tcp: add accessors to read/set tp->snd_cwnd
authorEric Dumazet <edumazet@google.com>
Tue, 5 Apr 2022 23:35:38 +0000 (16:35 -0700)
committerJakub Kicinski <kuba@kernel.org>
Wed, 6 Apr 2022 19:05:41 +0000 (12:05 -0700)
We had various bugs over the years with code
breaking the assumption that tp->snd_cwnd is greater
than zero.

Lately, syzbot reported the WARN_ON_ONCE(!tp->prior_cwnd) added
in commit 8b8a321ff72c ("tcp: fix zero cwnd in tcp_cwnd_reduction")
can trigger, and without a repro we would have to spend
considerable time finding the bug.

Instead of complaining too late, we want to catch where
and when tp->snd_cwnd is set to an illegal value.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Suggested-by: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Link: https://lore.kernel.org/r/20220405233538.947344-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
27 files changed:
include/net/tcp.h
include/trace/events/tcp.h
net/core/filter.c
net/ipv4/tcp.c
net/ipv4/tcp_bbr.c
net/ipv4/tcp_bic.c
net/ipv4/tcp_cdg.c
net/ipv4/tcp_cong.c
net/ipv4/tcp_cubic.c
net/ipv4/tcp_dctcp.c
net/ipv4/tcp_highspeed.c
net/ipv4/tcp_htcp.c
net/ipv4/tcp_hybla.c
net/ipv4/tcp_illinois.c
net/ipv4/tcp_input.c
net/ipv4/tcp_ipv4.c
net/ipv4/tcp_lp.c
net/ipv4/tcp_metrics.c
net/ipv4/tcp_nv.c
net/ipv4/tcp_output.c
net/ipv4/tcp_rate.c
net/ipv4/tcp_scalable.c
net/ipv4/tcp_vegas.c
net/ipv4/tcp_veno.c
net/ipv4/tcp_westwood.c
net/ipv4/tcp_yeah.c
net/ipv6/tcp_ipv6.c

index 70ca4a5e330a2002acffd7ddd4f685a758c7fbc4..54aa2f02bb6300b39e889469aa5504d886785ef6 100644 (file)
@@ -1207,9 +1207,20 @@ static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
 
 #define TCP_INFINITE_SSTHRESH  0x7fffffff
 
+static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp)
+{
+       return tp->snd_cwnd;
+}
+
+static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val)
+{
+       WARN_ON_ONCE((int)val <= 0);
+       tp->snd_cwnd = val;
+}
+
 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
 {
-       return tp->snd_cwnd < tp->snd_ssthresh;
+       return tcp_snd_cwnd(tp) < tp->snd_ssthresh;
 }
 
 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
@@ -1235,8 +1246,8 @@ static inline __u32 tcp_current_ssthresh(const struct sock *sk)
                return tp->snd_ssthresh;
        else
                return max(tp->snd_ssthresh,
-                          ((tp->snd_cwnd >> 1) +
-                           (tp->snd_cwnd >> 2)));
+                          ((tcp_snd_cwnd(tp) >> 1) +
+                           (tcp_snd_cwnd(tp) >> 2)));
 }
 
 /* Use define here intentionally to get WARN_ON location shown at the caller */
@@ -1278,7 +1289,7 @@ static inline bool tcp_is_cwnd_limited(const struct sock *sk)
 
        /* If in slow start, ensure cwnd grows to twice what was ACKed. */
        if (tcp_in_slow_start(tp))
-               return tp->snd_cwnd < 2 * tp->max_packets_out;
+               return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out;
 
        return tp->is_cwnd_limited;
 }
index 521059d8dc0a67087527e8a9aa2338e81d86bd37..edcd6369de1029f307dd0fec09e35e9f5723d190 100644 (file)
@@ -279,7 +279,7 @@ TRACE_EVENT(tcp_probe,
                __entry->data_len = skb->len - __tcp_hdrlen(th);
                __entry->snd_nxt = tp->snd_nxt;
                __entry->snd_una = tp->snd_una;
-               __entry->snd_cwnd = tp->snd_cwnd;
+               __entry->snd_cwnd = tcp_snd_cwnd(tp);
                __entry->snd_wnd = tp->snd_wnd;
                __entry->rcv_wnd = tp->rcv_wnd;
                __entry->ssthresh = tcp_current_ssthresh(sk);
index a7044e98765ec5e5b55724527aa61068ccaec20a..29986eda285d06b2ebd344b325a9b6f0de17507a 100644 (file)
@@ -5173,7 +5173,7 @@ static int _bpf_setsockopt(struct sock *sk, int level, int optname,
                                if (val <= 0 || tp->data_segs_out > tp->syn_data)
                                        ret = -EINVAL;
                                else
-                                       tp->snd_cwnd = val;
+                                       tcp_snd_cwnd_set(tp, val);
                                break;
                        case TCP_BPF_SNDCWND_CLAMP:
                                if (val <= 0) {
index cf18fbcbf123a864608a9603bfe215def9e4b70e..e31cf137c6140f76f838b4a0dcddf9f104ad653b 100644 (file)
@@ -429,7 +429,7 @@ void tcp_init_sock(struct sock *sk)
         * algorithms that we must have the following bandaid to talk
         * efficiently to them.  -DaveM
         */
-       tp->snd_cwnd = TCP_INIT_CWND;
+       tcp_snd_cwnd_set(tp, TCP_INIT_CWND);
 
        /* There's a bubble in the pipe until at least the first ACK. */
        tp->app_limited = ~0U;
@@ -3033,7 +3033,7 @@ int tcp_disconnect(struct sock *sk, int flags)
        icsk->icsk_rto_min = TCP_RTO_MIN;
        icsk->icsk_delack_max = TCP_DELACK_MAX;
        tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
-       tp->snd_cwnd = TCP_INIT_CWND;
+       tcp_snd_cwnd_set(tp, TCP_INIT_CWND);
        tp->snd_cwnd_cnt = 0;
        tp->window_clamp = 0;
        tp->delivered = 0;
@@ -3744,7 +3744,7 @@ void tcp_get_info(struct sock *sk, struct tcp_info *info)
        info->tcpi_max_pacing_rate = rate64;
 
        info->tcpi_reordering = tp->reordering;
-       info->tcpi_snd_cwnd = tp->snd_cwnd;
+       info->tcpi_snd_cwnd = tcp_snd_cwnd(tp);
 
        if (info->tcpi_state == TCP_LISTEN) {
                /* listeners aliased fields :
@@ -3915,7 +3915,7 @@ struct sk_buff *tcp_get_timestamping_opt_stats(const struct sock *sk,
        rate64 = tcp_compute_delivery_rate(tp);
        nla_put_u64_64bit(stats, TCP_NLA_DELIVERY_RATE, rate64, TCP_NLA_PAD);
 
-       nla_put_u32(stats, TCP_NLA_SND_CWND, tp->snd_cwnd);
+       nla_put_u32(stats, TCP_NLA_SND_CWND, tcp_snd_cwnd(tp));
        nla_put_u32(stats, TCP_NLA_REORDERING, tp->reordering);
        nla_put_u32(stats, TCP_NLA_MIN_RTT, tcp_min_rtt(tp));
 
index 02e8626ccb278705b8c611332e47fe5b7c732cf5..c7d30a3bbd81d27e16e800ec446569b93a4123ba 100644 (file)
@@ -276,7 +276,7 @@ static void bbr_init_pacing_rate_from_rtt(struct sock *sk)
        } else {                         /* no RTT sample yet */
                rtt_us = USEC_PER_MSEC;  /* use nominal default RTT */
        }
-       bw = (u64)tp->snd_cwnd * BW_UNIT;
+       bw = (u64)tcp_snd_cwnd(tp) * BW_UNIT;
        do_div(bw, rtt_us);
        sk->sk_pacing_rate = bbr_bw_to_pacing_rate(sk, bw, bbr_high_gain);
 }
@@ -323,9 +323,9 @@ static void bbr_save_cwnd(struct sock *sk)
        struct bbr *bbr = inet_csk_ca(sk);
 
        if (bbr->prev_ca_state < TCP_CA_Recovery && bbr->mode != BBR_PROBE_RTT)
-               bbr->prior_cwnd = tp->snd_cwnd;  /* this cwnd is good enough */
+               bbr->prior_cwnd = tcp_snd_cwnd(tp);  /* this cwnd is good enough */
        else  /* loss recovery or BBR_PROBE_RTT have temporarily cut cwnd */
-               bbr->prior_cwnd = max(bbr->prior_cwnd, tp->snd_cwnd);
+               bbr->prior_cwnd = max(bbr->prior_cwnd, tcp_snd_cwnd(tp));
 }
 
 static void bbr_cwnd_event(struct sock *sk, enum tcp_ca_event event)
@@ -482,7 +482,7 @@ static bool bbr_set_cwnd_to_recover_or_restore(
        struct tcp_sock *tp = tcp_sk(sk);
        struct bbr *bbr = inet_csk_ca(sk);
        u8 prev_state = bbr->prev_ca_state, state = inet_csk(sk)->icsk_ca_state;
-       u32 cwnd = tp->snd_cwnd;
+       u32 cwnd = tcp_snd_cwnd(tp);
 
