1 /* $NetBSD: if_tun.c,v 1.14 1994/06/29 06:36:25 cgd Exp $ */
3 * SPDX-License-Identifier: BSD-2-Clause
5 * Copyright (C) 1999-2000 by Maksim Yevmenkin <m_evmenkin@yahoo.com>
7 * Copyright (c) 2019 Kyle Evans <kevans@FreeBSD.org>
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * -------------------------------------------------------------------------
34 * Copyright (c) 1988, Julian Onions <jpo@cs.nott.ac.uk>
35 * Nottingham University 1987.
37 * This source may be freely distributed, however I would be interested
38 * in any changes that are made.
40 * This driver takes packets off the IP i/f and hands them up to a
41 * user process to have its wicked way with. This driver has it's
42 * roots in a similar driver written by Phil Cockcroft (formerly) at
43 * UCL. This driver is based much more on read/write/poll mode of
48 #include "opt_inet6.h"
50 #include <sys/param.h>
54 #include <sys/systm.h>
57 #include <sys/module.h>
58 #include <sys/socket.h>
59 #include <sys/eventhandler.h>
60 #include <sys/fcntl.h>
61 #include <sys/filio.h>
62 #include <sys/sockio.h>
64 #include <sys/syslog.h>
65 #include <sys/ttycom.h>
67 #include <sys/selinfo.h>
68 #include <sys/signalvar.h>
69 #include <sys/filedesc.h>
70 #include <sys/kernel.h>
71 #include <sys/sysctl.h>
74 #include <sys/malloc.h>
75 #include <sys/random.h>
76 #include <sys/ctype.h>
78 #include <net/ethernet.h>
80 #include <net/if_var.h>
81 #include <net/if_clone.h>
82 #include <net/if_dl.h>
83 #include <net/if_media.h>
84 #include <net/if_private.h>
85 #include <net/if_types.h>
86 #include <net/if_vlan_var.h>
87 #include <net/netisr.h>
88 #include <net/route.h>
90 #include <netinet/in.h>
92 #include <netinet/ip.h>
95 #include <netinet/ip6.h>
96 #include <netinet6/ip6_var.h>
98 #include <netinet/udp.h>
99 #include <netinet/tcp.h>
100 #include <netinet/tcp_lro.h>
102 #include <net/if_tap.h>
103 #include <net/if_tun.h>
105 #include <dev/virtio/network/virtio_net.h>
107 #include <sys/queue.h>
108 #include <sys/condvar.h>
109 #include <security/mac/mac_framework.h>
111 struct tuntap_driver;
114 * tun_list is protected by global tunmtx. Other mutable fields are
115 * protected by tun->tun_mtx, or by their owning subsystem. tun_dev is
116 * static for the duration of a tunnel interface.
118 struct tuntap_softc {
119 TAILQ_ENTRY(tuntap_softc) tun_list;
120 struct cdev *tun_alias;
121 struct cdev *tun_dev;
122 u_short tun_flags; /* misc flags */
123 #define TUN_OPEN 0x0001
124 #define TUN_INITED 0x0002
125 #define TUN_UNUSED1 0x0008
126 #define TUN_UNUSED2 0x0010
127 #define TUN_LMODE 0x0020
128 #define TUN_RWAIT 0x0040
129 #define TUN_ASYNC 0x0080
130 #define TUN_IFHEAD 0x0100
131 #define TUN_DYING 0x0200
132 #define TUN_L2 0x0400
133 #define TUN_VMNET 0x0800
135 #define TUN_DRIVER_IDENT_MASK (TUN_L2 | TUN_VMNET)
136 #define TUN_READY (TUN_OPEN | TUN_INITED)
138 pid_t tun_pid; /* owning pid */
139 struct ifnet *tun_ifp; /* the interface */
140 struct sigio *tun_sigio; /* async I/O info */
141 struct tuntap_driver *tun_drv; /* appropriate driver */
142 struct selinfo tun_rsel; /* read select */
143 struct mtx tun_mtx; /* softc field mutex */
144 struct cv tun_cv; /* for ref'd dev destroy */
145 struct ether_addr tun_ether; /* remote address */
146 int tun_busy; /* busy count */
147 int tun_vhdrlen; /* virtio-net header length */
148 struct lro_ctrl tun_lro; /* for TCP LRO */
149 bool tun_lro_ready; /* TCP LRO initialized */
151 #define TUN2IFP(sc) ((sc)->tun_ifp)
153 #define TUNDEBUG if (tundebug) if_printf
155 #define TUN_LOCK(tp) mtx_lock(&(tp)->tun_mtx)
156 #define TUN_UNLOCK(tp) mtx_unlock(&(tp)->tun_mtx)
157 #define TUN_LOCK_ASSERT(tp) mtx_assert(&(tp)->tun_mtx, MA_OWNED);
159 #define TUN_VMIO_FLAG_MASK 0x0fff
162 * Interface capabilities of a tap device that supports the virtio-net
165 #define TAP_VNET_HDR_CAPS (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6 \
166 | IFCAP_VLAN_HWCSUM \
167 | IFCAP_TSO | IFCAP_LRO \
170 #define TAP_ALL_OFFLOAD (CSUM_TSO | CSUM_TCP | CSUM_UDP |\
171 CSUM_TCP_IPV6 | CSUM_UDP_IPV6)
174 * All mutable global variables in if_tun are locked using tunmtx, with
175 * the exception of tundebug, which is used unlocked, and the drivers' *clones,
176 * which are static after setup.
