xref: /illumos-gate/usr/src/uts/common/xen/io/xnbo.c (revision 71269a2275bf5a143dad6461eee2710a344e7261)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /*
28  * Xen network backend - mac client edition.
29  *
30  * A driver that sits above an existing GLDv3/Nemo MAC driver and
31  * relays packets to/from that driver from/to a guest domain.
32  */
33 
34 #include "xnb.h"
35 
36 #include <sys/sunddi.h>
37 #include <sys/modctl.h>
38 #include <sys/strsubr.h>
39 #include <sys/mac.h>
40 #include <net/if.h>
41 #include <sys/dlpi.h>
42 #include <sys/pattr.h>
43 #include <xen/sys/xenbus_impl.h>
44 #include <xen/sys/xendev.h>
45 
46 typedef struct xnbo {
47 	mac_handle_t		o_mh;
48 	mac_rx_handle_t		o_mrh;
49 	const mac_txinfo_t	*o_mtx;
50 	mac_notify_handle_t	o_mnh;
51 	boolean_t		o_running;
52 	boolean_t		o_promiscuous;
53 	uint32_t		o_hcksum_capab;
54 } xnbo_t;
55 
56 static void xnbo_close_mac(xnbo_t *);
57 
58 /*
59  * Packets from the peer come here.  We pass them to the mac device.
60  */
61 static void
62 xnbo_to_mac(xnb_t *xnbp, mblk_t *mp)
63 {
64 	xnbo_t *xnbop = xnbp->xnb_flavour_data;
65 
66 	ASSERT(mp != NULL);
67 
68 	if (!xnbop->o_running) {
69 		xnbp->xnb_stat_tx_too_early++;
70 		goto fail;
71 	}
72 
73 	mp = xnbop->o_mtx->mt_fn(xnbop->o_mtx->mt_arg, mp);
74 
75 	if (mp != NULL) {
76 		xnbp->xnb_stat_mac_full++;
77 		goto fail;
78 	}
79 
80 	return;
81 
82 fail:
83 	freemsgchain(mp);
84 }
85 
86 static mblk_t *
87 xnbo_cksum_from_peer(xnb_t *xnbp, mblk_t *mp, uint16_t flags)
88 {
89 	xnbo_t *xnbop = xnbp->xnb_flavour_data;
90 
91 	ASSERT(mp->b_next == NULL);
92 
93 	if ((flags & NETTXF_csum_blank) != 0) {
94 		/*
95 		 * It would be nice to ASSERT that xnbp->xnb_cksum_offload
96 		 * is TRUE here, but some peers insist on assuming
97 		 * that it is available even when they have been told
98 		 * otherwise.
99 		 *
100 		 * The checksum in the packet is blank.  Determine
101 		 * whether we can do hardware offload and, if so,
102 		 * update the flags on the mblk according.  If not,
103 		 * calculate and insert the checksum using software.
104 		 */
105 		mp = xnb_process_cksum_flags(xnbp, mp,
106 		    xnbop->o_hcksum_capab);
107 	}
108 
109 	return (mp);
110 }
111 
112 static uint16_t
113 xnbo_cksum_to_peer(xnb_t *xnbp, mblk_t *mp)
114 {
115 	uint16_t r = 0;
116 
117 	/*
118 	 * We might also check for HCK_PARTIALCKSUM here and,
119 	 * providing that the partial checksum covers the TCP/UDP
120 	 * payload, return NETRXF_data_validated.
121 	 *
122 	 * It seems that it's probably not worthwhile, as even MAC
123 	 * devices which advertise HCKSUM_INET_PARTIAL in their
124 	 * capabilities tend to use HCK_FULLCKSUM on the receive side
125 	 * - they are actually saying that in the output path the
126 	 * caller must use HCK_PARTIALCKSUM.
127 	 */
128 
129 	if (xnbp->xnb_cksum_offload) {
130 		uint32_t pflags, csum;
131 
132 		/*
133 		 * XXPV dme: Pull in improved hcksum_retrieve() from
134 		 * Crossbow, which gives back the csum in the seventh
135 		 * argument for HCK_FULLCKSUM.
