summaryrefslogtreecommitdiff
path: root/net/ipv4
diff options
context:
space:
mode:
authorEric Dumazet <edumazet@google.com>2018-03-31 12:58:49 -0700
committerDavid S. Miller <davem@davemloft.net>2018-03-31 23:25:39 -0400
commit648700f76b03b7e8149d13cc2bdb3355035258a9 (patch)
tree713b21f58b23762ad32bdf3ea26b373e287d8247 /net/ipv4
parentae6da1f503abb5a5081f9f6c4a6881de97830f3e (diff)
inet: frags: use rhashtables for reassembly units
Some applications still rely on IP fragmentation, and to be fair linux reassembly unit is not working under any serious load. It uses static hash tables of 1024 buckets, and up to 128 items per bucket (!!!) A work queue is supposed to garbage collect items when host is under memory pressure, and doing a hash rebuild, changing seed used in hash computations. This work queue blocks softirqs for up to 25 ms when doing a hash rebuild, occurring every 5 seconds if host is under fire. Then there is the problem of sharing this hash table for all netns. It is time to switch to rhashtables, and allocate one of them per netns to speedup netns dismantle, since this is a critical metric these days. Lookup is now using RCU. A followup patch will even remove the refcount hold/release left from prior implementation and save a couple of atomic operations. Before this patch, 16 cpus (16 RX queue NIC) could not handle more than 1 Mpps frags DDOS. After the patch, I reach 9 Mpps without any tuning, and can use up to 2GB of storage for the fragments (exact number depends on frags being evicted after timeout) $ grep FRAG /proc/net/sockstat FRAG: inuse 1966916 memory 2140004608 A followup patch will change the limits for 64bit arches. Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Florian Westphal <fw@strlen.de> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Alexander Aring <alex.aring@gmail.com> Cc: Stefan Schmidt <stefan@osg.samsung.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat (limited to 'net/ipv4')
-rw-r--r--net/ipv4/inet_fragment.c344
-rw-r--r--net/ipv4/ip_fragment.c112
2 files changed, 115 insertions, 341 deletions
diff --git a/net/ipv4/inet_fragment.c b/net/ipv4/inet_fragment.c
index 1ac69f65d0de..ebb8f411e0db 100644
--- a/net/ipv4/inet_fragment.c
+++ b/net/ipv4/inet_fragment.c
@@ -25,12 +25,6 @@
#include <net/inet_frag.h>
#include <net/inet_ecn.h>
-#define INETFRAGS_EVICT_BUCKETS 128
-#define INETFRAGS_EVICT_MAX 512
-
-/* don't rebuild inetfrag table with new secret more often than this */
-#define INETFRAGS_MIN_REBUILD_INTERVAL (5 * HZ)
-
/* Given the OR values of all fragments, apply RFC 3168 5.3 requirements
* Value : 0xff if frame should be dropped.
* 0 or INET_ECN_CE value, to be ORed in to final iph->tos field
@@ -52,157 +46,8 @@ const u8 ip_frag_ecn_table[16] = {
};
EXPORT_SYMBOL(ip_frag_ecn_table);
-static unsigned int
-inet_frag_hashfn(const struct inet_frags *f, const struct inet_frag_queue *q)
-{
- return f->hashfn(q) & (INETFRAGS_HASHSZ - 1);
-}
-
-static bool inet_frag_may_rebuild(struct inet_frags *f)
-{
- return time_after(jiffies,
- f->last_rebuild_jiffies + INETFRAGS_MIN_REBUILD_INTERVAL);
-}
-
-static void inet_frag_secret_rebuild(struct inet_frags *f)
-{
- int i;
-
- write_seqlock_bh(&f->rnd_seqlock);
-
- if (!inet_frag_may_rebuild(f))
- goto out;
-
- get_random_bytes(&f->rnd, sizeof(u32));
-
- for (i = 0; i < INETFRAGS_HASHSZ; i++) {
- struct inet_frag_bucket *hb;
- struct inet_frag_queue *q;
- struct hlist_node *n;
-
- hb = &f->hash[i];
- spin_lock(&hb->chain_lock);
-
- hlist_for_each_entry_safe(q, n, &hb->chain, list) {
- unsigned int hval = inet_frag_hashfn(f, q);
-
- if (hval != i) {
- struct inet_frag_bucket *hb_dest;
-
- hlist_del(&q->list);
-
- /* Relink to new hash chain. */
- hb_dest = &f->hash[hval];
-
- /* This is the only place where we take
- * another chain_lock while already holding
- * one. As this will not run concurrently,
- * we cannot deadlock on hb_dest lock below, if its
- * already locked it will be released soon since
- * other caller cannot be waiting for hb lock
- * that we've taken above.
