xref: /netbsd-src/sys/rump/net/lib/libvirtif/if_virt.c (revision e89934bbf778a6d6d6894877c4da59d0c7835b0f)
1 /*	$NetBSD: if_virt.c,v 1.54 2016/12/15 09:28:07 ozaki-r Exp $	*/
2 
3 /*
4  * Copyright (c) 2008, 2013 Antti Kantee.  All Rights Reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS
16  * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
17  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
18  * DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
21  * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __KERNEL_RCSID(0, "$NetBSD: if_virt.c,v 1.54 2016/12/15 09:28:07 ozaki-r Exp $");
30 
31 #include <sys/param.h>
32 #include <sys/kernel.h>
33 #include <sys/kmem.h>
34 #include <sys/cprng.h>
35 #include <sys/module.h>
36 
37 #include <net/bpf.h>
38 #include <net/if.h>
39 #include <net/if_dl.h>
40 #include <net/if_ether.h>
41 
42 #include <netinet/in.h>
43 #include <netinet/in_var.h>
44 
45 #include "if_virt.h"
46 #include "virtif_user.h"
47 
48 /*
49  * Virtual interface.  Uses hypercalls to shovel packets back
50  * and forth.  The exact method for shoveling depends on the
51  * hypercall implementation.
52  */
53 
54 static int	virtif_init(struct ifnet *);
55 static int	virtif_ioctl(struct ifnet *, u_long, void *);
56 static void	virtif_start(struct ifnet *);
57 static void	virtif_stop(struct ifnet *, int);
58 
59 struct virtif_sc {
60 	struct ethercom sc_ec;
61 	struct virtif_user *sc_viu;
62 
63 	int sc_num;
64 	char *sc_linkstr;
65 };
66 
67 static int  virtif_clone(struct if_clone *, int);
68 static int  virtif_unclone(struct ifnet *);
69 
70 struct if_clone VIF_CLONER =
71     IF_CLONE_INITIALIZER(VIF_NAME, virtif_clone, virtif_unclone);
72 
73 static int
74 virtif_create(struct ifnet *ifp)
75 {
76 	uint8_t enaddr[ETHER_ADDR_LEN] = { 0xb2, 0x0a, 0x00, 0x0b, 0x0e, 0x01 };
77 	char enaddrstr[3*ETHER_ADDR_LEN];
78 	struct virtif_sc *sc = ifp->if_softc;
79 	int error;
80 
81 	if (sc->sc_viu)
82 		panic("%s: already created", ifp->if_xname);
83 
84 	enaddr[2] = cprng_fast32() & 0xff;
85 	enaddr[5] = sc->sc_num & 0xff;
86 
87 	if ((error = VIFHYPER_CREATE(sc->sc_linkstr,
88 	    sc, enaddr, &sc->sc_viu)) != 0) {
89 		printf("VIFHYPER_CREATE failed: %d\n", error);
90 		return error;
91 	}
92 
93 	ether_ifattach(ifp, enaddr);
94 	ether_snprintf(enaddrstr, sizeof(enaddrstr), enaddr);
95 	aprint_normal_ifnet(ifp, "Ethernet address %s\n", enaddrstr);
96 
97 	IFQ_SET_READY(&ifp->if_snd);
98 
99 	return 0;
100 }
101 
102 static int
103 virtif_clone(struct if_clone *ifc, int num)
104 {
105 	struct virtif_sc *sc;
106 	struct ifnet *ifp;
107 	int error = 0;
108 
109 	sc = kmem_zalloc(sizeof(*sc), KM_SLEEP);
110 	sc->sc_num = num;
111 	ifp = &sc->sc_ec.ec_if;
112 
113 	if_initname(ifp, VIF_NAME, num);
114 	ifp->if_softc = sc;
115 
116 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
117 	ifp->if_init = virtif_init;
118 	ifp->if_ioctl = virtif_ioctl;
119 	ifp->if_start = virtif_start;
120 	ifp->if_stop = virtif_stop;
121 	ifp->if_mtu = ETHERMTU;
122 	ifp->if_dlt = DLT_EN10MB;
123 
124 	if_initialize(ifp);
125 	if_register(ifp);
126 
127 #ifndef RUMP_VIF_LINKSTR
128 	/*
129 	 * if the underlying interface does not expect linkstr, we can
130 	 * create everything now.  Otherwise, we need to wait for
131 	 * SIOCSLINKSTR.
