Lines Matching +full:tx +full:- +full:ts +full:- +full:routes
1 // SPDX-License-Identifier: GPL-2.0-or-later
12 * Corey Minyard <wf-rch!minyard@relay.EU.net>
13 * Florian La Roche, <flla@stud.uni-sb.de>
25 * sk->inuse=1 and was trying to connect
51 * Alan Cox : sk->keepopen now seems to work
53 * Alan Cox : Fixed assorted sk->rqueue->next errors
63 * Alan Cox : FIN with no memory -> CRASH
113 * Alan Cox : Changed the semantics of sk->socket to
164 * Matt Day : poll()->select() match BSD precisely on error
166 * Marc Tamsky : Various sk->prot->retransmits and
167 * sk->retransmits misupdating fixed.
172 * Alan Cox : Look up device on a retransmit - routes may
214 * waiting for final ack in three-way handshake.
338 val--; in tcp_enter_memory_pressure()
353 jiffies_to_msecs(jiffies - val)); in tcp_leave_memory_pressure()
384 while (--retrans) { in retrans_to_secs()
396 u32 rate = READ_ONCE(tp->rate_delivered); in tcp_compute_delivery_rate()
397 u32 intv = READ_ONCE(tp->rate_interval_us); in tcp_compute_delivery_rate()
401 rate64 = (u64)rate * tp->mss_cache * USEC_PER_SEC; in tcp_compute_delivery_rate()
407 /* Address-family independent initialization for a tcp_sock.
417 tp->out_of_order_queue = RB_ROOT; in tcp_init_sock()
418 sk->tcp_rtx_queue = RB_ROOT; in tcp_init_sock()
420 INIT_LIST_HEAD(&tp->tsq_node); in tcp_init_sock()
421 INIT_LIST_HEAD(&tp->tsorted_sent_queue); in tcp_init_sock()
423 icsk->icsk_rto = TCP_TIMEOUT_INIT; in tcp_init_sock()
424 icsk->icsk_rto_min = TCP_RTO_MIN; in tcp_init_sock()
425 icsk->icsk_delack_max = TCP_DELACK_MAX; in tcp_init_sock()
426 tp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT); in tcp_init_sock()
427 minmax_reset(&tp->rtt_min, tcp_jiffies32, ~0U); in tcp_init_sock()
430 * initial SYN frame in their delayed-ACK and congestion control in tcp_init_sock()
432 * efficiently to them. -DaveM in tcp_init_sock()
437 tp->app_limited = ~0U; in tcp_init_sock()
438 tp->rate_app_limited = 1; in tcp_init_sock()
440 /* See draft-stevens-tcpca-spec-01 for discussion of the in tcp_init_sock()
443 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH; in tcp_init_sock()
444 tp->snd_cwnd_clamp = ~0; in tcp_init_sock()
445 tp->mss_cache = TCP_MSS_DEFAULT; in tcp_init_sock()
447 tp->reordering = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reordering); in tcp_init_sock()
450 tp->tsoffset = 0; in tcp_init_sock()
451 tp->rack.reo_wnd_steps = 1; in tcp_init_sock()
453 sk->sk_write_space = sk_stream_write_space; in tcp_init_sock()
456 icsk->icsk_sync_mss = tcp_sync_mss; in tcp_init_sock()
458 WRITE_ONCE(sk->sk_sndbuf, READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_wmem[1])); in tcp_init_sock()
459 WRITE_ONCE(sk->sk_rcvbuf, READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[1])); in tcp_init_sock()
462 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags); in tcp_init_sock()
475 sock_tx_timestamp(sk, tsflags, &shinfo->tx_flags); in tcp_tx_timestamp()
477 tcb->txstamp_ack = 1; in tcp_tx_timestamp()
479 shinfo->tskey = TCP_SKB_CB(skb)->seq + skb->len - 1; in tcp_tx_timestamp()
500 struct sock *sk = sock->sk; in tcp_poll()
529 * Check-me. in tcp_poll()
534 * in state CLOSE_WAIT. One solution is evident --- to set EPOLLHUP in tcp_poll()
540 * why EPOLLHUP is incompatible with EPOLLOUT. --ANK in tcp_poll()
543 * blocking on fresh not-connected or disconnected socket. --ANK in tcp_poll()
545 shutdown = READ_ONCE(sk->sk_shutdown); in tcp_poll()
553 (state != TCP_SYN_RECV || rcu_access_pointer(tp->fastopen_rsk))) { in tcp_poll()
555 u16 urg_data = READ_ONCE(tp->urg_data); in tcp_poll()
558 READ_ONCE(tp->urg_seq) == READ_ONCE(tp->copied_seq) && in tcp_poll()
570 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); in tcp_poll()
596 if (READ_ONCE(sk->sk_err) || in tcp_poll()
597 !skb_queue_empty_lockless(&sk->sk_error_queue)) in tcp_poll()
612 if (sk->sk_state == TCP_LISTEN) in tcp_ioctl()
613 return -EINVAL; in tcp_ioctl()
620 answ = READ_ONCE(tp->urg_data) && in tcp_ioctl()
621 READ_ONCE(tp->urg_seq) == READ_ONCE(tp->copied_seq); in tcp_ioctl()
624 if (sk->sk_state == TCP_LISTEN) in tcp_ioctl()
625 return -EINVAL; in tcp_ioctl()
627 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) in tcp_ioctl()
630 answ = READ_ONCE(tp->write_seq) - tp->snd_una; in tcp_ioctl()
633 if (sk->sk_state == TCP_LISTEN) in tcp_ioctl()
634 return -EINVAL; in tcp_ioctl()
636 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) in tcp_ioctl()
639 answ = READ_ONCE(tp->write_seq) - in tcp_ioctl()
640 READ_ONCE(tp->snd_nxt); in tcp_ioctl()
643 return -ENOIOCTLCMD; in tcp_ioctl()
653 TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH; in tcp_mark_push()
654 tp->pushed_seq = tp->write_seq; in tcp_mark_push()
659 return after(tp->write_seq, tp->pushed_seq + (tp->max_window >> 1)); in forced_push()
667 tcb->seq = tcb->end_seq = tp->write_seq; in tcp_skb_entail()
668 tcb->tcp_flags = TCPHDR_ACK; in tcp_skb_entail()
671 sk_wmem_queued_add(sk, skb->truesize); in tcp_skb_entail()
672 sk_mem_charge(sk, skb->truesize); in tcp_skb_entail()
673 if (tp->nonagle & TCP_NAGLE_PUSH) in tcp_skb_entail()
674 tp->nonagle &= ~TCP_NAGLE_PUSH; in tcp_skb_entail()
682 tp->snd_up = tp->write_seq; in tcp_mark_urg()
687 * Because TX completion will happen shortly, it gives a chance
693 * or if TX completion was delayed after we processed ACK packet.
698 return skb->len < size_goal && in tcp_should_autocork()
699 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_autocorking) && in tcp_should_autocork()
701 refcount_read(&sk->sk_wmem_alloc) > skb->truesize && in tcp_should_autocork()
722 if (!test_bit(TSQ_THROTTLED, &sk->sk_tsq_flags)) { in tcp_push()
724 set_bit(TSQ_THROTTLED, &sk->sk_tsq_flags); in tcp_push()
727 /* It is possible TX completion already happened in tcp_push()
730 if (refcount_read(&sk->sk_wmem_alloc) > skb->truesize) in tcp_push()
743 struct tcp_splice_state *tss = rd_desc->arg.data; in tcp_splice_data_recv()
746 ret = skb_splice_bits(skb, skb->sk, offset, tss->pipe, in tcp_splice_data_recv()
747 min(rd_desc->count, len), tss->flags); in tcp_splice_data_recv()
749 rd_desc->count -= ret; in tcp_splice_data_recv()
758 .count = tss->len, in __tcp_splice_read()
765 * tcp_splice_read - splice data from TCP socket to a pipe
780 struct sock *sk = sock->sk; in tcp_splice_read()
795 return -ESPIPE; in tcp_splice_read()
801 timeo = sock_rcvtimeo(sk, sock->file->f_flags & O_NONBLOCK); in tcp_splice_read()
811 if (sk->sk_err) { in tcp_splice_read()
815 if (sk->sk_shutdown & RCV_SHUTDOWN) in tcp_splice_read()
817 if (sk->sk_state == TCP_CLOSE) { in tcp_splice_read()
822 ret = -ENOTCONN; in tcp_splice_read()
826 ret = -EAGAIN; in tcp_splice_read()
833 if (!skb_queue_empty(&sk->sk_receive_queue)) in tcp_splice_read()
844 tss.len -= ret; in tcp_splice_read()
852 if (sk->sk_err || sk->sk_state == TCP_CLOSE || in tcp_splice_read()
853 (sk->sk_shutdown & RCV_SHUTDOWN) || in tcp_splice_read()
876 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); in tcp_stream_alloc_skb()
879 sk_forced_mem_schedule(sk, skb->truesize); in tcp_stream_alloc_skb()
881 mem_scheduled = sk_wmem_schedule(sk, skb->truesize); in tcp_stream_alloc_skb()
885 skb->ip_summed = CHECKSUM_PARTIAL; in tcp_stream_alloc_skb()
886 INIT_LIST_HEAD(&skb->tcp_tsorted_anchor); in tcp_stream_alloc_skb()
891 sk->sk_prot->enter_memory_pressure(sk); in tcp_stream_alloc_skb()
907 new_size_goal = tcp_bound_to_half_wnd(tp, sk->sk_gso_max_size); in tcp_xmit_size_goal()
910 size_goal = tp->gso_segs * mss_now; in tcp_xmit_size_goal()
913 tp->gso_segs = min_t(u16, new_size_goal / mss_now, in tcp_xmit_size_goal()
914 sk->sk_gso_max_segs); in tcp_xmit_size_goal()
915 size_goal = tp->gso_segs * mss_now; in tcp_xmit_size_goal()
941 if (skb && TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) { in tcp_remove_empty_skb()
953 u32 extra = skb->truesize - in tcp_downgrade_zcopy_pure()
957 return -ENOMEM; in tcp_downgrade_zcopy_pure()
960 skb_shinfo(skb)->flags &= ~SKBFL_PURE_ZEROCOPY; in tcp_downgrade_zcopy_pure()
974 * Use whatever is left in sk->sk_forward_alloc and tcp_wmem[0] in tcp_wmem_schedule()
977 left = sock_net(sk)->ipv4.sysctl_tcp_wmem[0] - sk->sk_wmem_queued; in tcp_wmem_schedule()
980 return min(copy, sk->sk_forward_alloc); in tcp_wmem_schedule()
985 if (tp->fastopen_req) { in tcp_free_fastopen_req()
986 kfree(tp->fastopen_req); in tcp_free_fastopen_req()
