Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0-or-later
0002 /*
0003  * INET     An implementation of the TCP/IP protocol suite for the LINUX
0004  *      operating system.  INET is implemented using the  BSD Socket
0005  *      interface as the means of communication with the user level.
0006  *
0007  *      Support for INET connection oriented protocols.
0008  *
0009  * Authors: See the TCP sources
0010  */
0011 
0012 #include <linux/module.h>
0013 #include <linux/jhash.h>
0014 
0015 #include <net/inet_connection_sock.h>
0016 #include <net/inet_hashtables.h>
0017 #include <net/inet_timewait_sock.h>
0018 #include <net/ip.h>
0019 #include <net/route.h>
0020 #include <net/tcp_states.h>
0021 #include <net/xfrm.h>
0022 #include <net/tcp.h>
0023 #include <net/sock_reuseport.h>
0024 #include <net/addrconf.h>
0025 
0026 #if IS_ENABLED(CONFIG_IPV6)
0027 /* match_sk*_wildcard == true:  IPV6_ADDR_ANY equals to any IPv6 addresses
0028  *              if IPv6 only, and any IPv4 addresses
0029  *              if not IPv6 only
0030  * match_sk*_wildcard == false: addresses must be exactly the same, i.e.
0031  *              IPV6_ADDR_ANY only equals to IPV6_ADDR_ANY,
0032  *              and 0.0.0.0 equals to 0.0.0.0 only
0033  */
0034 static bool ipv6_rcv_saddr_equal(const struct in6_addr *sk1_rcv_saddr6,
0035                  const struct in6_addr *sk2_rcv_saddr6,
0036                  __be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
0037                  bool sk1_ipv6only, bool sk2_ipv6only,
0038                  bool match_sk1_wildcard,
0039                  bool match_sk2_wildcard)
0040 {
0041     int addr_type = ipv6_addr_type(sk1_rcv_saddr6);
0042     int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
0043 
0044     /* if both are mapped, treat as IPv4 */
0045     if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED) {
0046         if (!sk2_ipv6only) {
0047             if (sk1_rcv_saddr == sk2_rcv_saddr)
0048                 return true;
0049             return (match_sk1_wildcard && !sk1_rcv_saddr) ||
0050                 (match_sk2_wildcard && !sk2_rcv_saddr);
0051         }
0052         return false;
0053     }
0054 
0055     if (addr_type == IPV6_ADDR_ANY && addr_type2 == IPV6_ADDR_ANY)
0056         return true;
0057 
0058     if (addr_type2 == IPV6_ADDR_ANY && match_sk2_wildcard &&
0059         !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
0060         return true;
0061 
0062     if (addr_type == IPV6_ADDR_ANY && match_sk1_wildcard &&
0063         !(sk1_ipv6only && addr_type2 == IPV6_ADDR_MAPPED))
0064         return true;
0065 
0066     if (sk2_rcv_saddr6 &&
0067         ipv6_addr_equal(sk1_rcv_saddr6, sk2_rcv_saddr6))
0068         return true;
0069 
0070     return false;
0071 }
0072 #endif
0073 
0074 /* match_sk*_wildcard == true:  0.0.0.0 equals to any IPv4 addresses
0075  * match_sk*_wildcard == false: addresses must be exactly the same, i.e.
0076  *              0.0.0.0 only equals to 0.0.0.0
0077  */
0078 static bool ipv4_rcv_saddr_equal(__be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
0079                  bool sk2_ipv6only, bool match_sk1_wildcard,
0080                  bool match_sk2_wildcard)
0081 {
0082     if (!sk2_ipv6only) {
0083         if (sk1_rcv_saddr == sk2_rcv_saddr)
0084             return true;
0085         return (match_sk1_wildcard && !sk1_rcv_saddr) ||
0086             (match_sk2_wildcard && !sk2_rcv_saddr);
0087     }
0088     return false;
0089 }
0090 
0091 bool inet_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2,
0092               bool match_wildcard)
0093 {
0094 #if IS_ENABLED(CONFIG_IPV6)
0095     if (sk->sk_family == AF_INET6)
0096         return ipv6_rcv_saddr_equal(&sk->sk_v6_rcv_saddr,
0097                         inet6_rcv_saddr(sk2),
0098                         sk->sk_rcv_saddr,
0099                         sk2->sk_rcv_saddr,
0100                         ipv6_only_sock(sk),
0101                         ipv6_only_sock(sk2),
0102                         match_wildcard,
0103                         match_wildcard);
0104 #endif
0105     return ipv4_rcv_saddr_equal(sk->sk_rcv_saddr, sk2->sk_rcv_saddr,
0106                     ipv6_only_sock(sk2), match_wildcard,
0107                     match_wildcard);
0108 }
0109 EXPORT_SYMBOL(inet_rcv_saddr_equal);
0110 
0111 bool inet_rcv_saddr_any(const struct sock *sk)
0112 {
0113 #if IS_ENABLED(CONFIG_IPV6)
0114     if (sk->sk_family == AF_INET6)
0115         return ipv6_addr_any(&sk->sk_v6_rcv_saddr);
0116 #endif
0117     return !sk->sk_rcv_saddr;
0118 }
0119 
0120 void inet_get_local_port_range(struct net *net, int *low, int *high)
0121 {
0122     unsigned int seq;
0123 
0124     do {
0125         seq = read_seqbegin(&net->ipv4.ip_local_ports.lock);
0126 
0127         *low = net->ipv4.ip_local_ports.range[0];
0128         *high = net->ipv4.ip_local_ports.range[1];
0129     } while (read_seqretry(&net->ipv4.ip_local_ports.lock, seq));
0130 }
0131 EXPORT_SYMBOL(inet_get_local_port_range);
0132 
0133 static int inet_csk_bind_conflict(const struct sock *sk,
0134                   const struct inet_bind_bucket *tb,
0135                   bool relax, bool reuseport_ok)
0136 {
0137     struct sock *sk2;
0138     bool reuseport_cb_ok;
0139     bool reuse = sk->sk_reuse;
0140     bool reuseport = !!sk->sk_reuseport;
0141     struct sock_reuseport *reuseport_cb;
0142     kuid_t uid = sock_i_uid((struct sock *)sk);
0143 
0144     rcu_read_lock();
0145     reuseport_cb = rcu_dereference(sk->sk_reuseport_cb);
0146     /* paired with WRITE_ONCE() in __reuseport_(add|detach)_closed_sock */
0147     reuseport_cb_ok = !reuseport_cb || READ_ONCE(reuseport_cb->num_closed_socks);
0148     rcu_read_unlock();
0149 
0150     /*
0151      * Unlike other sk lookup places we do not check
0152      * for sk_net here, since _all_ the socks listed
0153      * in tb->owners list belong to the same net - the
0154      * one this bucket belongs to.