132 	 */
133 #define LINKSTRNUMLEN 16
134 	sc->sc_linkstr = kmem_alloc(LINKSTRNUMLEN, KM_SLEEP);
135 	snprintf(sc->sc_linkstr, LINKSTRNUMLEN, "%d", sc->sc_num);
136 #undef LINKSTRNUMLEN
137 	error = virtif_create(ifp);
138 	if (error) {
139 		if_detach(ifp);
140 		kmem_free(sc, sizeof(*sc));
141 		ifp->if_softc = NULL;
142 	}
143 #endif /* !RUMP_VIF_LINKSTR */
144 
145 	return error;
146 }
147 
148 static int
149 virtif_unclone(struct ifnet *ifp)
150 {
151 	struct virtif_sc *sc = ifp->if_softc;
152 	int rv;
153 
154 	if (ifp->if_flags & IFF_UP)
155 		return EBUSY;
156 
157 	if ((rv = VIFHYPER_DYING(sc->sc_viu)) != 0)
158 		return rv;
159 
160 	virtif_stop(ifp, 1);
161 	if_down(ifp);
162 
163 	VIFHYPER_DESTROY(sc->sc_viu);
164 
165 	kmem_free(sc, sizeof(*sc));
166 
167 	ether_ifdetach(ifp);
168 	if_detach(ifp);
169 
170 	return 0;
171 }
172 
173 static int
174 virtif_init(struct ifnet *ifp)
175 {
176 	struct virtif_sc *sc = ifp->if_softc;
177 
178 	if (sc->sc_viu == NULL)
179 		return ENXIO;
180 
181 	ifp->if_flags |= IFF_RUNNING;
182 	return 0;
183 }
184 
185 static int
186 virtif_ioctl(struct ifnet *ifp, u_long cmd, void *data)
187 {
188 	struct virtif_sc *sc = ifp->if_softc;
189 	int rv;
190 
191 	switch (cmd) {
192 #ifdef RUMP_VIF_LINKSTR
193 	struct ifdrv *ifd;
194 	size_t linkstrlen;
195 
196 #ifndef RUMP_VIF_LINKSTRMAX
197 #define RUMP_VIF_LINKSTRMAX 4096
198 #endif
199 
200 	case SIOCGLINKSTR:
201 		ifd = data;
202 
203 		if (!sc->sc_linkstr) {
204 			rv = ENOENT;
205 			break;
206 		}
207 		linkstrlen = strlen(sc->sc_linkstr)+1;
208 
209 		if (ifd->ifd_cmd == IFLINKSTR_QUERYLEN) {
210 			ifd->ifd_len = linkstrlen;
211 			rv = 0;
212 			break;
213 		}
214 		if (ifd->ifd_cmd != 0) {
215 			rv = ENOTTY;
216 			break;
217 		}
218 
219 		rv = copyoutstr(sc->sc_linkstr,
220 		    ifd->ifd_data, MIN(ifd->ifd_len,linkstrlen), NULL);
221 		break;
222 	case SIOCSLINKSTR:
223 		if (ifp->if_flags & IFF_UP) {
224 			rv = EBUSY;
225 			break;
226 		}
227 
228 		ifd = data;
229 
230 		if (ifd->ifd_cmd == IFLINKSTR_UNSET) {
231 			panic("unset linkstr not implemented");
232 		} else if (ifd->ifd_cmd != 0) {
233 			rv = ENOTTY;
234 			break;
235 		} else if (sc->sc_linkstr) {
236 			rv = EBUSY;
237 			break;
238 		}
239 
240 		if (ifd->ifd_len > RUMP_VIF_LINKSTRMAX) {
241 			rv = E2BIG;
242 			break;
243 		} else if (ifd->ifd_len < 1) {
244 			rv = EINVAL;
245 			break;
246 		}
247 
248 
249 		sc->sc_linkstr = kmem_alloc(ifd->ifd_len, KM_SLEEP);
250 		rv = copyinstr(ifd->ifd_data, sc->sc_linkstr,
251 		    ifd->ifd_len, NULL);
252 		if (rv) {
253 			kmem_free(sc->sc_linkstr, ifd->ifd_len);
254 			break;
255 		}
256 
257 		rv = virtif_create(ifp);
258 		if (rv) {
259 			kmem_free(sc->sc_linkstr, ifd->ifd_len);
260 		}
261 		break;
262 #endif /* RUMP_VIF_LINKSTR */
263 	default:
264 		if (!sc->sc_linkstr)
265 			rv = ENXIO;
266 		else
267 			rv = ether_ioctl(ifp, cmd, data);
268 		if (rv == ENETRESET)
269 			rv = 0;
270 		break;
271 	}
272 
273 	return rv;
274 }
275 
276 /*
277  * Output packets in-context until outgoing queue is empty.
278  * Leave responsibility of choosing whether or not to drop the
279  * kernel lock to VIPHYPER_SEND().
