aboutsummaryrefslogtreecommitdiff
path: root/sys/netinet
diff options
context:
space:
mode:
Diffstat (limited to 'sys/netinet')
-rw-r--r--sys/netinet/in_pcb.c13
-rw-r--r--sys/netinet/in_proto.c2
-rw-r--r--sys/netinet/libalias/alias_db.c2
-rw-r--r--sys/netinet/raw_ip.c4
-rw-r--r--sys/netinet/tcp_stacks/bbr.c8
-rw-r--r--sys/netinet/tcp_stacks/rack.c284
-rw-r--r--sys/netinet/tcp_stacks/tcp_rack.h1
-rw-r--r--sys/netinet/tcp_subr.c2
-rw-r--r--sys/netinet/tcp_syncache.c16
-rw-r--r--sys/netinet/tcp_timer.c7
-rw-r--r--sys/netinet/udp_usrreq.c49
-rw-r--r--sys/netinet/udp_var.h1
12 files changed, 118 insertions, 271 deletions
diff --git a/sys/netinet/in_pcb.c b/sys/netinet/in_pcb.c
index dbe48242381d..712ff28768dc 100644
--- a/sys/netinet/in_pcb.c
+++ b/sys/netinet/in_pcb.c
@@ -2665,10 +2665,13 @@ in_pcbinshash(struct inpcb *inp)
INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
/*
- * Add entry to load balance group.
- * Only do this if SO_REUSEPORT_LB is set.
+ * Ignore SO_REUSEPORT_LB if the socket is connected. Really this case
+ * should be an error, but for UDP sockets it is not, and some
+ * applications erroneously set it on connected UDP sockets, so we can't
+ * change this without breaking compatibility.
*/
- if ((inp->inp_socket->so_options & SO_REUSEPORT_LB) != 0) {
+ if (!connected &&
+ (inp->inp_socket->so_options & SO_REUSEPORT_LB) != 0) {
int error = in_pcbinslbgrouphash(inp, M_NODOM);
if (error != 0)
return (error);
@@ -2770,6 +2773,10 @@ in_pcbrehash(struct inpcb *inp)
connected = !in_nullhost(inp->inp_faddr);
}
+ /* See the comment in in_pcbinshash(). */
+ if (connected && (inp->inp_flags & INP_INLBGROUP) != 0)
+ in_pcbremlbgrouphash(inp);
+
/*
* When rehashing, the caller must ensure that either the new or the old
* foreign address was unspecified.
diff --git a/sys/netinet/in_proto.c b/sys/netinet/in_proto.c
index db46da6022c5..42a6cf0b5810 100644
--- a/sys/netinet/in_proto.c
+++ b/sys/netinet/in_proto.c
@@ -108,6 +108,8 @@ SYSCTL_NODE(_net_inet, IPPROTO_ICMP, icmp, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
"ICMP");
SYSCTL_NODE(_net_inet, IPPROTO_UDP, udp, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
"UDP");
+SYSCTL_NODE(_net_inet, IPPROTO_UDPLITE, udplite, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
+ "UDP-Lite");
SYSCTL_NODE(_net_inet, IPPROTO_TCP, tcp, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
"TCP");
#if defined(SCTP) || defined(SCTP_SUPPORT)
diff --git a/sys/netinet/libalias/alias_db.c b/sys/netinet/libalias/alias_db.c
index c143d74a2f45..41f0a328daec 100644
--- a/sys/netinet/libalias/alias_db.c
+++ b/sys/netinet/libalias/alias_db.c
@@ -2181,7 +2181,7 @@ LibAliasInit(struct libalias *la)
#undef malloc /* XXX: ugly */
la = malloc(sizeof *la, M_ALIAS, M_WAITOK | M_ZERO);
#else
- la = calloc(sizeof *la, 1);
+ la = calloc(1, sizeof *la);
if (la == NULL)
return (la);
#endif
diff --git a/sys/netinet/raw_ip.c b/sys/netinet/raw_ip.c
index 66070faf97e9..bfe608be6b36 100644
--- a/sys/netinet/raw_ip.c
+++ b/sys/netinet/raw_ip.c
@@ -680,7 +680,6 @@ rip_ctloutput(struct socket *so, struct sockopt *sopt)
break;
case IP_DUMMYNET3: /* generic dummynet v.3 functions */
- case IP_DUMMYNET_GET:
if (ip_dn_ctl_ptr != NULL)
error = ip_dn_ctl_ptr(sopt);
else
@@ -747,9 +746,6 @@ rip_ctloutput(struct socket *so, struct sockopt *sopt)
break;
case IP_DUMMYNET3: /* generic dummynet v.3 functions */
- case IP_DUMMYNET_CONFIGURE:
- case IP_DUMMYNET_DEL:
- case IP_DUMMYNET_FLUSH:
if (ip_dn_ctl_ptr != NULL)
error = ip_dn_ctl_ptr(sopt);
else
diff --git a/sys/netinet/tcp_stacks/bbr.c b/sys/netinet/tcp_stacks/bbr.c
index 66983edcdd73..10383bc0801e 100644
--- a/sys/netinet/tcp_stacks/bbr.c
+++ b/sys/netinet/tcp_stacks/bbr.c
@@ -477,7 +477,7 @@ bbr_log_rtt_shrinks(struct tcp_bbr *bbr, uint32_t cts, uint32_t applied,
uint16_t set);
static struct bbr_sendmap *
bbr_find_lowest_rsm(struct tcp_bbr *bbr);
-static __inline uint32_t
+static inline uint32_t
bbr_get_rtt(struct tcp_bbr *bbr, int32_t rtt_type);
static void
bbr_log_to_start(struct tcp_bbr *bbr, uint32_t cts, uint32_t to, int32_t pacing_delay,
@@ -1841,7 +1841,7 @@ bbr_counter_destroy(void)
}
-static __inline void
+static inline void
bbr_fill_in_logging_data(struct tcp_bbr *bbr, struct tcp_log_bbr *l, uint32_t cts)
{
memset(l, 0, sizeof(union tcp_log_stackspecific));
@@ -4206,7 +4206,7 @@ bbr_calc_thresh_tlp(struct tcpcb *tp, struct tcp_bbr *bbr,
/*
* Return one of three RTTs to use (in microseconds).