0155      */
0156 
0157     sk_for_each_bound(sk2, &tb->owners) {
0158         int bound_dev_if2;
0159 
0160         if (sk == sk2)
0161             continue;
0162         bound_dev_if2 = READ_ONCE(sk2->sk_bound_dev_if);
0163         if ((!sk->sk_bound_dev_if ||
0164              !bound_dev_if2 ||
0165              sk->sk_bound_dev_if == bound_dev_if2)) {
0166             if (reuse && sk2->sk_reuse &&
0167                 sk2->sk_state != TCP_LISTEN) {
0168                 if ((!relax ||
0169                      (!reuseport_ok &&
0170                       reuseport && sk2->sk_reuseport &&
0171                       reuseport_cb_ok &&
0172                       (sk2->sk_state == TCP_TIME_WAIT ||
0173                        uid_eq(uid, sock_i_uid(sk2))))) &&
0174                     inet_rcv_saddr_equal(sk, sk2, true))
0175                     break;
0176             } else if (!reuseport_ok ||
0177                    !reuseport || !sk2->sk_reuseport ||
0178                    !reuseport_cb_ok ||
0179                    (sk2->sk_state != TCP_TIME_WAIT &&
0180                     !uid_eq(uid, sock_i_uid(sk2)))) {
0181                 if (inet_rcv_saddr_equal(sk, sk2, true))
0182                     break;
0183             }
0184         }
0185     }
0186     return sk2 != NULL;
0187 }
0188 
0189 /*
0190  * Find an open port number for the socket.  Returns with the
0191  * inet_bind_hashbucket lock held.
0192  */
0193 static struct inet_bind_hashbucket *
0194 inet_csk_find_open_port(struct sock *sk, struct inet_bind_bucket **tb_ret, int *port_ret)
0195 {
0196     struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
0197     int port = 0;
0198     struct inet_bind_hashbucket *head;
0199     struct net *net = sock_net(sk);
0200     bool relax = false;
0201     int i, low, high, attempt_half;
0202     struct inet_bind_bucket *tb;
0203     u32 remaining, offset;
0204     int l3mdev;
0205 
0206     l3mdev = inet_sk_bound_l3mdev(sk);
0207 ports_exhausted:
0208     attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
0209 other_half_scan:
0210     inet_get_local_port_range(net, &low, &high);
0211     high++; /* [32768, 60999] -> [32768, 61000[ */
0212     if (high - low < 4)
0213         attempt_half = 0;
0214     if (attempt_half) {
0215         int half = low + (((high - low) >> 2) << 1);
0216 
0217         if (attempt_half == 1)
0218             high = half;
0219         else
0220             low = half;
0221     }
0222     remaining = high - low;
0223     if (likely(remaining > 1))
0224         remaining &= ~1U;
0225 
0226     offset = prandom_u32() % remaining;
0227     /* __inet_hash_connect() favors ports having @low parity
0228      * We do the opposite to not pollute connect() users.
0229      */
0230     offset |= 1U;
0231 
0232 other_parity_scan:
0233     port = low + offset;
0234     for (i = 0; i < remaining; i += 2, port += 2) {
0235         if (unlikely(port >= high))
0236             port -= remaining;
0237         if (inet_is_local_reserved_port(net, port))
0238             continue;
0239         head = &hinfo->bhash[inet_bhashfn(net, port,
0240                           hinfo->bhash_size)];
0241         spin_lock_bh(&head->lock);
0242         inet_bind_bucket_for_each(tb, &head->chain)
0243             if (net_eq(ib_net(tb), net) && tb->l3mdev == l3mdev &&
0244                 tb->port == port) {
0245                 if (!inet_csk_bind_conflict(sk, tb, relax, false))
0246                     goto success;
0247                 goto next_port;
0248             }
0249         tb = NULL;
0250         goto success;
0251 next_port:
0252         spin_unlock_bh(&head->lock);
0253         cond_resched();
0254     }
0255 
0256     offset--;
0257     if (!(offset & 1))
0258         goto other_parity_scan;
0259 
0260     if (attempt_half == 1) {
0261         /* OK we now try the upper half of the range */
0262         attempt_half = 2;
0263         goto other_half_scan;
0264     }
0265 
0266     if (READ_ONCE(net->ipv4.sysctl_ip_autobind_reuse) && !relax) {
0267         /* We still have a chance to connect to different destinations */
0268         relax = true;
0269         goto ports_exhausted;
0270     }
0271     return NULL;
0272 success:
0273     *port_ret = port;
0274     *tb_ret = tb;
0275     return head;
0276 }
0277 
0278 static inline int sk_reuseport_match(struct inet_bind_bucket *tb,
0279                      struct sock *sk)
0280 {
0281     kuid_t uid = sock_i_uid(sk);
0282 
0283     if (tb->fastreuseport <= 0)
0284         return 0;
0285     if (!sk->sk_reuseport)
0286         return 0;
0287     if (rcu_access_pointer(sk->sk_reuseport_cb))
0288         return 0;
0289     if (!uid_eq(tb->fastuid, uid))
0290         return 0;
0291     /* We only need to check the rcv_saddr if this tb was once marked
0292      * without fastreuseport and then was reset, as we can only know that
0293      * the fast_*rcv_saddr doesn't have any conflicts with the socks on the
0294      * owners list.
