xref: /illumos-gate/usr/src/uts/common/io/zcons.c (revision 581cede61ac9c14d8d4ea452562a567189eead78)
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  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 
27 /*
28  * Zone Console Driver.
29  *
30  * This driver, derived from the pts/ptm drivers, is the pseudo console driver
31  * for system zones.  Its implementation is straightforward.  Each instance
32  * of the driver represents a global-zone/local-zone pair (this maps in a
33  * straightforward way to the commonly used terminal notion of "master side"
34  * and "slave side", and we use that terminology throughout).
35  *
36  * Instances of zcons are onlined as children of /pseudo/zconsnex@1/
37  * by zoneadmd in userland, using the devctl framework; thus the driver
38  * does not need to maintain any sort of "admin" node.
39  *
40  * The driver shuttles I/O from master side to slave side and back.  In a break
41  * from the pts/ptm semantics, if one side is not open, I/O directed towards
42  * it will simply be discarded.  This is so that if zoneadmd is not holding
43  * the master side console open (i.e. it has died somehow), processes in
44  * the zone do not experience any errors and I/O to the console does not
45  * hang.
46  *
47  * TODO: we may want to revisit the other direction; i.e. we may want
48  * zoneadmd to be able to detect whether no zone processes are holding the
49  * console open, an unusual situation.
50  *
51  *
52  *
53  * MASTER SIDE IOCTLS
54  *
55  * The ZC_HOLDSLAVE and ZC_RELEASESLAVE ioctls instruct the master side of the
56  * console to hold and release a reference to the slave side's vnode.  They are
57  * meant to be issued by zoneadmd after the console device node is created and
58  * before it is destroyed so that the slave's STREAMS anchor, ptem, is
59  * preserved when ttymon starts popping STREAMS modules from within the
60  * associated zone.  This guarantees that the zone console will always have
61  * terminal semantics while the zone is running.
62  *
63  * Here is the issue: the ptem module is anchored in the zone console
64  * (slave side) so that processes within the associated non-global zone will
65  * fail to pop it off, thus ensuring that the slave will retain terminal
66  * semantics.  When a process attempts to pop the anchor off of a stream, the
67  * STREAMS subsystem checks whether the calling process' zone is the same as
68  * that of the process that pushed the anchor onto the stream and cancels the
69  * pop if they differ.  zoneadmd used to hold an open file descriptor for the
70  * slave while the associated non-global zone ran, thus ensuring that the
71  * slave's STREAMS anchor would never be popped from within the non-global zone
72  * (because zoneadmd runs in the global zone).  However, this file descriptor
73  * was removed to make zone console management more robust.  sad(7D) is now
74  * used to automatically set up the slave's STREAMS modules when the zone
75  * console is freshly opened within the associated non-global zone.  However,
76  * when a process within the non-global zone freshly opens the zone console, the
77  * anchor is pushed from within the non-global zone, making it possible for
78  * processes within the non-global zone (e.g., ttymon) to pop the anchor and
79  * destroy the zone console's terminal semantics.
80  *
81  * One solution is to make the zcons device hold the slave open while the
82  * associated non-global zone runs so that the STREAMS anchor will always be
83  * associated with the global zone.  Unfortunately, the slave cannot be opened
84  * from within the zcons driver because the driver is not reentrant: it has
85  * an outer STREAMS perimeter.  Therefore, the next best option is for zcons to
86  * provide an ioctl interface to zoneadmd to manage holding and releasing
87  * the slave side of the console.  It is sufficient to hold the slave side's
88  * vnode and bump the associated snode's reference count to preserve the slave's
89  * STREAMS configuration while the associated zone runs, so that's what the
90  * ioctls do.
91  *
92  *
93  * ZC_HOLDSLAVE
94  *
95  * This ioctl takes a file descriptor as an argument.  It effectively gets a
96  * reference to the slave side's minor node's vnode and bumps the associated
97  * snode's reference count.  The vnode reference is stored in the zcons device
98  * node's soft state.  This ioctl succeeds if the given file descriptor refers
99  * to the slave side's minor node or if there is already a reference to the
100  * slave side's minor node's vnode in the device's soft state.
101  *
102  *
103  * ZC_RELEASESLAVE
104  *
105  * This ioctl takes a file descriptor as an argument.  It effectively releases
106  * the vnode reference stored in the zcons device node's soft state (which was
107  * previously acquired via ZC_HOLDSLAVE) and decrements the reference count of
108  * the snode associated with the vnode.  This ioctl succeeds if the given file
109  * descriptor refers to the slave side's minor node or if no reference to the
110  * slave side's minor node's vnode is stored in the device's soft state.
111  *
112  *
113  * Note that the file descriptor arguments for both ioctls must be cast to
114  * integers of pointer width.
115  *
116  * Here's how the dance between zcons and zoneadmd works:
117  *
118  *     Zone boot:
119  *     1.  While booting the zone, zoneadmd creates an instance of zcons.
