Revert "net: Add a second bind table hashed by port and address"
authorJoanne Koong <joannelkoong@gmail.com>
Wed, 15 Jun 2022 19:32:13 +0000 (12:32 -0700)
committerJakub Kicinski <kuba@kernel.org>
Thu, 16 Jun 2022 18:07:59 +0000 (11:07 -0700)
This reverts:

commit d5a42de8bdbe ("net: Add a second bind table hashed by port and address")
commit 538aaf9b2383 ("selftests: Add test for timing a bind request to a port with a populated bhash entry")
Link: https://lore.kernel.org/netdev/20220520001834.2247810-1-kuba@kernel.org/
There are a few things that need to be fixed here:
* Updating bhash2 in cases where the socket's rcv saddr changes
* Adding bhash2 hashbucket locks

Links to syzbot reports:
https://lore.kernel.org/netdev/00000000000022208805e0df247a@google.com/
https://lore.kernel.org/netdev/0000000000003f33bc05dfaf44fe@google.com/

Fixes: d5a42de8bdbe ("net: Add a second bind table hashed by port and address")
Reported-by: syzbot+015d756bbd1f8b5c8f09@syzkaller.appspotmail.com
Reported-by: syzbot+98fd2d1422063b0f8c44@syzkaller.appspotmail.com
Reported-by: syzbot+0a847a982613c6438fba@syzkaller.appspotmail.com
Signed-off-by: Joanne Koong <joannelkoong@gmail.com>
Link: https://lore.kernel.org/r/20220615193213.2419568-1-joannelkoong@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
include/net/inet_connection_sock.h
include/net/inet_hashtables.h
include/net/sock.h
net/dccp/proto.c
net/ipv4/inet_connection_sock.c
net/ipv4/inet_hashtables.c
net/ipv4/tcp.c
tools/testing/selftests/net/.gitignore
tools/testing/selftests/net/Makefile
tools/testing/selftests/net/bind_bhash_test.c [deleted file]

index 077cd73..85cd695 100644 (file)
@@ -25,7 +25,6 @@
 #undef INET_CSK_CLEAR_TIMERS
 
 struct inet_bind_bucket;
-struct inet_bind2_bucket;
 struct tcp_congestion_ops;
 
 /*
@@ -58,7 +57,6 @@ struct inet_connection_sock_af_ops {
  *
  * @icsk_accept_queue:    FIFO of established children
  * @icsk_bind_hash:       Bind node
- * @icsk_bind2_hash:      Bind node in the bhash2 table
  * @icsk_timeout:         Timeout
  * @icsk_retransmit_timer: Resend (no ack)
  * @icsk_rto:             Retransmit timeout
@@ -85,7 +83,6 @@ struct inet_connection_sock {
        struct inet_sock          icsk_inet;
        struct request_sock_queue icsk_accept_queue;
        struct inet_bind_bucket   *icsk_bind_hash;
-       struct inet_bind2_bucket  *icsk_bind2_hash;
        unsigned long             icsk_timeout;
        struct timer_list         icsk_retransmit_timer;
        struct timer_list         icsk_delack_timer;
index a0887b7..ebfa3df 100644 (file)
@@ -90,32 +90,11 @@ struct inet_bind_bucket {
        struct hlist_head       owners;
 };
 
-struct inet_bind2_bucket {
-       possible_net_t          ib_net;
-       int                     l3mdev;
-       unsigned short          port;
-       union {
-#if IS_ENABLED(CONFIG_IPV6)
-               struct in6_addr         v6_rcv_saddr;
-#endif
-               __be32                  rcv_saddr;
-       };
-       /* Node in the inet2_bind_hashbucket chain */
-       struct hlist_node       node;
-       /* List of sockets hashed to this bucket */
-       struct hlist_head       owners;
-};
-
 static inline struct net *ib_net(struct inet_bind_bucket *ib)
 {
        return read_pnet(&ib->ib_net);
 }
 
-static inline struct net *ib2_net(struct inet_bind2_bucket *ib)
-{
-       return read_pnet(&ib->ib_net);
-}
-
 #define inet_bind_bucket_for_each(tb, head) \
        hlist_for_each_entry(tb, head, node)
 
@@ -124,15 +103,6 @@ struct inet_bind_hashbucket {
        struct hlist_head       chain;
 };
 
-/* This is synchronized using the inet_bind_hashbucket's spinlock.
- * Instead of having separate spinlocks, the inet_bind2_hashbucket can share
- * the inet_bind_hashbucket's given that in every case where the bhash2 table
- * is useful, a lookup in the bhash table also occurs.
- */
-struct inet_bind2_hashbucket {
-       struct hlist_head       chain;
-};
-
 /* Sockets can be hashed in established or listening table.
  * We must use different 'nulls' end-of-chain value for all hash buckets :
  * A socket might transition from ESTABLISH to LISTEN state without
@@ -164,12 +134,6 @@ struct inet_hashinfo {
         */
        struct kmem_cache               *bind_bucket_cachep;
        struct inet_bind_hashbucket     *bhash;
-       /* The 2nd binding table hashed by port and address.
-        * This is used primarily for expediting the resolution of bind
-        * conflicts.
-        */
-       struct kmem_cache               *bind2_bucket_cachep;
-       struct inet_bind2_hashbucket    *bhash2;
        unsigned int                    bhash_size;
 
        /* The 2nd listener table hashed by local port and address */
@@ -229,36 +193,6 @@ inet_bind_bucket_create(struct kmem_cache *cachep, struct net *net,
 void inet_bind_bucket_destroy(struct kmem_cache *cachep,
                              struct inet_bind_bucket *tb);
 
