xref: /titanic_52/usr/src/uts/sun4v/io/vdc.c (revision a93a1f58a8763fa69172980b98e3d24720c1136e)
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 2007 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 /*
30  * LDoms virtual disk client (vdc) device driver
31  *
32  * This driver runs on a guest logical domain and communicates with the virtual
33  * disk server (vds) driver running on the service domain which is exporting
34  * virtualized "disks" to the guest logical domain.
35  *
36  * The driver can be divided into four sections:
37  *
38  * 1) generic device driver housekeeping
39  *	_init, _fini, attach, detach, ops structures, etc.
40  *
41  * 2) communication channel setup
42  *	Setup the communications link over the LDC channel that vdc uses to
43  *	talk to the vDisk server. Initialise the descriptor ring which
44  *	allows the LDC clients to transfer data via memory mappings.
45  *
46  * 3) Support exported to upper layers (filesystems, etc)
47  *	The upper layers call into vdc via strategy(9E) and DKIO(7I)
48  *	ioctl calls. vdc will copy the data to be written to the descriptor
49  *	ring or maps the buffer to store the data read by the vDisk
50  *	server into the descriptor ring. It then sends a message to the
51  *	vDisk server requesting it to complete the operation.
52  *
53  * 4) Handling responses from vDisk server.
54  *	The vDisk server will ACK some or all of the messages vdc sends to it
55  *	(this is configured during the handshake). Upon receipt of an ACK
56  *	vdc will check the descriptor ring and signal to the upper layer
57  *	code waiting on the IO.
58  */
59 
60 #include <sys/atomic.h>
61 #include <sys/conf.h>
62 #include <sys/disp.h>
63 #include <sys/ddi.h>
64 #include <sys/dkio.h>
65 #include <sys/efi_partition.h>
66 #include <sys/fcntl.h>
67 #include <sys/file.h>
68 #include <sys/mach_descrip.h>
69 #include <sys/modctl.h>
70 #include <sys/mdeg.h>
71 #include <sys/note.h>
72 #include <sys/open.h>
73 #include <sys/sdt.h>
74 #include <sys/stat.h>
75 #include <sys/sunddi.h>
76 #include <sys/types.h>
77 #include <sys/promif.h>
78 #include <sys/vtoc.h>
79 #include <sys/archsystm.h>
80 #include <sys/sysmacros.h>
81 
82 #include <sys/cdio.h>
83 #include <sys/dktp/fdisk.h>
84 #include <sys/dktp/dadkio.h>
85 #include <sys/scsi/generic/sense.h>
86 #include <sys/scsi/impl/uscsi.h>	/* Needed for defn of USCSICMD ioctl */
87 
88 #include <sys/ldoms.h>
89 #include <sys/ldc.h>
90 #include <sys/vio_common.h>
91 #include <sys/vio_mailbox.h>
92 #include <sys/vdsk_common.h>
93 #include <sys/vdsk_mailbox.h>
94 #include <sys/vdc.h>
95 
96 /*
97  * function prototypes
98  */
99 
100 /* standard driver functions */
101 static int	vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred);
102 static int	vdc_close(dev_t dev, int flag, int otyp, cred_t *cred);
103 static int	vdc_strategy(struct buf *buf);
104 static int	vdc_print(dev_t dev, char *str);
105 static int	vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk);
106 static int	vdc_read(dev_t dev, struct uio *uio, cred_t *cred);
107 static int	vdc_write(dev_t dev, struct uio *uio, cred_t *cred);
108 static int	vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode,
109 			cred_t *credp, int *rvalp);
110 static int	vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred);
111 static int	vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred);
112 
113 static int	vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,
114 			void *arg, void **resultp);
115 static int	vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd);
116 static int	vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd);
117 
118 /* setup */
119 static void	vdc_min(struct buf *bufp);
120 static int	vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen);
121 static int	vdc_do_ldc_init(vdc_t *vdc, md_t *mdp, mde_cookie_t vd_node);
122 static int	vdc_start_ldc_connection(vdc_t *vdc);
123 static int	vdc_create_device_nodes(vdc_t *vdc);
124 static int	vdc_create_device_nodes_efi(vdc_t *vdc);
125 static int	vdc_create_device_nodes_vtoc(vdc_t *vdc);
126 static int	vdc_create_device_nodes_props(vdc_t *vdc);
127 static int	vdc_get_md_node(dev_info_t *dip, md_t **mdpp,
128 		    mde_cookie_t *vd_nodep, mde_cookie_t *vd_portp);
129 static int	vdc_get_ldc_id(md_t *, mde_cookie_t, uint64_t *);
130 static int	vdc_do_ldc_up(vdc_t *vdc);
131 static void	vdc_terminate_ldc(vdc_t *vdc);
132 static int	vdc_init_descriptor_ring(vdc_t *vdc);
133 static void	vdc_destroy_descriptor_ring(vdc_t *vdc);
134 static int	vdc_setup_devid(vdc_t *vdc);
135 static void	vdc_store_label_efi(vdc_t *vdc, struct dk_gpt *efi);
136 static void	vdc_store_label_vtoc(vdc_t *, struct dk_geom *, struct vtoc *);
137 static void	vdc_store_label_unk(vdc_t *vdc);
138 static boolean_t vdc_is_opened(vdc_t *vdc);
139 
140 /* handshake with vds */
141 static int		vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver);
142 static int		vdc_ver_negotiation(vdc_t *vdcp);
143 static int		vdc_init_attr_negotiation(vdc_t *vdc);
144 static int		vdc_attr_negotiation(vdc_t *vdcp);
145 static int		vdc_init_dring_negotiate(vdc_t *vdc);
146 static int		vdc_dring_negotiation(vdc_t *vdcp);
147 static int		vdc_send_rdx(vdc_t *vdcp);
148 static int		vdc_rdx_exchange(vdc_t *vdcp);
149 static boolean_t	vdc_is_supported_version(vio_ver_msg_t *ver_msg);
150 
151 /* processing incoming messages from vDisk server */
152 static void	vdc_process_msg_thread(vdc_t *vdc);
153 static int	vdc_recv(vdc_t *vdc, vio_msg_t *msgp, size_t *nbytesp);
154 
155 static uint_t	vdc_handle_cb(uint64_t event, caddr_t arg);
156 static int	vdc_process_data_msg(vdc_t *vdc, vio_msg_t *msg);
157 static int	vdc_process_err_msg(vdc_t *vdc, vio_msg_t msg);
158 static int	vdc_handle_ver_msg(vdc_t *vdc, vio_ver_msg_t *ver_msg);
159 static int	vdc_handle_attr_msg(vdc_t *vdc, vd_attr_msg_t *attr_msg);
160 static int	vdc_handle_dring_reg_msg(vdc_t *vdc, vio_dring_reg_msg_t *msg);
161 static int 	vdc_send_request(vdc_t *vdcp, int operation,
162 		    caddr_t addr, size_t nbytes, int slice, diskaddr_t offset,
163 		    int cb_type, void *cb_arg, vio_desc_direction_t dir);
164 static int	vdc_map_to_shared_dring(vdc_t *vdcp, int idx);
165 static int 	vdc_populate_descriptor(vdc_t *vdcp, int operation,
166 		    caddr_t addr, size_t nbytes, int slice, diskaddr_t offset,
167 		    int cb_type, void *cb_arg, vio_desc_direction_t dir);
168 static int 	vdc_do_sync_op(vdc_t *vdcp, int operation,
169 		    caddr_t addr, size_t nbytes, int slice, diskaddr_t offset,
170 		    int cb_type, void *cb_arg, vio_desc_direction_t dir);
171 
172 static int	vdc_wait_for_response(vdc_t *vdcp, vio_msg_t *msgp);
173 static int	vdc_drain_response(vdc_t *vdcp);
174 static int	vdc_depopulate_descriptor(vdc_t *vdc, uint_t idx);
175 static int	vdc_populate_mem_hdl(vdc_t *vdcp, vdc_local_desc_t *ldep);
176 static int	vdc_verify_seq_num(vdc_t *vdc, vio_dring_msg_t *dring_msg);
177 
178 /* dkio */
179 static int	vd_process_ioctl(dev_t dev, int cmd, caddr_t arg, int mode);
180 static void	vdc_create_fake_geometry(vdc_t *vdc);
181 static int	vdc_validate_geometry(vdc_t *vdc);
182 static void	vdc_validate(vdc_t *vdc);
183 static void	vdc_validate_task(void *arg);
184 static int	vdc_null_copy_func(vdc_t *vdc, void *from, void *to,
185 		    int mode, int dir);
186 static int	vdc_get_wce_convert(vdc_t *vdc, void *from, void *to,
187 		    int mode, int dir);
188 static int	vdc_set_wce_convert(vdc_t *vdc, void *from, void *to,
189 		    int mode, int dir);
190 static int	vdc_get_vtoc_convert(vdc_t *vdc, void *from, void *to,
191 		    int mode, int dir);
192 static int	vdc_set_vtoc_convert(vdc_t *vdc, void *from, void *to,
193 		    int mode, int dir);
194 static int	vdc_get_geom_convert(vdc_t *vdc, void *from, void *to,
195 		    int mode, int dir);
196 static int	vdc_set_geom_convert(vdc_t *vdc, void *from, void *to,
197 		    int mode, int dir);
198 static int	vdc_uscsicmd_convert(vdc_t *vdc, void *from, void *to,
199 		    int mode, int dir);
200 static int	vdc_get_efi_convert(vdc_t *vdc, void *from, void *to,
201 		    int mode, int dir);
202 static int	vdc_set_efi_convert(vdc_t *vdc, void *from, void *to,
203 		    int mode, int dir);
204 
205 /*
206  * Module variables
207  */
208 
209 /*
210  * Tunable variables to control how long vdc waits before timing out on
211  * various operations
212  */
213 static int	vdc_retries = 10;
214 static int	vdc_hshake_retries = 3;
215 
216 static int	vdc_timeout = 0; /* units: seconds */
217 
218 /* calculated from 'vdc_usec_timeout' during attach */
219 static uint64_t	vdc_hz_timeout;				/* units: Hz */
220 static uint64_t	vdc_usec_timeout = 30 * MICROSEC;	/* 30s units: ns */
221 
222 static uint64_t vdc_hz_min_ldc_delay;
223 static uint64_t vdc_min_timeout_ldc = 1 * MILLISEC;
224 static uint64_t vdc_hz_max_ldc_delay;
225 static uint64_t vdc_max_timeout_ldc = 100 * MILLISEC;
226 
227 static uint64_t vdc_ldc_read_init_delay = 1 * MILLISEC;
228 static uint64_t vdc_ldc_read_max_delay = 100 * MILLISEC;
229 
230 /* values for dumping - need to run in a tighter loop */
231 static uint64_t	vdc_usec_timeout_dump = 100 * MILLISEC;	/* 0.1s units: ns */
232 static int	vdc_dump_retries = 100;
233 
234 /* Count of the number of vdc instances attached */
235 static volatile uint32_t	vdc_instance_count = 0;
236 
237 /* Soft state pointer */
238 static void	*vdc_state;
239 
240 /*
241  * Controlling the verbosity of the error/debug messages
242  *
243  * vdc_msglevel - controls level of messages
244  * vdc_matchinst - 64-bit variable where each bit corresponds
245  *                 to the vdc instance the vdc_msglevel applies.
246  */
247 int		vdc_msglevel = 0x0;
248 uint64_t	vdc_matchinst = 0ull;
249 
250 /*
251  * Supported vDisk protocol version pairs.
252  *
253  * The first array entry is the latest and preferred version.