0295      */
0296     if (tb->fastreuseport == FASTREUSEPORT_ANY)
0297         return 1;
0298 #if IS_ENABLED(CONFIG_IPV6)
0299     if (tb->fast_sk_family == AF_INET6)
0300         return ipv6_rcv_saddr_equal(&tb->fast_v6_rcv_saddr,
0301                         inet6_rcv_saddr(sk),
0302                         tb->fast_rcv_saddr,
0303                         sk->sk_rcv_saddr,
0304                         tb->fast_ipv6_only,
0305                         ipv6_only_sock(sk), true, false);
0306 #endif
0307     return ipv4_rcv_saddr_equal(tb->fast_rcv_saddr, sk->sk_rcv_saddr,
0308                     ipv6_only_sock(sk), true, false);
0309 }
0310 
0311 void inet_csk_update_fastreuse(struct inet_bind_bucket *tb,
0312                    struct sock *sk)
0313 {
0314     kuid_t uid = sock_i_uid(sk);
0315     bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
0316 
0317     if (hlist_empty(&tb->owners)) {
0318         tb->fastreuse = reuse;
0319         if (sk->sk_reuseport) {
0320             tb->fastreuseport = FASTREUSEPORT_ANY;
0321             tb->fastuid = uid;
0322             tb->fast_rcv_saddr = sk->sk_rcv_saddr;
0323             tb->fast_ipv6_only = ipv6_only_sock(sk);
0324             tb->fast_sk_family = sk->sk_family;
0325 #if IS_ENABLED(CONFIG_IPV6)
0326             tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
0327 #endif
0328         } else {
0329             tb->fastreuseport = 0;
0330         }
0331     } else {
0332         if (!reuse)
0333             tb->fastreuse = 0;
0334         if (sk->sk_reuseport) {
0335             /* We didn't match or we don't have fastreuseport set on
0336              * the tb, but we have sk_reuseport set on this socket
0337              * and we know that there are no bind conflicts with
0338              * this socket in this tb, so reset our tb's reuseport
0339              * settings so that any subsequent sockets that match
0340              * our current socket will be put on the fast path.
0341              *
0342              * If we reset we need to set FASTREUSEPORT_STRICT so we
0343              * do extra checking for all subsequent sk_reuseport
0344              * socks.
0345              */
0346             if (!sk_reuseport_match(tb, sk)) {
0347                 tb->fastreuseport = FASTREUSEPORT_STRICT;
0348                 tb->fastuid = uid;
0349                 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
0350                 tb->fast_ipv6_only = ipv6_only_sock(sk);
0351                 tb->fast_sk_family = sk->sk_family;
0352 #if IS_ENABLED(CONFIG_IPV6)
0353                 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
0354 #endif
0355             }
0356         } else {
0357             tb->fastreuseport = 0;
0358         }
0359     }
0360 }
0361 
0362 /* Obtain a reference to a local port for the given sock,
0363  * if snum is zero it means select any available local port.
0364  * We try to allocate an odd port (and leave even ports for connect())
0365  */
0366 int inet_csk_get_port(struct sock *sk, unsigned short snum)
0367 {
0368     bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
0369     struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
0370     int ret = 1, port = snum;
0371     struct inet_bind_hashbucket *head;
0372     struct net *net = sock_net(sk);
0373     struct inet_bind_bucket *tb = NULL;
0374     int l3mdev;
0375 
0376     l3mdev = inet_sk_bound_l3mdev(sk);
0377 
0378     if (!port) {
0379         head = inet_csk_find_open_port(sk, &tb, &port);
0380         if (!head)
0381             return ret;
0382         if (!tb)
0383             goto tb_not_found;
0384         goto success;
0385     }
0386     head = &hinfo->bhash[inet_bhashfn(net, port,
0387                       hinfo->bhash_size)];
0388     spin_lock_bh(&head->lock);
0389     inet_bind_bucket_for_each(tb, &head->chain)
0390         if (net_eq(ib_net(tb), net) && tb->l3mdev == l3mdev &&
0391             tb->port == port)
0392             goto tb_found;
0393 tb_not_found:
0394     tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
0395                      net, head, port, l3mdev);
0396     if (!tb)
0397         goto fail_unlock;
0398 tb_found:
0399     if (!hlist_empty(&tb->owners)) {
0400         if (sk->sk_reuse == SK_FORCE_REUSE)
0401             goto success;
0402 
0403         if ((tb->fastreuse > 0 && reuse) ||
0404             sk_reuseport_match(tb, sk))
0405             goto success;
0406         if (inet_csk_bind_conflict(sk, tb, true, true))
0407             goto fail_unlock;
0408     }
0409 success:
0410     inet_csk_update_fastreuse(tb, sk);
0411 
0412     if (!inet_csk(sk)->icsk_bind_hash)
0413         inet_bind_hash(sk, tb, port);
0414     WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
0415     ret = 0;
0416 
0417 fail_unlock:
0418     spin_unlock_bh(&head->lock);
0419     return ret;
0420 }
0421 EXPORT_SYMBOL_GPL(inet_csk_get_port);
0422 
0423 /*
0424  * Wait for an incoming connection, avoid race conditions. This must be called
0425  * with the socket locked.
0426  */
0427 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
0428 {
0429     struct inet_connection_sock *icsk = inet_csk(sk);
0430     DEFINE_WAIT(wait);
0431     int err;
0432 
0433     /*
0434      * True wake-one mechanism for incoming connections: only
0435      * one process gets woken up, not the 'whole herd'.
0436      * Since we do not 'race & poll' for established sockets
0437      * anymore, the common case will execute the loop only once.
0438      *
0439      * Subtle issue: "add_wait_queue_exclusive()" will be added
0440      * after any current non-exclusive waiters, and we know that
0441      * it will always _stay_ after any new non-exclusive waiters
0442      * because all non-exclusive waiters are added at the
0443      * beginning of the wait-queue. As such, it's ok to "drop"
0444      * our exclusiveness temporarily when we get woken up without
0445      * having to remove and re-insert us on the wait queue.
0446      */
0447     for (;;) {
0448         prepare_to_wait_exclusive(sk_sleep(sk), &wait,
0449                       TASK_INTERRUPTIBLE);
0450         release_sock(sk);
0451         if (reqsk_queue_empty(&icsk->icsk_accept_queue))
0452             timeo = schedule_timeout(timeo);
0453         sched_annotate_sleep();
0454         lock_sock(sk);
0455         err = 0;
0456         if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
0457             break;
0458         err = -EINVAL;
0459         if (sk->sk_state != TCP_LISTEN)
0460             break;
0461         err = sock_intr_errno(timeo);
0462         if (signal_pending(current))
0463             break;
0464         err = -EAGAIN;
0465         if (!timeo)
0466             break;
0467     }
0468     finish_wait(sk_sleep(sk), &wait);
0469     return err;
0470 }
0471 
0472 /*
0473  * This will accept the next outstanding connection.