-static inline bool check_bind_bucket_match(struct inet_bind_bucket *tb,
-                                          struct net *net,
-                                          const unsigned short port,
-                                          int l3mdev)
-{
-       return net_eq(ib_net(tb), net) && tb->port == port &&
-               tb->l3mdev == l3mdev;
-}
-
-struct inet_bind2_bucket *
-inet_bind2_bucket_create(struct kmem_cache *cachep, struct net *net,
-                        struct inet_bind2_hashbucket *head,
-                        const unsigned short port, int l3mdev,
-                        const struct sock *sk);
-
-void inet_bind2_bucket_destroy(struct kmem_cache *cachep,
-                              struct inet_bind2_bucket *tb);
-
-struct inet_bind2_bucket *
-inet_bind2_bucket_find(struct inet_hashinfo *hinfo, struct net *net,
-                      const unsigned short port, int l3mdev,
-                      struct sock *sk,
-                      struct inet_bind2_hashbucket **head);
-
-bool check_bind2_bucket_match_nulladdr(struct inet_bind2_bucket *tb,
-                                      struct net *net,
-                                      const unsigned short port,
-                                      int l3mdev,
-                                      const struct sock *sk);
-
 static inline u32 inet_bhashfn(const struct net *net, const __u16 lport,
                               const u32 bhash_size)
 {
@@ -266,7 +200,7 @@ static inline u32 inet_bhashfn(const struct net *net, const __u16 lport,
 }
 
 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
-                   struct inet_bind2_bucket *tb2, const unsigned short snum);
+                   const unsigned short snum);
 
 /* Caller must disable local BH processing. */
 int __inet_inherit_port(const struct sock *sk, struct sock *child);
index c585ef6..72ca97c 100644 (file)
@@ -348,7 +348,6 @@ struct sk_filter;
   *    @sk_txtime_report_errors: set report errors mode for SO_TXTIME
   *    @sk_txtime_unused: unused txtime flags
   *    @ns_tracker: tracker for netns reference
-  *    @sk_bind2_node: bind node in the bhash2 table
   */
 struct sock {
        /*
@@ -538,7 +537,6 @@ struct sock {
 #endif
        struct rcu_head         sk_rcu;
        netns_tracker           ns_tracker;
-       struct hlist_node       sk_bind2_node;
 };
 
 enum sk_pacing {
@@ -819,16 +817,6 @@ static inline void sk_add_bind_node(struct sock *sk,
        hlist_add_head(&sk->sk_bind_node, list);
 }
 
-static inline void __sk_del_bind2_node(struct sock *sk)
-{
-       __hlist_del(&sk->sk_bind2_node);
-}
-
-static inline void sk_add_bind2_node(struct sock *sk, struct hlist_head *list)
-{
-       hlist_add_head(&sk->sk_bind2_node, list);
-}
-
 #define sk_for_each(__sk, list) \
        hlist_for_each_entry(__sk, list, sk_node)
 #define sk_for_each_rcu(__sk, list) \
@@ -846,8 +834,6 @@ static inline void sk_add_bind2_node(struct sock *sk, struct hlist_head *list)
        hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
 #define sk_for_each_bound(__sk, list) \
        hlist_for_each_entry(__sk, list, sk_bind_node)
-#define sk_for_each_bound_bhash2(__sk, list) \
-       hlist_for_each_entry(__sk, list, sk_bind2_node)
 
 /**
  * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
index 2e78458..eb8e128 100644 (file)
@@ -1120,12 +1120,6 @@ static int __init dccp_init(void)
                                  SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
        if (!dccp_hashinfo.bind_bucket_cachep)
                goto out_free_hashinfo2;
-       dccp_hashinfo.bind2_bucket_cachep =
-               kmem_cache_create("dccp_bind2_bucket",
-                                 sizeof(struct inet_bind2_bucket), 0,
-                                 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
-       if (!dccp_hashinfo.bind2_bucket_cachep)
-               goto out_free_bind_bucket_cachep;
 
        /*
         * Size and allocate the main established and bind bucket
@@ -1156,7 +1150,7 @@ static int __init dccp_init(void)
 
        if (!dccp_hashinfo.ehash) {
                DCCP_CRIT("Failed to allocate DCCP established hash table");
-               goto out_free_bind2_bucket_cachep;
+               goto out_free_bind_bucket_cachep;
        }
 
        for (i = 0; i <= dccp_hashinfo.ehash_mask; i++)
@@ -1182,23 +1176,14 @@ static int __init dccp_init(void)
                goto out_free_dccp_locks;
        }
 
-       dccp_hashinfo.bhash2 = (struct inet_bind2_hashbucket *)
-               __get_free_pages(GFP_ATOMIC | __GFP_NOWARN, bhash_order);
-
-       if (!dccp_hashinfo.bhash2) {
-               DCCP_CRIT("Failed to allocate DCCP bind2 hash table");
-               goto out_free_dccp_bhash;
-       }
-
        for (i = 0; i < dccp_hashinfo.bhash_size; i++) {
                spin_lock_init(&dccp_hashinfo.bhash[i].lock);
                INIT_HLIST_HEAD(&dccp_hashinfo.bhash[i].chain);
-               INIT_HLIST_HEAD(&dccp_hashinfo.bhash2[i].chain);
        }
 
        rc = dccp_mib_init();
        if (rc)
-               goto out_free_dccp_bhash2;
+               goto out_free_dccp_bhash;
 
        rc = dccp_ackvec_init();
        if (rc)
@@ -1222,38 +1207,30 @@ out_ackvec_exit:
        dccp_ackvec_exit();
 out_free_dccp_mib:
        dccp_mib_exit();
-out_free_dccp_bhash2:
-       free_pages((unsigned long)dccp_hashinfo.bhash2, bhash_order);
 out_free_dccp_bhash:
        free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
 out_free_dccp_locks:
        inet_ehash_locks_free(&dccp_hashinfo);
 out_free_dccp_ehash:
        free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order);
-out_free_bind2_bucket_cachep:
-       kmem_cache_destroy(dccp_hashinfo.bind2_bucket_cachep);
 out_free_bind_bucket_cachep:
        kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
 out_free_hashinfo2:
        inet_hashinfo2_free_mod(&dccp_hashinfo);
 out_fail:
        dccp_hashinfo.bhash = NULL;
-       dccp_hashinfo.bhash2 = NULL;
        dccp_hashinfo.ehash = NULL;
        dccp_hashinfo.bind_bucket_cachep = NULL;
-       dccp_hashinfo.bind2_bucket_cachep = NULL;
        return rc;
 }
 