254  */
255 static const vio_ver_t	vdc_version[] = {{1, 0}};
256 
257 static struct cb_ops vdc_cb_ops = {
258 	vdc_open,	/* cb_open */
259 	vdc_close,	/* cb_close */
260 	vdc_strategy,	/* cb_strategy */
261 	vdc_print,	/* cb_print */
262 	vdc_dump,	/* cb_dump */
263 	vdc_read,	/* cb_read */
264 	vdc_write,	/* cb_write */
265 	vdc_ioctl,	/* cb_ioctl */
266 	nodev,		/* cb_devmap */
267 	nodev,		/* cb_mmap */
268 	nodev,		/* cb_segmap */
269 	nochpoll,	/* cb_chpoll */
270 	ddi_prop_op,	/* cb_prop_op */
271 	NULL,		/* cb_str */
272 	D_MP | D_64BIT,	/* cb_flag */
273 	CB_REV,		/* cb_rev */
274 	vdc_aread,	/* cb_aread */
275 	vdc_awrite	/* cb_awrite */
276 };
277 
278 static struct dev_ops vdc_ops = {
279 	DEVO_REV,	/* devo_rev */
280 	0,		/* devo_refcnt */
281 	vdc_getinfo,	/* devo_getinfo */
282 	nulldev,	/* devo_identify */
283 	nulldev,	/* devo_probe */
284 	vdc_attach,	/* devo_attach */
285 	vdc_detach,	/* devo_detach */
286 	nodev,		/* devo_reset */
287 	&vdc_cb_ops,	/* devo_cb_ops */
288 	NULL,		/* devo_bus_ops */
289 	nulldev		/* devo_power */
290 };
291 
292 static struct modldrv modldrv = {
293 	&mod_driverops,
294 	"virtual disk client",
295 	&vdc_ops,
296 };
297 
298 static struct modlinkage modlinkage = {
299 	MODREV_1,
300 	&modldrv,
301 	NULL
302 };
303 
304 /* -------------------------------------------------------------------------- */
305 
306 /*
307  * Device Driver housekeeping and setup
308  */
309 
310 int
311 _init(void)
312 {
313 	int	status;
314 
315 	if ((status = ddi_soft_state_init(&vdc_state, sizeof (vdc_t), 1)) != 0)
316 		return (status);
317 	if ((status = mod_install(&modlinkage)) != 0)
318 		ddi_soft_state_fini(&vdc_state);
319 	vdc_efi_init(vd_process_ioctl);
320 	return (status);
321 }
322 
323 int
324 _info(struct modinfo *modinfop)
325 {
326 	return (mod_info(&modlinkage, modinfop));
327 }
328 
329 int
330 _fini(void)
331 {
332 	int	status;
333 
334 	if ((status = mod_remove(&modlinkage)) != 0)
335 		return (status);
336 	vdc_efi_fini();
337 	ddi_soft_state_fini(&vdc_state);
338 	return (0);
339 }
340 
341 static int
342 vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,  void *arg, void **resultp)
343 {
344 	_NOTE(ARGUNUSED(dip))
345 
346 	int	instance = VDCUNIT((dev_t)arg);
347 	vdc_t	*vdc = NULL;
348 
349 	switch (cmd) {
350 	case DDI_INFO_DEVT2DEVINFO:
351 		if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
352 			*resultp = NULL;
353 			return (DDI_FAILURE);
354 		}
355 		*resultp = vdc->dip;
356 		return (DDI_SUCCESS);
357 	case DDI_INFO_DEVT2INSTANCE:
358 		*resultp = (void *)(uintptr_t)instance;
359 		return (DDI_SUCCESS);
360 	default:
361 		*resultp = NULL;
362 		return (DDI_FAILURE);
363 	}
364 }
365 
366 static int
367 vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
368 {
369 	int	instance;
370 	int	rv;
371 	vdc_t	*vdc = NULL;
372 
373 	switch (cmd) {
374 	case DDI_DETACH:
375 		/* the real work happens below */
376 		break;
377 	case DDI_SUSPEND:
378 		/* nothing to do for this non-device */
379 		return (DDI_SUCCESS);
380 	default:
381 		return (DDI_FAILURE);
382 	}
383 
384 	ASSERT(cmd == DDI_DETACH);
385 	instance = ddi_get_instance(dip);
386 	DMSGX(1, "[%d] Entered\n", instance);
387 
388 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
389 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
390 		return (DDI_FAILURE);
391 	}
392 
393 	if (vdc_is_opened(vdc)) {
394 		DMSG(vdc, 0, "[%d] Cannot detach: device is open", instance);
395 		return (DDI_FAILURE);
396 	}
397 
398 	if (vdc->dkio_flush_pending) {
399 		DMSG(vdc, 0,
400 		    "[%d] Cannot detach: %d outstanding DKIO flushes\n",
401 		    instance, vdc->dkio_flush_pending);
402 		return (DDI_FAILURE);
403 	}
404 
405 	if (vdc->validate_pending) {
406 		DMSG(vdc, 0,
407 		    "[%d] Cannot detach: %d outstanding validate request\n",
408 		    instance, vdc->validate_pending);
409 		return (DDI_FAILURE);
410 	}
411 
412 	DMSG(vdc, 0, "[%d] proceeding...\n", instance);
413 
414 	/* mark instance as detaching */
415 	vdc->lifecycle	= VDC_LC_DETACHING;
416 
417 	/*
418 	 * try and disable callbacks to prevent another handshake
419 	 */
420 	rv = ldc_set_cb_mode(vdc->ldc_handle, LDC_CB_DISABLE);
421 	DMSG(vdc, 0, "callback disabled (rv=%d)\n", rv);
422 
423 	if (vdc->initialized & VDC_THREAD) {
424 		mutex_enter(&vdc->read_lock);
425 		if ((vdc->read_state == VDC_READ_WAITING) ||
426 		    (vdc->read_state == VDC_READ_RESET)) {
427 			vdc->read_state = VDC_READ_RESET;
428 			cv_signal(&vdc->read_cv);
429 		}
430 
431 		mutex_exit(&vdc->read_lock);
432 
433 		/* wake up any thread waiting for connection to come online */
434 		mutex_enter(&vdc->lock);
435 		if (vdc->state == VDC_STATE_INIT_WAITING) {
436 			DMSG(vdc, 0,
437 			    "[%d] write reset - move to resetting state...\n",
438 			    instance);
439 			vdc->state = VDC_STATE_RESETTING;
440 			cv_signal(&vdc->initwait_cv);
441 		}
442 		mutex_exit(&vdc->lock);
443 
444 		/* now wait until state transitions to VDC_STATE_DETACH */
445 		thread_join(vdc->msg_proc_thr->t_did);
446 		ASSERT(vdc->state == VDC_STATE_DETACH);
447 		DMSG(vdc, 0, "[%d] Reset thread exit and join ..\n",
448 		    vdc->instance);
449 	}
450 
451 	mutex_enter(&vdc->lock);
452 
453 	if (vdc->initialized & VDC_DRING)
454 		vdc_destroy_descriptor_ring(vdc);
455 
456 	if (vdc->initialized & VDC_LDC)
457 		vdc_terminate_ldc(vdc);
458 
459 	mutex_exit(&vdc->lock);
460 
461 	if (vdc->initialized & VDC_MINOR) {
462 		ddi_prop_remove_all(dip);
463 		ddi_remove_minor_node(dip, NULL);
464 	}
465 
466 	if (vdc->initialized & VDC_LOCKS) {
467 		mutex_destroy(&vdc->lock);
468 		mutex_destroy(&vdc->read_lock);
469 		cv_destroy(&vdc->initwait_cv);
470 		cv_destroy(&vdc->dring_free_cv);
471 		cv_destroy(&vdc->membind_cv);
472 		cv_destroy(&vdc->sync_pending_cv);
473 		cv_destroy(&vdc->sync_blocked_cv);
474 		cv_destroy(&vdc->read_cv);
475 		cv_destroy(&vdc->running_cv);
476 	}
477 
478 	if (vdc->minfo)
479 		kmem_free(vdc->minfo, sizeof (struct dk_minfo));
480 
481 	if (vdc->cinfo)
482 		kmem_free(vdc->cinfo, sizeof (struct dk_cinfo));
483 
484 	if (vdc->vtoc)
485 		kmem_free(vdc->vtoc, sizeof (struct vtoc));
486 
487 	if (vdc->geom)
488 		kmem_free(vdc->geom, sizeof (struct dk_geom));
489 
490 	if (vdc->devid) {
491 		ddi_devid_unregister(dip);
492 		ddi_devid_free(vdc->devid);
493 	}
494 
495 	if (vdc->initialized & VDC_SOFT_STATE)
496 		ddi_soft_state_free(vdc_state, instance);
497 
498 	DMSG(vdc, 0, "[%d] End %p\n", instance, (void *)vdc);
499 
500 	return (DDI_SUCCESS);
501 }
502 
503 
504 static int
505 vdc_do_attach(dev_info_t *dip)
506 {
507 	int		instance;
508 	vdc_t		*vdc = NULL;
509 	int		status;
510 	md_t		*mdp;
511 	mde_cookie_t	vd_node, vd_port;
512 
513 	ASSERT(dip != NULL);
514 
515 	instance = ddi_get_instance(dip);
516 	if (ddi_soft_state_zalloc(vdc_state, instance) != DDI_SUCCESS) {
517 		cmn_err(CE_NOTE, "[%d] Couldn't alloc state structure",
518 		    instance);
519 		return (DDI_FAILURE);
520 	}
521 
522 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
523 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
524 		return (DDI_FAILURE);
525 	}
526 
527 	/*
528 	 * We assign the value to initialized in this case to zero out the
529 	 * variable and then set bits in it to indicate what has been done
530 	 */
531 	vdc->initialized = VDC_SOFT_STATE;
532 
533 	vdc_hz_timeout = drv_usectohz(vdc_usec_timeout);
534 
535 	vdc_hz_min_ldc_delay = drv_usectohz(vdc_min_timeout_ldc);
536 	vdc_hz_max_ldc_delay = drv_usectohz(vdc_max_timeout_ldc);
537 
538 	vdc->dip	= dip;
539 	vdc->instance	= instance;
540 	vdc->vdisk_type	= VD_DISK_TYPE_UNK;
541 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
542 	vdc->state	= VDC_STATE_INIT;
543 	vdc->lifecycle	= VDC_LC_ATTACHING;
544 	vdc->ldc_state	= 0;
545 	vdc->session_id = 0;
546 	vdc->block_size = DEV_BSIZE;
547 	vdc->max_xfer_sz = maxphys / DEV_BSIZE;
548 
549 	vdc->vtoc = NULL;
550 	vdc->geom = NULL;
551 	vdc->cinfo = NULL;
552 	vdc->minfo = NULL;
553 
554 	mutex_init(&vdc->lock, NULL, MUTEX_DRIVER, NULL);
555 	cv_init(&vdc->initwait_cv, NULL, CV_DRIVER, NULL);
556 	cv_init(&vdc->dring_free_cv, NULL, CV_DRIVER, NULL);
557 	cv_init(&vdc->membind_cv, NULL, CV_DRIVER, NULL);
558 	cv_init(&vdc->running_cv, NULL, CV_DRIVER, NULL);
559 
560 	vdc->threads_pending = 0;
561 	vdc->sync_op_pending = B_FALSE;
562 	vdc->sync_op_blocked = B_FALSE;
563 	cv_init(&vdc->sync_pending_cv, NULL, CV_DRIVER, NULL);
564 	cv_init(&vdc->sync_blocked_cv, NULL, CV_DRIVER, NULL);
565 
566 	/* init blocking msg read functionality */
567 	mutex_init(&vdc->read_lock, NULL, MUTEX_DRIVER, NULL);
568 	cv_init(&vdc->read_cv, NULL, CV_DRIVER, NULL);
569 	vdc->read_state = VDC_READ_IDLE;
570 
571 	vdc->initialized |= VDC_LOCKS;
572 
573 	/* get device and port MD node for this disk instance */
574 	if (vdc_get_md_node(dip, &mdp, &vd_node, &vd_port) != 0) {
575 		cmn_err(CE_NOTE, "[%d] Could not get machine description node",
576 		    instance);
577 		return (DDI_FAILURE);
578 	}
579 
580 	/* set the connection timeout */
581 	if (vd_port == NULL || (md_get_prop_val(mdp, vd_port,
582 	    VDC_MD_TIMEOUT, &vdc->ctimeout) != 0)) {
583 		vdc->ctimeout = 0;
584 	}
585 
586 	/* initialise LDC channel which will be used to communicate with vds */
587 	status = vdc_do_ldc_init(vdc, mdp, vd_node);
588 
589 	(void) md_fini_handle(mdp);
590 
591 	if (status != 0) {
592 		cmn_err(CE_NOTE, "[%d] Couldn't initialize LDC", instance);
593 		goto return_status;
594 	}
595 
596 	/* initialize the thread responsible for managing state with server */
597 	vdc->msg_proc_thr = thread_create(NULL, 0, vdc_process_msg_thread,
598 	    vdc, 0, &p0, TS_RUN, minclsyspri);
599 	if (vdc->msg_proc_thr == NULL) {
600 		cmn_err(CE_NOTE, "[%d] Failed to create msg processing thread",
601 		    instance);
602 		return (DDI_FAILURE);
603 	}
604 
605 	vdc->initialized |= VDC_THREAD;
606 
607 	atomic_inc_32(&vdc_instance_count);
608 
609 	/*
610 	 * Check the disk label. This will send requests and do the handshake.
611 	 * We don't really care about the disk label now. What we really need is
612 	 * the handshake do be done so that we know the type of the disk (slice
613 	 * or full disk) and the appropriate device nodes can be created.
614 	 */
615 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
616 	vdc->vtoc = kmem_zalloc(sizeof (struct vtoc), KM_SLEEP);
617 	vdc->geom = kmem_zalloc(sizeof (struct dk_geom), KM_SLEEP);
618 
619 	mutex_enter(&vdc->lock);
620 	(void) vdc_validate_geometry(vdc);
621 	mutex_exit(&vdc->lock);
622 
623 	/*
624 	 * Now that we have the device info we can create the
625 	 * device nodes and properties
626 	 */
627 	status = vdc_create_device_nodes(vdc);
628 	if (status) {
629 		DMSG(vdc, 0, "[%d] Failed to create device nodes",
630 		    instance);
631 		goto return_status;
632 	}
633 	status = vdc_create_device_nodes_props(vdc);
634 	if (status) {
635 		DMSG(vdc, 0, "[%d] Failed to create device nodes"
636 		    " properties (%d)", instance, status);
637 		goto return_status;
638 	}
639 
640 	/*
641 	 * Setup devid
642 	 */
643 	if (vdc_setup_devid(vdc)) {
644 		DMSG(vdc, 0, "[%d] No device id available\n", instance);
645 	}
646 
647 	ddi_report_dev(dip);
648 	vdc->lifecycle	= VDC_LC_ONLINE;
649 	DMSG(vdc, 0, "[%d] Attach tasks successful\n", instance);
650 
651 return_status:
652 	DMSG(vdc, 0, "[%d] Attach completed\n", instance);
653 	return (status);
654 }
655 
656 static int
657 vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
658 {
659 	int	status;
660 
661 	switch (cmd) {
662 	case DDI_ATTACH:
663 		if ((status = vdc_do_attach(dip)) != 0)
664 			(void) vdc_detach(dip, DDI_DETACH);
665 		return (status);
666 	case DDI_RESUME:
667 		/* nothing to do for this non-device */
668 		return (DDI_SUCCESS);
669 	default:
670 		return (DDI_FAILURE);
671 	}
672 }
673 
674 static int
675 vdc_do_ldc_init(vdc_t *vdc, md_t *mdp, mde_cookie_t vd_node)
676 {
677 	int			status = 0;
678 	ldc_status_t		ldc_state;
679 	ldc_attr_t		ldc_attr;
680 	uint64_t		ldc_id = 0;
681 
682 	ASSERT(vdc != NULL);
683 
684 	vdc->initialized |= VDC_LDC;
685 
686 	if ((status = vdc_get_ldc_id(mdp, vd_node, &ldc_id)) != 0) {
687 		DMSG(vdc, 0, "[%d] Failed to get LDC channel ID property",
688 		    vdc->instance);
689 		return (EIO);
690 	}
691 
692 	DMSGX(0, "[%d] LDC id is 0x%lx\n", vdc->instance, ldc_id);
693 
694 	vdc->ldc_id = ldc_id;
695 
696 	ldc_attr.devclass = LDC_DEV_BLK;
697 	ldc_attr.instance = vdc->instance;
698 	ldc_attr.mode = LDC_MODE_UNRELIABLE;	/* unreliable transport */
699 	ldc_attr.mtu = VD_LDC_MTU;
700 
701 	if ((vdc->initialized & VDC_LDC_INIT) == 0) {
702 		status = ldc_init(ldc_id, &ldc_attr, &vdc->ldc_handle);
703 		if (status != 0) {
704 			DMSG(vdc, 0, "[%d] ldc_init(chan %ld) returned %d",
705 			    vdc->instance, ldc_id, status);
706 			return (status);
707 		}
708 		vdc->initialized |= VDC_LDC_INIT;
709 	}
710 	status = ldc_status(vdc->ldc_handle, &ldc_state);
711 	if (status != 0) {
712 		DMSG(vdc, 0, "[%d] Cannot discover LDC status [err=%d]",
713 		    vdc->instance, status);
714 		return (status);
715 	}
716 	vdc->ldc_state = ldc_state;
717 
718 	if ((vdc->initialized & VDC_LDC_CB) == 0) {
719 		status = ldc_reg_callback(vdc->ldc_handle, vdc_handle_cb,
720 		    (caddr_t)vdc);
721 		if (status != 0) {
722 			DMSG(vdc, 0, "[%d] LDC callback reg. failed (%d)",
723 			    vdc->instance, status);
724 			return (status);
725 		}
726 		vdc->initialized |= VDC_LDC_CB;
727 	}
728 
729 	vdc->initialized |= VDC_LDC;
730 
731 	/*
732 	 * At this stage we have initialised LDC, we will now try and open
733 	 * the connection.
