Lines Matching +full:tp +full:- +full:sensitive +full:- +full:adjust
1 // SPDX-License-Identifier: GPL-2.0-only
12 * Corey Minyard <wf-rch!minyard@relay.EU.net>
13 * Florian La Roche, <flla@stud.uni-sb.de>
33 * Cacophonix Gaul : draft-minshall-nagle-01
53 void tcp_mstamp_refresh(struct tcp_sock *tp) in tcp_mstamp_refresh() argument
57 tp->tcp_clock_cache = val; in tcp_mstamp_refresh()
58 tp->tcp_mstamp = div_u64(val, NSEC_PER_USEC); in tcp_mstamp_refresh()
68 struct tcp_sock *tp = tcp_sk(sk); in tcp_event_new_data_sent() local
69 unsigned int prior_packets = tp->packets_out; in tcp_event_new_data_sent()
71 WRITE_ONCE(tp->snd_nxt, TCP_SKB_CB(skb)->end_seq); in tcp_event_new_data_sent()
73 __skb_unlink(skb, &sk->sk_write_queue); in tcp_event_new_data_sent()
74 tcp_rbtree_insert(&sk->tcp_rtx_queue, skb); in tcp_event_new_data_sent()
76 if (tp->highest_sack == NULL) in tcp_event_new_data_sent()
77 tp->highest_sack = skb; in tcp_event_new_data_sent()
79 tp->packets_out += tcp_skb_pcount(skb); in tcp_event_new_data_sent()
80 if (!prior_packets || icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) in tcp_event_new_data_sent()
91 * Using SND.UNA we will fail to open window, SND.NXT is out of window. :-(
97 const struct tcp_sock *tp = tcp_sk(sk); in tcp_acceptable_seq() local
99 if (!before(tcp_wnd_end(tp), tp->snd_nxt) || in tcp_acceptable_seq()
100 (tp->rx_opt.wscale_ok && in tcp_acceptable_seq()
101 ((tp->snd_nxt - tcp_wnd_end(tp)) < (1 << tp->rx_opt.rcv_wscale)))) in tcp_acceptable_seq()
102 return tp->snd_nxt; in tcp_acceptable_seq()
104 return tcp_wnd_end(tp); in tcp_acceptable_seq()
108 * RFC1122, RFC1063, draft-ietf-tcpimpl-pmtud-01 state that:
111 * 2. Ideally, it is maximal possible segment size i.e. 65535-40.
123 struct tcp_sock *tp = tcp_sk(sk); in tcp_advertise_mss() local
125 int mss = tp->advmss; in tcp_advertise_mss()
132 tp->advmss = mss; in tcp_advertise_mss()
144 struct tcp_sock *tp = tcp_sk(sk); in tcp_cwnd_restart() local
145 u32 restart_cwnd = tcp_init_cwnd(tp, __sk_dst_get(sk)); in tcp_cwnd_restart()
146 u32 cwnd = tp->snd_cwnd; in tcp_cwnd_restart()
150 tp->snd_ssthresh = tcp_current_ssthresh(sk); in tcp_cwnd_restart()
153 while ((delta -= inet_csk(sk)->icsk_rto) > 0 && cwnd > restart_cwnd) in tcp_cwnd_restart()
155 tp->snd_cwnd = max(cwnd, restart_cwnd); in tcp_cwnd_restart()
156 tp->snd_cwnd_stamp = tcp_jiffies32; in tcp_cwnd_restart()
157 tp->snd_cwnd_used = 0; in tcp_cwnd_restart()
161 static void tcp_event_data_sent(struct tcp_sock *tp, in tcp_event_data_sent() argument
167 if (tcp_packets_in_flight(tp) == 0) in tcp_event_data_sent()
170 tp->lsndtime = now; in tcp_event_data_sent()
175 if ((u32)(now - icsk->icsk_ack.lrcvtime) < icsk->icsk_ack.ato) in tcp_event_data_sent()
183 struct tcp_sock *tp = tcp_sk(sk); in tcp_event_ack_sent() local
185 if (unlikely(tp->compressed_ack)) { in tcp_event_ack_sent()
187 tp->compressed_ack); in tcp_event_ack_sent()
188 tp->compressed_ack = 0; in tcp_event_ack_sent()
189 if (hrtimer_try_to_cancel(&tp->compressed_ack_timer) == 1) in tcp_event_ack_sent()
193 if (unlikely(rcv_nxt != tp->rcv_nxt)) in tcp_event_ack_sent()
201 * will be offered. Store the results in the tp structure.
225 * we will truncate our initial window offering to 32K-1 in tcp_select_initial_window()
230 if (sock_net(sk)->ipv4.sysctl_tcp_workaround_signed_windows) in tcp_select_initial_window()
241 space = max_t(u32, space, READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[2])); in tcp_select_initial_window()
244 *rcv_wscale = clamp_t(int, ilog2(space) - 15, in tcp_select_initial_window()
254 * value can be stuffed directly into th->window for an outgoing
259 struct tcp_sock *tp = tcp_sk(sk); in tcp_select_window() local
260 u32 old_win = tp->rcv_wnd; in tcp_select_window()
261 u32 cur_win = tcp_receive_window(tp); in tcp_select_window()
269 * window in time. --DaveM in tcp_select_window()
276 new_win = ALIGN(cur_win, 1 << tp->rx_opt.rcv_wscale); in tcp_select_window()
278 tp->rcv_wnd = new_win; in tcp_select_window()
279 tp->rcv_wup = tp->rcv_nxt; in tcp_select_window()
284 if (!tp->rx_opt.rcv_wscale && in tcp_select_window()
285 sock_net(sk)->ipv4.sysctl_tcp_workaround_signed_windows) in tcp_select_window()
288 new_win = min(new_win, (65535U << tp->rx_opt.rcv_wscale)); in tcp_select_window()
291 new_win >>= tp->rx_opt.rcv_wscale; in tcp_select_window()
295 tp->pred_flags = 0; in tcp_select_window()
306 /* Packet ECN state for a SYN-ACK */
309 const struct tcp_sock *tp = tcp_sk(sk); in tcp_ecn_send_synack() local
311 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_CWR; in tcp_ecn_send_synack()
312 if (!(tp->ecn_flags & TCP_ECN_OK)) in tcp_ecn_send_synack()
313 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_ECE; in tcp_ecn_send_synack()
322 struct tcp_sock *tp = tcp_sk(sk); in tcp_ecn_send_syn() local
324 bool use_ecn = sock_net(sk)->ipv4.sysctl_tcp_ecn == 1 || in tcp_ecn_send_syn()
334 tp->ecn_flags = 0; in tcp_ecn_send_syn()
337 TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ECE | TCPHDR_CWR; in tcp_ecn_send_syn()
338 tp->ecn_flags = TCP_ECN_OK; in tcp_ecn_send_syn()
346 if (sock_net(sk)->ipv4.sysctl_tcp_ecn_fallback) in tcp_ecn_clear_syn()
347 /* tp->ecn_flags are cleared at a later point in time when in tcp_ecn_clear_syn()
350 TCP_SKB_CB(skb)->tcp_flags &= ~(TCPHDR_ECE | TCPHDR_CWR); in tcp_ecn_clear_syn()
356 if (inet_rsk(req)->ecn_ok) in tcp_ecn_make_synack()
357 th->ece = 1; in tcp_ecn_make_synack()
366 struct tcp_sock *tp = tcp_sk(sk); in tcp_ecn_send() local
368 if (tp->ecn_flags & TCP_ECN_OK) { in tcp_ecn_send()
369 /* Not-retransmitted data segment: set ECT and inject CWR. */ in tcp_ecn_send()
370 if (skb->len != tcp_header_len && in tcp_ecn_send()
371 !before(TCP_SKB_CB(skb)->seq, tp->snd_nxt)) { in tcp_ecn_send()
373 if (tp->ecn_flags & TCP_ECN_QUEUE_CWR) { in tcp_ecn_send()
374 tp->ecn_flags &= ~TCP_ECN_QUEUE_CWR; in tcp_ecn_send()
375 th->cwr = 1; in tcp_ecn_send()
376 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN; in tcp_ecn_send()
382 if (tp->ecn_flags & TCP_ECN_DEMAND_CWR) in tcp_ecn_send()
383 th->ece = 1; in tcp_ecn_send()
387 /* Constructs common control bits of non-data skb. If SYN/FIN is present,
392 skb->ip_summed = CHECKSUM_PARTIAL; in tcp_init_nondata_skb()
394 TCP_SKB_CB(skb)->tcp_flags = flags; in tcp_init_nondata_skb()
395 TCP_SKB_CB(skb)->sacked = 0; in tcp_init_nondata_skb()
399 TCP_SKB_CB(skb)->seq = seq; in tcp_init_nondata_skb()
402 TCP_SKB_CB(skb)->end_seq = seq; in tcp_init_nondata_skb()
405 static inline bool tcp_urg_mode(const struct tcp_sock *tp) in tcp_urg_mode() argument
407 return tp->snd_una != tp->snd_up; in tcp_urg_mode()
449 if (unlikely(OPTION_MPTCP & opts->options)) in mptcp_options_write()
450 mptcp_write_options(ptr, &opts->mptcp); in mptcp_options_write()
496 * Thus, "req" is passed here and the cgroup-bpf-progs in bpf_skops_hdr_opt_len()
501 * consistent between fastopen and non-fastopen on in bpf_skops_hdr_opt_len()
524 opts->bpf_opt_len = *remaining - sock_ops.remaining_opt_len; in bpf_skops_hdr_opt_len()
526 opts->bpf_opt_len = (opts->bpf_opt_len + 3) & ~3; in bpf_skops_hdr_opt_len()
528 *remaining -= opts->bpf_opt_len; in bpf_skops_hdr_opt_len()
537 u8 first_opt_off, nr_written, max_opt_len = opts->bpf_opt_len; in bpf_skops_write_hdr_opt()
560 first_opt_off = tcp_hdrlen(skb) - max_opt_len; in bpf_skops_write_hdr_opt()
568 nr_written = max_opt_len - sock_ops.remaining_opt_len; in bpf_skops_write_hdr_opt()
571 memset(skb->data + first_opt_off + nr_written, TCPOPT_NOP, in bpf_skops_write_hdr_opt()
572 max_opt_len - nr_written); in bpf_skops_write_hdr_opt()
595 * Beware: Something in the Internet is very sensitive to the ordering of
597 * Luckily we can at least blame others for their non-compliance but from
598 * inter-operability perspective it seems that we're somewhat stuck with
606 static void tcp_options_write(__be32 *ptr, struct tcp_sock *tp, in tcp_options_write() argument
609 u16 options = opts->options; /* mungable copy */ in tcp_options_write()
615 opts->hash_location = (__u8 *)ptr; in tcp_options_write()
619 if (unlikely(opts->mss)) { in tcp_options_write()
622 opts->mss); in tcp_options_write()
638 *ptr++ = htonl(opts->tsval); in tcp_options_write()
639 *ptr++ = htonl(opts->tsecr); in tcp_options_write()
653 opts->ws); in tcp_options_write()
656 if (unlikely(opts->num_sack_blocks)) { in tcp_options_write()
657 struct tcp_sack_block *sp = tp->rx_opt.dsack ? in tcp_options_write()
658 tp->duplicate_sack : tp->selective_acks; in tcp_options_write()
664 (TCPOLEN_SACK_BASE + (opts->num_sack_blocks * in tcp_options_write()
667 for (this_sack = 0; this_sack < opts->num_sack_blocks; in tcp_options_write()