178 static struct mtx tunmtx;
179 static eventhandler_tag arrival_tag;
180 static eventhandler_tag clone_tag;
181 static const char tunname[] = "tun";
182 static const char tapname[] = "tap";
183 static const char vmnetname[] = "vmnet";
184 static MALLOC_DEFINE(M_TUN, tunname, "Tunnel Interface");
185 static int tundebug = 0;
186 static int tundclone = 1;
187 static int tap_allow_uopen = 0; /* allow user devfs cloning */
188 static int tapuponopen = 0; /* IFF_UP on open() */
189 static int tapdclone = 1; /* enable devfs cloning */
191 static TAILQ_HEAD(,tuntap_softc) tunhead = TAILQ_HEAD_INITIALIZER(tunhead);
192 SYSCTL_INT(_debug, OID_AUTO, if_tun_debug, CTLFLAG_RW, &tundebug, 0, "");
194 static struct sx tun_ioctl_sx;
195 SX_SYSINIT(tun_ioctl_sx, &tun_ioctl_sx, "tun_ioctl");
197 SYSCTL_DECL(_net_link);
199 static SYSCTL_NODE(_net_link, OID_AUTO, tun, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
200 "IP tunnel software network interface");
201 SYSCTL_INT(_net_link_tun, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tundclone, 0,
202 "Enable legacy devfs interface creation");
205 static SYSCTL_NODE(_net_link, OID_AUTO, tap, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
206 "Ethernet tunnel software network interface");
207 SYSCTL_INT(_net_link_tap, OID_AUTO, user_open, CTLFLAG_RW, &tap_allow_uopen, 0,
208 "Enable legacy devfs interface creation for all users");
209 SYSCTL_INT(_net_link_tap, OID_AUTO, up_on_open, CTLFLAG_RW, &tapuponopen, 0,
210 "Bring interface up when /dev/tap is opened");
211 SYSCTL_INT(_net_link_tap, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tapdclone, 0,
212 "Enable legacy devfs interface creation");
213 SYSCTL_INT(_net_link_tap, OID_AUTO, debug, CTLFLAG_RW, &tundebug, 0, "");
215 static int tun_create_device(struct tuntap_driver *drv, int unit,
216 struct ucred *cr, struct cdev **dev, const char *name);
217 static int tun_busy_locked(struct tuntap_softc *tp);
218 static void tun_unbusy_locked(struct tuntap_softc *tp);
219 static int tun_busy(struct tuntap_softc *tp);
220 static void tun_unbusy(struct tuntap_softc *tp);
222 static int tuntap_name2info(const char *name, int *unit, int *flags);
223 static void tunclone(void *arg, struct ucred *cred, char *name,
224 int namelen, struct cdev **dev);
225 static void tuncreate(struct cdev *dev);
226 static void tundtor(void *data);
227 static void tunrename(void *arg, struct ifnet *ifp);
228 static int tunifioctl(struct ifnet *, u_long, caddr_t);
229 static void tuninit(struct ifnet *);
230 static void tunifinit(void *xtp);
231 static int tuntapmodevent(module_t, int, void *);
232 static int tunoutput(struct ifnet *, struct mbuf *,
233 const struct sockaddr *, struct route *ro);
234 static void tunstart(struct ifnet *);
235 static void tunstart_l2(struct ifnet *);
237 static int tun_clone_match(struct if_clone *ifc, const char *name);
238 static int tap_clone_match(struct if_clone *ifc, const char *name);
239 static int vmnet_clone_match(struct if_clone *ifc, const char *name);
240 static int tun_clone_create(struct if_clone *, char *, size_t,
241 struct ifc_data *, struct ifnet **);
242 static int tun_clone_destroy(struct if_clone *, struct ifnet *, uint32_t);
243 static void tun_vnethdr_set(struct ifnet *ifp, int vhdrlen);
245 static d_open_t tunopen;
246 static d_read_t tunread;
247 static d_write_t tunwrite;
248 static d_ioctl_t tunioctl;
249 static d_poll_t tunpoll;
250 static d_kqfilter_t tunkqfilter;
252 static int tunkqread(struct knote *, long);
253 static int tunkqwrite(struct knote *, long);
254 static void tunkqdetach(struct knote *);
256 static struct filterops tun_read_filterops = {
259 .f_detach = tunkqdetach,
260 .f_event = tunkqread,
263 static struct filterops tun_write_filterops = {
266 .f_detach = tunkqdetach,
267 .f_event = tunkqwrite,
270 static struct tuntap_driver {
271 struct cdevsw cdevsw;
273 struct unrhdr *unrhdr;
274 struct clonedevs *clones;
275 ifc_match_f *clone_match_fn;
276 ifc_create_f *clone_create_fn;
277 ifc_destroy_f *clone_destroy_fn;
278 } tuntap_drivers[] = {
282 .d_version = D_VERSION,
283 .d_flags = D_NEEDMINOR,
289 .d_kqfilter = tunkqfilter,
292 .clone_match_fn = tun_clone_match,
293 .clone_create_fn = tun_clone_create,
294 .clone_destroy_fn = tun_clone_destroy,
297 .ident_flags = TUN_L2,
299 .d_version = D_VERSION,
300 .d_flags = D_NEEDMINOR,
306 .d_kqfilter = tunkqfilter,
309 .clone_match_fn = tap_clone_match,
310 .clone_create_fn = tun_clone_create,
311 .clone_destroy_fn = tun_clone_destroy,
314 .ident_flags = TUN_L2 | TUN_VMNET,
316 .d_version = D_VERSION,
317 .d_flags = D_NEEDMINOR,
323 .d_kqfilter = tunkqfilter,
326 .clone_match_fn = vmnet_clone_match,
327 .clone_create_fn = tun_clone_create,
328 .clone_destroy_fn = tun_clone_destroy,
332 struct tuntap_driver_cloner {
333 SLIST_ENTRY(tuntap_driver_cloner) link;
334 struct tuntap_driver *drv;
335 struct if_clone *cloner;
338 VNET_DEFINE_STATIC(SLIST_HEAD(, tuntap_driver_cloner), tuntap_driver_cloners) =
339 SLIST_HEAD_INITIALIZER(tuntap_driver_cloners);
341 #define V_tuntap_driver_cloners VNET(tuntap_driver_cloners)
344 * Mechanism for marking a tunnel device as busy so that we can safely do some
345 * orthogonal operations (such as operations on devices) without racing against
346 * tun_destroy. tun_destroy will wait on the condvar if we're at all busy or
347 * open, to be woken up when the condition is alleviated.
350 tun_busy_locked(struct tuntap_softc *tp)
354 if ((tp->tun_flags & TUN_DYING) != 0) {
356 * Perhaps unintuitive, but the device is busy going away.
357 * Other interpretations of EBUSY from tun_busy make little
358 * sense, since making a busy device even more busy doesn't
359 * sound like a problem.
369 tun_unbusy_locked(struct tuntap_softc *tp)
373 KASSERT(tp->tun_busy != 0, ("tun_unbusy: called for non-busy tunnel"));
376 /* Wake up anything that may be waiting on our busy tunnel. */
377 if (tp->tun_busy == 0)
378 cv_broadcast(&tp->tun_cv);
382 tun_busy(struct tuntap_softc *tp)
387 ret = tun_busy_locked(tp);
393 tun_unbusy(struct tuntap_softc *tp)
397 tun_unbusy_locked(tp);
402 * Sets unit and/or flags given the device name. Must be called with correct
406 tuntap_name2info(const char *name, int *outunit, int *outflags)
408 struct tuntap_driver *drv;
409 struct tuntap_driver_cloner *drvc;
418 * Needed for dev_stdclone, but dev_stdclone will not modify, it just
419 * wants to be able to pass back a char * through the second param. We
420 * will always set that as NULL here, so we'll fake it.
422 dname = __DECONST(char *, name);
425 KASSERT(!SLIST_EMPTY(&V_tuntap_driver_cloners),
426 ("tuntap_driver_cloners failed to initialize"));
427 SLIST_FOREACH(drvc, &V_tuntap_driver_cloners, link) {
428 KASSERT(drvc->drv != NULL,
429 ("tuntap_driver_cloners entry not properly initialized"));
432 if (strcmp(name, drv->cdevsw.d_name) == 0) {
435 flags = drv->ident_flags;
439 if (dev_stdclone(dname, NULL, drv->cdevsw.d_name, &unit) == 1) {
441 flags = drv->ident_flags;
451 if (outflags != NULL)
457 * Get driver information from a set of flags specified. Masks the identifying
458 * part of the flags and compares it against all of the available
459 * tuntap_drivers. Must be called with correct vnet context.