987 tp->fastopen_req = NULL; in tcp_free_fastopen_req()
996 struct sockaddr *uaddr = msg->msg_name; in tcp_sendmsg_fastopen()
999 if (!(READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen) & in tcp_sendmsg_fastopen()
1001 (uaddr && msg->msg_namelen >= sizeof(uaddr->sa_family) && in tcp_sendmsg_fastopen()
1002 uaddr->sa_family == AF_UNSPEC)) in tcp_sendmsg_fastopen()
1003 return -EOPNOTSUPP; in tcp_sendmsg_fastopen()
1004 if (tp->fastopen_req) in tcp_sendmsg_fastopen()
1005 return -EALREADY; /* Another Fast Open is in progress */ in tcp_sendmsg_fastopen()
1007 tp->fastopen_req = kzalloc(sizeof(struct tcp_fastopen_request), in tcp_sendmsg_fastopen()
1008 sk->sk_allocation); in tcp_sendmsg_fastopen()
1009 if (unlikely(!tp->fastopen_req)) in tcp_sendmsg_fastopen()
1010 return -ENOBUFS; in tcp_sendmsg_fastopen()
1011 tp->fastopen_req->data = msg; in tcp_sendmsg_fastopen()
1012 tp->fastopen_req->size = size; in tcp_sendmsg_fastopen()
1013 tp->fastopen_req->uarg = uarg; in tcp_sendmsg_fastopen()
1020 inet->inet_dport = 0; in tcp_sendmsg_fastopen()
1021 sk->sk_route_caps = 0; in tcp_sendmsg_fastopen()
1024 flags = (msg->msg_flags & MSG_DONTWAIT) ? O_NONBLOCK : 0; in tcp_sendmsg_fastopen()
1025 err = __inet_stream_connect(sk->sk_socket, uaddr, in tcp_sendmsg_fastopen()
1026 msg->msg_namelen, flags, 1); in tcp_sendmsg_fastopen()
1030 if (tp->fastopen_req) { in tcp_sendmsg_fastopen()
1031 *copied = tp->fastopen_req->copied; in tcp_sendmsg_fastopen()
1050 flags = msg->msg_flags; in tcp_sendmsg_locked()
1053 if (msg->msg_ubuf) { in tcp_sendmsg_locked()
1054 uarg = msg->msg_ubuf; in tcp_sendmsg_locked()
1055 if (sk->sk_route_caps & NETIF_F_SG) in tcp_sendmsg_locked()
1061 err = -ENOBUFS; in tcp_sendmsg_locked()
1064 if (sk->sk_route_caps & NETIF_F_SG) in tcp_sendmsg_locked()
1067 uarg_to_msgzc(uarg)->zerocopy = 0; in tcp_sendmsg_locked()
1069 } else if (unlikely(msg->msg_flags & MSG_SPLICE_PAGES) && size) { in tcp_sendmsg_locked()
1070 if (sk->sk_route_caps & NETIF_F_SG) in tcp_sendmsg_locked()
1076 !tp->repair) { in tcp_sendmsg_locked()
1078 if (err == -EINPROGRESS && copied_syn > 0) in tcp_sendmsg_locked()
1086 tcp_rate_check_app_limited(sk); /* is sending application-limited? */ in tcp_sendmsg_locked()
1092 if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) && in tcp_sendmsg_locked()
1099 if (unlikely(tp->repair)) { in tcp_sendmsg_locked()
1100 if (tp->repair_queue == TCP_RECV_QUEUE) { in tcp_sendmsg_locked()
1105 err = -EINVAL; in tcp_sendmsg_locked()
1106 if (tp->repair_queue == TCP_NO_QUEUE) in tcp_sendmsg_locked()
1113 if (msg->msg_controllen) { in tcp_sendmsg_locked()
1116 err = -EINVAL; in tcp_sendmsg_locked()
1130 err = -EPIPE; in tcp_sendmsg_locked()
1131 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) in tcp_sendmsg_locked()
1139 copy = size_goal - skb->len; in tcp_sendmsg_locked()
1154 skb = tcp_stream_alloc_skb(sk, sk->sk_allocation, in tcp_sendmsg_locked()
1162 skb->decrypted = !!(flags & MSG_SENDPAGE_DECRYPTED); in tcp_sendmsg_locked()
1171 if (tp->repair) in tcp_sendmsg_locked()
1172 TCP_SKB_CB(skb)->sacked |= TCPCB_REPAIRED; in tcp_sendmsg_locked()
1181 int i = skb_shinfo(skb)->nr_frags; in tcp_sendmsg_locked()
1187 if (!skb_can_coalesce(skb, i, pfrag->page, in tcp_sendmsg_locked()
1188 pfrag->offset)) { in tcp_sendmsg_locked()
1196 copy = min_t(int, copy, pfrag->size - pfrag->offset); in tcp_sendmsg_locked()
1208 err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb, in tcp_sendmsg_locked()
1209 pfrag->page, in tcp_sendmsg_locked()
1210 pfrag->offset, in tcp_sendmsg_locked()
1217 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); in tcp_sendmsg_locked()
1219 skb_fill_page_desc(skb, i, pfrag->page, in tcp_sendmsg_locked()
1220 pfrag->offset, copy); in tcp_sendmsg_locked()
1221 page_ref_inc(pfrag->page); in tcp_sendmsg_locked()
1223 pfrag->offset += copy; in tcp_sendmsg_locked()
1228 if (!skb->len) in tcp_sendmsg_locked()
1229 skb_shinfo(skb)->flags |= SKBFL_PURE_ZEROCOPY; in tcp_sendmsg_locked()
1238 if (err == -EMSGSIZE || err == -EEXIST) { in tcp_sendmsg_locked()
1253 err = skb_splice_from_iter(skb, &msg->msg_iter, copy, in tcp_sendmsg_locked()
1254 sk->sk_allocation); in tcp_sendmsg_locked()
1256 if (err == -EMSGSIZE) { in tcp_sendmsg_locked()
1265 skb_shinfo(skb)->flags |= SKBFL_SHARED_FRAG; in tcp_sendmsg_locked()
1272 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH; in tcp_sendmsg_locked()
1274 WRITE_ONCE(tp->write_seq, tp->write_seq + copy); in tcp_sendmsg_locked()
1275 TCP_SKB_CB(skb)->end_seq += copy; in tcp_sendmsg_locked()
1281 TCP_SKB_CB(skb)->eor = 1; in tcp_sendmsg_locked()
1285 if (skb->len < size_goal || (flags & MSG_OOB) || unlikely(tp->repair)) in tcp_sendmsg_locked()
1296 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); in tcp_sendmsg_locked()
1312 tcp_push(sk, flags, mss_now, tp->nonagle, size_goal); in tcp_sendmsg_locked()
1315 /* msg->msg_ubuf is pinned by the caller so we don't take extra refs */ in tcp_sendmsg_locked()
1316 if (uarg && !msg->msg_ubuf) in tcp_sendmsg_locked()
1326 /* msg->msg_ubuf is pinned by the caller so we don't take extra refs */ in tcp_sendmsg_locked()
1327 if (uarg && !msg->msg_ubuf) in tcp_sendmsg_locked()
1331 if (unlikely(tcp_rtx_and_write_queues_empty(sk) && err == -EAGAIN)) { in tcp_sendmsg_locked()
1332 sk->sk_write_space(sk); in tcp_sendmsg_locked()
1353 struct sock *sk = sock->sk; in tcp_splice_eof()
1362 tcp_push(sk, 0, mss_now, tp->nonagle, size_goal); in tcp_splice_eof()
1377 if (sock_flag(sk, SOCK_URGINLINE) || !tp->urg_data || in tcp_recv_urg()
1378 tp->urg_data == TCP_URG_READ) in tcp_recv_urg()
1379 return -EINVAL; /* Yes this is right ! */ in tcp_recv_urg()
1381 if (sk->sk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DONE)) in tcp_recv_urg()
1382 return -ENOTCONN; in tcp_recv_urg()
1384 if (tp->urg_data & TCP_URG_VALID) { in tcp_recv_urg()
1386 char c = tp->urg_data; in tcp_recv_urg()
1389 WRITE_ONCE(tp->urg_data, TCP_URG_READ); in tcp_recv_urg()
1392 msg->msg_flags |= MSG_OOB; in tcp_recv_urg()
1399 msg->msg_flags |= MSG_TRUNC; in tcp_recv_urg()
1401 return err ? -EFAULT : len; in tcp_recv_urg()
1404 if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN)) in tcp_recv_urg()
1411 * Mike <pall@rz.uni-karlsruhe.de> in tcp_recv_urg()
1413 return -EAGAIN; in tcp_recv_urg()
1421 /* XXX -- need to support SO_PEEK_OFF */ in tcp_peek_sndq()
1423 skb_rbtree_walk(skb, &sk->tcp_rtx_queue) { in tcp_peek_sndq()
1424 err = skb_copy_datagram_msg(skb, 0, msg, skb->len); in tcp_peek_sndq()
1427 copied += skb->len; in tcp_peek_sndq()
1430 skb_queue_walk(&sk->sk_write_queue, skb) { in tcp_peek_sndq()
1431 err = skb_copy_datagram_msg(skb, 0, msg, skb->len); in tcp_peek_sndq()
1435 copied += skb->len; in tcp_peek_sndq()
1455 if (/* Once-per-two-segments ACK was not sent by tcp_input.c */ in __tcp_cleanup_rbuf()
1456 tp->rcv_nxt - tp->rcv_wup > icsk->icsk_ack.rcv_mss || in __tcp_cleanup_rbuf()
1464 ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED2) || in __tcp_cleanup_rbuf()
1465 ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED) && in __tcp_cleanup_rbuf()
1467 !atomic_read(&sk->sk_rmem_alloc))) in __tcp_cleanup_rbuf()
1471 /* We send an ACK if we can now advertise a non-zero window in __tcp_cleanup_rbuf()
1477 if (copied > 0 && !time_to_ack && !(sk->sk_shutdown & RCV_SHUTDOWN)) { in __tcp_cleanup_rbuf()
1481 if (2*rcv_window_now <= tp->window_clamp) { in __tcp_cleanup_rbuf()
1499 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue); in tcp_cleanup_rbuf()
1502 WARN(skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq), in tcp_cleanup_rbuf()
1504 tp->copied_seq, TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt); in tcp_cleanup_rbuf()
1510 __skb_unlink(skb, &sk->sk_receive_queue); in tcp_eat_recv_skb()
1511 if (likely(skb->destructor == sock_rfree)) { in tcp_eat_recv_skb()
1513 skb->destructor = NULL; in tcp_eat_recv_skb()
1514 skb->sk = NULL; in tcp_eat_recv_skb()
1525 while ((skb = skb_peek(&sk->sk_receive_queue)) != NULL) { in tcp_recv_skb()
1526 offset = seq - TCP_SKB_CB(skb)->seq; in tcp_recv_skb()
1527 if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { in tcp_recv_skb()
1529 offset--; in tcp_recv_skb()
1531 if (offset < skb->len || (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)) { in tcp_recv_skb()
1550 * - It is assumed that the socket was locked by the caller.
