aboutsummaryrefslogtreecommitdiff
path: root/sys/net
diff options
context:
space:
mode:
Diffstat (limited to 'sys/net')
-rw-r--r--sys/net/if_bridgevar.h2
-rw-r--r--sys/net/if_clone.h2
-rw-r--r--sys/net/if_pfsync.h4
-rw-r--r--sys/net/if_tap.h2
-rw-r--r--sys/net/if_tun.h2
-rw-r--r--sys/net/if_tuntap.c88
-rw-r--r--sys/net/iflib.c92
-rw-r--r--sys/net/pfvar.h69
8 files changed, 172 insertions, 89 deletions
diff --git a/sys/net/if_bridgevar.h b/sys/net/if_bridgevar.h
index b0f579f688ac..5ed8c19f3128 100644
--- a/sys/net/if_bridgevar.h
+++ b/sys/net/if_bridgevar.h
@@ -159,7 +159,7 @@ struct ifbreq {
uint32_t ifbr_addrexceeded; /* member if addr violations */
ether_vlanid_t ifbr_pvid; /* member if PVID */
uint16_t ifbr_vlanproto; /* member if VLAN protocol */
- uint8_t pad[32];
+ uint8_t pad[28];
};
/* BRDGGIFFLAGS, BRDGSIFFLAGS */
diff --git a/sys/net/if_clone.h b/sys/net/if_clone.h
index 5a74ffa1cc2f..d780e49af25f 100644
--- a/sys/net/if_clone.h
+++ b/sys/net/if_clone.h
@@ -153,7 +153,7 @@ int if_clone_destroy(const char *);
int if_clone_list(struct if_clonereq *);
void if_clone_restoregroup(struct ifnet *);
-/* The below interfaces are used only by epair(4). */
+/* The below interfaces are used only by epair(4) and tun(4)/tap(4). */
void if_clone_addif(struct if_clone *, struct ifnet *);
int if_clone_destroyif(struct if_clone *, struct ifnet *);
diff --git a/sys/net/if_pfsync.h b/sys/net/if_pfsync.h
index 1efc220aa8e1..e99df0b85ccf 100644
--- a/sys/net/if_pfsync.h
+++ b/sys/net/if_pfsync.h
@@ -160,8 +160,8 @@ struct pfsync_ins_ack {
struct pfsync_upd_c {
u_int64_t id;
- struct pfsync_state_peer src;
- struct pfsync_state_peer dst;
+ struct pf_state_peer_export src;
+ struct pf_state_peer_export dst;
u_int32_t creatorid;
u_int32_t expire;
u_int8_t timeout;
diff --git a/sys/net/if_tap.h b/sys/net/if_tap.h
index d84cd2eba6f3..8297b8d9e3d2 100644
--- a/sys/net/if_tap.h
+++ b/sys/net/if_tap.h
@@ -57,6 +57,8 @@
#define TAPGIFNAME TUNGIFNAME
#define TAPSVNETHDR _IOW('t', 91, int)
#define TAPGVNETHDR _IOR('t', 94, int)
+#define TAPSTRANSIENT TUNSTRANSIENT
+#define TAPGTRANSIENT TUNGTRANSIENT
/* VMware ioctl's */
#define VMIO_SIOCSIFFLAGS _IOWINT('V', 0)
diff --git a/sys/net/if_tun.h b/sys/net/if_tun.h
index a8fb61db45a2..ccdc25944823 100644
--- a/sys/net/if_tun.h
+++ b/sys/net/if_tun.h
@@ -43,5 +43,7 @@ struct tuninfo {
#define TUNSIFPID _IO('t', 95)
#define TUNSIFHEAD _IOW('t', 96, int)
#define TUNGIFHEAD _IOR('t', 97, int)
+#define TUNSTRANSIENT _IOW('t', 98, int)
+#define TUNGTRANSIENT _IOR('t', 99, int)
#endif /* !_NET_IF_TUN_H_ */
diff --git a/sys/net/if_tuntap.c b/sys/net/if_tuntap.c
index 5e6f65c04b2f..c8dbb6aa8893 100644
--- a/sys/net/if_tuntap.c
+++ b/sys/net/if_tuntap.c
@@ -132,6 +132,7 @@ struct tuntap_softc {
#define TUN_DYING 0x0200
#define TUN_L2 0x0400
#define TUN_VMNET 0x0800
+#define TUN_TRANSIENT 0x1000
#define TUN_DRIVER_IDENT_MASK (TUN_L2 | TUN_VMNET)
#define TUN_READY (TUN_OPEN | TUN_INITED)
@@ -443,6 +444,18 @@ tuntap_name2info(const char *name, int *outunit, int *outflags)
return (0);
}
+static struct if_clone *
+tuntap_cloner_from_flags(int tun_flags)
+{
+
+ for (u_int i = 0; i < NDRV; i++)
+ if ((tun_flags & TUN_DRIVER_IDENT_MASK) ==
+ tuntap_drivers[i].ident_flags)
+ return (V_tuntap_driver_cloners[i]);
+
+ return (NULL);
+}
+
/*
* Get driver information from a set of flags specified. Masks the identifying
* part of the flags and compares it against all of the available
@@ -615,19 +628,39 @@ out:
CURVNET_RESTORE();
}
-static void
-tun_destroy(struct tuntap_softc *tp)
+static int
+tun_destroy(struct tuntap_softc *tp, bool may_intr)
{
+ int error;
TUN_LOCK(tp);
+
+ /*
+ * Transient tunnels may have set TUN_DYING if we're being destroyed as
+ * a result of the last close, which we'll allow.