120  *     2.  zoneadmd opens the master and slave sides of the new zone console
121  *         and issues the ZC_HOLDSLAVE ioctl on the master side, passing its
122  *         file descriptor for the slave side as the ioctl argument.
123  *     3.  zcons holds the slave side's vnode, bumps the snode's reference
124  *         count, and stores a pointer to the vnode in the device's soft
125  *         state.
126  *     4.  zoneadmd closes the master and slave sides and continues to boot
127  *         the zone.
128  *
129  *     Zone halt:
130  *     1.  While halting the zone, zoneadmd opens the master and slave sides
131  *         of the zone's console and issues the ZC_RELEASESLAVE ioctl on the
132  *         master side, passing its file descriptor for the slave side as the
133  *         ioctl argument.
134  *     2.  zcons decrements the slave side's snode's reference count, releases
135  *         the slave's vnode, and eliminates its reference to the vnode in the
136  *         device's soft state.
137  *     3.  zoneadmd closes the master and slave sides.
138  *     4.  zoneadmd destroys the zcons device and continues to halt the zone.
139  *
140  * It is necessary for zoneadmd to hold the slave open while issuing
141  * ZC_RELEASESLAVE because zcons might otherwise release the last reference to
142  * the slave's vnode.  If it does, then specfs will panic because it will expect
143  * that the STREAMS configuration for the vnode was destroyed, which VN_RELE
144  * doesn't do.  Forcing zoneadmd to hold the slave open guarantees that zcons
145  * won't release the vnode's last reference.  zoneadmd will properly destroy the
146  * vnode and the snode when it closes the file descriptor.
147  *
148  * Technically, any process that can access the master side can issue these
149  * ioctls, but they should be treated as private interfaces for zoneadmd.
150  */
151 
152 #include <sys/types.h>
153 #include <sys/cmn_err.h>
154 #include <sys/conf.h>
155 #include <sys/cred.h>
156 #include <sys/ddi.h>
157 #include <sys/debug.h>
158 #include <sys/devops.h>
159 #include <sys/errno.h>
160 #include <sys/file.h>
161 #include <sys/kstr.h>
162 #include <sys/modctl.h>
163 #include <sys/param.h>
164 #include <sys/stat.h>
165 #include <sys/stream.h>
166 #include <sys/stropts.h>
167 #include <sys/strsun.h>
168 #include <sys/sunddi.h>
169 #include <sys/sysmacros.h>
170 #include <sys/systm.h>
171 #include <sys/types.h>
172 #include <sys/zcons.h>
173 #include <sys/vnode.h>
174 #include <sys/fs/snode.h>
175 
176 static int zc_getinfo(dev_info_t *, ddi_info_cmd_t, void *, void **);
177 static int zc_attach(dev_info_t *, ddi_attach_cmd_t);
178 static int zc_detach(dev_info_t *, ddi_detach_cmd_t);
179 
180 static int zc_open(queue_t *, dev_t *, int, int, cred_t *);
181 static int zc_close(queue_t *, int, cred_t *);
182 static void zc_wput(queue_t *, mblk_t *);
183 static void zc_rsrv(queue_t *);
184 static void zc_wsrv(queue_t *);
185 
186 /*
187  * The instance number is encoded in the dev_t in the minor number; the lowest
188  * bit of the minor number is used to track the master vs. slave side of the
189  * virtual console.  The rest of the bits in the minor number are the instance.
190  */
191 #define	ZC_MASTER_MINOR		0
192 #define	ZC_SLAVE_MINOR		1
193 
194 #define	ZC_INSTANCE(x)		(getminor((x)) >> 1)
195 #define	ZC_NODE(x)		(getminor((x)) & 0x01)
196 
197 /*
198  * This macro converts a zc_state_t pointer to the associated slave minor node's
199  * dev_t.
200  */
201 #define	ZC_STATE_TO_SLAVEDEV(x)	(makedevice(ddi_driver_major((x)->zc_devinfo), \
202 	(minor_t)(ddi_get_instance((x)->zc_devinfo) << 1 | ZC_SLAVE_MINOR)))
203 
204 int zcons_debug = 0;
205 #define	DBG(a)   if (zcons_debug) cmn_err(CE_NOTE, a)
206 #define	DBG1(a, b)   if (zcons_debug) cmn_err(CE_NOTE, a, b)
207 
208 
209 /*
210  * Zone Console Pseudo Terminal Module: stream data structure definitions
211  */
212 static struct module_info zc_info = {
213 	31337,	/* c0z we r hAx0rs */
214 	"zcons",
215 	0,
216 	INFPSZ,
217 	2048,
218 	128
219 };
220 
221 static struct qinit zc_rinit = {
222 	NULL,
223 	(int (*)()) zc_rsrv,
224 	zc_open,
225 	zc_close,
226 	NULL,
227 	&zc_info,
228 	NULL
229 };
230 
231 static struct qinit zc_winit = {
232 	(int (*)()) zc_wput,
233 	(int (*)()) zc_wsrv,
234 	NULL,
235 	NULL,
236 	NULL,
237 	&zc_info,
238 	NULL
239 };
240 
241 static struct streamtab zc_tab_info = {
242 	&zc_rinit,
243 	&zc_winit,
244 	NULL,
245 	NULL
246 };
247 
248 #define	ZC_CONF_FLAG	(D_MP | D_MTQPAIR | D_MTOUTPERIM | D_MTOCEXCL)
249 
250 /*
251  * this will define (struct cb_ops cb_zc_ops) and (struct dev_ops zc_ops)
252  */
253 DDI_DEFINE_STREAM_OPS(zc_ops, nulldev, nulldev,	zc_attach, zc_detach, nodev, \
254 	zc_getinfo, ZC_CONF_FLAG, &zc_tab_info, ddi_quiesce_not_needed);
255 
256 /*
257  * Module linkage information for the kernel.