        /* An ACK for P pkts should release at most 2*P packets. We do this
         * in two steps. First, here we deduct the number of lost packets.
@@ -520,7 +520,7 @@ static void bbr_set_cwnd(struct sock *sk, const struct rate_sample *rs,
 {
        struct tcp_sock *tp = tcp_sk(sk);
        struct bbr *bbr = inet_csk_ca(sk);
-       u32 cwnd = tp->snd_cwnd, target_cwnd = 0;
+       u32 cwnd = tcp_snd_cwnd(tp), target_cwnd = 0;
 
        if (!acked)
                goto done;  /* no packet fully ACKed; just apply caps */
@@ -544,9 +544,9 @@ static void bbr_set_cwnd(struct sock *sk, const struct rate_sample *rs,
        cwnd = max(cwnd, bbr_cwnd_min_target);
 
 done:
-       tp->snd_cwnd = min(cwnd, tp->snd_cwnd_clamp);   /* apply global cap */
+       tcp_snd_cwnd_set(tp, min(cwnd, tp->snd_cwnd_clamp));    /* apply global cap */
        if (bbr->mode == BBR_PROBE_RTT)  /* drain queue, refresh min_rtt */
-               tp->snd_cwnd = min(tp->snd_cwnd, bbr_cwnd_min_target);
+               tcp_snd_cwnd_set(tp, min(tcp_snd_cwnd(tp), bbr_cwnd_min_target));
 }
 
 /* End cycle phase if it's time and/or we hit the phase's in-flight target. */
@@ -856,7 +856,7 @@ static void bbr_update_ack_aggregation(struct sock *sk,
        bbr->ack_epoch_acked = min_t(u32, 0xFFFFF,
                                     bbr->ack_epoch_acked + rs->acked_sacked);
        extra_acked = bbr->ack_epoch_acked - expected_acked;
-       extra_acked = min(extra_acked, tp->snd_cwnd);
+       extra_acked = min(extra_acked, tcp_snd_cwnd(tp));
        if (extra_acked > bbr->extra_acked[bbr->extra_acked_win_idx])
                bbr->extra_acked[bbr->extra_acked_win_idx] = extra_acked;
 }
@@ -914,7 +914,7 @@ static void bbr_check_probe_rtt_done(struct sock *sk)
                return;
 
        bbr->min_rtt_stamp = tcp_jiffies32;  /* wait a while until PROBE_RTT */
-       tp->snd_cwnd = max(tp->snd_cwnd, bbr->prior_cwnd);
+       tcp_snd_cwnd_set(tp, max(tcp_snd_cwnd(tp), bbr->prior_cwnd));
        bbr_reset_mode(sk);
 }
 
@@ -1093,7 +1093,7 @@ static u32 bbr_undo_cwnd(struct sock *sk)
        bbr->full_bw = 0;   /* spurious slow-down; reset full pipe detection */
        bbr->full_bw_cnt = 0;
        bbr_reset_lt_bw_sampling(sk);
-       return tcp_sk(sk)->snd_cwnd;
+       return tcp_snd_cwnd(tcp_sk(sk));
 }
 
 /* Entering loss recovery, so save cwnd for when we exit or undo recovery. */
index f5f588b1f6e9dcf2b37c465d46518af0b673af55..58358bf92e1b8ac43c07789dac9f6031fa2e03dd 100644 (file)
@@ -150,7 +150,7 @@ static void bictcp_cong_avoid(struct sock *sk, u32 ack, u32 acked)
                if (!acked)
                        return;
        }
-       bictcp_update(ca, tp->snd_cwnd);
+       bictcp_update(ca, tcp_snd_cwnd(tp));
        tcp_cong_avoid_ai(tp, ca->cnt, acked);
 }
 
@@ -166,16 +166,16 @@ static u32 bictcp_recalc_ssthresh(struct sock *sk)
        ca->epoch_start = 0;    /* end of epoch */
 
        /* Wmax and fast convergence */
-       if (tp->snd_cwnd < ca->last_max_cwnd && fast_convergence)
-               ca->last_max_cwnd = (tp->snd_cwnd * (BICTCP_BETA_SCALE + beta))
+       if (tcp_snd_cwnd(tp) < ca->last_max_cwnd && fast_convergence)
+               ca->last_max_cwnd = (tcp_snd_cwnd(tp) * (BICTCP_BETA_SCALE + beta))
                        / (2 * BICTCP_BETA_SCALE);
        else
-               ca->last_max_cwnd = tp->snd_cwnd;
+               ca->last_max_cwnd = tcp_snd_cwnd(tp);
 
-       if (tp->snd_cwnd <= low_window)
-               return max(tp->snd_cwnd >> 1U, 2U);
+       if (tcp_snd_cwnd(tp) <= low_window)
+               return max(tcp_snd_cwnd(tp) >> 1U, 2U);
        else
-               return max((tp->snd_cwnd * beta) / BICTCP_BETA_SCALE, 2U);
+               return max((tcp_snd_cwnd(tp) * beta) / BICTCP_BETA_SCALE, 2U);
 }
 
 static void bictcp_state(struct sock *sk, u8 new_state)
index 709d23801823935b4eb2d3121b6e7ca1bdb08972..ddc7ba0554bddaa5df2fffdb61faba1f3cfbde5c 100644 (file)
@@ -161,8 +161,8 @@ static void tcp_cdg_hystart_update(struct sock *sk)
                                              LINUX_MIB_TCPHYSTARTTRAINDETECT);
                                NET_ADD_STATS(sock_net(sk),
                                              LINUX_MIB_TCPHYSTARTTRAINCWND,
-                                             tp->snd_cwnd);
-                               tp->snd_ssthresh = tp->snd_cwnd;
+                                             tcp_snd_cwnd(tp));
+                               tp->snd_ssthresh = tcp_snd_cwnd(tp);
                                return;
                        }
                }
@@ -180,8 +180,8 @@ static void tcp_cdg_hystart_update(struct sock *sk)
                                              LINUX_MIB_TCPHYSTARTDELAYDETECT);
                                NET_ADD_STATS(sock_net(sk),
                                              LINUX_MIB_TCPHYSTARTDELAYCWND,
-                                             tp->snd_cwnd);
-                               tp->snd_ssthresh = tp->snd_cwnd;
+                                             tcp_snd_cwnd(tp));
+                               tp->snd_ssthresh = tcp_snd_cwnd(tp);
                        }
                }
        }
@@ -252,7 +252,7 @@ static bool tcp_cdg_backoff(struct sock *sk, u32 grad)
                        return false;
        }
 
-       ca->shadow_wnd = max(ca->shadow_wnd, tp->snd_cwnd);
+       ca->shadow_wnd = max(ca->shadow_wnd, tcp_snd_cwnd(tp));
        ca->state = CDG_BACKOFF;
        tcp_enter_cwr(sk);
        return true;
@@ -285,14 +285,14 @@ static void tcp_cdg_cong_avoid(struct sock *sk, u32 ack, u32 acked)
        }
 
        if (!tcp_is_cwnd_limited(sk)) {
-               ca->shadow_wnd = min(ca->shadow_wnd, tp->snd_cwnd);
+               ca->shadow_wnd = min(ca->shadow_wnd, tcp_snd_cwnd(tp));
                return;
        }
 
-       prior_snd_cwnd = tp->snd_cwnd;
+       prior_snd_cwnd = tcp_snd_cwnd(tp);
        tcp_reno_cong_avoid(sk, ack, acked);
 
-       incr = tp->snd_cwnd - prior_snd_cwnd;
+       incr = tcp_snd_cwnd(tp) - prior_snd_cwnd;
        ca->shadow_wnd = max(ca->shadow_wnd, ca->shadow_wnd + incr);
 }
 
@@ -331,15 +331,15 @@ static u32 tcp_cdg_ssthresh(struct sock *sk)
        struct tcp_sock *tp = tcp_sk(sk);
 
        if (ca->state == CDG_BACKOFF)
-               return max(2U, (tp->snd_cwnd * min(1024U, backoff_beta)) >> 10);
+               return max(2U, (tcp_snd_cwnd(tp) * min(1024U, backoff_beta)) >> 10);
 
        if (ca->state == CDG_NONFULL && use_tolerance)
-               return tp->snd_cwnd;
+               return tcp_snd_cwnd(tp);
 
-       ca->shadow_wnd = min(ca->shadow_wnd >> 1, tp->snd_cwnd);
+       ca->shadow_wnd = min(ca->shadow_wnd >> 1, tcp_snd_cwnd(tp));
        if (use_shadow)
-               return max3(2U, ca->shadow_wnd, tp->snd_cwnd >> 1);
-       return max(2U, tp->snd_cwnd >> 1);
+               return max3(2U, ca->shadow_wnd, tcp_snd_cwnd(tp) >> 1);
+       return max(2U, tcp_snd_cwnd(tp) >> 1);
 }
 
 static void tcp_cdg_cwnd_event(struct sock *sk, const enum tcp_ca_event ev)
@@ -357,7 +357,7 @@ static void tcp_cdg_cwnd_event(struct sock *sk, const enum tcp_ca_event ev)
 
                ca->gradients = gradients;
                ca->rtt_seq = tp->snd_nxt;
-               ca->shadow_wnd = tp->snd_cwnd;
+               ca->shadow_wnd = tcp_snd_cwnd(tp);
                break;
        case CA_EVENT_COMPLETE_CWR:
                ca->state = CDG_UNKNOWN;
@@ -380,7 +380,7 @@ static void tcp_cdg_init(struct sock *sk)
                ca->gradients = kcalloc(window, sizeof(ca->gradients[0]),
                                        GFP_NOWAIT | __GFP_NOWARN);
        ca->rtt_seq = tp->snd_nxt;
-       ca->shadow_wnd = tp->snd_cwnd;
+       ca->shadow_wnd = tcp_snd_cwnd(tp);
 }
 
 static void tcp_cdg_release(struct sock *sk)
index dc95572163df322c97bd7181c420c1ceb6355eec..d854bcfb990607d88e241bfaab8192119e4488f2 100644 (file)
@@ -393,10 +393,10 @@ int tcp_set_congestion_control(struct sock *sk, const char *name, bool load,
  */
 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked)
 {
-       u32 cwnd = min(tp->snd_cwnd + acked, tp->snd_ssthresh);
+       u32 cwnd = min(tcp_snd_cwnd(tp) + acked, tp->snd_ssthresh);
 