*/
-static __inline uint32_t
+static inline uint32_t
bbr_get_rtt(struct tcp_bbr *bbr, int32_t rtt_type)
{
uint32_t f_rtt;
@@ -4370,7 +4370,7 @@ bbr_timeout_rack(struct tcpcb *tp, struct tcp_bbr *bbr, uint32_t cts)
return (0);
}
-static __inline void
+static inline void
bbr_clone_rsm(struct tcp_bbr *bbr, struct bbr_sendmap *nrsm, struct bbr_sendmap *rsm, uint32_t start)
{
int idx;
diff --git a/sys/netinet/tcp_stacks/rack.c b/sys/netinet/tcp_stacks/rack.c
index c7962b57a69e..9ed26d5a617b 100644
--- a/sys/netinet/tcp_stacks/rack.c
+++ b/sys/netinet/tcp_stacks/rack.c
@@ -204,10 +204,6 @@ static int32_t rack_dnd_default = 0; /* For rr_conf = 3, what is the default fo
static int32_t rack_rxt_controls = 0;
static int32_t rack_fill_cw_state = 0;
static uint8_t rack_req_measurements = 1;
-/* Attack threshold detections */
-static uint32_t rack_highest_sack_thresh_seen = 0;
-static uint32_t rack_highest_move_thresh_seen = 0;
-static uint32_t rack_merge_out_sacks_on_attack = 0;
static int32_t rack_enable_hw_pacing = 0; /* Due to CCSP keep it off by default */
static int32_t rack_hw_rate_caps = 0; /* 1; */
static int32_t rack_hw_rate_cap_per = 0; /* 0 -- off */
@@ -223,7 +219,6 @@ static int32_t rack_default_pacing_divisor = 250;
static uint16_t rack_pacing_min_seg = 0;
static int32_t rack_timely_off = 0;
-static uint32_t sad_seg_size_per = 800; /* 80.0 % */
static int32_t rack_pkt_delay = 1000;
static int32_t rack_send_a_lot_in_prr = 1;
static int32_t rack_min_to = 1000; /* Number of microsecond min timeout */
@@ -399,18 +394,6 @@ counter_u64_t rack_extended_rfo;
counter_u64_t rack_sack_proc_all;
counter_u64_t rack_sack_proc_short;
counter_u64_t rack_sack_proc_restart;
-counter_u64_t rack_sack_attacks_detected;
-counter_u64_t rack_sack_attacks_reversed;
-counter_u64_t rack_sack_attacks_suspect;
-counter_u64_t rack_sack_used_next_merge;
-counter_u64_t rack_sack_splits;
-counter_u64_t rack_sack_used_prev_merge;
-counter_u64_t rack_sack_skipped_acked;
-counter_u64_t rack_ack_total;
-counter_u64_t rack_express_sack;
-counter_u64_t rack_sack_total;
-counter_u64_t rack_move_none;
-counter_u64_t rack_move_some;
counter_u64_t rack_input_idle_reduces;
counter_u64_t rack_collapsed_win;
@@ -834,18 +817,6 @@ sysctl_rack_clear(SYSCTL_HANDLER_ARGS)
counter_u64_zero(rack_rxt_clamps_cwnd_uniq);
counter_u64_zero(rack_multi_single_eq);
counter_u64_zero(rack_proc_non_comp_ack);
- counter_u64_zero(rack_sack_attacks_detected);
- counter_u64_zero(rack_sack_attacks_reversed);
- counter_u64_zero(rack_sack_attacks_suspect);
- counter_u64_zero(rack_sack_used_next_merge);
- counter_u64_zero(rack_sack_used_prev_merge);
- counter_u64_zero(rack_sack_splits);
- counter_u64_zero(rack_sack_skipped_acked);
- counter_u64_zero(rack_ack_total);
- counter_u64_zero(rack_express_sack);
- counter_u64_zero(rack_sack_total);
- counter_u64_zero(rack_move_none);
- counter_u64_zero(rack_move_some);
counter_u64_zero(rack_try_scwnd);
counter_u64_zero(rack_collapsed_win);
counter_u64_zero(rack_collapsed_win_rxt);
@@ -872,7 +843,6 @@ static void
rack_init_sysctls(void)
{
struct sysctl_oid *rack_counters;
- struct sysctl_oid *rack_attack;
struct sysctl_oid *rack_pacing;
struct sysctl_oid *rack_timely;
struct sysctl_oid *rack_timers;
@@ -883,12 +853,6 @@ rack_init_sysctls(void)
struct sysctl_oid *rack_probertt;
struct sysctl_oid *rack_hw_pacing;
- rack_attack = SYSCTL_ADD_NODE(&rack_sysctl_ctx,
- SYSCTL_CHILDREN(rack_sysctl_root),
- OID_AUTO,
- "sack_attack",
- CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
- "Rack Sack Attack Counters and Controls");
rack_counters = SYSCTL_ADD_NODE(&rack_sysctl_ctx,