258  */
259 
260 static struct modldrv modldrv = {
261 	&mod_driverops, 	/* Type of module (this is a pseudo driver) */
262 	"Zone console driver",	/* description of module */
263 	&zc_ops			/* driver ops */
264 };
265 
266 static struct modlinkage modlinkage = {
267 	MODREV_1,
268 	&modldrv,
269 	NULL
270 };
271 
272 typedef struct zc_state {
273 	dev_info_t *zc_devinfo;
274 	queue_t *zc_master_rdq;
275 	queue_t *zc_slave_rdq;
276 	vnode_t *zc_slave_vnode;
277 	int zc_state;
278 } zc_state_t;
279 
280 #define	ZC_STATE_MOPEN	0x01
281 #define	ZC_STATE_SOPEN	0x02
282 
283 static void *zc_soft_state;
284 
285 /*
286  * List of STREAMS modules that should be pushed onto every slave instance.
287  */
288 static char *zcons_mods[] = {
289 	"ptem",
290 	"ldterm",
291 	"ttcompat",
292 	NULL
293 };
294 
295 int
296 _init(void)
297 {
298 	int err;
299 
300 	if ((err = ddi_soft_state_init(&zc_soft_state,
301 	    sizeof (zc_state_t), 0)) != 0) {
302 		return (err);
303 	}
304 
305 	if ((err = mod_install(&modlinkage)) != 0)
306 		ddi_soft_state_fini(zc_soft_state);
307 
308 	return (err);
309 }
310 
311 
312 int
313 _fini(void)
314 {
315 	int err;
316 
317 	if ((err = mod_remove(&modlinkage)) != 0) {
318 		return (err);
319 	}
320 
321 	ddi_soft_state_fini(&zc_soft_state);
322 	return (0);
323 }
324 
325 int
326 _info(struct modinfo *modinfop)
327 {
328 	return (mod_info(&modlinkage, modinfop));
329 }
330 
331 /*
332  * This is a convenience function that clears a device's autopush configuration.
333  * It is meant to be used on the slave side of the console.  Unlike calling
334  * kstr_autopush() directly, this function outputs a warning via cmn_err() if
335  * kstr_autopush() fails.  'dip' must be non-NULL in debug builds.  Both
336  * 'major' and 'minor' must be valid.
337  */
338 static void
339 zc_clearautopush(dev_info_t *dip, major_t major, minor_t minor)
340 {
341 	char *devicepathp;
342 
343 	if (kstr_autopush(CLR_AUTOPUSH, &major, &minor, NULL, NULL, NULL) !=
344 	    0 && zcons_debug != 0) {
345 		devicepathp = (char *)kmem_alloc(MAXPATHLEN * sizeof (char),
346 		    KM_SLEEP);
347 		(void) ddi_pathname(dip, devicepathp);
348 		cmn_err(CE_NOTE, "zc_detach: could not clear sad configuration "
349 		    "for device %s\n", devicepathp);
350 		kmem_free(devicepathp, MAXPATHLEN * sizeof (char));
351 	}
352 }
353 
354 static int
355 zc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
356 {
357 	zc_state_t *zcs;
358 	int instance;
359 	major_t major;
360 	minor_t minor;
361 	minor_t lastminor;
362 	uint_t anchorindex;
363 
364 	if (cmd != DDI_ATTACH)
365 		return (DDI_FAILURE);
366 
367 	instance = ddi_get_instance(dip);
368 	if (ddi_soft_state_zalloc(zc_soft_state, instance) != DDI_SUCCESS)
369 		return (DDI_FAILURE);
370 
371 	/*
372 	 * Set up sad(7D) so that the necessary STREAMS modules will be in place
373 	 * when the slave is opened.  A wrinkle is that 'ptem' must be anchored
374 	 * in place (see streamio(7i)) because we always want the console to
375 	 * have terminal semantics.