673 tp->rx_opt.dsack = 0; in tcp_options_write()
677 struct tcp_fastopen_cookie *foc = opts->fastopen_cookie; in tcp_options_write()
681 if (foc->exp) { in tcp_options_write()
682 len = TCPOLEN_EXP_FASTOPEN_BASE + foc->len; in tcp_options_write()
687 len = TCPOLEN_FASTOPEN_BASE + foc->len; in tcp_options_write()
692 memcpy(p, foc->val, foc->len); in tcp_options_write()
694 p[foc->len] = TCPOPT_NOP; in tcp_options_write()
695 p[foc->len + 1] = TCPOPT_NOP; in tcp_options_write()
705 static void smc_set_option(const struct tcp_sock *tp, in smc_set_option() argument
711 if (tp->syn_smc) { in smc_set_option()
713 opts->options |= OPTION_SMC; in smc_set_option()
714 *remaining -= TCPOLEN_EXP_SMC_BASE_ALIGNED; in smc_set_option()
721 static void smc_set_option_cond(const struct tcp_sock *tp, in smc_set_option_cond() argument
728 if (tp->syn_smc && ireq->smc_ok) { in smc_set_option_cond()
730 opts->options |= OPTION_SMC; in smc_set_option_cond()
731 *remaining -= TCPOLEN_EXP_SMC_BASE_ALIGNED; in smc_set_option_cond()
745 if (mptcp_synack_options(req, &size, &opts->mptcp)) { in mptcp_set_option_cond()
747 opts->options |= OPTION_MPTCP; in mptcp_set_option_cond()
748 *remaining -= size; in mptcp_set_option_cond()
761 struct tcp_sock *tp = tcp_sk(sk); in tcp_syn_options() local
763 struct tcp_fastopen_request *fastopen = tp->fastopen_req; in tcp_syn_options()
768 rcu_access_pointer(tp->md5sig_info)) { in tcp_syn_options()
769 *md5 = tp->af_specific->md5_lookup(sk, sk); in tcp_syn_options()
771 opts->options |= OPTION_MD5; in tcp_syn_options()
772 remaining -= TCPOLEN_MD5SIG_ALIGNED; in tcp_syn_options()
779 * advertised. But we subtract them from tp->mss_cache so that in tcp_syn_options()
786 opts->mss = tcp_advertise_mss(sk); in tcp_syn_options()
787 remaining -= TCPOLEN_MSS_ALIGNED; in tcp_syn_options()
789 if (likely(READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_timestamps) && !*md5)) { in tcp_syn_options()
790 opts->options |= OPTION_TS; in tcp_syn_options()
791 opts->tsval = tcp_skb_timestamp(skb) + tp->tsoffset; in tcp_syn_options()
792 opts->tsecr = tp->rx_opt.ts_recent; in tcp_syn_options()
793 remaining -= TCPOLEN_TSTAMP_ALIGNED; in tcp_syn_options()
795 if (likely(READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_window_scaling))) { in tcp_syn_options()
796 opts->ws = tp->rx_opt.rcv_wscale; in tcp_syn_options()
797 opts->options |= OPTION_WSCALE; in tcp_syn_options()
798 remaining -= TCPOLEN_WSCALE_ALIGNED; in tcp_syn_options()
800 if (likely(READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_sack))) { in tcp_syn_options()
801 opts->options |= OPTION_SACK_ADVERTISE; in tcp_syn_options()
802 if (unlikely(!(OPTION_TS & opts->options))) in tcp_syn_options()
803 remaining -= TCPOLEN_SACKPERM_ALIGNED; in tcp_syn_options()
806 if (fastopen && fastopen->cookie.len >= 0) { in tcp_syn_options()
807 u32 need = fastopen->cookie.len; in tcp_syn_options()
809 need += fastopen->cookie.exp ? TCPOLEN_EXP_FASTOPEN_BASE : in tcp_syn_options()
813 opts->options |= OPTION_FAST_OPEN_COOKIE; in tcp_syn_options()
814 opts->fastopen_cookie = &fastopen->cookie; in tcp_syn_options()
815 remaining -= need; in tcp_syn_options()
816 tp->syn_fastopen = 1; in tcp_syn_options()
817 tp->syn_fastopen_exp = fastopen->cookie.exp ? 1 : 0; in tcp_syn_options()
821 smc_set_option(tp, opts, &remaining); in tcp_syn_options()
826 if (mptcp_syn_options(sk, skb, &size, &opts->mptcp)) { in tcp_syn_options()
827 opts->options |= OPTION_MPTCP; in tcp_syn_options()
828 remaining -= size; in tcp_syn_options()
834 return MAX_TCP_OPTION_SPACE - remaining; in tcp_syn_options()
837 /* Set up TCP options for SYN-ACKs. */
852 opts->options |= OPTION_MD5; in tcp_synack_options()
853 remaining -= TCPOLEN_MD5SIG_ALIGNED; in tcp_synack_options()
861 ireq->tstamp_ok &= !ireq->sack_ok; in tcp_synack_options()
866 opts->mss = mss; in tcp_synack_options()
867 remaining -= TCPOLEN_MSS_ALIGNED; in tcp_synack_options()
869 if (likely(ireq->wscale_ok)) { in tcp_synack_options()
870 opts->ws = ireq->rcv_wscale; in tcp_synack_options()
871 opts->options |= OPTION_WSCALE; in tcp_synack_options()
872 remaining -= TCPOLEN_WSCALE_ALIGNED; in tcp_synack_options()
874 if (likely(ireq->tstamp_ok)) { in tcp_synack_options()
875 opts->options |= OPTION_TS; in tcp_synack_options()
876 opts->tsval = tcp_skb_timestamp(skb) + tcp_rsk(req)->ts_off; in tcp_synack_options()
877 opts->tsecr = req->ts_recent; in tcp_synack_options()
878 remaining -= TCPOLEN_TSTAMP_ALIGNED; in tcp_synack_options()
880 if (likely(ireq->sack_ok)) { in tcp_synack_options()
881 opts->options |= OPTION_SACK_ADVERTISE; in tcp_synack_options()
882 if (unlikely(!ireq->tstamp_ok)) in tcp_synack_options()
883 remaining -= TCPOLEN_SACKPERM_ALIGNED; in tcp_synack_options()
885 if (foc != NULL && foc->len >= 0) { in tcp_synack_options()
886 u32 need = foc->len; in tcp_synack_options()
888 need += foc->exp ? TCPOLEN_EXP_FASTOPEN_BASE : in tcp_synack_options()
892 opts->options |= OPTION_FAST_OPEN_COOKIE; in tcp_synack_options()
893 opts->fastopen_cookie = foc; in tcp_synack_options()
894 remaining -= need; in tcp_synack_options()
905 return MAX_TCP_OPTION_SPACE - remaining; in tcp_synack_options()
915 struct tcp_sock *tp = tcp_sk(sk); in tcp_established_options() local
919 opts->options = 0; in tcp_established_options()
924 rcu_access_pointer(tp->md5sig_info)) { in tcp_established_options()
925 *md5 = tp->af_specific->md5_lookup(sk, sk); in tcp_established_options()
927 opts->options |= OPTION_MD5; in tcp_established_options()
933 if (likely(tp->rx_opt.tstamp_ok)) { in tcp_established_options()
934 opts->options |= OPTION_TS; in tcp_established_options()
935 opts->tsval = skb ? tcp_skb_timestamp(skb) + tp->tsoffset : 0; in tcp_established_options()
936 opts->tsecr = tp->rx_opt.ts_recent; in tcp_established_options()
947 unsigned int remaining = MAX_TCP_OPTION_SPACE - size; in tcp_established_options()
951 &opts->mptcp)) { in tcp_established_options()
952 opts->options |= OPTION_MPTCP; in tcp_established_options()
957 eff_sacks = tp->rx_opt.num_sacks + tp->rx_opt.dsack; in tcp_established_options()
959 const unsigned int remaining = MAX_TCP_OPTION_SPACE - size; in tcp_established_options()
964 opts->num_sack_blocks = in tcp_established_options()
966 (remaining - TCPOLEN_SACK_BASE_ALIGNED) / in tcp_established_options()
970 opts->num_sack_blocks * TCPOLEN_SACK_PERBLOCK; in tcp_established_options()
973 if (unlikely(BPF_SOCK_OPS_TEST_FLAG(tp, in tcp_established_options()
975 unsigned int remaining = MAX_TCP_OPTION_SPACE - size; in tcp_established_options()
979 size = MAX_TCP_OPTION_SPACE - remaining; in tcp_established_options()
994 * The invariant being skb->truesize subtracted from sk->sk_wmem_alloc
1008 if ((1 << sk->sk_state) & in tcp_tsq_write()
1011 struct tcp_sock *tp = tcp_sk(sk); in tcp_tsq_write() local
1013 if (tp->lost_out > tp->retrans_out && in tcp_tsq_write()
1014 tp->snd_cwnd > tcp_packets_in_flight(tp)) { in tcp_tsq_write()
1015 tcp_mstamp_refresh(tp); in tcp_tsq_write()
1019 tcp_write_xmit(sk, tcp_current_mss(sk), tp->nonagle, in tcp_tsq_write()
1029 else if (!test_and_set_bit(TCP_TSQ_DEFERRED, &sk->sk_tsq_flags)) in tcp_tsq_handler()
1036 * transferring tsq->head because tcp_wfree() might
1045 struct tcp_sock *tp; in tcp_tasklet_func() local
1049 list_splice_init(&tsq->head, &list); in tcp_tasklet_func()
1053 tp = list_entry(q, struct tcp_sock, tsq_node); in tcp_tasklet_func()
1054 list_del(&tp->tsq_node); in tcp_tasklet_func()
1056 sk = (struct sock *)tp; in tcp_tasklet_func()
1058 clear_bit(TSQ_QUEUED, &sk->sk_tsq_flags); in tcp_tasklet_func()
1070 * tcp_release_cb - tcp release_sock() callback
1082 flags = sk->sk_tsq_flags; in tcp_release_cb()
1086 } while (cmpxchg(&sk->sk_tsq_flags, flags, nflags) != flags); in tcp_release_cb()
1096 * 3) socket owned by us (sk->sk_lock.owned == 1) in tcp_release_cb()
1112 inet_csk(sk)->icsk_af_ops->mtu_reduced(sk); in tcp_release_cb()
1125 INIT_LIST_HEAD(&tsq->head); in tcp_tasklet_init()
1126 tasklet_init(&tsq->tasklet, in tcp_tasklet_init()
1139 struct sock *sk = skb->sk; in tcp_wfree()
1140 struct tcp_sock *tp = tcp_sk(sk); in tcp_wfree() local
1146 WARN_ON(refcount_sub_and_test(skb->truesize - 1, &sk->sk_wmem_alloc)); in tcp_wfree()
1151 * - less callbacks to tcp_write_xmit(), reducing stress (batches) in tcp_wfree()
1152 * - chance for incoming ACK (processed by another cpu maybe) in tcp_wfree()
1153 * to migrate this flow (skb->ooo_okay will be eventually set) in tcp_wfree()
1155 if (refcount_read(&sk->sk_wmem_alloc) >= SKB_TRUESIZE(1) && this_cpu_ksoftirqd() == current) in tcp_wfree()
1158 for (oval = READ_ONCE(sk->sk_tsq_flags);; oval = nval) { in tcp_wfree()
1166 nval = cmpxchg(&sk->sk_tsq_flags, oval, nval); in tcp_wfree()