0474  */
0475 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err, bool kern)
0476 {
0477     struct inet_connection_sock *icsk = inet_csk(sk);
0478     struct request_sock_queue *queue = &icsk->icsk_accept_queue;
0479     struct request_sock *req;
0480     struct sock *newsk;
0481     int error;
0482 
0483     lock_sock(sk);
0484 
0485     /* We need to make sure that this socket is listening,
0486      * and that it has something pending.
0487      */
0488     error = -EINVAL;
0489     if (sk->sk_state != TCP_LISTEN)
0490         goto out_err;
0491 
0492     /* Find already established connection */
0493     if (reqsk_queue_empty(queue)) {
0494         long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
0495 
0496         /* If this is a non blocking socket don't sleep */
0497         error = -EAGAIN;
0498         if (!timeo)
0499             goto out_err;
0500 
0501         error = inet_csk_wait_for_connect(sk, timeo);
0502         if (error)
0503             goto out_err;
0504     }
0505     req = reqsk_queue_remove(queue, sk);
0506     newsk = req->sk;
0507 
0508     if (sk->sk_protocol == IPPROTO_TCP &&
0509         tcp_rsk(req)->tfo_listener) {
0510         spin_lock_bh(&queue->fastopenq.lock);
0511         if (tcp_rsk(req)->tfo_listener) {
0512             /* We are still waiting for the final ACK from 3WHS
0513              * so can't free req now. Instead, we set req->sk to
0514              * NULL to signify that the child socket is taken
0515              * so reqsk_fastopen_remove() will free the req
0516              * when 3WHS finishes (or is aborted).
0517              */
0518             req->sk = NULL;
0519             req = NULL;
0520         }
0521         spin_unlock_bh(&queue->fastopenq.lock);
0522     }
0523 
0524 out:
0525     release_sock(sk);
0526     if (newsk && mem_cgroup_sockets_enabled) {
0527         int amt;
0528 
0529         /* atomically get the memory usage, set and charge the
0530          * newsk->sk_memcg.
0531          */
0532         lock_sock(newsk);
0533 
0534         /* The socket has not been accepted yet, no need to look at
0535          * newsk->sk_wmem_queued.
0536          */
0537         amt = sk_mem_pages(newsk->sk_forward_alloc +
0538                    atomic_read(&newsk->sk_rmem_alloc));
0539         mem_cgroup_sk_alloc(newsk);
0540         if (newsk->sk_memcg && amt)
0541             mem_cgroup_charge_skmem(newsk->sk_memcg, amt,
0542                         GFP_KERNEL | __GFP_NOFAIL);
0543 
0544         release_sock(newsk);
0545     }
0546     if (req)
0547         reqsk_put(req);
0548     return newsk;
0549 out_err:
0550     newsk = NULL;
0551     req = NULL;
0552     *err = error;
0553     goto out;
0554 }
0555 EXPORT_SYMBOL(inet_csk_accept);
0556 
0557 /*
0558  * Using different timers for retransmit, delayed acks and probes
0559  * We may wish use just one timer maintaining a list of expire jiffies
0560  * to optimize.
0561  */
0562 void inet_csk_init_xmit_timers(struct sock *sk,
0563                    void (*retransmit_handler)(struct timer_list *t),
0564                    void (*delack_handler)(struct timer_list *t),
0565                    void (*keepalive_handler)(struct timer_list *t))
0566 {
0567     struct inet_connection_sock *icsk = inet_csk(sk);
0568 
0569     timer_setup(&icsk->icsk_retransmit_timer, retransmit_handler, 0);
0570     timer_setup(&icsk->icsk_delack_timer, delack_handler, 0);
0571     timer_setup(&sk->sk_timer, keepalive_handler, 0);
0572     icsk->icsk_pending = icsk->icsk_ack.pending = 0;
0573 }
0574 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
0575 
0576 void inet_csk_clear_xmit_timers(struct sock *sk)
0577 {
0578     struct inet_connection_sock *icsk = inet_csk(sk);
0579 
0580     icsk->icsk_pending = icsk->icsk_ack.pending = 0;
0581 
0582     sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
0583     sk_stop_timer(sk, &icsk->icsk_delack_timer);
0584     sk_stop_timer(sk, &sk->sk_timer);
0585 }
0586 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
0587 
0588 void inet_csk_delete_keepalive_timer(struct sock *sk)
0589 {
0590     sk_stop_timer(sk, &sk->sk_timer);
0591 }
0592 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
0593 
0594 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
0595 {
0596     sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
0597 }
0598 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
0599 
0600 struct dst_entry *inet_csk_route_req(const struct sock *sk,
0601                      struct flowi4 *fl4,
0602                      const struct request_sock *req)
0603 {
0604     const struct inet_request_sock *ireq = inet_rsk(req);
0605     struct net *net = read_pnet(&ireq->ireq_net);
0606     struct ip_options_rcu *opt;
0607     struct rtable *rt;
0608 
0609     rcu_read_lock();
0610     opt = rcu_dereference(ireq->ireq_opt);
0611 
0612     flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
0613                RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
0614                sk->sk_protocol, inet_sk_flowi_flags(sk),
0615                (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
0616                ireq->ir_loc_addr, ireq->ir_rmt_port,
0617                htons(ireq->ir_num), sk->sk_uid);
0618     security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
0619     rt = ip_route_output_flow(net, fl4, sk);
0620     if (IS_ERR(rt))
0621         goto no_route;
0622     if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
0623         goto route_err;
0624     rcu_read_unlock();
0625     return &rt->dst;
0626 
0627 route_err:
0628     ip_rt_put(rt);
0629 no_route:
0630     rcu_read_unlock();
0631     __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
0632     return NULL;
0633 }
0634 EXPORT_SYMBOL_GPL(inet_csk_route_req);
0635 
0636 struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
0637                         struct sock *newsk,
0638                         const struct request_sock *req)
0639 {
0640     const struct inet_request_sock *ireq = inet_rsk(req);
0641     struct net *net = read_pnet(&ireq->ireq_net);
0642     struct inet_sock *newinet = inet_sk(newsk);
0643     struct ip_options_rcu *opt;
0644     struct flowi4 *fl4;
0645     struct rtable *rt;
0646 
0647     opt = rcu_dereference(ireq->ireq_opt);
0648     fl4 = &newinet->cork.fl.u.ip4;
0649 