- */
- spin_lock_nested(&hb_dest->chain_lock,
- SINGLE_DEPTH_NESTING);
- hlist_add_head(&q->list, &hb_dest->chain);
- spin_unlock(&hb_dest->chain_lock);
- }
- }
- spin_unlock(&hb->chain_lock);
- }
-
- f->rebuild = false;
- f->last_rebuild_jiffies = jiffies;
-out:
- write_sequnlock_bh(&f->rnd_seqlock);
-}
-
-static bool inet_fragq_should_evict(const struct inet_frag_queue *q)
-{
- if (!hlist_unhashed(&q->list_evictor))
- return false;
-
- return q->net->low_thresh == 0 ||
- frag_mem_limit(q->net) >= q->net->low_thresh;
-}
-
-static unsigned int
-inet_evict_bucket(struct inet_frags *f, struct inet_frag_bucket *hb)
-{
- struct inet_frag_queue *fq;
- struct hlist_node *n;
- unsigned int evicted = 0;
- HLIST_HEAD(expired);
-
- spin_lock(&hb->chain_lock);
-
- hlist_for_each_entry_safe(fq, n, &hb->chain, list) {
- if (!inet_fragq_should_evict(fq))
- continue;
-
- if (!del_timer(&fq->timer))
- continue;
-
- hlist_add_head(&fq->list_evictor, &expired);
- ++evicted;
- }
-
- spin_unlock(&hb->chain_lock);
-
- hlist_for_each_entry_safe(fq, n, &expired, list_evictor)
- f->frag_expire(&fq->timer);
-
- return evicted;
-}
-
-static void inet_frag_worker(struct work_struct *work)
-{
- unsigned int budget = INETFRAGS_EVICT_BUCKETS;
- unsigned int i, evicted = 0;
- struct inet_frags *f;
-
- f = container_of(work, struct inet_frags, frags_work);
-
- BUILD_BUG_ON(INETFRAGS_EVICT_BUCKETS >= INETFRAGS_HASHSZ);
-
- local_bh_disable();
-
- for (i = READ_ONCE(f->next_bucket); budget; --budget) {
- evicted += inet_evict_bucket(f, &f->hash[i]);
- i = (i + 1) & (INETFRAGS_HASHSZ - 1);
- if (evicted > INETFRAGS_EVICT_MAX)
- break;
- }
-
- f->next_bucket = i;
-
- local_bh_enable();
-
- if (f->rebuild && inet_frag_may_rebuild(f))
- inet_frag_secret_rebuild(f);
-}
-
-static void inet_frag_schedule_worker(struct inet_frags *f)
-{
- if (unlikely(!work_pending(&f->frags_work)))
- schedule_work(&f->frags_work);
-}
-
int inet_frags_init(struct inet_frags *f)
{
- int i;
-
- INIT_WORK(&f->frags_work, inet_frag_worker);
-
- for (i = 0; i < INETFRAGS_HASHSZ; i++) {
- struct inet_frag_bucket *hb = &f->hash[i];
-
- spin_lock_init(&hb->chain_lock);
- INIT_HLIST_HEAD(&hb->chain);
- }
-
- seqlock_init(&f->rnd_seqlock);
- f->last_rebuild_jiffies = 0;
f->frags_cachep = kmem_cache_create(f->frags_cache_name, f->qsize, 0, 0,
NULL);
if (!f->frags_cachep)
@@ -214,66 +59,42 @@ EXPORT_SYMBOL(inet_frags_init);
void inet_frags_fini(struct inet_frags *f)
{
- cancel_work_sync(&f->frags_work);
+ /* We must wait that all inet_frag_destroy_rcu() have completed. */
+ rcu_barrier();
+
kmem_cache_destroy(f->frags_cachep);
+ f->frags_cachep = NULL;
}
EXPORT_SYMBOL(inet_frags_fini);
-void inet_frags_exit_net(struct netns_frags *nf)
-{
- struct inet_frags *f =nf->f;
- unsigned int seq;
- int i;
-
- nf->low_thresh = 0;
-
-evict_again:
- local_bh_disable();
- seq = read_seqbegin(&f->rnd_seqlock);
-
- for (i = 0; i < INETFRAGS_HASHSZ ; i++)
- inet_evict_bucket(f, &f->hash[i]);
-
- local_bh_enable();
- cond_resched();
-
- if (read_seqretry(&f->rnd_seqlock, seq) ||
- sum_frag_mem_limit(nf))
- goto evict_again;
-}
-EXPORT_SYMBOL(inet_frags_exit_net);
-
-static struct inet_frag_bucket *
-get_frag_bucket_locked(struct inet_frag_queue *fq, struct inet_frags *f)
-__acquires(hb->chain_lock)
+static void inet_frags_free_cb(void *ptr, void *arg)
{
- struct inet_frag_bucket *hb;
- unsigned int seq, hash;
+ struct inet_frag_queue *fq = ptr;
- restart:
- seq = read_seqbegin(&f->rnd_seqlock);
-
- hash = inet_frag_hashfn(f, fq);
- hb = &f->hash[hash];
+ /* If we can not cancel the timer, it means this frag_queue
+ * is already disappearing, we have nothing to do.
+ * Otherwise, we own a refcount until the end of this function.
+ */
+ if (!del_timer(&fq->timer))
+ return;
- spin_lock(&hb->chain_lock);
- if (read_seqretry(&f->rnd_seqlock, seq)) {
- spin_unlock(&hb->chain_lock);
- goto restart;
+ spin_lock_bh(&fq->lock);
+ if (!(fq->flags & INET_FRAG_COMPLETE)) {
+ fq->flags |= INET_FRAG_COMPLETE;
+ refcount_dec(&fq->refcnt);
}
+ spin_unlock_bh(&fq->lock);
- return hb;
+ inet_frag_put(fq);
}
-static inline void fq_unlink(struct inet_frag_queue *fq)
+void inet_frags_exit_net(struct netns_frags *nf)
{
- struct inet_frag_bucket *hb;
+ nf->low_thresh = 0; /* prevent creation of new frags */
- hb = get_frag_bucket_locked(fq, fq->net->f);
- hlist_del(&fq->list);
- fq->flags |= INET_FRAG_COMPLETE;
- spin_unlock(&hb->chain_lock);
+ rhashtable_free_and_destroy(&nf->rhashtable, inet_frags_free_cb, NULL);
}
+EXPORT_SYMBOL(inet_frags_exit_net);
void inet_frag_kill(struct inet_frag_queue *fq)
{
@@ -281,12 +102,26 @@ void inet_frag_kill(struct inet_frag_queue *fq)
refcount_dec(&fq->refcnt);
if (!(fq->flags & INET_FRAG_COMPLETE)) {
- fq_unlink(fq);
+ struct netns_frags *nf = fq->net;
+
+ fq->flags |= INET_FRAG_COMPLETE;
+ rhashtable_remove_fast(&nf->rhashtable, &fq->node, nf->f->rhash_params);
refcount_dec(&fq->refcnt);
}
}
EXPORT_SYMBOL(inet_frag_kill);
+static void inet_frag_destroy_rcu(struct rcu_head *head)
+{
+ struct inet_frag_queue *q = container_of(head, struct inet_frag_queue,
+ rcu);
+ struct inet_frags *f = q->net->f;
+
+ if (f->destructor)
+ f->destructor(q);
+ kmem_cache_free(f->frags_cachep, q);
+}
+
void inet_frag_destroy(struct inet_frag_queue *q)
{
struct sk_buff *fp;
@@ -310,59 +145,20 @@ void inet_frag_destroy(struct inet_frag_queue *q)
}
sum = sum_truesize + f->qsize;
- if (f->destructor)
- f->destructor(q);
- kmem_cache_free(f->frags_cachep, q);
+ call_rcu(&q->rcu, inet_frag_destroy_rcu);
sub_frag_mem_limit(nf, sum);
}
EXPORT_SYMBOL(inet_frag_destroy);
-static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf,
- struct inet_frag_queue *qp_in,