136 		 */
137 		hcksum_retrieve(mp, NULL, NULL, NULL, NULL,
138 		    NULL, NULL, &pflags);
139 		csum = DB_CKSUM16(mp);
140 
141 		/*
142 		 * If the MAC driver has asserted that the checksum is
143 		 * good, let the peer know.
144 		 */
145 		if (((pflags & HCK_FULLCKSUM) != 0) &&
146 		    (((pflags & HCK_FULLCKSUM_OK) != 0) ||
147 		    (csum == 0xffff)))
148 			r |= NETRXF_data_validated;
149 	}
150 
151 	return (r);
152 }
153 
154 /*
155  * Packets from the mac device come here.  We pass them to the peer.
156  */
157 /*ARGSUSED*/
158 static void
159 xnbo_from_mac(void *arg, mac_resource_handle_t mrh, mblk_t *mp)
160 {
161 	xnb_t *xnbp = arg;
162 
163 	mp = xnb_copy_to_peer(xnbp, mp);
164 
165 	if (mp != NULL)
166 		freemsgchain(mp);
167 }
168 
169 /*
170  * Packets from the mac device come here. We pass them to the peer if
171  * the destination mac address matches or it's a multicast/broadcast
172  * address.
173  */
174 /*ARGSUSED*/
175 static void
176 xnbo_from_mac_filter(void *arg, mac_resource_handle_t mrh, mblk_t *mp)
177 {
178 	xnb_t *xnbp = arg;
179 	xnbo_t *xnbop = xnbp->xnb_flavour_data;
180 	mblk_t *next, *keep, *keep_head, *free, *free_head;
181 
182 	keep = keep_head = free = free_head = NULL;
183 
184 #define	ADD(list, bp)				\
185 	if (list != NULL)			\
186 		list->b_next = bp;		\
187 	else					\
188 		list##_head = bp;		\
189 	list = bp;
190 
191 	for (; mp != NULL; mp = next) {
192 		mac_header_info_t hdr_info;
193 
194 		next = mp->b_next;
195 		mp->b_next = NULL;
196 
197 		if (mac_header_info(xnbop->o_mh, mp, &hdr_info) != 0) {
198 			ADD(free, mp);
199 			continue;
200 		}
201 
202 		if ((hdr_info.mhi_dsttype == MAC_ADDRTYPE_BROADCAST) ||
203 		    (hdr_info.mhi_dsttype == MAC_ADDRTYPE_MULTICAST)) {
204 			ADD(keep, mp);
205 			continue;
206 		}
207 
208 		if (bcmp(hdr_info.mhi_daddr, xnbp->xnb_mac_addr,
209 		    sizeof (xnbp->xnb_mac_addr)) == 0) {
210 			ADD(keep, mp);
211 			continue;
212 		}
213 
214 		ADD(free, mp);
215 	}
216 #undef	ADD
217 
218 	if (keep_head != NULL)
219 		xnbo_from_mac(xnbp, mrh, keep_head);
220 
221 	if (free_head != NULL)
222 		freemsgchain(free_head);
223 }
224 
225 static void
226 xnbo_notify(void *arg, mac_notify_type_t type)
227 {
228 	xnb_t *xnbp = arg;
229 	xnbo_t *xnbop = xnbp->xnb_flavour_data;
230 
231 	switch (type) {
232 	case MAC_NOTE_PROMISC:
233 		xnbop->o_mtx = mac_tx_get(xnbop->o_mh);
234 		break;
235 	}
236 }
237 
238 static boolean_t
239 xnbo_open_mac(xnb_t *xnbp, char *mac)
240 {
241 	xnbo_t *xnbop = xnbp->xnb_flavour_data;
242 	int err, need_rx_filter, need_setphysaddr, need_promiscuous;
243 	const mac_info_t *mi;
244 	char *xsname;
245 	void (*rx_fn)(void *, mac_resource_handle_t, mblk_t *);