 static void __exit dccp_fini(void)
 {
-       int bhash_order = get_order(dccp_hashinfo.bhash_size *
-                                   sizeof(struct inet_bind_hashbucket));
-
        ccid_cleanup_builtins();
        dccp_mib_exit();
-       free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
-       free_pages((unsigned long)dccp_hashinfo.bhash2, bhash_order);
+       free_pages((unsigned long)dccp_hashinfo.bhash,
+                  get_order(dccp_hashinfo.bhash_size *
+                            sizeof(struct inet_bind_hashbucket)));
        free_pages((unsigned long)dccp_hashinfo.ehash,
                   get_order((dccp_hashinfo.ehash_mask + 1) *
                             sizeof(struct inet_ehash_bucket)));
index c0b7e6c..53f5f95 100644 (file)
@@ -117,32 +117,6 @@ bool inet_rcv_saddr_any(const struct sock *sk)
        return !sk->sk_rcv_saddr;
 }
 
-static bool use_bhash2_on_bind(const struct sock *sk)
-{
-#if IS_ENABLED(CONFIG_IPV6)
-       int addr_type;
-
-       if (sk->sk_family == AF_INET6) {
-               addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr);
-               return addr_type != IPV6_ADDR_ANY &&
-                       addr_type != IPV6_ADDR_MAPPED;
-       }
-#endif
-       return sk->sk_rcv_saddr != htonl(INADDR_ANY);
-}
-
-static u32 get_bhash2_nulladdr_hash(const struct sock *sk, struct net *net,
-                                   int port)
-{
-#if IS_ENABLED(CONFIG_IPV6)
-       struct in6_addr nulladdr = {};
-
-       if (sk->sk_family == AF_INET6)
-               return ipv6_portaddr_hash(net, &nulladdr, port);
-#endif
-       return ipv4_portaddr_hash(net, 0, port);
-}
-
 void inet_get_local_port_range(struct net *net, int *low, int *high)
 {
        unsigned int seq;
@@ -156,71 +130,16 @@ void inet_get_local_port_range(struct net *net, int *low, int *high)
 }
 EXPORT_SYMBOL(inet_get_local_port_range);
 
-static bool bind_conflict_exist(const struct sock *sk, struct sock *sk2,
-                               kuid_t sk_uid, bool relax,
-                               bool reuseport_cb_ok, bool reuseport_ok)
-{
-       int bound_dev_if2;
-
-       if (sk == sk2)
-               return false;
-
-       bound_dev_if2 = READ_ONCE(sk2->sk_bound_dev_if);
-
-       if (!sk->sk_bound_dev_if || !bound_dev_if2 ||
-           sk->sk_bound_dev_if == bound_dev_if2) {
-               if (sk->sk_reuse && sk2->sk_reuse &&
-                   sk2->sk_state != TCP_LISTEN) {
-                       if (!relax || (!reuseport_ok && sk->sk_reuseport &&
-                                      sk2->sk_reuseport && reuseport_cb_ok &&
-                                      (sk2->sk_state == TCP_TIME_WAIT ||
-                                       uid_eq(sk_uid, sock_i_uid(sk2)))))
-                               return true;
-               } else if (!reuseport_ok || !sk->sk_reuseport ||
-                          !sk2->sk_reuseport || !reuseport_cb_ok ||
-                          (sk2->sk_state != TCP_TIME_WAIT &&
-                           !uid_eq(sk_uid, sock_i_uid(sk2)))) {
-                       return true;
-               }
-       }
-       return false;
-}
-
-static bool check_bhash2_conflict(const struct sock *sk,
-                                 struct inet_bind2_bucket *tb2, kuid_t sk_uid,
-                                 bool relax, bool reuseport_cb_ok,
-                                 bool reuseport_ok)
-{
-       struct sock *sk2;
-
-       sk_for_each_bound_bhash2(sk2, &tb2->owners) {
-               if (sk->sk_family == AF_INET && ipv6_only_sock(sk2))
-                       continue;
-
-               if (bind_conflict_exist(sk, sk2, sk_uid, relax,
-                                       reuseport_cb_ok, reuseport_ok))
-                       return true;
-       }
-       return false;
-}
-
-/* This should be called only when the corresponding inet_bind_bucket spinlock
- * is held
- */
-static int inet_csk_bind_conflict(const struct sock *sk, int port,
-                                 struct inet_bind_bucket *tb,
-                                 struct inet_bind2_bucket *tb2, /* may be null */
+static int inet_csk_bind_conflict(const struct sock *sk,
+                                 const struct inet_bind_bucket *tb,
                                  bool relax, bool reuseport_ok)
 {
-       struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
-       kuid_t uid = sock_i_uid((struct sock *)sk);
-       struct sock_reuseport *reuseport_cb;
-       struct inet_bind2_hashbucket *head2;
-       bool reuseport_cb_ok;
        struct sock *sk2;
-       struct net *net;
-       int l3mdev;
-       u32 hash;
+       bool reuseport_cb_ok;
+       bool reuse = sk->sk_reuse;
+       bool reuseport = !!sk->sk_reuseport;
+       struct sock_reuseport *reuseport_cb;
+       kuid_t uid = sock_i_uid((struct sock *)sk);
 
        rcu_read_lock();
        reuseport_cb = rcu_dereference(sk->sk_reuseport_cb);
@@ -231,42 +150,40 @@ static int inet_csk_bind_conflict(const struct sock *sk, int port,
        /*
         * Unlike other sk lookup places we do not check
         * for sk_net here, since _all_ the socks listed
-        * in tb->owners and tb2->owners list belong
-        * to the same net
+        * in tb->owners list belong to the same net - the
+        * one this bucket belongs to.
         */
 