1173 empty = list_empty(&tsq->head); in tcp_wfree()
1174 list_add(&tp->tsq_node, &tsq->head); in tcp_wfree()
1176 tasklet_schedule(&tsq->tasklet); in tcp_wfree()
1189 struct tcp_sock *tp = container_of(timer, struct tcp_sock, pacing_timer); in tcp_pace_kick() local
1190 struct sock *sk = (struct sock *)tp; in tcp_pace_kick()
1201 struct tcp_sock *tp = tcp_sk(sk); in tcp_update_skb_after_send() local
1203 if (sk->sk_pacing_status != SK_PACING_NONE) { in tcp_update_skb_after_send()
1204 unsigned long rate = sk->sk_pacing_rate; in tcp_update_skb_after_send()
1207 * Note that tp->data_segs_out overflows after 2^32 packets, in tcp_update_skb_after_send()
1210 if (rate != ~0UL && rate && tp->data_segs_out >= 10) { in tcp_update_skb_after_send()
1211 u64 len_ns = div64_ul((u64)skb->len * NSEC_PER_SEC, rate); in tcp_update_skb_after_send()
1212 u64 credit = tp->tcp_wstamp_ns - prior_wstamp; in tcp_update_skb_after_send()
1215 len_ns -= min_t(u64, len_ns / 2, credit); in tcp_update_skb_after_send()
1216 tp->tcp_wstamp_ns += len_ns; in tcp_update_skb_after_send()
1219 list_move_tail(&skb->tcp_tsorted_anchor, &tp->tsorted_sent_queue); in tcp_update_skb_after_send()
1242 struct tcp_sock *tp; in __tcp_transmit_skb() local
1253 tp = tcp_sk(sk); in __tcp_transmit_skb()
1254 prior_wstamp = tp->tcp_wstamp_ns; in __tcp_transmit_skb()
1255 tp->tcp_wstamp_ns = max(tp->tcp_wstamp_ns, tp->tcp_clock_cache); in __tcp_transmit_skb()
1256 skb->skb_mstamp_ns = tp->tcp_wstamp_ns; in __tcp_transmit_skb()
1258 TCP_SKB_CB(skb)->tx.in_flight = TCP_SKB_CB(skb)->end_seq in __tcp_transmit_skb()
1259 - tp->snd_una; in __tcp_transmit_skb()
1270 return -ENOBUFS; in __tcp_transmit_skb()
1271 /* retransmit skbs might have a non zero value in skb->dev in __tcp_transmit_skb()
1272 * because skb->dev is aliased with skb->rbnode.rb_left in __tcp_transmit_skb()
1274 skb->dev = NULL; in __tcp_transmit_skb()
1281 if (unlikely(tcb->tcp_flags & TCPHDR_SYN)) { in __tcp_transmit_skb()
1290 * and in this case it is better to delay the delivery of 1-MSS in __tcp_transmit_skb()
1295 tcb->tcp_flags |= TCPHDR_PSH; in __tcp_transmit_skb()
1303 * TODO: Ideally, in-flight pure ACK packets should not matter here. in __tcp_transmit_skb()
1304 * One way to get this would be to set skb->truesize = 2 on them. in __tcp_transmit_skb()
1306 skb->ooo_okay = sk_wmem_alloc_get(sk) < SKB_TRUESIZE(1); in __tcp_transmit_skb()
1313 skb->pfmemalloc = 0; in __tcp_transmit_skb()
1319 skb->sk = sk; in __tcp_transmit_skb()
1320 skb->destructor = skb_is_tcp_pure_ack(skb) ? __sock_wfree : tcp_wfree; in __tcp_transmit_skb()
1322 refcount_add(skb->truesize, &sk->sk_wmem_alloc); in __tcp_transmit_skb()
1324 skb_set_dst_pending_confirm(skb, sk->sk_dst_pending_confirm); in __tcp_transmit_skb()
1327 th = (struct tcphdr *)skb->data; in __tcp_transmit_skb()
1328 th->source = inet->inet_sport; in __tcp_transmit_skb()
1329 th->dest = inet->inet_dport; in __tcp_transmit_skb()
1330 th->seq = htonl(tcb->seq); in __tcp_transmit_skb()
1331 th->ack_seq = htonl(rcv_nxt); in __tcp_transmit_skb()
1333 tcb->tcp_flags); in __tcp_transmit_skb()
1335 th->check = 0; in __tcp_transmit_skb()
1336 th->urg_ptr = 0; in __tcp_transmit_skb()
1339 if (unlikely(tcp_urg_mode(tp) && before(tcb->seq, tp->snd_up))) { in __tcp_transmit_skb()
1340 if (before(tp->snd_up, tcb->seq + 0x10000)) { in __tcp_transmit_skb()
1341 th->urg_ptr = htons(tp->snd_up - tcb->seq); in __tcp_transmit_skb()
1342 th->urg = 1; in __tcp_transmit_skb()
1343 } else if (after(tcb->seq + 0xFFFF, tp->snd_nxt)) { in __tcp_transmit_skb()
1344 th->urg_ptr = htons(0xFFFF); in __tcp_transmit_skb()
1345 th->urg = 1; in __tcp_transmit_skb()
1349 tcp_options_write((__be32 *)(th + 1), tp, &opts); in __tcp_transmit_skb()
1350 skb_shinfo(skb)->gso_type = sk->sk_gso_type; in __tcp_transmit_skb()
1351 if (likely(!(tcb->tcp_flags & TCPHDR_SYN))) { in __tcp_transmit_skb()
1352 th->window = htons(tcp_select_window(sk)); in __tcp_transmit_skb()
1358 th->window = htons(min(tp->rcv_wnd, 65535U)); in __tcp_transmit_skb()
1364 tp->af_specific->calc_md5_hash(opts.hash_location, in __tcp_transmit_skb()
1372 INDIRECT_CALL_INET(icsk->icsk_af_ops->send_check, in __tcp_transmit_skb()
1376 if (likely(tcb->tcp_flags & TCPHDR_ACK)) in __tcp_transmit_skb()
1379 if (skb->len != tcp_header_size) { in __tcp_transmit_skb()
1380 tcp_event_data_sent(tp, sk); in __tcp_transmit_skb()
1381 tp->data_segs_out += tcp_skb_pcount(skb); in __tcp_transmit_skb()
1382 tp->bytes_sent += skb->len - tcp_header_size; in __tcp_transmit_skb()
1385 if (after(tcb->end_seq, tp->snd_nxt) || tcb->seq == tcb->end_seq) in __tcp_transmit_skb()
1389 tp->segs_out += tcp_skb_pcount(skb); in __tcp_transmit_skb()
1390 /* OK, its time to fill skb_shinfo(skb)->gso_{segs|size} */ in __tcp_transmit_skb()
1391 skb_shinfo(skb)->gso_segs = tcp_skb_pcount(skb); in __tcp_transmit_skb()
1392 skb_shinfo(skb)->gso_size = tcp_skb_mss(skb); in __tcp_transmit_skb()
1394 /* Leave earliest departure time in skb->tstamp (skb->skb_mstamp_ns) */ in __tcp_transmit_skb()
1397 memset(skb->cb, 0, max(sizeof(struct inet_skb_parm), in __tcp_transmit_skb()
1400 tcp_add_tx_delay(skb, tp); in __tcp_transmit_skb()
1402 err = INDIRECT_CALL_INET(icsk->icsk_af_ops->queue_xmit, in __tcp_transmit_skb()
1404 sk, skb, &inet->cork.fl); in __tcp_transmit_skb()
1421 tcp_sk(sk)->rcv_nxt); in tcp_transmit_skb()
1431 struct tcp_sock *tp = tcp_sk(sk); in tcp_queue_skb() local
1434 WRITE_ONCE(tp->write_seq, TCP_SKB_CB(skb)->end_seq); in tcp_queue_skb()
1437 sk_wmem_queued_add(sk, skb->truesize); in tcp_queue_skb()
1438 sk_mem_charge(sk, skb->truesize); in tcp_queue_skb()
1444 if (skb->len <= mss_now) { in tcp_set_skb_tso_segs()
1446 * non-TSO case. in tcp_set_skb_tso_segs()
1449 TCP_SKB_CB(skb)->tcp_gso_size = 0; in tcp_set_skb_tso_segs()
1451 tcp_skb_pcount_set(skb, DIV_ROUND_UP(skb->len, mss_now)); in tcp_set_skb_tso_segs()
1452 TCP_SKB_CB(skb)->tcp_gso_size = mss_now; in tcp_set_skb_tso_segs()
1461 struct tcp_sock *tp = tcp_sk(sk); in tcp_adjust_pcount() local
1463 tp->packets_out -= decr; in tcp_adjust_pcount()
1465 if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED) in tcp_adjust_pcount()
1466 tp->sacked_out -= decr; in tcp_adjust_pcount()
1467 if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) in tcp_adjust_pcount()
1468 tp->retrans_out -= decr; in tcp_adjust_pcount()
1469 if (TCP_SKB_CB(skb)->sacked & TCPCB_LOST) in tcp_adjust_pcount()
1470 tp->lost_out -= decr; in tcp_adjust_pcount()
1473 if (tcp_is_reno(tp) && decr > 0) in tcp_adjust_pcount()
1474 tp->sacked_out -= min_t(u32, tp->sacked_out, decr); in tcp_adjust_pcount()
1476 if (tp->lost_skb_hint && in tcp_adjust_pcount()
1477 before(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(tp->lost_skb_hint)->seq) && in tcp_adjust_pcount()
1478 (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)) in tcp_adjust_pcount()
1479 tp->lost_cnt_hint -= decr; in tcp_adjust_pcount()
1481 tcp_verify_left_out(tp); in tcp_adjust_pcount()
1486 return TCP_SKB_CB(skb)->txstamp_ack || in tcp_has_tx_tstamp()
1487 (skb_shinfo(skb)->tx_flags & SKBTX_ANY_TSTAMP); in tcp_has_tx_tstamp()
1495 !before(shinfo->tskey, TCP_SKB_CB(skb2)->seq)) { in tcp_fragment_tstamp()
1497 u8 tsflags = shinfo->tx_flags & SKBTX_ANY_TSTAMP; in tcp_fragment_tstamp()
1499 shinfo->tx_flags &= ~tsflags; in tcp_fragment_tstamp()
1500 shinfo2->tx_flags |= tsflags; in tcp_fragment_tstamp()
1501 swap(shinfo->tskey, shinfo2->tskey); in tcp_fragment_tstamp()
1502 TCP_SKB_CB(skb2)->txstamp_ack = TCP_SKB_CB(skb)->txstamp_ack; in tcp_fragment_tstamp()
1503 TCP_SKB_CB(skb)->txstamp_ack = 0; in tcp_fragment_tstamp()
1509 TCP_SKB_CB(skb2)->eor = TCP_SKB_CB(skb)->eor; in tcp_skb_fragment_eor()
1510 TCP_SKB_CB(skb)->eor = 0; in tcp_skb_fragment_eor()
1520 __skb_queue_after(&sk->sk_write_queue, skb, buff); in tcp_insert_write_queue_after()
1522 tcp_rbtree_insert(&sk->tcp_rtx_queue, buff); in tcp_insert_write_queue_after()
1534 struct tcp_sock *tp = tcp_sk(sk); in tcp_fragment() local
1541 if (WARN_ON(len > skb->len)) in tcp_fragment()
1542 return -EINVAL; in tcp_fragment()
1544 nsize = skb_headlen(skb) - len; in tcp_fragment()
1553 limit = sk->sk_sndbuf + 2 * SKB_TRUESIZE(GSO_MAX_SIZE); in tcp_fragment()
1554 if (unlikely((sk->sk_wmem_queued >> 1) > limit && in tcp_fragment()
1559 return -ENOMEM; in tcp_fragment()
1563 return -ENOMEM; in tcp_fragment()
1568 return -ENOMEM; /* We'll just try again later. */ in tcp_fragment()
1571 sk_wmem_queued_add(sk, buff->truesize); in tcp_fragment()
1572 sk_mem_charge(sk, buff->truesize); in tcp_fragment()
1573 nlen = skb->len - len - nsize; in tcp_fragment()
1574 buff->truesize += nlen; in tcp_fragment()