246 	uint_t max_sdu;
247 
248 	xsname = xvdi_get_xsname(xnbp->xnb_devinfo);
249 
250 	if ((err = mac_open_by_linkname(mac, &xnbop->o_mh)) != 0) {
251 		cmn_err(CE_WARN, "xnbo_open_mac: "
252 		    "cannot open mac for link %s (%d)", mac, err);
253 		return (B_FALSE);
254 	}
255 	ASSERT(xnbop->o_mh != NULL);
256 
257 	mi = mac_info(xnbop->o_mh);
258 	ASSERT(mi != NULL);
259 
260 	if (mi->mi_media != DL_ETHER) {
261 		cmn_err(CE_WARN, "xnbo_open_mac: "
262 		    "device is not DL_ETHER (%d)", mi->mi_media);
263 		xnbo_close_mac(xnbop);
264 		return (B_FALSE);
265 	}
266 	if (mi->mi_media != mi->mi_nativemedia) {
267 		cmn_err(CE_WARN, "xnbo_open_mac: "
268 		    "device media and native media mismatch (%d != %d)",
269 		    mi->mi_media, mi->mi_nativemedia);
270 		xnbo_close_mac(xnbop);
271 		return (B_FALSE);
272 	}
273 
274 	mac_sdu_get(xnbop->o_mh, NULL, &max_sdu);
275 	if (max_sdu > XNBMAXPKT) {
276 		cmn_err(CE_WARN, "xnbo_open_mac: mac device SDU too big (%d)",
277 		    max_sdu);
278 		xnbo_close_mac(xnbop);
279 		return (B_FALSE);
280 	}
281 
282 	xnbop->o_mnh = mac_notify_add(xnbop->o_mh, xnbo_notify, xnbp);
283 	ASSERT(xnbop->o_mnh != NULL);
284 
285 	/*
286 	 * Should the receive path filter packets from the downstream
287 	 * NIC before passing them to the peer? The default is "no".
288 	 */
289 	if (xenbus_scanf(XBT_NULL, xsname,
290 	    "SUNW-need-rx-filter", "%d", &need_rx_filter) != 0)
291 		need_rx_filter = 0;
292 	if (need_rx_filter > 0)
293 		rx_fn = xnbo_from_mac_filter;
294 	else
295 		rx_fn = xnbo_from_mac;
296 
297 	xnbop->o_mrh = mac_rx_add(xnbop->o_mh, rx_fn, xnbp);
298 	ASSERT(xnbop->o_mrh != NULL);
299 
300 	xnbop->o_mtx = mac_tx_get(xnbop->o_mh);
301 	ASSERT(xnbop->o_mtx != NULL);
302 
303 	if (!mac_capab_get(xnbop->o_mh, MAC_CAPAB_HCKSUM,
304 	    &xnbop->o_hcksum_capab))
305 		xnbop->o_hcksum_capab = 0;
306 
307 	/*
308 	 * Should we set the physical address of the underlying NIC
309 	 * to match that assigned to the peer? The default is "no".
310 	 */
311 	if (xenbus_scanf(XBT_NULL, xsname,
312 	    "SUNW-need-set-physaddr", "%d", &need_setphysaddr) != 0)
313 		need_setphysaddr = 0;
314 	if (need_setphysaddr > 0) {
315 		struct ether_addr ea;
316 
317 		err = mac_unicst_set(xnbop->o_mh, xnbp->xnb_mac_addr);
318 		/* Warn, but continue on. */
319 		if (err != 0) {
320 			bcopy(xnbp->xnb_mac_addr, ea.ether_addr_octet,
321 			    ETHERADDRL);
322 			cmn_err(CE_WARN, "xnbo_open_mac: "
323 			    "cannot set MAC address of %s to "
324 			    "%s: %d", mac, ether_sprintf(&ea),
325 			    err);
326 		}
327 	}
328 
329 	/*
330 	 * Should we set the underlying NIC into promiscuous mode? The
331 	 * default is "no".