734 	 */
735 	if (vdc->ldc_state == LDC_INIT) {
736 		status = ldc_open(vdc->ldc_handle);
737 		if (status != 0) {
738 			DMSG(vdc, 0, "[%d] ldc_open(chan %ld) returned %d",
739 			    vdc->instance, vdc->ldc_id, status);
740 			return (status);
741 		}
742 		vdc->initialized |= VDC_LDC_OPEN;
743 	}
744 
745 	return (status);
746 }
747 
748 static int
749 vdc_start_ldc_connection(vdc_t *vdc)
750 {
751 	int		status = 0;
752 
753 	ASSERT(vdc != NULL);
754 
755 	ASSERT(MUTEX_HELD(&vdc->lock));
756 
757 	status = vdc_do_ldc_up(vdc);
758 
759 	DMSG(vdc, 0, "[%d] Finished bringing up LDC\n", vdc->instance);
760 
761 	return (status);
762 }
763 
764 static int
765 vdc_stop_ldc_connection(vdc_t *vdcp)
766 {
767 	int	status;
768 
769 	DMSG(vdcp, 0, ": Resetting connection to vDisk server : state %d\n",
770 	    vdcp->state);
771 
772 	status = ldc_down(vdcp->ldc_handle);
773 	DMSG(vdcp, 0, "ldc_down() = %d\n", status);
774 
775 	vdcp->initialized &= ~VDC_HANDSHAKE;
776 	DMSG(vdcp, 0, "initialized=%x\n", vdcp->initialized);
777 
778 	return (status);
779 }
780 
781 static int
782 vdc_create_device_nodes_efi(vdc_t *vdc)
783 {
784 	ddi_remove_minor_node(vdc->dip, "h");
785 	ddi_remove_minor_node(vdc->dip, "h,raw");
786 
787 	if (ddi_create_minor_node(vdc->dip, "wd", S_IFBLK,
788 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
789 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
790 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'wd'",
791 		    vdc->instance);
792 		return (EIO);
793 	}
794 
795 	/* if any device node is created we set this flag */
796 	vdc->initialized |= VDC_MINOR;
797 
798 	if (ddi_create_minor_node(vdc->dip, "wd,raw", S_IFCHR,
799 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
800 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
801 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'wd,raw'",
802 		    vdc->instance);
803 		return (EIO);
804 	}
805 
806 	return (0);
807 }
808 
809 static int
810 vdc_create_device_nodes_vtoc(vdc_t *vdc)
811 {
812 	ddi_remove_minor_node(vdc->dip, "wd");
813 	ddi_remove_minor_node(vdc->dip, "wd,raw");
814 
815 	if (ddi_create_minor_node(vdc->dip, "h", S_IFBLK,
816 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
817 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
818 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'h'",
819 		    vdc->instance);
820 		return (EIO);
821 	}
822 
823 	/* if any device node is created we set this flag */
824 	vdc->initialized |= VDC_MINOR;
825 
826 	if (ddi_create_minor_node(vdc->dip, "h,raw", S_IFCHR,
827 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
828 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
829 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'h,raw'",
830 		    vdc->instance);
831 		return (EIO);
832 	}
833 
834 	return (0);
835 }
836 
837 /*
838  * Function:
839  *	vdc_create_device_nodes
840  *
841  * Description:
842  *	This function creates the block and character device nodes under
843  *	/devices along with the node properties. It is called as part of
844  *	the attach(9E) of the instance during the handshake with vds after
845  *	vds has sent the attributes to vdc.
846  *
847  *	If the device is of type VD_DISK_TYPE_SLICE then the minor node
848  *	of 2 is used in keeping with the Solaris convention that slice 2
849  *	refers to a whole disk. Slices start at 'a'
850  *
851  * Parameters:
852  *	vdc 		- soft state pointer
853  *
854  * Return Values
855  *	0		- Success
856  *	EIO		- Failed to create node
857  *	EINVAL		- Unknown type of disk exported
858  */
859 static int
860 vdc_create_device_nodes(vdc_t *vdc)
861 {
862 	char		name[sizeof ("s,raw")];
863 	dev_info_t	*dip = NULL;
864 	int		instance, status;
865 	int		num_slices = 1;
866 	int		i;
867 
868 	ASSERT(vdc != NULL);
869 
870 	instance = vdc->instance;
871 	dip = vdc->dip;
872 
873 	switch (vdc->vdisk_type) {
874 	case VD_DISK_TYPE_DISK:
875 		num_slices = V_NUMPAR;
876 		break;
877 	case VD_DISK_TYPE_SLICE:
878 		num_slices = 1;
879 		break;
880 	case VD_DISK_TYPE_UNK:
881 	default:
882 		return (EINVAL);
883 	}
884 
885 	/*
886 	 * Minor nodes are different for EFI disks: EFI disks do not have
887 	 * a minor node 'g' for the minor number corresponding to slice
888 	 * VD_EFI_WD_SLICE (slice 7) instead they have a minor node 'wd'
889 	 * representing the whole disk.
890 	 */
891 	for (i = 0; i < num_slices; i++) {
892 
893 		if (i == VD_EFI_WD_SLICE) {
894 			if (vdc->vdisk_label == VD_DISK_LABEL_EFI)
895 				status = vdc_create_device_nodes_efi(vdc);
896 			else
897 				status = vdc_create_device_nodes_vtoc(vdc);
898 			if (status != 0)
899 				return (status);
900 			continue;
901 		}
902 
903 		(void) snprintf(name, sizeof (name), "%c", 'a' + i);
904 		if (ddi_create_minor_node(dip, name, S_IFBLK,
905 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
906 			cmn_err(CE_NOTE, "[%d] Couldn't add block node '%s'",
907 			    instance, name);
908 			return (EIO);
909 		}
910 
911 		/* if any device node is created we set this flag */
912 		vdc->initialized |= VDC_MINOR;
913 
914 		(void) snprintf(name, sizeof (name), "%c%s", 'a' + i, ",raw");
915 
916 		if (ddi_create_minor_node(dip, name, S_IFCHR,
917 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
918 			cmn_err(CE_NOTE, "[%d] Couldn't add raw node '%s'",
919 			    instance, name);
920 			return (EIO);
921 		}
922 	}
923 
924 	return (0);
925 }
926 
927 /*
928  * Function:
929  *	vdc_create_device_nodes_props
930  *
931  * Description:
932  *	This function creates the block and character device nodes under
933  *	/devices along with the node properties. It is called as part of
934  *	the attach(9E) of the instance during the handshake with vds after
935  *	vds has sent the attributes to vdc.
936  *
937  * Parameters:
938  *	vdc 		- soft state pointer
939  *
940  * Return Values
941  *	0		- Success
942  *	EIO		- Failed to create device node property
943  *	EINVAL		- Unknown type of disk exported
944  */
945 static int
946 vdc_create_device_nodes_props(vdc_t *vdc)
947 {
948 	dev_info_t	*dip = NULL;
949 	int		instance;
950 	int		num_slices = 1;
951 	int64_t		size = 0;
952 	dev_t		dev;
953 	int		rv;
954 	int		i;
955 
956 	ASSERT(vdc != NULL);
957 	ASSERT(vdc->vtoc != NULL);
958 
959 	instance = vdc->instance;
960 	dip = vdc->dip;
961 
962 	switch (vdc->vdisk_type) {
963 	case VD_DISK_TYPE_DISK:
964 		num_slices = V_NUMPAR;
965 		break;
966 	case VD_DISK_TYPE_SLICE:
967 		num_slices = 1;
968 		break;
969 	case VD_DISK_TYPE_UNK:
970 	default:
971 		return (EINVAL);
972 	}
973 
974 	if (vdc->vdisk_label == VD_DISK_LABEL_UNK) {
975 		/* remove all properties */
976 		for (i = 0; i < num_slices; i++) {
977 			dev = makedevice(ddi_driver_major(dip),
978 			    VD_MAKE_DEV(instance, i));
979 			(void) ddi_prop_remove(dev, dip, VDC_SIZE_PROP_NAME);
980 			(void) ddi_prop_remove(dev, dip, VDC_NBLOCKS_PROP_NAME);
981 		}
982 		return (0);
983 	}
984 
985 	for (i = 0; i < num_slices; i++) {
986 		dev = makedevice(ddi_driver_major(dip),
987 		    VD_MAKE_DEV(instance, i));
988 
989 		size = vdc->vtoc->v_part[i].p_size * vdc->vtoc->v_sectorsz;
990 		DMSG(vdc, 0, "[%d] sz %ld (%ld Mb)  p_size %lx\n",
991 		    instance, size, size / (1024 * 1024),
992 		    vdc->vtoc->v_part[i].p_size);
993 
994 		rv = ddi_prop_update_int64(dev, dip, VDC_SIZE_PROP_NAME, size);
995 		if (rv != DDI_PROP_SUCCESS) {
996 			cmn_err(CE_NOTE, "[%d] Couldn't add '%s' prop of [%ld]",
997 			    instance, VDC_SIZE_PROP_NAME, size);
998 			return (EIO);
999 		}
1000 
1001 		rv = ddi_prop_update_int64(dev, dip, VDC_NBLOCKS_PROP_NAME,
1002 		    lbtodb(size));
1003 		if (rv != DDI_PROP_SUCCESS) {
1004 			cmn_err(CE_NOTE, "[%d] Couldn't add '%s' prop [%llu]",
1005 			    instance, VDC_NBLOCKS_PROP_NAME, lbtodb(size));
1006 			return (EIO);
1007 		}
1008 	}
1009 
1010 	return (0);
1011 }
1012 
1013 /*
1014  * Function:
1015  *	vdc_is_opened
1016  *
1017  * Description:
1018  *	This function checks if any slice of a given virtual disk is
1019  *	currently opened.
1020  *
1021  * Parameters:
1022  *	vdc 		- soft state pointer
1023  *
1024  * Return Values
1025  *	B_TRUE		- at least one slice is opened.
1026  *	B_FALSE		- no slice is opened.
1027  */
1028 static boolean_t
1029 vdc_is_opened(vdc_t *vdc)
1030 {
1031 	int i, nslices;
1032 
1033 	switch (vdc->vdisk_type) {
1034 	case VD_DISK_TYPE_DISK:
1035 		nslices = V_NUMPAR;
1036 		break;
1037 	case VD_DISK_TYPE_SLICE:
1038 		nslices = 1;
1039 		break;
1040 	case VD_DISK_TYPE_UNK:
1041 	default:
1042 		ASSERT(0);
1043 	}
1044 
1045 	/* check if there's any layered open */
1046 	for (i = 0; i < nslices; i++) {
1047 		if (vdc->open_lyr[i] > 0)
1048 			return (B_TRUE);
1049 	}
1050 
1051 	/* check if there is any other kind of open */
1052 	for (i = 0; i < OTYPCNT; i++) {
1053 		if (vdc->open[i] != 0)
1054 			return (B_TRUE);
1055 	}
1056 
1057 	return (B_FALSE);
1058 }
1059 
1060 static int
1061 vdc_mark_opened(vdc_t *vdc, int slice, int flag, int otyp)
1062 {
1063 	uint8_t slicemask;
1064 	int i;
1065 
1066 	ASSERT(otyp < OTYPCNT);
1067 	ASSERT(slice < V_NUMPAR);
1068 	ASSERT(MUTEX_HELD(&vdc->lock));
1069 
1070 	slicemask = 1 << slice;
1071 
1072 	/* check if slice is already exclusively opened */
1073 	if (vdc->open_excl & slicemask)
1074 		return (EBUSY);
1075 
1076 	/* if open exclusive, check if slice is already opened */
1077 	if (flag & FEXCL) {
1078 		if (vdc->open_lyr[slice] > 0)
1079 			return (EBUSY);
1080 		for (i = 0; i < OTYPCNT; i++) {
1081 			if (vdc->open[i] & slicemask)
1082 				return (EBUSY);
1083 		}
1084 		vdc->open_excl |= slicemask;
1085 	}
1086 
1087 	/* mark slice as opened */
1088 	if (otyp == OTYP_LYR) {
1089 		vdc->open_lyr[slice]++;
1090 	} else {
1091 		vdc->open[otyp] |= slicemask;
1092 	}
1093 
1094 	return (0);
1095 }
1096 
1097 static void
1098 vdc_mark_closed(vdc_t *vdc, int slice, int flag, int otyp)
1099 {
1100 	uint8_t slicemask;
1101 
1102 	ASSERT(otyp < OTYPCNT);
1103 	ASSERT(slice < V_NUMPAR);
1104 	ASSERT(MUTEX_HELD(&vdc->lock));
1105 
1106 	slicemask = 1 << slice;
1107 
1108 	if (otyp == OTYP_LYR) {
1109 		ASSERT(vdc->open_lyr[slice] > 0);
1110 		vdc->open_lyr[slice]--;
1111 	} else {
1112 		vdc->open[otyp] &= ~slicemask;
1113 	}
1114 
1115 	if (flag & FEXCL)
1116 		vdc->open_excl &= ~slicemask;
1117 }
1118 
1119 static int
1120 vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred)
1121 {
1122 	_NOTE(ARGUNUSED(cred))
1123 
1124 	int	instance;
1125 	int	slice, status = 0;
1126 	vdc_t	*vdc;
1127 
1128 	ASSERT(dev != NULL);
1129 	instance = VDCUNIT(*dev);
1130 
1131 	if (otyp >= OTYPCNT)
1132 		return (EINVAL);
1133 
1134 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1135 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1136 		return (ENXIO);
1137 	}
1138 
1139 	DMSG(vdc, 0, "minor = %d flag = %x, otyp = %x\n",
1140 	    getminor(*dev), flag, otyp);
1141 
1142 	slice = VDCPART(*dev);
1143 
1144 	mutex_enter(&vdc->lock);
1145 
1146 	status = vdc_mark_opened(vdc, slice, flag, otyp);
1147 
1148 	if (status != 0) {
1149 		mutex_exit(&vdc->lock);
1150 		return (status);
1151 	}
1152 
1153 	if (flag & (FNDELAY | FNONBLOCK)) {
1154 
1155 		/* don't resubmit a validate request if there's already one */
1156 		if (vdc->validate_pending > 0) {
1157 			mutex_exit(&vdc->lock);
1158 			return (0);
1159 		}
1160 
1161 		/* call vdc_validate() asynchronously to avoid blocking */
1162 		if (taskq_dispatch(system_taskq, vdc_validate_task,
1163 		    (void *)vdc, TQ_NOSLEEP) == NULL) {
1164 			vdc_mark_closed(vdc, slice, flag, otyp);
1165 			mutex_exit(&vdc->lock);
1166 			return (ENXIO);
1167 		}
1168 
1169 		vdc->validate_pending++;
1170 		mutex_exit(&vdc->lock);
1171 		return (0);
1172 	}
1173 
1174 	mutex_exit(&vdc->lock);
1175 
1176 	vdc_validate(vdc);
1177 
1178 	mutex_enter(&vdc->lock);
1179 
1180 	if (vdc->vdisk_label == VD_DISK_LABEL_UNK ||
1181 	    vdc->vtoc->v_part[slice].p_size == 0) {
1182 		vdc_mark_closed(vdc, slice, flag, otyp);
1183 		status = EIO;
1184 	}
1185 
1186 	mutex_exit(&vdc->lock);
1187 
1188 	return (status);
1189 }
1190 
1191 static int
1192 vdc_close(dev_t dev, int flag, int otyp, cred_t *cred)
1193 {
1194 	_NOTE(ARGUNUSED(cred))
1195 
1196 	int	instance;
1197 	int	slice;
1198 	int	rv;
1199 	vdc_t	*vdc;
1200 
1201 	instance = VDCUNIT(dev);
1202 
1203 	if (otyp >= OTYPCNT)
1204 		return (EINVAL);
1205 
1206 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1207 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1208 		return (ENXIO);
1209 	}
1210 
1211 	DMSG(vdc, 0, "[%d] flag = %x, otyp = %x\n", instance, flag, otyp);
1212 
1213 	slice = VDCPART(dev);
1214 
1215 	/*
1216 	 * Attempt to flush the W$ on a close operation. If this is
1217 	 * not a supported IOCTL command or the backing device is read-only
1218 	 * do not fail the close operation.