1575 skb->truesize -= nlen; in tcp_fragment()
1578 TCP_SKB_CB(buff)->seq = TCP_SKB_CB(skb)->seq + len; in tcp_fragment()
1579 TCP_SKB_CB(buff)->end_seq = TCP_SKB_CB(skb)->end_seq; in tcp_fragment()
1580 TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(buff)->seq; in tcp_fragment()
1583 flags = TCP_SKB_CB(skb)->tcp_flags; in tcp_fragment()
1584 TCP_SKB_CB(skb)->tcp_flags = flags & ~(TCPHDR_FIN | TCPHDR_PSH); in tcp_fragment()
1585 TCP_SKB_CB(buff)->tcp_flags = flags; in tcp_fragment()
1586 TCP_SKB_CB(buff)->sacked = TCP_SKB_CB(skb)->sacked; in tcp_fragment()
1591 buff->ip_summed = CHECKSUM_PARTIAL; in tcp_fragment()
1593 buff->tstamp = skb->tstamp; in tcp_fragment()
1603 TCP_SKB_CB(buff)->tx = TCP_SKB_CB(skb)->tx; in tcp_fragment()
1606 * adjust the various packet counters. in tcp_fragment()
1608 if (!before(tp->snd_nxt, TCP_SKB_CB(buff)->end_seq)) { in tcp_fragment()
1609 int diff = old_factor - tcp_skb_pcount(skb) - in tcp_fragment()
1620 list_add(&buff->tcp_tsorted_anchor, &skb->tcp_tsorted_anchor); in tcp_fragment()
1636 len -= eat; in __pskb_trim_head()
1643 for (i = 0; i < shinfo->nr_frags; i++) { in __pskb_trim_head()
1644 int size = skb_frag_size(&shinfo->frags[i]); in __pskb_trim_head()
1648 eat -= size; in __pskb_trim_head()
1650 shinfo->frags[k] = shinfo->frags[i]; in __pskb_trim_head()
1652 skb_frag_off_add(&shinfo->frags[k], eat); in __pskb_trim_head()
1653 skb_frag_size_sub(&shinfo->frags[k], eat); in __pskb_trim_head()
1659 shinfo->nr_frags = k; in __pskb_trim_head()
1661 skb->data_len -= len; in __pskb_trim_head()
1662 skb->len = skb->data_len; in __pskb_trim_head()
1672 return -ENOMEM; in tcp_trim_head()
1676 TCP_SKB_CB(skb)->seq += len; in tcp_trim_head()
1677 skb->ip_summed = CHECKSUM_PARTIAL; in tcp_trim_head()
1680 skb->truesize -= delta_truesize; in tcp_trim_head()
1681 sk_wmem_queued_add(sk, -delta_truesize); in tcp_trim_head()
1685 /* Any change of skb->len requires recalculation of tso factor. */ in tcp_trim_head()
1695 const struct tcp_sock *tp = tcp_sk(sk); in __tcp_mtu_to_mss() local
1700 It is MMS_S - sizeof(tcphdr) of rfc1122 in __tcp_mtu_to_mss()
1702 mss_now = pmtu - icsk->icsk_af_ops->net_header_len - sizeof(struct tcphdr); in __tcp_mtu_to_mss()
1705 if (icsk->icsk_af_ops->net_frag_header_len) { in __tcp_mtu_to_mss()
1709 mss_now -= icsk->icsk_af_ops->net_frag_header_len; in __tcp_mtu_to_mss()
1713 if (mss_now > tp->rx_opt.mss_clamp) in __tcp_mtu_to_mss()
1714 mss_now = tp->rx_opt.mss_clamp; in __tcp_mtu_to_mss()
1717 mss_now -= icsk->icsk_ext_hdr_len; in __tcp_mtu_to_mss()
1721 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_min_snd_mss)); in __tcp_mtu_to_mss()
1729 return __tcp_mtu_to_mss(sk, pmtu) - in tcp_mtu_to_mss()
1730 (tcp_sk(sk)->tcp_header_len - sizeof(struct tcphdr)); in tcp_mtu_to_mss()
1737 const struct tcp_sock *tp = tcp_sk(sk); in tcp_mss_to_mtu() local
1742 tp->tcp_header_len + in tcp_mss_to_mtu()
1743 icsk->icsk_ext_hdr_len + in tcp_mss_to_mtu()
1744 icsk->icsk_af_ops->net_header_len; in tcp_mss_to_mtu()
1747 if (icsk->icsk_af_ops->net_frag_header_len) { in tcp_mss_to_mtu()
1751 mtu += icsk->icsk_af_ops->net_frag_header_len; in tcp_mss_to_mtu()
1760 struct tcp_sock *tp = tcp_sk(sk); in tcp_mtup_init() local
1764 icsk->icsk_mtup.enabled = READ_ONCE(net->ipv4.sysctl_tcp_mtu_probing) > 1; in tcp_mtup_init()
1765 icsk->icsk_mtup.search_high = tp->rx_opt.mss_clamp + sizeof(struct tcphdr) + in tcp_mtup_init()
1766 icsk->icsk_af_ops->net_header_len; in tcp_mtup_init()
1767 icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, READ_ONCE(net->ipv4.sysctl_tcp_base_mss)); in tcp_mtup_init()
1768 icsk->icsk_mtup.probe_size = 0; in tcp_mtup_init()
1769 if (icsk->icsk_mtup.enabled) in tcp_mtup_init()
1770 icsk->icsk_mtup.probe_timestamp = tcp_jiffies32; in tcp_mtup_init()
1776 tp->rx_opt.user_mss is mss set by user by TCP_MAXSEG. It does NOT counts
1779 tp->rx_opt.mss_clamp is mss negotiated at connection setup.
1783 inet_csk(sk)->icsk_pmtu_cookie is last pmtu, seen by this function.
1785 tp->mss_cache is current effective sending mss, including
1788 tp->rx_opt.mss_clamp.
1793 NOTE2. inet_csk(sk)->icsk_pmtu_cookie and tp->mss_cache
1794 are READ ONLY outside this function. --ANK (980731)
1798 struct tcp_sock *tp = tcp_sk(sk); in tcp_sync_mss() local
1802 if (icsk->icsk_mtup.search_high > pmtu) in tcp_sync_mss()
1803 icsk->icsk_mtup.search_high = pmtu; in tcp_sync_mss()
1806 mss_now = tcp_bound_to_half_wnd(tp, mss_now); in tcp_sync_mss()
1809 icsk->icsk_pmtu_cookie = pmtu; in tcp_sync_mss()
1810 if (icsk->icsk_mtup.enabled) in tcp_sync_mss()
1811 mss_now = min(mss_now, tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low)); in tcp_sync_mss()
1812 tp->mss_cache = mss_now; in tcp_sync_mss()
1823 const struct tcp_sock *tp = tcp_sk(sk); in tcp_current_mss() local
1830 mss_now = tp->mss_cache; in tcp_current_mss()
1834 if (mtu != inet_csk(sk)->icsk_pmtu_cookie) in tcp_current_mss()
1840 /* The mss_cache is sized based on tp->tcp_header_len, which assumes in tcp_current_mss()
1843 * we have to adjust mss_now correspondingly */ in tcp_current_mss()
1844 if (header_len != tp->tcp_header_len) { in tcp_current_mss()
1845 int delta = (int) header_len - tp->tcp_header_len; in tcp_current_mss()
1846 mss_now -= delta; in tcp_current_mss()
1852 /* RFC2861, slow part. Adjust cwnd, after it was not full during one rto.
1858 struct tcp_sock *tp = tcp_sk(sk); in tcp_cwnd_application_limited() local
1860 if (inet_csk(sk)->icsk_ca_state == TCP_CA_Open && in tcp_cwnd_application_limited()
1861 sk->sk_socket && !test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { in tcp_cwnd_application_limited()
1863 u32 init_win = tcp_init_cwnd(tp, __sk_dst_get(sk)); in tcp_cwnd_application_limited()
1864 u32 win_used = max(tp->snd_cwnd_used, init_win); in tcp_cwnd_application_limited()
1865 if (win_used < tp->snd_cwnd) { in tcp_cwnd_application_limited()
1866 tp->snd_ssthresh = tcp_current_ssthresh(sk); in tcp_cwnd_application_limited()
1867 tp->snd_cwnd = (tp->snd_cwnd + win_used) >> 1; in tcp_cwnd_application_limited()
1869 tp->snd_cwnd_used = 0; in tcp_cwnd_application_limited()
1871 tp->snd_cwnd_stamp = tcp_jiffies32; in tcp_cwnd_application_limited()
1876 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; in tcp_cwnd_validate()
1877 struct tcp_sock *tp = tcp_sk(sk); in tcp_cwnd_validate() local
1880 * is fully utilized. If cwnd-limited then remember that fact for the in tcp_cwnd_validate()
1881 * current window. If not cwnd-limited then track the maximum number of in tcp_cwnd_validate()
1882 * outstanding packets in the current window. (If cwnd-limited then we in tcp_cwnd_validate()
1883 * chose to not update tp->max_packets_out to avoid an extra else in tcp_cwnd_validate()
1886 if (!before(tp->snd_una, tp->cwnd_usage_seq) || in tcp_cwnd_validate()
1888 (!tp->is_cwnd_limited && in tcp_cwnd_validate()
1889 tp->packets_out > tp->max_packets_out)) { in tcp_cwnd_validate()
1890 tp->is_cwnd_limited = is_cwnd_limited; in tcp_cwnd_validate()
1891 tp->max_packets_out = tp->packets_out; in tcp_cwnd_validate()
1892 tp->cwnd_usage_seq = tp->snd_nxt; in tcp_cwnd_validate()
1897 tp->snd_cwnd_used = 0; in tcp_cwnd_validate()
1898 tp->snd_cwnd_stamp = tcp_jiffies32; in tcp_cwnd_validate()
1901 if (tp->packets_out > tp->snd_cwnd_used) in tcp_cwnd_validate()
1902 tp->snd_cwnd_used = tp->packets_out; in tcp_cwnd_validate()
1904 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) && in tcp_cwnd_validate()
1905 (s32)(tcp_jiffies32 - tp->snd_cwnd_stamp) >= inet_csk(sk)->icsk_rto && in tcp_cwnd_validate()
1906 !ca_ops->cong_control) in tcp_cwnd_validate()
1916 if (tcp_write_queue_empty(sk) && sk->sk_socket && in tcp_cwnd_validate()
1917 test_bit(SOCK_NOSPACE, &sk->sk_socket->flags) && in tcp_cwnd_validate()
1918 (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) in tcp_cwnd_validate()
1924 static bool tcp_minshall_check(const struct tcp_sock *tp) in tcp_minshall_check() argument
1926 return after(tp->snd_sml, tp->snd_una) && in tcp_minshall_check()
1927 !after(tp->snd_sml, tp->snd_nxt); in tcp_minshall_check()
1931 * Note that a TSO packet might end with a sub-mss segment
1933 * if ((skb->len % mss) != 0)
1934 * tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
1936 * skb_pcount = skb->len / mss_now
1938 static void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss_now, in tcp_minshall_update() argument
1941 if (skb->len < tcp_skb_pcount(skb) * mss_now) in tcp_minshall_update()
1942 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; in tcp_minshall_update()
1952 static bool tcp_nagle_check(bool partial, const struct tcp_sock *tp, in tcp_nagle_check() argument
1957 (!nonagle && tp->packets_out && tcp_minshall_check(tp))); in tcp_nagle_check()