0650     flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
0651                RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
0652                sk->sk_protocol, inet_sk_flowi_flags(sk),
0653                (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
0654                ireq->ir_loc_addr, ireq->ir_rmt_port,
0655                htons(ireq->ir_num), sk->sk_uid);
0656     security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
0657     rt = ip_route_output_flow(net, fl4, sk);
0658     if (IS_ERR(rt))
0659         goto no_route;
0660     if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
0661         goto route_err;
0662     return &rt->dst;
0663 
0664 route_err:
0665     ip_rt_put(rt);
0666 no_route:
0667     __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
0668     return NULL;
0669 }
0670 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
0671 
0672 /* Decide when to expire the request and when to resend SYN-ACK */
0673 static void syn_ack_recalc(struct request_sock *req,
0674                const int max_syn_ack_retries,
0675                const u8 rskq_defer_accept,
0676                int *expire, int *resend)
0677 {
0678     if (!rskq_defer_accept) {
0679         *expire = req->num_timeout >= max_syn_ack_retries;
0680         *resend = 1;
0681         return;
0682     }
0683     *expire = req->num_timeout >= max_syn_ack_retries &&
0684           (!inet_rsk(req)->acked || req->num_timeout >= rskq_defer_accept);
0685     /* Do not resend while waiting for data after ACK,
0686      * start to resend on end of deferring period to give
0687      * last chance for data or ACK to create established socket.
0688      */
0689     *resend = !inet_rsk(req)->acked ||
0690           req->num_timeout >= rskq_defer_accept - 1;
0691 }
0692 
0693 int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req)
0694 {
0695     int err = req->rsk_ops->rtx_syn_ack(parent, req);
0696 
0697     if (!err)
0698         req->num_retrans++;
0699     return err;
0700 }
0701 EXPORT_SYMBOL(inet_rtx_syn_ack);
0702 
0703 static struct request_sock *inet_reqsk_clone(struct request_sock *req,
0704                          struct sock *sk)
0705 {
0706     struct sock *req_sk, *nreq_sk;
0707     struct request_sock *nreq;
0708 
0709     nreq = kmem_cache_alloc(req->rsk_ops->slab, GFP_ATOMIC | __GFP_NOWARN);
0710     if (!nreq) {
0711         __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
0712 
0713         /* paired with refcount_inc_not_zero() in reuseport_migrate_sock() */
0714         sock_put(sk);
0715         return NULL;
0716     }
0717 
0718     req_sk = req_to_sk(req);
0719     nreq_sk = req_to_sk(nreq);
0720 
0721     memcpy(nreq_sk, req_sk,
0722            offsetof(struct sock, sk_dontcopy_begin));
0723     memcpy(&nreq_sk->sk_dontcopy_end, &req_sk->sk_dontcopy_end,
0724            req->rsk_ops->obj_size - offsetof(struct sock, sk_dontcopy_end));
0725 
0726     sk_node_init(&nreq_sk->sk_node);
0727     nreq_sk->sk_tx_queue_mapping = req_sk->sk_tx_queue_mapping;
0728 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
0729     nreq_sk->sk_rx_queue_mapping = req_sk->sk_rx_queue_mapping;
0730 #endif
0731     nreq_sk->sk_incoming_cpu = req_sk->sk_incoming_cpu;
0732 
0733     nreq->rsk_listener = sk;
0734 
0735     /* We need not acquire fastopenq->lock
0736      * because the child socket is locked in inet_csk_listen_stop().
0737      */
0738     if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(nreq)->tfo_listener)
0739         rcu_assign_pointer(tcp_sk(nreq->sk)->fastopen_rsk, nreq);
0740 
0741     return nreq;
0742 }
0743 
0744 static void reqsk_queue_migrated(struct request_sock_queue *queue,
0745                  const struct request_sock *req)
0746 {
0747     if (req->num_timeout == 0)
0748         atomic_inc(&queue->young);
0749     atomic_inc(&queue->qlen);
0750 }
0751 
0752 static void reqsk_migrate_reset(struct request_sock *req)
0753 {
0754     req->saved_syn = NULL;
0755 #if IS_ENABLED(CONFIG_IPV6)
0756     inet_rsk(req)->ipv6_opt = NULL;
0757     inet_rsk(req)->pktopts = NULL;
0758 #else
0759     inet_rsk(req)->ireq_opt = NULL;
0760 #endif
0761 }
0762 
0763 /* return true if req was found in the ehash table */
0764 static bool reqsk_queue_unlink(struct request_sock *req)
0765 {
0766     struct inet_hashinfo *hashinfo = req_to_sk(req)->sk_prot->h.hashinfo;
0767     bool found = false;
0768 
0769     if (sk_hashed(req_to_sk(req))) {
0770         spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
0771 
0772         spin_lock(lock);
0773         found = __sk_nulls_del_node_init_rcu(req_to_sk(req));
0774         spin_unlock(lock);
0775     }
0776     if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
0777         reqsk_put(req);
0778     return found;
0779 }
0780 
0781 bool inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
0782 {
0783     bool unlinked = reqsk_queue_unlink(req);
0784 
0785     if (unlinked) {
0786         reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
0787         reqsk_put(req);
0788     }
0789     return unlinked;
0790 }
0791 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
0792 
0793 void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
0794 {
0795     inet_csk_reqsk_queue_drop(sk, req);
0796     reqsk_put(req);
0797 }
0798 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
0799 
0800 static void reqsk_timer_handler(struct timer_list *t)
0801 {
0802     struct request_sock *req = from_timer(req, t, rsk_timer);
0803     struct request_sock *nreq = NULL, *oreq = req;
0804     struct sock *sk_listener = req->rsk_listener;
0805     struct inet_connection_sock *icsk;
0806     struct request_sock_queue *queue;
0807     struct net *net;
0808     int max_syn_ack_retries, qlen, expire = 0, resend = 0;
0809 
0810     if (inet_sk_state_load(sk_listener) != TCP_LISTEN) {
0811         struct sock *nsk;
0812 
0813         nsk = reuseport_migrate_sock(sk_listener, req_to_sk(req), NULL);
0814         if (!nsk)
0815             goto drop;
0816 
0817         nreq = inet_reqsk_clone(req, nsk);
0818         if (!nreq)
0819             goto drop;
0820 
0821         /* The new timer for the cloned req can decrease the 2
0822          * by calling inet_csk_reqsk_queue_drop_and_put(), so
0823          * hold another count to prevent use-after-free and
0824          * call reqsk_put() just before return.