1219 	 */
1220 	rv = vd_process_ioctl(dev, DKIOCFLUSHWRITECACHE, NULL, FKIOCTL);
1221 
1222 	if (rv != 0 && rv != ENOTSUP && rv != ENOTTY && rv != EROFS) {
1223 		DMSG(vdc, 0, "[%d] flush failed with error %d on close\n",
1224 		    instance, rv);
1225 		return (EIO);
1226 	}
1227 
1228 	mutex_enter(&vdc->lock);
1229 	vdc_mark_closed(vdc, slice, flag, otyp);
1230 	mutex_exit(&vdc->lock);
1231 
1232 	return (0);
1233 }
1234 
1235 static int
1236 vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
1237 {
1238 	_NOTE(ARGUNUSED(credp))
1239 	_NOTE(ARGUNUSED(rvalp))
1240 
1241 	return (vd_process_ioctl(dev, cmd, (caddr_t)arg, mode));
1242 }
1243 
1244 static int
1245 vdc_print(dev_t dev, char *str)
1246 {
1247 	cmn_err(CE_NOTE, "vdc%d:  %s", VDCUNIT(dev), str);
1248 	return (0);
1249 }
1250 
1251 static int
1252 vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk)
1253 {
1254 	int	rv;
1255 	size_t	nbytes = nblk * DEV_BSIZE;
1256 	int	instance = VDCUNIT(dev);
1257 	vdc_t	*vdc = NULL;
1258 
1259 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1260 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1261 		return (ENXIO);
1262 	}
1263 
1264 	DMSG(vdc, 2, "[%d] dump %ld bytes at block 0x%lx : addr=0x%p\n",
1265 	    instance, nbytes, blkno, (void *)addr);
1266 	rv = vdc_send_request(vdc, VD_OP_BWRITE, addr, nbytes,
1267 	    VDCPART(dev), blkno, CB_STRATEGY, 0, VIO_write_dir);
1268 	if (rv) {
1269 		DMSG(vdc, 0, "Failed to do a disk dump (err=%d)\n", rv);
1270 		return (rv);
1271 	}
1272 
1273 	if (ddi_in_panic())
1274 		(void) vdc_drain_response(vdc);
1275 
1276 	DMSG(vdc, 0, "[%d] End\n", instance);
1277 
1278 	return (0);
1279 }
1280 
1281 /* -------------------------------------------------------------------------- */
1282 
1283 /*
1284  * Disk access routines
1285  *
1286  */
1287 
1288 /*
1289  * vdc_strategy()
1290  *
1291  * Return Value:
1292  *	0:	As per strategy(9E), the strategy() function must return 0
1293  *		[ bioerror(9f) sets b_flags to the proper error code ]
1294  */
1295 static int
1296 vdc_strategy(struct buf *buf)
1297 {
1298 	int	rv = -1;
1299 	vdc_t	*vdc = NULL;
1300 	int	instance = VDCUNIT(buf->b_edev);
1301 	int	op = (buf->b_flags & B_READ) ? VD_OP_BREAD : VD_OP_BWRITE;
1302 	int	slice;
1303 
1304 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1305 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1306 		bioerror(buf, ENXIO);
1307 		biodone(buf);
1308 		return (0);
1309 	}
1310 
1311 	DMSG(vdc, 2, "[%d] %s %ld bytes at block %llx : b_addr=0x%p\n",
1312 	    instance, (buf->b_flags & B_READ) ? "Read" : "Write",
1313 	    buf->b_bcount, buf->b_lblkno, (void *)buf->b_un.b_addr);
1314 	DTRACE_IO2(vstart, buf_t *, buf, vdc_t *, vdc);
1315 
1316 	bp_mapin(buf);
1317 
1318 	if ((long)buf->b_private == VD_SLICE_NONE) {
1319 		/* I/O using an absolute disk offset */
1320 		slice = VD_SLICE_NONE;
1321 	} else {
1322 		slice = VDCPART(buf->b_edev);
1323 	}
1324 
1325 	rv = vdc_send_request(vdc, op, (caddr_t)buf->b_un.b_addr,
1326 	    buf->b_bcount, slice, buf->b_lblkno,
1327 	    CB_STRATEGY, buf, (op == VD_OP_BREAD) ? VIO_read_dir :
1328 	    VIO_write_dir);
1329 
1330 	/*
1331 	 * If the request was successfully sent, the strategy call returns and
1332 	 * the ACK handler calls the bioxxx functions when the vDisk server is
1333 	 * done.
1334 	 */
1335 	if (rv) {
1336 		DMSG(vdc, 0, "Failed to read/write (err=%d)\n", rv);
1337 		bioerror(buf, rv);
1338 		biodone(buf);
1339 	}
1340 
1341 	return (0);
1342 }
1343 
1344 /*
1345  * Function:
1346  *	vdc_min
1347  *
1348  * Description:
1349  *	Routine to limit the size of a data transfer. Used in
1350  *	conjunction with physio(9F).
1351  *
1352  * Arguments:
1353  *	bp - pointer to the indicated buf(9S) struct.
1354  *
1355  */
1356 static void
1357 vdc_min(struct buf *bufp)
1358 {
1359 	vdc_t	*vdc = NULL;
1360 	int	instance = VDCUNIT(bufp->b_edev);
1361 
1362 	vdc = ddi_get_soft_state(vdc_state, instance);
1363 	VERIFY(vdc != NULL);
1364 
1365 	if (bufp->b_bcount > (vdc->max_xfer_sz * vdc->block_size)) {
1366 		bufp->b_bcount = vdc->max_xfer_sz * vdc->block_size;
1367 	}
1368 }
1369 
1370 static int
1371 vdc_read(dev_t dev, struct uio *uio, cred_t *cred)
1372 {
1373 	_NOTE(ARGUNUSED(cred))
1374 
1375 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1376 	return (physio(vdc_strategy, NULL, dev, B_READ, vdc_min, uio));
1377 }
1378 
1379 static int
1380 vdc_write(dev_t dev, struct uio *uio, cred_t *cred)
1381 {
1382 	_NOTE(ARGUNUSED(cred))
1383 
1384 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1385 	return (physio(vdc_strategy, NULL, dev, B_WRITE, vdc_min, uio));
1386 }
1387 
1388 static int
1389 vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred)
1390 {
1391 	_NOTE(ARGUNUSED(cred))
1392 
1393 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1394 	return (aphysio(vdc_strategy, anocancel, dev, B_READ, vdc_min, aio));
1395 }
1396 
1397 static int
1398 vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred)
1399 {
1400 	_NOTE(ARGUNUSED(cred))
1401 
1402 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1403 	return (aphysio(vdc_strategy, anocancel, dev, B_WRITE, vdc_min, aio));
1404 }
1405 
1406 
1407 /* -------------------------------------------------------------------------- */
1408 
1409 /*
1410  * Handshake support
1411  */
1412 
1413 
1414 /*
1415  * Function:
1416  *	vdc_init_ver_negotiation()
1417  *
1418  * Description:
1419  *
1420  * Arguments:
1421  *	vdc	- soft state pointer for this instance of the device driver.
1422  *
1423  * Return Code:
1424  *	0	- Success
1425  */
1426 static int
1427 vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver)
1428 {
1429 	vio_ver_msg_t	pkt;
1430 	size_t		msglen = sizeof (pkt);
1431 	int		status = -1;
1432 
1433 	ASSERT(vdc != NULL);
1434 	ASSERT(mutex_owned(&vdc->lock));
1435 
1436 	DMSG(vdc, 0, "[%d] Entered.\n", vdc->instance);
1437 
1438 	/*
1439 	 * set the Session ID to a unique value
1440 	 * (the lower 32 bits of the clock tick)
1441 	 */
1442 	vdc->session_id = ((uint32_t)gettick() & 0xffffffff);
1443 	DMSG(vdc, 0, "[%d] Set SID to 0x%lx\n", vdc->instance, vdc->session_id);
1444 
1445 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1446 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1447 	pkt.tag.vio_subtype_env = VIO_VER_INFO;
1448 	pkt.tag.vio_sid = vdc->session_id;
1449 	pkt.dev_class = VDEV_DISK;
1450 	pkt.ver_major = ver.major;
1451 	pkt.ver_minor = ver.minor;
1452 
1453 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1454 	DMSG(vdc, 0, "[%d] Ver info sent (status = %d)\n",
1455 	    vdc->instance, status);
1456 	if ((status != 0) || (msglen != sizeof (vio_ver_msg_t))) {
1457 		DMSG(vdc, 0, "[%d] Failed to send Ver negotiation info: "
1458 		    "id(%lx) rv(%d) size(%ld)", vdc->instance, vdc->ldc_handle,
1459 		    status, msglen);
1460 		if (msglen != sizeof (vio_ver_msg_t))
1461 			status = ENOMSG;
1462 	}
1463 
1464 	return (status);
1465 }
1466 
1467 /*
1468  * Function:
1469  *	vdc_ver_negotiation()
1470  *
1471  * Description:
1472  *
1473  * Arguments:
1474  *	vdcp	- soft state pointer for this instance of the device driver.
1475  *
1476  * Return Code:
1477  *	0	- Success
1478  */
1479 static int
1480 vdc_ver_negotiation(vdc_t *vdcp)
1481 {
1482 	vio_msg_t vio_msg;
1483 	int status;
1484 
1485 	if (status = vdc_init_ver_negotiation(vdcp, vdc_version[0]))
1486 		return (status);
1487 
1488 	/* release lock and wait for response */
1489 	mutex_exit(&vdcp->lock);
1490 	status = vdc_wait_for_response(vdcp, &vio_msg);
1491 	mutex_enter(&vdcp->lock);
1492 	if (status) {
1493 		DMSG(vdcp, 0,
1494 		    "[%d] Failed waiting for Ver negotiation response, rv(%d)",
1495 		    vdcp->instance, status);
1496 		return (status);
1497 	}
1498 
1499 	/* check type and sub_type ... */
1500 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1501 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1502 		DMSG(vdcp, 0, "[%d] Invalid ver negotiation response\n",
1503 		    vdcp->instance);
1504 		return (EPROTO);
1505 	}
1506 
1507 	return (vdc_handle_ver_msg(vdcp, (vio_ver_msg_t *)&vio_msg));
1508 }
1509 
1510 /*
1511  * Function:
1512  *	vdc_init_attr_negotiation()
1513  *
1514  * Description:
1515  *
1516  * Arguments:
1517  *	vdc	- soft state pointer for this instance of the device driver.
1518  *
1519  * Return Code:
1520  *	0	- Success
1521  */
1522 static int
1523 vdc_init_attr_negotiation(vdc_t *vdc)
1524 {
1525 	vd_attr_msg_t	pkt;
1526 	size_t		msglen = sizeof (pkt);
1527 	int		status;
1528 
1529 	ASSERT(vdc != NULL);
1530 	ASSERT(mutex_owned(&vdc->lock));
1531 
1532 	DMSG(vdc, 0, "[%d] entered\n", vdc->instance);
1533 
1534 	/* fill in tag */
1535 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1536 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1537 	pkt.tag.vio_subtype_env = VIO_ATTR_INFO;
1538 	pkt.tag.vio_sid = vdc->session_id;
1539 	/* fill in payload */
1540 	pkt.max_xfer_sz = vdc->max_xfer_sz;
1541 	pkt.vdisk_block_size = vdc->block_size;
1542 	pkt.xfer_mode = VIO_DRING_MODE;
1543 	pkt.operations = 0;	/* server will set bits of valid operations */
1544 	pkt.vdisk_type = 0;	/* server will set to valid device type */
1545 	pkt.vdisk_size = 0;	/* server will set to valid size */
1546 
1547 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1548 	DMSG(vdc, 0, "Attr info sent (status = %d)\n", status);
1549 
1550 	if ((status != 0) || (msglen != sizeof (vio_ver_msg_t))) {
1551 		DMSG(vdc, 0, "[%d] Failed to send Attr negotiation info: "
1552 		    "id(%lx) rv(%d) size(%ld)", vdc->instance, vdc->ldc_handle,
1553 		    status, msglen);
1554 		if (msglen != sizeof (vio_ver_msg_t))
1555 			status = ENOMSG;
1556 	}
1557 
1558 	return (status);
1559 }
1560 
1561 /*
1562  * Function:
1563  *	vdc_attr_negotiation()
1564  *
1565  * Description:
1566  *
1567  * Arguments:
1568  *	vdc	- soft state pointer for this instance of the device driver.
1569  *
1570  * Return Code:
1571  *	0	- Success
1572  */
1573 static int
1574 vdc_attr_negotiation(vdc_t *vdcp)
1575 {
1576 	int status;
1577 	vio_msg_t vio_msg;
1578 
1579 	if (status = vdc_init_attr_negotiation(vdcp))
1580 		return (status);
1581 
1582 	/* release lock and wait for response */
1583 	mutex_exit(&vdcp->lock);
1584 	status = vdc_wait_for_response(vdcp, &vio_msg);
1585 	mutex_enter(&vdcp->lock);
1586 	if (status) {
1587 		DMSG(vdcp, 0,
1588 		    "[%d] Failed waiting for Attr negotiation response, rv(%d)",
1589 		    vdcp->instance, status);
1590 		return (status);
1591 	}
1592 
1593 	/* check type and sub_type ... */
1594 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1595 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1596 		DMSG(vdcp, 0, "[%d] Invalid attr negotiation response\n",
1597 		    vdcp->instance);
1598 		return (EPROTO);
1599 	}
1600 
1601 	return (vdc_handle_attr_msg(vdcp, (vd_attr_msg_t *)&vio_msg));
1602 }
1603 
1604 
1605 /*
1606  * Function:
1607  *	vdc_init_dring_negotiate()
1608  *
1609  * Description:
1610  *
1611  * Arguments:
1612  *	vdc	- soft state pointer for this instance of the device driver.
1613  *
1614  * Return Code:
1615  *	0	- Success
1616  */
1617 static int
1618 vdc_init_dring_negotiate(vdc_t *vdc)
1619 {
1620 	vio_dring_reg_msg_t	pkt;
1621 	size_t			msglen = sizeof (pkt);
1622 	int			status = -1;
1623 	int			retry;
1624 	int			nretries = 10;
1625 
1626 	ASSERT(vdc != NULL);
1627 	ASSERT(mutex_owned(&vdc->lock));
1628 
1629 	for (retry = 0; retry < nretries; retry++) {
1630 		status = vdc_init_descriptor_ring(vdc);
1631 		if (status != EAGAIN)
1632 			break;
1633 		drv_usecwait(vdc_min_timeout_ldc);
1634 	}
1635 
1636 	if (status != 0) {
1637 		DMSG(vdc, 0, "[%d] Failed to init DRing (status = %d)\n",
1638 		    vdc->instance, status);
1639 		return (status);
1640 	}
1641 
1642 	DMSG(vdc, 0, "[%d] Init of descriptor ring completed (status = %d)\n",
1643 	    vdc->instance, status);
1644 
1645 	/* fill in tag */
1646 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1647 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1648 	pkt.tag.vio_subtype_env = VIO_DRING_REG;
1649 	pkt.tag.vio_sid = vdc->session_id;
1650 	/* fill in payload */
1651 	pkt.dring_ident = 0;
1652 	pkt.num_descriptors = vdc->dring_len;
1653 	pkt.descriptor_size = vdc->dring_entry_size;
1654 	pkt.options = (VIO_TX_DRING | VIO_RX_DRING);
1655 	pkt.ncookies = vdc->dring_cookie_count;
1656 	pkt.cookie[0] = vdc->dring_cookie[0];	/* for now just one cookie */
1657 
1658 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1659 	if (status != 0) {
1660 		DMSG(vdc, 0, "[%d] Failed to register DRing (err = %d)",
1661 		    vdc->instance, status);
1662 	}
1663 
1664 	return (status);
1665 }
1666 
1667 
1668 /*
1669  * Function:
1670  *	vdc_dring_negotiation()
1671  *
1672  * Description:
1673  *
1674  * Arguments:
1675  *	vdc	- soft state pointer for this instance of the device driver.