SYSCTL_CHILDREN(rack_sysctl_root),
OID_AUTO,
@@ -1535,11 +1499,6 @@ rack_init_sysctls(void)
"Do not disturb default for rack_rrr = 3");
SYSCTL_ADD_S32(&rack_sysctl_ctx,
SYSCTL_CHILDREN(rack_misc),
- OID_AUTO, "sad_seg_per", CTLFLAG_RW,
- &sad_seg_size_per, 800,
- "Percentage of segment size needed in a sack 800 = 80.0?");
- SYSCTL_ADD_S32(&rack_sysctl_ctx,
- SYSCTL_CHILDREN(rack_misc),
OID_AUTO, "rxt_controls", CTLFLAG_RW,
&rack_rxt_controls, 0,
"Retransmit sending size controls (valid values 0, 1, 2 default=1)?");
@@ -1619,85 +1578,6 @@ rack_init_sysctls(void)
&rack_autosndbuf_inc, 20,
"What percentage should rack scale up its snd buffer by?");
-
- /* Sack Attacker detection stuff */
- SYSCTL_ADD_U32(&rack_sysctl_ctx,
- SYSCTL_CHILDREN(rack_attack),
- OID_AUTO, "merge_out", CTLFLAG_RW,
- &rack_merge_out_sacks_on_attack, 0,
- "Do we merge the sendmap when we decide we are being attacked?");
-
- SYSCTL_ADD_U32(&rack_sysctl_ctx,
- SYSCTL_CHILDREN(rack_attack),
- OID_AUTO, "detect_highsackratio", CTLFLAG_RW,
- &rack_highest_sack_thresh_seen, 0,
- "Highest sack to ack ratio seen");
- SYSCTL_ADD_U32(&rack_sysctl_ctx,
- SYSCTL_CHILDREN(rack_attack),
- OID_AUTO, "detect_highmoveratio", CTLFLAG_RW,
- &rack_highest_move_thresh_seen, 0,
- "Highest move to non-move ratio seen");
- rack_ack_total = counter_u64_alloc(M_WAITOK);
- SYSCTL_ADD_COUNTER_U64(&rack_sysctl_ctx,
- SYSCTL_CHILDREN(rack_attack),
- OID_AUTO, "acktotal", CTLFLAG_RD,
- &rack_ack_total,
- "Total number of Ack's");
- rack_express_sack = counter_u64_alloc(M_WAITOK);
- SYSCTL_ADD_COUNTER_U64(&rack_sysctl_ctx,
- SYSCTL_CHILDREN(rack_attack),
- OID_AUTO, "exp_sacktotal", CTLFLAG_RD,
- &rack_express_sack,
- "Total expresss number of Sack's");
- rack_sack_total = counter_u64_alloc(M_WAITOK);
- SYSCTL_ADD_COUNTER_U64(&rack_sysctl_ctx,
- SYSCTL_CHILDREN(rack_attack),
- OID_AUTO, "sacktotal", CTLFLAG_RD,
- &rack_sack_total,
- "Total number of SACKs");
- rack_move_none = counter_u64_alloc(M_WAITOK);
- SYSCTL_ADD_COUNTER_U64(&rack_sysctl_ctx,
- SYSCTL_CHILDREN(rack_attack),
- OID_AUTO, "move_none", CTLFLAG_RD,
- &rack_move_none,
- "Total number of SACK index reuse of positions under threshold");
- rack_move_some = counter_u64_alloc(M_WAITOK);
- SYSCTL_ADD_COUNTER_U64(&rack_sysctl_ctx,
- SYSCTL_CHILDREN(rack_attack),
- OID_AUTO, "move_some", CTLFLAG_RD,
- &rack_move_some,
- "Total number of SACK index reuse of positions over threshold");
- rack_sack_attacks_detected = counter_u64_alloc(M_WAITOK);
- SYSCTL_ADD_COUNTER_U64(&rack_sysctl_ctx,
- SYSCTL_CHILDREN(rack_attack),
- OID_AUTO, "attacks", CTLFLAG_RD,
- &rack_sack_attacks_detected,
- "Total number of SACK attackers that had sack disabled");
- rack_sack_attacks_reversed = counter_u64_alloc(M_WAITOK);
- SYSCTL_ADD_COUNTER_U64(&rack_sysctl_ctx,
- SYSCTL_CHILDREN(rack_attack),
- OID_AUTO, "reversed", CTLFLAG_RD,
- &rack_sack_attacks_reversed,
- "Total number of SACK attackers that were later determined false positive");
- rack_sack_attacks_suspect = counter_u64_alloc(M_WAITOK);
- SYSCTL_ADD_COUNTER_U64(&rack_sysctl_ctx,
- SYSCTL_CHILDREN(rack_attack),
- OID_AUTO, "suspect", CTLFLAG_RD,
- &rack_sack_attacks_suspect,
- "Total number of SACKs that triggered early detection");
-
- rack_sack_used_next_merge = counter_u64_alloc(M_WAITOK);
- SYSCTL_ADD_COUNTER_U64(&rack_sysctl_ctx,
- SYSCTL_CHILDREN(rack_attack),
- OID_AUTO, "nextmerge", CTLFLAG_RD,
- &rack_sack_used_next_merge,
- "Total number of times we used the next merge");
- rack_sack_used_prev_merge = counter_u64_alloc(M_WAITOK);
- SYSCTL_ADD_COUNTER_U64(&rack_sysctl_ctx,
- SYSCTL_CHILDREN(rack_attack),