1551 * - The routine does not block.
1552 * - At present, there is no support for reading OOB data
1566 if (sk->sk_state == TCP_LISTEN) in __tcp_read_sock()
1567 return -ENOTCONN; in __tcp_read_sock()
1569 if (offset < skb->len) { in __tcp_read_sock()
1573 len = skb->len - offset; in __tcp_read_sock()
1575 if (unlikely(tp->urg_data)) { in __tcp_read_sock()
1576 u32 urg_offset = tp->urg_seq - seq; in __tcp_read_sock()
1599 skb = tcp_recv_skb(sk, seq - 1, &offset); in __tcp_read_sock()
1605 if (offset + 1 != skb->len) in __tcp_read_sock()
1608 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) { in __tcp_read_sock()
1614 if (!desc->count) in __tcp_read_sock()
1638 &tcp_sk(sk)->copied_seq); in tcp_read_sock()
1654 if (sk->sk_state == TCP_LISTEN) in tcp_read_skb()
1655 return -ENOTCONN; in tcp_read_skb()
1657 while ((skb = skb_peek(&sk->sk_receive_queue)) != NULL) { in tcp_read_skb()
1661 __skb_unlink(skb, &sk->sk_receive_queue); in tcp_read_skb()
1663 tcp_flags = TCP_SKB_CB(skb)->tcp_flags; in tcp_read_skb()
1682 u32 seq = tp->copied_seq; in tcp_read_done()
1687 if (sk->sk_state == TCP_LISTEN) in tcp_read_done()
1694 used = min_t(size_t, skb->len - offset, left); in tcp_read_done()
1696 left -= used; in tcp_read_done()
1698 if (skb->len > offset + used) in tcp_read_done()
1701 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) { in tcp_read_done()
1708 WRITE_ONCE(tp->copied_seq, seq); in tcp_read_done()
1714 tcp_cleanup_rbuf(sk, len - left); in tcp_read_done()
1720 return tcp_inq(sock->sk); in tcp_peek_len()
1729 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK) in tcp_set_rcvlowat()
1730 cap = sk->sk_rcvbuf >> 1; in tcp_set_rcvlowat()
1732 cap = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[2]) >> 1; in tcp_set_rcvlowat()
1734 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1); in tcp_set_rcvlowat()
1739 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK) in tcp_set_rcvlowat()
1743 if (space > sk->sk_rcvbuf) { in tcp_set_rcvlowat()
1744 WRITE_ONCE(sk->sk_rcvbuf, space); in tcp_set_rcvlowat()
1745 WRITE_ONCE(tcp_sk(sk)->window_clamp, val); in tcp_set_rcvlowat()
1754 if (skb->tstamp) in tcp_update_recv_tstamps()
1755 tss->ts[0] = ktime_to_timespec64(skb->tstamp); in tcp_update_recv_tstamps()
1757 tss->ts[0] = (struct timespec64) {0}; in tcp_update_recv_tstamps()
1759 if (skb_hwtstamps(skb)->hwtstamp) in tcp_update_recv_tstamps()
1760 tss->ts[2] = ktime_to_timespec64(skb_hwtstamps(skb)->hwtstamp); in tcp_update_recv_tstamps()
1762 tss->ts[2] = (struct timespec64) {0}; in tcp_update_recv_tstamps()
1772 if (vma->vm_flags & (VM_WRITE | VM_EXEC)) in tcp_mmap()
1773 return -EPERM; in tcp_mmap()
1779 vma->vm_ops = &tcp_vm_ops; in tcp_mmap()
1789 if (unlikely(offset_skb >= skb->len)) in skb_advance_to_frag()
1792 offset_skb -= skb_headlen(skb); in skb_advance_to_frag()
1796 frag = skb_shinfo(skb)->frags; in skb_advance_to_frag()
1802 offset_skb -= skb_frag_size(frag); in skb_advance_to_frag()
1818 if (PageCompound(page) || page->mapping) in can_map_frag()
1848 zc->recv_skip_hint = skb->len - offset; in tcp_zerocopy_set_hint_for_skb()
1859 if (frag == &info->frags[info->nr_frags - 1]) in tcp_zerocopy_set_hint_for_skb()
1863 partial_frag_remainder = skb_frag_size(frag) - frag_offset; in tcp_zerocopy_set_hint_for_skb()
1864 zc->recv_skip_hint -= partial_frag_remainder; in tcp_zerocopy_set_hint_for_skb()
1872 mappable_offset = find_next_mappable_frag(frag, zc->recv_skip_hint); in tcp_zerocopy_set_hint_for_skb()
1873 zc->recv_skip_hint = mappable_offset + partial_frag_remainder; in tcp_zerocopy_set_hint_for_skb()
1883 unsigned long copy_address = (unsigned long)zc->copybuf_address; in receive_fallback_to_copy()
1888 zc->length = 0; in receive_fallback_to_copy()
1889 zc->recv_skip_hint = 0; in receive_fallback_to_copy()
1891 if (copy_address != zc->copybuf_address) in receive_fallback_to_copy()
1892 return -EINVAL; in receive_fallback_to_copy()
1900 tss, &zc->msg_flags); in receive_fallback_to_copy()
1904 zc->copybuf_len = err; in receive_fallback_to_copy()
1905 if (likely(zc->copybuf_len)) { in receive_fallback_to_copy()
1909 skb = tcp_recv_skb(sk, tcp_sk(sk)->copied_seq, &offset); in receive_fallback_to_copy()
1920 unsigned long copy_address = (unsigned long)zc->copybuf_address; in tcp_copy_straggler_data()
1925 if (copy_address != zc->copybuf_address) in tcp_copy_straggler_data()
1926 return -EINVAL; in tcp_copy_straggler_data()
1935 zc->recv_skip_hint -= copylen; in tcp_copy_straggler_data()
1948 u32 offset, copylen = min_t(u32, copybuf_len, zc->recv_skip_hint); in tcp_zc_handle_leftover()
1954 offset = *seq - TCP_SKB_CB(skb)->seq; in tcp_zc_handle_leftover()
1957 if (TCP_SKB_CB(skb)->has_rxtstamp) { in tcp_zc_handle_leftover()
1959 zc->msg_flags |= TCP_CMSG_TS; in tcp_zc_handle_leftover()
1963 zc->copybuf_len = tcp_copy_straggler_data(zc, skb, copylen, &offset, in tcp_zc_handle_leftover()
1965 return zc->copybuf_len < 0 ? 0 : copylen; in tcp_zc_handle_leftover()
1979 if (err == -EBUSY && in tcp_zerocopy_vm_insert_batch_error()
1980 zc->flags & TCP_RECEIVE_ZEROCOPY_FLAG_TLB_CLEAN_HINT) { in tcp_zerocopy_vm_insert_batch_error()
1983 maybe_zap_len = total_bytes_to_map - /* All bytes to map */ in tcp_zerocopy_vm_insert_batch_error()
1998 bytes_mapped = PAGE_SIZE * (leftover_pages - pages_remaining); in tcp_zerocopy_vm_insert_batch_error()
2010 *length -= bytes_not_mapped; in tcp_zerocopy_vm_insert_batch_error()
2011 zc->recv_skip_hint += bytes_not_mapped; in tcp_zerocopy_vm_insert_batch_error()
2031 pages_mapped = pages_to_map - (unsigned int)pages_remaining; in tcp_zerocopy_vm_insert_batch()
2056 msg_control_addr = (unsigned long)zc->msg_control; in tcp_zc_finalize_rx_tstamp()
2059 (__kernel_size_t)zc->msg_controllen; in tcp_zc_finalize_rx_tstamp()
2063 zc->msg_flags = 0; in tcp_zc_finalize_rx_tstamp()
2064 if (zc->msg_control == msg_control_addr && in tcp_zc_finalize_rx_tstamp()
2065 zc->msg_controllen == cmsg_dummy.msg_controllen) { in tcp_zc_finalize_rx_tstamp()
2067 zc->msg_control = (__u64) in tcp_zc_finalize_rx_tstamp()
2069 zc->msg_controllen = in tcp_zc_finalize_rx_tstamp()
2071 zc->msg_flags = (__u32)cmsg_dummy.msg_flags; in tcp_zc_finalize_rx_tstamp()
2082 if (vma->vm_ops != &tcp_vm_ops) { in find_tcp_vma()
2092 if (!vma || vma->vm_ops != &tcp_vm_ops) { in find_tcp_vma()
2106 unsigned long address = (unsigned long)zc->address; in tcp_zerocopy_receive()
2108 s32 copybuf_len = zc->copybuf_len; in tcp_zerocopy_receive()
2114 u32 seq = tp->copied_seq; in tcp_zerocopy_receive()
2120 zc->copybuf_len = 0; in tcp_zerocopy_receive()
2121 zc->msg_flags = 0; in tcp_zerocopy_receive()
2123 if (address & (PAGE_SIZE - 1) || address != zc->address) in tcp_zerocopy_receive()
2124 return -EINVAL; in tcp_zerocopy_receive()
2126 if (sk->sk_state == TCP_LISTEN) in tcp_zerocopy_receive()
2127 return -ENOTCONN; in tcp_zerocopy_receive()
2135 zc->length = 0; in tcp_zerocopy_receive()
2136 zc->recv_skip_hint = inq; in tcp_zerocopy_receive()
2138 return -EIO; in tcp_zerocopy_receive()
2142 vma = find_tcp_vma(current->mm, address, &mmap_locked); in tcp_zerocopy_receive()
2144 return -EINVAL; in tcp_zerocopy_receive()
2146 vma_len = min_t(unsigned long, zc->length, vma->vm_end - address); in tcp_zerocopy_receive()
2148 total_bytes_to_map = avail_len & ~(PAGE_SIZE - 1); in tcp_zerocopy_receive()