1676  *
1677  * Return Code:
1678  *	0	- Success
1679  */
1680 static int
1681 vdc_dring_negotiation(vdc_t *vdcp)
1682 {
1683 	int status;
1684 	vio_msg_t vio_msg;
1685 
1686 	if (status = vdc_init_dring_negotiate(vdcp))
1687 		return (status);
1688 
1689 	/* release lock and wait for response */
1690 	mutex_exit(&vdcp->lock);
1691 	status = vdc_wait_for_response(vdcp, &vio_msg);
1692 	mutex_enter(&vdcp->lock);
1693 	if (status) {
1694 		DMSG(vdcp, 0,
1695 		    "[%d] Failed waiting for Dring negotiation response,"
1696 		    " rv(%d)", vdcp->instance, status);
1697 		return (status);
1698 	}
1699 
1700 	/* check type and sub_type ... */
1701 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1702 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1703 		DMSG(vdcp, 0, "[%d] Invalid Dring negotiation response\n",
1704 		    vdcp->instance);
1705 		return (EPROTO);
1706 	}
1707 
1708 	return (vdc_handle_dring_reg_msg(vdcp,
1709 	    (vio_dring_reg_msg_t *)&vio_msg));
1710 }
1711 
1712 
1713 /*
1714  * Function:
1715  *	vdc_send_rdx()
1716  *
1717  * Description:
1718  *
1719  * Arguments:
1720  *	vdc	- soft state pointer for this instance of the device driver.
1721  *
1722  * Return Code:
1723  *	0	- Success
1724  */
1725 static int
1726 vdc_send_rdx(vdc_t *vdcp)
1727 {
1728 	vio_msg_t	msg;
1729 	size_t		msglen = sizeof (vio_msg_t);
1730 	int		status;
1731 
1732 	/*
1733 	 * Send an RDX message to vds to indicate we are ready
1734 	 * to send data
1735 	 */
1736 	msg.tag.vio_msgtype = VIO_TYPE_CTRL;
1737 	msg.tag.vio_subtype = VIO_SUBTYPE_INFO;
1738 	msg.tag.vio_subtype_env = VIO_RDX;
1739 	msg.tag.vio_sid = vdcp->session_id;
1740 	status = vdc_send(vdcp, (caddr_t)&msg, &msglen);
1741 	if (status != 0) {
1742 		DMSG(vdcp, 0, "[%d] Failed to send RDX message (%d)",
1743 		    vdcp->instance, status);
1744 	}
1745 
1746 	return (status);
1747 }
1748 
1749 /*
1750  * Function:
1751  *	vdc_handle_rdx()
1752  *
1753  * Description:
1754  *
1755  * Arguments:
1756  *	vdc	- soft state pointer for this instance of the device driver.
1757  *	msgp	- received msg
1758  *
1759  * Return Code:
1760  *	0	- Success
1761  */
1762 static int
1763 vdc_handle_rdx(vdc_t *vdcp, vio_rdx_msg_t *msgp)
1764 {
1765 	_NOTE(ARGUNUSED(vdcp))
1766 	_NOTE(ARGUNUSED(msgp))
1767 
1768 	ASSERT(msgp->tag.vio_msgtype == VIO_TYPE_CTRL);
1769 	ASSERT(msgp->tag.vio_subtype == VIO_SUBTYPE_ACK);
1770 	ASSERT(msgp->tag.vio_subtype_env == VIO_RDX);
1771 
1772 	DMSG(vdcp, 1, "[%d] Got an RDX msg", vdcp->instance);
1773 
1774 	return (0);
1775 }
1776 
1777 /*
1778  * Function:
1779  *	vdc_rdx_exchange()
1780  *
1781  * Description:
1782  *
1783  * Arguments:
1784  *	vdc	- soft state pointer for this instance of the device driver.
1785  *
1786  * Return Code:
1787  *	0	- Success
1788  */
1789 static int
1790 vdc_rdx_exchange(vdc_t *vdcp)
1791 {
1792 	int status;
1793 	vio_msg_t vio_msg;
1794 
1795 	if (status = vdc_send_rdx(vdcp))
1796 		return (status);
1797 
1798 	/* release lock and wait for response */
1799 	mutex_exit(&vdcp->lock);
1800 	status = vdc_wait_for_response(vdcp, &vio_msg);
1801 	mutex_enter(&vdcp->lock);
1802 	if (status) {
1803 		DMSG(vdcp, 0, "[%d] Failed waiting for RDX response, rv(%d)",
1804 		    vdcp->instance, status);
1805 		return (status);
1806 	}
1807 
1808 	/* check type and sub_type ... */
1809 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1810 	    vio_msg.tag.vio_subtype != VIO_SUBTYPE_ACK) {
1811 		DMSG(vdcp, 0, "[%d] Invalid RDX response\n", vdcp->instance);
1812 		return (EPROTO);
1813 	}
1814 
1815 	return (vdc_handle_rdx(vdcp, (vio_rdx_msg_t *)&vio_msg));
1816 }
1817 
1818 
1819 /* -------------------------------------------------------------------------- */
1820 
1821 /*
1822  * LDC helper routines
1823  */
1824 
1825 static int
1826 vdc_recv(vdc_t *vdc, vio_msg_t *msgp, size_t *nbytesp)
1827 {
1828 	int		status;
1829 	boolean_t	q_has_pkts = B_FALSE;
1830 	int		delay_time;
1831 	size_t		len;
1832 
1833 	mutex_enter(&vdc->read_lock);
1834 
1835 	if (vdc->read_state == VDC_READ_IDLE)
1836 		vdc->read_state = VDC_READ_WAITING;
1837 
1838 	while (vdc->read_state != VDC_READ_PENDING) {
1839 
1840 		/* detect if the connection has been reset */
1841 		if (vdc->read_state == VDC_READ_RESET) {
1842 			status = ECONNRESET;
1843 			goto done;
1844 		}
1845 
1846 		cv_wait(&vdc->read_cv, &vdc->read_lock);
1847 	}
1848 
1849 	/*
1850 	 * Until we get a blocking ldc read we have to retry
1851 	 * until the entire LDC message has arrived before
1852 	 * ldc_read() will succeed. Note we also bail out if
1853 	 * the channel is reset or goes away.
1854 	 */
1855 	delay_time = vdc_ldc_read_init_delay;
1856 loop:
1857 	len = *nbytesp;
1858 	status = ldc_read(vdc->ldc_handle, (caddr_t)msgp, &len);
1859 	switch (status) {
1860 	case EAGAIN:
1861 		delay_time *= 2;
1862 		if (delay_time >= vdc_ldc_read_max_delay)
1863 			delay_time = vdc_ldc_read_max_delay;
1864 		delay(delay_time);
1865 		goto loop;
1866 
1867 	case 0:
1868 		if (len == 0) {
1869 			DMSG(vdc, 0, "[%d] ldc_read returned 0 bytes with "
1870 			    "no error!\n", vdc->instance);
1871 			goto loop;
1872 		}
1873 
1874 		*nbytesp = len;
1875 
1876 		/*
1877 		 * If there are pending messages, leave the
1878 		 * read state as pending. Otherwise, set the state
1879 		 * back to idle.
1880 		 */
1881 		status = ldc_chkq(vdc->ldc_handle, &q_has_pkts);
1882 		if (status == 0 && !q_has_pkts)
1883 			vdc->read_state = VDC_READ_IDLE;
1884 
1885 		break;
1886 	default:
1887 		DMSG(vdc, 0, "ldc_read returned %d\n", status);
1888 		break;
1889 	}
1890 
1891 done:
1892 	mutex_exit(&vdc->read_lock);
1893 
1894 	return (status);
1895 }
1896 
1897 
1898 
1899 #ifdef DEBUG
1900 void
1901 vdc_decode_tag(vdc_t *vdcp, vio_msg_t *msg)
1902 {
1903 	char *ms, *ss, *ses;
1904 	switch (msg->tag.vio_msgtype) {
1905 #define	Q(_s)	case _s : ms = #_s; break;
1906 	Q(VIO_TYPE_CTRL)
1907 	Q(VIO_TYPE_DATA)
1908 	Q(VIO_TYPE_ERR)
1909 #undef Q
1910 	default: ms = "unknown"; break;
1911 	}
1912 
1913 	switch (msg->tag.vio_subtype) {
1914 #define	Q(_s)	case _s : ss = #_s; break;
1915 	Q(VIO_SUBTYPE_INFO)
1916 	Q(VIO_SUBTYPE_ACK)
1917 	Q(VIO_SUBTYPE_NACK)
1918 #undef Q
1919 	default: ss = "unknown"; break;
1920 	}
1921 
1922 	switch (msg->tag.vio_subtype_env) {
1923 #define	Q(_s)	case _s : ses = #_s; break;
1924 	Q(VIO_VER_INFO)
1925 	Q(VIO_ATTR_INFO)
1926 	Q(VIO_DRING_REG)
1927 	Q(VIO_DRING_UNREG)
1928 	Q(VIO_RDX)
1929 	Q(VIO_PKT_DATA)
1930 	Q(VIO_DESC_DATA)
1931 	Q(VIO_DRING_DATA)
1932 #undef Q
1933 	default: ses = "unknown"; break;
1934 	}
1935 
1936 	DMSG(vdcp, 3, "(%x/%x/%x) message : (%s/%s/%s)\n",
1937 	    msg->tag.vio_msgtype, msg->tag.vio_subtype,
1938 	    msg->tag.vio_subtype_env, ms, ss, ses);
1939 }
1940 #endif
1941 
1942 /*
1943  * Function:
1944  *	vdc_send()
1945  *
1946  * Description:
1947  *	The function encapsulates the call to write a message using LDC.
1948  *	If LDC indicates that the call failed due to the queue being full,
1949  *	we retry the ldc_write() [ up to 'vdc_retries' time ], otherwise
1950  *	we return the error returned by LDC.
1951  *
1952  * Arguments:
1953  *	ldc_handle	- LDC handle for the channel this instance of vdc uses
1954  *	pkt		- address of LDC message to be sent
1955  *	msglen		- the size of the message being sent. When the function
1956  *			  returns, this contains the number of bytes written.
1957  *
1958  * Return Code:
1959  *	0		- Success.
1960  *	EINVAL		- pkt or msglen were NULL
1961  *	ECONNRESET	- The connection was not up.
1962  *	EWOULDBLOCK	- LDC queue is full
1963  *	xxx		- other error codes returned by ldc_write
1964  */
1965 static int
1966 vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen)
1967 {
1968 	size_t	size = 0;
1969 	int	status = 0;
1970 	clock_t delay_ticks;
1971 
1972 	ASSERT(vdc != NULL);
1973 	ASSERT(mutex_owned(&vdc->lock));
1974 	ASSERT(msglen != NULL);
1975 	ASSERT(*msglen != 0);
1976 
1977 #ifdef DEBUG
1978 	vdc_decode_tag(vdc, (vio_msg_t *)pkt);
1979 #endif
1980 	/*
1981 	 * Wait indefinitely to send if channel
1982 	 * is busy, but bail out if we succeed or
1983 	 * if the channel closes or is reset.
1984 	 */
1985 	delay_ticks = vdc_hz_min_ldc_delay;
1986 	do {
1987 		size = *msglen;
1988 		status = ldc_write(vdc->ldc_handle, pkt, &size);
1989 		if (status == EWOULDBLOCK) {
1990 			delay(delay_ticks);
1991 			/* geometric backoff */
1992 			delay_ticks *= 2;
1993 			if (delay_ticks > vdc_hz_max_ldc_delay)
1994 				delay_ticks = vdc_hz_max_ldc_delay;
1995 		}
1996 	} while (status == EWOULDBLOCK);
1997 
1998 	/* if LDC had serious issues --- reset vdc state */
1999 	if (status == EIO || status == ECONNRESET) {
2000 		/* LDC had serious issues --- reset vdc state */
2001 		mutex_enter(&vdc->read_lock);
2002 		if ((vdc->read_state == VDC_READ_WAITING) ||
2003 		    (vdc->read_state == VDC_READ_RESET))
2004 			cv_signal(&vdc->read_cv);
2005 		vdc->read_state = VDC_READ_RESET;
2006 		mutex_exit(&vdc->read_lock);
2007 
2008 		/* wake up any waiters in the reset thread */
2009 		if (vdc->state == VDC_STATE_INIT_WAITING) {
2010 			DMSG(vdc, 0, "[%d] write reset - "
2011 			    "vdc is resetting ..\n", vdc->instance);
2012 			vdc->state = VDC_STATE_RESETTING;
2013 			cv_signal(&vdc->initwait_cv);
2014 		}
2015 
2016 		return (ECONNRESET);
2017 	}
2018 
2019 	/* return the last size written */
2020 	*msglen = size;
2021 
2022 	return (status);
2023 }
2024 
2025 /*
2026  * Function:
2027  *	vdc_get_md_node
2028  *
2029  * Description:
2030  *	Get the MD, the device node and the port node for the given
2031  *	disk instance. The caller is responsible for cleaning up the
2032  *	reference to the returned MD (mdpp) by calling md_fini_handle().
2033  *
2034  * Arguments:
2035  *	dip	- dev info pointer for this instance of the device driver.
2036  *	mdpp	- the returned MD.
2037  *	vd_nodep - the returned device node.
2038  *	vd_portp - the returned port node. The returned port node is NULL
2039  *		   if no port node is found.
2040  *
2041  * Return Code:
2042  *	0	- Success.
2043  *	ENOENT	- Expected node or property did not exist.
2044  *	ENXIO	- Unexpected error communicating with MD framework
2045  */
2046 static int
2047 vdc_get_md_node(dev_info_t *dip, md_t **mdpp, mde_cookie_t *vd_nodep,
2048     mde_cookie_t *vd_portp)
2049 {
2050 	int		status = ENOENT;
2051 	char		*node_name = NULL;
2052 	md_t		*mdp = NULL;
2053 	int		num_nodes;
2054 	int		num_vdevs;
2055 	int		num_vports;
2056 	mde_cookie_t	rootnode;
2057 	mde_cookie_t	*listp = NULL;
2058 	boolean_t	found_inst = B_FALSE;
2059 	int		listsz;
2060 	int		idx;
2061 	uint64_t	md_inst;
2062 	int		obp_inst;
2063 	int		instance = ddi_get_instance(dip);
2064 
2065 	/*
2066 	 * Get the OBP instance number for comparison with the MD instance
2067 	 *
2068 	 * The "cfg-handle" property of a vdc node in an MD contains the MD's
2069 	 * notion of "instance", or unique identifier, for that node; OBP
2070 	 * stores the value of the "cfg-handle" MD property as the value of
2071 	 * the "reg" property on the node in the device tree it builds from
2072 	 * the MD and passes to Solaris.  Thus, we look up the devinfo node's
2073 	 * "reg" property value to uniquely identify this device instance.
2074 	 * If the "reg" property cannot be found, the device tree state is
2075 	 * presumably so broken that there is no point in continuing.
2076 	 */
2077 	if (!ddi_prop_exists(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS, OBP_REG)) {
2078 		cmn_err(CE_WARN, "'%s' property does not exist", OBP_REG);
2079 		return (ENOENT);
2080 	}
2081 	obp_inst = ddi_prop_get_int(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
2082 	    OBP_REG, -1);
2083 	DMSGX(1, "[%d] OBP inst=%d\n", instance, obp_inst);
2084 
2085 	/*
2086 	 * We now walk the MD nodes to find the node for this vdisk.
2087 	 */
2088 	if ((mdp = md_get_handle()) == NULL) {
2089 		cmn_err(CE_WARN, "unable to init machine description");
2090 		return (ENXIO);
2091 	}
2092 
2093 	num_nodes = md_node_count(mdp);
2094 	ASSERT(num_nodes > 0);
2095 
2096 	listsz = num_nodes * sizeof (mde_cookie_t);
2097 
2098 	/* allocate memory for nodes */
2099 	listp = kmem_zalloc(listsz, KM_SLEEP);
2100 
2101 	rootnode = md_root_node(mdp);
2102 	ASSERT(rootnode != MDE_INVAL_ELEM_COOKIE);
2103 
2104 	/*
2105 	 * Search for all the virtual devices, we will then check to see which
2106 	 * ones are disk nodes.
