xref: /netbsd-src/sys/rump/net/lib/libvirtif/if_virt.c (revision 76c7fc5f6b13ed0b1508e6b313e88e59977ed78e)
1 /*	$NetBSD: if_virt.c,v 1.57 2018/06/26 06:48:03 msaitoh 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 1.57 2018/06/26 06:48:03 msaitoh 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 	error = if_initialize(ifp);
125 	if (error != 0) {
126 		aprint_error("%s: if_initialize failed(%d)\n", ifp->if_xname,
127 		    error);
128 		goto fail_1;
129 	}
130 
131 	if_register(ifp);
132 
133 #ifndef RUMP_VIF_LINKSTR
134 	/*
135 	 * if the underlying interface does not expect linkstr, we can
136 	 * create everything now.  Otherwise, we need to wait for
137 	 * SIOCSLINKSTR.
138 	 */
139 #define LINKSTRNUMLEN 16
140 	sc->sc_linkstr = kmem_alloc(LINKSTRNUMLEN, KM_SLEEP);
141 	if (sc->sc_linkstr == NULL) {
142 		error = ENOMEM;
143 		goto fail_2;
144 	}
145 	snprintf(sc->sc_linkstr, LINKSTRNUMLEN, "%d", sc->sc_num);
146 	error = virtif_create(ifp);
147 	if (error) {
148 fail_2:
149 		if_detach(ifp);
150 		if (sc->sc_linkstr != NULL)
151 			kmem_free(sc->sc_linkstr, LINKSTRNUMLEN);
152 #undef LINKSTRNUMLEN
153 fail_1:
154 		kmem_free(sc, sizeof(*sc));
155 		ifp->if_softc = NULL;
156 	}
157 #endif /* !RUMP_VIF_LINKSTR */
158 
159 	return error;
160 }
161 
162 static int
163 virtif_unclone(struct ifnet *ifp)
164 {
165 	struct virtif_sc *sc = ifp->if_softc;
166 	int rv;
167 
168 	if (ifp->if_flags & IFF_UP)
169 		return EBUSY;
170 
171 	if ((rv = VIFHYPER_DYING(sc->sc_viu)) != 0)
172 		return rv;
173 
174 	virtif_stop(ifp, 1);
175 	if_down(ifp);
176 
177 	VIFHYPER_DESTROY(sc->sc_viu);
178 
179 	kmem_free(sc, sizeof(*sc));
180 
181 	ether_ifdetach(ifp);
182 	if_detach(ifp);
183 
184 	return 0;
185 }
186 
187 static int
188 virtif_init(struct ifnet *ifp)
189 {
190 	struct virtif_sc *sc = ifp->if_softc;
191 
192 	if (sc->sc_viu == NULL)
193 		return ENXIO;
194 
195 	ifp->if_flags |= IFF_RUNNING;
196 	return 0;
197 }
198 
199 static int
200 virtif_ioctl(struct ifnet *ifp, u_long cmd, void *data)
201 {
202 	struct virtif_sc *sc = ifp->if_softc;
203 	int rv;
204 
205 	switch (cmd) {
206 #ifdef RUMP_VIF_LINKSTR
207 	struct ifdrv *ifd;
208 	size_t linkstrlen;
209 
210 #ifndef RUMP_VIF_LINKSTRMAX
211 #define RUMP_VIF_LINKSTRMAX 4096
212 #endif
213 
214 	case SIOCGLINKSTR:
215 		ifd = data;
216 
217 		if (!sc->sc_linkstr) {
218 			rv = ENOENT;
219 			break;
220 		}
221 		linkstrlen = strlen(sc->sc_linkstr)+1;
222 
223 		if (ifd->ifd_cmd == IFLINKSTR_QUERYLEN) {
224 			ifd->ifd_len = linkstrlen;
225 			rv = 0;
226 			break;
227 		}
228 		if (ifd->ifd_cmd != 0) {
229 			rv = ENOTTY;
230 			break;
231 		}
232 
233 		rv = copyoutstr(sc->sc_linkstr,
234 		    ifd->ifd_data, MIN(ifd->ifd_len,linkstrlen), NULL);
235 		break;
236 	case SIOCSLINKSTR:
237 		if (ifp->if_flags & IFF_UP) {
238 			rv = EBUSY;
239 			break;
240 		}
241 
242 		ifd = data;
243 
244 		if (ifd->ifd_cmd == IFLINKSTR_UNSET) {
245 			panic("unset linkstr not implemented");
246 		} else if (ifd->ifd_cmd != 0) {
247 			rv = ENOTTY;
248 			break;
249 		} else if (sc->sc_linkstr) {
250 			rv = EBUSY;
251 			break;
252 		}
253 
254 		if (ifd->ifd_len > RUMP_VIF_LINKSTRMAX) {
255 			rv = E2BIG;
256 			break;
257 		} else if (ifd->ifd_len < 1) {
258 			rv = EINVAL;
259 			break;
260 		}
261 
262 
263 		sc->sc_linkstr = kmem_alloc(ifd->ifd_len, KM_SLEEP);
264 		rv = copyinstr(ifd->ifd_data, sc->sc_linkstr,
265 		    ifd->ifd_len, NULL);
266 		if (rv) {
267 			kmem_free(sc->sc_linkstr, ifd->ifd_len);
268 			break;
269 		}
270 
271 		rv = virtif_create(ifp);
272 		if (rv) {
273 			kmem_free(sc->sc_linkstr, ifd->ifd_len);
274 		}
275 		break;
276 #endif /* RUMP_VIF_LINKSTR */
277 	default:
278 		if (!sc->sc_linkstr)
279 			rv = ENXIO;
280 		else
281 			rv = ether_ioctl(ifp, cmd, data);
282 		if (rv == ENETRESET)
283 			rv = 0;
284 		break;
285 	}
286 
287 	return rv;
288 }
289 
290 /*
291  * Output packets in-context until outgoing queue is empty.