-       acked -= cwnd - tp->snd_cwnd;
-       tp->snd_cwnd = min(cwnd, tp->snd_cwnd_clamp);
+       acked -= cwnd - tcp_snd_cwnd(tp);
+       tcp_snd_cwnd_set(tp, min(cwnd, tp->snd_cwnd_clamp));
 
        return acked;
 }
@@ -410,7 +410,7 @@ void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked)
        /* If credits accumulated at a higher w, apply them gently now. */
        if (tp->snd_cwnd_cnt >= w) {
                tp->snd_cwnd_cnt = 0;
-               tp->snd_cwnd++;
+               tcp_snd_cwnd_set(tp, tcp_snd_cwnd(tp) + 1);
        }
 
        tp->snd_cwnd_cnt += acked;
@@ -418,9 +418,9 @@ void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked)
                u32 delta = tp->snd_cwnd_cnt / w;
 
                tp->snd_cwnd_cnt -= delta * w;
-               tp->snd_cwnd += delta;
+               tcp_snd_cwnd_set(tp, tcp_snd_cwnd(tp) + delta);
        }
-       tp->snd_cwnd = min(tp->snd_cwnd, tp->snd_cwnd_clamp);
+       tcp_snd_cwnd_set(tp, min(tcp_snd_cwnd(tp), tp->snd_cwnd_clamp));
 }
 EXPORT_SYMBOL_GPL(tcp_cong_avoid_ai);
 
@@ -445,7 +445,7 @@ void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked)
                        return;
        }
        /* In dangerous area, increase slowly. */
-       tcp_cong_avoid_ai(tp, tp->snd_cwnd, acked);
+       tcp_cong_avoid_ai(tp, tcp_snd_cwnd(tp), acked);
 }
 EXPORT_SYMBOL_GPL(tcp_reno_cong_avoid);
 
@@ -454,7 +454,7 @@ u32 tcp_reno_ssthresh(struct sock *sk)
 {
        const struct tcp_sock *tp = tcp_sk(sk);
 
-       return max(tp->snd_cwnd >> 1U, 2U);
+       return max(tcp_snd_cwnd(tp) >> 1U, 2U);
 }
 EXPORT_SYMBOL_GPL(tcp_reno_ssthresh);
 
@@ -462,7 +462,7 @@ u32 tcp_reno_undo_cwnd(struct sock *sk)
 {
        const struct tcp_sock *tp = tcp_sk(sk);
 
-       return max(tp->snd_cwnd, tp->prior_cwnd);
+       return max(tcp_snd_cwnd(tp), tp->prior_cwnd);
 }
 EXPORT_SYMBOL_GPL(tcp_reno_undo_cwnd);
 
index 24d562dd62254d6e50dd08236f8967400d81e1ea..b0918839bee7cf0264ec3bbcdfc1417daa86d197 100644 (file)
@@ -334,7 +334,7 @@ static void cubictcp_cong_avoid(struct sock *sk, u32 ack, u32 acked)
                if (!acked)
                        return;
        }
-       bictcp_update(ca, tp->snd_cwnd, acked);
+       bictcp_update(ca, tcp_snd_cwnd(tp), acked);
        tcp_cong_avoid_ai(tp, ca->cnt, acked);
 }
 
@@ -346,13 +346,13 @@ static u32 cubictcp_recalc_ssthresh(struct sock *sk)
        ca->epoch_start = 0;    /* end of epoch */
 
        /* Wmax and fast convergence */
-       if (tp->snd_cwnd < ca->last_max_cwnd && fast_convergence)
-               ca->last_max_cwnd = (tp->snd_cwnd * (BICTCP_BETA_SCALE + beta))
+       if (tcp_snd_cwnd(tp) < ca->last_max_cwnd && fast_convergence)
+               ca->last_max_cwnd = (tcp_snd_cwnd(tp) * (BICTCP_BETA_SCALE + beta))
                        / (2 * BICTCP_BETA_SCALE);
        else
-               ca->last_max_cwnd = tp->snd_cwnd;
+               ca->last_max_cwnd = tcp_snd_cwnd(tp);
 
-       return max((tp->snd_cwnd * beta) / BICTCP_BETA_SCALE, 2U);
+       return max((tcp_snd_cwnd(tp) * beta) / BICTCP_BETA_SCALE, 2U);
 }
 
 static void cubictcp_state(struct sock *sk, u8 new_state)
@@ -413,13 +413,13 @@ static void hystart_update(struct sock *sk, u32 delay)
                                ca->found = 1;
                                pr_debug("hystart_ack_train (%u > %u) delay_min %u (+ ack_delay %u) cwnd %u\n",
                                         now - ca->round_start, threshold,
-                                        ca->delay_min, hystart_ack_delay(sk), tp->snd_cwnd);
+                                        ca->delay_min, hystart_ack_delay(sk), tcp_snd_cwnd(tp));
                                NET_INC_STATS(sock_net(sk),
                                              LINUX_MIB_TCPHYSTARTTRAINDETECT);
                                NET_ADD_STATS(sock_net(sk),
                                              LINUX_MIB_TCPHYSTARTTRAINCWND,
-                                             tp->snd_cwnd);
-                               tp->snd_ssthresh = tp->snd_cwnd;
+                                             tcp_snd_cwnd(tp));
+                               tp->snd_ssthresh = tcp_snd_cwnd(tp);
                        }
                }
        }
@@ -438,8 +438,8 @@ static void hystart_update(struct sock *sk, u32 delay)
                                              LINUX_MIB_TCPHYSTARTDELAYDETECT);
                                NET_ADD_STATS(sock_net(sk),
                                              LINUX_MIB_TCPHYSTARTDELAYCWND,
-                                             tp->snd_cwnd);
-                               tp->snd_ssthresh = tp->snd_cwnd;
+                                             tcp_snd_cwnd(tp));
+                               tp->snd_ssthresh = tcp_snd_cwnd(tp);
                        }
                }
        }
@@ -469,7 +469,7 @@ static void cubictcp_acked(struct sock *sk, const struct ack_sample *sample)
 
        /* hystart triggers when cwnd is larger than some threshold */
        if (!ca->found && tcp_in_slow_start(tp) && hystart &&
-           tp->snd_cwnd >= hystart_low_window)
+           tcp_snd_cwnd(tp) >= hystart_low_window)
                hystart_update(sk, delay);
 }
 
index 1943a6630341c67f98a43dc1052321fe9edebfaa..ab034a4e9324a1116c40185811e7597ff8dea06d 100644 (file)
@@ -106,8 +106,8 @@ static u32 dctcp_ssthresh(struct sock *sk)
        struct dctcp *ca = inet_csk_ca(sk);
        struct tcp_sock *tp = tcp_sk(sk);
 
-       ca->loss_cwnd = tp->snd_cwnd;
-       return max(tp->snd_cwnd - ((tp->snd_cwnd * ca->dctcp_alpha) >> 11U), 2U);
+       ca->loss_cwnd = tcp_snd_cwnd(tp);
+       return max(tcp_snd_cwnd(tp) - ((tcp_snd_cwnd(tp) * ca->dctcp_alpha) >> 11U), 2U);
 }
 
 static void dctcp_update_alpha(struct sock *sk, u32 flags)
@@ -148,8 +148,8 @@ static void dctcp_react_to_loss(struct sock *sk)
        struct dctcp *ca = inet_csk_ca(sk);
        struct tcp_sock *tp = tcp_sk(sk);
 
-       ca->loss_cwnd = tp->snd_cwnd;
-       tp->snd_ssthresh = max(tp->snd_cwnd >> 1U, 2U);
+       ca->loss_cwnd = tcp_snd_cwnd(tp);
+       tp->snd_ssthresh = max(tcp_snd_cwnd(tp) >> 1U, 2U);
 }
 
 static void dctcp_state(struct sock *sk, u8 new_state)
@@ -211,8 +211,9 @@ static size_t dctcp_get_info(struct sock *sk, u32 ext, int *attr,
 static u32 dctcp_cwnd_undo(struct sock *sk)
 {
        const struct dctcp *ca = inet_csk_ca(sk);
+       struct tcp_sock *tp = tcp_sk(sk);
 
-       return max(tcp_sk(sk)->snd_cwnd, ca->loss_cwnd);
+       return max(tcp_snd_cwnd(tp), ca->loss_cwnd);
 }
 
 static struct tcp_congestion_ops dctcp __read_mostly = {
index 349069d6cd0aabbdb1e4c10c95ec0cf887b7b56b..c6de5ce79ad3c582618f3ea62310bfb257b1350f 100644 (file)
@@ -127,22 +127,22 @@ static void hstcp_cong_avoid(struct sock *sk, u32 ack, u32 acked)
                 *     snd_cwnd <=
                 *     hstcp_aimd_vals[ca->ai].cwnd
                 */
-               if (tp->snd_cwnd > hstcp_aimd_vals[ca->ai].cwnd) {
-                       while (tp->snd_cwnd > hstcp_aimd_vals[ca->ai].cwnd &&
+               if (tcp_snd_cwnd(tp) > hstcp_aimd_vals[ca->ai].cwnd) {
+                       while (tcp_snd_cwnd(tp) > hstcp_aimd_vals[ca->ai].cwnd &&
                               ca->ai < HSTCP_AIMD_MAX - 1)
                                ca->ai++;
-               } else if (ca->ai && tp->snd_cwnd <= hstcp_aimd_vals[ca->ai-1].cwnd) {
-                       while (ca->ai && tp->snd_cwnd <= hstcp_aimd_vals[ca->ai-1].cwnd)
+               } else if (ca->ai && tcp_snd_cwnd(tp) <= hstcp_aimd_vals[ca->ai-1].cwnd) {
+                       while (ca->ai && tcp_snd_cwnd(tp) <= hstcp_aimd_vals[ca->ai-1].cwnd)
                                ca->ai--;
                }
 