- OID_AUTO, "prevmerge", CTLFLAG_RD,
- &rack_sack_used_prev_merge,
- "Total number of times we used the prev merge");
/* Counters */
rack_total_bytes = counter_u64_alloc(M_WAITOK);
SYSCTL_ADD_COUNTER_U64(&rack_sysctl_ctx,
@@ -1908,18 +1788,6 @@ rack_init_sysctls(void)
OID_AUTO, "sack_short", CTLFLAG_RD,
&rack_sack_proc_short,
"Total times we took shortcut for sack processing");
- rack_sack_skipped_acked = counter_u64_alloc(M_WAITOK);
- SYSCTL_ADD_COUNTER_U64(&rack_sysctl_ctx,
- SYSCTL_CHILDREN(rack_attack),
- OID_AUTO, "skipacked", CTLFLAG_RD,
- &rack_sack_skipped_acked,
- "Total number of times we skipped previously sacked");
- rack_sack_splits = counter_u64_alloc(M_WAITOK);
- SYSCTL_ADD_COUNTER_U64(&rack_sysctl_ctx,
- SYSCTL_CHILDREN(rack_attack),
- OID_AUTO, "ofsplit", CTLFLAG_RD,
- &rack_sack_splits,
- "Total number of times we did the old fashion tree split");
rack_input_idle_reduces = counter_u64_alloc(M_WAITOK);
SYSCTL_ADD_COUNTER_U64(&rack_sysctl_ctx,
SYSCTL_CHILDREN(rack_counters),
@@ -3319,16 +3187,6 @@ rack_counter_destroy(void)
counter_u64_free(rack_hw_pace_lost);
counter_u64_free(rack_non_fto_send);
counter_u64_free(rack_extended_rfo);
- counter_u64_free(rack_ack_total);
- counter_u64_free(rack_express_sack);
- counter_u64_free(rack_sack_total);
- counter_u64_free(rack_move_none);
- counter_u64_free(rack_move_some);
- counter_u64_free(rack_sack_attacks_detected);
- counter_u64_free(rack_sack_attacks_reversed);
- counter_u64_free(rack_sack_attacks_suspect);
- counter_u64_free(rack_sack_used_next_merge);
- counter_u64_free(rack_sack_used_prev_merge);
counter_u64_free(rack_tlp_tot);
counter_u64_free(rack_tlp_newdata);
counter_u64_free(rack_tlp_retran);
@@ -3351,8 +3209,6 @@ rack_counter_destroy(void)
counter_u64_free(rack_sack_proc_all);
counter_u64_free(rack_sack_proc_restart);
counter_u64_free(rack_sack_proc_short);
- counter_u64_free(rack_sack_skipped_acked);
- counter_u64_free(rack_sack_splits);
counter_u64_free(rack_input_idle_reduces);
counter_u64_free(rack_collapsed_win);
counter_u64_free(rack_collapsed_win_rxt);
@@ -4730,7 +4586,7 @@ rack_make_timely_judgement(struct tcp_rack *rack, uint32_t rtt, int32_t rtt_diff
return (timely_says);
}
-static __inline int
+static inline int
rack_in_gp_window(struct tcpcb *tp, struct rack_sendmap *rsm)
{
if (SEQ_GEQ(rsm->r_start, tp->gput_seq) &&
@@ -4767,7 +4623,7 @@ rack_in_gp_window(struct tcpcb *tp, struct rack_sendmap *rsm)
return (0);
}
-static __inline void
+static inline void
rack_mark_in_gp_win(struct tcpcb *tp, struct rack_sendmap *rsm)
{
@@ -4784,7 +4640,7 @@ rack_mark_in_gp_win(struct tcpcb *tp, struct rack_sendmap *rsm)
rsm->r_flags &= ~RACK_IN_GP_WIN;
}
-static __inline void
+static inline void
rack_clear_gp_marks(struct tcpcb *tp, struct tcp_rack *rack)
{
/* A GP measurement is ending, clear all marks on the send map*/
@@ -4802,7 +4658,7 @@ rack_clear_gp_marks(struct tcpcb *tp, struct tcp_rack *rack)
}
-static __inline void
+static inline void
rack_tend_gp_marks(struct tcpcb *tp, struct tcp_rack *rack)
{
struct rack_sendmap *rsm = NULL;
@@ -6864,6 +6720,18 @@ rack_mark_lost(struct tcpcb *tp,
}
}
+static inline void
+rack_mark_nolonger_lost(struct tcp_rack *rack, struct rack_sendmap *rsm)
+{
+ KASSERT((rack->r_ctl.rc_considered_lost >= (rsm->r_end - rsm->r_start)),
+ ("rsm:%p rack:%p rc_considered_lost goes negative", rsm, rack));
+ rsm->r_flags &= ~RACK_WAS_LOST;
+ if (rack->r_ctl.rc_considered_lost >= (rsm->r_end - rsm->r_start))
+ rack->r_ctl.rc_considered_lost -= rsm->r_end - rsm->r_start;
+ else
+ rack->r_ctl.rc_considered_lost = 0;
+}
+
/*
* RACK Timer, here we simply do logging and house keeping.