461 static struct tuntap_driver *
462 tuntap_driver_from_flags(int tun_flags)
464 struct tuntap_driver *drv;
465 struct tuntap_driver_cloner *drvc;
467 KASSERT(!SLIST_EMPTY(&V_tuntap_driver_cloners),
468 ("tuntap_driver_cloners failed to initialize"));
469 SLIST_FOREACH(drvc, &V_tuntap_driver_cloners, link) {
470 KASSERT(drvc->drv != NULL,
471 ("tuntap_driver_cloners entry not properly initialized"));
473 if ((tun_flags & TUN_DRIVER_IDENT_MASK) == drv->ident_flags)
481 tun_clone_match(struct if_clone *ifc, const char *name)
485 if (tuntap_name2info(name, NULL, &tunflags) == 0) {
486 if ((tunflags & TUN_L2) == 0)
494 tap_clone_match(struct if_clone *ifc, const char *name)
498 if (tuntap_name2info(name, NULL, &tunflags) == 0) {
499 if ((tunflags & (TUN_L2 | TUN_VMNET)) == TUN_L2)
507 vmnet_clone_match(struct if_clone *ifc, const char *name)
511 if (tuntap_name2info(name, NULL, &tunflags) == 0) {
512 if ((tunflags & TUN_VMNET) != 0)
520 tun_clone_create(struct if_clone *ifc, char *name, size_t len,
521 struct ifc_data *ifd, struct ifnet **ifpp)
523 struct tuntap_driver *drv;
525 int err, i, tunflags, unit;
528 /* The name here tells us exactly what we're creating */
529 err = tuntap_name2info(name, &unit, &tunflags);
533 drv = tuntap_driver_from_flags(tunflags);
538 /* If this unit number is still available that's okay. */
539 if (alloc_unr_specific(drv->unrhdr, unit) == -1)
542 unit = alloc_unr(drv->unrhdr);
545 snprintf(name, IFNAMSIZ, "%s%d", drv->cdevsw.d_name, unit);
547 /* find any existing device, or allocate new unit number */
549 i = clone_create(&drv->clones, &drv->cdevsw, &unit, &dev, 0);
550 /* No preexisting struct cdev *, create one */
552 i = tun_create_device(drv, unit, NULL, &dev, name);
556 struct tuntap_softc *tp = dev->si_drv1;
564 tunclone(void *arg, struct ucred *cred, char *name, int namelen,
567 char devname[SPECNAMELEN + 1];
568 struct tuntap_driver *drv;
569 int append_unit, i, u, tunflags;
576 CURVNET_SET(CRED_TO_VNET(cred));
577 if (tuntap_name2info(name, &u, &tunflags) != 0)
578 goto out; /* Not recognized */
580 if (u != -1 && u > IF_MAXUNIT)
581 goto out; /* Unit number too high */
583 mayclone = priv_check_cred(cred, PRIV_NET_IFCREATE) == 0;
584 if ((tunflags & TUN_L2) != 0) {
585 /* tap/vmnet allow user open with a sysctl */
586 mayclone = (mayclone || tap_allow_uopen) && tapdclone;
588 mayclone = mayclone && tundclone;
592 * If tun cloning is enabled, only the superuser can create an
603 drv = tuntap_driver_from_flags(tunflags);
607 /* find any existing device, or allocate new unit number */
608 i = clone_create(&drv->clones, &drv->cdevsw, &u, dev, 0);
611 namelen = snprintf(devname, sizeof(devname), "%s%d",
616 i = tun_create_device(drv, u, cred, dev, name);
620 if_clone_create(name, namelen, NULL);
627 tun_destroy(struct tuntap_softc *tp)
631 tp->tun_flags |= TUN_DYING;
632 if (tp->tun_busy != 0)
633 cv_wait_unlock(&tp->tun_cv, &tp->tun_mtx);
637 CURVNET_SET(TUN2IFP(tp)->if_vnet);
639 /* destroy_dev will take care of any alias. */
640 destroy_dev(tp->tun_dev);
641 seldrain(&tp->tun_rsel);
642 knlist_clear(&tp->tun_rsel.si_note, 0);
643 knlist_destroy(&tp->tun_rsel.si_note);
644 if ((tp->tun_flags & TUN_L2) != 0) {
645 ether_ifdetach(TUN2IFP(tp));
647 bpfdetach(TUN2IFP(tp));
648 if_detach(TUN2IFP(tp));
650 sx_xlock(&tun_ioctl_sx);
651 TUN2IFP(tp)->if_softc = NULL;
652 sx_xunlock(&tun_ioctl_sx);
653 free_unr(tp->tun_drv->unrhdr, TUN2IFP(tp)->if_dunit);
654 if_free(TUN2IFP(tp));
655 mtx_destroy(&tp->tun_mtx);
656 cv_destroy(&tp->tun_cv);
662 tun_clone_destroy(struct if_clone *ifc __unused, struct ifnet *ifp, uint32_t flags)
664 struct tuntap_softc *tp = ifp->if_softc;
667 TAILQ_REMOVE(&tunhead, tp, tun_list);
675 vnet_tun_init(const void *unused __unused)
677 struct tuntap_driver *drv;
678 struct tuntap_driver_cloner *drvc;
681 for (i = 0; i < nitems(tuntap_drivers); ++i) {
682 drv = &tuntap_drivers[i];
683 drvc = malloc(sizeof(*drvc), M_TUN, M_WAITOK | M_ZERO);
686 struct if_clone_addreq req = {
687 .match_f = drv->clone_match_fn,
688 .create_f = drv->clone_create_fn,
689 .destroy_f = drv->clone_destroy_fn,
691 drvc->cloner = ifc_attach_cloner(drv->cdevsw.d_name, &req);
692 SLIST_INSERT_HEAD(&V_tuntap_driver_cloners, drvc, link);
695 VNET_SYSINIT(vnet_tun_init, SI_SUB_PROTO_IF, SI_ORDER_ANY,
696 vnet_tun_init, NULL);
699 vnet_tun_uninit(const void *unused __unused)
701 struct tuntap_driver_cloner *drvc;
703 while (!SLIST_EMPTY(&V_tuntap_driver_cloners)) {
704 drvc = SLIST_FIRST(&V_tuntap_driver_cloners);
705 SLIST_REMOVE_HEAD(&V_tuntap_driver_cloners, link);
707 if_clone_detach(drvc->cloner);
711 VNET_SYSUNINIT(vnet_tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY,
712 vnet_tun_uninit, NULL);
715 tun_uninit(const void *unused __unused)
717 struct tuntap_driver *drv;
718 struct tuntap_softc *tp;
721 EVENTHANDLER_DEREGISTER(ifnet_arrival_event, arrival_tag);
722 EVENTHANDLER_DEREGISTER(dev_clone, clone_tag);
725 while ((tp = TAILQ_FIRST(&tunhead)) != NULL) {
726 TAILQ_REMOVE(&tunhead, tp, tun_list);
732 for (i = 0; i < nitems(tuntap_drivers); ++i) {
733 drv = &tuntap_drivers[i];
734 delete_unrhdr(drv->unrhdr);
735 clone_cleanup(&drv->clones);
737 mtx_destroy(&tunmtx);
739 SYSUNINIT(tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY, tun_uninit, NULL);
741 static struct tuntap_driver *
742 tuntap_driver_from_ifnet(const struct ifnet *ifp)
744 struct tuntap_driver *drv;
750 for (i = 0; i < nitems(tuntap_drivers); ++i) {
751 drv = &tuntap_drivers[i];
752 if (strcmp(ifp->if_dname, drv->cdevsw.d_name) == 0)