1969 sk->sk_pacing_rate >> READ_ONCE(sk->sk_pacing_shift), in tcp_tso_autosize()
1970 sk->sk_gso_max_size - 1 - MAX_TCP_HEADER); in tcp_tso_autosize()
1987 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; in tcp_tso_segs()
1990 min_tso = ca_ops->min_tso_segs ? in tcp_tso_segs()
1991 ca_ops->min_tso_segs(sk) : in tcp_tso_segs()
1992 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_min_tso_segs); in tcp_tso_segs()
1995 return min_t(u32, tso_segs, sk->sk_gso_max_segs); in tcp_tso_segs()
2005 const struct tcp_sock *tp = tcp_sk(sk); in tcp_mss_split_point() local
2008 window = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq; in tcp_mss_split_point()
2014 needed = min(skb->len, window); in tcp_mss_split_point()
2024 if (tcp_nagle_check(partial != 0, tp, nonagle)) in tcp_mss_split_point()
2025 return needed - partial; in tcp_mss_split_point()
2033 static inline unsigned int tcp_cwnd_test(const struct tcp_sock *tp, in tcp_cwnd_test() argument
2039 if ((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) && in tcp_cwnd_test()
2043 in_flight = tcp_packets_in_flight(tp); in tcp_cwnd_test()
2044 cwnd = tp->snd_cwnd; in tcp_cwnd_test()
2052 return min(halfcwnd, cwnd - in_flight); in tcp_cwnd_test()
2074 static inline bool tcp_nagle_test(const struct tcp_sock *tp, const struct sk_buff *skb, in tcp_nagle_test() argument
2087 if (tcp_urg_mode(tp) || (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)) in tcp_nagle_test()
2090 if (!tcp_nagle_check(skb->len < cur_mss, tp, nonagle)) in tcp_nagle_test()
2097 static bool tcp_snd_wnd_test(const struct tcp_sock *tp, in tcp_snd_wnd_test() argument
2101 u32 end_seq = TCP_SKB_CB(skb)->end_seq; in tcp_snd_wnd_test()
2103 if (skb->len > cur_mss) in tcp_snd_wnd_test()
2104 end_seq = TCP_SKB_CB(skb)->seq + cur_mss; in tcp_snd_wnd_test()
2106 return !after(end_seq, tcp_wnd_end(tp)); in tcp_snd_wnd_test()
2113 * know that all the data is in scatter-gather pages, and that the
2119 int nlen = skb->len - len; in tso_fragment()
2124 if (skb->len != skb->data_len) in tso_fragment()
2130 return -ENOMEM; in tso_fragment()
2133 sk_wmem_queued_add(sk, buff->truesize); in tso_fragment()
2134 sk_mem_charge(sk, buff->truesize); in tso_fragment()
2135 buff->truesize += nlen; in tso_fragment()
2136 skb->truesize -= nlen; in tso_fragment()
2139 TCP_SKB_CB(buff)->seq = TCP_SKB_CB(skb)->seq + len; in tso_fragment()
2140 TCP_SKB_CB(buff)->end_seq = TCP_SKB_CB(skb)->end_seq; in tso_fragment()
2141 TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(buff)->seq; in tso_fragment()
2144 flags = TCP_SKB_CB(skb)->tcp_flags; in tso_fragment()
2145 TCP_SKB_CB(skb)->tcp_flags = flags & ~(TCPHDR_FIN | TCPHDR_PSH); in tso_fragment()
2146 TCP_SKB_CB(buff)->tcp_flags = flags; in tso_fragment()
2149 TCP_SKB_CB(buff)->sacked = 0; in tso_fragment()
2153 buff->ip_summed = CHECKSUM_PARTIAL; in tso_fragment()
2180 struct tcp_sock *tp = tcp_sk(sk); in tcp_tso_should_defer() local
2185 if (icsk->icsk_ca_state >= TCP_CA_Recovery) in tcp_tso_should_defer()
2190 * Note that tp->tcp_wstamp_ns can be in the future if we have in tcp_tso_should_defer()
2193 delta = tp->tcp_clock_cache - tp->tcp_wstamp_ns - NSEC_PER_MSEC; in tcp_tso_should_defer()
2197 in_flight = tcp_packets_in_flight(tp); in tcp_tso_should_defer()
2200 BUG_ON(tp->snd_cwnd <= in_flight); in tcp_tso_should_defer()
2202 send_win = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq; in tcp_tso_should_defer()
2205 cong_win = (tp->snd_cwnd - in_flight) * tp->mss_cache; in tcp_tso_should_defer()
2209 /* If a full-sized TSO skb can be sent, do it. */ in tcp_tso_should_defer()
2210 if (limit >= max_segs * tp->mss_cache) in tcp_tso_should_defer()
2214 if ((skb != tcp_write_queue_tail(sk)) && (limit >= skb->len)) in tcp_tso_should_defer()
2217 win_divisor = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_tso_win_divisor); in tcp_tso_should_defer()
2219 u32 chunk = min(tp->snd_wnd, tp->snd_cwnd * tp->mss_cache); in tcp_tso_should_defer()
2233 if (limit > tcp_max_tso_deferred_mss(tp) * tp->mss_cache) in tcp_tso_should_defer()
2241 delta = tp->tcp_clock_cache - head->tstamp; in tcp_tso_should_defer()
2243 if ((s64)(delta - (u64)NSEC_PER_USEC * (tp->srtt_us >> 4)) < 0) in tcp_tso_should_defer()
2248 * 1) We are cwnd-limited in tcp_tso_should_defer()
2249 * 2) We are rwnd-limited in tcp_tso_should_defer()
2253 if (cong_win <= skb->len) { in tcp_tso_should_defer()
2258 if (send_win <= skb->len) { in tcp_tso_should_defer()
2265 if ((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) || in tcp_tso_should_defer()
2266 TCP_SKB_CB(skb)->eor) in tcp_tso_should_defer()
2278 struct tcp_sock *tp = tcp_sk(sk); in tcp_mtu_check_reprobe() local
2283 interval = READ_ONCE(net->ipv4.sysctl_tcp_probe_interval); in tcp_mtu_check_reprobe()
2284 delta = tcp_jiffies32 - icsk->icsk_mtup.probe_timestamp; in tcp_mtu_check_reprobe()
2289 icsk->icsk_mtup.probe_size = 0; in tcp_mtu_check_reprobe()
2290 icsk->icsk_mtup.search_high = tp->rx_opt.mss_clamp + in tcp_mtu_check_reprobe()
2292 icsk->icsk_af_ops->net_header_len; in tcp_mtu_check_reprobe()
2293 icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss); in tcp_mtu_check_reprobe()
2296 icsk->icsk_mtup.probe_timestamp = tcp_jiffies32; in tcp_mtu_check_reprobe()
2306 if (len <= skb->len) in tcp_can_coalesce_send_queue_head()
2309 if (unlikely(TCP_SKB_CB(skb)->eor) || tcp_has_tx_tstamp(skb)) in tcp_can_coalesce_send_queue_head()
2312 len -= skb->len; in tcp_can_coalesce_send_queue_head()
2325 * -1 otherwise
2330 struct tcp_sock *tp = tcp_sk(sk); in tcp_mtu_probe() local
2344 if (likely(!icsk->icsk_mtup.enabled || in tcp_mtu_probe()
2345 icsk->icsk_mtup.probe_size || in tcp_mtu_probe()
2346 inet_csk(sk)->icsk_ca_state != TCP_CA_Open || in tcp_mtu_probe()
2347 tp->snd_cwnd < 11 || in tcp_mtu_probe()
2348 tp->rx_opt.num_sacks || tp->rx_opt.dsack)) in tcp_mtu_probe()
2349 return -1; in tcp_mtu_probe()
2352 * and current mss_clamp. if (search_high - search_low) in tcp_mtu_probe()
2356 probe_size = tcp_mtu_to_mss(sk, (icsk->icsk_mtup.search_high + in tcp_mtu_probe()
2357 icsk->icsk_mtup.search_low) >> 1); in tcp_mtu_probe()
2358 size_needed = probe_size + (tp->reordering + 1) * tp->mss_cache; in tcp_mtu_probe()
2359 interval = icsk->icsk_mtup.search_high - icsk->icsk_mtup.search_low; in tcp_mtu_probe()
2364 if (probe_size > tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_high) || in tcp_mtu_probe()
2365 interval < READ_ONCE(net->ipv4.sysctl_tcp_probe_threshold)) { in tcp_mtu_probe()
2370 return -1; in tcp_mtu_probe()
2374 if (tp->write_seq - tp->snd_nxt < size_needed) in tcp_mtu_probe()
2375 return -1; in tcp_mtu_probe()
2377 if (tp->snd_wnd < size_needed) in tcp_mtu_probe()
2378 return -1; in tcp_mtu_probe()
2379 if (after(tp->snd_nxt + size_needed, tcp_wnd_end(tp))) in tcp_mtu_probe()
2383 if (tcp_packets_in_flight(tp) + 2 > tp->snd_cwnd) { in tcp_mtu_probe()
2384 if (!tcp_packets_in_flight(tp)) in tcp_mtu_probe()
2385 return -1; in tcp_mtu_probe()
2391 return -1; in tcp_mtu_probe()
2396 return -1; in tcp_mtu_probe()
2397 sk_wmem_queued_add(sk, nskb->truesize); in tcp_mtu_probe()
2398 sk_mem_charge(sk, nskb->truesize); in tcp_mtu_probe()
2403 TCP_SKB_CB(nskb)->seq = TCP_SKB_CB(skb)->seq; in tcp_mtu_probe()
2404 TCP_SKB_CB(nskb)->end_seq = TCP_SKB_CB(skb)->seq + probe_size; in tcp_mtu_probe()
2405 TCP_SKB_CB(nskb)->tcp_flags = TCPHDR_ACK; in tcp_mtu_probe()
2406 TCP_SKB_CB(nskb)->sacked = 0; in tcp_mtu_probe()
2407 nskb->csum = 0; in tcp_mtu_probe()
2408 nskb->ip_summed = CHECKSUM_PARTIAL; in tcp_mtu_probe()
2415 copy = min_t(int, skb->len, probe_size - len); in tcp_mtu_probe()
2418 if (skb->len <= copy) { in tcp_mtu_probe()
2421 TCP_SKB_CB(nskb)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags; in tcp_mtu_probe()
2425 TCP_SKB_CB(nskb)->eor = TCP_SKB_CB(skb)->eor; in tcp_mtu_probe()
2430 TCP_SKB_CB(nskb)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags & in tcp_mtu_probe()
2432 if (!skb_shinfo(skb)->nr_frags) { in tcp_mtu_probe()
2438 TCP_SKB_CB(skb)->seq += copy; in tcp_mtu_probe()
2446 tcp_init_tso_segs(nskb, nskb->len); in tcp_mtu_probe()
2449 * be resegmented into mss-sized pieces by tcp_write_xmit(). in tcp_mtu_probe()
2454 tp->snd_cwnd--; in tcp_mtu_probe()
2457 icsk->icsk_mtup.probe_size = tcp_mss_to_mtu(sk, nskb->len); in tcp_mtu_probe()
2458 tp->mtu_probe.probe_seq_start = TCP_SKB_CB(nskb)->seq; in tcp_mtu_probe()
2459 tp->mtu_probe.probe_seq_end = TCP_SKB_CB(nskb)->end_seq; in tcp_mtu_probe()
2464 return -1; in tcp_mtu_probe()
2469 struct tcp_sock *tp = tcp_sk(sk); in tcp_pacing_check() local
2474 if (tp->tcp_wstamp_ns <= tp->tcp_clock_cache) in tcp_pacing_check()
2477 if (!hrtimer_is_queued(&tp->pacing_timer)) { in tcp_pacing_check()