2150 if (!(zc->flags & TCP_RECEIVE_ZEROCOPY_FLAG_TLB_CLEAN_HINT)) in tcp_zerocopy_receive()
2153 zc->length = total_bytes_to_map; in tcp_zerocopy_receive()
2154 zc->recv_skip_hint = 0; in tcp_zerocopy_receive()
2156 zc->length = avail_len; in tcp_zerocopy_receive()
2157 zc->recv_skip_hint = avail_len; in tcp_zerocopy_receive()
2160 while (length + PAGE_SIZE <= zc->length) { in tcp_zerocopy_receive()
2164 if (zc->recv_skip_hint < PAGE_SIZE) { in tcp_zerocopy_receive()
2168 if (zc->recv_skip_hint > 0) in tcp_zerocopy_receive()
2170 skb = skb->next; in tcp_zerocopy_receive()
2171 offset = seq - TCP_SKB_CB(skb)->seq; in tcp_zerocopy_receive()
2176 if (TCP_SKB_CB(skb)->has_rxtstamp) { in tcp_zerocopy_receive()
2178 zc->msg_flags |= TCP_CMSG_TS; in tcp_zerocopy_receive()
2180 zc->recv_skip_hint = skb->len - offset; in tcp_zerocopy_receive()
2187 zc->recv_skip_hint); in tcp_zerocopy_receive()
2189 zc->recv_skip_hint = mappable_offset; in tcp_zerocopy_receive()
2196 zc->recv_skip_hint -= PAGE_SIZE; in tcp_zerocopy_receive()
2199 zc->recv_skip_hint < PAGE_SIZE) { in tcp_zerocopy_receive()
2220 mmap_read_unlock(current->mm); in tcp_zerocopy_receive()
2228 WRITE_ONCE(tp->copied_seq, seq); in tcp_zerocopy_receive()
2235 if (length == zc->length) in tcp_zerocopy_receive()
2236 zc->recv_skip_hint = 0; in tcp_zerocopy_receive()
2238 if (!zc->recv_skip_hint && sock_flag(sk, SOCK_DONE)) in tcp_zerocopy_receive()
2239 ret = -EIO; in tcp_zerocopy_receive()
2241 zc->length = length; in tcp_zerocopy_receive()
2253 if (tss->ts[0].tv_sec || tss->ts[0].tv_nsec) { in tcp_recv_timestamp()
2258 .tv_sec = tss->ts[0].tv_sec, in tcp_recv_timestamp()
2259 .tv_nsec = tss->ts[0].tv_nsec, in tcp_recv_timestamp()
2265 .tv_sec = tss->ts[0].tv_sec, in tcp_recv_timestamp()
2266 .tv_nsec = tss->ts[0].tv_nsec, in tcp_recv_timestamp()
2274 .tv_sec = tss->ts[0].tv_sec, in tcp_recv_timestamp()
2275 .tv_usec = tss->ts[0].tv_nsec / 1000, in tcp_recv_timestamp()
2281 .tv_sec = tss->ts[0].tv_sec, in tcp_recv_timestamp()
2282 .tv_usec = tss->ts[0].tv_nsec / 1000, in tcp_recv_timestamp()
2290 if (READ_ONCE(sk->sk_tsflags) & SOF_TIMESTAMPING_SOFTWARE) in tcp_recv_timestamp()
2293 tss->ts[0] = (struct timespec64) {0}; in tcp_recv_timestamp()
2296 if (tss->ts[2].tv_sec || tss->ts[2].tv_nsec) { in tcp_recv_timestamp()
2297 if (READ_ONCE(sk->sk_tsflags) & SOF_TIMESTAMPING_RAW_HARDWARE) in tcp_recv_timestamp()
2300 tss->ts[2] = (struct timespec64) {0}; in tcp_recv_timestamp()
2304 tss->ts[1] = (struct timespec64) {0}; in tcp_recv_timestamp()
2315 u32 copied_seq = READ_ONCE(tp->copied_seq); in tcp_inq_hint()
2316 u32 rcv_nxt = READ_ONCE(tp->rcv_nxt); in tcp_inq_hint()
2319 inq = rcv_nxt - copied_seq; in tcp_inq_hint()
2320 if (unlikely(inq < 0 || copied_seq != READ_ONCE(tp->copied_seq))) { in tcp_inq_hint()
2322 inq = tp->rcv_nxt - tp->copied_seq; in tcp_inq_hint()
2325 /* After receiving a FIN, tell the user-space to continue reading in tcp_inq_hint()
2326 * by returning a non-zero inq. in tcp_inq_hint()
2337 * tricks with *seq access order and skb->users are not required.
2356 err = -ENOTCONN; in tcp_recvmsg_locked()
2357 if (sk->sk_state == TCP_LISTEN) in tcp_recvmsg_locked()
2360 if (tp->recvmsg_inq) { in tcp_recvmsg_locked()
2362 msg->msg_get_inq = 1; in tcp_recvmsg_locked()
2370 if (unlikely(tp->repair)) { in tcp_recvmsg_locked()
2371 err = -EPERM; in tcp_recvmsg_locked()
2375 if (tp->repair_queue == TCP_SEND_QUEUE) in tcp_recvmsg_locked()
2378 err = -EINVAL; in tcp_recvmsg_locked()
2379 if (tp->repair_queue == TCP_NO_QUEUE) in tcp_recvmsg_locked()
2382 /* 'common' recv queue MSG_PEEK-ing */ in tcp_recvmsg_locked()
2385 seq = &tp->copied_seq; in tcp_recvmsg_locked()
2387 peek_seq = tp->copied_seq; in tcp_recvmsg_locked()
2397 if (unlikely(tp->urg_data) && tp->urg_seq == *seq) { in tcp_recvmsg_locked()
2401 copied = timeo ? sock_intr_errno(timeo) : -EAGAIN; in tcp_recvmsg_locked()
2408 last = skb_peek_tail(&sk->sk_receive_queue); in tcp_recvmsg_locked()
2409 skb_queue_walk(&sk->sk_receive_queue, skb) { in tcp_recvmsg_locked()
2414 if (WARN(before(*seq, TCP_SKB_CB(skb)->seq), in tcp_recvmsg_locked()
2416 *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, in tcp_recvmsg_locked()
2420 offset = *seq - TCP_SKB_CB(skb)->seq; in tcp_recvmsg_locked()
2421 if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { in tcp_recvmsg_locked()
2423 offset--; in tcp_recvmsg_locked()
2425 if (offset < skb->len) in tcp_recvmsg_locked()
2427 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) in tcp_recvmsg_locked()
2431 *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, flags); in tcp_recvmsg_locked()
2436 if (copied >= target && !READ_ONCE(sk->sk_backlog.tail)) in tcp_recvmsg_locked()
2441 sk->sk_err || in tcp_recvmsg_locked()
2442 sk->sk_state == TCP_CLOSE || in tcp_recvmsg_locked()
2443 (sk->sk_shutdown & RCV_SHUTDOWN) || in tcp_recvmsg_locked()
2450 if (sk->sk_err) { in tcp_recvmsg_locked()
2455 if (sk->sk_shutdown & RCV_SHUTDOWN) in tcp_recvmsg_locked()
2458 if (sk->sk_state == TCP_CLOSE) { in tcp_recvmsg_locked()
2462 copied = -ENOTCONN; in tcp_recvmsg_locked()
2467 copied = -EAGAIN; in tcp_recvmsg_locked()
2490 (peek_seq - copied - urg_hole != tp->copied_seq)) { in tcp_recvmsg_locked()
2492 current->comm, in tcp_recvmsg_locked()
2494 peek_seq = tp->copied_seq; in tcp_recvmsg_locked()
2500 used = skb->len - offset; in tcp_recvmsg_locked()
2505 if (unlikely(tp->urg_data)) { in tcp_recvmsg_locked()
2506 u32 urg_offset = tp->urg_seq - *seq; in tcp_recvmsg_locked()
2513 used--; in tcp_recvmsg_locked()
2527 copied = -EFAULT; in tcp_recvmsg_locked()
2534 len -= used; in tcp_recvmsg_locked()
2539 if (unlikely(tp->urg_data) && after(tp->copied_seq, tp->urg_seq)) { in tcp_recvmsg_locked()
2540 WRITE_ONCE(tp->urg_data, 0); in tcp_recvmsg_locked()
2544 if (TCP_SKB_CB(skb)->has_rxtstamp) { in tcp_recvmsg_locked()
2549 if (used + offset < skb->len) in tcp_recvmsg_locked()
2552 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) in tcp_recvmsg_locked()
2567 * on connected socket. I was just happy when found this 8) --ANK in tcp_recvmsg_locked()
2596 skb_queue_empty_lockless(&sk->sk_receive_queue) && in tcp_recvmsg()
2597 sk->sk_state == TCP_ESTABLISHED) in tcp_recvmsg()
2604 if ((cmsg_flags || msg->msg_get_inq) && ret >= 0) { in tcp_recvmsg()
2607 if (msg->msg_get_inq) { in tcp_recvmsg()
2608 msg->msg_inq = tcp_inq_hint(sk); in tcp_recvmsg()
2611 sizeof(msg->msg_inq), &msg->msg_inq); in tcp_recvmsg()
2620 int oldstate = sk->sk_state; in tcp_set_state()
2649 * above-mentioned anonymous enum in the vmlinux DWARF and hence BTF in tcp_set_state()
2671 sk->sk_prot->unhash(sk); in tcp_set_state()
2672 if (inet_csk(sk)->icsk_bind_hash && in tcp_set_state()
2673 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) in tcp_set_state()
2714 int next = (int)new_state[sk->sk_state]; in tcp_close_state()
2737 if ((1 << sk->sk_state) & in tcp_shutdown()
2783 net_info_ratelimited("out of memory -- consider tuning tcp_mem\n"); in tcp_check_oom()
2793 WRITE_ONCE(sk->sk_shutdown, SHUTDOWN_MASK); in __tcp_close()
2795 if (sk->sk_state == TCP_LISTEN) { in __tcp_close()
2805 * descriptor close, not protocol-sourced closes, because the in __tcp_close()
2808 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) { in __tcp_close()
2809 u32 len = TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq; in __tcp_close()