+ */
+ MPASS((tp->tun_flags & (TUN_DYING | TUN_TRANSIENT)) != TUN_DYING);
tp->tun_flags |= TUN_DYING;
- if (tp->tun_busy != 0)
- cv_wait_unlock(&tp->tun_cv, &tp->tun_mtx);
- else
- TUN_UNLOCK(tp);
+ error = 0;
+ while (tp->tun_busy != 0) {
+ if (may_intr)
+ error = cv_wait_sig(&tp->tun_cv, &tp->tun_mtx);
+ else
+ cv_wait(&tp->tun_cv, &tp->tun_mtx);
+ if (error != 0) {
+ tp->tun_flags &= ~TUN_DYING;
+ TUN_UNLOCK(tp);
+ return (error);
+ }
+ }
+ TUN_UNLOCK(tp);
CURVNET_SET(TUN2IFP(tp)->if_vnet);
+ mtx_lock(&tunmtx);
+ TAILQ_REMOVE(&tunhead, tp, tun_list);
+ mtx_unlock(&tunmtx);
+
/* destroy_dev will take care of any alias. */
destroy_dev(tp->tun_dev);
seldrain(&tp->tun_rsel);
@@ -648,6 +681,8 @@ tun_destroy(struct tuntap_softc *tp)
cv_destroy(&tp->tun_cv);
free(tp, M_TUN);
CURVNET_RESTORE();
+
+ return (0);
}
static int
@@ -655,12 +690,7 @@ tun_clone_destroy(struct if_clone *ifc __unused, struct ifnet *ifp, uint32_t fla
{
struct tuntap_softc *tp = ifp->if_softc;
- mtx_lock(&tunmtx);
- TAILQ_REMOVE(&tunhead, tp, tun_list);
- mtx_unlock(&tunmtx);
- tun_destroy(tp);
-
- return (0);
+ return (tun_destroy(tp, true));
}
static void
@@ -702,9 +732,9 @@ tun_uninit(const void *unused __unused)
mtx_lock(&tunmtx);
while ((tp = TAILQ_FIRST(&tunhead)) != NULL) {
- TAILQ_REMOVE(&tunhead, tp, tun_list);
mtx_unlock(&tunmtx);
- tun_destroy(tp);
+ /* tun_destroy() will remove it from the tailq. */
+ tun_destroy(tp, false);
mtx_lock(&tunmtx);
}
mtx_unlock(&tunmtx);
@@ -1217,6 +1247,23 @@ out:
tun_vnethdr_set(ifp, 0);
tun_unbusy_locked(tp);
+ if ((tp->tun_flags & TUN_TRANSIENT) != 0) {
+ struct if_clone *cloner;
+ int error __diagused;
+
+ /* Mark it busy so that nothing can re-open it. */
+ tp->tun_flags |= TUN_DYING;
+ TUN_UNLOCK(tp);
+
+ CURVNET_SET_QUIET(ifp->if_home_vnet);
+ cloner = tuntap_cloner_from_flags(tp->tun_flags);
+ CURVNET_RESTORE();
+
+ error = if_clone_destroyif(cloner, ifp);
+ MPASS(error == 0 || error == EINTR || error == ERESTART);
+ return;
+ }
+
TUN_UNLOCK(tp);
}
@@ -1668,6 +1715,19 @@ tunioctl(struct cdev *dev, u_long cmd, caddr_t data, int flag,
case TUNGDEBUG:
*(int *)data = tundebug;
break;
+ case TUNSTRANSIENT:
+ TUN_LOCK(tp);
+ if (*(int *)data)
+ tp->tun_flags |= TUN_TRANSIENT;
+ else
+ tp->tun_flags &= ~TUN_TRANSIENT;
+ TUN_UNLOCK(tp);
+ break;
+ case TUNGTRANSIENT:
+ TUN_LOCK(tp);
+ *(int *)data = (tp->tun_flags & TUN_TRANSIENT) != 0;
+ TUN_UNLOCK(tp);
+ break;
case FIONBIO:
break;
case FIOASYNC:
diff --git a/sys/net/iflib.c b/sys/net/iflib.c
index 2b43f6f19051..98c59e5de988 100644
--- a/sys/net/iflib.c
+++ b/sys/net/iflib.c
@@ -142,6 +142,7 @@ struct iflib_ctx;
static void iru_init(if_rxd_update_t iru, iflib_rxq_t rxq, uint8_t flid);
