const struct nlattr *attr) __read_mostly;
EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
-DEFINE_SPINLOCK(nf_conntrack_lock);
-EXPORT_SYMBOL_GPL(nf_conntrack_lock);
+__cacheline_aligned_in_smp spinlock_t nf_conntrack_locks[CONNTRACK_LOCKS];
+EXPORT_SYMBOL_GPL(nf_conntrack_locks);
__cacheline_aligned_in_smp DEFINE_SPINLOCK(nf_conntrack_expect_lock);
EXPORT_SYMBOL_GPL(nf_conntrack_expect_lock);
+static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2)
+{
+ h1 %= CONNTRACK_LOCKS;
+ h2 %= CONNTRACK_LOCKS;
+ spin_unlock(&nf_conntrack_locks[h1]);
+ if (h1 != h2)
+ spin_unlock(&nf_conntrack_locks[h2]);
+}
+
+/* return true if we need to recompute hashes (in case hash table was resized) */
+static bool nf_conntrack_double_lock(struct net *net, unsigned int h1,
+ unsigned int h2, unsigned int sequence)
+{
+ h1 %= CONNTRACK_LOCKS;
+ h2 %= CONNTRACK_LOCKS;
+ if (h1 <= h2) {
+ spin_lock(&nf_conntrack_locks[h1]);
+ if (h1 != h2)
+ spin_lock_nested(&nf_conntrack_locks[h2],
+ SINGLE_DEPTH_NESTING);
+ } else {
+ spin_lock(&nf_conntrack_locks[h2]);
+ spin_lock_nested(&nf_conntrack_locks[h1],
+ SINGLE_DEPTH_NESTING);
+ }
+ if (read_seqcount_retry(&net->ct.generation, sequence)) {
+ nf_conntrack_double_unlock(h1, h2);
+ return true;
+ }
+ return false;
+}
+
+static void nf_conntrack_all_lock(void)
+{
+ int i;
+
+ for (i = 0; i < CONNTRACK_LOCKS; i++)
+ spin_lock_nested(&nf_conntrack_locks[i], i);
+}
+
+static void nf_conntrack_all_unlock(void)
+{
+ int i;
+
+ for (i = 0; i < CONNTRACK_LOCKS; i++)
+ spin_unlock(&nf_conntrack_locks[i]);
+}
+
unsigned int nf_conntrack_htable_size __read_mostly;
EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
static void nf_ct_delete_from_lists(struct nf_conn *ct)
{
struct net *net = nf_ct_net(ct);
+ unsigned int hash, reply_hash;
+ u16 zone = nf_ct_zone(ct);
+ unsigned int sequence;
nf_ct_helper_destroy(ct);
- spin_lock_bh(&nf_conntrack_lock);
- /* Inside lock so preempt is disabled on module removal path.
- * Otherwise we can get spurious warnings. */
- NF_CT_STAT_INC(net, delete_list);
+
+ local_bh_disable();
+ do {
+ sequence = read_seqcount_begin(&net->ct.generation);
+ hash = hash_conntrack(net, zone,
+ &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
+ reply_hash = hash_conntrack(net, zone,
+ &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
+ } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
+
clean_from_lists(ct);
+ nf_conntrack_double_unlock(hash, reply_hash);
+
nf_ct_add_to_dying_list(ct);
- spin_unlock_bh(&nf_conntrack_lock);
+
+ NF_CT_STAT_INC(net, delete_list);
+ local_bh_enable();
}
static void death_by_event(unsigned long ul_conntrack)
* Warning :
* - Caller must take a reference on returned object
* and recheck nf_ct_tuple_equal(tuple, &h->tuple)
- * OR
- * - Caller must lock nf_conntrack_lock before calling this function
*/
static struct nf_conntrack_tuple_hash *
____nf_conntrack_find(struct net *net, u16 zone,
struct nf_conntrack_tuple_hash *h;
struct hlist_nulls_node *n;
u16 zone;
+ unsigned int sequence;
zone = nf_ct_zone(ct);
- hash = hash_conntrack(net, zone,
- &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
- reply_hash = hash_conntrack(net, zone,
- &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
- spin_lock_bh(&nf_conntrack_lock);
+ local_bh_disable();
+ do {
+ sequence = read_seqcount_begin(&net->ct.generation);
+ hash = hash_conntrack(net, zone,
+ &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
+ reply_hash = hash_conntrack(net, zone,
+ &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
+ } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
/* See if there's one in the list already, including reverse */
hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
/* The caller holds a reference to this object */
atomic_set(&ct->ct_general.use, 2);
__nf_conntrack_hash_insert(ct, hash, reply_hash);
+ nf_conntrack_double_unlock(hash, reply_hash);
NF_CT_STAT_INC(net, insert);
- spin_unlock_bh(&nf_conntrack_lock);
-
+ local_bh_enable();
return 0;
out:
+ nf_conntrack_double_unlock(hash, reply_hash);
NF_CT_STAT_INC(net, insert_failed);
- spin_unlock_bh(&nf_conntrack_lock);
+ local_bh_enable();
return -EEXIST;
}
EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
enum ip_conntrack_info ctinfo;
struct net *net;
u16 zone;
+ unsigned int sequence;
ct = nf_ct_get(skb, &ctinfo);
net = nf_ct_net(ct);
return NF_ACCEPT;
zone = nf_ct_zone(ct);
- /* reuse the hash saved before */
- hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
- hash = hash_bucket(hash, net);
- reply_hash = hash_conntrack(net, zone,
- &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
+ local_bh_disable();
+
+ do {
+ sequence = read_seqcount_begin(&net->ct.generation);
+ /* reuse the hash saved before */
+ hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
+ hash = hash_bucket(hash, net);
+ reply_hash = hash_conntrack(net, zone,
+ &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
+
+ } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
/* We're not in hash table, and we refuse to set up related
- connections for unconfirmed conns. But packet copies and
- REJECT will give spurious warnings here. */
+ * connections for unconfirmed conns. But packet copies and
+ * REJECT will give spurious warnings here.
+ */
/* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
/* No external references means no one else could have
- confirmed us. */
+ * confirmed us.
+ */
NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
pr_debug("Confirming conntrack %p\n", ct);
-
- spin_lock_bh(&nf_conntrack_lock);
-
/* We have to check the DYING flag inside the lock to prevent
a race against nf_ct_get_next_corpse() possibly called from
user context, else we insert an already 'dead' hash, blocking
further use of that particular connection -JM */
if (unlikely(nf_ct_is_dying(ct))) {
- spin_unlock_bh(&nf_conntrack_lock);
+ nf_conntrack_double_unlock(hash, reply_hash);
+ local_bh_enable();
return NF_ACCEPT;
}
* stores are visible.
*/
__nf_conntrack_hash_insert(ct, hash, reply_hash);
+ nf_conntrack_double_unlock(hash, reply_hash);
NF_CT_STAT_INC(net, insert);
- spin_unlock_bh(&nf_conntrack_lock);
+ local_bh_enable();
help = nfct_help(ct);
if (help && help->helper)
return NF_ACCEPT;
out:
+ nf_conntrack_double_unlock(hash, reply_hash);
NF_CT_STAT_INC(net, insert_failed);
- spin_unlock_bh(&nf_conntrack_lock);
+ local_bh_enable();
return NF_DROP;
}
EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
/* There's a small race here where we may free a just-assured
connection. Too bad: we're in trouble anyway. */
-static noinline int early_drop(struct net *net, unsigned int hash)
+static noinline int early_drop(struct net *net, unsigned int _hash)
{
/* Use oldest entry, which is roughly LRU */
struct nf_conntrack_tuple_hash *h;
struct nf_conn *ct = NULL, *tmp;
struct hlist_nulls_node *n;
- unsigned int i, cnt = 0;
+ unsigned int i = 0, cnt = 0;
int dropped = 0;
+ unsigned int hash, sequence;
+ spinlock_t *lockp;
- rcu_read_lock();
- for (i = 0; i < net->ct.htable_size; i++) {
+ local_bh_disable();
+restart:
+ sequence = read_seqcount_begin(&net->ct.generation);
+ hash = hash_bucket(_hash, net);
+ for (; i < net->ct.htable_size; i++) {
+ lockp = &nf_conntrack_locks[hash % CONNTRACK_LOCKS];
+ spin_lock(lockp);
+ if (read_seqcount_retry(&net->ct.generation, sequence)) {
+ spin_unlock(lockp);
+ goto restart;
+ }
hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
hnnode) {
tmp = nf_ct_tuplehash_to_ctrack(h);
- if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
+ if (!test_bit(IPS_ASSURED_BIT, &tmp->status) &&
+ !nf_ct_is_dying(tmp) &&
+ atomic_inc_not_zero(&tmp->ct_general.use)) {
ct = tmp;
+ break;
+ }
cnt++;
}
- if (ct != NULL) {
- if (likely(!nf_ct_is_dying(ct) &&
- atomic_inc_not_zero(&ct->ct_general.use)))
- break;
- else
- ct = NULL;
- }
+ hash = (hash + 1) % net->ct.htable_size;
+ spin_unlock(lockp);
- if (cnt >= NF_CT_EVICTION_RANGE)
+ if (ct || cnt >= NF_CT_EVICTION_RANGE)
break;
- hash = (hash + 1) % net->ct.htable_size;
}
- rcu_read_unlock();
+ local_bh_enable();
if (!ct)
return dropped;
if (nf_conntrack_max &&
unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
- if (!early_drop(net, hash_bucket(hash, net))) {
+ if (!early_drop(net, hash)) {
atomic_dec(&net->ct.count);
net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
return ERR_PTR(-ENOMEM);
struct nf_conn *ct;
struct hlist_nulls_node *n;
int cpu;
+ spinlock_t *lockp;
- spin_lock_bh(&nf_conntrack_lock);
for (; *bucket < net->ct.htable_size; (*bucket)++) {
- hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
- if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
- continue;
- ct = nf_ct_tuplehash_to_ctrack(h);
- if (iter(ct, data))
- goto found;
+ lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS];
+ local_bh_disable();
+ spin_lock(lockp);
+ if (*bucket < net->ct.htable_size) {
+ hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
+ if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
+ continue;
+ ct = nf_ct_tuplehash_to_ctrack(h);
+ if (iter(ct, data))
+ goto found;
+ }
}
+ spin_unlock(lockp);
+ local_bh_enable();
}
- spin_unlock_bh(&nf_conntrack_lock);
for_each_possible_cpu(cpu) {
struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
return NULL;
found:
atomic_inc(&ct->ct_general.use);
- spin_unlock_bh(&nf_conntrack_lock);
+ spin_unlock(lockp);
+ local_bh_enable();
return ct;
}
if (!hash)
return -ENOMEM;
+ local_bh_disable();
+ nf_conntrack_all_lock();
+ write_seqcount_begin(&init_net.ct.generation);
+
/* Lookups in the old hash might happen in parallel, which means we
* might get false negatives during connection lookup. New connections
* created because of a false negative won't make it into the hash
- * though since that required taking the lock.
+ * though since that required taking the locks.
*/
- spin_lock_bh(&nf_conntrack_lock);
+
for (i = 0; i < init_net.ct.htable_size; i++) {
while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
h = hlist_nulls_entry(init_net.ct.hash[i].first,
init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
init_net.ct.hash = hash;
- spin_unlock_bh(&nf_conntrack_lock);
+
+ write_seqcount_end(&init_net.ct.generation);
+ nf_conntrack_all_unlock();
+ local_bh_enable();
nf_ct_free_hashtable(old_hash, old_size);
return 0;
int nf_conntrack_init_start(void)
{
int max_factor = 8;
- int ret, cpu;
+ int i, ret, cpu;
+
+ for (i = 0; i < ARRAY_SIZE(nf_conntrack_locks); i++)
+ spin_lock_init(&nf_conntrack_locks[i]);
/* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
* machine has 512 buckets. >= 1GB machines have 16384 buckets. */
projects / karo-tx-linux.git / blobdiff