                /* Do additive increase */
-               if (tp->snd_cwnd < tp->snd_cwnd_clamp) {
+               if (tcp_snd_cwnd(tp) < tp->snd_cwnd_clamp) {
                        /* cwnd = cwnd + a(w) / cwnd */
                        tp->snd_cwnd_cnt += ca->ai + 1;
-                       if (tp->snd_cwnd_cnt >= tp->snd_cwnd) {
-                               tp->snd_cwnd_cnt -= tp->snd_cwnd;
-                               tp->snd_cwnd++;
+                       if (tp->snd_cwnd_cnt >= tcp_snd_cwnd(tp)) {
+                               tp->snd_cwnd_cnt -= tcp_snd_cwnd(tp);
+                               tcp_snd_cwnd_set(tp, tcp_snd_cwnd(tp) + 1);
                        }
                }
        }
@@ -154,7 +154,7 @@ static u32 hstcp_ssthresh(struct sock *sk)
        struct hstcp *ca = inet_csk_ca(sk);
 
        /* Do multiplicative decrease */
-       return max(tp->snd_cwnd - ((tp->snd_cwnd * hstcp_aimd_vals[ca->ai].md) >> 8), 2U);
+       return max(tcp_snd_cwnd(tp) - ((tcp_snd_cwnd(tp) * hstcp_aimd_vals[ca->ai].md) >> 8), 2U);
 }
 
 static struct tcp_congestion_ops tcp_highspeed __read_mostly = {
index 55adcfcf96feaca1032198c1931e569ee80b3668..52b1f2665dfae81f63440c63f088232bd247fb7b 100644 (file)
@@ -124,7 +124,7 @@ static void measure_achieved_throughput(struct sock *sk,
 
        ca->packetcount += sample->pkts_acked;
 
-       if (ca->packetcount >= tp->snd_cwnd - (ca->alpha >> 7 ? : 1) &&
+       if (ca->packetcount >= tcp_snd_cwnd(tp) - (ca->alpha >> 7 ? : 1) &&
            now - ca->lasttime >= ca->minRTT &&
            ca->minRTT > 0) {
                __u32 cur_Bi = ca->packetcount * HZ / (now - ca->lasttime);
@@ -225,7 +225,7 @@ static u32 htcp_recalc_ssthresh(struct sock *sk)
        const struct htcp *ca = inet_csk_ca(sk);
 
        htcp_param_update(sk);
-       return max((tp->snd_cwnd * ca->beta) >> 7, 2U);
+       return max((tcp_snd_cwnd(tp) * ca->beta) >> 7, 2U);
 }
 
 static void htcp_cong_avoid(struct sock *sk, u32 ack, u32 acked)
@@ -242,9 +242,9 @@ static void htcp_cong_avoid(struct sock *sk, u32 ack, u32 acked)
                /* In dangerous area, increase slowly.
                 * In theory this is tp->snd_cwnd += alpha / tp->snd_cwnd
                 */
-               if ((tp->snd_cwnd_cnt * ca->alpha)>>7 >= tp->snd_cwnd) {
-                       if (tp->snd_cwnd < tp->snd_cwnd_clamp)
-                               tp->snd_cwnd++;
+               if ((tp->snd_cwnd_cnt * ca->alpha)>>7 >= tcp_snd_cwnd(tp)) {
+                       if (tcp_snd_cwnd(tp) < tp->snd_cwnd_clamp)
+                               tcp_snd_cwnd_set(tp, tcp_snd_cwnd(tp) + 1);
                        tp->snd_cwnd_cnt = 0;
                        htcp_alpha_update(ca);
                } else
index be39327e04e6c33cc7bc1ec26f3745c1c59cc70a..abd7d91807e542dee0a32913274de958b731e013 100644 (file)
@@ -54,7 +54,7 @@ static void hybla_init(struct sock *sk)
        ca->rho2_7ls = 0;
        ca->snd_cwnd_cents = 0;
        ca->hybla_en = true;
-       tp->snd_cwnd = 2;
+       tcp_snd_cwnd_set(tp, 2);
        tp->snd_cwnd_clamp = 65535;
 
        /* 1st Rho measurement based on initial srtt */
@@ -62,7 +62,7 @@ static void hybla_init(struct sock *sk)
 
        /* set minimum rtt as this is the 1st ever seen */
        ca->minrtt_us = tp->srtt_us;
-       tp->snd_cwnd = ca->rho;
+       tcp_snd_cwnd_set(tp, ca->rho);
 }
 
 static void hybla_state(struct sock *sk, u8 ca_state)
@@ -137,31 +137,31 @@ static void hybla_cong_avoid(struct sock *sk, u32 ack, u32 acked)
                 * as long as increment is estimated as (rho<<7)/window
                 * it already is <<7 and we can easily count its fractions.
                 */
-               increment = ca->rho2_7ls / tp->snd_cwnd;
+               increment = ca->rho2_7ls / tcp_snd_cwnd(tp);
                if (increment < 128)
                        tp->snd_cwnd_cnt++;
        }
 
        odd = increment % 128;
-       tp->snd_cwnd += increment >> 7;
+       tcp_snd_cwnd_set(tp, tcp_snd_cwnd(tp) + (increment >> 7));
        ca->snd_cwnd_cents += odd;
 
        /* check when fractions goes >=128 and increase cwnd by 1. */
        while (ca->snd_cwnd_cents >= 128) {
-               tp->snd_cwnd++;
+               tcp_snd_cwnd_set(tp, tcp_snd_cwnd(tp) + 1);
                ca->snd_cwnd_cents -= 128;
                tp->snd_cwnd_cnt = 0;
        }
        /* check when cwnd has not been incremented for a while */
-       if (increment == 0 && odd == 0 && tp->snd_cwnd_cnt >= tp->snd_cwnd) {
-               tp->snd_cwnd++;
+       if (increment == 0 && odd == 0 && tp->snd_cwnd_cnt >= tcp_snd_cwnd(tp)) {
+               tcp_snd_cwnd_set(tp, tcp_snd_cwnd(tp) + 1);
                tp->snd_cwnd_cnt = 0;
        }
        /* clamp down slowstart cwnd to ssthresh value. */
        if (is_slowstart)
-               tp->snd_cwnd = min(tp->snd_cwnd, tp->snd_ssthresh);
+               tcp_snd_cwnd_set(tp, min(tcp_snd_cwnd(tp), tp->snd_ssthresh));
 
-       tp->snd_cwnd = min_t(u32, tp->snd_cwnd, tp->snd_cwnd_clamp);
+       tcp_snd_cwnd_set(tp, min(tcp_snd_cwnd(tp), tp->snd_cwnd_clamp));
 }
 
 static struct tcp_congestion_ops tcp_hybla __read_mostly = {
index 00e54873213e87566eda45d8f5fbbe4c69f78262..c0c81a2c77faebc4fa55ae387adf738a20406827 100644 (file)
@@ -224,7 +224,7 @@ static void update_params(struct sock *sk)
        struct tcp_sock *tp = tcp_sk(sk);
        struct illinois *ca = inet_csk_ca(sk);
 
-       if (tp->snd_cwnd < win_thresh) {
+       if (tcp_snd_cwnd(tp) < win_thresh) {
                ca->alpha = ALPHA_BASE;
                ca->beta = BETA_BASE;
        } else if (ca->cnt_rtt > 0) {
@@ -284,9 +284,9 @@ static void tcp_illinois_cong_avoid(struct sock *sk, u32 ack, u32 acked)
                 * tp->snd_cwnd += alpha/tp->snd_cwnd
                */
                delta = (tp->snd_cwnd_cnt * ca->alpha) >> ALPHA_SHIFT;
-               if (delta >= tp->snd_cwnd) {
-                       tp->snd_cwnd = min(tp->snd_cwnd + delta / tp->snd_cwnd,
-                                          (u32)tp->snd_cwnd_clamp);
+               if (delta >= tcp_snd_cwnd(tp)) {
+                       tcp_snd_cwnd_set(tp, min(tcp_snd_cwnd(tp) + delta / tcp_snd_cwnd(tp),
+                                                (u32)tp->snd_cwnd_clamp));
                        tp->snd_cwnd_cnt = 0;
                }
        }
@@ -296,9 +296,11 @@ static u32 tcp_illinois_ssthresh(struct sock *sk)
 {
        struct tcp_sock *tp = tcp_sk(sk);
        struct illinois *ca = inet_csk_ca(sk);
+       u32 decr;
 
        /* Multiplicative decrease */
-       return max(tp->snd_cwnd - ((tp->snd_cwnd * ca->beta) >> BETA_SHIFT), 2U);
+       decr = (tcp_snd_cwnd(tp) * ca->beta) >> BETA_SHIFT;
+       return max(tcp_snd_cwnd(tp) - decr, 2U);
 }
 
 /* Extract info for Tcp socket info provided via netlink. */
index 2088f93fa37b5fb9110e7933242a27bd4009990e..1595b76ea2bea195b8896f4cd533beb14b56dd96 100644 (file)
@@ -414,7 +414,7 @@ static void tcp_sndbuf_expand(struct sock *sk)
        per_mss = roundup_pow_of_two(per_mss) +
                  SKB_DATA_ALIGN(sizeof(struct sk_buff));
 