-       if (!use_bhash2_on_bind(sk)) {
-               sk_for_each_bound(sk2, &tb->owners)
-                       if (bind_conflict_exist(sk, sk2, uid, relax,
-                                               reuseport_cb_ok, reuseport_ok) &&
-                           inet_rcv_saddr_equal(sk, sk2, true))
-                               return true;
+       sk_for_each_bound(sk2, &tb->owners) {
+               int bound_dev_if2;
 
-               return false;
+               if (sk == sk2)
+                       continue;
+               bound_dev_if2 = READ_ONCE(sk2->sk_bound_dev_if);
+               if ((!sk->sk_bound_dev_if ||
+                    !bound_dev_if2 ||
+                    sk->sk_bound_dev_if == bound_dev_if2)) {
+                       if (reuse && sk2->sk_reuse &&
+                           sk2->sk_state != TCP_LISTEN) {
+                               if ((!relax ||
+                                    (!reuseport_ok &&
+                                     reuseport && sk2->sk_reuseport &&
+                                     reuseport_cb_ok &&
+                                     (sk2->sk_state == TCP_TIME_WAIT ||
+                                      uid_eq(uid, sock_i_uid(sk2))))) &&
+                                   inet_rcv_saddr_equal(sk, sk2, true))
+                                       break;
+                       } else if (!reuseport_ok ||
+                                  !reuseport || !sk2->sk_reuseport ||
+                                  !reuseport_cb_ok ||
+                                  (sk2->sk_state != TCP_TIME_WAIT &&
+                                   !uid_eq(uid, sock_i_uid(sk2)))) {
+                               if (inet_rcv_saddr_equal(sk, sk2, true))
+                                       break;
+                       }
+               }
        }
-
-       if (tb2 && check_bhash2_conflict(sk, tb2, uid, relax, reuseport_cb_ok,
-                                        reuseport_ok))
-               return true;
-
-       net = sock_net(sk);
-
-       /* check there's no conflict with an existing IPV6_ADDR_ANY (if ipv6) or
-        * INADDR_ANY (if ipv4) socket.
-        */
-       hash = get_bhash2_nulladdr_hash(sk, net, port);
-       head2 = &hinfo->bhash2[hash & (hinfo->bhash_size - 1)];
-
-       l3mdev = inet_sk_bound_l3mdev(sk);
-       inet_bind_bucket_for_each(tb2, &head2->chain)
-               if (check_bind2_bucket_match_nulladdr(tb2, net, port, l3mdev, sk))
-                       break;
-
-       if (tb2 && check_bhash2_conflict(sk, tb2, uid, relax, reuseport_cb_ok,
-                                        reuseport_ok))
-               return true;
-
-       return false;
+       return sk2 != NULL;
 }
 
 /*
@@ -274,20 +191,16 @@ static int inet_csk_bind_conflict(const struct sock *sk, int port,
  * inet_bind_hashbucket lock held.
  */
 static struct inet_bind_hashbucket *
-inet_csk_find_open_port(struct sock *sk, struct inet_bind_bucket **tb_ret,
-                       struct inet_bind2_bucket **tb2_ret,
-                       struct inet_bind2_hashbucket **head2_ret, int *port_ret)
+inet_csk_find_open_port(struct sock *sk, struct inet_bind_bucket **tb_ret, int *port_ret)
 {
        struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
-       struct inet_bind2_hashbucket *head2;
+       int port = 0;
        struct inet_bind_hashbucket *head;
        struct net *net = sock_net(sk);
+       bool relax = false;
        int i, low, high, attempt_half;
-       struct inet_bind2_bucket *tb2;
        struct inet_bind_bucket *tb;
        u32 remaining, offset;
-       bool relax = false;
-       int port = 0;
        int l3mdev;
 
        l3mdev = inet_sk_bound_l3mdev(sk);
@@ -326,12 +239,10 @@ other_parity_scan:
                head = &hinfo->bhash[inet_bhashfn(net, port,
                                                  hinfo->bhash_size)];
                spin_lock_bh(&head->lock);
-               tb2 = inet_bind2_bucket_find(hinfo, net, port, l3mdev, sk,
-                                            &head2);
                inet_bind_bucket_for_each(tb, &head->chain)
-                       if (check_bind_bucket_match(tb, net, port, l3mdev)) {
-                               if (!inet_csk_bind_conflict(sk, port, tb, tb2,
-                                                           relax, false))
+                       if (net_eq(ib_net(tb), net) && tb->l3mdev == l3mdev &&
+                           tb->port == port) {
+                               if (!inet_csk_bind_conflict(sk, tb, relax, false))
                                        goto success;
                                goto next_port;
                        }
@@ -361,8 +272,6 @@ next_port:
 success:
        *port_ret = port;
        *tb_ret = tb;
-       *tb2_ret = tb2;
-       *head2_ret = head2;
        return head;
 }
 