2107 	 */
2108 	num_vdevs = md_scan_dag(mdp, rootnode,
2109 	    md_find_name(mdp, VDC_MD_VDEV_NAME),
2110 	    md_find_name(mdp, "fwd"), listp);
2111 
2112 	if (num_vdevs <= 0) {
2113 		cmn_err(CE_NOTE, "No '%s' node found", VDC_MD_VDEV_NAME);
2114 		status = ENOENT;
2115 		goto done;
2116 	}
2117 
2118 	DMSGX(1, "[%d] num_vdevs=%d\n", instance, num_vdevs);
2119 	for (idx = 0; idx < num_vdevs; idx++) {
2120 		status = md_get_prop_str(mdp, listp[idx], "name", &node_name);
2121 		if ((status != 0) || (node_name == NULL)) {
2122 			cmn_err(CE_NOTE, "Unable to get name of node type '%s'"
2123 			    ": err %d", VDC_MD_VDEV_NAME, status);
2124 			continue;
2125 		}
2126 
2127 		DMSGX(1, "[%d] Found node '%s'\n", instance, node_name);
2128 		if (strcmp(VDC_MD_DISK_NAME, node_name) == 0) {
2129 			status = md_get_prop_val(mdp, listp[idx],
2130 			    VDC_MD_CFG_HDL, &md_inst);
2131 			DMSGX(1, "[%d] vdc inst in MD=%lx\n",
2132 			    instance, md_inst);
2133 			if ((status == 0) && (md_inst == obp_inst)) {
2134 				found_inst = B_TRUE;
2135 				break;
2136 			}
2137 		}
2138 	}
2139 
2140 	if (!found_inst) {
2141 		DMSGX(0, "Unable to find correct '%s' node", VDC_MD_DISK_NAME);
2142 		status = ENOENT;
2143 		goto done;
2144 	}
2145 	DMSGX(0, "[%d] MD inst=%lx\n", instance, md_inst);
2146 
2147 	*vd_nodep = listp[idx];
2148 	*mdpp = mdp;
2149 
2150 	num_vports = md_scan_dag(mdp, *vd_nodep,
2151 	    md_find_name(mdp, VDC_MD_PORT_NAME),
2152 	    md_find_name(mdp, "fwd"), listp);
2153 
2154 	if (num_vports != 1) {
2155 		DMSGX(0, "Expected 1 '%s' node for '%s' port, found %d\n",
2156 		    VDC_MD_PORT_NAME, VDC_MD_VDEV_NAME, num_vports);
2157 	}
2158 
2159 	*vd_portp = (num_vports == 0)? NULL: listp[0];
2160 
2161 done:
2162 	kmem_free(listp, listsz);
2163 	return (status);
2164 }
2165 
2166 /*
2167  * Function:
2168  *	vdc_get_ldc_id()
2169  *
2170  * Description:
2171  *	This function gets the 'ldc-id' for this particular instance of vdc.
2172  *	The id returned is the guest domain channel endpoint LDC uses for
2173  *	communication with vds.
2174  *
2175  * Arguments:
2176  *	mdp	- pointer to the machine description.
2177  *	vd_node	- the vdisk element from the MD.
2178  *	ldc_id	- pointer to variable used to return the 'ldc-id' found.
2179  *
2180  * Return Code:
2181  *	0	- Success.
2182  *	ENOENT	- Expected node or property did not exist.
2183  */
2184 static int
2185 vdc_get_ldc_id(md_t *mdp, mde_cookie_t vd_node, uint64_t *ldc_id)
2186 {
2187 	mde_cookie_t	*chanp = NULL;
2188 	int		listsz;
2189 	int		num_chans;
2190 	int		num_nodes;
2191 	int		status = 0;
2192 
2193 	num_nodes = md_node_count(mdp);
2194 	ASSERT(num_nodes > 0);
2195 
2196 	listsz = num_nodes * sizeof (mde_cookie_t);
2197 
2198 	/* allocate memory for nodes */
2199 	chanp = kmem_zalloc(listsz, KM_SLEEP);
2200 
2201 	/* get the channels for this node */
2202 	num_chans = md_scan_dag(mdp, vd_node,
2203 	    md_find_name(mdp, VDC_MD_CHAN_NAME),
2204 	    md_find_name(mdp, "fwd"), chanp);
2205 
2206 	/* expecting at least one channel */
2207 	if (num_chans <= 0) {
2208 		cmn_err(CE_NOTE, "No '%s' node for '%s' port",
2209 		    VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME);
2210 		status = ENOENT;
2211 		goto done;
2212 
2213 	} else if (num_chans != 1) {
2214 		DMSGX(0, "Expected 1 '%s' node for '%s' port, found %d\n",
2215 		    VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME, num_chans);
2216 	}
2217 
2218 	/*
2219 	 * We use the first channel found (index 0), irrespective of how
2220 	 * many are there in total.
2221 	 */
2222 	if (md_get_prop_val(mdp, chanp[0], VDC_MD_ID, ldc_id) != 0) {
2223 		cmn_err(CE_NOTE, "Channel '%s' property not found", VDC_MD_ID);
2224 		status = ENOENT;
2225 	}
2226 
2227 done:
2228 	kmem_free(chanp, listsz);
2229 	return (status);
2230 }
2231 
2232 static int
2233 vdc_do_ldc_up(vdc_t *vdc)
2234 {
2235 	int		status;
2236 	ldc_status_t	ldc_state;
2237 
2238 	DMSG(vdc, 0, "[%d] Bringing up channel %lx\n",
2239 	    vdc->instance, vdc->ldc_id);
2240 
2241 	if (vdc->lifecycle == VDC_LC_DETACHING)
2242 		return (EINVAL);
2243 
2244 	if ((status = ldc_up(vdc->ldc_handle)) != 0) {
2245 		switch (status) {
2246 		case ECONNREFUSED:	/* listener not ready at other end */
2247 			DMSG(vdc, 0, "[%d] ldc_up(%lx,...) return %d\n",
2248 			    vdc->instance, vdc->ldc_id, status);
2249 			status = 0;
2250 			break;
2251 		default:
2252 			DMSG(vdc, 0, "[%d] Failed to bring up LDC: "
2253 			    "channel=%ld, err=%d", vdc->instance, vdc->ldc_id,
2254 			    status);
2255 			break;
2256 		}
2257 	}
2258 
2259 	if (ldc_status(vdc->ldc_handle, &ldc_state) == 0) {
2260 		vdc->ldc_state = ldc_state;
2261 		if (ldc_state == LDC_UP) {
2262 			DMSG(vdc, 0, "[%d] LDC channel already up\n",
2263 			    vdc->instance);
2264 			vdc->seq_num = 1;
2265 			vdc->seq_num_reply = 0;
2266 		}
2267 	}
2268 
2269 	return (status);
2270 }
2271 
2272 /*
2273  * Function:
2274  *	vdc_terminate_ldc()
2275  *
2276  * Description:
2277  *
2278  * Arguments:
2279  *	vdc	- soft state pointer for this instance of the device driver.
2280  *
2281  * Return Code:
2282  *	None
2283  */
2284 static void
2285 vdc_terminate_ldc(vdc_t *vdc)
2286 {
2287 	int	instance = ddi_get_instance(vdc->dip);
2288 
2289 	ASSERT(vdc != NULL);
2290 	ASSERT(mutex_owned(&vdc->lock));
2291 
2292 	DMSG(vdc, 0, "[%d] initialized=%x\n", instance, vdc->initialized);
2293 
2294 	if (vdc->initialized & VDC_LDC_OPEN) {
2295 		DMSG(vdc, 0, "[%d] ldc_close()\n", instance);
2296 		(void) ldc_close(vdc->ldc_handle);
2297 	}
2298 	if (vdc->initialized & VDC_LDC_CB) {
2299 		DMSG(vdc, 0, "[%d] ldc_unreg_callback()\n", instance);
2300 		(void) ldc_unreg_callback(vdc->ldc_handle);
2301 	}
2302 	if (vdc->initialized & VDC_LDC) {
2303 		DMSG(vdc, 0, "[%d] ldc_fini()\n", instance);
2304 		(void) ldc_fini(vdc->ldc_handle);
2305 		vdc->ldc_handle = NULL;
2306 	}
2307 
2308 	vdc->initialized &= ~(VDC_LDC | VDC_LDC_CB | VDC_LDC_OPEN);
2309 }
2310 
2311 /* -------------------------------------------------------------------------- */
2312 
2313 /*
2314  * Descriptor Ring helper routines
2315  */
2316 
2317 /*
2318  * Function:
2319  *	vdc_init_descriptor_ring()
2320  *
2321  * Description:
2322  *
2323  * Arguments:
2324  *	vdc	- soft state pointer for this instance of the device driver.
2325  *
2326  * Return Code:
2327  *	0	- Success
2328  */
2329 static int
2330 vdc_init_descriptor_ring(vdc_t *vdc)
2331 {
2332 	vd_dring_entry_t	*dep = NULL;	/* DRing Entry pointer */
2333 	int	status = 0;
2334 	int	i;
2335 
2336 	DMSG(vdc, 0, "[%d] initialized=%x\n", vdc->instance, vdc->initialized);
2337 
2338 	ASSERT(vdc != NULL);
2339 	ASSERT(mutex_owned(&vdc->lock));
2340 	ASSERT(vdc->ldc_handle != NULL);
2341 
2342 	/* ensure we have enough room to store max sized block */
2343 	ASSERT(maxphys <= VD_MAX_BLOCK_SIZE);
2344 
2345 	if ((vdc->initialized & VDC_DRING_INIT) == 0) {
2346 		DMSG(vdc, 0, "[%d] ldc_mem_dring_create\n", vdc->instance);
2347 		/*
2348 		 * Calculate the maximum block size we can transmit using one
2349 		 * Descriptor Ring entry from the attributes returned by the
2350 		 * vDisk server. This is subject to a minimum of 'maxphys'
2351 		 * as we do not have the capability to split requests over
2352 		 * multiple DRing entries.
2353 		 */
2354 		if ((vdc->max_xfer_sz * vdc->block_size) < maxphys) {
2355 			DMSG(vdc, 0, "[%d] using minimum DRing size\n",
2356 			    vdc->instance);
2357 			vdc->dring_max_cookies = maxphys / PAGESIZE;
2358 		} else {
2359 			vdc->dring_max_cookies =
2360 			    (vdc->max_xfer_sz * vdc->block_size) / PAGESIZE;
2361 		}
2362 		vdc->dring_entry_size = (sizeof (vd_dring_entry_t) +
2363 		    (sizeof (ldc_mem_cookie_t) *
2364 		    (vdc->dring_max_cookies - 1)));
2365 		vdc->dring_len = VD_DRING_LEN;
2366 
2367 		status = ldc_mem_dring_create(vdc->dring_len,
2368 		    vdc->dring_entry_size, &vdc->ldc_dring_hdl);
2369 		if ((vdc->ldc_dring_hdl == NULL) || (status != 0)) {
2370 			DMSG(vdc, 0, "[%d] Descriptor ring creation failed",
2371 			    vdc->instance);
2372 			return (status);
2373 		}
2374 		vdc->initialized |= VDC_DRING_INIT;
2375 	}
2376 
2377 	if ((vdc->initialized & VDC_DRING_BOUND) == 0) {
2378 		DMSG(vdc, 0, "[%d] ldc_mem_dring_bind\n", vdc->instance);
2379 		vdc->dring_cookie =
2380 		    kmem_zalloc(sizeof (ldc_mem_cookie_t), KM_SLEEP);
2381 
2382 		status = ldc_mem_dring_bind(vdc->ldc_handle, vdc->ldc_dring_hdl,
2383 		    LDC_SHADOW_MAP|LDC_DIRECT_MAP, LDC_MEM_RW,
2384 		    &vdc->dring_cookie[0],
2385 		    &vdc->dring_cookie_count);
2386 		if (status != 0) {
2387 			DMSG(vdc, 0, "[%d] Failed to bind descriptor ring "
2388 			    "(%lx) to channel (%lx) status=%d\n",
2389 			    vdc->instance, vdc->ldc_dring_hdl,
2390 			    vdc->ldc_handle, status);
2391 			return (status);
2392 		}
2393 		ASSERT(vdc->dring_cookie_count == 1);
2394 		vdc->initialized |= VDC_DRING_BOUND;
2395 	}
2396 
2397 	status = ldc_mem_dring_info(vdc->ldc_dring_hdl, &vdc->dring_mem_info);
2398 	if (status != 0) {
2399 		DMSG(vdc, 0,
2400 		    "[%d] Failed to get info for descriptor ring (%lx)\n",
2401 		    vdc->instance, vdc->ldc_dring_hdl);
2402 		return (status);
2403 	}
2404 
2405 	if ((vdc->initialized & VDC_DRING_LOCAL) == 0) {
2406 		DMSG(vdc, 0, "[%d] local dring\n", vdc->instance);
2407 
2408 		/* Allocate the local copy of this dring */
2409 		vdc->local_dring =
2410 		    kmem_zalloc(vdc->dring_len * sizeof (vdc_local_desc_t),
2411 		    KM_SLEEP);
2412 		vdc->initialized |= VDC_DRING_LOCAL;
2413 	}
2414 
2415 	/*
2416 	 * Mark all DRing entries as free and initialize the private
2417 	 * descriptor's memory handles. If any entry is initialized,
2418 	 * we need to free it later so we set the bit in 'initialized'
2419 	 * at the start.
2420 	 */
2421 	vdc->initialized |= VDC_DRING_ENTRY;
2422 	for (i = 0; i < vdc->dring_len; i++) {
2423 		dep = VDC_GET_DRING_ENTRY_PTR(vdc, i);
2424 		dep->hdr.dstate = VIO_DESC_FREE;
2425 
2426 		status = ldc_mem_alloc_handle(vdc->ldc_handle,
2427 		    &vdc->local_dring[i].desc_mhdl);
2428 		if (status != 0) {
2429 			DMSG(vdc, 0, "![%d] Failed to alloc mem handle for"
2430 			    " descriptor %d", vdc->instance, i);
2431 			return (status);
2432 		}
2433 		vdc->local_dring[i].is_free = B_TRUE;
2434 		vdc->local_dring[i].dep = dep;
2435 	}
2436 
2437 	/* Initialize the starting index */
2438 	vdc->dring_curr_idx = 0;
2439 
2440 	return (status);
2441 }
2442 
2443 /*
2444  * Function:
2445  *	vdc_destroy_descriptor_ring()
2446  *
2447  * Description:
2448  *
2449  * Arguments:
2450  *	vdc	- soft state pointer for this instance of the device driver.
2451  *
2452  * Return Code:
2453  *	None
2454  */
2455 static void
2456 vdc_destroy_descriptor_ring(vdc_t *vdc)
2457 {
2458 	vdc_local_desc_t	*ldep = NULL;	/* Local Dring Entry Pointer */
2459 	ldc_mem_handle_t	mhdl = NULL;
2460 	ldc_mem_info_t		minfo;
2461 	int			status = -1;
2462 	int			i;	/* loop */
2463 
2464 	ASSERT(vdc != NULL);
2465 	ASSERT(mutex_owned(&vdc->lock));
2466 
2467 	DMSG(vdc, 0, "[%d] Entered\n", vdc->instance);
2468 
2469 	if (vdc->initialized & VDC_DRING_ENTRY) {
2470 		DMSG(vdc, 0,
2471 		    "[%d] Removing Local DRing entries\n", vdc->instance);
2472 		for (i = 0; i < vdc->dring_len; i++) {
2473 			ldep = &vdc->local_dring[i];
2474 			mhdl = ldep->desc_mhdl;
2475 
2476 			if (mhdl == NULL)
2477 				continue;
2478 
2479 			if ((status = ldc_mem_info(mhdl, &minfo)) != 0) {
2480 				DMSG(vdc, 0,
2481 				    "ldc_mem_info returned an error: %d\n",
2482 				    status);
2483 
2484 				/*
2485 				 * This must mean that the mem handle
2486 				 * is not valid. Clear it out so that
2487 				 * no one tries to use it.