-       nr_segs = max_t(u32, TCP_INIT_CWND, tp->snd_cwnd);
+       nr_segs = max_t(u32, TCP_INIT_CWND, tcp_snd_cwnd(tp));
        nr_segs = max_t(u32, nr_segs, tp->reordering + 1);
 
        /* Fast Recovery (RFC 5681 3.2) :
@@ -909,12 +909,12 @@ static void tcp_update_pacing_rate(struct sock *sk)
         *       If snd_cwnd >= (tp->snd_ssthresh / 2), we are approaching
         *       end of slow start and should slow down.
         */
-       if (tp->snd_cwnd < tp->snd_ssthresh / 2)
+       if (tcp_snd_cwnd(tp) < tp->snd_ssthresh / 2)
                rate *= sock_net(sk)->ipv4.sysctl_tcp_pacing_ss_ratio;
        else
                rate *= sock_net(sk)->ipv4.sysctl_tcp_pacing_ca_ratio;
 
-       rate *= max(tp->snd_cwnd, tp->packets_out);
+       rate *= max(tcp_snd_cwnd(tp), tp->packets_out);
 
        if (likely(tp->srtt_us))
                do_div(rate, tp->srtt_us);
@@ -2147,12 +2147,12 @@ void tcp_enter_loss(struct sock *sk)
            !after(tp->high_seq, tp->snd_una) ||
            (icsk->icsk_ca_state == TCP_CA_Loss && !icsk->icsk_retransmits)) {
                tp->prior_ssthresh = tcp_current_ssthresh(sk);
-               tp->prior_cwnd = tp->snd_cwnd;
+               tp->prior_cwnd = tcp_snd_cwnd(tp);
                tp->snd_ssthresh = icsk->icsk_ca_ops->ssthresh(sk);
                tcp_ca_event(sk, CA_EVENT_LOSS);
                tcp_init_undo(tp);
        }
-       tp->snd_cwnd       = tcp_packets_in_flight(tp) + 1;
+       tcp_snd_cwnd_set(tp, tcp_packets_in_flight(tp) + 1);
        tp->snd_cwnd_cnt   = 0;
        tp->snd_cwnd_stamp = tcp_jiffies32;
 
@@ -2458,7 +2458,7 @@ static void DBGUNDO(struct sock *sk, const char *msg)
                pr_debug("Undo %s %pI4/%u c%u l%u ss%u/%u p%u\n",
                         msg,
                         &inet->inet_daddr, ntohs(inet->inet_dport),
-                        tp->snd_cwnd, tcp_left_out(tp),
+                        tcp_snd_cwnd(tp), tcp_left_out(tp),
                         tp->snd_ssthresh, tp->prior_ssthresh,
                         tp->packets_out);
        }
@@ -2467,7 +2467,7 @@ static void DBGUNDO(struct sock *sk, const char *msg)
                pr_debug("Undo %s %pI6/%u c%u l%u ss%u/%u p%u\n",
                         msg,
                         &sk->sk_v6_daddr, ntohs(inet->inet_dport),
-                        tp->snd_cwnd, tcp_left_out(tp),
+                        tcp_snd_cwnd(tp), tcp_left_out(tp),
                         tp->snd_ssthresh, tp->prior_ssthresh,
                         tp->packets_out);
        }
@@ -2492,7 +2492,7 @@ static void tcp_undo_cwnd_reduction(struct sock *sk, bool unmark_loss)
        if (tp->prior_ssthresh) {
                const struct inet_connection_sock *icsk = inet_csk(sk);
 
-               tp->snd_cwnd = icsk->icsk_ca_ops->undo_cwnd(sk);
+               tcp_snd_cwnd_set(tp, icsk->icsk_ca_ops->undo_cwnd(sk));
 
                if (tp->prior_ssthresh > tp->snd_ssthresh) {
                        tp->snd_ssthresh = tp->prior_ssthresh;
@@ -2599,7 +2599,7 @@ static void tcp_init_cwnd_reduction(struct sock *sk)
        tp->high_seq = tp->snd_nxt;
        tp->tlp_high_seq = 0;
        tp->snd_cwnd_cnt = 0;
-       tp->prior_cwnd = tp->snd_cwnd;
+       tp->prior_cwnd = tcp_snd_cwnd(tp);
        tp->prr_delivered = 0;
        tp->prr_out = 0;
        tp->snd_ssthresh = inet_csk(sk)->icsk_ca_ops->ssthresh(sk);
@@ -2629,7 +2629,7 @@ void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost,
        }
        /* Force a fast retransmit upon entering fast recovery */
        sndcnt = max(sndcnt, (tp->prr_out ? 0 : 1));
-       tp->snd_cwnd = tcp_packets_in_flight(tp) + sndcnt;
+       tcp_snd_cwnd_set(tp, tcp_packets_in_flight(tp) + sndcnt);
 }
 
 static inline void tcp_end_cwnd_reduction(struct sock *sk)
@@ -2642,7 +2642,7 @@ static inline void tcp_end_cwnd_reduction(struct sock *sk)
        /* Reset cwnd to ssthresh in CWR or Recovery (unless it's undone) */
        if (tp->snd_ssthresh < TCP_INFINITE_SSTHRESH &&
            (inet_csk(sk)->icsk_ca_state == TCP_CA_CWR || tp->undo_marker)) {
-               tp->snd_cwnd = tp->snd_ssthresh;
+               tcp_snd_cwnd_set(tp, tp->snd_ssthresh);
                tp->snd_cwnd_stamp = tcp_jiffies32;
        }
        tcp_ca_event(sk, CA_EVENT_COMPLETE_CWR);
@@ -2709,9 +2709,9 @@ static void tcp_mtup_probe_success(struct sock *sk)
 
        /* FIXME: breaks with very large cwnd */
        tp->prior_ssthresh = tcp_current_ssthresh(sk);
-       tp->snd_cwnd = tp->snd_cwnd *
-                      tcp_mss_to_mtu(sk, tp->mss_cache) /
-                      icsk->icsk_mtup.probe_size;
+       tcp_snd_cwnd_set(tp, tcp_snd_cwnd(tp) *
+                            tcp_mss_to_mtu(sk, tp->mss_cache) /
+                            icsk->icsk_mtup.probe_size);
        tp->snd_cwnd_cnt = 0;
        tp->snd_cwnd_stamp = tcp_jiffies32;
        tp->snd_ssthresh = tcp_current_ssthresh(sk);
@@ -3034,7 +3034,7 @@ static void tcp_fastretrans_alert(struct sock *sk, const u32 prior_snd_una,
                    tp->snd_una == tp->mtu_probe.probe_seq_start) {
                        tcp_mtup_probe_failed(sk);
                        /* Restores the reduction we did in tcp_mtup_probe() */
-                       tp->snd_cwnd++;
+                       tcp_snd_cwnd_set(tp, tcp_snd_cwnd(tp) + 1);
                        tcp_simple_retransmit(sk);
                        return;
                }
@@ -5436,7 +5436,7 @@ static bool tcp_should_expand_sndbuf(struct sock *sk)
                return false;
 
        /* If we filled the congestion window, do not expand.  */
-       if (tcp_packets_in_flight(tp) >= tp->snd_cwnd)
+       if (tcp_packets_in_flight(tp) >= tcp_snd_cwnd(tp))
                return false;
 
        return true;
@@ -5998,9 +5998,9 @@ void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb)
         * retransmission has occurred.
         */
        if (tp->total_retrans > 1 && tp->undo_marker)
-               tp->snd_cwnd = 1;
+               tcp_snd_cwnd_set(tp, 1);
        else
-               tp->snd_cwnd = tcp_init_cwnd(tp, __sk_dst_get(sk));
+               tcp_snd_cwnd_set(tp, tcp_init_cwnd(tp, __sk_dst_get(sk)));
        tp->snd_cwnd_stamp = tcp_jiffies32;
 
        bpf_skops_established(sk, bpf_op, skb);
index f9cec624068dfa1d218357d7e88c89459d7d54f4..157265aecbedd1761de2d892b5a54f4a0cfe1912 100644 (file)
@@ -2621,7 +2621,7 @@ static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
                jiffies_to_clock_t(icsk->icsk_rto),
                jiffies_to_clock_t(icsk->icsk_ack.ato),
                (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sk),
-               tp->snd_cwnd,
+               tcp_snd_cwnd(tp),
                state == TCP_LISTEN ?
                    fastopenq->max_qlen :
                    (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
index 82b36ec3f2f82b3f5c33f3cd21fda34a4d015a4f..ae36780977d2762066cdd59e40116d1240492b90 100644 (file)
@@ -297,7 +297,7 @@ static void tcp_lp_pkts_acked(struct sock *sk, const struct ack_sample *sample)
                lp->flag &= ~LP_WITHIN_THR;
 
        pr_debug("TCP-LP: %05o|%5u|%5u|%15u|%15u|%15u\n", lp->flag,
-                tp->snd_cwnd, lp->remote_hz, lp->owd_min, lp->owd_max,
+                tcp_snd_cwnd(tp), lp->remote_hz, lp->owd_min, lp->owd_max,
                 lp->sowd >> 3);
 
        if (lp->flag & LP_WITHIN_THR)
@@ -313,12 +313,12 @@ static void tcp_lp_pkts_acked(struct sock *sk, const struct ack_sample *sample)
        /* happened within inference
         * drop snd_cwnd into 1 */
        if (lp->flag & LP_WITHIN_INF)
-               tp->snd_cwnd = 1U;
+               tcp_snd_cwnd_set(tp, 1U);
 
