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/tcp_stacks/bbr.c8
-rw-r--r--sys/netinet/tcp_stacks/rack.c83
-rw-r--r--sys/netinet/tcp_subr.c2
-rw-r--r--sys/netinet/tcp_syncache.c2
-rw-r--r--sys/netinet/udp_usrreq.c20
-rw-r--r--sys/netinet/udp_var.h1
8 files changed, 70 insertions, 61 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/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..50077abdfd86 100644
--- a/sys/netinet/tcp_stacks/rack.c
+++ b/sys/netinet/tcp_stacks/rack.c
@@ -4730,7 +4730,7 @@ 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 +4767,7 @@ 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 +4784,7 @@ 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 +4802,7 @@ 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 +6864,18 @@ 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 +7017,7 @@ 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 +8142,7 @@ 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 +9498,11 @@ 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));
@@ -9659,16 +9670,11 @@ 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 */
@@ -9851,6 +9857,10 @@ 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));
@@ -9952,16 +9962,10 @@ 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);
@@ -10362,13 +10366,7 @@ 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 +10474,7 @@ 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
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..3cb538f7054d 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) {
diff --git a/sys/netinet/udp_usrreq.c b/sys/netinet/udp_usrreq.c
index cea8a916679b..04d01099d54a 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)
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 2025-10-23 12:54:11 +0000