@@ -458,81 +367,54 @@ int inet_csk_get_port(struct sock *sk, unsigned short snum)
 {
        bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
        struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
-       bool bhash_created = false, bhash2_created = false;
-       struct inet_bind2_bucket *tb2 = NULL;
-       struct inet_bind2_hashbucket *head2;
-       struct inet_bind_bucket *tb = NULL;
+       int ret = 1, port = snum;
        struct inet_bind_hashbucket *head;
        struct net *net = sock_net(sk);
-       int ret = 1, port = snum;
-       bool found_port = false;
+       struct inet_bind_bucket *tb = NULL;
        int l3mdev;
 
        l3mdev = inet_sk_bound_l3mdev(sk);
 
        if (!port) {
-               head = inet_csk_find_open_port(sk, &tb, &tb2, &head2, &port);
+               head = inet_csk_find_open_port(sk, &tb, &port);
                if (!head)
                        return ret;
-               if (tb && tb2)
-                       goto success;
-               found_port = true;
-       } else {
-               head = &hinfo->bhash[inet_bhashfn(net, port,
-                                                 hinfo->bhash_size)];
-               spin_lock_bh(&head->lock);
-               inet_bind_bucket_for_each(tb, &head->chain)
-                       if (check_bind_bucket_match(tb, net, port, l3mdev))
-                               break;
-
-               tb2 = inet_bind2_bucket_find(hinfo, net, port, l3mdev, sk,
-                                            &head2);
-       }
-
-       if (!tb) {
-               tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep, net,
-                                            head, port, l3mdev);
                if (!tb)
-                       goto fail_unlock;
-               bhash_created = true;
-       }
-
-       if (!tb2) {
-               tb2 = inet_bind2_bucket_create(hinfo->bind2_bucket_cachep,
-                                              net, head2, port, l3mdev, sk);
-               if (!tb2)
-                       goto fail_unlock;
-               bhash2_created = true;
+                       goto tb_not_found;
+               goto success;
        }
-
-       /* If we had to find an open port, we already checked for conflicts */
-       if (!found_port && !hlist_empty(&tb->owners)) {
+       head = &hinfo->bhash[inet_bhashfn(net, port,
+                                         hinfo->bhash_size)];
+       spin_lock_bh(&head->lock);
+       inet_bind_bucket_for_each(tb, &head->chain)
+               if (net_eq(ib_net(tb), net) && tb->l3mdev == l3mdev &&
+                   tb->port == port)
+                       goto tb_found;
+tb_not_found:
+       tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
+                                    net, head, port, l3mdev);
+       if (!tb)
+               goto fail_unlock;
+tb_found:
+       if (!hlist_empty(&tb->owners)) {
                if (sk->sk_reuse == SK_FORCE_REUSE)
                        goto success;
 
                if ((tb->fastreuse > 0 && reuse) ||
                    sk_reuseport_match(tb, sk))
                        goto success;
-               if (inet_csk_bind_conflict(sk, port, tb, tb2, true, true))
+               if (inet_csk_bind_conflict(sk, tb, true, true))
                        goto fail_unlock;
        }
 success:
        inet_csk_update_fastreuse(tb, sk);
 
        if (!inet_csk(sk)->icsk_bind_hash)
-               inet_bind_hash(sk, tb, tb2, port);
+               inet_bind_hash(sk, tb, port);
        WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
-       WARN_ON(inet_csk(sk)->icsk_bind2_hash != tb2);
        ret = 0;
 
 fail_unlock:
-       if (ret) {
-               if (bhash_created)
-                       inet_bind_bucket_destroy(hinfo->bind_bucket_cachep, tb);
-               if (bhash2_created)
-                       inet_bind2_bucket_destroy(hinfo->bind2_bucket_cachep,
-                                                 tb2);
-       }
        spin_unlock_bh(&head->lock);
        return ret;
 }
@@ -1079,7 +961,6 @@ struct sock *inet_csk_clone_lock(const struct sock *sk,
 
                inet_sk_set_state(newsk, TCP_SYN_RECV);
                newicsk->icsk_bind_hash = NULL;
-               newicsk->icsk_bind2_hash = NULL;
 
                inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
                inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
index 545f91b..b9d995b 100644 (file)
@@ -81,41 +81,6 @@ struct inet_bind_bucket *inet_bind_bucket_create(struct kmem_cache *cachep,
        return tb;
 }
 
-struct inet_bind2_bucket *inet_bind2_bucket_create(struct kmem_cache *cachep,
-                                                  struct net *net,
-                                                  struct inet_bind2_hashbucket *head,
-                                                  const unsigned short port,
-                                                  int l3mdev,
-                                                  const struct sock *sk)
-{
-       struct inet_bind2_bucket *tb = kmem_cache_alloc(cachep, GFP_ATOMIC);
-
-       if (tb) {
-               write_pnet(&tb->ib_net, net);
-               tb->l3mdev    = l3mdev;
-               tb->port      = port;
-#if IS_ENABLED(CONFIG_IPV6)
-               if (sk->sk_family == AF_INET6)
-                       tb->v6_rcv_saddr = sk->sk_v6_rcv_saddr;
-               else
-#endif
-                       tb->rcv_saddr = sk->sk_rcv_saddr;
-               INIT_HLIST_HEAD(&tb->owners);
-               hlist_add_head(&tb->node, &head->chain);
-       }
-       return tb;
-}
-
-static bool bind2_bucket_addr_match(struct inet_bind2_bucket *tb2, struct sock *sk)
-{
-#if IS_ENABLED(CONFIG_IPV6)
-       if (sk->sk_family == AF_INET6)
-               return ipv6_addr_equal(&tb2->v6_rcv_saddr,
-                                      &sk->sk_v6_rcv_saddr);
-#endif
-       return tb2->rcv_saddr == sk->sk_rcv_saddr;
-}
-
 /*
  * Caller must hold hashbucket lock for this tb with local BH disabled
  */
@@ -127,25 +92,12 @@ void inet_bind_bucket_destroy(struct kmem_cache *cachep, struct inet_bind_bucket
        }
 }
 
-/* Caller must hold the lock for the corresponding hashbucket in the bhash table
- * with local BH disabled
- */
-void inet_bind2_bucket_destroy(struct kmem_cache *cachep, struct inet_bind2_bucket *tb)
-{
-       if (hlist_empty(&tb->owners)) {
-               __hlist_del(&tb->node);
-               kmem_cache_free(cachep, tb);
-       }
-}
-
 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
-                   struct inet_bind2_bucket *tb2, const unsigned short snum)
+                   const unsigned short snum)
 {
        inet_sk(sk)->inet_num = snum;
        sk_add_bind_node(sk, &tb->owners);
        inet_csk(sk)->icsk_bind_hash = tb;
-       sk_add_bind2_node(sk, &tb2->owners);
-       inet_csk(sk)->icsk_bind2_hash = tb2;
 }
 