        /* happened after inference
         * cut snd_cwnd into half */
        else
-               tp->snd_cwnd = max(tp->snd_cwnd >> 1U, 1U);
+               tcp_snd_cwnd_set(tp, max(tcp_snd_cwnd(tp) >> 1U, 1U));
 
        /* record this drop time */
        lp->last_drop = now;
index 0588b004ddac1636be7ef1c5e57f5910dbfd8756..7029b0e98edb285102dcab4521f296511a70dc57 100644 (file)
@@ -388,15 +388,15 @@ void tcp_update_metrics(struct sock *sk)
                if (!net->ipv4.sysctl_tcp_no_ssthresh_metrics_save &&
                    !tcp_metric_locked(tm, TCP_METRIC_SSTHRESH)) {
                        val = tcp_metric_get(tm, TCP_METRIC_SSTHRESH);
-                       if (val && (tp->snd_cwnd >> 1) > val)
+                       if (val && (tcp_snd_cwnd(tp) >> 1) > val)
                                tcp_metric_set(tm, TCP_METRIC_SSTHRESH,
-                                              tp->snd_cwnd >> 1);
+                                              tcp_snd_cwnd(tp) >> 1);
                }
                if (!tcp_metric_locked(tm, TCP_METRIC_CWND)) {
                        val = tcp_metric_get(tm, TCP_METRIC_CWND);
-                       if (tp->snd_cwnd > val)
+                       if (tcp_snd_cwnd(tp) > val)
                                tcp_metric_set(tm, TCP_METRIC_CWND,
-                                              tp->snd_cwnd);
+                                              tcp_snd_cwnd(tp));
                }
        } else if (!tcp_in_slow_start(tp) &&
                   icsk->icsk_ca_state == TCP_CA_Open) {
@@ -404,10 +404,10 @@ void tcp_update_metrics(struct sock *sk)
                if (!net->ipv4.sysctl_tcp_no_ssthresh_metrics_save &&
                    !tcp_metric_locked(tm, TCP_METRIC_SSTHRESH))
                        tcp_metric_set(tm, TCP_METRIC_SSTHRESH,
-                                      max(tp->snd_cwnd >> 1, tp->snd_ssthresh));
+                                      max(tcp_snd_cwnd(tp) >> 1, tp->snd_ssthresh));
                if (!tcp_metric_locked(tm, TCP_METRIC_CWND)) {
                        val = tcp_metric_get(tm, TCP_METRIC_CWND);
-                       tcp_metric_set(tm, TCP_METRIC_CWND, (val + tp->snd_cwnd) >> 1);
+                       tcp_metric_set(tm, TCP_METRIC_CWND, (val + tcp_snd_cwnd(tp)) >> 1);
                }
        } else {
                /* Else slow start did not finish, cwnd is non-sense,
index ab552356bdba8f92f414aec55a5a9cd93b7975b2..a60662f4bdf92cba1d92a3eedd7c607d1537d7f2 100644 (file)
@@ -197,10 +197,10 @@ static void tcpnv_cong_avoid(struct sock *sk, u32 ack, u32 acked)
        }
 
        if (ca->cwnd_growth_factor < 0) {
-               cnt = tp->snd_cwnd << -ca->cwnd_growth_factor;
+               cnt = tcp_snd_cwnd(tp) << -ca->cwnd_growth_factor;
                tcp_cong_avoid_ai(tp, cnt, acked);
        } else {
-               cnt = max(4U, tp->snd_cwnd >> ca->cwnd_growth_factor);
+               cnt = max(4U, tcp_snd_cwnd(tp) >> ca->cwnd_growth_factor);
                tcp_cong_avoid_ai(tp, cnt, acked);
        }
 }
@@ -209,7 +209,7 @@ static u32 tcpnv_recalc_ssthresh(struct sock *sk)
 {
        const struct tcp_sock *tp = tcp_sk(sk);
 
-       return max((tp->snd_cwnd * nv_loss_dec_factor) >> 10, 2U);
+       return max((tcp_snd_cwnd(tp) * nv_loss_dec_factor) >> 10, 2U);
 }
 
 static void tcpnv_state(struct sock *sk, u8 new_state)
@@ -257,7 +257,7 @@ static void tcpnv_acked(struct sock *sk, const struct ack_sample *sample)
                return;
 
        /* Stop cwnd growth if we were in catch up mode */
-       if (ca->nv_catchup && tp->snd_cwnd >= nv_min_cwnd) {
+       if (ca->nv_catchup && tcp_snd_cwnd(tp) >= nv_min_cwnd) {
                ca->nv_catchup = 0;
                ca->nv_allow_cwnd_growth = 0;
        }
@@ -371,7 +371,7 @@ static void tcpnv_acked(struct sock *sk, const struct ack_sample *sample)
                 * if cwnd < max_win, grow cwnd
                 * else leave the same
                 */
-               if (tp->snd_cwnd > max_win) {
+               if (tcp_snd_cwnd(tp) > max_win) {
                        /* there is congestion, check that it is ok
                         * to make a CA decision
                         * 1. We should have at least nv_dec_eval_min_calls
@@ -398,20 +398,20 @@ static void tcpnv_acked(struct sock *sk, const struct ack_sample *sample)
                        ca->nv_allow_cwnd_growth = 0;
                        tp->snd_ssthresh =
                                (nv_ssthresh_factor * max_win) >> 3;
-                       if (tp->snd_cwnd - max_win > 2) {
+                       if (tcp_snd_cwnd(tp) - max_win > 2) {
                                /* gap > 2, we do exponential cwnd decrease */
                                int dec;
 
-                               dec = max(2U, ((tp->snd_cwnd - max_win) *
+                               dec = max(2U, ((tcp_snd_cwnd(tp) - max_win) *
                                               nv_cong_dec_mult) >> 7);
-                               tp->snd_cwnd -= dec;
+                               tcp_snd_cwnd_set(tp, tcp_snd_cwnd(tp) - dec);
                        } else if (nv_cong_dec_mult > 0) {
-                               tp->snd_cwnd = max_win;
+                               tcp_snd_cwnd_set(tp, max_win);
                        }
                        if (ca->cwnd_growth_factor > 0)
                                ca->cwnd_growth_factor = 0;
                        ca->nv_no_cong_cnt = 0;
-               } else if (tp->snd_cwnd <= max_win - nv_pad_buffer) {
+               } else if (tcp_snd_cwnd(tp) <= max_win - nv_pad_buffer) {
                        /* There is no congestion, grow cwnd if allowed*/
                        if (ca->nv_eval_call_cnt < nv_inc_eval_min_calls)
                                return;
@@ -444,8 +444,8 @@ static void tcpnv_acked(struct sock *sk, const struct ack_sample *sample)
                 * (it wasn't before, if it is now is because nv
                 *  decreased it).
                 */
-               if (tp->snd_cwnd < nv_min_cwnd)
-                       tp->snd_cwnd = nv_min_cwnd;
+               if (tcp_snd_cwnd(tp) < nv_min_cwnd)
+                       tcp_snd_cwnd_set(tp, nv_min_cwnd);
        }
 }
 
index 9ede847f4199844c5884e3f62ea450562072a0a7..c221f3bce975ef17189d6250a6b52f5c80f142c8 100644 (file)
@@ -142,7 +142,7 @@ void tcp_cwnd_restart(struct sock *sk, s32 delta)
 {
        struct tcp_sock *tp = tcp_sk(sk);
        u32 restart_cwnd = tcp_init_cwnd(tp, __sk_dst_get(sk));
-       u32 cwnd = tp->snd_cwnd;
+       u32 cwnd = tcp_snd_cwnd(tp);
 
        tcp_ca_event(sk, CA_EVENT_CWND_RESTART);
 
@@ -151,7 +151,7 @@ void tcp_cwnd_restart(struct sock *sk, s32 delta)
 
        while ((delta -= inet_csk(sk)->icsk_rto) > 0 && cwnd > restart_cwnd)
                cwnd >>= 1;
-       tp->snd_cwnd = max(cwnd, restart_cwnd);
+       tcp_snd_cwnd_set(tp, max(cwnd, restart_cwnd));
        tp->snd_cwnd_stamp = tcp_jiffies32;
        tp->snd_cwnd_used = 0;
 }
@@ -1013,7 +1013,7 @@ static void tcp_tsq_write(struct sock *sk)
                struct tcp_sock *tp = tcp_sk(sk);
 
                if (tp->lost_out > tp->retrans_out &&
-                   tp->snd_cwnd > tcp_packets_in_flight(tp)) {
+                   tcp_snd_cwnd(tp) > tcp_packets_in_flight(tp)) {
                        tcp_mstamp_refresh(tp);
                        tcp_xmit_retransmit_queue(sk);
                }
@@ -1860,9 +1860,9 @@ static void tcp_cwnd_application_limited(struct sock *sk)
                /* Limited by application or receiver window. */
                u32 init_win = tcp_init_cwnd(tp, __sk_dst_get(sk));
                u32 win_used = max(tp->snd_cwnd_used, init_win);
-               if (win_used < tp->snd_cwnd) {
+               if (win_used < tcp_snd_cwnd(tp)) {
                        tp->snd_ssthresh = tcp_current_ssthresh(sk);
-                       tp->snd_cwnd = (tp->snd_cwnd + win_used) >> 1;
+                       tcp_snd_cwnd_set(tp, (tcp_snd_cwnd(tp) + win_used) >> 1);
                }
                tp->snd_cwnd_used = 0;
        }
@@ -2043,7 +2043,7 @@ static inline unsigned int tcp_cwnd_test(const struct tcp_sock *tp,
                return 1;
 
        in_flight = tcp_packets_in_flight(tp);
-       cwnd = tp->snd_cwnd;
+       cwnd = tcp_snd_cwnd(tp);
        if (in_flight >= cwnd)
                return 0;
 