0825          */
0826         refcount_set(&nreq->rsk_refcnt, 2 + 1);
0827         timer_setup(&nreq->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
0828         reqsk_queue_migrated(&inet_csk(nsk)->icsk_accept_queue, req);
0829 
0830         req = nreq;
0831         sk_listener = nsk;
0832     }
0833 
0834     icsk = inet_csk(sk_listener);
0835     net = sock_net(sk_listener);
0836     max_syn_ack_retries = icsk->icsk_syn_retries ? :
0837         READ_ONCE(net->ipv4.sysctl_tcp_synack_retries);
0838     /* Normally all the openreqs are young and become mature
0839      * (i.e. converted to established socket) for first timeout.
0840      * If synack was not acknowledged for 1 second, it means
0841      * one of the following things: synack was lost, ack was lost,
0842      * rtt is high or nobody planned to ack (i.e. synflood).
0843      * When server is a bit loaded, queue is populated with old
0844      * open requests, reducing effective size of queue.
0845      * When server is well loaded, queue size reduces to zero
0846      * after several minutes of work. It is not synflood,
0847      * it is normal operation. The solution is pruning
0848      * too old entries overriding normal timeout, when
0849      * situation becomes dangerous.
0850      *
0851      * Essentially, we reserve half of room for young
0852      * embrions; and abort old ones without pity, if old
0853      * ones are about to clog our table.
0854      */
0855     queue = &icsk->icsk_accept_queue;
0856     qlen = reqsk_queue_len(queue);
0857     if ((qlen << 1) > max(8U, READ_ONCE(sk_listener->sk_max_ack_backlog))) {
0858         int young = reqsk_queue_len_young(queue) << 1;
0859 
0860         while (max_syn_ack_retries > 2) {
0861             if (qlen < young)
0862                 break;
0863             max_syn_ack_retries--;
0864             young <<= 1;
0865         }
0866     }
0867     syn_ack_recalc(req, max_syn_ack_retries, READ_ONCE(queue->rskq_defer_accept),
0868                &expire, &resend);
0869     req->rsk_ops->syn_ack_timeout(req);
0870     if (!expire &&
0871         (!resend ||
0872          !inet_rtx_syn_ack(sk_listener, req) ||
0873          inet_rsk(req)->acked)) {
0874         if (req->num_timeout++ == 0)
0875             atomic_dec(&queue->young);
0876         mod_timer(&req->rsk_timer, jiffies + reqsk_timeout(req, TCP_RTO_MAX));
0877 
0878         if (!nreq)
0879             return;
0880 
0881         if (!inet_ehash_insert(req_to_sk(nreq), req_to_sk(oreq), NULL)) {
0882             /* delete timer */
0883             inet_csk_reqsk_queue_drop(sk_listener, nreq);
0884             goto no_ownership;
0885         }
0886 
0887         __NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQSUCCESS);
0888         reqsk_migrate_reset(oreq);
0889         reqsk_queue_removed(&inet_csk(oreq->rsk_listener)->icsk_accept_queue, oreq);
0890         reqsk_put(oreq);
0891 
0892         reqsk_put(nreq);
0893         return;
0894     }
0895 
0896     /* Even if we can clone the req, we may need not retransmit any more
0897      * SYN+ACKs (nreq->num_timeout > max_syn_ack_retries, etc), or another
0898      * CPU may win the "own_req" race so that inet_ehash_insert() fails.
0899      */
0900     if (nreq) {
0901         __NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQFAILURE);
0902 no_ownership:
0903         reqsk_migrate_reset(nreq);
0904         reqsk_queue_removed(queue, nreq);
0905         __reqsk_free(nreq);
0906     }
0907 
0908 drop:
0909     inet_csk_reqsk_queue_drop_and_put(oreq->rsk_listener, oreq);
0910 }
0911 
0912 static void reqsk_queue_hash_req(struct request_sock *req,
0913                  unsigned long timeout)
0914 {
0915     timer_setup(&req->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
0916     mod_timer(&req->rsk_timer, jiffies + timeout);
0917 
0918     inet_ehash_insert(req_to_sk(req), NULL, NULL);
0919     /* before letting lookups find us, make sure all req fields
0920      * are committed to memory and refcnt initialized.