376 	 */
377 	minor = instance << 1 | ZC_SLAVE_MINOR;
378 	lastminor = 0;
379 	major = ddi_driver_major(dip);
380 	anchorindex = 1;
381 	if (kstr_autopush(SET_AUTOPUSH, &major, &minor, &lastminor,
382 	    &anchorindex, zcons_mods) != 0)
383 		goto commonfail;
384 
385 	/*
386 	 * Create the master and slave minor nodes.
387 	 */
388 	if ((ddi_create_minor_node(dip, ZCONS_SLAVE_NAME, S_IFCHR, minor,
389 	    DDI_PSEUDO, 0) == DDI_FAILURE) ||
390 	    (ddi_create_minor_node(dip, ZCONS_MASTER_NAME, S_IFCHR,
391 	    instance << 1 | ZC_MASTER_MINOR, DDI_PSEUDO, 0) == DDI_FAILURE))
392 		goto failwithautopush;
393 
394 	VERIFY((zcs = ddi_get_soft_state(zc_soft_state, instance)) != NULL);
395 	zcs->zc_devinfo = dip;
396 	return (DDI_SUCCESS);
397 
398 failwithautopush:
399 	zc_clearautopush(dip, major, minor);
400 	ddi_remove_minor_node(dip, NULL);
401 commonfail:
402 	ddi_soft_state_free(zc_soft_state, instance);
403 	return (DDI_FAILURE);
404 }
405 
406 static int
407 zc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
408 {
409 	zc_state_t *zcs;
410 	int instance;
411 
412 	if (cmd != DDI_DETACH)
413 		return (DDI_FAILURE);
414 
415 	instance = ddi_get_instance(dip);
416 	if ((zcs = ddi_get_soft_state(zc_soft_state, instance)) == NULL)
417 		return (DDI_FAILURE);
418 
419 	if ((zcs->zc_state & ZC_STATE_MOPEN) ||
420 	    (zcs->zc_state & ZC_STATE_SOPEN)) {
421 		DBG1("zc_detach: device (dip=%p) still open\n", (void *)dip);
422 		return (DDI_FAILURE);
423 	}
424 
425 	/*
426 	 * Clear the sad configuration so that reattaching doesn't fail to
427 	 * set up sad configuration.
428 	 */
429 	zc_clearautopush(dip, ddi_driver_major(dip), instance << 1 |
430 	    ZC_SLAVE_MINOR);
431 
432 	ddi_remove_minor_node(dip, NULL);
433 	ddi_soft_state_free(zc_soft_state, instance);
434 
435 	return (DDI_SUCCESS);
436 }
437 
438 /*
439  * zc_getinfo()
440  *	getinfo(9e) entrypoint.
441  */
442 /*ARGSUSED*/
443 static int
444 zc_getinfo(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result)
445 {
446 	zc_state_t *zcs;
447 	int instance = ZC_INSTANCE((dev_t)arg);
448 
449 	switch (infocmd) {
450 	case DDI_INFO_DEVT2DEVINFO:
451 		if ((zcs = ddi_get_soft_state(zc_soft_state, instance)) == NULL)
452 			return (DDI_FAILURE);
453 		*result = zcs->zc_devinfo;
454 		return (DDI_SUCCESS);
455 	case DDI_INFO_DEVT2INSTANCE:
456 		*result = (void *)(uintptr_t)instance;
457 		return (DDI_SUCCESS);
458 	}
459 	return (DDI_FAILURE);
460 }
461 
462 /*
463  * Return the equivalent queue from the other side of the relationship.
464  * e.g.: given the slave's write queue, return the master's write queue.
465  */
466 static queue_t *
467 zc_switch(queue_t *qp)
468 {
469 	zc_state_t *zcs = qp->q_ptr;
470 	ASSERT(zcs != NULL);
471 
472 	if (qp == zcs->zc_master_rdq)
473 		return (zcs->zc_slave_rdq);
474 	else if (OTHERQ(qp) == zcs->zc_master_rdq && zcs->zc_slave_rdq != NULL)
475 		return (OTHERQ(zcs->zc_slave_rdq));
476 	else if (qp == zcs->zc_slave_rdq)
477 		return (zcs->zc_master_rdq);
478 	else if (OTHERQ(qp) == zcs->zc_slave_rdq && zcs->zc_master_rdq != NULL)
479 		return (OTHERQ(zcs->zc_master_rdq));
480 	else
481 		return (NULL);
482 }
483 
484 /*
485  * For debugging and outputting messages.  Returns the name of the side of
486  * the relationship associated with this queue.