2811 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) in __tcp_close()
2812 len--; in __tcp_close()
2817 /* If socket has been already reset (e.g. in tcp_reset()) - kill it. */ in __tcp_close()
2818 if (sk->sk_state == TCP_CLOSE) in __tcp_close()
2825 * advertise a zero window, then kill -9 the FTP client, wheee... in __tcp_close()
2828 if (unlikely(tcp_sk(sk)->repair)) { in __tcp_close()
2829 sk->sk_prot->disconnect(sk, 0); in __tcp_close()
2834 tcp_send_active_reset(sk, sk->sk_allocation); in __tcp_close()
2835 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) { in __tcp_close()
2837 sk->sk_prot->disconnect(sk, 0); in __tcp_close()
2844 /* RED-PEN. Formally speaking, we have broken TCP state in __tcp_close()
2847 * TCP_ESTABLISHED -> TCP_FIN_WAIT1 in __tcp_close()
2848 * TCP_SYN_RECV -> TCP_FIN_WAIT1 (it is difficult) in __tcp_close()
2849 * TCP_CLOSE_WAIT -> TCP_LAST_ACK in __tcp_close()
2855 * if Linux state is FIN-WAIT-1, but FIN is still not sent. in __tcp_close()
2858 * we enter time-wait state, when it is not required really in __tcp_close()
2863 * --ANK in __tcp_close()
2864 * XXX (TFO) - To start off we don't support SYN+ACK+FIN in __tcp_close()
2866 * probably need API support or TCP_CORK SYN-ACK until in __tcp_close()
2875 state = sk->sk_state; in __tcp_close()
2887 if (state != TCP_CLOSE && sk->sk_state == TCP_CLOSE) in __tcp_close()
2894 * our end. If they send after that then tough - BUT: long enough in __tcp_close()
2895 * that we won't make the old 4*rto = almost no time - whoops in __tcp_close()
2901 * linger2 option. --ANK in __tcp_close()
2904 if (sk->sk_state == TCP_FIN_WAIT2) { in __tcp_close()
2906 if (READ_ONCE(tp->linger2) < 0) { in __tcp_close()
2916 tmo - TCP_TIMEWAIT_LEN); in __tcp_close()
2923 if (sk->sk_state != TCP_CLOSE) { in __tcp_close()
2935 if (sk->sk_state == TCP_CLOSE) { in __tcp_close()
2938 req = rcu_dereference_protected(tcp_sk(sk)->fastopen_rsk, in __tcp_close()
2940 /* We could get here with a non-NULL req if the socket is in __tcp_close()
2960 if (!sk->sk_net_refcnt) in tcp_close()
2977 struct rb_node *p = rb_first(&sk->tcp_rtx_queue); in tcp_rtx_queue_purge()
2979 tcp_sk(sk)->highest_sack = NULL; in tcp_rtx_queue_purge()
2985 * list_del(&skb->tcp_tsorted_anchor) in tcp_rtx_queue_purge()
2997 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) { in tcp_write_queue_purge()
3002 INIT_LIST_HEAD(&tcp_sk(sk)->tsorted_sent_queue); in tcp_write_queue_purge()
3004 tcp_sk(sk)->packets_out = 0; in tcp_write_queue_purge()
3005 inet_csk(sk)->icsk_backoff = 0; in tcp_write_queue_purge()
3013 int old_state = sk->sk_state; in tcp_disconnect()
3022 } else if (unlikely(tp->repair)) { in tcp_disconnect()
3023 WRITE_ONCE(sk->sk_err, ECONNABORTED); in tcp_disconnect()
3025 (tp->snd_nxt != tp->write_seq && in tcp_disconnect()
3031 WRITE_ONCE(sk->sk_err, ECONNRESET); in tcp_disconnect()
3033 WRITE_ONCE(sk->sk_err, ECONNRESET); in tcp_disconnect()
3036 __skb_queue_purge(&sk->sk_receive_queue); in tcp_disconnect()
3037 WRITE_ONCE(tp->copied_seq, tp->rcv_nxt); in tcp_disconnect()
3038 WRITE_ONCE(tp->urg_data, 0); in tcp_disconnect()
3041 skb_rbtree_purge(&tp->out_of_order_queue); in tcp_disconnect()
3043 inet->inet_dport = 0; in tcp_disconnect()
3047 WRITE_ONCE(sk->sk_shutdown, 0); in tcp_disconnect()
3049 tp->srtt_us = 0; in tcp_disconnect()
3050 tp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT); in tcp_disconnect()
3051 tp->rcv_rtt_last_tsecr = 0; in tcp_disconnect()
3053 seq = tp->write_seq + tp->max_window + 2; in tcp_disconnect()
3056 WRITE_ONCE(tp->write_seq, seq); in tcp_disconnect()
3058 icsk->icsk_backoff = 0; in tcp_disconnect()
3059 icsk->icsk_probes_out = 0; in tcp_disconnect()
3060 icsk->icsk_probes_tstamp = 0; in tcp_disconnect()
3061 icsk->icsk_rto = TCP_TIMEOUT_INIT; in tcp_disconnect()
3062 icsk->icsk_rto_min = TCP_RTO_MIN; in tcp_disconnect()
3063 icsk->icsk_delack_max = TCP_DELACK_MAX; in tcp_disconnect()
3064 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH; in tcp_disconnect()
3066 tp->snd_cwnd_cnt = 0; in tcp_disconnect()
3067 tp->is_cwnd_limited = 0; in tcp_disconnect()
3068 tp->max_packets_out = 0; in tcp_disconnect()
3069 tp->window_clamp = 0; in tcp_disconnect()
3070 tp->delivered = 0; in tcp_disconnect()
3071 tp->delivered_ce = 0; in tcp_disconnect()
3072 if (icsk->icsk_ca_ops->release) in tcp_disconnect()
3073 icsk->icsk_ca_ops->release(sk); in tcp_disconnect()
3074 memset(icsk->icsk_ca_priv, 0, sizeof(icsk->icsk_ca_priv)); in tcp_disconnect()
3075 icsk->icsk_ca_initialized = 0; in tcp_disconnect()
3077 tp->is_sack_reneg = 0; in tcp_disconnect()
3079 tp->total_retrans = 0; in tcp_disconnect()
3084 icsk->icsk_ack.rcv_mss = TCP_MIN_MSS; in tcp_disconnect()
3085 memset(&tp->rx_opt, 0, sizeof(tp->rx_opt)); in tcp_disconnect()
3087 dst_release(unrcu_pointer(xchg(&sk->sk_rx_dst, NULL))); in tcp_disconnect()
3089 tp->compressed_ack = 0; in tcp_disconnect()
3090 tp->segs_in = 0; in tcp_disconnect()
3091 tp->segs_out = 0; in tcp_disconnect()
3092 tp->bytes_sent = 0; in tcp_disconnect()
3093 tp->bytes_acked = 0; in tcp_disconnect()
3094 tp->bytes_received = 0; in tcp_disconnect()
3095 tp->bytes_retrans = 0; in tcp_disconnect()
3096 tp->data_segs_in = 0; in tcp_disconnect()
3097 tp->data_segs_out = 0; in tcp_disconnect()
3098 tp->duplicate_sack[0].start_seq = 0; in tcp_disconnect()
3099 tp->duplicate_sack[0].end_seq = 0; in tcp_disconnect()
3100 tp->dsack_dups = 0; in tcp_disconnect()
3101 tp->reord_seen = 0; in tcp_disconnect()
3102 tp->retrans_out = 0; in tcp_disconnect()
3103 tp->sacked_out = 0; in tcp_disconnect()
3104 tp->tlp_high_seq = 0; in tcp_disconnect()
3105 tp->last_oow_ack_time = 0; in tcp_disconnect()
3106 tp->plb_rehash = 0; in tcp_disconnect()
3108 tp->app_limited = ~0U; in tcp_disconnect()
3109 tp->rate_app_limited = 1; in tcp_disconnect()
3110 tp->rack.mstamp = 0; in tcp_disconnect()
3111 tp->rack.advanced = 0; in tcp_disconnect()
3112 tp->rack.reo_wnd_steps = 1; in tcp_disconnect()
3113 tp->rack.last_delivered = 0; in tcp_disconnect()
3114 tp->rack.reo_wnd_persist = 0; in tcp_disconnect()
3115 tp->rack.dsack_seen = 0; in tcp_disconnect()
3116 tp->syn_data_acked = 0; in tcp_disconnect()
3117 tp->rx_opt.saw_tstamp = 0; in tcp_disconnect()
3118 tp->rx_opt.dsack = 0; in tcp_disconnect()
3119 tp->rx_opt.num_sacks = 0; in tcp_disconnect()
3120 tp->rcv_ooopack = 0; in tcp_disconnect()
3126 tp->fastopen_client_fail = 0; in tcp_disconnect()
3128 WARN_ON(inet->inet_num && !icsk->icsk_bind_hash); in tcp_disconnect()
3130 if (sk->sk_frag.page) { in tcp_disconnect()
3131 put_page(sk->sk_frag.page); in tcp_disconnect()
3132 sk->sk_frag.page = NULL; in tcp_disconnect()
3133 sk->sk_frag.offset = 0; in tcp_disconnect()
3142 return sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN) && in tcp_can_repair_sock()
3143 (sk->sk_state != TCP_LISTEN); in tcp_can_repair_sock()
3150 if (!tp->repair) in tcp_repair_set_window()
3151 return -EPERM; in tcp_repair_set_window()
3154 return -EINVAL; in tcp_repair_set_window()
3157 return -EFAULT; in tcp_repair_set_window()
3160 return -EINVAL; in tcp_repair_set_window()
3162 if (after(opt.snd_wl1, tp->rcv_nxt + opt.rcv_wnd)) in tcp_repair_set_window()
3163 return -EINVAL; in tcp_repair_set_window()
3165 if (after(opt.rcv_wup, tp->rcv_nxt)) in tcp_repair_set_window()
3166 return -EINVAL; in tcp_repair_set_window()
3168 tp->snd_wl1 = opt.snd_wl1; in tcp_repair_set_window()