2478 hrtimer_start(&tp->pacing_timer, in tcp_pacing_check()
2479 ns_to_ktime(tp->tcp_wstamp_ns), in tcp_pacing_check()
2490 * - better RTT estimation and ACK scheduling
2491 * - faster recovery
2492 * - high rates
2503 2 * skb->truesize, in tcp_small_queue_check()
2504 sk->sk_pacing_rate >> READ_ONCE(sk->sk_pacing_shift)); in tcp_small_queue_check()
2505 if (sk->sk_pacing_status == SK_PACING_NONE) in tcp_small_queue_check()
2507 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_limit_output_bytes)); in tcp_small_queue_check()
2511 tcp_sk(sk)->tcp_tx_delay) { in tcp_small_queue_check()
2512 u64 extra_bytes = (u64)sk->sk_pacing_rate * tcp_sk(sk)->tcp_tx_delay; in tcp_small_queue_check()
2515 * approximate our needs assuming an ~100% skb->truesize overhead. in tcp_small_queue_check()
2519 extra_bytes >>= (20 - 1); in tcp_small_queue_check()
2522 if (refcount_read(&sk->sk_wmem_alloc) > limit) { in tcp_small_queue_check()
2531 set_bit(TSQ_THROTTLED, &sk->sk_tsq_flags); in tcp_small_queue_check()
2537 if (refcount_read(&sk->sk_wmem_alloc) > limit) in tcp_small_queue_check()
2543 static void tcp_chrono_set(struct tcp_sock *tp, const enum tcp_chrono new) in tcp_chrono_set() argument
2546 enum tcp_chrono old = tp->chrono_type; in tcp_chrono_set()
2549 tp->chrono_stat[old - 1] += now - tp->chrono_start; in tcp_chrono_set()
2550 tp->chrono_start = now; in tcp_chrono_set()
2551 tp->chrono_type = new; in tcp_chrono_set()
2556 struct tcp_sock *tp = tcp_sk(sk); in tcp_chrono_start() local
2563 if (type > tp->chrono_type) in tcp_chrono_start()
2564 tcp_chrono_set(tp, type); in tcp_chrono_start()
2569 struct tcp_sock *tp = tcp_sk(sk); in tcp_chrono_stop() local
2580 tcp_chrono_set(tp, TCP_CHRONO_UNSPEC); in tcp_chrono_stop()
2581 else if (type == tp->chrono_type) in tcp_chrono_stop()
2582 tcp_chrono_set(tp, TCP_CHRONO_BUSY); in tcp_chrono_stop()
2590 * snd_up-64k-mss .. snd_up cannot be large. However, taking into
2602 struct tcp_sock *tp = tcp_sk(sk); in tcp_write_xmit() local
2612 tcp_mstamp_refresh(tp); in tcp_write_xmit()
2627 if (unlikely(tp->repair) && tp->repair_queue == TCP_SEND_QUEUE) { in tcp_write_xmit()
2629 skb->skb_mstamp_ns = tp->tcp_wstamp_ns = tp->tcp_clock_cache; in tcp_write_xmit()
2630 list_move_tail(&skb->tcp_tsorted_anchor, &tp->tsorted_sent_queue); in tcp_write_xmit()
2641 cwnd_quota = tcp_cwnd_test(tp, skb); in tcp_write_xmit()
2650 if (unlikely(!tcp_snd_wnd_test(tp, skb, mss_now))) { in tcp_write_xmit()
2656 if (unlikely(!tcp_nagle_test(tp, skb, mss_now, in tcp_write_xmit()
2668 if (tso_segs > 1 && !tcp_urg_mode(tp)) in tcp_write_xmit()
2675 if (skb->len > limit && in tcp_write_xmit()
2684 * We do not want to send a pure-ack packet and have in tcp_write_xmit()
2687 if (TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq) in tcp_write_xmit()
2699 tcp_minshall_update(tp, mss_now, skb); in tcp_write_xmit()
2711 is_cwnd_limited |= (tcp_packets_in_flight(tp) >= tp->snd_cwnd); in tcp_write_xmit()
2717 tp->prr_out += sent_pkts; in tcp_write_xmit()
2724 return !tp->packets_out && !tcp_write_queue_empty(sk); in tcp_write_xmit()
2730 struct tcp_sock *tp = tcp_sk(sk); in tcp_schedule_loss_probe() local
2737 if (rcu_access_pointer(tp->fastopen_rsk)) in tcp_schedule_loss_probe()
2740 early_retrans = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_early_retrans); in tcp_schedule_loss_probe()
2745 !tp->packets_out || !tcp_is_sack(tp) || in tcp_schedule_loss_probe()
2746 (icsk->icsk_ca_state != TCP_CA_Open && in tcp_schedule_loss_probe()
2747 icsk->icsk_ca_state != TCP_CA_CWR)) in tcp_schedule_loss_probe()
2754 if (tp->srtt_us) { in tcp_schedule_loss_probe()
2755 timeout = usecs_to_jiffies(tp->srtt_us >> 2); in tcp_schedule_loss_probe()
2756 if (tp->packets_out == 1) in tcp_schedule_loss_probe()
2766 jiffies_to_usecs(inet_csk(sk)->icsk_rto) : in tcp_schedule_loss_probe()
2795 struct tcp_sock *tp = tcp_sk(sk); in tcp_send_loss_probe() local
2801 if (tp->tlp_high_seq) in tcp_send_loss_probe()
2804 tp->tlp_retrans = 0; in tcp_send_loss_probe()
2806 if (skb && tcp_snd_wnd_test(tp, skb, mss)) { in tcp_send_loss_probe()
2807 pcount = tp->packets_out; in tcp_send_loss_probe()
2809 if (tp->packets_out > pcount) in tcp_send_loss_probe()
2813 skb = skb_rb_last(&sk->tcp_rtx_queue); in tcp_send_loss_probe()
2815 WARN_ONCE(tp->packets_out, in tcp_send_loss_probe()
2817 tp->packets_out, sk->sk_state, tp->snd_cwnd, mss); in tcp_send_loss_probe()
2818 inet_csk(sk)->icsk_pending = 0; in tcp_send_loss_probe()
2829 if ((pcount > 1) && (skb->len > (pcount - 1) * mss)) { in tcp_send_loss_probe()
2831 (pcount - 1) * mss, mss, in tcp_send_loss_probe()
2843 tp->tlp_retrans = 1; in tcp_send_loss_probe()
2847 tp->tlp_high_seq = tp->snd_nxt; in tcp_send_loss_probe()
2851 inet_csk(sk)->icsk_pending = 0; in tcp_send_loss_probe()
2867 if (unlikely(sk->sk_state == TCP_CLOSE)) in __tcp_push_pending_frames()
2882 BUG_ON(!skb || skb->len < mss_now); in tcp_push_one()
2884 tcp_write_xmit(sk, mss_now, TCP_NAGLE_PUSH, 1, sk->sk_allocation); in tcp_push_one()
2896 * RCV.BUFF - RCV.USER - RCV.WINDOW >= min(1/2 RCV.BUFF, MSS)"
2902 * since header prediction assumes th->window stays fixed.
2904 * Strictly speaking, keeping th->window fixed violates the receiver
2936 * Note, we don't "adjust" for TIMESTAMP or SACK option bytes.
2942 struct tcp_sock *tp = tcp_sk(sk); in __tcp_select_window() local
2947 * fluctuations. --SAW 1998/11/1 in __tcp_select_window()
2949 int mss = icsk->icsk_ack.rcv_mss; in __tcp_select_window()
2957 full_space = min_t(int, tp->window_clamp, allowed_space); in __tcp_select_window()
2965 icsk->icsk_ack.quick = 0; in __tcp_select_window()
2968 tp->rcv_ssthresh = min(tp->rcv_ssthresh, in __tcp_select_window()
2969 4U * tp->advmss); in __tcp_select_window()
2974 free_space = round_down(free_space, 1 << tp->rx_opt.rcv_wscale); in __tcp_select_window()
2977 * of the maximum allowed, try to move to zero-window, else in __tcp_select_window()
2987 if (free_space > tp->rcv_ssthresh) in __tcp_select_window()
2988 free_space = tp->rcv_ssthresh; in __tcp_select_window()
2993 if (tp->rx_opt.rcv_wscale) { in __tcp_select_window()
3000 window = ALIGN(window, (1 << tp->rx_opt.rcv_wscale)); in __tcp_select_window()
3002 window = tp->rcv_wnd; in __tcp_select_window()
3011 if (window <= free_space - mss || window > free_space) in __tcp_select_window()
3029 shinfo->tx_flags |= next_shinfo->tx_flags & SKBTX_ANY_TSTAMP; in tcp_skb_collapse_tstamp()
3030 shinfo->tskey = next_shinfo->tskey; in tcp_skb_collapse_tstamp()
3031 TCP_SKB_CB(skb)->txstamp_ack |= in tcp_skb_collapse_tstamp()
3032 TCP_SKB_CB(next_skb)->txstamp_ack; in tcp_skb_collapse_tstamp()
3039 struct tcp_sock *tp = tcp_sk(sk); in tcp_collapse_retrans() local
3043 next_skb_size = next_skb->len; in tcp_collapse_retrans()
3057 TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(next_skb)->end_seq; in tcp_collapse_retrans()
3060 TCP_SKB_CB(skb)->tcp_flags |= TCP_SKB_CB(next_skb)->tcp_flags; in tcp_collapse_retrans()
3065 TCP_SKB_CB(skb)->sacked |= TCP_SKB_CB(next_skb)->sacked & TCPCB_EVER_RETRANS; in tcp_collapse_retrans()
3066 TCP_SKB_CB(skb)->eor = TCP_SKB_CB(next_skb)->eor; in tcp_collapse_retrans()
3069 tcp_clear_retrans_hints_partial(tp); in tcp_collapse_retrans()
3070 if (next_skb == tp->retransmit_skb_hint) in tcp_collapse_retrans()
3071 tp->retransmit_skb_hint = skb; in tcp_collapse_retrans()
3089 if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED) in tcp_can_collapse()
3101 struct tcp_sock *tp = tcp_sk(sk); in tcp_retrans_try_collapse() local
3105 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_retrans_collapse)) in tcp_retrans_try_collapse()
3107 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN) in tcp_retrans_try_collapse()
3117 space -= skb->len; in tcp_retrans_try_collapse()
3127 if (after(TCP_SKB_CB(skb)->end_seq, tcp_wnd_end(tp))) in tcp_retrans_try_collapse()
3136 * state updates are done by the caller. Returns non-zero if an
3142 struct tcp_sock *tp = tcp_sk(sk); in __tcp_retransmit_skb() local
3148 if (icsk->icsk_mtup.probe_size) in __tcp_retransmit_skb()
3149 icsk->icsk_mtup.probe_size = 0; in __tcp_retransmit_skb()
3154 if (refcount_read(&sk->sk_wmem_alloc) > in __tcp_retransmit_skb()
3155 min_t(u32, sk->sk_wmem_queued + (sk->sk_wmem_queued >> 2), in __tcp_retransmit_skb()
3156 sk->sk_sndbuf)) in __tcp_retransmit_skb()
3157 return -EAGAIN; in __tcp_retransmit_skb()
3160 return -EBUSY; in __tcp_retransmit_skb()
3162 if (before(TCP_SKB_CB(skb)->seq, tp->snd_una)) { in __tcp_retransmit_skb()
3163 if (unlikely(before(TCP_SKB_CB(skb)->end_seq, tp->snd_una))) { in __tcp_retransmit_skb()
3165 return -EINVAL; in __tcp_retransmit_skb()