487  */
488 static const char *
489 zc_side(queue_t *qp)
490 {
491 	zc_state_t *zcs = qp->q_ptr;
492 	ASSERT(zcs != NULL);
493 
494 	if (qp == zcs->zc_master_rdq ||
495 	    OTHERQ(qp) == zcs->zc_master_rdq) {
496 		return ("master");
497 	}
498 	ASSERT(qp == zcs->zc_slave_rdq || OTHERQ(qp) == zcs->zc_slave_rdq);
499 	return ("slave");
500 }
501 
502 /*ARGSUSED*/
503 static int
504 zc_master_open(zc_state_t *zcs,
505     queue_t	*rqp,	/* pointer to the read side queue */
506     dev_t	*devp,	/* pointer to stream tail's dev */
507     int		oflag,	/* the user open(2) supplied flags */
508     int		sflag,	/* open state flag */
509     cred_t	*credp)	/* credentials */
510 {
511 	mblk_t *mop;
512 	struct stroptions *sop;
513 
514 	/*
515 	 * Enforce exclusivity on the master side; the only consumer should
516 	 * be the zoneadmd for the zone.
517 	 */
518 	if ((zcs->zc_state & ZC_STATE_MOPEN) != 0)
519 		return (EBUSY);
520 
521 	if ((mop = allocb(sizeof (struct stroptions), BPRI_MED)) == NULL) {
522 		DBG("zc_master_open(): mop allocation failed\n");
523 		return (ENOMEM);
524 	}
525 
526 	zcs->zc_state |= ZC_STATE_MOPEN;
527 
528 	/*
529 	 * q_ptr stores driver private data; stash the soft state data on both
530 	 * read and write sides of the queue.
531 	 */
532 	WR(rqp)->q_ptr = rqp->q_ptr = zcs;
533 	qprocson(rqp);
534 
535 	/*
536 	 * Following qprocson(), the master side is fully plumbed into the
537 	 * STREAM and may send/receive messages.  Setting zcs->zc_master_rdq
538 	 * will allow the slave to send messages to us (the master).
539 	 * This cannot occur before qprocson() because the master is not
540 	 * ready to process them until that point.
541 	 */
542 	zcs->zc_master_rdq = rqp;
543 
544 	/*
545 	 * set up hi/lo water marks on stream head read queue and add
546 	 * controlling tty as needed.
547 	 */
548 	mop->b_datap->db_type = M_SETOPTS;
549 	mop->b_wptr += sizeof (struct stroptions);
550 	sop = (struct stroptions *)(void *)mop->b_rptr;
551 	if (oflag & FNOCTTY)
552 		sop->so_flags = SO_HIWAT | SO_LOWAT;
553 	else
554 		sop->so_flags = SO_HIWAT | SO_LOWAT | SO_ISTTY;
555 	sop->so_hiwat = 512;
556 	sop->so_lowat = 256;
557 	putnext(rqp, mop);
558 
559 	return (0);
560 }
561 
562 /*ARGSUSED*/
563 static int
564 zc_slave_open(zc_state_t *zcs,
565     queue_t	*rqp,	/* pointer to the read side queue */
566     dev_t	*devp,	/* pointer to stream tail's dev */
567     int		oflag,	/* the user open(2) supplied flags */
568     int		sflag,	/* open state flag */
569     cred_t	*credp)	/* credentials */
570 {
571 	mblk_t *mop;
572 	struct stroptions *sop;
573 
574 	/*
575 	 * The slave side can be opened as many times as needed.
576 	 */
577 	if ((zcs->zc_state & ZC_STATE_SOPEN) != 0) {
578 		ASSERT((rqp != NULL) && (WR(rqp)->q_ptr == zcs));
579 		return (0);
580 	}
581 
582 	if ((mop = allocb(sizeof (struct stroptions), BPRI_MED)) == NULL) {
583 		DBG("zc_slave_open(): mop allocation failed\n");
584 		return (ENOMEM);
585 	}
586 
587 	zcs->zc_state |= ZC_STATE_SOPEN;
588 
589 	/*
590 	 * q_ptr stores driver private data; stash the soft state data on both
591 	 * read and write sides of the queue.
592 	 */
593 	WR(rqp)->q_ptr = rqp->q_ptr = zcs;
594 
595 	qprocson(rqp);
596 
597 	/*
598 	 * Must follow qprocson(), since we aren't ready to process until then.
599 	 */
600 	zcs->zc_slave_rdq = rqp;
601 
602 	/*
603 	 * set up hi/lo water marks on stream head read queue and add
604 	 * controlling tty as needed.
605 	 */
606 	mop->b_datap->db_type = M_SETOPTS;
607 	mop->b_wptr += sizeof (struct stroptions);
608 	sop = (struct stroptions *)(void *)mop->b_rptr;
609 	sop->so_flags = SO_HIWAT | SO_LOWAT | SO_ISTTY;
610 	sop->so_hiwat = 512;
611 	sop->so_lowat = 256;
612 	putnext(rqp, mop);
613 
614 	return (0);
615 }
616 
617 /*
618  * open(9e) entrypoint; checks sflag, and rejects anything unordinary.