@@ -2196,12 +2196,12 @@ static bool tcp_tso_should_defer(struct sock *sk, struct sk_buff *skb,
        in_flight = tcp_packets_in_flight(tp);
 
        BUG_ON(tcp_skb_pcount(skb) <= 1);
-       BUG_ON(tp->snd_cwnd <= in_flight);
+       BUG_ON(tcp_snd_cwnd(tp) <= in_flight);
 
        send_win = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq;
 
        /* From in_flight test above, we know that cwnd > in_flight.  */
-       cong_win = (tp->snd_cwnd - in_flight) * tp->mss_cache;
+       cong_win = (tcp_snd_cwnd(tp) - in_flight) * tp->mss_cache;
 
        limit = min(send_win, cong_win);
 
@@ -2215,7 +2215,7 @@ static bool tcp_tso_should_defer(struct sock *sk, struct sk_buff *skb,
 
        win_divisor = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_tso_win_divisor);
        if (win_divisor) {
-               u32 chunk = min(tp->snd_wnd, tp->snd_cwnd * tp->mss_cache);
+               u32 chunk = min(tp->snd_wnd, tcp_snd_cwnd(tp) * tp->mss_cache);
 
                /* If at least some fraction of a window is available,
                 * just use it.
@@ -2345,7 +2345,7 @@ static int tcp_mtu_probe(struct sock *sk)
        if (likely(!icsk->icsk_mtup.enabled ||
                   icsk->icsk_mtup.probe_size ||
                   inet_csk(sk)->icsk_ca_state != TCP_CA_Open ||
-                  tp->snd_cwnd < 11 ||
+                  tcp_snd_cwnd(tp) < 11 ||
                   tp->rx_opt.num_sacks || tp->rx_opt.dsack))
                return -1;
 
@@ -2381,7 +2381,7 @@ static int tcp_mtu_probe(struct sock *sk)
                return 0;
 
        /* Do we need to wait to drain cwnd? With none in flight, don't stall */
-       if (tcp_packets_in_flight(tp) + 2 > tp->snd_cwnd) {
+       if (tcp_packets_in_flight(tp) + 2 > tcp_snd_cwnd(tp)) {
                if (!tcp_packets_in_flight(tp))
                        return -1;
                else
@@ -2450,7 +2450,7 @@ static int tcp_mtu_probe(struct sock *sk)
        if (!tcp_transmit_skb(sk, nskb, 1, GFP_ATOMIC)) {
                /* Decrement cwnd here because we are sending
                 * effectively two packets. */
-               tp->snd_cwnd--;
+               tcp_snd_cwnd_set(tp, tcp_snd_cwnd(tp) - 1);
                tcp_event_new_data_sent(sk, nskb);
 
                icsk->icsk_mtup.probe_size = tcp_mss_to_mtu(sk, nskb->len);
@@ -2708,7 +2708,7 @@ repair:
        else
                tcp_chrono_stop(sk, TCP_CHRONO_RWND_LIMITED);
 
-       is_cwnd_limited |= (tcp_packets_in_flight(tp) >= tp->snd_cwnd);
+       is_cwnd_limited |= (tcp_packets_in_flight(tp) >= tcp_snd_cwnd(tp));
        if (likely(sent_pkts || is_cwnd_limited))
                tcp_cwnd_validate(sk, is_cwnd_limited);
 
@@ -2818,7 +2818,7 @@ void tcp_send_loss_probe(struct sock *sk)
        if (unlikely(!skb)) {
                WARN_ONCE(tp->packets_out,
                          "invalid inflight: %u state %u cwnd %u mss %d\n",
-                         tp->packets_out, sk->sk_state, tp->snd_cwnd, mss);
+                         tp->packets_out, sk->sk_state, tcp_snd_cwnd(tp), mss);
                inet_csk(sk)->icsk_pending = 0;
                return;
        }
@@ -3302,7 +3302,7 @@ void tcp_xmit_retransmit_queue(struct sock *sk)
                if (!hole)
                        tp->retransmit_skb_hint = skb;
 
-               segs = tp->snd_cwnd - tcp_packets_in_flight(tp);
+               segs = tcp_snd_cwnd(tp) - tcp_packets_in_flight(tp);
                if (segs <= 0)
                        break;
                sacked = TCP_SKB_CB(skb)->sacked;
index fbab921670cc91a121f2ab8cd7aa6ecfd3748535..617b8187c03d9c339958089258ab30e149364eef 100644 (file)
@@ -195,7 +195,7 @@ void tcp_rate_check_app_limited(struct sock *sk)
            /* Nothing in sending host's qdisc queues or NIC tx queue. */
            sk_wmem_alloc_get(sk) < SKB_TRUESIZE(1) &&
            /* We are not limited by CWND. */
-           tcp_packets_in_flight(tp) < tp->snd_cwnd &&
+           tcp_packets_in_flight(tp) < tcp_snd_cwnd(tp) &&
            /* All lost packets have been retransmitted. */
            tp->lost_out <= tp->retrans_out)
                tp->app_limited =
index 5842081bc8a25b53bc10745688d97ac59a0ee200..862b96248a92dc1b7fa15d688d31ae043fc446a7 100644 (file)
@@ -27,7 +27,7 @@ static void tcp_scalable_cong_avoid(struct sock *sk, u32 ack, u32 acked)
                if (!acked)
                        return;
        }
-       tcp_cong_avoid_ai(tp, min(tp->snd_cwnd, TCP_SCALABLE_AI_CNT),
+       tcp_cong_avoid_ai(tp, min(tcp_snd_cwnd(tp), TCP_SCALABLE_AI_CNT),
                          acked);
 }
 
@@ -35,7 +35,7 @@ static u32 tcp_scalable_ssthresh(struct sock *sk)
 {
        const struct tcp_sock *tp = tcp_sk(sk);
 
-       return max(tp->snd_cwnd - (tp->snd_cwnd>>TCP_SCALABLE_MD_SCALE), 2U);
+       return max(tcp_snd_cwnd(tp) - (tcp_snd_cwnd(tp)>>TCP_SCALABLE_MD_SCALE), 2U);
 }
 
 static struct tcp_congestion_ops tcp_scalable __read_mostly = {
index c8003c8aad2c00835a57d6084e788fb453a96026..786848ad37ea8d5f9bd817666181905f3f6ec9d4 100644 (file)
@@ -159,7 +159,7 @@ EXPORT_SYMBOL_GPL(tcp_vegas_cwnd_event);
 
 static inline u32 tcp_vegas_ssthresh(struct tcp_sock *tp)
 {
-       return  min(tp->snd_ssthresh, tp->snd_cwnd);
+       return  min(tp->snd_ssthresh, tcp_snd_cwnd(tp));
 }
 
 static void tcp_vegas_cong_avoid(struct sock *sk, u32 ack, u32 acked)
@@ -217,14 +217,14 @@ static void tcp_vegas_cong_avoid(struct sock *sk, u32 ack, u32 acked)
                         * This is:
                         *     (actual rate in segments) * baseRTT
                         */
-                       target_cwnd = (u64)tp->snd_cwnd * vegas->baseRTT;
+                       target_cwnd = (u64)tcp_snd_cwnd(tp) * vegas->baseRTT;
                        do_div(target_cwnd, rtt);
 
                        /* Calculate the difference between the window we had,
                         * and the window we would like to have. This quantity
                         * is the "Diff" from the Arizona Vegas papers.
                         */
-                       diff = tp->snd_cwnd * (rtt-vegas->baseRTT) / vegas->baseRTT;
+                       diff = tcp_snd_cwnd(tp) * (rtt-vegas->baseRTT) / vegas->baseRTT;
 
                        if (diff > gamma && tcp_in_slow_start(tp)) {
                                /* Going too fast. Time to slow down
@@ -238,7 +238,8 @@ static void tcp_vegas_cong_avoid(struct sock *sk, u32 ack, u32 acked)
                                 * truncation robs us of full link
                                 * utilization.
                                 */
-                               tp->snd_cwnd = min(tp->snd_cwnd, (u32)target_cwnd+1);
+                               tcp_snd_cwnd_set(tp, min(tcp_snd_cwnd(tp),
+                                                        (u32)target_cwnd + 1));
                                tp->snd_ssthresh = tcp_vegas_ssthresh(tp);
 
                        } else if (tcp_in_slow_start(tp)) {
@@ -254,14 +255,14 @@ static void tcp_vegas_cong_avoid(struct sock *sk, u32 ack, u32 acked)
                                        /* The old window was too fast, so
                                         * we slow down.
                                         */
-                                       tp->snd_cwnd--;
+                                       tcp_snd_cwnd_set(tp, tcp_snd_cwnd(tp) - 1);
                                        tp->snd_ssthresh
                                                = tcp_vegas_ssthresh(tp);
                                } else if (diff < alpha) {
                                        /* We don't have enough extra packets
                                         * in the network, so speed up.
                                         */
-                                       tp->snd_cwnd++;
+                                       tcp_snd_cwnd_set(tp, tcp_snd_cwnd(tp) + 1);
                                } else {
                                        /* Sending just as fast as we
                                         * should be.
@@ -269,10 +270,10 @@ static void tcp_vegas_cong_avoid(struct sock *sk, u32 ack, u32 acked)
                                }
                        }
 
-                       if (tp->snd_cwnd < 2)
-                               tp->snd_cwnd = 2;
-                       else if (tp->snd_cwnd > tp->snd_cwnd_clamp)
-                               tp->snd_cwnd = tp->snd_cwnd_clamp;
+                       if (tcp_snd_cwnd(tp) < 2)
+                               tcp_snd_cwnd_set(tp, 2);
+                       else if (tcp_snd_cwnd(tp) > tp->snd_cwnd_clamp)
+                               tcp_snd_cwnd_set(tp, tp->snd_cwnd_clamp);
 