3167 if (tcp_trim_head(sk, skb, tp->snd_una - TCP_SKB_CB(skb)->seq)) in __tcp_retransmit_skb()
3168 return -ENOMEM; in __tcp_retransmit_skb()
3171 if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk)) in __tcp_retransmit_skb()
3172 return -EHOSTUNREACH; /* Routing failure or similar. */ in __tcp_retransmit_skb()
3175 avail_wnd = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq; in __tcp_retransmit_skb()
3183 if (TCP_SKB_CB(skb)->seq != tp->snd_una) in __tcp_retransmit_skb()
3184 return -EAGAIN; in __tcp_retransmit_skb()
3194 if (skb->len > len) { in __tcp_retransmit_skb()
3197 return -ENOMEM; /* We'll try again later. */ in __tcp_retransmit_skb()
3200 return -ENOMEM; in __tcp_retransmit_skb()
3204 diff -= tcp_skb_pcount(skb); in __tcp_retransmit_skb()
3208 if (skb->len < avail_wnd) in __tcp_retransmit_skb()
3213 if ((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN_ECN) == TCPHDR_SYN_ECN) in __tcp_retransmit_skb()
3219 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN) in __tcp_retransmit_skb()
3221 tp->total_retrans += segs; in __tcp_retransmit_skb()
3222 tp->bytes_retrans += skb->len; in __tcp_retransmit_skb()
3224 /* make sure skb->data is aligned on arches that require it in __tcp_retransmit_skb()
3225 * and check if ack-trimming & collapsing extended the headroom in __tcp_retransmit_skb()
3228 if (unlikely((NET_IP_ALIGN && ((unsigned long)skb->data & 3)) || in __tcp_retransmit_skb()
3235 nskb->dev = NULL; in __tcp_retransmit_skb()
3238 err = -ENOBUFS; in __tcp_retransmit_skb()
3243 tcp_update_skb_after_send(sk, skb, tp->tcp_wstamp_ns); in __tcp_retransmit_skb()
3253 TCP_SKB_CB(skb)->sacked |= TCPCB_EVER_RETRANS; in __tcp_retransmit_skb()
3255 if (BPF_SOCK_OPS_TEST_FLAG(tp, BPF_SOCK_OPS_RETRANS_CB_FLAG)) in __tcp_retransmit_skb()
3257 TCP_SKB_CB(skb)->seq, segs, err); in __tcp_retransmit_skb()
3261 } else if (err != -EBUSY) { in __tcp_retransmit_skb()
3269 struct tcp_sock *tp = tcp_sk(sk); in tcp_retransmit_skb() local
3274 if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) { in tcp_retransmit_skb()
3278 TCP_SKB_CB(skb)->sacked |= TCPCB_RETRANS; in tcp_retransmit_skb()
3279 tp->retrans_out += tcp_skb_pcount(skb); in tcp_retransmit_skb()
3283 if (!tp->retrans_stamp) in tcp_retransmit_skb()
3284 tp->retrans_stamp = tcp_skb_timestamp(skb); in tcp_retransmit_skb()
3286 if (tp->undo_retrans < 0) in tcp_retransmit_skb()
3287 tp->undo_retrans = 0; in tcp_retransmit_skb()
3288 tp->undo_retrans += tcp_skb_pcount(skb); in tcp_retransmit_skb()
3301 struct tcp_sock *tp = tcp_sk(sk); in tcp_xmit_retransmit_queue() local
3306 if (!tp->packets_out) in tcp_xmit_retransmit_queue()
3310 skb = tp->retransmit_skb_hint ?: rtx_head; in tcp_xmit_retransmit_queue()
3321 tp->retransmit_skb_hint = skb; in tcp_xmit_retransmit_queue()
3323 segs = tp->snd_cwnd - tcp_packets_in_flight(tp); in tcp_xmit_retransmit_queue()
3326 sacked = TCP_SKB_CB(skb)->sacked; in tcp_xmit_retransmit_queue()
3332 if (tp->retrans_out >= tp->lost_out) { in tcp_xmit_retransmit_queue()
3340 if (icsk->icsk_ca_state != TCP_CA_Loss) in tcp_xmit_retransmit_queue()
3358 tp->prr_out += tcp_skb_pcount(skb); in tcp_xmit_retransmit_queue()
3361 icsk->icsk_pending != ICSK_TIME_REO_TIMEOUT) in tcp_xmit_retransmit_queue()
3367 inet_csk(sk)->icsk_rto, in tcp_xmit_retransmit_queue()
3382 delta = size - sk->sk_forward_alloc; in sk_forced_mem_schedule()
3386 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM; in sk_forced_mem_schedule()
3389 if (mem_cgroup_sockets_enabled && sk->sk_memcg) in sk_forced_mem_schedule()
3390 mem_cgroup_charge_skmem(sk->sk_memcg, amt); in sk_forced_mem_schedule()
3399 struct tcp_sock *tp = tcp_sk(sk); in tcp_send_fin() local
3408 tskb = skb_rb_last(&sk->tcp_rtx_queue); in tcp_send_fin()
3411 TCP_SKB_CB(tskb)->tcp_flags |= TCPHDR_FIN; in tcp_send_fin()
3412 TCP_SKB_CB(tskb)->end_seq++; in tcp_send_fin()
3413 tp->write_seq++; in tcp_send_fin()
3417 * We need to set tp->snd_nxt to the value it would have in tcp_send_fin()
3419 * does not change tp->snd_nxt. in tcp_send_fin()
3421 WRITE_ONCE(tp->snd_nxt, tp->snd_nxt + 1); in tcp_send_fin()
3425 skb = alloc_skb_fclone(MAX_TCP_HEADER, sk->sk_allocation); in tcp_send_fin()
3429 INIT_LIST_HEAD(&skb->tcp_tsorted_anchor); in tcp_send_fin()
3431 sk_forced_mem_schedule(sk, skb->truesize); in tcp_send_fin()
3433 tcp_init_nondata_skb(skb, tp->write_seq, in tcp_send_fin()
3443 * by RFC 2525, section 2.17. -DaveM
3473 /* Send a crossed SYN-ACK during socket establishment.
3484 if (!skb || !(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { in tcp_send_synack()
3486 return -EFAULT; in tcp_send_synack()
3488 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK)) { in tcp_send_synack()
3496 return -ENOMEM; in tcp_send_synack()
3497 INIT_LIST_HEAD(&nskb->tcp_tsorted_anchor); in tcp_send_synack()
3501 tcp_rbtree_insert(&sk->tcp_rtx_queue, nskb); in tcp_send_synack()
3502 sk_wmem_queued_add(sk, nskb->truesize); in tcp_send_synack()
3503 sk_mem_charge(sk, nskb->truesize); in tcp_send_synack()
3507 TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ACK; in tcp_send_synack()
3514 * tcp_make_synack - Allocate one skb and build a SYNACK packet.
3530 const struct tcp_sock *tp = tcp_sk(sk); in tcp_make_synack() local
3559 * sk->sk_wmem_alloc in an atomic, we can promote to rw. in tcp_make_synack()
3566 mss = tcp_mss_clamp(tp, dst_metric_advmss(dst)); in tcp_make_synack()
3571 if (unlikely(synack_type == TCP_SYNACK_COOKIE && ireq->tstamp_ok)) in tcp_make_synack()
3572 skb->skb_mstamp_ns = cookie_init_timestamp(req, now); in tcp_make_synack()
3576 skb->skb_mstamp_ns = now; in tcp_make_synack()
3577 if (!tcp_rsk(req)->snt_synack) /* Timestamp first SYNACK */ in tcp_make_synack()
3578 tcp_rsk(req)->snt_synack = tcp_skb_timestamp_us(skb); in tcp_make_synack()
3583 md5 = tcp_rsk(req)->af_specific->req_md5_lookup(sk, req_to_sk(req)); in tcp_make_synack()
3585 skb_set_hash(skb, tcp_rsk(req)->txhash, PKT_HASH_TYPE_L4); in tcp_make_synack()
3587 TCP_SKB_CB(skb)->tcp_flags = TCPHDR_SYN | TCPHDR_ACK; in tcp_make_synack()
3595 th = (struct tcphdr *)skb->data; in tcp_make_synack()
3597 th->syn = 1; in tcp_make_synack()
3598 th->ack = 1; in tcp_make_synack()
3600 th->source = htons(ireq->ir_num); in tcp_make_synack()
3601 th->dest = ireq->ir_rmt_port; in tcp_make_synack()
3602 skb->mark = ireq->ir_mark; in tcp_make_synack()
3603 skb->ip_summed = CHECKSUM_PARTIAL; in tcp_make_synack()
3604 th->seq = htonl(tcp_rsk(req)->snt_isn); in tcp_make_synack()
3606 th->ack_seq = htonl(tcp_rsk(req)->rcv_nxt); in tcp_make_synack()
3609 th->window = htons(min(req->rsk_rcv_wnd, 65535U)); in tcp_make_synack()
3611 th->doff = (tcp_header_size >> 2); in tcp_make_synack()
3615 /* Okay, we have all we need - do the md5 hash if needed */ in tcp_make_synack()
3617 tcp_rsk(req)->af_specific->calc_md5_hash(opts.hash_location, in tcp_make_synack()
3625 skb->skb_mstamp_ns = now; in tcp_make_synack()
3626 tcp_add_tx_delay(skb, tp); in tcp_make_synack()
3643 if (likely(ca && bpf_try_module_get(ca, ca->owner))) { in tcp_ca_dst_init()
3644 bpf_module_put(icsk->icsk_ca_ops, icsk->icsk_ca_ops->owner); in tcp_ca_dst_init()
3645 icsk->icsk_ca_dst_locked = tcp_ca_dst_locked(dst); in tcp_ca_dst_init()
3646 icsk->icsk_ca_ops = ca; in tcp_ca_dst_init()
3655 struct tcp_sock *tp = tcp_sk(sk); in tcp_connect_init() local
3662 tp->tcp_header_len = sizeof(struct tcphdr); in tcp_connect_init()
3663 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_timestamps)) in tcp_connect_init()
3664 tp->tcp_header_len += TCPOLEN_TSTAMP_ALIGNED; in tcp_connect_init()
3667 if (tp->af_specific->md5_lookup(sk, sk)) in tcp_connect_init()
3668 tp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED; in tcp_connect_init()
3672 if (tp->rx_opt.user_mss) in tcp_connect_init()
3673 tp->rx_opt.mss_clamp = tp->rx_opt.user_mss; in tcp_connect_init()
3674 tp->max_window = 0; in tcp_connect_init()
3680 if (!tp->window_clamp) in tcp_connect_init()
3681 tp->window_clamp = dst_metric(dst, RTAX_WINDOW); in tcp_connect_init()
3682 tp->advmss = tcp_mss_clamp(tp, dst_metric_advmss(dst)); in tcp_connect_init()
3687 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK && in tcp_connect_init()
3688 (tp->window_clamp > tcp_full_space(sk) || tp->window_clamp == 0)) in tcp_connect_init()
3689 tp->window_clamp = tcp_full_space(sk); in tcp_connect_init()
3696 tp->advmss - (tp->rx_opt.ts_recent_stamp ? tp->tcp_header_len - sizeof(struct tcphdr) : 0), in tcp_connect_init()
3697 &tp->rcv_wnd, in tcp_connect_init()
3698 &tp->window_clamp, in tcp_connect_init()
3699 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_window_scaling), in tcp_connect_init()
3703 tp->rx_opt.rcv_wscale = rcv_wscale; in tcp_connect_init()