3169 tp->snd_wnd = opt.snd_wnd; in tcp_repair_set_window()
3170 tp->max_window = opt.max_window; in tcp_repair_set_window()
3172 tp->rcv_wnd = opt.rcv_wnd; in tcp_repair_set_window()
3173 tp->rcv_wup = opt.rcv_wup; in tcp_repair_set_window()
3187 return -EFAULT; in tcp_repair_options_est()
3190 len -= sizeof(opt); in tcp_repair_options_est()
3194 tp->rx_opt.mss_clamp = opt.opt_val; in tcp_repair_options_est()
3203 return -EFBIG; in tcp_repair_options_est()
3205 tp->rx_opt.snd_wscale = snd_wscale; in tcp_repair_options_est()
3206 tp->rx_opt.rcv_wscale = rcv_wscale; in tcp_repair_options_est()
3207 tp->rx_opt.wscale_ok = 1; in tcp_repair_options_est()
3212 return -EINVAL; in tcp_repair_options_est()
3214 tp->rx_opt.sack_ok |= TCP_SACK_SEEN; in tcp_repair_options_est()
3218 return -EINVAL; in tcp_repair_options_est()
3220 tp->rx_opt.tstamp_ok = 1; in tcp_repair_options_est()
3243 /* When set indicates to always queue non-full frames. Later the user clears
3257 tp->nonagle |= TCP_NAGLE_CORK; in __tcp_sock_set_cork()
3259 tp->nonagle &= ~TCP_NAGLE_CORK; in __tcp_sock_set_cork()
3260 if (tp->nonagle & TCP_NAGLE_OFF) in __tcp_sock_set_cork()
3261 tp->nonagle |= TCP_NAGLE_PUSH; in __tcp_sock_set_cork()
3283 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH; in __tcp_sock_set_nodelay()
3286 tcp_sk(sk)->nonagle &= ~TCP_NAGLE_OFF; in __tcp_sock_set_nodelay()
3306 if ((1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) && in __tcp_sock_set_quickack()
3308 inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_PUSHED; in __tcp_sock_set_quickack()
3326 return -EINVAL; in tcp_sock_set_syncnt()
3328 WRITE_ONCE(inet_csk(sk)->icsk_syn_retries, val); in tcp_sock_set_syncnt()
3339 return -EINVAL; in tcp_sock_set_user_timeout()
3341 WRITE_ONCE(inet_csk(sk)->icsk_user_timeout, val); in tcp_sock_set_user_timeout()
3351 return -EINVAL; in tcp_sock_set_keepidle_locked()
3354 WRITE_ONCE(tp->keepalive_time, val * HZ); in tcp_sock_set_keepidle_locked()
3356 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) { in tcp_sock_set_keepidle_locked()
3359 if (tp->keepalive_time > elapsed) in tcp_sock_set_keepidle_locked()
3360 elapsed = tp->keepalive_time - elapsed; in tcp_sock_set_keepidle_locked()
3383 return -EINVAL; in tcp_sock_set_keepintvl()
3385 WRITE_ONCE(tcp_sk(sk)->keepalive_intvl, val * HZ); in tcp_sock_set_keepintvl()
3393 return -EINVAL; in tcp_sock_set_keepcnt()
3396 WRITE_ONCE(tcp_sk(sk)->keepalive_probes, val); in tcp_sock_set_keepcnt()
3406 if (sk->sk_state != TCP_CLOSE) in tcp_set_window_clamp()
3407 return -EINVAL; in tcp_set_window_clamp()
3408 WRITE_ONCE(tp->window_clamp, 0); in tcp_set_window_clamp()
3410 u32 new_rcv_ssthresh, old_window_clamp = tp->window_clamp; in tcp_set_window_clamp()
3417 WRITE_ONCE(tp->window_clamp, new_window_clamp); in tcp_set_window_clamp()
3422 __tcp_adjust_rcv_ssthresh(sk, tp->window_clamp); in tcp_set_window_clamp()
3425 new_rcv_ssthresh = min(tp->rcv_wnd, tp->window_clamp); in tcp_set_window_clamp()
3426 tp->rcv_ssthresh = max(new_rcv_ssthresh, in tcp_set_window_clamp()
3427 tp->rcv_ssthresh); in tcp_set_window_clamp()
3451 return -EINVAL; in do_tcp_setsockopt()
3454 min_t(long, TCP_CA_NAME_MAX-1, optlen)); in do_tcp_setsockopt()
3456 return -EFAULT; in do_tcp_setsockopt()
3461 sockopt_ns_capable(sock_net(sk)->user_ns, in do_tcp_setsockopt()
3470 return -EINVAL; in do_tcp_setsockopt()
3473 min_t(long, TCP_ULP_NAME_MAX - 1, in do_tcp_setsockopt()
3476 return -EFAULT; in do_tcp_setsockopt()
3493 return -EINVAL; in do_tcp_setsockopt()
3496 return -EFAULT; in do_tcp_setsockopt()
3509 return -EINVAL; in do_tcp_setsockopt()
3512 return -EFAULT; in do_tcp_setsockopt()
3526 WRITE_ONCE(tp->linger2, -1); in do_tcp_setsockopt()
3528 WRITE_ONCE(tp->linger2, TCP_FIN_TIMEOUT_MAX); in do_tcp_setsockopt()
3530 WRITE_ONCE(tp->linger2, val * HZ); in do_tcp_setsockopt()
3534 WRITE_ONCE(icsk->icsk_accept_queue.rskq_defer_accept, in do_tcp_setsockopt()
3549 err = -EINVAL; in do_tcp_setsockopt()
3552 tp->rx_opt.user_mss = val; in do_tcp_setsockopt()
3561 err = -EINVAL; in do_tcp_setsockopt()
3563 tp->thin_lto = val; in do_tcp_setsockopt()
3568 err = -EINVAL; in do_tcp_setsockopt()
3573 err = -EPERM; in do_tcp_setsockopt()
3575 tp->repair = 1; in do_tcp_setsockopt()
3576 sk->sk_reuse = SK_FORCE_REUSE; in do_tcp_setsockopt()
3577 tp->repair_queue = TCP_NO_QUEUE; in do_tcp_setsockopt()
3579 tp->repair = 0; in do_tcp_setsockopt()
3580 sk->sk_reuse = SK_NO_REUSE; in do_tcp_setsockopt()
3583 tp->repair = 0; in do_tcp_setsockopt()
3584 sk->sk_reuse = SK_NO_REUSE; in do_tcp_setsockopt()
3586 err = -EINVAL; in do_tcp_setsockopt()
3591 if (!tp->repair) in do_tcp_setsockopt()
3592 err = -EPERM; in do_tcp_setsockopt()
3594 tp->repair_queue = val; in do_tcp_setsockopt()
3596 err = -EINVAL; in do_tcp_setsockopt()
3600 if (sk->sk_state != TCP_CLOSE) { in do_tcp_setsockopt()
3601 err = -EPERM; in do_tcp_setsockopt()
3602 } else if (tp->repair_queue == TCP_SEND_QUEUE) { in do_tcp_setsockopt()
3604 err = -EPERM; in do_tcp_setsockopt()
3606 WRITE_ONCE(tp->write_seq, val); in do_tcp_setsockopt()
3607 } else if (tp->repair_queue == TCP_RECV_QUEUE) { in do_tcp_setsockopt()
3608 if (tp->rcv_nxt != tp->copied_seq) { in do_tcp_setsockopt()
3609 err = -EPERM; in do_tcp_setsockopt()
3611 WRITE_ONCE(tp->rcv_nxt, val); in do_tcp_setsockopt()
3612 WRITE_ONCE(tp->copied_seq, val); in do_tcp_setsockopt()
3615 err = -EINVAL; in do_tcp_setsockopt()
3620 if (!tp->repair) in do_tcp_setsockopt()
3621 err = -EINVAL; in do_tcp_setsockopt()
3622 else if (sk->sk_state == TCP_ESTABLISHED && !tp->bytes_sent) in do_tcp_setsockopt()
3625 err = -EPERM; in do_tcp_setsockopt()
3638 err = -EINVAL; in do_tcp_setsockopt()
3640 tp->save_syn = val; in do_tcp_setsockopt()
3654 err = tp->af_specific->md5_parse(sk, optname, optval, optlen); in do_tcp_setsockopt()
3658 if (val >= 0 && ((1 << sk->sk_state) & (TCPF_CLOSE | in do_tcp_setsockopt()
3664 err = -EINVAL; in do_tcp_setsockopt()
3669 err = -EINVAL; in do_tcp_setsockopt()
3670 } else if (READ_ONCE(net->ipv4.sysctl_tcp_fastopen) & in do_tcp_setsockopt()
3672 if (sk->sk_state == TCP_CLOSE) in do_tcp_setsockopt()
3673 tp->fastopen_connect = val; in do_tcp_setsockopt()
3675 err = -EINVAL; in do_tcp_setsockopt()
3677 err = -EOPNOTSUPP; in do_tcp_setsockopt()
3682 err = -EINVAL; in do_tcp_setsockopt()
3683 else if (!((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) in do_tcp_setsockopt()
3684 err = -EINVAL; in do_tcp_setsockopt()
3686 tp->fastopen_no_cookie = val; in do_tcp_setsockopt()
3689 if (!tp->repair) in do_tcp_setsockopt()
3690 err = -EPERM; in do_tcp_setsockopt()
3692 WRITE_ONCE(tp->tsoffset, val - tcp_time_stamp_raw()); in do_tcp_setsockopt()
3698 WRITE_ONCE(tp->notsent_lowat, val); in do_tcp_setsockopt()
3699 sk->sk_write_space(sk); in do_tcp_setsockopt()
3703 err = -EINVAL; in do_tcp_setsockopt()
3705 tp->recvmsg_inq = val; in do_tcp_setsockopt()
3710 WRITE_ONCE(tp->tcp_tx_delay, val); in do_tcp_setsockopt()
3713 err = -ENOPROTOOPT; in do_tcp_setsockopt()
3728 return READ_ONCE(icsk->icsk_af_ops)->setsockopt(sk, level, optname, in tcp_setsockopt()
3741 stats[i] = tp->chrono_stat[i - 1]; in tcp_get_info_chrono_stats()
3742 if (i == tp->chrono_type) in tcp_get_info_chrono_stats()
3743 stats[i] += tcp_jiffies32 - tp->chrono_start; in tcp_get_info_chrono_stats()