- struct inet_frags *f,
- void *arg)
-{
- struct inet_frag_bucket *hb = get_frag_bucket_locked(qp_in, f);
- struct inet_frag_queue *qp;
-
-#ifdef CONFIG_SMP
- /* With SMP race we have to recheck hash table, because
- * such entry could have been created on other cpu before
- * we acquired hash bucket lock.
- */
- hlist_for_each_entry(qp, &hb->chain, list) {
- if (qp->net == nf && f->match(qp, arg)) {
- refcount_inc(&qp->refcnt);
- spin_unlock(&hb->chain_lock);
- qp_in->flags |= INET_FRAG_COMPLETE;
- inet_frag_put(qp_in);
- return qp;
- }
- }
-#endif
- qp = qp_in;
- if (!mod_timer(&qp->timer, jiffies + nf->timeout))
- refcount_inc(&qp->refcnt);
-
- refcount_inc(&qp->refcnt);
- hlist_add_head(&qp->list, &hb->chain);
-
- spin_unlock(&hb->chain_lock);
-
- return qp;
-}
-
static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf,
struct inet_frags *f,
void *arg)
{
struct inet_frag_queue *q;
- if (!nf->high_thresh || frag_mem_limit(nf) > nf->high_thresh) {
- inet_frag_schedule_worker(f);
+ if (!nf->high_thresh || frag_mem_limit(nf) > nf->high_thresh)
return NULL;
- }
q = kmem_cache_zalloc(f->frags_cachep, GFP_ATOMIC);
if (!q)
@@ -374,59 +170,52 @@ static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf,
timer_setup(&q->timer, f->frag_expire, 0);
spin_lock_init(&q->lock);
- refcount_set(&q->refcnt, 1);
+ refcount_set(&q->refcnt, 3);
return q;
}
static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf,
- struct inet_frags *f,
void *arg)
{
+ struct inet_frags *f = nf->f;
struct inet_frag_queue *q;
+ int err;
q = inet_frag_alloc(nf, f, arg);
if (!q)
return NULL;
- return inet_frag_intern(nf, q, f, arg);
+ mod_timer(&q->timer, jiffies + nf->timeout);
+
+ err = rhashtable_insert_fast(&nf->rhashtable, &q->node,
+ f->rhash_params);
+ if (err < 0) {
+ q->flags |= INET_FRAG_COMPLETE;
+ inet_frag_kill(q);
+ inet_frag_destroy(q);
+ return NULL;
+ }
+ return q;
}
-struct inet_frag_queue *inet_frag_find(struct netns_frags *nf,
- struct inet_frags *f, void *key,
- unsigned int hash)
+/* TODO : call from rcu_read_lock() and no longer use refcount_inc_not_zero() */
+struct inet_frag_queue *inet_frag_find(struct netns_frags *nf, void *key)
{
- struct inet_frag_bucket *hb;
- struct inet_frag_queue *q;
- int depth = 0;
-
- if (frag_mem_limit(nf) > nf->low_thresh)
- inet_frag_schedule_worker(f);
-
- hash &= (INETFRAGS_HASHSZ - 1);
- hb = &f->hash[hash];
-
- spin_lock(&hb->chain_lock);
- hlist_for_each_entry(q, &hb->chain, list) {
- if (q->net == nf && f->match(q, key)) {
- refcount_inc(&q->refcnt);
- spin_unlock(&hb->chain_lock);
- return q;
- }
- depth++;
- }
- spin_unlock(&hb->chain_lock);
+ struct inet_frag_queue *fq;
- if (depth <= INETFRAGS_MAXDEPTH)
- return inet_frag_create(nf, f, key);
+ rcu_read_lock();
- if (inet_frag_may_rebuild(f)) {
- if (!f->rebuild)
- f->rebuild = true;