 /*
@@ -157,7 +109,6 @@ static void __inet_put_port(struct sock *sk)
        const int bhash = inet_bhashfn(sock_net(sk), inet_sk(sk)->inet_num,
                        hashinfo->bhash_size);
        struct inet_bind_hashbucket *head = &hashinfo->bhash[bhash];
-       struct inet_bind2_bucket *tb2;
        struct inet_bind_bucket *tb;
 
        spin_lock(&head->lock);
@@ -166,13 +117,6 @@ static void __inet_put_port(struct sock *sk)
        inet_csk(sk)->icsk_bind_hash = NULL;
        inet_sk(sk)->inet_num = 0;
        inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb);
-
-       if (inet_csk(sk)->icsk_bind2_hash) {
-               tb2 = inet_csk(sk)->icsk_bind2_hash;
-               __sk_del_bind2_node(sk);
-               inet_csk(sk)->icsk_bind2_hash = NULL;
-               inet_bind2_bucket_destroy(hashinfo->bind2_bucket_cachep, tb2);
-       }
        spin_unlock(&head->lock);
 }
 
@@ -189,19 +133,14 @@ int __inet_inherit_port(const struct sock *sk, struct sock *child)
        struct inet_hashinfo *table = sk->sk_prot->h.hashinfo;
        unsigned short port = inet_sk(child)->inet_num;
        const int bhash = inet_bhashfn(sock_net(sk), port,
-                                      table->bhash_size);
+                       table->bhash_size);
        struct inet_bind_hashbucket *head = &table->bhash[bhash];
-       struct inet_bind2_hashbucket *head_bhash2;
-       bool created_inet_bind_bucket = false;
-       struct net *net = sock_net(sk);
-       struct inet_bind2_bucket *tb2;
        struct inet_bind_bucket *tb;
        int l3mdev;
 
        spin_lock(&head->lock);
        tb = inet_csk(sk)->icsk_bind_hash;
-       tb2 = inet_csk(sk)->icsk_bind2_hash;
-       if (unlikely(!tb || !tb2)) {
+       if (unlikely(!tb)) {
                spin_unlock(&head->lock);
                return -ENOENT;
        }
@@ -214,45 +153,25 @@ int __inet_inherit_port(const struct sock *sk, struct sock *child)
                 * as that of the child socket. We have to look up or
                 * create a new bind bucket for the child here. */
                inet_bind_bucket_for_each(tb, &head->chain) {
-                       if (check_bind_bucket_match(tb, net, port, l3mdev))
+                       if (net_eq(ib_net(tb), sock_net(sk)) &&
+                           tb->l3mdev == l3mdev && tb->port == port)
                                break;
                }
                if (!tb) {
                        tb = inet_bind_bucket_create(table->bind_bucket_cachep,
-                                                    net, head, port, l3mdev);
+                                                    sock_net(sk), head, port,
+                                                    l3mdev);
                        if (!tb) {
                                spin_unlock(&head->lock);
                                return -ENOMEM;
                        }
-                       created_inet_bind_bucket = true;
                }
                inet_csk_update_fastreuse(tb, child);
-
-               goto bhash2_find;
-       } else if (!bind2_bucket_addr_match(tb2, child)) {
-               l3mdev = inet_sk_bound_l3mdev(sk);
-
-bhash2_find:
-               tb2 = inet_bind2_bucket_find(table, net, port, l3mdev, child,
-                                            &head_bhash2);
-               if (!tb2) {
-                       tb2 = inet_bind2_bucket_create(table->bind2_bucket_cachep,
-                                                      net, head_bhash2, port,
-                                                      l3mdev, child);
-                       if (!tb2)
-                               goto error;
-               }
        }
-       inet_bind_hash(child, tb, tb2, port);
+       inet_bind_hash(child, tb, port);
        spin_unlock(&head->lock);
 
        return 0;
-
-error:
-       if (created_inet_bind_bucket)
-               inet_bind_bucket_destroy(table->bind_bucket_cachep, tb);
-       spin_unlock(&head->lock);
-       return -ENOMEM;
 }
 EXPORT_SYMBOL_GPL(__inet_inherit_port);
 
@@ -756,76 +675,6 @@ void inet_unhash(struct sock *sk)
 }
 EXPORT_SYMBOL_GPL(inet_unhash);
 