292  * Leave responsibility of choosing whether or not to drop the
293  * kernel lock to VIPHYPER_SEND().
294  */
295 #define LB_SH 32
296 static void
297 virtif_start(struct ifnet *ifp)
298 {
299 	struct virtif_sc *sc = ifp->if_softc;
300 	struct mbuf *m, *m0;
301 	struct iovec io[LB_SH];
302 	int i;
303 
304 	ifp->if_flags |= IFF_OACTIVE;
305 
306 	for (;;) {
307 		IF_DEQUEUE(&ifp->if_snd, m0);
308 		if (!m0) {
309 			break;
310 		}
311 
312 		m = m0;
313 		for (i = 0; i < LB_SH && m; ) {
314 			if (m->m_len) {
315 				io[i].iov_base = mtod(m, void *);
316 				io[i].iov_len = m->m_len;
317 				i++;
318 			}
319 			m = m->m_next;
320 		}
321 		if (i == LB_SH && m)
322 			panic("lazy bum");
323 		bpf_mtap(ifp, m0, BPF_D_OUT);
324 
325 		VIFHYPER_SEND(sc->sc_viu, io, i);
326 
327 		m_freem(m0);
328 		ifp->if_opackets++;
329 	}
330 
331 	ifp->if_flags &= ~IFF_OACTIVE;
332 }
333 
334 static void
335 virtif_stop(struct ifnet *ifp, int disable)
336 {
337 
338 	/* XXX: VIFHYPER_STOP() */
339 
340 	ifp->if_flags &= ~IFF_RUNNING;
341 }
342 
343 void
344 VIF_DELIVERPKT(struct virtif_sc *sc, struct iovec *iov, size_t iovlen)
345 {
346 	struct ifnet *ifp = &sc->sc_ec.ec_if;
347 	struct ether_header *eth;
348 	struct mbuf *m;
349 	size_t i;
350 	int off, olen;
351 	bool passup;
352 	const int align
353 	    = ALIGN(sizeof(struct ether_header)) - sizeof(struct ether_header);
354 
355 	if ((ifp->if_flags & IFF_RUNNING) == 0)
356 		return;
357 
358 	m = m_gethdr(M_NOWAIT, MT_DATA);
359 	if (m == NULL)
360 		return; /* drop packet */
361 	m->m_len = m->m_pkthdr.len = 0;
362 
363 	for (i = 0, off = align; i < iovlen; i++) {
364 		olen = m->m_pkthdr.len;
365 		m_copyback(m, off, iov[i].iov_len, iov[i].iov_base);
366 		off += iov[i].iov_len;
367 		if (olen + off != m->m_pkthdr.len) {
368 			aprint_verbose_ifnet(ifp, "m_copyback failed\n");
369 			m_freem(m);
370 			return;
371 		}
372 	}
373 	m->m_data += align;
374 	m->m_pkthdr.len -= align;
375 	m->m_len -= align;
376 
377 	eth = mtod(m, struct ether_header *);
378 	if (memcmp(eth->ether_dhost, CLLADDR(ifp->if_sadl),
379 	    ETHER_ADDR_LEN) == 0) {
380 		passup = true;
381 	} else if (ETHER_IS_MULTICAST(eth->ether_dhost)) {
382 		passup = true;
383 	} else if (ifp->if_flags & IFF_PROMISC) {
384 		m->m_flags |= M_PROMISC;
385 		passup = true;
386 	} else {
387 		passup = false;
388 	}
389 
390 	if (passup) {
391 		int bound;
392 		m_set_rcvif(m, ifp);
393 		KERNEL_LOCK(1, NULL);
394 		/* Prevent LWP migrations between CPUs for psref(9) */
395 		bound = curlwp_bind();
396 		if_input(ifp, m);
397 		curlwp_bindx(bound);
398 		KERNEL_UNLOCK_LAST(NULL);
399 	} else {
400 		m_freem(m);
401 	}
402 	m = NULL;
403 }
404 
405 /*
406  * The following ensures that no two modules using if_virt end up with
407  * the same module name.  MODULE() and modcmd wrapped in ... bad mojo.
408  */
409 #define VIF_MOJO(x) MODULE(MODULE_CLASS_DRIVER,x,NULL);
410 #define VIF_MODULE() VIF_MOJO(VIF_BASENAME(if_virt_,VIRTIF_BASE))
411 #define VIF_MODCMD VIF_BASENAME3(if_virt_,VIRTIF_BASE,_modcmd)
412 VIF_MODULE();
413 static int
414 VIF_MODCMD(modcmd_t cmd, void *opaque)
415 {
416 	int error = 0;
417 
418 	switch (cmd) {
419 	case MODULE_CMD_INIT:
420 		if_clone_attach(&VIF_CLONER);
421 		break;
422 	case MODULE_CMD_FINI:
423 		/*
424 		 * not sure if interfaces are refcounted
425 		 * and properly protected
426 		 */
427 #if 0
428 		if_clone_detach(&VIF_CLONER);
429 #else
430 		error = ENOTTY;
431 #endif
432 		break;
433 	default:
434 		error = ENOTTY;
435 	}
436 	return error;
437 }
438