- inet_frag_schedule_worker(f);
+ fq = rhashtable_lookup(&nf->rhashtable, key, nf->f->rhash_params);
+ if (fq) {
+ if (!refcount_inc_not_zero(&fq->refcnt))
+ fq = NULL;
+ rcu_read_unlock();
+ return fq;
}
+ rcu_read_unlock();
- return ERR_PTR(-ENOBUFS);
+ return inet_frag_create(nf, key);
}
EXPORT_SYMBOL(inet_frag_find);
@@ -434,8 +223,7 @@ void inet_frag_maybe_warn_overflow(struct inet_frag_queue *q,
const char *prefix)
{
static const char msg[] = "inet_frag_find: Fragment hash bucket"
- " list length grew over limit " __stringify(INETFRAGS_MAXDEPTH)
- ". Dropping fragment.\n";
+ " list length grew over limit. Dropping fragment.\n";
if (PTR_ERR(q) == -ENOBUFS)
net_dbg_ratelimited("%s%s", prefix, msg);
diff --git a/net/ipv4/ip_fragment.c b/net/ipv4/ip_fragment.c
index 1a3bc85d6f5e..4021820db6f2 100644
--- a/net/ipv4/ip_fragment.c
+++ b/net/ipv4/ip_fragment.c
@@ -69,15 +69,9 @@ struct ipfrag_skb_cb
struct ipq {
struct inet_frag_queue q;
- u32 user;
- __be32 saddr;
- __be32 daddr;
- __be16 id;
- u8 protocol;
u8 ecn; /* RFC3168 support */
u16 max_df_size; /* largest frag with DF set seen */
int iif;
- int vif; /* L3 master device index */
unsigned int rid;
struct inet_peer *peer;
};
@@ -97,41 +91,6 @@ int ip_frag_mem(struct net *net)
static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
struct net_device *dev);
-struct ip4_create_arg {
- struct iphdr *iph;
- u32 user;
- int vif;
-};
-
-static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
-{
- net_get_random_once(&ip4_frags.rnd, sizeof(ip4_frags.rnd));
- return jhash_3words((__force u32)id << 16 | prot,
- (__force u32)saddr, (__force u32)daddr,
- ip4_frags.rnd);
-}
-
-static unsigned int ip4_hashfn(const struct inet_frag_queue *q)
-{
- const struct ipq *ipq;
-
- ipq = container_of(q, struct ipq, q);
- return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
-}
-
-static bool ip4_frag_match(const struct inet_frag_queue *q, const void *a)
-{
- const struct ipq *qp;
- const struct ip4_create_arg *arg = a;
-
- qp = container_of(q, struct ipq, q);
- return qp->id == arg->iph->id &&
- qp->saddr == arg->iph->saddr &&
- qp->daddr == arg->iph->daddr &&
- qp->protocol == arg->iph->protocol &&
- qp->user == arg->user &&
- qp->vif == arg->vif;
-}
static void ip4_frag_init(struct inet_frag_queue *q, const void *a)
{
@@ -140,17 +99,12 @@ static void ip4_frag_init(struct inet_frag_queue *q, const void *a)
frags);
struct net *net = container_of(ipv4, struct net, ipv4);
- const struct ip4_create_arg *arg = a;
+ const struct frag_v4_compare_key *key = a;
- qp->protocol = arg->iph->protocol;
- qp->id = arg->iph->id;
- qp->ecn = ip4_frag_ecn(arg->iph->tos);
- qp->saddr = arg->iph->saddr;
- qp->daddr = arg->iph->daddr;
- qp->vif = arg->vif;
- qp->user = arg->user;
+ q->key.v4 = *key;
+ qp->ecn = 0;
qp->peer = q->net->max_dist ?
- inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, arg->vif, 1) :
+ inet_getpeer_v4(net->ipv4.peers, key->saddr, key->vif, 1) :
NULL;
}
@@ -234,7 +188,7 @@ static void ip_expire(struct timer_list *t)
/* Only an end host needs to send an ICMP
* "Fragment Reassembly Timeout" message, per RFC792.
*/
- if (frag_expire_skip_icmp(qp->user) &&
+ if (frag_expire_skip_icmp(qp->q.key.v4.user) &&
(skb_rtable(head)->rt_type != RTN_LOCAL))
goto out;
@@ -262,17 +216,17 @@ out_rcu_unlock:
static struct ipq *ip_find(struct net *net, struct iphdr *iph,
u32 user, int vif)
{
+ struct frag_v4_compare_key key = {
+ .saddr = iph->saddr,
+ .daddr = iph->daddr,
+ .user = user,
+ .vif = vif,
+ .id = iph->id,
+ .protocol = iph->protocol,
+ };
struct inet_frag_queue *q;
- struct ip4_create_arg arg;
- unsigned int hash;
-
- arg.iph = iph;
- arg.user = user;
- arg.vif = vif;
- hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
-
- q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
+ q = inet_frag_find(&net->ipv4.frags, &key);
if (IS_ERR_OR_NULL(q)) {
inet_frag_maybe_warn_overflow(q, pr_fmt());
return NULL;
@@ -656,7 +610,7 @@ out_nomem:
err = -ENOMEM;
goto out_fail;
out_oversize:
- net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
+ net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->q.key.v4.saddr);
out_fail:
__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
return err;
@@ -894,15 +848,47 @@ static struct pernet_operations ip4_frags_ops = {
.exit = ipv4_frags_exit_net,
};
+
+static u32 ip4_key_hashfn(const void *data, u32 len, u32 seed)
+{
+ return jhash2(data,
+ sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
+}
+
+static u32 ip4_obj_hashfn(const void *data, u32 len, u32 seed)
+{
+ const struct inet_frag_queue *fq = data;
+
+ return jhash2((const u32 *)&fq->key.v4,
+ sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
+}
+
+static int ip4_obj_cmpfn(struct rhashtable_compare_arg *arg, const void *ptr)
+{
+ const struct frag_v4_compare_key *key = arg->key;
+ const struct inet_frag_queue *fq = ptr;
+
+ return !!memcmp(&fq->key, key, sizeof(*key));
+}
+
+static const struct rhashtable_params ip4_rhash_params = {
+ .head_offset = offsetof(struct inet_frag_queue, node),
+ .key_offset = offsetof(struct inet_frag_queue, key),
+ .key_len = sizeof(struct frag_v4_compare_key),
+ .hashfn = ip4_key_hashfn,
+ .obj_hashfn = ip4_obj_hashfn,
+ .obj_cmpfn = ip4_obj_cmpfn,
+ .automatic_shrinking = true,
+};
+
void __init ipfrag_init(void)
{
- ip4_frags.hashfn = ip4_hashfn;
ip4_frags.constructor = ip4_frag_init;
ip4_frags.destructor = ip4_frag_free;
ip4_frags.qsize = sizeof(struct ipq);
- ip4_frags.match = ip4_frag_match;
ip4_frags.frag_expire = ip_expire;
ip4_frags.frags_cache_name = ip_frag_cache_name;
+ ip4_frags.rhash_params = ip4_rhash_params;
if (inet_frags_init(&ip4_frags))
panic("IP: failed to allocate ip4_frags cache\n");
ip4_frags_ctl_register();