619  */
620 static int
621 zc_open(queue_t *rqp,		/* pointer to the read side queue */
622 	dev_t   *devp,		/* pointer to stream tail's dev */
623 	int	oflag,		/* the user open(2) supplied flags */
624 	int	sflag,		/* open state flag */
625 	cred_t  *credp)		/* credentials */
626 {
627 	int instance = ZC_INSTANCE(*devp);
628 	int ret;
629 	zc_state_t *zcs;
630 
631 	if (sflag != 0)
632 		return (EINVAL);
633 
634 	if ((zcs = ddi_get_soft_state(zc_soft_state, instance)) == NULL)
635 		return (ENXIO);
636 
637 	switch (ZC_NODE(*devp)) {
638 	case ZC_MASTER_MINOR:
639 		ret = zc_master_open(zcs, rqp, devp, oflag, sflag, credp);
640 		break;
641 	case ZC_SLAVE_MINOR:
642 		ret = zc_slave_open(zcs, rqp, devp, oflag, sflag, credp);
643 		break;
644 	default:
645 		ret = ENXIO;
646 		break;
647 	}
648 
649 	return (ret);
650 }
651 
652 /*
653  * close(9e) entrypoint.
654  */
655 /*ARGSUSED1*/
656 static int
657 zc_close(queue_t *rqp, int flag, cred_t *credp)
658 {
659 	queue_t *wqp;
660 	mblk_t	*bp;
661 	zc_state_t *zcs;
662 
663 	zcs = (zc_state_t *)rqp->q_ptr;
664 
665 	if (rqp == zcs->zc_master_rdq) {
666 		DBG("Closing master side");
667 
668 		zcs->zc_master_rdq = NULL;
669 		zcs->zc_state &= ~ZC_STATE_MOPEN;
670 
671 		/*
672 		 * qenable slave side write queue so that it can flush
673 		 * its messages as master's read queue is going away
674 		 */
675 		if (zcs->zc_slave_rdq != NULL) {
676 			qenable(WR(zcs->zc_slave_rdq));
677 		}
678 
679 		qprocsoff(rqp);
680 		WR(rqp)->q_ptr = rqp->q_ptr = NULL;
681 
682 	} else if (rqp == zcs->zc_slave_rdq) {
683 
684 		DBG("Closing slave side");
685 		zcs->zc_state &= ~ZC_STATE_SOPEN;
686 		zcs->zc_slave_rdq = NULL;
687 
688 		wqp = WR(rqp);
689 		while ((bp = getq(wqp)) != NULL) {
690 			if (zcs->zc_master_rdq != NULL)
691 				putnext(zcs->zc_master_rdq, bp);
692 			else if (bp->b_datap->db_type == M_IOCTL)
693 				miocnak(wqp, bp, 0, 0);
694 			else
695 				freemsg(bp);
696 		}
697 
698 		/*
699 		 * Qenable master side write queue so that it can flush its
700 		 * messages as slaves's read queue is going away.
701 		 */
702 		if (zcs->zc_master_rdq != NULL)
703 			qenable(WR(zcs->zc_master_rdq));
704 
705 		qprocsoff(rqp);
706 		WR(rqp)->q_ptr = rqp->q_ptr = NULL;
707 	}
708 
709 	return (0);
710 }
711 
712 static void
713 handle_mflush(queue_t *qp, mblk_t *mp)
714 {
715 	mblk_t *nmp;
716 	DBG1("M_FLUSH on %s side", zc_side(qp));
717 
718 	if (*mp->b_rptr & FLUSHW) {
719 		DBG1("M_FLUSH, FLUSHW, %s side", zc_side(qp));
720 		flushq(qp, FLUSHDATA);
721 		*mp->b_rptr &= ~FLUSHW;
722 		if ((*mp->b_rptr & FLUSHR) == 0) {
723 			/*
724 			 * FLUSHW only. Change to FLUSHR and putnext other side,
725 			 * then we are done.
726 			 */
727 			*mp->b_rptr |= FLUSHR;
728 			if (zc_switch(RD(qp)) != NULL) {
729 				putnext(zc_switch(RD(qp)), mp);
730 				return;
731 			}
732 		} else if ((zc_switch(RD(qp)) != NULL) &&
733 		    (nmp = copyb(mp)) != NULL) {
734 			/*
735 			 * It is a FLUSHRW; we copy the mblk and send
736 			 * it to the other side, since we still need to use
737 			 * the mblk in FLUSHR processing, below.
738 			 */
739 			putnext(zc_switch(RD(qp)), nmp);
740 		}
741 	}
742 
743 	if (*mp->b_rptr & FLUSHR) {
744 		DBG("qreply(qp) turning FLUSHR around\n");
745 		qreply(qp, mp);
746 		return;
747 	}
748 	freemsg(mp);
749 }
750 
751 /*
752  * wput(9E) is symmetric for master and slave sides, so this handles both
753  * without splitting the codepath.  (The only exception to this is the
754  * processing of zcons ioctls, which is restricted to the master side.)