3704 tp->rcv_ssthresh = tp->rcv_wnd; in tcp_connect_init()
3706 sk->sk_err = 0; in tcp_connect_init()
3708 tp->snd_wnd = 0; in tcp_connect_init()
3709 tcp_init_wl(tp, 0); in tcp_connect_init()
3711 tp->snd_una = tp->write_seq; in tcp_connect_init()
3712 tp->snd_sml = tp->write_seq; in tcp_connect_init()
3713 tp->snd_up = tp->write_seq; in tcp_connect_init()
3714 WRITE_ONCE(tp->snd_nxt, tp->write_seq); in tcp_connect_init()
3716 if (likely(!tp->repair)) in tcp_connect_init()
3717 tp->rcv_nxt = 0; in tcp_connect_init()
3719 tp->rcv_tstamp = tcp_jiffies32; in tcp_connect_init()
3720 tp->rcv_wup = tp->rcv_nxt; in tcp_connect_init()
3721 WRITE_ONCE(tp->copied_seq, tp->rcv_nxt); in tcp_connect_init()
3723 inet_csk(sk)->icsk_rto = tcp_timeout_init(sk); in tcp_connect_init()
3724 inet_csk(sk)->icsk_retransmits = 0; in tcp_connect_init()
3725 tcp_clear_retrans(tp); in tcp_connect_init()
3730 struct tcp_sock *tp = tcp_sk(sk); in tcp_connect_queue_skb() local
3733 tcb->end_seq += skb->len; in tcp_connect_queue_skb()
3735 sk_wmem_queued_add(sk, skb->truesize); in tcp_connect_queue_skb()
3736 sk_mem_charge(sk, skb->truesize); in tcp_connect_queue_skb()
3737 WRITE_ONCE(tp->write_seq, tcb->end_seq); in tcp_connect_queue_skb()
3738 tp->packets_out += tcp_skb_pcount(skb); in tcp_connect_queue_skb()
3742 * queue a data-only packet after the regular SYN, such that regular SYNs
3743 * are retransmitted on timeouts. Also if the remote SYN-ACK acknowledges
3751 struct tcp_sock *tp = tcp_sk(sk); in tcp_send_syn_data() local
3752 struct tcp_fastopen_request *fo = tp->fastopen_req; in tcp_send_syn_data()
3756 tp->rx_opt.mss_clamp = tp->advmss; /* If MSS is not cached */ in tcp_send_syn_data()
3757 if (!tcp_fastopen_cookie_check(sk, &tp->rx_opt.mss_clamp, &fo->cookie)) in tcp_send_syn_data()
3760 /* MSS for SYN-data is based on cached MSS and bounded by PMTU and in tcp_send_syn_data()
3761 * user-MSS. Reserve maximum option space for middleboxes that add in tcp_send_syn_data()
3764 tp->rx_opt.mss_clamp = tcp_mss_clamp(tp, tp->rx_opt.mss_clamp); in tcp_send_syn_data()
3766 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); in tcp_send_syn_data()
3768 space = __tcp_mtu_to_mss(sk, icsk->icsk_pmtu_cookie) - in tcp_send_syn_data()
3771 space = min_t(size_t, space, fo->size); in tcp_send_syn_data()
3773 /* limit to order-0 allocations */ in tcp_send_syn_data()
3776 syn_data = sk_stream_alloc_skb(sk, space, sk->sk_allocation, false); in tcp_send_syn_data()
3779 syn_data->ip_summed = CHECKSUM_PARTIAL; in tcp_send_syn_data()
3780 memcpy(syn_data->cb, syn->cb, sizeof(syn->cb)); in tcp_send_syn_data()
3783 &fo->data->msg_iter); in tcp_send_syn_data()
3793 skb_zcopy_set(syn_data, fo->uarg, NULL); in tcp_send_syn_data()
3796 if (space == fo->size) in tcp_send_syn_data()
3797 fo->data = NULL; in tcp_send_syn_data()
3798 fo->copied = space; in tcp_send_syn_data()
3801 if (syn_data->len) in tcp_send_syn_data()
3804 err = tcp_transmit_skb(sk, syn_data, 1, sk->sk_allocation); in tcp_send_syn_data()
3806 syn->skb_mstamp_ns = syn_data->skb_mstamp_ns; in tcp_send_syn_data()
3813 TCP_SKB_CB(syn_data)->seq++; in tcp_send_syn_data()
3814 TCP_SKB_CB(syn_data)->tcp_flags = TCPHDR_ACK | TCPHDR_PSH; in tcp_send_syn_data()
3816 tp->syn_data = (fo->copied > 0); in tcp_send_syn_data()
3817 tcp_rbtree_insert(&sk->tcp_rtx_queue, syn_data); in tcp_send_syn_data()
3823 __skb_queue_tail(&sk->sk_write_queue, syn_data); in tcp_send_syn_data()
3824 tp->packets_out -= tcp_skb_pcount(syn_data); in tcp_send_syn_data()
3828 if (fo->cookie.len > 0) in tcp_send_syn_data()
3829 fo->cookie.len = 0; in tcp_send_syn_data()
3830 err = tcp_transmit_skb(sk, syn, 1, sk->sk_allocation); in tcp_send_syn_data()
3832 tp->syn_fastopen = 0; in tcp_send_syn_data()
3834 fo->cookie.len = -1; /* Exclude Fast Open option for SYN retries */ in tcp_send_syn_data()
3841 struct tcp_sock *tp = tcp_sk(sk); in tcp_connect() local
3847 if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk)) in tcp_connect()
3848 return -EHOSTUNREACH; /* Routing failure or similar. */ in tcp_connect()
3852 if (unlikely(tp->repair)) { in tcp_connect()
3857 buff = sk_stream_alloc_skb(sk, 0, sk->sk_allocation, true); in tcp_connect()
3859 return -ENOBUFS; in tcp_connect()
3861 tcp_init_nondata_skb(buff, tp->write_seq++, TCPHDR_SYN); in tcp_connect()
3862 tcp_mstamp_refresh(tp); in tcp_connect()
3863 tp->retrans_stamp = tcp_time_stamp(tp); in tcp_connect()
3866 tcp_rbtree_insert(&sk->tcp_rtx_queue, buff); in tcp_connect()
3869 err = tp->fastopen_req ? tcp_send_syn_data(sk, buff) : in tcp_connect()
3870 tcp_transmit_skb(sk, buff, 1, sk->sk_allocation); in tcp_connect()
3871 if (err == -ECONNREFUSED) in tcp_connect()
3874 /* We change tp->snd_nxt after the tcp_transmit_skb() call in tcp_connect()
3877 WRITE_ONCE(tp->snd_nxt, tp->write_seq); in tcp_connect()
3878 tp->pushed_seq = tp->write_seq; in tcp_connect()
3881 WRITE_ONCE(tp->snd_nxt, TCP_SKB_CB(buff)->seq); in tcp_connect()
3882 tp->pushed_seq = TCP_SKB_CB(buff)->seq; in tcp_connect()
3888 inet_csk(sk)->icsk_rto, TCP_RTO_MAX); in tcp_connect()
3900 int ato = icsk->icsk_ack.ato; in tcp_send_delayed_ack()
3904 const struct tcp_sock *tp = tcp_sk(sk); in tcp_send_delayed_ack() local
3908 (icsk->icsk_ack.pending & ICSK_ACK_PUSHED)) in tcp_send_delayed_ack()
3914 * Do not use inet_csk(sk)->icsk_rto here, use results of rtt measurements in tcp_send_delayed_ack()
3917 if (tp->srtt_us) { in tcp_send_delayed_ack()
3918 int rtt = max_t(int, usecs_to_jiffies(tp->srtt_us >> 3), in tcp_send_delayed_ack()
3928 ato = min_t(u32, ato, inet_csk(sk)->icsk_delack_max); in tcp_send_delayed_ack()
3934 if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) { in tcp_send_delayed_ack()
3936 if (time_before_eq(icsk->icsk_ack.timeout, jiffies + (ato >> 2))) { in tcp_send_delayed_ack()
3941 if (!time_before(timeout, icsk->icsk_ack.timeout)) in tcp_send_delayed_ack()
3942 timeout = icsk->icsk_ack.timeout; in tcp_send_delayed_ack()
3944 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER; in tcp_send_delayed_ack()
3945 icsk->icsk_ack.timeout = timeout; in tcp_send_delayed_ack()
3946 sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout); in tcp_send_delayed_ack()
3955 if (sk->sk_state == TCP_CLOSE) in __tcp_send_ack()
3968 delay = TCP_DELACK_MAX << icsk->icsk_ack.retry; in __tcp_send_ack()
3970 icsk->icsk_ack.retry++; in __tcp_send_ack()
3972 icsk->icsk_ack.ato = TCP_ATO_MIN; in __tcp_send_ack()
3994 __tcp_send_ack(sk, tcp_sk(sk)->rcv_nxt); in tcp_send_ack()
4004 * Current solution: to send TWO zero-length segments in urgent mode:
4006 * out-of-date with SND.UNA-1 to probe window.
4010 struct tcp_sock *tp = tcp_sk(sk); in tcp_xmit_probe_skb() local
4017 return -1; in tcp_xmit_probe_skb()
4025 tcp_init_nondata_skb(skb, tp->snd_una - !urgent, TCPHDR_ACK); in tcp_xmit_probe_skb()
4033 if (sk->sk_state == TCP_ESTABLISHED) { in tcp_send_window_probe()
4034 tcp_sk(sk)->snd_wl1 = tcp_sk(sk)->rcv_nxt - 1; in tcp_send_window_probe()
4043 struct tcp_sock *tp = tcp_sk(sk); in tcp_write_wakeup() local
4046 if (sk->sk_state == TCP_CLOSE) in tcp_write_wakeup()
4047 return -1; in tcp_write_wakeup()
4050 if (skb && before(TCP_SKB_CB(skb)->seq, tcp_wnd_end(tp))) { in tcp_write_wakeup()
4053 unsigned int seg_size = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq; in tcp_write_wakeup()
4055 if (before(tp->pushed_seq, TCP_SKB_CB(skb)->end_seq)) in tcp_write_wakeup()
4056 tp->pushed_seq = TCP_SKB_CB(skb)->end_seq; in tcp_write_wakeup()
4062 if (seg_size < TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq || in tcp_write_wakeup()
4063 skb->len > mss) { in tcp_write_wakeup()
4065 TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH; in tcp_write_wakeup()
4068 return -1; in tcp_write_wakeup()
4072 TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH; in tcp_write_wakeup()
4078 if (between(tp->snd_up, tp->snd_una + 1, tp->snd_una + 0xFFFF)) in tcp_write_wakeup()
4090 struct tcp_sock *tp = tcp_sk(sk); in tcp_send_probe0() local
4097 if (tp->packets_out || tcp_write_queue_empty(sk)) { in tcp_send_probe0()
4099 icsk->icsk_probes_out = 0; in tcp_send_probe0()
4100 icsk->icsk_backoff = 0; in tcp_send_probe0()
4101 icsk->icsk_probes_tstamp = 0; in tcp_send_probe0()
4105 icsk->icsk_probes_out++; in tcp_send_probe0()
4107 if (icsk->icsk_backoff < READ_ONCE(net->ipv4.sysctl_tcp_retries2)) in tcp_send_probe0()
4108 icsk->icsk_backoff++; in tcp_send_probe0()
4123 const struct tcp_request_sock_ops *af_ops = tcp_rsk(req)->af_specific; in tcp_rtx_synack()
4127 tcp_rsk(req)->txhash = net_tx_rndhash(); in tcp_rtx_synack()
4128 res = af_ops->send_synack(sk, NULL, &fl, req, NULL, TCP_SYNACK_NORMAL, in tcp_rtx_synack()
4134 tcp_sk(sk)->total_retrans++; in tcp_rtx_synack()