* the normal rack_output() function will call the
@@ -7005,7 +6873,7 @@ rack_setup_offset_for_rsm(struct tcp_rack *rack, struct rack_sendmap *src_rsm, s
rsm->orig_t_space = M_TRAILINGROOM(rsm->m);
}
-static __inline void
+static inline void
rack_clone_rsm(struct tcp_rack *rack, struct rack_sendmap *nrsm,
struct rack_sendmap *rsm, uint32_t start)
{
@@ -8130,13 +7998,7 @@ rack_update_rsm(struct tcpcb *tp, struct tcp_rack *rack,
* remove the lost desgination and reduce the
* bytes considered lost.
*/
- rsm->r_flags &= ~RACK_WAS_LOST;
- KASSERT((rack->r_ctl.rc_considered_lost >= (rsm->r_end - rsm->r_start)),
- ("rsm:%p rack:%p rc_considered_lost goes negative", rsm, rack));
- if (rack->r_ctl.rc_considered_lost >= (rsm->r_end - rsm->r_start))
- rack->r_ctl.rc_considered_lost -= rsm->r_end - rsm->r_start;
- else
- rack->r_ctl.rc_considered_lost = 0;
+ rack_mark_nolonger_lost(rack, rsm);
}
idx = rsm->r_rtr_cnt - 1;
rsm->r_tim_lastsent[idx] = ts;
@@ -9492,6 +9354,11 @@ do_rest_ofb:
if (rsm->r_flags & RACK_WAS_LOST) {
int my_chg;
+ /*
+ * Note here we do not use our rack_mark_nolonger_lost() function
+ * since we are moving our data pointer around and the
+ * ack'ed side is already not considered lost.
+ */
my_chg = (nrsm->r_end - nrsm->r_start);
KASSERT((rack->r_ctl.rc_considered_lost >= my_chg),
("rsm:%p rack:%p rc_considered_lost goes negative", rsm, rack));
@@ -9531,7 +9398,6 @@ do_rest_ofb:
goto out;
}
rack_log_map_chg(tp, rack, &stack_map, rsm, next, MAP_SACK_M1, end, __LINE__);
- counter_u64_add(rack_sack_used_next_merge, 1);
/* Postion for the next block */
start = next->r_end;
rsm = tqhash_next(rack->r_ctl.tqh, next);
@@ -9563,7 +9429,6 @@ do_rest_ofb:
*/
goto out;
}
- counter_u64_add(rack_sack_splits, 1);
rack_clone_rsm(rack, nrsm, rsm, start);
rsm->r_just_ret = 0;
#ifndef INVARIANTS
@@ -9585,7 +9450,6 @@ do_rest_ofb:
}
} else {
/* Already sacked this piece */
- counter_u64_add(rack_sack_skipped_acked, 1);
if (end == rsm->r_end) {
/* Done with block */
rsm = tqhash_next(rack->r_ctl.tqh, rsm);
@@ -9659,16 +9523,11 @@ do_rest_ofb:
changed += (rsm->r_end - rsm->r_start);
/* You get a count for acking a whole segment or more */
if (rsm->r_flags & RACK_WAS_LOST) {
- int my_chg;
-
- my_chg = (rsm->r_end - rsm->r_start);
- rsm->r_flags &= ~RACK_WAS_LOST;
- KASSERT((rack->r_ctl.rc_considered_lost >= my_chg),
- ("rsm:%p rack:%p rc_considered_lost goes negative", rsm, rack));
- if (my_chg <= rack->r_ctl.rc_considered_lost)
- rack->r_ctl.rc_considered_lost -= my_chg;
- else
- rack->r_ctl.rc_considered_lost = 0;
+ /*
+ * Here we can use the inline function since
+ * the rsm is truly marked lost and now no longer lost.
+ */
+ rack_mark_nolonger_lost(rack, rsm);
}
rack->r_ctl.rc_sacked += (rsm->r_end - rsm->r_start);
if (rsm->r_in_tmap) /* should be true */
@@ -9690,8 +9549,6 @@ do_rest_ofb:
rsm->r_in_tmap = 0;
}
rack_log_map_chg(tp, rack, NULL, rsm, NULL, MAP_SACK_M3, end, __LINE__);
- } else {
- counter_u64_add(rack_sack_skipped_acked, 1);
}
if (end == rsm->r_end) {
/* This block only - done, setup for next */
@@ -9851,6 +9708,10 @@ do_rest_ofb:
if (rsm->r_flags & RACK_WAS_LOST) {
int my_chg;
+ /*
+ * Note here we are using hookery again so we can't
+ * use our rack_mark_nolonger_lost() function.
+ */
my_chg = (nrsm->r_end - nrsm->r_start);
KASSERT((rack->r_ctl.rc_considered_lost >= my_chg),
("rsm:%p rack:%p rc_considered_lost goes negative", rsm, rack));
@@ -9866,7 +9727,6 @@ do_rest_ofb:
}
rack_log_map_chg(tp, rack, prev, &stack_map, rsm, MAP_SACK_M4, end, __LINE__);
rsm = prev;
- counter_u64_add(rack_sack_used_prev_merge, 1);
} else {
/**
* This is the case where our previous
@@ -9931,7 +9791,6 @@ do_rest_ofb:
* rsm |---| (acked)
* nrsm |------| (not acked)
*/
- counter_u64_add(rack_sack_splits, 1);
rack_clone_rsm(rack, nrsm, rsm, end);
rsm->r_flags &= (~RACK_HAS_FIN);
rsm->r_just_ret = 0;
@@ -9952,16 +9811,10 @@ do_rest_ofb:
rack_update_rtt(tp, rack, rsm, to, cts, SACKED, 0);
changed += (rsm->r_end - rsm->r_start);
if (rsm->r_flags & RACK_WAS_LOST) {
- int my_chg;
-
- my_chg = (rsm->r_end - rsm->r_start);
- rsm->r_flags &= ~RACK_WAS_LOST;
- KASSERT((rack->r_ctl.rc_considered_lost >= my_chg),
- ("rsm:%p rack:%p rc_considered_lost goes negative", rsm, rack));
- if (my_chg <= rack->r_ctl.rc_considered_lost)
- rack->r_ctl.rc_considered_lost -= my_chg;
- else
- rack->r_ctl.rc_considered_lost = 0;
+ /*
+ * Here it is safe to use our function.