755  *
756  * zc_wput() looks at the other side; if there is no process holding that
757  * side open, it frees the message.  This prevents processes from hanging
758  * if no one is holding open the console.  Otherwise, it putnext's high
759  * priority messages, putnext's normal messages if possible, and otherwise
760  * enqueues the messages; in the case that something is enqueued, wsrv(9E)
761  * will take care of eventually shuttling I/O to the other side.
762  */
763 static void
764 zc_wput(queue_t *qp, mblk_t *mp)
765 {
766 	unsigned char type = mp->b_datap->db_type;
767 	zc_state_t *zcs;
768 	struct iocblk *iocbp;
769 	file_t *slave_filep;
770 	struct snode *slave_snodep;
771 	int slave_fd;
772 
773 	ASSERT(qp->q_ptr);
774 
775 	DBG1("entering zc_wput, %s side", zc_side(qp));
776 
777 	/*
778 	 * Process zcons ioctl messages if qp is the master console's write
779 	 * queue.
780 	 */
781 	zcs = (zc_state_t *)qp->q_ptr;
782 	if (zcs->zc_master_rdq != NULL && qp == WR(zcs->zc_master_rdq) &&
783 	    type == M_IOCTL) {
784 		iocbp = (struct iocblk *)(void *)mp->b_rptr;
785 		switch (iocbp->ioc_cmd) {
786 		case ZC_HOLDSLAVE:
787 			/*
788 			 * Hold the slave's vnode and increment the refcount
789 			 * of the snode.  If the vnode is already held, then
790 			 * indicate success.
791 			 */
792 			if (iocbp->ioc_count != TRANSPARENT) {
793 				miocack(qp, mp, 0, EINVAL);
794 				return;
795 			}
796 			if (zcs->zc_slave_vnode != NULL) {
797 				miocack(qp, mp, 0, 0);
798 				return;
799 			}
800 
801 			/*
802 			 * The calling process must pass a file descriptor for
803 			 * the slave device.
804 			 */
805 			slave_fd =
806 			    (int)(intptr_t)*(caddr_t *)(void *)mp->b_cont->
807 			    b_rptr;
808 			slave_filep = getf(slave_fd);
809 			if (slave_filep == NULL) {
810 				miocack(qp, mp, 0, EINVAL);
811 				return;
812 			}
813 			if (ZC_STATE_TO_SLAVEDEV(zcs) !=
814 			    slave_filep->f_vnode->v_rdev) {
815 				releasef(slave_fd);
816 				miocack(qp, mp, 0, EINVAL);
817 				return;
818 			}
819 
820 			/*
821 			 * Get a reference to the slave's vnode.  Also bump the
822 			 * reference count on the associated snode.
823 			 */
824 			ASSERT(vn_matchops(slave_filep->f_vnode,
825 			    spec_getvnodeops()));
826 			zcs->zc_slave_vnode = slave_filep->f_vnode;
827 			VN_HOLD(zcs->zc_slave_vnode);
828 			slave_snodep = VTOCS(zcs->zc_slave_vnode);
829 			mutex_enter(&slave_snodep->s_lock);
830 			++slave_snodep->s_count;
831 			mutex_exit(&slave_snodep->s_lock);
832 			releasef(slave_fd);
833 			miocack(qp, mp, 0, 0);
834 			return;
835 		case ZC_RELEASESLAVE:
836 			/*
837 			 * Release the master's handle on the slave's vnode.
838 			 * If there isn't a handle for the vnode, then indicate
839 			 * success.
840 			 */
841 			if (iocbp->ioc_count != TRANSPARENT) {
842 				miocack(qp, mp, 0, EINVAL);
843 				return;
844 			}
845 			if (zcs->zc_slave_vnode == NULL) {
846 				miocack(qp, mp, 0, 0);
847 				return;
848 			}
849 
850 			/*
851 			 * The process that passed the ioctl must have provided
852 			 * a file descriptor for the slave device.  Make sure
853 			 * this is correct.
854 			 */
855 			slave_fd =
856 			    (int)(intptr_t)*(caddr_t *)(void *)mp->b_cont->
857 			    b_rptr;
858 			slave_filep = getf(slave_fd);
859 			if (slave_filep == NULL) {
860 				miocack(qp, mp, 0, EINVAL);
861 				return;
862 			}
863 			if (zcs->zc_slave_vnode->v_rdev !=
864 			    slave_filep->f_vnode->v_rdev) {
865 				releasef(slave_fd);
866 				miocack(qp, mp, 0, EINVAL);
867 				return;
868 			}
869 
870 			/*
871 			 * Decrement the snode's reference count and release the
872 			 * vnode.