760 tuntapmodevent(module_t mod, int type, void *data)
762 struct tuntap_driver *drv;
767 mtx_init(&tunmtx, "tunmtx", NULL, MTX_DEF);
768 for (i = 0; i < nitems(tuntap_drivers); ++i) {
769 drv = &tuntap_drivers[i];
770 clone_setup(&drv->clones);
771 drv->unrhdr = new_unrhdr(0, IF_MAXUNIT, &tunmtx);
773 arrival_tag = EVENTHANDLER_REGISTER(ifnet_arrival_event,
775 if (arrival_tag == NULL)
777 clone_tag = EVENTHANDLER_REGISTER(dev_clone, tunclone, 0, 1000);
778 if (clone_tag == NULL)
782 /* See tun_uninit, so it's done after the vnet_sysuninit() */
790 static moduledata_t tuntap_mod = {
796 /* We'll only ever have these two, so no need for a macro. */
797 static moduledata_t tun_mod = { "if_tun", NULL, 0 };
798 static moduledata_t tap_mod = { "if_tap", NULL, 0 };
800 DECLARE_MODULE(if_tuntap, tuntap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
801 MODULE_VERSION(if_tuntap, 1);
802 DECLARE_MODULE(if_tun, tun_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
803 MODULE_VERSION(if_tun, 1);
804 DECLARE_MODULE(if_tap, tap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
805 MODULE_VERSION(if_tap, 1);
808 tun_create_device(struct tuntap_driver *drv, int unit, struct ucred *cr,
809 struct cdev **dev, const char *name)
811 struct make_dev_args args;
812 struct tuntap_softc *tp;
815 tp = malloc(sizeof(*tp), M_TUN, M_WAITOK | M_ZERO);
816 mtx_init(&tp->tun_mtx, "tun_mtx", NULL, MTX_DEF);
817 cv_init(&tp->tun_cv, "tun_condvar");
818 tp->tun_flags = drv->ident_flags;
821 make_dev_args_init(&args);
823 args.mda_flags = MAKEDEV_REF | MAKEDEV_CHECKNAME;
824 args.mda_devsw = &drv->cdevsw;
826 args.mda_uid = UID_UUCP;
827 args.mda_gid = GID_DIALER;
828 args.mda_mode = 0600;
829 args.mda_unit = unit;
830 args.mda_si_drv1 = tp;
831 error = make_dev_s(&args, dev, "%s", name);
837 KASSERT((*dev)->si_drv1 != NULL,
838 ("Failed to set si_drv1 at %s creation", name));
840 knlist_init_mtx(&tp->tun_rsel.si_note, &tp->tun_mtx);
842 TAILQ_INSERT_TAIL(&tunhead, tp, tun_list);
848 tunstart(struct ifnet *ifp)
850 struct tuntap_softc *tp = ifp->if_softc;
853 TUNDEBUG(ifp, "starting\n");
854 if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
855 IFQ_LOCK(&ifp->if_snd);
856 IFQ_POLL_NOLOCK(&ifp->if_snd, m);
858 IFQ_UNLOCK(&ifp->if_snd);
861 IFQ_UNLOCK(&ifp->if_snd);
865 if (tp->tun_flags & TUN_RWAIT) {
866 tp->tun_flags &= ~TUN_RWAIT;
869 selwakeuppri(&tp->tun_rsel, PZERO + 1);
870 KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
871 if (tp->tun_flags & TUN_ASYNC && tp->tun_sigio) {
873 pgsigio(&tp->tun_sigio, SIGIO, 0);
881 * queue packets from higher level ready to put out
884 tunstart_l2(struct ifnet *ifp)
886 struct tuntap_softc *tp = ifp->if_softc;
888 TUNDEBUG(ifp, "starting\n");
891 * do not junk pending output if we are in VMnet mode.
892 * XXX: can this do any harm because of queue overflow?
896 if (((tp->tun_flags & TUN_VMNET) == 0) &&
897 ((tp->tun_flags & TUN_READY) != TUN_READY)) {
901 TUNDEBUG(ifp, "not ready, tun_flags = 0x%x\n", tp->tun_flags);
904 IF_DEQUEUE(&ifp->if_snd, m);
907 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
916 ifp->if_drv_flags |= IFF_DRV_OACTIVE;
918 if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
919 if (tp->tun_flags & TUN_RWAIT) {
920 tp->tun_flags &= ~TUN_RWAIT;
924 if ((tp->tun_flags & TUN_ASYNC) && (tp->tun_sigio != NULL)) {
926 pgsigio(&tp->tun_sigio, SIGIO, 0);
930 selwakeuppri(&tp->tun_rsel, PZERO+1);
931 KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
932 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); /* obytes are counted in ether_output */
935 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
940 tap_transmit(struct ifnet *ifp, struct mbuf *m)
945 IFQ_HANDOFF(ifp, m, error);
949 /* XXX: should return an error code so it can fail. */
951 tuncreate(struct cdev *dev)
953 struct tuntap_driver *drv;
954 struct tuntap_softc *tp;
956 struct ether_addr eaddr;
962 ("si_drv1 should have been initialized at creation"));
965 iflags = IFF_MULTICAST;
966 if ((tp->tun_flags & TUN_L2) != 0) {
968 iflags |= IFF_BROADCAST | IFF_SIMPLEX;
971 iflags |= IFF_POINTOPOINT;
973 ifp = tp->tun_ifp = if_alloc(type);
975 panic("%s%d: failed to if_alloc() interface.\n",
976 drv->cdevsw.d_name, dev2unit(dev));
978 if_initname(ifp, drv->cdevsw.d_name, dev2unit(dev));
979 ifp->if_ioctl = tunifioctl;
980 ifp->if_flags = iflags;
981 IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
982 ifp->if_capabilities |= IFCAP_LINKSTATE;
983 if ((tp->tun_flags & TUN_L2) != 0)
984 ifp->if_capabilities |=
985 IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO;
986 ifp->if_capenable |= IFCAP_LINKSTATE;
988 if ((tp->tun_flags & TUN_L2) != 0) {
989 ifp->if_init = tunifinit;
990 ifp->if_start = tunstart_l2;
991 ifp->if_transmit = tap_transmit;
992 ifp->if_qflush = if_qflush;
994 ether_gen_addr(ifp, &eaddr);
995 ether_ifattach(ifp, eaddr.octet);
997 ifp->if_mtu = TUNMTU;
998 ifp->if_start = tunstart;
999 ifp->if_output = tunoutput;
1001 ifp->if_snd.ifq_drv_maxlen = 0;
1002 IFQ_SET_READY(&ifp->if_snd);
1005 bpfattach(ifp, DLT_NULL, sizeof(u_int32_t));
1009 tp->tun_flags |= TUN_INITED;
1012 TUNDEBUG(ifp, "interface %s is created, minor = %#x\n",
1013 ifp->if_xname, dev2unit(dev));
1017 tunrename(void *arg __unused, struct ifnet *ifp)
1019 struct tuntap_softc *tp;
1022 if ((ifp->if_flags & IFF_RENAMING) == 0)
1025 if (tuntap_driver_from_ifnet(ifp) == NULL)
1029 * We need to grab the ioctl sx long enough to make sure the softc is
1030 * still there. If it is, we can safely try to busy the tun device.