                        tp->snd_ssthresh = tcp_current_ssthresh(sk);
                }
index cd50a61c9976d78dbcc5fc27b8f687ee2f3fbebe..366ff6f214b2ee746cecdf10353e4624252478e0 100644 (file)
@@ -146,11 +146,11 @@ static void tcp_veno_cong_avoid(struct sock *sk, u32 ack, u32 acked)
 
                rtt = veno->minrtt;
 
-               target_cwnd = (u64)tp->snd_cwnd * veno->basertt;
+               target_cwnd = (u64)tcp_snd_cwnd(tp) * veno->basertt;
                target_cwnd <<= V_PARAM_SHIFT;
                do_div(target_cwnd, rtt);
 
-               veno->diff = (tp->snd_cwnd << V_PARAM_SHIFT) - target_cwnd;
+               veno->diff = (tcp_snd_cwnd(tp) << V_PARAM_SHIFT) - target_cwnd;
 
                if (tcp_in_slow_start(tp)) {
                        /* Slow start. */
@@ -164,15 +164,15 @@ static void tcp_veno_cong_avoid(struct sock *sk, u32 ack, u32 acked)
                        /* In the "non-congestive state", increase cwnd
                         * every rtt.
                         */
-                       tcp_cong_avoid_ai(tp, tp->snd_cwnd, acked);
+                       tcp_cong_avoid_ai(tp, tcp_snd_cwnd(tp), acked);
                } else {
                        /* In the "congestive state", increase cwnd
                         * every other rtt.
                         */
-                       if (tp->snd_cwnd_cnt >= tp->snd_cwnd) {
+                       if (tp->snd_cwnd_cnt >= tcp_snd_cwnd(tp)) {
                                if (veno->inc &&
-                                   tp->snd_cwnd < tp->snd_cwnd_clamp) {
-                                       tp->snd_cwnd++;
+                                   tcp_snd_cwnd(tp) < tp->snd_cwnd_clamp) {
+                                       tcp_snd_cwnd_set(tp, tcp_snd_cwnd(tp) + 1);
                                        veno->inc = 0;
                                } else
                                        veno->inc = 1;
@@ -181,10 +181,10 @@ static void tcp_veno_cong_avoid(struct sock *sk, u32 ack, u32 acked)
                                tp->snd_cwnd_cnt += acked;
                }
 done:
-               if (tp->snd_cwnd < 2)
-                       tp->snd_cwnd = 2;
-               else if (tp->snd_cwnd > tp->snd_cwnd_clamp)
-                       tp->snd_cwnd = tp->snd_cwnd_clamp;
+               if (tcp_snd_cwnd(tp) < 2)
+                       tcp_snd_cwnd_set(tp, 2);
+               else if (tcp_snd_cwnd(tp) > tp->snd_cwnd_clamp)
+                       tcp_snd_cwnd_set(tp, tp->snd_cwnd_clamp);
        }
        /* Wipe the slate clean for the next rtt. */
        /* veno->cntrtt = 0; */
@@ -199,10 +199,10 @@ static u32 tcp_veno_ssthresh(struct sock *sk)
 
        if (veno->diff < beta)
                /* in "non-congestive state", cut cwnd by 1/5 */
-               return max(tp->snd_cwnd * 4 / 5, 2U);
+               return max(tcp_snd_cwnd(tp) * 4 / 5, 2U);
        else
                /* in "congestive state", cut cwnd by 1/2 */
-               return max(tp->snd_cwnd >> 1U, 2U);
+               return max(tcp_snd_cwnd(tp) >> 1U, 2U);
 }
 
 static struct tcp_congestion_ops tcp_veno __read_mostly = {
index b2e05c4cea00fd7b524e061ca9d095fd9f656e64..c6e97141eef2591c5ab50f4058a5377e0855313f 100644 (file)
@@ -244,7 +244,8 @@ static void tcp_westwood_event(struct sock *sk, enum tcp_ca_event event)
 
        switch (event) {
        case CA_EVENT_COMPLETE_CWR:
-               tp->snd_cwnd = tp->snd_ssthresh = tcp_westwood_bw_rttmin(sk);
+               tp->snd_ssthresh = tcp_westwood_bw_rttmin(sk);
+               tcp_snd_cwnd_set(tp, tp->snd_ssthresh);
                break;
        case CA_EVENT_LOSS:
                tp->snd_ssthresh = tcp_westwood_bw_rttmin(sk);
index 07c4c93b9fdb65a513d99532ed7c066cec7d89f4..18b07ff5d20e6c5eefe9ab54c6b8e429f01d64b9 100644 (file)
@@ -71,11 +71,11 @@ static void tcp_yeah_cong_avoid(struct sock *sk, u32 ack, u32 acked)
 
        if (!yeah->doing_reno_now) {
                /* Scalable */
-               tcp_cong_avoid_ai(tp, min(tp->snd_cwnd, TCP_SCALABLE_AI_CNT),
+               tcp_cong_avoid_ai(tp, min(tcp_snd_cwnd(tp), TCP_SCALABLE_AI_CNT),
                                  acked);
        } else {
                /* Reno */
-               tcp_cong_avoid_ai(tp, tp->snd_cwnd, acked);
+               tcp_cong_avoid_ai(tp, tcp_snd_cwnd(tp), acked);
        }
 
        /* The key players are v_vegas.beg_snd_una and v_beg_snd_nxt.
@@ -130,7 +130,7 @@ do_vegas:
                        /* Compute excess number of packets above bandwidth
                         * Avoid doing full 64 bit divide.
                         */
-                       bw = tp->snd_cwnd;
+                       bw = tcp_snd_cwnd(tp);
                        bw *= rtt - yeah->vegas.baseRTT;
                        do_div(bw, rtt);
                        queue = bw;
@@ -138,20 +138,20 @@ do_vegas:
                        if (queue > TCP_YEAH_ALPHA ||
                            rtt - yeah->vegas.baseRTT > (yeah->vegas.baseRTT / TCP_YEAH_PHY)) {
                                if (queue > TCP_YEAH_ALPHA &&
-                                   tp->snd_cwnd > yeah->reno_count) {
+                                   tcp_snd_cwnd(tp) > yeah->reno_count) {
                                        u32 reduction = min(queue / TCP_YEAH_GAMMA ,
-                                                           tp->snd_cwnd >> TCP_YEAH_EPSILON);
+                                                           tcp_snd_cwnd(tp) >> TCP_YEAH_EPSILON);
 
-                                       tp->snd_cwnd -= reduction;
+                                       tcp_snd_cwnd_set(tp, tcp_snd_cwnd(tp) - reduction);
 
-                                       tp->snd_cwnd = max(tp->snd_cwnd,
-                                                          yeah->reno_count);
+                                       tcp_snd_cwnd_set(tp, max(tcp_snd_cwnd(tp),
+                                                                yeah->reno_count));
 
-                                       tp->snd_ssthresh = tp->snd_cwnd;
+                                       tp->snd_ssthresh = tcp_snd_cwnd(tp);
                                }
 
                                if (yeah->reno_count <= 2)
-                                       yeah->reno_count = max(tp->snd_cwnd>>1, 2U);
+                                       yeah->reno_count = max(tcp_snd_cwnd(tp)>>1, 2U);
                                else
                                        yeah->reno_count++;
 
@@ -176,7 +176,7 @@ do_vegas:
                 */
                yeah->vegas.beg_snd_una  = yeah->vegas.beg_snd_nxt;
                yeah->vegas.beg_snd_nxt  = tp->snd_nxt;
-               yeah->vegas.beg_snd_cwnd = tp->snd_cwnd;
+               yeah->vegas.beg_snd_cwnd = tcp_snd_cwnd(tp);
 
                /* Wipe the slate clean for the next RTT. */
                yeah->vegas.cntRTT = 0;
@@ -193,16 +193,16 @@ static u32 tcp_yeah_ssthresh(struct sock *sk)
        if (yeah->doing_reno_now < TCP_YEAH_RHO) {
                reduction = yeah->lastQ;
 
-               reduction = min(reduction, max(tp->snd_cwnd>>1, 2U));
+               reduction = min(reduction, max(tcp_snd_cwnd(tp)>>1, 2U));
 
-               reduction = max(reduction, tp->snd_cwnd >> TCP_YEAH_DELTA);
+               reduction = max(reduction, tcp_snd_cwnd(tp) >> TCP_YEAH_DELTA);
        } else
-               reduction = max(tp->snd_cwnd>>1, 2U);
+               reduction = max(tcp_snd_cwnd(tp)>>1, 2U);
 
        yeah->fast_count = 0;
        yeah->reno_count = max(yeah->reno_count>>1, 2U);
 
-       return max_t(int, tp->snd_cwnd - reduction, 2);
+       return max_t(int, tcp_snd_cwnd(tp) - reduction, 2);
 }
 
 static struct tcp_congestion_ops tcp_yeah __read_mostly = {
index 13678d3908fac9990e5b0c0df87fa4cca685baaf..782df529ff69e0b2fc5f9d12fc72538b7041a392 100644 (file)
@@ -2044,7 +2044,7 @@ static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
                   jiffies_to_clock_t(icsk->icsk_rto),
                   jiffies_to_clock_t(icsk->icsk_ack.ato),
                   (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sp),
-                  tp->snd_cwnd,
+                  tcp_snd_cwnd(tp),
                   state == TCP_LISTEN ?
                        fastopenq->max_qlen :
                        (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)