performed
again.
8) convert peers.lock from rwlock to a spinlock
9) Remove SLAB_HWCACHE_ALIGN when peer_cachep is created, because
rcu_head adds 16 bytes on 64bit arches, doubling effective size (64 ->
128 bytes)
In a future patch, this is probably possible to revert this part, if rcu
field is put in an union to share space with rid, ip_id_count, tcp_ts &
tcp_ts_stamp. These fields being manipulated only with refcnt > 0.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-15 12:23:14 +04:00
|
|
|
}
|
2011-03-05 01:33:59 +03:00
|
|
|
write_sequnlock_bh(&base->lock);
|
2005-04-17 02:20:36 +04:00
|
|
|
|
|
|
|
return p;
|
|
|
|
}
|
2010-11-30 23:27:11 +03:00
|
|
|
EXPORT_SYMBOL_GPL(inet_getpeer);
|
2010-11-30 22:41:59 +03:00
|
|
|
|
2006-10-13 08:21:06 +04:00
|
|
|
void inet_putpeer(struct inet_peer *p)
|
|
|
|
{
|
2011-06-08 17:35:34 +04:00
|
|
|
p->dtime = (__u32)jiffies;
|
2014-03-17 21:06:10 +04:00
|
|
|
smp_mb__before_atomic();
|
2011-06-08 17:35:34 +04:00
|
|
|
atomic_dec(&p->refcnt);
|
2006-10-13 08:21:06 +04:00
|
|
|
}
|
2010-11-30 23:27:11 +03:00
|
|
|
EXPORT_SYMBOL_GPL(inet_putpeer);
|
2011-02-05 02:55:25 +03:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Check transmit rate limitation for given message.
|
|
|
|
* The rate information is held in the inet_peer entries now.
|
|
|
|
* This function is generic and could be used for other purposes
|
|
|
|
* too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
|
|
|
|
*
|
|
|
|
* Note that the same inet_peer fields are modified by functions in
|
|
|
|
* route.c too, but these work for packet destinations while xrlim_allow
|
|
|
|
* works for icmp destinations. This means the rate limiting information
|
|
|
|
* for one "ip object" is shared - and these ICMPs are twice limited:
|
|
|
|
* by source and by destination.
|
|
|
|
*
|
|
|
|
* RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
|
|
|
|
* SHOULD allow setting of rate limits
|
|
|
|
*
|
|
|
|
* Shared between ICMPv4 and ICMPv6.
|
|
|
|
*/
|
|
|
|
#define XRLIM_BURST_FACTOR 6
|
|
|
|
bool inet_peer_xrlim_allow(struct inet_peer *peer, int timeout)
|
|
|
|
{
|
|
|
|
unsigned long now, token;
|
|
|
|
bool rc = false;
|
|
|
|
|
|
|
|
if (!peer)
|
|
|
|
return true;
|
|
|
|
|
|
|
|
token = peer->rate_tokens;
|
|
|
|
now = jiffies;
|
|
|
|
token += now - peer->rate_last;
|
|
|
|
peer->rate_last = now;
|
|
|
|
if (token > XRLIM_BURST_FACTOR * timeout)
|
|
|
|
token = XRLIM_BURST_FACTOR * timeout;
|
|
|
|
if (token >= timeout) {
|
|
|
|
token -= timeout;
|
|
|
|
rc = true;
|
|
|
|
}
|
|
|
|
peer->rate_tokens = token;
|
|
|
|
return rc;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(inet_peer_xrlim_allow);
|
2012-03-07 01:20:26 +04:00
|
|
|
|
2012-06-05 07:00:18 +04:00
|
|
|
static void inetpeer_inval_rcu(struct rcu_head *head)
|
|
|
|
{
|
|
|
|
struct inet_peer *p = container_of(head, struct inet_peer, gc_rcu);
|
|
|
|
|
|
|
|
spin_lock_bh(&gc_lock);
|
|
|
|
list_add_tail(&p->gc_list, &gc_list);
|
|
|
|
spin_unlock_bh(&gc_lock);
|
|
|
|
|
|
|
|
schedule_delayed_work(&gc_work, gc_delay);
|
|
|
|
}
|
|
|
|
|
2012-06-10 03:32:41 +04:00
|
|
|
void inetpeer_invalidate_tree(struct inet_peer_base *base)
|
2012-03-07 01:20:26 +04:00
|
|
|
{
|
2012-06-20 08:02:10 +04:00
|
|
|
struct inet_peer *root;
|
2012-03-07 01:20:26 +04:00
|
|
|
|
|
|
|
write_seqlock_bh(&base->lock);
|
|
|
|
|
2012-06-20 08:02:10 +04:00
|
|
|
root = rcu_deref_locked(base->root, base);
|
|
|
|
if (root != peer_avl_empty) {
|
|
|
|
base->root = peer_avl_empty_rcu;
|
2012-03-07 01:20:26 +04:00
|
|
|
base->total = 0;
|
2012-06-20 08:02:10 +04:00
|
|
|
call_rcu(&root->gc_rcu, inetpeer_inval_rcu);
|
2012-03-07 01:20:26 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
write_sequnlock_bh(&base->lock);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(inetpeer_invalidate_tree);
|