1031 * The busy may fail if the device is currently dying, in which case
1032 * we do nothing. If it doesn't fail, the busy count stops the device
1033 * from dying until we've created the alias (that will then be
1034 * subsequently destroyed).
1036 sx_xlock(&tun_ioctl_sx);
1039 sx_xunlock(&tun_ioctl_sx);
1042 error = tun_busy(tp);
1043 sx_xunlock(&tun_ioctl_sx);
1046 if (tp->tun_alias != NULL) {
1047 destroy_dev(tp->tun_alias);
1048 tp->tun_alias = NULL;
1051 if (strcmp(ifp->if_xname, tp->tun_dev->si_name) == 0)
1055 * Failure's ok, aliases are created on a best effort basis. If a
1056 * tun user/consumer decides to rename the interface to conflict with
1057 * another device (non-ifnet) on the system, we will assume they know
1058 * what they are doing. make_dev_alias_p won't touch tun_alias on
1059 * failure, so we use it but ignore the return value.
1061 make_dev_alias_p(MAKEDEV_CHECKNAME, &tp->tun_alias, tp->tun_dev, "%s",
1068 tunopen(struct cdev *dev, int flag, int mode, struct thread *td)
1071 struct tuntap_softc *tp;
1072 int error __diagused, tunflags;
1075 CURVNET_SET(TD_TO_VNET(td));
1076 error = tuntap_name2info(dev->si_name, NULL, &tunflags);
1079 return (error); /* Shouldn't happen */
1084 ("si_drv1 should have been initialized at creation"));
1087 if ((tp->tun_flags & TUN_INITED) == 0) {
1092 if ((tp->tun_flags & (TUN_OPEN | TUN_DYING)) != 0) {
1098 error = tun_busy_locked(tp);
1099 KASSERT(error == 0, ("Must be able to busy an unopen tunnel"));
1102 if ((tp->tun_flags & TUN_L2) != 0) {
1103 bcopy(IF_LLADDR(ifp), tp->tun_ether.octet,
1104 sizeof(tp->tun_ether.octet));
1106 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1107 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1110 ifp->if_flags |= IFF_UP;
1113 tp->tun_pid = td->td_proc->p_pid;
1114 tp->tun_flags |= TUN_OPEN;
1116 if_link_state_change(ifp, LINK_STATE_UP);
1117 TUNDEBUG(ifp, "open\n");
1121 * This can fail with either ENOENT or EBUSY. This is in the middle of
1122 * d_open, so ENOENT should not be possible. EBUSY is possible, but
1123 * the only cdevpriv dtor being set will be tundtor and the softc being
1124 * passed is constant for a given cdev. We ignore the possible error
1125 * because of this as either "unlikely" or "not actually a problem."
1127 (void)devfs_set_cdevpriv(tp, tundtor);
1133 * tundtor - tear down the device - mark i/f down & delete
1140 struct tuntap_softc *tp;
1151 * Realistically, we can't be obstinate here. This only means that the
1152 * tuntap device was closed out of order, and the last closer wasn't the
1153 * controller. These are still good to know about, though, as software
1154 * should avoid multiple processes with a tuntap device open and
1155 * ill-defined transfer of control (e.g., handoff, TUNSIFPID, close in
1158 if (p->p_pid != tp->tun_pid) {
1160 "pid %d (%s), %s: tun/tap protocol violation, non-controlling process closed last.\n",
1161 p->p_pid, p->p_comm, tp->tun_dev->si_name);
1165 * junk all pending output
1167 CURVNET_SET(ifp->if_vnet);
1170 if ((tp->tun_flags & TUN_L2) != 0) {
1172 IF_DRAIN(&ifp->if_snd);
1174 IFQ_PURGE(&ifp->if_snd);
1177 /* For vmnet, we won't do most of the address/route bits */
1178 if ((tp->tun_flags & TUN_VMNET) != 0 ||
1179 (l2tun && (ifp->if_flags & IFF_LINK0) != 0))
1181 #if defined(INET) || defined(INET6)
1182 if (l2tun && tp->tun_lro_ready) {
1183 TUNDEBUG (ifp, "LRO disabled\n");
1184 tcp_lro_free(&tp->tun_lro);
1185 tp->tun_lro_ready = false;
1188 if (ifp->if_flags & IFF_UP) {
1194 /* Delete all addresses and routes which reference this interface. */
1195 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1196 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1203 if_link_state_change(ifp, LINK_STATE_DOWN);
1206 funsetown(&tp->tun_sigio);
1207 selwakeuppri(&tp->tun_rsel, PZERO + 1);
1208 KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
1209 TUNDEBUG (ifp, "closed\n");
1210 tp->tun_flags &= ~TUN_OPEN;
1212 tun_vnethdr_set(ifp, 0);
1214 tun_unbusy_locked(tp);
1219 tuninit(struct ifnet *ifp)
1221 struct tuntap_softc *tp = ifp->if_softc;
1223 TUNDEBUG(ifp, "tuninit\n");
1226 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1227 if ((tp->tun_flags & TUN_L2) == 0) {
1228 ifp->if_flags |= IFF_UP;
1229 getmicrotime(&ifp->if_lastchange);
1232 #if defined(INET) || defined(INET6)
1233 if (tcp_lro_init(&tp->tun_lro) == 0) {
1234 TUNDEBUG(ifp, "LRO enabled\n");
1235 tp->tun_lro.ifp = ifp;
1236 tp->tun_lro_ready = true;
1238 TUNDEBUG(ifp, "Could not enable LRO\n");
1239 tp->tun_lro_ready = false;
1242 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1244 /* attempt to start output */
1251 * Used only for l2 tunnel.