0921      */
0922     smp_wmb();
0923     refcount_set(&req->rsk_refcnt, 2 + 1);
0924 }
0925 
0926 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
0927                    unsigned long timeout)
0928 {
0929     reqsk_queue_hash_req(req, timeout);
0930     inet_csk_reqsk_queue_added(sk);
0931 }
0932 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
0933 
0934 static void inet_clone_ulp(const struct request_sock *req, struct sock *newsk,
0935                const gfp_t priority)
0936 {
0937     struct inet_connection_sock *icsk = inet_csk(newsk);
0938 
0939     if (!icsk->icsk_ulp_ops)
0940         return;
0941 
0942     if (icsk->icsk_ulp_ops->clone)
0943         icsk->icsk_ulp_ops->clone(req, newsk, priority);
0944 }
0945 
0946 /**
0947  *  inet_csk_clone_lock - clone an inet socket, and lock its clone
0948  *  @sk: the socket to clone
0949  *  @req: request_sock
0950  *  @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
0951  *
0952  *  Caller must unlock socket even in error path (bh_unlock_sock(newsk))
0953  */
0954 struct sock *inet_csk_clone_lock(const struct sock *sk,
0955                  const struct request_sock *req,
0956                  const gfp_t priority)
0957 {
0958     struct sock *newsk = sk_clone_lock(sk, priority);
0959 
0960     if (newsk) {
0961         struct inet_connection_sock *newicsk = inet_csk(newsk);
0962 
0963         inet_sk_set_state(newsk, TCP_SYN_RECV);
0964         newicsk->icsk_bind_hash = NULL;
0965 
0966         inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
0967         inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
0968         inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
0969 
0970         /* listeners have SOCK_RCU_FREE, not the children */
0971         sock_reset_flag(newsk, SOCK_RCU_FREE);
0972 
0973         inet_sk(newsk)->mc_list = NULL;
0974 
0975         newsk->sk_mark = inet_rsk(req)->ir_mark;
0976         atomic64_set(&newsk->sk_cookie,
0977                  atomic64_read(&inet_rsk(req)->ir_cookie));
0978 
0979         newicsk->icsk_retransmits = 0;
0980         newicsk->icsk_backoff     = 0;
0981         newicsk->icsk_probes_out  = 0;
0982         newicsk->icsk_probes_tstamp = 0;
0983 
0984         /* Deinitialize accept_queue to trap illegal accesses. */
0985         memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
0986 
0987         inet_clone_ulp(req, newsk, priority);
0988 
0989         security_inet_csk_clone(newsk, req);
0990     }
0991     return newsk;
0992 }
0993 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
0994 
0995 /*
0996  * At this point, there should be no process reference to this
0997  * socket, and thus no user references at all.  Therefore we
0998  * can assume the socket waitqueue is inactive and nobody will
0999  * try to jump onto it.
1000  */
1001 void inet_csk_destroy_sock(struct sock *sk)
1002 {
1003     WARN_ON(sk->sk_state != TCP_CLOSE);
1004     WARN_ON(!sock_flag(sk, SOCK_DEAD));
1005 
1006     /* It cannot be in hash table! */
1007     WARN_ON(!sk_unhashed(sk));
1008 
1009     /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
1010     WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
1011 
1012     sk->sk_prot->destroy(sk);
1013 
1014     sk_stream_kill_queues(sk);
1015 
1016     xfrm_sk_free_policy(sk);
1017 
1018     sk_refcnt_debug_release(sk);
1019 
1020     this_cpu_dec(*sk->sk_prot->orphan_count);
1021 
1022     sock_put(sk);
1023 }
1024 EXPORT_SYMBOL(inet_csk_destroy_sock);
1025 
1026 /* This function allows to force a closure of a socket after the call to
1027  * tcp/dccp_create_openreq_child().
1028  */
1029 void inet_csk_prepare_forced_close(struct sock *sk)
1030     __releases(&sk->sk_lock.slock)
1031 {
1032     /* sk_clone_lock locked the socket and set refcnt to 2 */
1033     bh_unlock_sock(sk);
1034     sock_put(sk);
1035     inet_csk_prepare_for_destroy_sock(sk);
1036     inet_sk(sk)->inet_num = 0;
1037 }
1038 EXPORT_SYMBOL(inet_csk_prepare_forced_close);
1039 
1040 int inet_csk_listen_start(struct sock *sk)
1041 {
1042     struct inet_connection_sock *icsk = inet_csk(sk);
1043     struct inet_sock *inet = inet_sk(sk);
1044     int err = -EADDRINUSE;
1045 
1046     reqsk_queue_alloc(&icsk->icsk_accept_queue);
1047 
1048     sk->sk_ack_backlog = 0;
1049     inet_csk_delack_init(sk);
1050 
1051     if (sk->sk_txrehash == SOCK_TXREHASH_DEFAULT)
1052         sk->sk_txrehash = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
1053 
1054     /* There is race window here: we announce ourselves listening,
1055      * but this transition is still not validated by get_port().
1056      * It is OK, because this socket enters to hash table only
1057      * after validation is complete.
1058      */
1059     inet_sk_state_store(sk, TCP_LISTEN);
1060     if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
1061         inet->inet_sport = htons(inet->inet_num);
1062 
1063         sk_dst_reset(sk);
1064         err = sk->sk_prot->hash(sk);
1065 
1066         if (likely(!err))
1067             return 0;
1068     }
1069 
1070     inet_sk_set_state(sk, TCP_CLOSE);
1071     return err;
1072 }
1073 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
1074 
1075 static void inet_child_forget(struct sock *sk, struct request_sock *req,
1076                   struct sock *child)
1077 {
1078     sk->sk_prot->disconnect(child, O_NONBLOCK);
1079 
1080     sock_orphan(child);
1081 
1082     this_cpu_inc(*sk->sk_prot->orphan_count);
1083 
1084     if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
1085         BUG_ON(rcu_access_pointer(tcp_sk(child)->fastopen_rsk) != req);
1086         BUG_ON(sk != req->rsk_listener);
1087 
1088         /* Paranoid, to prevent race condition if
1089          * an inbound pkt destined for child is
1090          * blocked by sock lock in tcp_v4_rcv().
1091          * Also to satisfy an assertion in
1092          * tcp_v4_destroy_sock().
1093          */
1094         RCU_INIT_POINTER(tcp_sk(child)->fastopen_rsk, NULL);
1095     }
1096     inet_csk_destroy_sock(child);
1097 }
1098 
1099 struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
1100                       struct request_sock *req,
1101                       struct sock *child)
1102 {
1103     struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
1104 
1105     spin_lock(&queue->rskq_lock);
1106     if (unlikely(sk->sk_state != TCP_LISTEN)) {
1107         inet_child_forget(sk, req, child);
1108         child = NULL;
1109     } else {
1110         req->sk = child;
1111         req->dl_next = NULL;
1112         if (queue->rskq_accept_head == NULL)
1113             WRITE_ONCE(queue->rskq_accept_head, req);
1114         else
1115             queue->rskq_accept_tail->dl_next = req;
1116         queue->rskq_accept_tail = req;
1117         sk_acceptq_added(sk);
1118     }
1119     spin_unlock(&queue->rskq_lock);
1120     return child;
1121 }
1122 EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
1123 
1124 struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
1125                      struct request_sock *req, bool own_req)
1126 {
1127     if (own_req) {
1128         inet_csk_reqsk_queue_drop(req->rsk_listener, req);
1129         reqsk_queue_removed(&inet_csk(req->rsk_listener)->icsk_accept_queue, req);
1130 
1131         if (sk != req->rsk_listener) {
1132             /* another listening sk has been selected,
1133              * migrate the req to it.