static void iflib_timer(void *arg);
static void iflib_tqg_detach(if_ctx_t ctx);
+static int iflib_simple_transmit(if_t ifp, struct mbuf *m);
typedef struct iflib_filter_info {
driver_filter_t *ifi_filter;
@@ -198,6 +199,7 @@ struct iflib_ctx {
uint8_t ifc_sysctl_use_logical_cores;
uint16_t ifc_sysctl_extra_msix_vectors;
bool ifc_cpus_are_physical_cores;
+ bool ifc_sysctl_simple_tx;
qidx_t ifc_sysctl_ntxds[8];
qidx_t ifc_sysctl_nrxds[8];
@@ -725,6 +727,7 @@ static void iflib_free_intr_mem(if_ctx_t ctx);
#ifndef __NO_STRICT_ALIGNMENT
static struct mbuf *iflib_fixup_rx(struct mbuf *m);
#endif
+static __inline int iflib_completed_tx_reclaim(iflib_txq_t txq, int thresh);
static SLIST_HEAD(cpu_offset_list, cpu_offset) cpu_offsets =
SLIST_HEAD_INITIALIZER(cpu_offsets);
@@ -2624,8 +2627,10 @@ iflib_stop(if_ctx_t ctx)
#endif /* DEV_NETMAP */
CALLOUT_UNLOCK(txq);
- /* clean any enqueued buffers */
- iflib_ifmp_purge(txq);
+ if (!ctx->ifc_sysctl_simple_tx) {
+ /* clean any enqueued buffers */
+ iflib_ifmp_purge(txq);
+ }
/* Free any existing tx buffers. */
for (j = 0; j < txq->ift_size; j++) {
iflib_txsd_free(ctx, txq, j);
@@ -3635,13 +3640,16 @@ defrag:
* cxgb
*/
if (__predict_false(nsegs + 2 > TXQ_AVAIL(txq))) {
- txq->ift_no_desc_avail++;
- bus_dmamap_unload(buf_tag, map);
- DBG_COUNTER_INC(encap_txq_avail_fail);
- DBG_COUNTER_INC(encap_txd_encap_fail);
- if ((txq->ift_task.gt_task.ta_flags & TASK_ENQUEUED) == 0)
- GROUPTASK_ENQUEUE(&txq->ift_task);
- return (ENOBUFS);
+ (void)iflib_completed_tx_reclaim(txq, RECLAIM_THRESH(ctx));
+ if (__predict_false(nsegs + 2 > TXQ_AVAIL(txq))) {
+ txq->ift_no_desc_avail++;
+ bus_dmamap_unload(buf_tag, map);
+ DBG_COUNTER_INC(encap_txq_avail_fail);
+ DBG_COUNTER_INC(encap_txd_encap_fail);
+ if ((txq->ift_task.gt_task.ta_flags & TASK_ENQUEUED) == 0)
+ GROUPTASK_ENQUEUE(&txq->ift_task);
+ return (ENOBUFS);
+ }
}
/*
* On Intel cards we can greatly reduce the number of TX interrupts
@@ -4014,6 +4022,12 @@ _task_fn_tx(void *context)
netmap_tx_irq(ifp, txq->ift_id))
goto skip_ifmp;
#endif
+ if (ctx->ifc_sysctl_simple_tx) {
+ mtx_lock(&txq->ift_mtx);
+ (void)iflib_completed_tx_reclaim(txq, RECLAIM_THRESH(ctx));
+ mtx_unlock(&txq->ift_mtx);
+ goto skip_ifmp;
+ }
#ifdef ALTQ
if (if_altq_is_enabled(ifp))
iflib_altq_if_start(ifp);
@@ -4027,9 +4041,8 @@ _task_fn_tx(void *context)
*/
if (abdicate)
ifmp_ring_check_drainage(txq->ift_br, TX_BATCH_SIZE);
-#ifdef DEV_NETMAP
+
skip_ifmp:
-#endif
if (ctx->ifc_flags & IFC_LEGACY)
IFDI_INTR_ENABLE(ctx);
else
@@ -5131,7 +5144,14 @@ iflib_device_register(device_t dev, void *sc, if_shared_ctx_t sctx, if_ctx_t *ct
scctx = &ctx->ifc_softc_ctx;
ifp = ctx->ifc_ifp;
-
+ if (ctx->ifc_sysctl_simple_tx) {
+#ifndef ALTQ