1254 tunifinit(void *xtp)
1256 struct tuntap_softc *tp;
1258 tp = (struct tuntap_softc *)xtp;
1259 tuninit(tp->tun_ifp);
1263 * To be called under TUN_LOCK. Update ifp->if_hwassist according to the
1264 * current value of ifp->if_capenable.
1267 tun_caps_changed(struct ifnet *ifp)
1269 uint64_t hwassist = 0;
1271 TUN_LOCK_ASSERT((struct tuntap_softc *)ifp->if_softc);
1272 if (ifp->if_capenable & IFCAP_TXCSUM)
1273 hwassist |= CSUM_TCP | CSUM_UDP;
1274 if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
1275 hwassist |= CSUM_TCP_IPV6
1277 if (ifp->if_capenable & IFCAP_TSO4)
1278 hwassist |= CSUM_IP_TSO;
1279 if (ifp->if_capenable & IFCAP_TSO6)
1280 hwassist |= CSUM_IP6_TSO;
1281 ifp->if_hwassist = hwassist;
1285 * To be called under TUN_LOCK. Update tp->tun_vhdrlen and adjust
1286 * if_capabilities and if_capenable as needed.
1289 tun_vnethdr_set(struct ifnet *ifp, int vhdrlen)
1291 struct tuntap_softc *tp = ifp->if_softc;
1293 TUN_LOCK_ASSERT(tp);
1295 if (tp->tun_vhdrlen == vhdrlen)
1299 * Update if_capabilities to reflect the
1300 * functionalities offered by the virtio-net
1304 ifp->if_capabilities |=
1307 ifp->if_capabilities &=
1310 * Disable any capabilities that we don't
1313 ifp->if_capenable &= ifp->if_capabilities;
1314 tun_caps_changed(ifp);
1315 tp->tun_vhdrlen = vhdrlen;
1317 TUNDEBUG(ifp, "vnet_hdr_len=%d, if_capabilities=%x\n",
1318 vhdrlen, ifp->if_capabilities);
1322 * Process an ioctl request.
1325 tunifioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1327 struct ifreq *ifr = (struct ifreq *)data;
1328 struct tuntap_softc *tp;
1330 struct ifmediareq *ifmr;
1331 int dummy, error = 0;
1335 sx_xlock(&tun_ioctl_sx);
1341 l2tun = (tp->tun_flags & TUN_L2) != 0;
1344 ifs = (struct ifstat *)data;
1347 snprintf(ifs->ascii, sizeof(ifs->ascii),
1348 "\tOpened by PID %d\n", tp->tun_pid);
1350 ifs->ascii[0] = '\0';
1355 error = ether_ioctl(ifp, cmd, data);
1359 TUNDEBUG(ifp, "address set\n");
1362 ifp->if_mtu = ifr->ifr_mtu;
1363 TUNDEBUG(ifp, "mtu set\n");
1375 ifmr = (struct ifmediareq *)data;
1376 dummy = ifmr->ifm_count;
1377 ifmr->ifm_count = 1;
1378 ifmr->ifm_status = IFM_AVALID;
1379 ifmr->ifm_active = IFM_ETHER | IFM_FDX | IFM_1000_T;
1380 if (tp->tun_flags & TUN_OPEN)
1381 ifmr->ifm_status |= IFM_ACTIVE;
1382 ifmr->ifm_current = ifmr->ifm_active;
1384 int media = IFM_ETHER;
1385 error = copyout(&media, ifmr->ifm_ulist, sizeof(int));
1390 ifp->if_capenable = ifr->ifr_reqcap;
1391 tun_caps_changed(ifp);
1393 VLAN_CAPABILITIES(ifp);
1397 error = ether_ioctl(ifp, cmd, data);
1403 sx_xunlock(&tun_ioctl_sx);
1408 * tunoutput - queue packets from higher level ready to put out.
1411 tunoutput(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst,
1414 struct tuntap_softc *tp = ifp->if_softc;
1415 u_short cached_tun_flags;
1419 TUNDEBUG (ifp, "tunoutput\n");
1422 error = mac_ifnet_check_transmit(ifp, m0);
1429 /* Could be unlocked read? */
1431 cached_tun_flags = tp->tun_flags;
1433 if ((cached_tun_flags & TUN_READY) != TUN_READY) {
1434 TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1439 if ((ifp->if_flags & IFF_UP) != IFF_UP) {
1444 /* BPF writes need to be handled specially. */
1445 if (dst->sa_family == AF_UNSPEC)
1446 bcopy(dst->sa_data, &af, sizeof(af));
1448 af = RO_GET_FAMILY(ro, dst);
1450 BPF_MTAP2(ifp, &af, sizeof(af), m0);
1452 /* prepend sockaddr? this may abort if the mbuf allocation fails */
1453 if (cached_tun_flags & TUN_LMODE) {
1454 /* allocate space for sockaddr */
1455 M_PREPEND(m0, dst->sa_len, M_NOWAIT);
1457 /* if allocation failed drop packet */
1459 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1460 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1463 bcopy(dst, m0->m_data, dst->sa_len);
1467 if (cached_tun_flags & TUN_IFHEAD) {
1468 /* Prepend the address family */
1469 M_PREPEND(m0, 4, M_NOWAIT);
1471 /* if allocation failed drop packet */
1473 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1474 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1477 *(u_int32_t *)m0->m_data = htonl(af);
1484 return (EAFNOSUPPORT);
1488 error = (ifp->if_transmit)(ifp, m0);
1491 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1496 * the cdevsw interface is now pretty minimal.