3748 info->tcpi_busy_time = total; in tcp_get_info_chrono_stats()
3749 info->tcpi_rwnd_limited = stats[TCP_CHRONO_RWND_LIMITED]; in tcp_get_info_chrono_stats()
3750 info->tcpi_sndbuf_limited = stats[TCP_CHRONO_SNDBUF_LIMITED]; in tcp_get_info_chrono_stats()
3764 if (sk->sk_type != SOCK_STREAM) in tcp_get_info()
3767 info->tcpi_state = inet_sk_state_load(sk); in tcp_get_info()
3770 rate = READ_ONCE(sk->sk_pacing_rate); in tcp_get_info()
3772 info->tcpi_pacing_rate = rate64; in tcp_get_info()
3774 rate = READ_ONCE(sk->sk_max_pacing_rate); in tcp_get_info()
3776 info->tcpi_max_pacing_rate = rate64; in tcp_get_info()
3778 info->tcpi_reordering = tp->reordering; in tcp_get_info()
3779 info->tcpi_snd_cwnd = tcp_snd_cwnd(tp); in tcp_get_info()
3781 if (info->tcpi_state == TCP_LISTEN) { in tcp_get_info()
3783 * tcpi_unacked -> Number of children ready for accept() in tcp_get_info()
3784 * tcpi_sacked -> max backlog in tcp_get_info()
3786 info->tcpi_unacked = READ_ONCE(sk->sk_ack_backlog); in tcp_get_info()
3787 info->tcpi_sacked = READ_ONCE(sk->sk_max_ack_backlog); in tcp_get_info()
3793 info->tcpi_ca_state = icsk->icsk_ca_state; in tcp_get_info()
3794 info->tcpi_retransmits = icsk->icsk_retransmits; in tcp_get_info()
3795 info->tcpi_probes = icsk->icsk_probes_out; in tcp_get_info()
3796 info->tcpi_backoff = icsk->icsk_backoff; in tcp_get_info()
3798 if (tp->rx_opt.tstamp_ok) in tcp_get_info()
3799 info->tcpi_options |= TCPI_OPT_TIMESTAMPS; in tcp_get_info()
3801 info->tcpi_options |= TCPI_OPT_SACK; in tcp_get_info()
3802 if (tp->rx_opt.wscale_ok) { in tcp_get_info()
3803 info->tcpi_options |= TCPI_OPT_WSCALE; in tcp_get_info()
3804 info->tcpi_snd_wscale = tp->rx_opt.snd_wscale; in tcp_get_info()
3805 info->tcpi_rcv_wscale = tp->rx_opt.rcv_wscale; in tcp_get_info()
3808 if (tp->ecn_flags & TCP_ECN_OK) in tcp_get_info()
3809 info->tcpi_options |= TCPI_OPT_ECN; in tcp_get_info()
3810 if (tp->ecn_flags & TCP_ECN_SEEN) in tcp_get_info()
3811 info->tcpi_options |= TCPI_OPT_ECN_SEEN; in tcp_get_info()
3812 if (tp->syn_data_acked) in tcp_get_info()
3813 info->tcpi_options |= TCPI_OPT_SYN_DATA; in tcp_get_info()
3815 info->tcpi_rto = jiffies_to_usecs(icsk->icsk_rto); in tcp_get_info()
3816 info->tcpi_ato = jiffies_to_usecs(min(icsk->icsk_ack.ato, in tcp_get_info()
3818 info->tcpi_snd_mss = tp->mss_cache; in tcp_get_info()
3819 info->tcpi_rcv_mss = icsk->icsk_ack.rcv_mss; in tcp_get_info()
3821 info->tcpi_unacked = tp->packets_out; in tcp_get_info()
3822 info->tcpi_sacked = tp->sacked_out; in tcp_get_info()
3824 info->tcpi_lost = tp->lost_out; in tcp_get_info()
3825 info->tcpi_retrans = tp->retrans_out; in tcp_get_info()
3828 info->tcpi_last_data_sent = jiffies_to_msecs(now - tp->lsndtime); in tcp_get_info()
3829 info->tcpi_last_data_recv = jiffies_to_msecs(now - icsk->icsk_ack.lrcvtime); in tcp_get_info()
3830 info->tcpi_last_ack_recv = jiffies_to_msecs(now - tp->rcv_tstamp); in tcp_get_info()
3832 info->tcpi_pmtu = icsk->icsk_pmtu_cookie; in tcp_get_info()
3833 info->tcpi_rcv_ssthresh = tp->rcv_ssthresh; in tcp_get_info()
3834 info->tcpi_rtt = tp->srtt_us >> 3; in tcp_get_info()
3835 info->tcpi_rttvar = tp->mdev_us >> 2; in tcp_get_info()
3836 info->tcpi_snd_ssthresh = tp->snd_ssthresh; in tcp_get_info()
3837 info->tcpi_advmss = tp->advmss; in tcp_get_info()
3839 info->tcpi_rcv_rtt = tp->rcv_rtt_est.rtt_us >> 3; in tcp_get_info()
3840 info->tcpi_rcv_space = tp->rcvq_space.space; in tcp_get_info()
3842 info->tcpi_total_retrans = tp->total_retrans; in tcp_get_info()
3844 info->tcpi_bytes_acked = tp->bytes_acked; in tcp_get_info()
3845 info->tcpi_bytes_received = tp->bytes_received; in tcp_get_info()
3846 info->tcpi_notsent_bytes = max_t(int, 0, tp->write_seq - tp->snd_nxt); in tcp_get_info()
3849 info->tcpi_segs_out = tp->segs_out; in tcp_get_info()
3852 info->tcpi_segs_in = READ_ONCE(tp->segs_in); in tcp_get_info()
3853 info->tcpi_data_segs_in = READ_ONCE(tp->data_segs_in); in tcp_get_info()
3855 info->tcpi_min_rtt = tcp_min_rtt(tp); in tcp_get_info()
3856 info->tcpi_data_segs_out = tp->data_segs_out; in tcp_get_info()
3858 info->tcpi_delivery_rate_app_limited = tp->rate_app_limited ? 1 : 0; in tcp_get_info()
3861 info->tcpi_delivery_rate = rate64; in tcp_get_info()
3862 info->tcpi_delivered = tp->delivered; in tcp_get_info()
3863 info->tcpi_delivered_ce = tp->delivered_ce; in tcp_get_info()
3864 info->tcpi_bytes_sent = tp->bytes_sent; in tcp_get_info()
3865 info->tcpi_bytes_retrans = tp->bytes_retrans; in tcp_get_info()
3866 info->tcpi_dsack_dups = tp->dsack_dups; in tcp_get_info()
3867 info->tcpi_reord_seen = tp->reord_seen; in tcp_get_info()
3868 info->tcpi_rcv_ooopack = tp->rcv_ooopack; in tcp_get_info()
3869 info->tcpi_snd_wnd = tp->snd_wnd; in tcp_get_info()
3870 info->tcpi_rcv_wnd = tp->rcv_wnd; in tcp_get_info()
3871 info->tcpi_rehash = tp->plb_rehash + tp->timeout_rehash; in tcp_get_info()
3872 info->tcpi_fastopen_client_fail = tp->fastopen_client_fail; in tcp_get_info()
3874 info->tcpi_total_rto = tp->total_rto; in tcp_get_info()
3875 info->tcpi_total_rto_recoveries = tp->total_rto_recoveries; in tcp_get_info()
3876 info->tcpi_total_rto_time = tp->total_rto_time; in tcp_get_info()
3877 if (tp->rto_stamp) { in tcp_get_info()
3878 info->tcpi_total_rto_time += tcp_time_stamp_raw() - in tcp_get_info()
3879 tp->rto_stamp; in tcp_get_info()
3922 if (skb->protocol == htons(ETH_P_IP)) in tcp_skb_ttl_or_hop_limit()
3923 return ip_hdr(skb)->ttl; in tcp_skb_ttl_or_hop_limit()
3924 else if (skb->protocol == htons(ETH_P_IPV6)) in tcp_skb_ttl_or_hop_limit()
3925 return ipv6_hdr(skb)->hop_limit; in tcp_skb_ttl_or_hop_limit()
3952 tp->data_segs_out, TCP_NLA_PAD); in tcp_get_timestamping_opt_stats()
3954 tp->total_retrans, TCP_NLA_PAD); in tcp_get_timestamping_opt_stats()
3956 rate = READ_ONCE(sk->sk_pacing_rate); in tcp_get_timestamping_opt_stats()
3964 nla_put_u32(stats, TCP_NLA_REORDERING, tp->reordering); in tcp_get_timestamping_opt_stats()
3967 nla_put_u8(stats, TCP_NLA_RECUR_RETRANS, inet_csk(sk)->icsk_retransmits); in tcp_get_timestamping_opt_stats()
3968 nla_put_u8(stats, TCP_NLA_DELIVERY_RATE_APP_LMT, !!tp->rate_app_limited); in tcp_get_timestamping_opt_stats()
3969 nla_put_u32(stats, TCP_NLA_SND_SSTHRESH, tp->snd_ssthresh); in tcp_get_timestamping_opt_stats()
3970 nla_put_u32(stats, TCP_NLA_DELIVERED, tp->delivered); in tcp_get_timestamping_opt_stats()
3971 nla_put_u32(stats, TCP_NLA_DELIVERED_CE, tp->delivered_ce); in tcp_get_timestamping_opt_stats()
3973 nla_put_u32(stats, TCP_NLA_SNDQ_SIZE, tp->write_seq - tp->snd_una); in tcp_get_timestamping_opt_stats()
3974 nla_put_u8(stats, TCP_NLA_CA_STATE, inet_csk(sk)->icsk_ca_state); in tcp_get_timestamping_opt_stats()
3976 nla_put_u64_64bit(stats, TCP_NLA_BYTES_SENT, tp->bytes_sent, in tcp_get_timestamping_opt_stats()
3978 nla_put_u64_64bit(stats, TCP_NLA_BYTES_RETRANS, tp->bytes_retrans, in tcp_get_timestamping_opt_stats()
3980 nla_put_u32(stats, TCP_NLA_DSACK_DUPS, tp->dsack_dups); in tcp_get_timestamping_opt_stats()
3981 nla_put_u32(stats, TCP_NLA_REORD_SEEN, tp->reord_seen); in tcp_get_timestamping_opt_stats()
3982 nla_put_u32(stats, TCP_NLA_SRTT, tp->srtt_us >> 3); in tcp_get_timestamping_opt_stats()
3983 nla_put_u16(stats, TCP_NLA_TIMEOUT_REHASH, tp->timeout_rehash); in tcp_get_timestamping_opt_stats()
3985 max_t(int, 0, tp->write_seq - tp->snd_nxt)); in tcp_get_timestamping_opt_stats()