280  */
281 #define LB_SH 32
282 static void
283 virtif_start(struct ifnet *ifp)
284 {
285 	struct virtif_sc *sc = ifp->if_softc;
286 	struct mbuf *m, *m0;
287 	struct iovec io[LB_SH];
288 	int i;
289 
290 	ifp->if_flags |= IFF_OACTIVE;
291 
292 	for (;;) {
293 		IF_DEQUEUE(&ifp->if_snd, m0);
294 		if (!m0) {
295 			break;
296 		}
297 
298 		m = m0;
299 		for (i = 0; i < LB_SH && m; ) {
300 			if (m->m_len) {
301 				io[i].iov_base = mtod(m, void *);
302 				io[i].iov_len = m->m_len;
303 				i++;
304 			}
305 			m = m->m_next;
306 		}
307 		if (i == LB_SH && m)
308 			panic("lazy bum");
309 		bpf_mtap(ifp, m0);
310 
311 		VIFHYPER_SEND(sc->sc_viu, io, i);
312 
313 		m_freem(m0);
314 		ifp->if_opackets++;
315 	}
316 
317 	ifp->if_flags &= ~IFF_OACTIVE;
318 }
319 
320 static void
321 virtif_stop(struct ifnet *ifp, int disable)
322 {
323 
324 	/* XXX: VIFHYPER_STOP() */
325 
326 	ifp->if_flags &= ~IFF_RUNNING;
327 }
328 
329 void
330 VIF_DELIVERPKT(struct virtif_sc *sc, struct iovec *iov, size_t iovlen)
331 {
332 	struct ifnet *ifp = &sc->sc_ec.ec_if;
333 	struct ether_header *eth;
334 	struct mbuf *m;
335 	size_t i;
336 	int off, olen;
337 	bool passup;
338 	const int align
339 	    = ALIGN(sizeof(struct ether_header)) - sizeof(struct ether_header);
340 
341 	if ((ifp->if_flags & IFF_RUNNING) == 0)
342 		return;
343 
344 	m = m_gethdr(M_NOWAIT, MT_DATA);
345 	if (m == NULL)
346 		return; /* drop packet */
347 	m->m_len = m->m_pkthdr.len = 0;
348 
349 	for (i = 0, off = align; i < iovlen; i++) {
350 		olen = m->m_pkthdr.len;
351 		m_copyback(m, off, iov[i].iov_len, iov[i].iov_base);
352 		off += iov[i].iov_len;
353 		if (olen + off != m->m_pkthdr.len) {
354 			aprint_verbose_ifnet(ifp, "m_copyback failed\n");
355 			m_freem(m);
356 			return;
357 		}
358 	}
359 	m->m_data += align;
360 	m->m_pkthdr.len -= align;
361 	m->m_len -= align;
362 
363 	eth = mtod(m, struct ether_header *);
364 	if (memcmp(eth->ether_dhost, CLLADDR(ifp->if_sadl),
365 	    ETHER_ADDR_LEN) == 0) {
366 		passup = true;
367 	} else if (ETHER_IS_MULTICAST(eth->ether_dhost)) {
368 		passup = true;
369 	} else if (ifp->if_flags & IFF_PROMISC) {
370 		m->m_flags |= M_PROMISC;
371 		passup = true;
372 	} else {
373 		passup = false;
374 	}
375 
376 	if (passup) {
377 		int bound;
378 		m_set_rcvif(m, ifp);
379 		KERNEL_LOCK(1, NULL);
380 		/* Prevent LWP migrations between CPUs for psref(9) */
381 		bound = curlwp_bind();
382 		if_input(ifp, m);
383 		curlwp_bindx(bound);
384 		KERNEL_UNLOCK_LAST(NULL);
385 	} else {
386 		m_freem(m);
387 	}
388 	m = NULL;
389 }
390 
391 /*
392  * The following ensures that no two modules using if_virt end up with
393  * the same module name.  MODULE() and modcmd wrapped in ... bad mojo.
394  */
395 #define VIF_MOJO(x) MODULE(MODULE_CLASS_DRIVER,x,NULL);
396 #define VIF_MODULE() VIF_MOJO(VIF_BASENAME(if_virt_,VIRTIF_BASE))
397 #define VIF_MODCMD VIF_BASENAME3(if_virt_,VIRTIF_BASE,_modcmd)
398 VIF_MODULE();
399 static int
400 VIF_MODCMD(modcmd_t cmd, void *opaque)
401 {
402 	int error = 0;
403 
404 	switch (cmd) {
405 	case MODULE_CMD_INIT:
406 		if_clone_attach(&VIF_CLONER);
407 		break;
408 	case MODULE_CMD_FINI:
409 		/*
410 		 * not sure if interfaces are refcounted
411 		 * and properly protected
412 		 */
413 #if 0
414 		if_clone_detach(&VIF_CLONER);
415 #else
416 		error = ENOTTY;
417 #endif
418 		break;
419 	default:
420 		error = ENOTTY;
421 	}
422 	return error;
423 }
424