+ if_settransmitfn(ifp, iflib_simple_transmit);
+ device_printf(dev, "using simple if_transmit\n");
+#else
+ device_printf(dev, "ALTQ prevents using simple if_transmit\n");
+#endif
+ }
iflib_reset_qvalues(ctx);
IFNET_WLOCK();
CTX_LOCK(ctx);
@@ -6766,6 +6786,9 @@ iflib_add_device_sysctl_pre(if_ctx_t ctx)
SYSCTL_ADD_CONST_STRING(ctx_list, oid_list, OID_AUTO, "driver_version",
CTLFLAG_RD, ctx->ifc_sctx->isc_driver_version, "driver version");
+ SYSCTL_ADD_BOOL(ctx_list, oid_list, OID_AUTO, "simple_tx",
+ CTLFLAG_RDTUN, &ctx->ifc_sysctl_simple_tx, 0,
+ "use simple tx ring");
SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "override_ntxqs",
CTLFLAG_RWTUN, &ctx->ifc_sysctl_ntxqs, 0,
"# of txqs to use, 0 => use default #");
@@ -7088,3 +7111,48 @@ iflib_debugnet_poll(if_t ifp, int count)
return (0);
}
#endif /* DEBUGNET */
+
+
+static inline iflib_txq_t
+iflib_simple_select_queue(if_ctx_t ctx, struct mbuf *m)
+{
+ int qidx;
+
+ if ((NTXQSETS(ctx) > 1) && M_HASHTYPE_GET(m))
+ qidx = QIDX(ctx, m);
+ else
+ qidx = NTXQSETS(ctx) + FIRST_QSET(ctx) - 1;
+ return (&ctx->ifc_txqs[qidx]);
+}
+
+static int
+iflib_simple_transmit(if_t ifp, struct mbuf *m)
+{
+ if_ctx_t ctx;
+ iflib_txq_t txq;
+ int error;
+ int bytes_sent = 0, pkt_sent = 0, mcast_sent = 0;
+
+
+ ctx = if_getsoftc(ifp);
+ if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
+ IFF_DRV_RUNNING)
+ return (EBUSY);
+ txq = iflib_simple_select_queue(ctx, m);
+ mtx_lock(&txq->ift_mtx);
+ error = iflib_encap(txq, &m);
+ if (error == 0) {
+ pkt_sent++;
+ bytes_sent += m->m_pkthdr.len;
+ mcast_sent += !!(m->m_flags & M_MCAST);
+ (void)iflib_txd_db_check(txq, true);
+ }
+ (void)iflib_completed_tx_reclaim(txq, RECLAIM_THRESH(ctx));
+ mtx_unlock(&txq->ift_mtx);
+ if_inc_counter(ifp, IFCOUNTER_OBYTES, bytes_sent);
+ if_inc_counter(ifp, IFCOUNTER_OPACKETS, pkt_sent);
+ if (mcast_sent)
+ if_inc_counter(ifp, IFCOUNTER_OMCASTS, mcast_sent);
+
+ return (error);
+}
diff --git a/sys/net/pfvar.h b/sys/net/pfvar.h
index c397f0b67896..d6c13470f2eb 100644
--- a/sys/net/pfvar.h
+++ b/sys/net/pfvar.h
@@ -1020,7 +1020,7 @@ struct pf_state_scrub_export {
#define PF_SCRUB_FLAG_VALID 0x01
uint8_t scrub_flag;
uint32_t pfss_ts_mod; /* timestamp modulation */
-};
+} __packed;
struct pf_state_key_export {
struct pf_addr addr[2];
@@ -1037,7 +1037,7 @@ struct pf_state_peer_export {
uint8_t state; /* active state level */
uint8_t wscale; /* window scaling factor */
uint8_t dummy[6];
-};
+} __packed;
_Static_assert(sizeof(struct pf_state_peer_export) == 32, "size incorrect");
struct pf_state_export {
@@ -1179,26 +1179,6 @@ struct pf_test_ctx {
* Unified state structures for pulling states out of the kernel
* used by pfsync(4) and the pf(4) ioctl.