-static bool check_bind2_bucket_match(struct inet_bind2_bucket *tb,
-                                    struct net *net, unsigned short port,
-                                    int l3mdev, struct sock *sk)
-{
-#if IS_ENABLED(CONFIG_IPV6)
-       if (sk->sk_family == AF_INET6)
-               return net_eq(ib2_net(tb), net) && tb->port == port &&
-                       tb->l3mdev == l3mdev &&
-                       ipv6_addr_equal(&tb->v6_rcv_saddr, &sk->sk_v6_rcv_saddr);
-       else
-#endif
-               return net_eq(ib2_net(tb), net) && tb->port == port &&
-                       tb->l3mdev == l3mdev && tb->rcv_saddr == sk->sk_rcv_saddr;
-}
-
-bool check_bind2_bucket_match_nulladdr(struct inet_bind2_bucket *tb,
-                                      struct net *net, const unsigned short port,
-                                      int l3mdev, const struct sock *sk)
-{
-#if IS_ENABLED(CONFIG_IPV6)
-       struct in6_addr nulladdr = {};
-
-       if (sk->sk_family == AF_INET6)
-               return net_eq(ib2_net(tb), net) && tb->port == port &&
-                       tb->l3mdev == l3mdev &&
-                       ipv6_addr_equal(&tb->v6_rcv_saddr, &nulladdr);
-       else
-#endif
-               return net_eq(ib2_net(tb), net) && tb->port == port &&
-                       tb->l3mdev == l3mdev && tb->rcv_saddr == 0;
-}
-
-static struct inet_bind2_hashbucket *
-inet_bhashfn_portaddr(struct inet_hashinfo *hinfo, const struct sock *sk,
-                     const struct net *net, unsigned short port)
-{
-       u32 hash;
-
-#if IS_ENABLED(CONFIG_IPV6)
-       if (sk->sk_family == AF_INET6)
-               hash = ipv6_portaddr_hash(net, &sk->sk_v6_rcv_saddr, port);
-       else
-#endif
-               hash = ipv4_portaddr_hash(net, sk->sk_rcv_saddr, port);
-       return &hinfo->bhash2[hash & (hinfo->bhash_size - 1)];
-}
-
-/* This should only be called when the spinlock for the socket's corresponding
- * bind_hashbucket is held
- */
-struct inet_bind2_bucket *
-inet_bind2_bucket_find(struct inet_hashinfo *hinfo, struct net *net,
-                      const unsigned short port, int l3mdev, struct sock *sk,
-                      struct inet_bind2_hashbucket **head)
-{
-       struct inet_bind2_bucket *bhash2 = NULL;
-       struct inet_bind2_hashbucket *h;
-
-       h = inet_bhashfn_portaddr(hinfo, sk, net, port);
-       inet_bind_bucket_for_each(bhash2, &h->chain) {
-               if (check_bind2_bucket_match(bhash2, net, port, l3mdev, sk))
-                       break;
-       }
-
-       if (head)
-               *head = h;
-
-       return bhash2;
-}
-
 /* RFC 6056 3.3.4.  Algorithm 4: Double-Hash Port Selection Algorithm
  * Note that we use 32bit integers (vs RFC 'short integers')
  * because 2^16 is not a multiple of num_ephemeral and this
@@ -846,13 +695,10 @@ int __inet_hash_connect(struct inet_timewait_death_row *death_row,
 {
        struct inet_hashinfo *hinfo = death_row->hashinfo;
        struct inet_timewait_sock *tw = NULL;
-       struct inet_bind2_hashbucket *head2;
        struct inet_bind_hashbucket *head;
        int port = inet_sk(sk)->inet_num;
        struct net *net = sock_net(sk);
-       struct inet_bind2_bucket *tb2;
        struct inet_bind_bucket *tb;
-       bool tb_created = false;
        u32 remaining, offset;
        int ret, i, low, high;
        int l3mdev;
@@ -909,7 +755,8 @@ other_parity_scan:
                 * the established check is already unique enough.
                 */
                inet_bind_bucket_for_each(tb, &head->chain) {
-                       if (check_bind_bucket_match(tb, net, port, l3mdev)) {
+                       if (net_eq(ib_net(tb), net) && tb->l3mdev == l3mdev &&
+                           tb->port == port) {
                                if (tb->fastreuse >= 0 ||
                                    tb->fastreuseport >= 0)
                                        goto next_port;
@@ -927,7 +774,6 @@ other_parity_scan:
                        spin_unlock_bh(&head->lock);
                        return -ENOMEM;
                }
-               tb_created = true;
                tb->fastreuse = -1;
                tb->fastreuseport = -1;
                goto ok;
@@ -943,17 +789,6 @@ next_port:
        return -EADDRNOTAVAIL;
 
 ok:
-       /* Find the corresponding tb2 bucket since we need to
-        * add the socket to the bhash2 table as well
-        */
-       tb2 = inet_bind2_bucket_find(hinfo, net, port, l3mdev, sk, &head2);
-       if (!tb2) {
-               tb2 = inet_bind2_bucket_create(hinfo->bind2_bucket_cachep, net,
-                                              head2, port, l3mdev, sk);
-               if (!tb2)
-                       goto error;
-       }
-
        /* Here we want to add a little bit of randomness to the next source
         * port that will be chosen. We use a max() with a random here so that
         * on low contention the randomness is maximal and on high contention
@@ -963,7 +798,7 @@ ok:
        WRITE_ONCE(table_perturb[index], READ_ONCE(table_perturb[index]) + i + 2);
 
        /* Head lock still held and bh's disabled */
-       inet_bind_hash(sk, tb, tb2, port);
+       inet_bind_hash(sk, tb, port);
        if (sk_unhashed(sk)) {
                inet_sk(sk)->inet_sport = htons(port);
                inet_ehash_nolisten(sk, (struct sock *)tw, NULL);
@@ -975,12 +810,6 @@ ok:
                inet_twsk_deschedule_put(tw);
        local_bh_enable();
        return 0;
-
-error:
-       if (tb_created)
-               inet_bind_bucket_destroy(hinfo->bind_bucket_cachep, tb);
-       spin_unlock_bh(&head->lock);
-       return -ENOMEM;
 }
 
 /*
index 9984d23..028513d 100644 (file)
@@ -4604,12 +4604,6 @@ void __init tcp_init(void)
                                  SLAB_HWCACHE_ALIGN | SLAB_PANIC |
                                  SLAB_ACCOUNT,
                                  NULL);
-       tcp_hashinfo.bind2_bucket_cachep =
-               kmem_cache_create("tcp_bind2_bucket",
-                                 sizeof(struct inet_bind2_bucket), 0,
-                                 SLAB_HWCACHE_ALIGN | SLAB_PANIC |
-                                 SLAB_ACCOUNT,
-                                 NULL);
 