2488 				 */
2489 				ldep->desc_mhdl = NULL;
2490 				continue;
2491 			}
2492 
2493 			if (minfo.status == LDC_BOUND) {
2494 				(void) ldc_mem_unbind_handle(mhdl);
2495 			}
2496 
2497 			(void) ldc_mem_free_handle(mhdl);
2498 
2499 			ldep->desc_mhdl = NULL;
2500 		}
2501 		vdc->initialized &= ~VDC_DRING_ENTRY;
2502 	}
2503 
2504 	if (vdc->initialized & VDC_DRING_LOCAL) {
2505 		DMSG(vdc, 0, "[%d] Freeing Local DRing\n", vdc->instance);
2506 		kmem_free(vdc->local_dring,
2507 		    vdc->dring_len * sizeof (vdc_local_desc_t));
2508 		vdc->initialized &= ~VDC_DRING_LOCAL;
2509 	}
2510 
2511 	if (vdc->initialized & VDC_DRING_BOUND) {
2512 		DMSG(vdc, 0, "[%d] Unbinding DRing\n", vdc->instance);
2513 		status = ldc_mem_dring_unbind(vdc->ldc_dring_hdl);
2514 		if (status == 0) {
2515 			vdc->initialized &= ~VDC_DRING_BOUND;
2516 		} else {
2517 			DMSG(vdc, 0, "[%d] Error %d unbinding DRing %lx",
2518 			    vdc->instance, status, vdc->ldc_dring_hdl);
2519 		}
2520 		kmem_free(vdc->dring_cookie, sizeof (ldc_mem_cookie_t));
2521 	}
2522 
2523 	if (vdc->initialized & VDC_DRING_INIT) {
2524 		DMSG(vdc, 0, "[%d] Destroying DRing\n", vdc->instance);
2525 		status = ldc_mem_dring_destroy(vdc->ldc_dring_hdl);
2526 		if (status == 0) {
2527 			vdc->ldc_dring_hdl = NULL;
2528 			bzero(&vdc->dring_mem_info, sizeof (ldc_mem_info_t));
2529 			vdc->initialized &= ~VDC_DRING_INIT;
2530 		} else {
2531 			DMSG(vdc, 0, "[%d] Error %d destroying DRing (%lx)",
2532 			    vdc->instance, status, vdc->ldc_dring_hdl);
2533 		}
2534 	}
2535 }
2536 
2537 /*
2538  * Function:
2539  *	vdc_map_to_shared_ring()
2540  *
2541  * Description:
2542  *	Copy contents of the local descriptor to the shared
2543  *	memory descriptor.
2544  *
2545  * Arguments:
2546  *	vdcp	- soft state pointer for this instance of the device driver.
2547  *	idx	- descriptor ring index
2548  *
2549  * Return Code:
2550  *	None
2551  */
2552 static int
2553 vdc_map_to_shared_dring(vdc_t *vdcp, int idx)
2554 {
2555 	vdc_local_desc_t	*ldep;
2556 	vd_dring_entry_t	*dep;
2557 	int			rv;
2558 
2559 	ldep = &(vdcp->local_dring[idx]);
2560 
2561 	/* for now leave in the old pop_mem_hdl stuff */
2562 	if (ldep->nbytes > 0) {
2563 		rv = vdc_populate_mem_hdl(vdcp, ldep);
2564 		if (rv) {
2565 			DMSG(vdcp, 0, "[%d] Cannot populate mem handle\n",
2566 			    vdcp->instance);
2567 			return (rv);
2568 		}
2569 	}
2570 
2571 	/*
2572 	 * fill in the data details into the DRing
2573 	 */
2574 	dep = ldep->dep;
2575 	ASSERT(dep != NULL);
2576 
2577 	dep->payload.req_id = VDC_GET_NEXT_REQ_ID(vdcp);
2578 	dep->payload.operation = ldep->operation;
2579 	dep->payload.addr = ldep->offset;
2580 	dep->payload.nbytes = ldep->nbytes;
2581 	dep->payload.status = (uint32_t)-1;	/* vds will set valid value */
2582 	dep->payload.slice = ldep->slice;
2583 	dep->hdr.dstate = VIO_DESC_READY;
2584 	dep->hdr.ack = 1;		/* request an ACK for every message */
2585 
2586 	return (0);
2587 }
2588 
2589 /*
2590  * Function:
2591  *	vdc_send_request
2592  *
2593  * Description:
2594  *	This routine writes the data to be transmitted to vds into the
2595  *	descriptor, notifies vds that the ring has been updated and
2596  *	then waits for the request to be processed.
2597  *
2598  * Arguments:
2599  *	vdcp	  - the soft state pointer
2600  *	operation - operation we want vds to perform (VD_OP_XXX)
2601  *	addr	  - address of data buf to be read/written.
2602  *	nbytes	  - number of bytes to read/write
2603  *	slice	  - the disk slice this request is for
2604  *	offset	  - relative disk offset
2605  *	cb_type   - type of call - STRATEGY or SYNC
2606  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
2607  *			. mode for ioctl(9e)
2608  *			. LP64 diskaddr_t (block I/O)
2609  *	dir	  - direction of operation (READ/WRITE/BOTH)
2610  *
2611  * Return Codes:
2612  *	0
2613  *	ENXIO
2614  */
2615 static int
2616 vdc_send_request(vdc_t *vdcp, int operation, caddr_t addr,
2617     size_t nbytes, int slice, diskaddr_t offset, int cb_type,
2618     void *cb_arg, vio_desc_direction_t dir)
2619 {
2620 	ASSERT(vdcp != NULL);
2621 	ASSERT(slice == VD_SLICE_NONE || slice < V_NUMPAR);
2622 
2623 	mutex_enter(&vdcp->lock);
2624 
2625 	do {
2626 		while (vdcp->state != VDC_STATE_RUNNING) {
2627 
2628 			/* return error if detaching */
2629 			if (vdcp->state == VDC_STATE_DETACH) {
2630 				mutex_exit(&vdcp->lock);
2631 				return (ENXIO);
2632 			}
2633 
2634 			/* fail request if connection timeout is reached */
2635 			if (vdcp->ctimeout_reached) {
2636 				mutex_exit(&vdcp->lock);
2637 				return (EIO);
2638 			}
2639 
2640 			cv_wait(&vdcp->running_cv, &vdcp->lock);
2641 		}
2642 
2643 	} while (vdc_populate_descriptor(vdcp, operation, addr,
2644 	    nbytes, slice, offset, cb_type, cb_arg, dir));
2645 
2646 	mutex_exit(&vdcp->lock);
2647 	return (0);
2648 }
2649 
2650 
2651 /*
2652  * Function:
2653  *	vdc_populate_descriptor
2654  *
2655  * Description:
2656  *	This routine writes the data to be transmitted to vds into the
2657  *	descriptor, notifies vds that the ring has been updated and
2658  *	then waits for the request to be processed.
2659  *
2660  * Arguments:
2661  *	vdcp	  - the soft state pointer
2662  *	operation - operation we want vds to perform (VD_OP_XXX)
2663  *	addr	  - address of data buf to be read/written.
2664  *	nbytes	  - number of bytes to read/write
2665  *	slice	  - the disk slice this request is for
2666  *	offset	  - relative disk offset
2667  *	cb_type   - type of call - STRATEGY or SYNC
2668  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
2669  *			. mode for ioctl(9e)
2670  *			. LP64 diskaddr_t (block I/O)
2671  *	dir	  - direction of operation (READ/WRITE/BOTH)
2672  *
2673  * Return Codes:
2674  *	0
2675  *	EAGAIN
2676  *		EFAULT
2677  *		ENXIO
2678  *		EIO
2679  */
2680 static int
2681 vdc_populate_descriptor(vdc_t *vdcp, int operation, caddr_t addr,
2682     size_t nbytes, int slice, diskaddr_t offset, int cb_type,
2683     void *cb_arg, vio_desc_direction_t dir)
2684 {
2685 	vdc_local_desc_t	*local_dep = NULL; /* Local Dring Pointer */
2686 	int			idx;		/* Index of DRing entry used */
2687 	int			next_idx;
2688 	vio_dring_msg_t		dmsg;
2689 	size_t			msglen;
2690 	int			rv;
2691 
2692 	ASSERT(MUTEX_HELD(&vdcp->lock));
2693 	vdcp->threads_pending++;
2694 loop:
2695 	DMSG(vdcp, 2, ": dring_curr_idx = %d\n", vdcp->dring_curr_idx);
2696 
2697 	/* Get next available D-Ring entry */
2698 	idx = vdcp->dring_curr_idx;
2699 	local_dep = &(vdcp->local_dring[idx]);
2700 
2701 	if (!local_dep->is_free) {
2702 		DMSG(vdcp, 2, "[%d]: dring full - waiting for space\n",
2703 		    vdcp->instance);
2704 		cv_wait(&vdcp->dring_free_cv, &vdcp->lock);
2705 		if (vdcp->state == VDC_STATE_RUNNING ||
2706 		    vdcp->state == VDC_STATE_HANDLE_PENDING) {
2707 			goto loop;
2708 		}
2709 		vdcp->threads_pending--;
2710 		return (ECONNRESET);
2711 	}
2712 
2713 	next_idx = idx + 1;
2714 	if (next_idx >= vdcp->dring_len)
2715 		next_idx = 0;
2716 	vdcp->dring_curr_idx = next_idx;
2717 
2718 	ASSERT(local_dep->is_free);
2719 
2720 	local_dep->operation = operation;
2721 	local_dep->addr = addr;
2722 	local_dep->nbytes = nbytes;
2723 	local_dep->slice = slice;
2724 	local_dep->offset = offset;
2725 	local_dep->cb_type = cb_type;
2726 	local_dep->cb_arg = cb_arg;
2727 	local_dep->dir = dir;
2728 
2729 	local_dep->is_free = B_FALSE;
2730 
2731 	rv = vdc_map_to_shared_dring(vdcp, idx);
2732 	if (rv) {
2733 		DMSG(vdcp, 0, "[%d]: cannot bind memory - waiting ..\n",
2734 		    vdcp->instance);
2735 		/* free the descriptor */
2736 		local_dep->is_free = B_TRUE;
2737 		vdcp->dring_curr_idx = idx;
2738 		cv_wait(&vdcp->membind_cv, &vdcp->lock);
2739 		if (vdcp->state == VDC_STATE_RUNNING ||
2740 		    vdcp->state == VDC_STATE_HANDLE_PENDING) {
2741 			goto loop;
2742 		}
2743 		vdcp->threads_pending--;
2744 		return (ECONNRESET);
2745 	}
2746 
2747 	/*
2748 	 * Send a msg with the DRing details to vds
2749 	 */
2750 	VIO_INIT_DRING_DATA_TAG(dmsg);
2751 	VDC_INIT_DRING_DATA_MSG_IDS(dmsg, vdcp);
2752 	dmsg.dring_ident = vdcp->dring_ident;
2753 	dmsg.start_idx = idx;
2754 	dmsg.end_idx = idx;
2755 	vdcp->seq_num++;
2756 
2757 	DTRACE_IO2(send, vio_dring_msg_t *, &dmsg, vdc_t *, vdcp);
2758 
2759 	DMSG(vdcp, 2, "ident=0x%lx, st=%u, end=%u, seq=%ld\n",
2760 	    vdcp->dring_ident, dmsg.start_idx, dmsg.end_idx, dmsg.seq_num);
2761 
2762 	/*
2763 	 * note we're still holding the lock here to
2764 	 * make sure the message goes out in order !!!...
2765 	 */
2766 	msglen = sizeof (dmsg);
2767 	rv = vdc_send(vdcp, (caddr_t)&dmsg, &msglen);
2768 	switch (rv) {
2769 	case ECONNRESET:
2770 		/*
2771 		 * vdc_send initiates the reset on failure.
2772 		 * Since the transaction has already been put
2773 		 * on the local dring, it will automatically get
2774 		 * retried when the channel is reset. Given that,
2775 		 * it is ok to just return success even though the
2776 		 * send failed.
2777 		 */
2778 		rv = 0;
2779 		break;
2780 
2781 	case 0: /* EOK */
2782 		DMSG(vdcp, 1, "sent via LDC: rv=%d\n", rv);
2783 		break;
2784 
2785 	default:
2786 		goto cleanup_and_exit;
2787 	}
2788 
2789 	vdcp->threads_pending--;
2790 	return (rv);
2791 
2792 cleanup_and_exit:
2793 	DMSG(vdcp, 0, "unexpected error, rv=%d\n", rv);
2794 	return (ENXIO);
2795 }
2796 
2797 /*
2798  * Function:
2799  *	vdc_do_sync_op
2800  *
2801  * Description:
2802  * 	Wrapper around vdc_populate_descriptor that blocks until the
2803  * 	response to the message is available.
2804  *
2805  * Arguments:
2806  *	vdcp	  - the soft state pointer
2807  *	operation - operation we want vds to perform (VD_OP_XXX)
2808  *	addr	  - address of data buf to be read/written.
2809  *	nbytes	  - number of bytes to read/write
2810  *	slice	  - the disk slice this request is for
2811  *	offset	  - relative disk offset
2812  *	cb_type   - type of call - STRATEGY or SYNC
2813  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
2814  *			. mode for ioctl(9e)
2815  *			. LP64 diskaddr_t (block I/O)
2816  *	dir	  - direction of operation (READ/WRITE/BOTH)
2817  *
2818  * Return Codes:
2819  *	0
2820  *	EAGAIN
2821  *		EFAULT
2822  *		ENXIO
2823  *		EIO
2824  */
2825 static int
2826 vdc_do_sync_op(vdc_t *vdcp, int operation, caddr_t addr, size_t nbytes,
2827     int slice, diskaddr_t offset, int cb_type, void *cb_arg,
2828     vio_desc_direction_t dir)
2829 {
2830 	int status;
2831 
2832 	ASSERT(cb_type == CB_SYNC);
2833 
2834 	/*
2835 	 * Grab the lock, if blocked wait until the server
2836 	 * response causes us to wake up again.
2837 	 */
2838 	mutex_enter(&vdcp->lock);
2839 	vdcp->sync_op_cnt++;
2840 	while (vdcp->sync_op_blocked && vdcp->state != VDC_STATE_DETACH)
2841 		cv_wait(&vdcp->sync_blocked_cv, &vdcp->lock);
2842 
2843 	if (vdcp->state == VDC_STATE_DETACH) {
2844 		cv_broadcast(&vdcp->sync_blocked_cv);
2845 		vdcp->sync_op_cnt--;
2846 		mutex_exit(&vdcp->lock);
2847 		return (ENXIO);
2848 	}
2849 
2850 	/* now block anyone other thread entering after us */
2851 	vdcp->sync_op_blocked = B_TRUE;
2852 	vdcp->sync_op_pending = B_TRUE;
2853 	mutex_exit(&vdcp->lock);
2854 
2855 	status = vdc_send_request(vdcp, operation, addr,
2856 	    nbytes, slice, offset, cb_type, cb_arg, dir);
2857 
2858 	mutex_enter(&vdcp->lock);
2859 
2860 	if (status != 0) {
2861 		vdcp->sync_op_pending = B_FALSE;
2862 	} else {
2863 		/*
2864 		 * block until our transaction completes.
2865 		 * Also anyone else waiting also gets to go next.