332 	 */
333 	if (xenbus_scanf(XBT_NULL, xsname,
334 	    "SUNW-need-promiscuous", "%d", &need_promiscuous) != 0)
335 		need_promiscuous = 0;
336 	if (need_promiscuous > 0) {
337 		err = mac_promisc_set(xnbop->o_mh, B_TRUE, MAC_DEVPROMISC);
338 		if (err != 0) {
339 			cmn_err(CE_WARN, "xnbo_open_mac: "
340 			    "cannot enable promiscuous mode of %s: %d",
341 			    mac, err);
342 			xnbo_close_mac(xnbop);
343 			return (B_FALSE);
344 		}
345 		xnbop->o_promiscuous = B_TRUE;
346 	}
347 
348 	if ((err = mac_start(xnbop->o_mh)) != 0) {
349 		cmn_err(CE_WARN, "xnbo_open_mac: "
350 		    "cannot start mac device (%d)", err);
351 		xnbo_close_mac(xnbop);
352 		return (B_FALSE);
353 	}
354 	xnbop->o_running = B_TRUE;
355 
356 	return (B_TRUE);
357 }
358 
359 /*
360  * xnb calls back here when the user-level hotplug code reports that
361  * the hotplug has successfully completed. For this flavour that means
362  * that the underlying MAC device that we will use is ready to be
363  * opened.
364  */
365 static boolean_t
366 xnbo_hotplug(xnb_t *xnbp)
367 {
368 	char *xsname;
369 	char mac[LIFNAMSIZ];
370 
371 	xsname = xvdi_get_xsname(xnbp->xnb_devinfo);
372 	if (xenbus_scanf(XBT_NULL, xsname, "nic", "%s", mac) != 0) {
373 		cmn_err(CE_WARN, "xnbo_hotplug: "
374 		    "cannot read nic name from %s", xsname);
375 		return (B_FALSE);
376 	}
377 
378 	return (xnbo_open_mac(xnbp, mac));
379 }
380 
381 static void
382 xnbo_close_mac(xnbo_t *xnbop)
383 {
384 	if (xnbop->o_mh == NULL)
385 		return;
386 
387 	if (xnbop->o_running) {
388 		mac_stop(xnbop->o_mh);
389 		xnbop->o_running = B_FALSE;
390 	}
391 
392 	if (xnbop->o_promiscuous) {
393 		(void) mac_promisc_set(xnbop->o_mh, B_FALSE,
394 		    MAC_DEVPROMISC);
395 		xnbop->o_promiscuous = B_FALSE;
396 	}
397 
398 	xnbop->o_mtx = NULL;
399 
400 	if (xnbop->o_mrh != NULL) {
401 		mac_rx_remove(xnbop->o_mh, xnbop->o_mrh, B_TRUE);
402 		xnbop->o_mrh = NULL;
403 	}
404 
405 	if (xnbop->o_mnh != NULL) {
406 		mac_notify_remove(xnbop->o_mh, xnbop->o_mnh);
407 		xnbop->o_mnh = NULL;
408 	}
409 
410 	mac_close(xnbop->o_mh);
411 	xnbop->o_mh = NULL;
412 }
413 
414 /*
415  * xnb calls back here when we successfully synchronize with the
416  * driver in the guest domain. In this flavour there is nothing to do as
417  * we open the underlying MAC device on successful hotplug completion.
418  */
419 /*ARGSUSED*/
420 static void
421 xnbo_connected(xnb_t *xnbp)
422 {
423 }
424 
425 /*
426  * xnb calls back here when the driver in the guest domain has closed
427  * down the inter-domain connection. We close the underlying MAC device.