1134              */
1135             struct request_sock *nreq;
1136 
1137             /* hold a refcnt for the nreq->rsk_listener
1138              * which is assigned in inet_reqsk_clone()
1139              */
1140             sock_hold(sk);
1141             nreq = inet_reqsk_clone(req, sk);
1142             if (!nreq) {
1143                 inet_child_forget(sk, req, child);
1144                 goto child_put;
1145             }
1146 
1147             refcount_set(&nreq->rsk_refcnt, 1);
1148             if (inet_csk_reqsk_queue_add(sk, nreq, child)) {
1149                 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQSUCCESS);
1150                 reqsk_migrate_reset(req);
1151                 reqsk_put(req);
1152                 return child;
1153             }
1154 
1155             __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
1156             reqsk_migrate_reset(nreq);
1157             __reqsk_free(nreq);
1158         } else if (inet_csk_reqsk_queue_add(sk, req, child)) {
1159             return child;
1160         }
1161     }
1162     /* Too bad, another child took ownership of the request, undo. */
1163 child_put:
1164     bh_unlock_sock(child);
1165     sock_put(child);
1166     return NULL;
1167 }
1168 EXPORT_SYMBOL(inet_csk_complete_hashdance);
1169 
1170 /*
1171  *  This routine closes sockets which have been at least partially
1172  *  opened, but not yet accepted.
1173  */
1174 void inet_csk_listen_stop(struct sock *sk)
1175 {
1176     struct inet_connection_sock *icsk = inet_csk(sk);
1177     struct request_sock_queue *queue = &icsk->icsk_accept_queue;
1178     struct request_sock *next, *req;
1179 
1180     /* Following specs, it would be better either to send FIN
1181      * (and enter FIN-WAIT-1, it is normal close)
1182      * or to send active reset (abort).
1183      * Certainly, it is pretty dangerous while synflood, but it is
1184      * bad justification for our negligence 8)
1185      * To be honest, we are not able to make either
1186      * of the variants now.         --ANK
1187      */
1188     while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
1189         struct sock *child = req->sk, *nsk;
1190         struct request_sock *nreq;
1191 
1192         local_bh_disable();
1193         bh_lock_sock(child);
1194         WARN_ON(sock_owned_by_user(child));
1195         sock_hold(child);
1196 
1197         nsk = reuseport_migrate_sock(sk, child, NULL);
1198         if (nsk) {
1199             nreq = inet_reqsk_clone(req, nsk);
1200             if (nreq) {
1201                 refcount_set(&nreq->rsk_refcnt, 1);
1202 
1203                 if (inet_csk_reqsk_queue_add(nsk, nreq, child)) {
1204                     __NET_INC_STATS(sock_net(nsk),
1205                             LINUX_MIB_TCPMIGRATEREQSUCCESS);
1206                     reqsk_migrate_reset(req);
1207                 } else {
1208                     __NET_INC_STATS(sock_net(nsk),
1209                             LINUX_MIB_TCPMIGRATEREQFAILURE);
1210                     reqsk_migrate_reset(nreq);
1211                     __reqsk_free(nreq);
1212                 }
1213 
1214                 /* inet_csk_reqsk_queue_add() has already
1215                  * called inet_child_forget() on failure case.
1216                  */
1217                 goto skip_child_forget;
1218             }
1219         }
1220 
1221         inet_child_forget(sk, req, child);
1222 skip_child_forget:
1223         reqsk_put(req);
1224         bh_unlock_sock(child);
1225         local_bh_enable();
1226         sock_put(child);
1227 
1228         cond_resched();
1229     }
1230     if (queue->fastopenq.rskq_rst_head) {
1231         /* Free all the reqs queued in rskq_rst_head. */
1232         spin_lock_bh(&queue->fastopenq.lock);
1233         req = queue->fastopenq.rskq_rst_head;
1234         queue->fastopenq.rskq_rst_head = NULL;
1235         spin_unlock_bh(&queue->fastopenq.lock);
1236         while (req != NULL) {
1237             next = req->dl_next;
1238             reqsk_put(req);
1239             req = next;
1240         }
1241     }
1242     WARN_ON_ONCE(sk->sk_ack_backlog);
1243 }
1244 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
1245 
1246 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
1247 {
1248     struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
1249     const struct inet_sock *inet = inet_sk(sk);
1250 
1251     sin->sin_family     = AF_INET;
1252     sin->sin_addr.s_addr    = inet->inet_daddr;
1253     sin->sin_port       = inet->inet_dport;
1254 }
1255 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
1256 
1257 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
1258 {
1259     const struct inet_sock *inet = inet_sk(sk);
1260     const struct ip_options_rcu *inet_opt;
1261     __be32 daddr = inet->inet_daddr;
1262     struct flowi4 *fl4;
1263     struct rtable *rt;
1264 
1265     rcu_read_lock();
1266     inet_opt = rcu_dereference(inet->inet_opt);
1267     if (inet_opt && inet_opt->opt.srr)
1268         daddr = inet_opt->opt.faddr;
1269     fl4 = &fl->u.ip4;
1270     rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
1271                    inet->inet_saddr, inet->inet_dport,
1272                    inet->inet_sport, sk->sk_protocol,
1273                    RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
1274     if (IS_ERR(rt))
1275         rt = NULL;
1276     if (rt)
1277         sk_setup_caps(sk, &rt->dst);
1278     rcu_read_unlock();
1279 
1280     return &rt->dst;
1281 }
1282 
1283 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
1284 {
1285     struct dst_entry *dst = __sk_dst_check(sk, 0);
1286     struct inet_sock *inet = inet_sk(sk);
1287 
1288     if (!dst) {
1289         dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1290         if (!dst)
1291             goto out;
1292     }
1293     dst->ops->update_pmtu(dst, sk, NULL, mtu, true);
1294 
1295     dst = __sk_dst_check(sk, 0);
1296     if (!dst)
1297         dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1298 out:
1299     return dst;
1300 }
1301 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!