*/
-struct pfsync_state_scrub {
- u_int16_t pfss_flags;
- u_int8_t pfss_ttl; /* stashed TTL */
-#define PFSYNC_SCRUB_FLAG_VALID 0x01
- u_int8_t scrub_flag;
- u_int32_t pfss_ts_mod; /* timestamp modulation */
-} __packed;
-
-struct pfsync_state_peer {
- struct pfsync_state_scrub scrub; /* state is scrubbed */
- u_int32_t seqlo; /* Max sequence number sent */
- u_int32_t seqhi; /* Max the other end ACKd + win */
- u_int32_t seqdiff; /* Sequence number modulator */
- u_int16_t max_win; /* largest window (pre scaling) */
- u_int16_t mss; /* Maximum segment size option */
- u_int8_t state; /* active state level */
- u_int8_t wscale; /* window scaling factor */
- u_int8_t pad[6];
-} __packed;
-
struct pfsync_state_key {
struct pf_addr addr[2];
u_int16_t port[2];
@@ -1208,8 +1188,8 @@ struct pfsync_state_1301 {
u_int64_t id;
char ifname[IFNAMSIZ];
struct pfsync_state_key key[2];
- struct pfsync_state_peer src;
- struct pfsync_state_peer dst;
+ struct pf_state_peer_export src;
+ struct pf_state_peer_export dst;
struct pf_addr rt_addr;
u_int32_t rule;
u_int32_t anchor;
@@ -1235,8 +1215,8 @@ struct pfsync_state_1400 {
u_int64_t id;
char ifname[IFNAMSIZ];
struct pfsync_state_key key[2];
- struct pfsync_state_peer src;
- struct pfsync_state_peer dst;
+ struct pf_state_peer_export src;
+ struct pf_state_peer_export dst;
struct pf_addr rt_addr;
u_int32_t rule;
u_int32_t anchor;
@@ -1323,39 +1303,10 @@ extern pflog_packet_t *pflog_packet_ptr;
/* for copies to/from network byte order */
/* ioctl interface also uses network byte order */
-#define pf_state_peer_hton(s,d) do { \
- (d)->seqlo = htonl((s)->seqlo); \
- (d)->seqhi = htonl((s)->seqhi); \
- (d)->seqdiff = htonl((s)->seqdiff); \
- (d)->max_win = htons((s)->max_win); \
- (d)->mss = htons((s)->mss); \
- (d)->state = (s)->state; \
- (d)->wscale = (s)->wscale; \
- if ((s)->scrub) { \
- (d)->scrub.pfss_flags = \
- htons((s)->scrub->pfss_flags & PFSS_TIMESTAMP); \
- (d)->scrub.pfss_ttl = (s)->scrub->pfss_ttl; \
- (d)->scrub.pfss_ts_mod = htonl((s)->scrub->pfss_ts_mod);\
- (d)->scrub.scrub_flag = PFSYNC_SCRUB_FLAG_VALID; \
- } \
-} while (0)
-
-#define pf_state_peer_ntoh(s,d) do { \
- (d)->seqlo = ntohl((s)->seqlo); \
- (d)->seqhi = ntohl((s)->seqhi); \
- (d)->seqdiff = ntohl((s)->seqdiff); \
- (d)->max_win = ntohs((s)->max_win); \
- (d)->mss = ntohs((s)->mss); \
- (d)->state = (s)->state; \
- (d)->wscale = (s)->wscale; \
- if ((s)->scrub.scrub_flag == PFSYNC_SCRUB_FLAG_VALID && \
- (d)->scrub != NULL) { \
- (d)->scrub->pfss_flags = \
- ntohs((s)->scrub.pfss_flags) & PFSS_TIMESTAMP; \
- (d)->scrub->pfss_ttl = (s)->scrub.pfss_ttl; \
- (d)->scrub->pfss_ts_mod = ntohl((s)->scrub.pfss_ts_mod);\
- } \
-} while (0)
+void pf_state_peer_hton(const struct pf_state_peer *,
+ struct pf_state_peer_export *);
+void pf_state_peer_ntoh(const struct pf_state_peer_export *,
+ struct pf_state_peer *);
#define pf_state_counter_hton(s,d) do { \
d[0] = htonl((s>>32)&0xffffffff); \
generated by cgit v1.2.3 (git 2.25.1) at 2025-08-23 17:04:31 +0000