+ */
+ rack_mark_nolonger_lost(rack, rsm);
}
rack->r_ctl.rc_sacked += (rsm->r_end - rsm->r_start);
@@ -9985,11 +9838,6 @@ do_rest_ofb:
rsm->r_in_tmap = 0;
}
}
- } else if (start != end){
- /*
- * The block was already acked.
- */
- counter_u64_add(rack_sack_skipped_acked, 1);
}
out:
if (rsm &&
@@ -10362,13 +10210,7 @@ more:
* and yet before retransmitting we get an ack
* which can happen due to reordering.
*/
- rsm->r_flags &= ~RACK_WAS_LOST;
- KASSERT((rack->r_ctl.rc_considered_lost >= (rsm->r_end - rsm->r_start)),
- ("rsm:%p rack:%p rc_considered_lost goes negative", rsm, rack));
- if (rack->r_ctl.rc_considered_lost >= (rsm->r_end - rsm->r_start))
- rack->r_ctl.rc_considered_lost -= rsm->r_end - rsm->r_start;
- else
- rack->r_ctl.rc_considered_lost = 0;
+ rack_mark_nolonger_lost(rack, rsm);
}
rack_log_map_chg(tp, rack, NULL, rsm, NULL, MAP_FREE, rsm->r_end, __LINE__);
rack->r_ctl.rc_holes_rxt -= rsm->r_rtr_bytes;
@@ -10476,12 +10318,7 @@ more:
* which can happen due to reordering. In this
* case its only a partial ack of the send.
*/
- KASSERT((rack->r_ctl.rc_considered_lost >= (th_ack - rsm->r_start)),
- ("rsm:%p rack:%p rc_considered_lost goes negative th_ack:%u", rsm, rack, th_ack));
- if (rack->r_ctl.rc_considered_lost >= (th_ack - rsm->r_start))
- rack->r_ctl.rc_considered_lost -= th_ack - rsm->r_start;
- else
- rack->r_ctl.rc_considered_lost = 0;
+ rack_mark_nolonger_lost(rack, rsm);
}
/*
* Clear the dup ack count for
@@ -10793,17 +10630,6 @@ rack_log_ack(struct tcpcb *tp, struct tcpopt *to, struct tcphdr *th, int entered
changed = 0;
th_ack = th->th_ack;
segsiz = ctf_fixed_maxseg(rack->rc_tp);
- if (BYTES_THIS_ACK(tp, th) >= segsiz) {
- /*
- * You only get credit for
- * MSS and greater (and you get extra
- * credit for larger cum-ack moves).
- */
- int ac;
-
- ac = BYTES_THIS_ACK(tp, th) / ctf_fixed_maxseg(rack->rc_tp);
- counter_u64_add(rack_ack_total, ac);
- }
if (SEQ_GT(th_ack, tp->snd_una)) {
rack_log_progress_event(rack, tp, ticks, PROGRESS_UPDATE, __LINE__);
tp->t_acktime = ticks;
@@ -10875,8 +10701,8 @@ rack_log_ack(struct tcpcb *tp, struct tcpopt *to, struct tcphdr *th, int entered
if (sacks_seen != NULL)
*sacks_seen = num_sack_blks;
if (num_sack_blks == 0) {
- /* Nothing to sack, but we need to update counts */
- goto out_with_totals;
+ /* Nothing to sack */
+ goto out;
}
/* Its a sack of some sort */
if (num_sack_blks < 2) {
@@ -10899,7 +10725,7 @@ rack_log_ack(struct tcpcb *tp, struct tcpopt *to, struct tcphdr *th, int entered
*/
again:
if (num_sack_blks == 0)
- goto out_with_totals;
+ goto out;
if (num_sack_blks > 1) {
for (i = 0; i < num_sack_blks; i++) {
for (j = i + 1; j < num_sack_blks; j++) {
@@ -10952,19 +10778,7 @@ do_sack_work:
changed += acked;
}
if (num_sack_blks == 1) {
- /*
- * This is what we would expect from
- * a normal implementation to happen
- * after we have retransmitted the FR,
- * i.e the sack-filter pushes down
- * to 1 block and the next to be retransmitted
- * is the sequence in the sack block (has more
- * are acked). Count this as ACK'd data to boost
- * up the chances of recovering any false positives.