1499 tunioctl(struct cdev *dev, u_long cmd, caddr_t data, int flag,
1502 struct ifreq ifr, *ifrp;
1503 struct tuntap_softc *tp = dev->si_drv1;
1504 struct ifnet *ifp = TUN2IFP(tp);
1505 struct tuninfo *tunp;
1506 int error, iflags, ival;
1509 l2tun = (tp->tun_flags & TUN_L2) != 0;
1511 /* tap specific ioctls */
1513 /* VMware/VMnet port ioctl's */
1514 #if defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD5) || \
1515 defined(COMPAT_FREEBSD4)
1517 ival = IOCPARM_IVAL(data);
1518 data = (caddr_t)&ival;
1521 case VMIO_SIOCSIFFLAGS: /* VMware/VMnet SIOCSIFFLAGS */
1522 iflags = *(int *)data;
1523 iflags &= TUN_VMIO_FLAG_MASK;
1524 iflags &= ~IFF_CANTCHANGE;
1528 ifp->if_flags = iflags |
1529 (ifp->if_flags & IFF_CANTCHANGE);
1533 case SIOCGIFADDR: /* get MAC address of the remote side */
1535 bcopy(&tp->tun_ether.octet, data,
1536 sizeof(tp->tun_ether.octet));
1540 case SIOCSIFADDR: /* set MAC address of the remote side */
1542 bcopy(data, &tp->tun_ether.octet,
1543 sizeof(tp->tun_ether.octet));
1548 ival = *(int *)data;
1550 ival != sizeof(struct virtio_net_hdr) &&
1551 ival != sizeof(struct virtio_net_hdr_mrg_rxbuf)) {
1555 tun_vnethdr_set(ifp, ival);
1561 *(int *)data = tp->tun_vhdrlen;
1567 /* Fall through to the common ioctls if unhandled */
1573 tp->tun_flags |= TUN_LMODE;
1574 tp->tun_flags &= ~TUN_IFHEAD;
1576 tp->tun_flags &= ~TUN_LMODE;
1583 tp->tun_flags |= TUN_IFHEAD;
1584 tp->tun_flags &= ~TUN_LMODE;
1586 tp->tun_flags &= ~TUN_IFHEAD;
1592 *(int *)data = (tp->tun_flags & TUN_IFHEAD) ? 1 : 0;
1597 /* deny this if UP */
1598 if (TUN2IFP(tp)->if_flags & IFF_UP)
1601 switch (*(int *)data & ~IFF_MULTICAST) {
1602 case IFF_POINTOPOINT:
1605 TUN2IFP(tp)->if_flags &=
1606 ~(IFF_BROADCAST|IFF_POINTOPOINT|IFF_MULTICAST);
1607 TUN2IFP(tp)->if_flags |= *(int *)data;
1618 tp->tun_pid = curthread->td_proc->p_pid;
1623 /* Fall through to the common ioctls if unhandled */
1628 ifrp = (struct ifreq *)data;
1629 strlcpy(ifrp->ifr_name, TUN2IFP(tp)->if_xname, IFNAMSIZ);
1633 tunp = (struct tuninfo *)data;
1634 if (TUN2IFP(tp)->if_type != tunp->type)
1635 return (EPROTOTYPE);
1637 if (TUN2IFP(tp)->if_mtu != tunp->mtu) {
1638 strlcpy(ifr.ifr_name, if_name(TUN2IFP(tp)), IFNAMSIZ);
1639 ifr.ifr_mtu = tunp->mtu;
1640 CURVNET_SET(TUN2IFP(tp)->if_vnet);
1641 error = ifhwioctl(SIOCSIFMTU, TUN2IFP(tp),
1649 TUN2IFP(tp)->if_baudrate = tunp->baudrate;
1653 tunp = (struct tuninfo *)data;
1655 tunp->mtu = TUN2IFP(tp)->if_mtu;
1656 tunp->type = TUN2IFP(tp)->if_type;
1657 tunp->baudrate = TUN2IFP(tp)->if_baudrate;
1661 tundebug = *(int *)data;
1664 *(int *)data = tundebug;
1671 tp->tun_flags |= TUN_ASYNC;
1673 tp->tun_flags &= ~TUN_ASYNC;
1677 if (!IFQ_IS_EMPTY(&TUN2IFP(tp)->if_snd)) {
1679 IFQ_LOCK(&TUN2IFP(tp)->if_snd);
1680 IFQ_POLL_NOLOCK(&TUN2IFP(tp)->if_snd, mb);
1681 for (*(int *)data = 0; mb != NULL; mb = mb->m_next)
1682 *(int *)data += mb->m_len;
1683 IFQ_UNLOCK(&TUN2IFP(tp)->if_snd);
1688 return (fsetown(*(int *)data, &tp->tun_sigio));
1691 *(int *)data = fgetown(&tp->tun_sigio);
1694 /* This is deprecated, FIOSETOWN should be used instead. */
1696 return (fsetown(-(*(int *)data), &tp->tun_sigio));
1698 /* This is deprecated, FIOGETOWN should be used instead. */
1700 *(int *)data = -fgetown(&tp->tun_sigio);
1710 * The cdevsw read interface - reads a packet at a time, or at
1711 * least as much of a packet as can be read.
1714 tunread(struct cdev *dev, struct uio *uio, int flag)
1716 struct tuntap_softc *tp = dev->si_drv1;
1717 struct ifnet *ifp = TUN2IFP(tp);
1722 TUNDEBUG (ifp, "read\n");
1724 if ((tp->tun_flags & TUN_READY) != TUN_READY) {
1726 TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1730 tp->tun_flags &= ~TUN_RWAIT;
1733 IFQ_DEQUEUE(&ifp->if_snd, m);
1736 if (flag & O_NONBLOCK) {
1738 return (EWOULDBLOCK);
1740 tp->tun_flags |= TUN_RWAIT;
1741 error = mtx_sleep(tp, &tp->tun_mtx, PCATCH | (PZERO + 1),
1750 len = min(tp->tun_vhdrlen, uio->uio_resid);
1752 struct virtio_net_hdr_mrg_rxbuf vhdr;
1754 bzero(&vhdr, sizeof(vhdr));
1755 if (m->m_pkthdr.csum_flags & TAP_ALL_OFFLOAD) {
1756 m = virtio_net_tx_offload(ifp, m, false, &vhdr.hdr);
1759 TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, "
1760 "gs %u, cs %u, co %u\n", vhdr.hdr.flags,
1761 vhdr.hdr.gso_type, vhdr.hdr.hdr_len,
1762 vhdr.hdr.gso_size, vhdr.hdr.csum_start,
1763 vhdr.hdr.csum_offset);
1764 error = uiomove(&vhdr, len, uio);
1767 while (m && uio->uio_resid > 0 && error == 0) {
1768 len = min(uio->uio_resid, m->m_len);
1770 error = uiomove(mtod(m, void *), len, uio);
1775 TUNDEBUG(ifp, "Dropping mbuf\n");
1782 tunwrite_l2(struct tuntap_softc *tp, struct mbuf *m,
1783 struct virtio_net_hdr_mrg_rxbuf *vhdr)
1785 struct epoch_tracker et;
1786 struct ether_header *eh;
1792 * Only pass a unicast frame to ether_input(), if it would
1793 * actually have been received by non-virtual hardware.