3986 nla_put_u64_64bit(stats, TCP_NLA_EDT, orig_skb->skb_mstamp_ns, in tcp_get_timestamping_opt_stats()
3992 nla_put_u32(stats, TCP_NLA_REHASH, tp->plb_rehash + tp->timeout_rehash); in tcp_get_timestamping_opt_stats()
4005 return -EFAULT; in do_tcp_getsockopt()
4008 return -EINVAL; in do_tcp_getsockopt()
4014 val = tp->mss_cache; in do_tcp_getsockopt()
4015 if (tp->rx_opt.user_mss && in do_tcp_getsockopt()
4016 ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) in do_tcp_getsockopt()
4017 val = tp->rx_opt.user_mss; in do_tcp_getsockopt()
4018 if (tp->repair) in do_tcp_getsockopt()
4019 val = tp->rx_opt.mss_clamp; in do_tcp_getsockopt()
4022 val = !!(tp->nonagle&TCP_NAGLE_OFF); in do_tcp_getsockopt()
4025 val = !!(tp->nonagle&TCP_NAGLE_CORK); in do_tcp_getsockopt()
4037 val = READ_ONCE(icsk->icsk_syn_retries) ? : in do_tcp_getsockopt()
4038 READ_ONCE(net->ipv4.sysctl_tcp_syn_retries); in do_tcp_getsockopt()
4041 val = READ_ONCE(tp->linger2); in do_tcp_getsockopt()
4043 val = (val ? : READ_ONCE(net->ipv4.sysctl_tcp_fin_timeout)) / HZ; in do_tcp_getsockopt()
4046 val = READ_ONCE(icsk->icsk_accept_queue.rskq_defer_accept); in do_tcp_getsockopt()
4051 val = READ_ONCE(tp->window_clamp); in do_tcp_getsockopt()
4057 return -EFAULT; in do_tcp_getsockopt()
4063 return -EFAULT; in do_tcp_getsockopt()
4065 return -EFAULT; in do_tcp_getsockopt()
4075 return -EFAULT; in do_tcp_getsockopt()
4077 ca_ops = icsk->icsk_ca_ops; in do_tcp_getsockopt()
4078 if (ca_ops && ca_ops->get_info) in do_tcp_getsockopt()
4079 sz = ca_ops->get_info(sk, ~0U, &attr, &info); in do_tcp_getsockopt()
4083 return -EFAULT; in do_tcp_getsockopt()
4085 return -EFAULT; in do_tcp_getsockopt()
4094 return -EFAULT; in do_tcp_getsockopt()
4097 return -EFAULT; in do_tcp_getsockopt()
4098 if (copy_to_sockptr(optval, icsk->icsk_ca_ops->name, len)) in do_tcp_getsockopt()
4099 return -EFAULT; in do_tcp_getsockopt()
4104 return -EFAULT; in do_tcp_getsockopt()
4106 if (!icsk->icsk_ulp_ops) { in do_tcp_getsockopt()
4109 return -EFAULT; in do_tcp_getsockopt()
4113 return -EFAULT; in do_tcp_getsockopt()
4114 if (copy_to_sockptr(optval, icsk->icsk_ulp_ops->name, len)) in do_tcp_getsockopt()
4115 return -EFAULT; in do_tcp_getsockopt()
4123 return -EFAULT; in do_tcp_getsockopt()
4129 return -EFAULT; in do_tcp_getsockopt()
4131 return -EFAULT; in do_tcp_getsockopt()
4135 val = tp->thin_lto; in do_tcp_getsockopt()
4143 val = tp->repair; in do_tcp_getsockopt()
4147 if (tp->repair) in do_tcp_getsockopt()
4148 val = tp->repair_queue; in do_tcp_getsockopt()
4150 return -EINVAL; in do_tcp_getsockopt()
4157 return -EFAULT; in do_tcp_getsockopt()
4160 return -EINVAL; in do_tcp_getsockopt()
4162 if (!tp->repair) in do_tcp_getsockopt()
4163 return -EPERM; in do_tcp_getsockopt()
4165 opt.snd_wl1 = tp->snd_wl1; in do_tcp_getsockopt()
4166 opt.snd_wnd = tp->snd_wnd; in do_tcp_getsockopt()
4167 opt.max_window = tp->max_window; in do_tcp_getsockopt()
4168 opt.rcv_wnd = tp->rcv_wnd; in do_tcp_getsockopt()
4169 opt.rcv_wup = tp->rcv_wup; in do_tcp_getsockopt()
4172 return -EFAULT; in do_tcp_getsockopt()
4176 if (tp->repair_queue == TCP_SEND_QUEUE) in do_tcp_getsockopt()
4177 val = tp->write_seq; in do_tcp_getsockopt()
4178 else if (tp->repair_queue == TCP_RECV_QUEUE) in do_tcp_getsockopt()
4179 val = tp->rcv_nxt; in do_tcp_getsockopt()
4181 return -EINVAL; in do_tcp_getsockopt()
4185 val = READ_ONCE(icsk->icsk_user_timeout); in do_tcp_getsockopt()
4189 val = READ_ONCE(icsk->icsk_accept_queue.fastopenq.max_qlen); in do_tcp_getsockopt()
4193 val = tp->fastopen_connect; in do_tcp_getsockopt()
4197 val = tp->fastopen_no_cookie; in do_tcp_getsockopt()
4201 val = READ_ONCE(tp->tcp_tx_delay); in do_tcp_getsockopt()
4205 val = tcp_time_stamp_raw() + READ_ONCE(tp->tsoffset); in do_tcp_getsockopt()
4208 val = READ_ONCE(tp->notsent_lowat); in do_tcp_getsockopt()
4211 val = tp->recvmsg_inq; in do_tcp_getsockopt()
4214 val = tp->save_syn; in do_tcp_getsockopt()
4218 return -EFAULT; in do_tcp_getsockopt()
4221 if (tp->saved_syn) { in do_tcp_getsockopt()
4222 if (len < tcp_saved_syn_len(tp->saved_syn)) { in do_tcp_getsockopt()
4223 len = tcp_saved_syn_len(tp->saved_syn); in do_tcp_getsockopt()
4226 return -EFAULT; in do_tcp_getsockopt()
4229 return -EINVAL; in do_tcp_getsockopt()
4231 len = tcp_saved_syn_len(tp->saved_syn); in do_tcp_getsockopt()
4234 return -EFAULT; in do_tcp_getsockopt()
4236 if (copy_to_sockptr(optval, tp->saved_syn->data, len)) { in do_tcp_getsockopt()
4238 return -EFAULT; in do_tcp_getsockopt()
4246 return -EFAULT; in do_tcp_getsockopt()
4257 return -EFAULT; in do_tcp_getsockopt()
4260 return -EINVAL; in do_tcp_getsockopt()
4263 len - sizeof(zc)); in do_tcp_getsockopt()
4265 return err == 0 ? -EINVAL : err; in do_tcp_getsockopt()
4268 return -EFAULT; in do_tcp_getsockopt()
4271 return -EFAULT; in do_tcp_getsockopt()
4273 return -EINVAL; in do_tcp_getsockopt()
4275 return -EINVAL; in do_tcp_getsockopt()
4311 err = -EFAULT; in do_tcp_getsockopt()
4316 return -ENOPROTOOPT; in do_tcp_getsockopt()
4320 return -EFAULT; in do_tcp_getsockopt()
4322 return -EFAULT; in do_tcp_getsockopt()
4345 return READ_ONCE(icsk->icsk_af_ops)->getsockopt(sk, level, optname, in tcp_getsockopt()
4418 * tcp_get_md5sig_pool - get md5sig_pool for this user
4444 struct ahash_request *req = hp->md5_req; in tcp_md5_hash_skb_data()
4447 skb_headlen(skb) - header_len : 0; in tcp_md5_hash_skb_data()
4458 for (i = 0; i < shi->nr_frags; ++i) { in tcp_md5_hash_skb_data()
4459 const skb_frag_t *f = &shi->frags[i]; in tcp_md5_hash_skb_data()
4480 u8 keylen = READ_ONCE(key->keylen); /* paired with WRITE_ONCE() in tcp_md5_do_add */ in tcp_md5_hash_key()
4483 sg_init_one(&sg, key->key, keylen); in tcp_md5_hash_key()
4484 ahash_request_set_crypt(hp->md5_req, &sg, NULL, keylen); in tcp_md5_hash_key()
4486 /* We use data_race() because tcp_md5_do_add() might change key->key under us */ in tcp_md5_hash_key()
4487 return data_race(crypto_ahash_update(hp->md5_req)); in tcp_md5_hash_key()
4520 /* We've parsed the options - do we have a hash? */ in tcp_inbound_md5_hash()
4536 * IPv4-mapped case. in tcp_inbound_md5_hash()
4543 genhash = tp->af_specific->calc_md5_hash(newhash, in tcp_inbound_md5_hash()
4550 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s L3 index %d\n", in tcp_inbound_md5_hash()
4551 saddr, ntohs(th->source), in tcp_inbound_md5_hash()
4552 daddr, ntohs(th->dest), in tcp_inbound_md5_hash()
4556 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u L3 index %d\n", in tcp_inbound_md5_hash()
4558 saddr, ntohs(th->source), in tcp_inbound_md5_hash()
4559 daddr, ntohs(th->dest), l3index); in tcp_inbound_md5_hash()
4577 req = rcu_dereference_protected(tcp_sk(sk)->fastopen_rsk, 1); in tcp_done()
4579 if (sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV) in tcp_done()
4587 WRITE_ONCE(sk->sk_shutdown, SHUTDOWN_MASK); in tcp_done()
4590 sk->sk_state_change(sk); in tcp_done()
4604 inet_csk_reqsk_queue_drop(req->rsk_listener, req); in tcp_abort()
4611 refcount_inc(&tw->tw_refcnt); in tcp_abort()
4624 if (sk->sk_state == TCP_CLOSE) { in tcp_abort()
4627 return -ENOENT; in tcp_abort()
4630 if (sk->sk_state == TCP_LISTEN) { in tcp_abort()
4639 if (tcp_need_reset(sk->sk_state)) in tcp_abort()
4755 /* Set per-socket limits to no more than 1/128 the pressure threshold */ in tcp_init()
4756 limit = nr_free_buffer_pages() << (PAGE_SHIFT - 7); in tcp_init()