- */
- counter_u64_add(rack_ack_total, (acked / ctf_fixed_maxseg(rack->rc_tp)));
- counter_u64_add(rack_express_sack, 1);
- goto out_with_totals;
+ goto out;
} else {
/*
* Start the loop through the
@@ -10973,7 +10787,6 @@ do_sack_work:
loop_start = 1;
}
}
- counter_u64_add(rack_sack_total, 1);
rsm = rack->r_ctl.rc_sacklast;
for (i = loop_start; i < num_sack_blks; i++) {
acked = rack_proc_sack_blk(tp, rack, &sack_blocks[i], to, &rsm, cts, segsiz);
@@ -10982,18 +10795,6 @@ do_sack_work:
changed += acked;
}
}
-out_with_totals:
- if (num_sack_blks > 1) {
- /*
- * You get an extra stroke if
- * you have more than one sack-blk, this
- * could be where we are skipping forward
- * and the sack-filter is still working, or
- * it could be an attacker constantly
- * moving us.
- */
- counter_u64_add(rack_move_some, 1);
- }
out:
if (changed) {
/* Something changed cancel the rack timer */
@@ -14713,7 +14514,6 @@ rack_init(struct tcpcb *tp, void **ptr)
rack->r_ctl.rc_prr_sendalot = rack_send_a_lot_in_prr;
rack->r_ctl.rc_min_to = rack_min_to;
microuptime(&rack->r_ctl.act_rcv_time);
- rack->r_ctl.rc_last_time_decay = rack->r_ctl.act_rcv_time;
rack->r_ctl.rack_per_of_gp_ss = rack_per_of_gp_ss;
if (rack_hw_up_only)
rack->r_up_only = 1;
diff --git a/sys/netinet/tcp_stacks/tcp_rack.h b/sys/netinet/tcp_stacks/tcp_rack.h
index 144b4fabf7eb..cac17d9aeb50 100644
--- a/sys/netinet/tcp_stacks/tcp_rack.h
+++ b/sys/netinet/tcp_stacks/tcp_rack.h
@@ -462,7 +462,6 @@ struct rack_control {
uint64_t rc_gp_output_ts; /* chg*/
uint64_t rc_gp_cumack_ts; /* chg*/
struct timeval act_rcv_time;
- struct timeval rc_last_time_decay; /* SAD time decay happened here */
uint64_t gp_bw;
uint64_t init_rate;
#ifdef NETFLIX_SHARED_CWND
diff --git a/sys/netinet/tcp_subr.c b/sys/netinet/tcp_subr.c
index c817c79881d6..b6f428b279b3 100644
--- a/sys/netinet/tcp_subr.c
+++ b/sys/netinet/tcp_subr.c
@@ -607,7 +607,7 @@ tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp,
}
}
m->m_pkthdr.tcp_tun_port = port = uh->uh_sport;
- bcopy(th, uh, m->m_len - off);
+ bcopy(th, uh, m->m_len - off - sizeof(struct udphdr));
m->m_len -= sizeof(struct udphdr);
m->m_pkthdr.len -= sizeof(struct udphdr);
/*
diff --git a/sys/netinet/tcp_syncache.c b/sys/netinet/tcp_syncache.c
index be20fb44a820..3a7755e9f09e 100644
--- a/sys/netinet/tcp_syncache.c
+++ b/sys/netinet/tcp_syncache.c
@@ -1168,7 +1168,7 @@ syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
/*
* If listening socket requested TCP digests, check that
* received ACK has signature and it is correct.
- * If not, drop the ACK and leave sc entry in th cache,
+ * If not, drop the ACK and leave sc entry in the cache,
* because SYN was received with correct signature.
*/
if (sc->sc_flags & SCF_SIGNATURE) {
@@ -1380,6 +1380,7 @@ syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
struct tcpcb *tp;
struct socket *rv = NULL;
struct syncache *sc = NULL;
+ struct ucred *cred;
struct syncache_head *sch;
struct mbuf *ipopts = NULL;
u_int ltflags;
@@ -1408,6 +1409,7 @@ syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
*/
KASSERT(SOLISTENING(so), ("%s: %p not listening", __func__, so));
tp = sototcpcb(so);
+ cred = V_tcp_syncache.see_other ? NULL : crhold(so->so_cred);
#ifdef INET6
if (inc->inc_flags & INC_ISIPV6) {
@@ -1636,16 +1638,16 @@ syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
/*
* sc_cred is only used in syncache_pcblist() to list TCP endpoints in
* TCPS_SYN_RECEIVED state when V_tcp_syncache.see_other is false.
- * Therefore, store the credentials and take a reference count only
- * when needed:
+ * Therefore, store the credentials only when needed:
* - sc is allocated from the zone and not using the on stack instance.
* - the sysctl variable net.inet.tcp.syncache.see_other is false.
* The reference count is decremented when a zone allocated sc is
* freed in syncache_free().