428  */
429 static void
430 xnbo_disconnected(xnb_t *xnbp)
431 {
432 	xnbo_close_mac(xnbp->xnb_flavour_data);
433 }
434 
435 static int
436 xnbo_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
437 {
438 	static xnb_flavour_t flavour = {
439 		xnbo_to_mac, xnbo_connected, xnbo_disconnected, xnbo_hotplug,
440 		xnbo_cksum_from_peer, xnbo_cksum_to_peer,
441 	};
442 	xnbo_t *xnbop;
443 
444 	switch (cmd) {
445 	case DDI_ATTACH:
446 		break;
447 	case DDI_RESUME:
448 		return (DDI_SUCCESS);
449 	default:
450 		return (DDI_FAILURE);
451 	}
452 
453 	xnbop = kmem_zalloc(sizeof (*xnbop), KM_SLEEP);
454 
455 	xnbop->o_mh = NULL;
456 	xnbop->o_mrh = NULL;
457 	xnbop->o_mtx = NULL;
458 	xnbop->o_running = B_FALSE;
459 	xnbop->o_hcksum_capab = 0;
460 
461 	if (xnb_attach(dip, &flavour, xnbop) != DDI_SUCCESS) {
462 		kmem_free(xnbop, sizeof (*xnbop));
463 		return (DDI_FAILURE);
464 	}
465 
466 	return (DDI_SUCCESS);
467 }
468 
469 static int
470 xnbo_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
471 {
472 	xnb_t *xnbp = ddi_get_driver_private(dip);
473 	xnbo_t *xnbop = xnbp->xnb_flavour_data;
474 
475 	switch (cmd) {
476 	case DDI_DETACH:
477 		break;
478 	case DDI_SUSPEND:
479 		return (DDI_SUCCESS);
480 	default:
481 		return (DDI_FAILURE);
482 	}
483 
484 	mutex_enter(&xnbp->xnb_tx_lock);
485 	mutex_enter(&xnbp->xnb_rx_lock);
486 
487 	if (!xnbp->xnb_detachable || xnbp->xnb_connected ||
488 	    (xnbp->xnb_tx_buf_count > 0)) {
489 		mutex_exit(&xnbp->xnb_rx_lock);
490 		mutex_exit(&xnbp->xnb_tx_lock);
491 
492 		return (DDI_FAILURE);
493 	}
494 
495 	mutex_exit(&xnbp->xnb_rx_lock);
496 	mutex_exit(&xnbp->xnb_tx_lock);
497 
498 	xnbo_close_mac(xnbop);
499 	kmem_free(xnbop, sizeof (*xnbop));
500 
501 	xnb_detach(dip);
502 
503 	return (DDI_SUCCESS);
504 }
505 
506 static struct cb_ops cb_ops = {
507 	nulldev,		/* open */
508 	nulldev,		/* close */
509 	nodev,			/* strategy */
510 	nodev,			/* print */
511 	nodev,			/* dump */
512 	nodev,			/* read */
513 	nodev,			/* write */
514 	nodev,			/* ioctl */
515 	nodev,			/* devmap */
516 	nodev,			/* mmap */
517 	nodev,			/* segmap */
518 	nochpoll,		/* poll */
519 	ddi_prop_op,		/* cb_prop_op */
520 	0,			/* streamtab  */
521 	D_NEW | D_MP | D_64BIT	/* Driver compatibility flag */
522 };
523 
524 static struct dev_ops ops = {
525 	DEVO_REV,		/* devo_rev */
526 	0,			/* devo_refcnt  */
527 	nulldev,		/* devo_getinfo */
528 	nulldev,		/* devo_identify */
529 	nulldev,		/* devo_probe */
530 	xnbo_attach,		/* devo_attach */
531 	xnbo_detach,		/* devo_detach */
532 	nodev,			/* devo_reset */
533 	&cb_ops,		/* devo_cb_ops */
534 	(struct bus_ops *)0,	/* devo_bus_ops */
535 	NULL,			/* devo_power */
536 	ddi_quiesce_not_needed,		/* devo_quiesce */
537 };
538 
539 static struct modldrv modldrv = {
540 	&mod_driverops, "xnbo driver", &ops,
541 };
542 
543 static struct modlinkage modlinkage = {
544 	MODREV_1, &modldrv, NULL
545 };
546 
547 int
548 _init(void)
549 {
550 	return (mod_install(&modlinkage));
551 }
552 
553 int
554 _info(struct modinfo *modinfop)
555 {
556 	return (mod_info(&modlinkage, modinfop));
557 }
558 
559 int
560 _fini(void)
561 {
562 	return (mod_remove(&modlinkage));
563 }
564