1795 if (m->m_len < sizeof(struct ether_header)) {
1800 eh = mtod(m, struct ether_header *);
1802 if ((ifp->if_flags & IFF_PROMISC) == 0 &&
1803 !ETHER_IS_MULTICAST(eh->ether_dhost) &&
1804 bcmp(eh->ether_dhost, IF_LLADDR(ifp), ETHER_ADDR_LEN) != 0) {
1810 if (virtio_net_rx_csum(m, &vhdr->hdr)) {
1815 switch (ntohs(eh->ether_type)) {
1818 if (ifp->if_capenable & IFCAP_RXCSUM) {
1819 m->m_pkthdr.csum_flags |=
1820 CSUM_IP_CHECKED | CSUM_IP_VALID |
1821 CSUM_DATA_VALID | CSUM_SCTP_VALID |
1823 m->m_pkthdr.csum_data = 0xffff;
1828 case ETHERTYPE_IPV6:
1829 if (ifp->if_capenable & IFCAP_RXCSUM_IPV6) {
1830 m->m_pkthdr.csum_flags |=
1831 CSUM_DATA_VALID_IPV6 | CSUM_SCTP_VALID |
1833 m->m_pkthdr.csum_data = 0xffff;
1840 /* Pass packet up to parent. */
1841 CURVNET_SET(ifp->if_vnet);
1842 NET_EPOCH_ENTER(et);
1843 #if defined(INET) || defined(INET6)
1844 if (tp->tun_lro_ready && ifp->if_capenable & IFCAP_LRO &&
1845 tcp_lro_rx(&tp->tun_lro, m, 0) == 0)
1846 tcp_lro_flush_all(&tp->tun_lro);
1849 (*ifp->if_input)(ifp, m);
1852 /* ibytes are counted in parent */
1853 if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1858 tunwrite_l3(struct tuntap_softc *tp, struct mbuf *m)
1860 struct epoch_tracker et;
1865 /* Could be unlocked read? */
1867 if (tp->tun_flags & TUN_IFHEAD) {
1869 if (m->m_len < sizeof(family) &&
1870 (m = m_pullup(m, sizeof(family))) == NULL)
1872 family = ntohl(*mtod(m, u_int32_t *));
1873 m_adj(m, sizeof(family));
1879 BPF_MTAP2(ifp, &family, sizeof(family), m);
1894 return (EAFNOSUPPORT);
1896 random_harvest_queue(m, sizeof(*m), RANDOM_NET_TUN);
1897 if_inc_counter(ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
1898 if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1899 CURVNET_SET(ifp->if_vnet);
1900 M_SETFIB(m, ifp->if_fib);
1901 NET_EPOCH_ENTER(et);
1902 netisr_dispatch(isr, m);
1909 * the cdevsw write interface - an atomic write is a packet - or else!
1912 tunwrite(struct cdev *dev, struct uio *uio, int flag)
1914 struct virtio_net_hdr_mrg_rxbuf vhdr;
1915 struct tuntap_softc *tp;
1919 int align, vhdrlen, error;
1924 TUNDEBUG(ifp, "tunwrite\n");
1925 if ((ifp->if_flags & IFF_UP) != IFF_UP)
1926 /* ignore silently */
1929 if (uio->uio_resid == 0)
1932 l2tun = (tp->tun_flags & TUN_L2) != 0;
1933 mru = l2tun ? TAPMRU : TUNMRU;
1934 vhdrlen = tp->tun_vhdrlen;
1937 align = ETHER_ALIGN;
1939 } else if ((tp->tun_flags & TUN_IFHEAD) != 0)
1940 mru += sizeof(uint32_t); /* family */
1941 if (uio->uio_resid < 0 || uio->uio_resid > mru) {
1942 TUNDEBUG(ifp, "len=%zd!\n", uio->uio_resid);
1947 error = uiomove(&vhdr, vhdrlen, uio);
1950 TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, "
1951 "gs %u, cs %u, co %u\n", vhdr.hdr.flags,
1952 vhdr.hdr.gso_type, vhdr.hdr.hdr_len,
1953 vhdr.hdr.gso_size, vhdr.hdr.csum_start,
1954 vhdr.hdr.csum_offset);
1957 if ((m = m_uiotombuf(uio, M_NOWAIT, 0, align, M_PKTHDR)) == NULL) {
1958 if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1962 m->m_pkthdr.rcvif = ifp;
1964 mac_ifnet_create_mbuf(ifp, m);
1968 return (tunwrite_l2(tp, m, vhdrlen > 0 ? &vhdr : NULL));
1970 return (tunwrite_l3(tp, m));
1974 * tunpoll - the poll interface, this is only useful on reads
1975 * really. The write detect always returns true, write never blocks
1976 * anyway, it either accepts the packet or drops it.
1979 tunpoll(struct cdev *dev, int events, struct thread *td)
1981 struct tuntap_softc *tp = dev->si_drv1;
1982 struct ifnet *ifp = TUN2IFP(tp);
1985 TUNDEBUG(ifp, "tunpoll\n");
1987 if (events & (POLLIN | POLLRDNORM)) {
1988 IFQ_LOCK(&ifp->if_snd);
1989 if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
1990 TUNDEBUG(ifp, "tunpoll q=%d\n", ifp->if_snd.ifq_len);
1991 revents |= events & (POLLIN | POLLRDNORM);
1993 TUNDEBUG(ifp, "tunpoll waiting\n");
1994 selrecord(td, &tp->tun_rsel);
1996 IFQ_UNLOCK(&ifp->if_snd);
1998 revents |= events & (POLLOUT | POLLWRNORM);
2004 * tunkqfilter - support for the kevent() system call.
2007 tunkqfilter(struct cdev *dev, struct knote *kn)
2009 struct tuntap_softc *tp = dev->si_drv1;
2010 struct ifnet *ifp = TUN2IFP(tp);
2012 switch(kn->kn_filter) {
2014 TUNDEBUG(ifp, "%s kqfilter: EVFILT_READ, minor = %#x\n",
2015 ifp->if_xname, dev2unit(dev));
2016 kn->kn_fop = &tun_read_filterops;
2020 TUNDEBUG(ifp, "%s kqfilter: EVFILT_WRITE, minor = %#x\n",
2021 ifp->if_xname, dev2unit(dev));
2022 kn->kn_fop = &tun_write_filterops;
2026 TUNDEBUG(ifp, "%s kqfilter: invalid filter, minor = %#x\n",
2027 ifp->if_xname, dev2unit(dev));
2032 knlist_add(&tp->tun_rsel.si_note, kn, 0);
2038 * Return true of there is data in the interface queue.
2041 tunkqread(struct knote *kn, long hint)
2044 struct tuntap_softc *tp = kn->kn_hook;
2045 struct cdev *dev = tp->tun_dev;
2046 struct ifnet *ifp = TUN2IFP(tp);
2048 if ((kn->kn_data = ifp->if_snd.ifq_len) > 0) {
2050 "%s have data in the queue. Len = %d, minor = %#x\n",
2051 ifp->if_xname, ifp->if_snd.ifq_len, dev2unit(dev));
2055 "%s waiting for data, minor = %#x\n", ifp->if_xname,
2064 * Always can write, always return MTU in kn->data.
2067 tunkqwrite(struct knote *kn, long hint)
2069 struct tuntap_softc *tp = kn->kn_hook;
2070 struct ifnet *ifp = TUN2IFP(tp);
2072 kn->kn_data = ifp->if_mtu;
2078 tunkqdetach(struct knote *kn)
2080 struct tuntap_softc *tp = kn->kn_hook;
2082 knlist_remove(&tp->tun_rsel.si_note, kn, 0);