*/
- if (sc != &scs && !V_tcp_syncache.see_other)
- sc->sc_cred = crhold(so->so_cred);
- else
+ if (sc != &scs && !V_tcp_syncache.see_other) {
+ sc->sc_cred = cred;
+ cred = NULL;
+ } else
sc->sc_cred = NULL;
sc->sc_port = port;
sc->sc_ipopts = ipopts;
@@ -1783,6 +1785,8 @@ donenoprobe:
tcp_fastopen_decrement_counter(tfo_pending);
tfo_expanded:
+ if (cred != NULL)
+ crfree(cred);
if (sc == NULL || sc == &scs) {
#ifdef MAC
mac_syncache_destroy(&maclabel);
diff --git a/sys/netinet/tcp_timer.c b/sys/netinet/tcp_timer.c
index 3b9fe7a317b0..57c57666fa3a 100644
--- a/sys/netinet/tcp_timer.c
+++ b/sys/netinet/tcp_timer.c
@@ -513,9 +513,12 @@ tcp_timer_persist(struct tcpcb *tp)
if (progdrop || (tp->t_rxtshift >= V_tcp_retries &&
(ticks - tp->t_rcvtime >= tcp_maxpersistidle ||
ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff))) {
- if (!progdrop)
+ if (progdrop) {
+ tcp_log_end_status(tp, TCP_EI_STATUS_PROGRESS);
+ } else {
TCPSTAT_INC(tcps_persistdrop);
- tcp_log_end_status(tp, TCP_EI_STATUS_PERSIST_MAX);
+ tcp_log_end_status(tp, TCP_EI_STATUS_PERSIST_MAX);
+ }
goto dropit;
}
/*
diff --git a/sys/netinet/udp_usrreq.c b/sys/netinet/udp_usrreq.c
index cea8a916679b..f1d952037d5a 100644
--- a/sys/netinet/udp_usrreq.c
+++ b/sys/netinet/udp_usrreq.c
@@ -787,7 +787,8 @@ udplite_ctlinput(struct icmp *icmp)
static int
udp_pcblist(SYSCTL_HANDLER_ARGS)
{
- struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_udbinfo,
+ struct inpcbinfo *pcbinfo = udp_get_inpcbinfo(arg2);
+ struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo,
INPLOOKUP_RLOCKPCB);
struct xinpgen xig;
struct inpcb *inp;
@@ -799,7 +800,7 @@ udp_pcblist(SYSCTL_HANDLER_ARGS)
if (req->oldptr == 0) {
int n;
- n = V_udbinfo.ipi_count;
+ n = pcbinfo->ipi_count;
n += imax(n / 8, 10);
req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
return (0);
@@ -810,8 +811,8 @@ udp_pcblist(SYSCTL_HANDLER_ARGS)
bzero(&xig, sizeof(xig));
xig.xig_len = sizeof xig;
- xig.xig_count = V_udbinfo.ipi_count;
- xig.xig_gen = V_udbinfo.ipi_gencnt;
+ xig.xig_count = pcbinfo->ipi_count;
+ xig.xig_gen = pcbinfo->ipi_gencnt;
xig.xig_sogen = so_gencnt;
error = SYSCTL_OUT(req, &xig, sizeof xig);
if (error)
@@ -838,9 +839,9 @@ udp_pcblist(SYSCTL_HANDLER_ARGS)
* that something happened while we were processing this
* request, and it might be necessary to retry.
*/
- xig.xig_gen = V_udbinfo.ipi_gencnt;
+ xig.xig_gen = pcbinfo->ipi_gencnt;
xig.xig_sogen = so_gencnt;
- xig.xig_count = V_udbinfo.ipi_count;
+ xig.xig_count = pcbinfo->ipi_count;
error = SYSCTL_OUT(req, &xig, sizeof xig);
}
@@ -848,10 +849,15 @@ udp_pcblist(SYSCTL_HANDLER_ARGS)
}
SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
- CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
+ CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, IPPROTO_UDP,
udp_pcblist, "S,xinpcb",
"List of active UDP sockets");
+SYSCTL_PROC(_net_inet_udplite, OID_AUTO, pcblist,
+ CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, IPPROTO_UDPLITE,
+ udp_pcblist, "S,xinpcb",
+ "List of active UDP-Lite sockets");
+
#ifdef INET
static int
udp_getcred(SYSCTL_HANDLER_ARGS)
@@ -1166,7 +1172,19 @@ udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
else
INP_RLOCK(inp);
NET_EPOCH_ENTER(et);
+#ifdef INET6
+ if ((flags & PRUS_IPV6) != 0) {
+ if ((inp->in6p_outputopts != NULL) &&
+ (inp->in6p_outputopts->ip6po_tclass != -1))
+ tos = (u_char)inp->in6p_outputopts->ip6po_tclass;
+ else
+ tos = 0;
+ } else {
+ tos = inp->inp_ip_tos;
+ }
+#else
tos = inp->inp_ip_tos;
+#endif
if (control != NULL) {
/*
* XXX: Currently, we assume all the optional information is
@@ -1190,6 +1208,23 @@ udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
error = udp_v4mapped_pktinfo(cm, &src, inp, flags);
if (error != 0)
break;
+ if (((flags & PRUS_IPV6) != 0) &&
+ (cm->cmsg_level == IPPROTO_IPV6) &&
+ (cm->cmsg_type == IPV6_TCLASS)) {
+ int tclass;
+
+ if (cm->cmsg_len != CMSG_LEN(sizeof(int))) {
+ error = EINVAL;
+ break;
+ }
+ tclass = *(int *)CMSG_DATA(cm);
+ if (tclass < -1 || tclass > 255) {
+ error = EINVAL;
+ break;
+ }
+ if (tclass != -1)
+ tos = (u_char)tclass;
+ }
#endif
if (cm->cmsg_level != IPPROTO_IP)
continue;
diff --git a/sys/netinet/udp_var.h b/sys/netinet/udp_var.h
index 3895f365db3c..3ae08fc0b8f0 100644
--- a/sys/netinet/udp_var.h
+++ b/sys/netinet/udp_var.h
@@ -147,6 +147,7 @@ void kmod_udpstat_inc(int statnum);
} while (0)
SYSCTL_DECL(_net_inet_udp);
+SYSCTL_DECL(_net_inet_udplite);
VNET_DECLARE(struct inpcbinfo, udbinfo);
VNET_DECLARE(struct inpcbinfo, ulitecbinfo);
generated by cgit v1.2.3 (git 2.25.1) at 2025-10-31 06:32:49 +0000