873 			 */
874 			ASSERT(vn_matchops(slave_filep->f_vnode,
875 			    spec_getvnodeops()));
876 			slave_snodep = VTOCS(zcs->zc_slave_vnode);
877 			mutex_enter(&slave_snodep->s_lock);
878 			--slave_snodep->s_count;
879 			mutex_exit(&slave_snodep->s_lock);
880 			VN_RELE(zcs->zc_slave_vnode);
881 			zcs->zc_slave_vnode = NULL;
882 			releasef(slave_fd);
883 			miocack(qp, mp, 0, 0);
884 			return;
885 		default:
886 			break;
887 		}
888 	}
889 
890 	if (zc_switch(RD(qp)) == NULL) {
891 		DBG1("wput to %s side (no one listening)", zc_side(qp));
892 		switch (type) {
893 		case M_FLUSH:
894 			handle_mflush(qp, mp);
895 			break;
896 		case M_IOCTL:
897 			miocnak(qp, mp, 0, 0);
898 			break;
899 		default:
900 			freemsg(mp);
901 			break;
902 		}
903 		return;
904 	}
905 
906 	if (type >= QPCTL) {
907 		DBG1("(hipri) wput, %s side", zc_side(qp));
908 		switch (type) {
909 		case M_READ:		/* supposedly from ldterm? */
910 			DBG("zc_wput: tossing M_READ\n");
911 			freemsg(mp);
912 			break;
913 		case M_FLUSH:
914 			handle_mflush(qp, mp);
915 			break;
916 		default:
917 			/*
918 			 * Put this to the other side.
919 			 */
920 			ASSERT(zc_switch(RD(qp)) != NULL);
921 			putnext(zc_switch(RD(qp)), mp);
922 			break;
923 		}
924 		DBG1("done (hipri) wput, %s side", zc_side(qp));
925 		return;
926 	}
927 
928 	/*
929 	 * Only putnext if there isn't already something in the queue.
930 	 * otherwise things would wind up out of order.
931 	 */
932 	if (qp->q_first == NULL && bcanputnext(RD(zc_switch(qp)), mp->b_band)) {
933 		DBG("wput: putting message to other side\n");
934 		putnext(RD(zc_switch(qp)), mp);
935 	} else {
936 		DBG("wput: putting msg onto queue\n");
937 		(void) putq(qp, mp);
938 	}
939 	DBG1("done wput, %s side", zc_side(qp));
940 }
941 
942 /*
943  * rsrv(9E) is symmetric for master and slave, so zc_rsrv() handles both
944  * without splitting up the codepath.
945  *
946  * Enable the write side of the partner.  This triggers the partner to send
947  * messages queued on its write side to this queue's read side.
948  */
949 static void
950 zc_rsrv(queue_t *qp)
951 {
952 	zc_state_t *zcs;
953 	zcs = (zc_state_t *)qp->q_ptr;
954 
955 	/*
956 	 * Care must be taken here, as either of the master or slave side
957 	 * qptr could be NULL.
958 	 */
959 	ASSERT(qp == zcs->zc_master_rdq || qp == zcs->zc_slave_rdq);
960 	if (zc_switch(qp) == NULL) {
961 		DBG("zc_rsrv: other side isn't listening\n");
962 		return;
963 	}
964 	qenable(WR(zc_switch(qp)));
965 }
966 
967 /*
968  * This routine is symmetric for master and slave, so it handles both without
969  * splitting up the codepath.
970  *
971  * If there are messages on this queue that can be sent to the other, send
972  * them via putnext(). Else, if queued messages cannot be sent, leave them
973  * on this queue.
974  */
975 static void
976 zc_wsrv(queue_t *qp)
977 {
978 	mblk_t *mp;
979 
980 	DBG1("zc_wsrv master (%s) side", zc_side(qp));
981 
982 	/*
983 	 * Partner has no read queue, so take the data, and throw it away.
984 	 */
985 	if (zc_switch(RD(qp)) == NULL) {
986 		DBG("zc_wsrv: other side isn't listening");
987 		while ((mp = getq(qp)) != NULL) {
988 			if (mp->b_datap->db_type == M_IOCTL)
989 				miocnak(qp, mp, 0, 0);
990 			else
991 				freemsg(mp);
992 		}
993 		flushq(qp, FLUSHALL);
994 		return;
995 	}
996 
997 	/*
998 	 * while there are messages on this write queue...
999 	 */
1000 	while ((mp = getq(qp)) != NULL) {
1001 		/*
1002 		 * Due to the way zc_wput is implemented, we should never
1003 		 * see a control message here.
1004 		 */
1005 		ASSERT(mp->b_datap->db_type < QPCTL);
1006 
1007 		if (bcanputnext(RD(zc_switch(qp)), mp->b_band)) {
1008 			DBG("wsrv: send message to other side\n");
1009 			putnext(RD(zc_switch(qp)), mp);
1010 		} else {
1011 			DBG("wsrv: putting msg back on queue\n");
1012 			(void) putbq(qp, mp);
1013 			break;
1014 		}
1015 	}
1016 }
1017