        /* Size and allocate the main established and bind bucket
         * hash tables.
@@ -4632,9 +4626,8 @@ void __init tcp_init(void)
        if (inet_ehash_locks_alloc(&tcp_hashinfo))
                panic("TCP: failed to alloc ehash_locks");
        tcp_hashinfo.bhash =
-               alloc_large_system_hash("TCP bind bhash tables",
-                                       sizeof(struct inet_bind_hashbucket) +
-                                       sizeof(struct inet_bind2_hashbucket),
+               alloc_large_system_hash("TCP bind",
+                                       sizeof(struct inet_bind_hashbucket),
                                        tcp_hashinfo.ehash_mask + 1,
                                        17, /* one slot per 128 KB of memory */
                                        0,
@@ -4643,12 +4636,9 @@ void __init tcp_init(void)
                                        0,
                                        64 * 1024);
        tcp_hashinfo.bhash_size = 1U << tcp_hashinfo.bhash_size;
-       tcp_hashinfo.bhash2 =
-               (struct inet_bind2_hashbucket *)(tcp_hashinfo.bhash + tcp_hashinfo.bhash_size);
        for (i = 0; i < tcp_hashinfo.bhash_size; i++) {
                spin_lock_init(&tcp_hashinfo.bhash[i].lock);
                INIT_HLIST_HEAD(&tcp_hashinfo.bhash[i].chain);
-               INIT_HLIST_HEAD(&tcp_hashinfo.bhash2[i].chain);
        }
 
 
index b984f8c..a29f796 100644 (file)
@@ -37,4 +37,3 @@ gro
 ioam6_parser
 toeplitz
 cmsg_sender
-bind_bhash_test
index 464df13..7ea54af 100644 (file)
@@ -59,7 +59,6 @@ TEST_GEN_FILES += toeplitz
 TEST_GEN_FILES += cmsg_sender
 TEST_GEN_FILES += stress_reuseport_listen
 TEST_PROGS += test_vxlan_vnifiltering.sh
-TEST_GEN_FILES += bind_bhash_test
 
 TEST_FILES := settings
 
@@ -70,5 +69,4 @@ include bpf/Makefile
 
 $(OUTPUT)/reuseport_bpf_numa: LDLIBS += -lnuma
 $(OUTPUT)/tcp_mmap: LDLIBS += -lpthread
-$(OUTPUT)/bind_bhash_test: LDLIBS += -lpthread
 $(OUTPUT)/tcp_inq: LDLIBS += -lpthread
diff --git a/tools/testing/selftests/net/bind_bhash_test.c b/tools/testing/selftests/net/bind_bhash_test.c
deleted file mode 100644 (file)
index 252e737..0000000
+++ /dev/null
@@ -1,119 +0,0 @@
-// SPDX-License-Identifier: GPL-2.0
-/*
- * This times how long it takes to bind to a port when the port already
- * has multiple sockets in its bhash table.
- *
- * In the setup(), we populate the port's bhash table with
- * MAX_THREADS * MAX_CONNECTIONS number of entries.
- */
-
-#include <unistd.h>
-#include <stdio.h>
-#include <netdb.h>
-#include <pthread.h>
-
-#define MAX_THREADS 600
-#define MAX_CONNECTIONS 40
-
-static const char *bind_addr = "::1";
-static const char *port;
-
-static int fd_array[MAX_THREADS][MAX_CONNECTIONS];
-
-static int bind_socket(int opt, const char *addr)
-{
-       struct addrinfo *res, hint = {};
-       int sock_fd, reuse = 1, err;
-
-       sock_fd = socket(AF_INET6, SOCK_STREAM, 0);
-       if (sock_fd < 0) {
-               perror("socket fd err");
-               return -1;
-       }
-
-       hint.ai_family = AF_INET6;
-       hint.ai_socktype = SOCK_STREAM;
-
-       err = getaddrinfo(addr, port, &hint, &res);
-       if (err) {
-               perror("getaddrinfo failed");
-               return -1;
-       }
-
-       if (opt) {
-               err = setsockopt(sock_fd, SOL_SOCKET, opt, &reuse, sizeof(reuse));
-               if (err) {
-                       perror("setsockopt failed");
-                       return -1;
-               }
-       }
-
-       err = bind(sock_fd, res->ai_addr, res->ai_addrlen);
-       if (err) {
-               perror("failed to bind to port");
-               return -1;
-       }
-
-       return sock_fd;
-}
-
-static void *setup(void *arg)
-{
-       int sock_fd, i;
-       int *array = (int *)arg;
-
-       for (i = 0; i < MAX_CONNECTIONS; i++) {
-               sock_fd = bind_socket(SO_REUSEADDR | SO_REUSEPORT, bind_addr);
-               if (sock_fd < 0)
-                       return NULL;
-               array[i] = sock_fd;
-       }
-
-       return NULL;
-}
-
-int main(int argc, const char *argv[])
-{
-       int listener_fd, sock_fd, i, j;
-       pthread_t tid[MAX_THREADS];
-       clock_t begin, end;
-
-       if (argc != 2) {
-               printf("Usage: listener <port>\n");
-               return -1;
-       }
-
-       port = argv[1];
-
-       listener_fd = bind_socket(SO_REUSEADDR | SO_REUSEPORT, bind_addr);
-       if (listen(listener_fd, 100) < 0) {
-               perror("listen failed");
-               return -1;
-       }
-
-       /* Set up threads to populate the bhash table entry for the port */
-       for (i = 0; i < MAX_THREADS; i++)
-               pthread_create(&tid[i], NULL, setup, fd_array[i]);
-
-       for (i = 0; i < MAX_THREADS; i++)
-               pthread_join(tid[i], NULL);
-
-       begin = clock();
-
-       /* Bind to the same port on a different address */
-       sock_fd  = bind_socket(0, "2001:0db8:0:f101::1");
-
-       end = clock();
-
-       printf("time spent = %f\n", (double)(end - begin) / CLOCKS_PER_SEC);
-
-       /* clean up */
-       close(sock_fd);
-       close(listener_fd);
-       for (i = 0; i < MAX_THREADS; i++) {
-               for (j = 0; i < MAX_THREADS; i++)
-                       close(fd_array[i][j]);
-       }
-
-       return 0;
-}