2866 		 */
2867 		while (vdcp->sync_op_pending && vdcp->state != VDC_STATE_DETACH)
2868 			cv_wait(&vdcp->sync_pending_cv, &vdcp->lock);
2869 
2870 		DMSG(vdcp, 2, ": operation returned %d\n",
2871 		    vdcp->sync_op_status);
2872 		if (vdcp->state == VDC_STATE_DETACH) {
2873 			vdcp->sync_op_pending = B_FALSE;
2874 			status = ENXIO;
2875 		} else {
2876 			status = vdcp->sync_op_status;
2877 		}
2878 	}
2879 
2880 	vdcp->sync_op_status = 0;
2881 	vdcp->sync_op_blocked = B_FALSE;
2882 	vdcp->sync_op_cnt--;
2883 
2884 	/* signal the next waiting thread */
2885 	cv_signal(&vdcp->sync_blocked_cv);
2886 	mutex_exit(&vdcp->lock);
2887 
2888 	return (status);
2889 }
2890 
2891 
2892 /*
2893  * Function:
2894  *	vdc_drain_response()
2895  *
2896  * Description:
2897  * 	When a guest is panicking, the completion of requests needs to be
2898  * 	handled differently because interrupts are disabled and vdc
2899  * 	will not get messages. We have to poll for the messages instead.
2900  *
2901  * Arguments:
2902  *	vdc	- soft state pointer for this instance of the device driver.
2903  *
2904  * Return Code:
2905  *	0	- Success
2906  */
2907 static int
2908 vdc_drain_response(vdc_t *vdc)
2909 {
2910 	int 			rv, idx, retries;
2911 	size_t			msglen;
2912 	vdc_local_desc_t 	*ldep = NULL;	/* Local Dring Entry Pointer */
2913 	vio_dring_msg_t		dmsg;
2914 
2915 	mutex_enter(&vdc->lock);
2916 
2917 	retries = 0;
2918 	for (;;) {
2919 		msglen = sizeof (dmsg);
2920 		rv = ldc_read(vdc->ldc_handle, (caddr_t)&dmsg, &msglen);
2921 		if (rv) {
2922 			rv = EINVAL;
2923 			break;
2924 		}
2925 
2926 		/*
2927 		 * if there are no packets wait and check again
2928 		 */
2929 		if ((rv == 0) && (msglen == 0)) {
2930 			if (retries++ > vdc_dump_retries) {
2931 				rv = EAGAIN;
2932 				break;
2933 			}
2934 
2935 			drv_usecwait(vdc_usec_timeout_dump);
2936 			continue;
2937 		}
2938 
2939 		/*
2940 		 * Ignore all messages that are not ACKs/NACKs to
2941 		 * DRing requests.
2942 		 */
2943 		if ((dmsg.tag.vio_msgtype != VIO_TYPE_DATA) ||
2944 		    (dmsg.tag.vio_subtype_env != VIO_DRING_DATA)) {
2945 			DMSG(vdc, 0, "discard pkt: type=%d sub=%d env=%d\n",
2946 			    dmsg.tag.vio_msgtype,
2947 			    dmsg.tag.vio_subtype,
2948 			    dmsg.tag.vio_subtype_env);
2949 			continue;
2950 		}
2951 
2952 		/*
2953 		 * set the appropriate return value for the current request.
2954 		 */
2955 		switch (dmsg.tag.vio_subtype) {
2956 		case VIO_SUBTYPE_ACK:
2957 			rv = 0;
2958 			break;
2959 		case VIO_SUBTYPE_NACK:
2960 			rv = EAGAIN;
2961 			break;
2962 		default:
2963 			continue;
2964 		}
2965 
2966 		idx = dmsg.start_idx;
2967 		if (idx >= vdc->dring_len) {
2968 			DMSG(vdc, 0, "[%d] Bogus ack data : start %d\n",
2969 			    vdc->instance, idx);
2970 			continue;
2971 		}
2972 		ldep = &vdc->local_dring[idx];
2973 		if (ldep->dep->hdr.dstate != VIO_DESC_DONE) {
2974 			DMSG(vdc, 0, "[%d] Entry @ %d - state !DONE %d\n",
2975 			    vdc->instance, idx, ldep->dep->hdr.dstate);
2976 			continue;
2977 		}
2978 
2979 		DMSG(vdc, 1, "[%d] Depopulating idx=%d state=%d\n",
2980 		    vdc->instance, idx, ldep->dep->hdr.dstate);
2981 		rv = vdc_depopulate_descriptor(vdc, idx);
2982 		if (rv) {
2983 			DMSG(vdc, 0,
2984 			    "[%d] Entry @ %d - depopulate failed ..\n",
2985 			    vdc->instance, idx);
2986 		}
2987 
2988 		/* if this is the last descriptor - break out of loop */
2989 		if ((idx + 1) % vdc->dring_len == vdc->dring_curr_idx)
2990 			break;
2991 	}
2992 
2993 	mutex_exit(&vdc->lock);
2994 	DMSG(vdc, 0, "End idx=%d\n", idx);
2995 
2996 	return (rv);
2997 }
2998 
2999 
3000 /*
3001  * Function:
3002  *	vdc_depopulate_descriptor()
3003  *
3004  * Description:
3005  *
3006  * Arguments:
3007  *	vdc	- soft state pointer for this instance of the device driver.
3008  *	idx	- Index of the Descriptor Ring entry being modified
3009  *
3010  * Return Code:
3011  *	0	- Success
3012  */
3013 static int
3014 vdc_depopulate_descriptor(vdc_t *vdc, uint_t idx)
3015 {
3016 	vd_dring_entry_t *dep = NULL;		/* Dring Entry Pointer */
3017 	vdc_local_desc_t *ldep = NULL;		/* Local Dring Entry Pointer */
3018 	int		status = ENXIO;
3019 	int		rv = 0;
3020 
3021 	ASSERT(vdc != NULL);
3022 	ASSERT(idx < vdc->dring_len);
3023 	ldep = &vdc->local_dring[idx];
3024 	ASSERT(ldep != NULL);
3025 	ASSERT(MUTEX_HELD(&vdc->lock));
3026 
3027 	DMSG(vdc, 2, ": idx = %d\n", idx);
3028 	dep = ldep->dep;
3029 	ASSERT(dep != NULL);
3030 	ASSERT((dep->hdr.dstate == VIO_DESC_DONE) ||
3031 	    (dep->payload.status == ECANCELED));
3032 
3033 	VDC_MARK_DRING_ENTRY_FREE(vdc, idx);
3034 
3035 	ldep->is_free = B_TRUE;
3036 	status = dep->payload.status;
3037 	DMSG(vdc, 2, ": is_free = %d : status = %d\n", ldep->is_free, status);
3038 
3039 	/*
3040 	 * If no buffers were used to transfer information to the server when
3041 	 * populating the descriptor then no memory handles need to be unbound
3042 	 * and we can return now.
3043 	 */
3044 	if (ldep->nbytes == 0) {
3045 		cv_signal(&vdc->dring_free_cv);
3046 		return (status);
3047 	}
3048 
3049 	/*
3050 	 * If the upper layer passed in a misaligned address we copied the
3051 	 * data into an aligned buffer before sending it to LDC - we now
3052 	 * copy it back to the original buffer.
3053 	 */
3054 	if (ldep->align_addr) {
3055 		ASSERT(ldep->addr != NULL);
3056 
3057 		if (dep->payload.nbytes > 0)
3058 			bcopy(ldep->align_addr, ldep->addr,
3059 			    dep->payload.nbytes);
3060 		kmem_free(ldep->align_addr,
3061 		    sizeof (caddr_t) * P2ROUNDUP(ldep->nbytes, 8));
3062 		ldep->align_addr = NULL;
3063 	}
3064 
3065 	rv = ldc_mem_unbind_handle(ldep->desc_mhdl);
3066 	if (rv != 0) {
3067 		DMSG(vdc, 0, "?[%d] unbind mhdl 0x%lx @ idx %d failed (%d)",
3068 		    vdc->instance, ldep->desc_mhdl, idx, rv);
3069 		/*
3070 		 * The error returned by the vDisk server is more informative
3071 		 * and thus has a higher priority but if it isn't set we ensure
3072 		 * that this function returns an error.
3073 		 */
3074 		if (status == 0)
3075 			status = EINVAL;
3076 	}
3077 
3078 	cv_signal(&vdc->membind_cv);
3079 	cv_signal(&vdc->dring_free_cv);
3080 
3081 	return (status);
3082 }
3083 
3084 /*
3085  * Function:
3086  *	vdc_populate_mem_hdl()
3087  *
3088  * Description:
3089  *
3090  * Arguments:
3091  *	vdc	- soft state pointer for this instance of the device driver.
3092  *	idx	- Index of the Descriptor Ring entry being modified
3093  *	addr	- virtual address being mapped in
3094  *	nybtes	- number of bytes in 'addr'
3095  *	operation - the vDisk operation being performed (VD_OP_xxx)
3096  *
3097  * Return Code:
3098  *	0	- Success
3099  */
3100 static int
3101 vdc_populate_mem_hdl(vdc_t *vdcp, vdc_local_desc_t *ldep)
3102 {
3103 	vd_dring_entry_t	*dep = NULL;
3104 	ldc_mem_handle_t	mhdl;
3105 	caddr_t			vaddr;
3106 	size_t			nbytes;
3107 	uint8_t			perm = LDC_MEM_RW;
3108 	uint8_t			maptype;
3109 	int			rv = 0;
3110 	int			i;
3111 
3112 	ASSERT(vdcp != NULL);
3113 
3114 	dep = ldep->dep;
3115 	mhdl = ldep->desc_mhdl;
3116 
3117 	switch (ldep->dir) {
3118 	case VIO_read_dir:
3119 		perm = LDC_MEM_W;
3120 		break;
3121 
3122 	case VIO_write_dir:
3123 		perm = LDC_MEM_R;
3124 		break;
3125 
3126 	case VIO_both_dir:
3127 		perm = LDC_MEM_RW;
3128 		break;
3129 
3130 	default:
3131 		ASSERT(0);	/* catch bad programming in vdc */
3132 	}
3133 
3134 	/*
3135 	 * LDC expects any addresses passed in to be 8-byte aligned. We need
3136 	 * to copy the contents of any misaligned buffers to a newly allocated
3137 	 * buffer and bind it instead (and copy the the contents back to the
3138 	 * original buffer passed in when depopulating the descriptor)
3139 	 */
3140 	vaddr = ldep->addr;
3141 	nbytes = ldep->nbytes;
3142 	if (((uint64_t)vaddr & 0x7) != 0) {
3143 		ASSERT(ldep->align_addr == NULL);
3144 		ldep->align_addr =
3145 		    kmem_alloc(sizeof (caddr_t) *
3146 		    P2ROUNDUP(nbytes, 8), KM_SLEEP);
3147 		DMSG(vdcp, 0, "[%d] Misaligned address %p reallocating "
3148 		    "(buf=%p nb=%ld op=%d)\n",
3149 		    vdcp->instance, (void *)vaddr, (void *)ldep->align_addr,
3150 		    nbytes, ldep->operation);
3151 		if (perm != LDC_MEM_W)
3152 			bcopy(vaddr, ldep->align_addr, nbytes);
3153 		vaddr = ldep->align_addr;
3154 	}
3155 
3156 	maptype = LDC_IO_MAP|LDC_SHADOW_MAP|LDC_DIRECT_MAP;
3157 	rv = ldc_mem_bind_handle(mhdl, vaddr, P2ROUNDUP(nbytes, 8),
3158 	    maptype, perm, &dep->payload.cookie[0], &dep->payload.ncookies);
3159 	DMSG(vdcp, 2, "[%d] bound mem handle; ncookies=%d\n",
3160 	    vdcp->instance, dep->payload.ncookies);
3161 	if (rv != 0) {
3162 		DMSG(vdcp, 0, "[%d] Failed to bind LDC memory handle "
3163 		    "(mhdl=%p, buf=%p, err=%d)\n",
3164 		    vdcp->instance, (void *)mhdl, (void *)vaddr, rv);
3165 		if (ldep->align_addr) {
3166 			kmem_free(ldep->align_addr,
3167 			    sizeof (caddr_t) * P2ROUNDUP(nbytes, 8));
3168 			ldep->align_addr = NULL;
3169 		}
3170 		return (EAGAIN);
3171 	}
3172 
3173 	/*
3174 	 * Get the other cookies (if any).
3175 	 */
3176 	for (i = 1; i < dep->payload.ncookies; i++) {
3177 		rv = ldc_mem_nextcookie(mhdl, &dep->payload.cookie[i]);
3178 		if (rv != 0) {
3179 			(void) ldc_mem_unbind_handle(mhdl);
3180 			DMSG(vdcp, 0, "?[%d] Failed to get next cookie "
3181 			    "(mhdl=%lx cnum=%d), err=%d",
3182 			    vdcp->instance, mhdl, i, rv);
3183 			if (ldep->align_addr) {
3184 				kmem_free(ldep->align_addr,
3185 				    sizeof (caddr_t) * ldep->nbytes);
3186 				ldep->align_addr = NULL;
3187 			}
3188 			return (EAGAIN);
3189 		}
3190 	}
3191 
3192 	return (rv);
3193 }
3194 
3195 /*
3196  * Interrupt handlers for messages from LDC
3197  */
3198 
3199 /*
3200  * Function:
3201  *	vdc_handle_cb()
3202  *
3203  * Description:
3204  *
3205  * Arguments:
3206  *	event	- Type of event (LDC_EVT_xxx) that triggered the callback
3207  *	arg	- soft state pointer for this instance of the device driver.
3208  *
3209  * Return Code:
3210  *	0	- Success
3211  */
3212 static uint_t
3213 vdc_handle_cb(uint64_t event, caddr_t arg)
3214 {
3215 	ldc_status_t	ldc_state;
3216 	int		rv = 0;
3217 
3218 	vdc_t	*vdc = (vdc_t *)(void *)arg;
3219 
3220 	ASSERT(vdc != NULL);
3221 
3222 	DMSG(vdc, 1, "evt=%lx seqID=%ld\n", event, vdc->seq_num);
3223 
3224 	/*
3225 	 * Depending on the type of event that triggered this callback,
3226 	 * we modify the handshake state or read the data.
3227 	 *
3228 	 * NOTE: not done as a switch() as event could be triggered by
3229 	 * a state change and a read request. Also the ordering	of the
3230 	 * check for the event types is deliberate.
3231 	 */
3232 	if (event & LDC_EVT_UP) {
3233 		DMSG(vdc, 0, "[%d] Received LDC_EVT_UP\n", vdc->instance);
3234 
3235 		mutex_enter(&vdc->lock);
3236 
3237 		/* get LDC state */
3238 		rv = ldc_status(vdc->ldc_handle, &ldc_state);
3239 		if (rv != 0) {
3240 			DMSG(vdc, 0, "[%d] Couldn't get LDC status %d",
3241 			    vdc->instance, rv);
3242 			return (LDC_SUCCESS);
3243 		}
3244 		if (vdc->ldc_state != LDC_UP && ldc_state == LDC_UP) {
3245 			/*
3246 			 * Reset the transaction sequence numbers when
3247 			 * LDC comes up. We then kick off the handshake
3248 			 * negotiation with the vDisk server.
3249 			 */
3250 			vdc->seq_num = 1;
3251 			vdc->seq_num_reply = 0;
3252 			vdc->ldc_state = ldc_state;
3253 			cv_signal(&vdc->initwait_cv);
3254 		}
3255 
3256 		mutex_exit(&vdc->lock);
3257 	}
3258 
3259 	if (event & LDC_EVT_READ) {
3260 		DMSG(vdc, 0, "[%d] Received LDC_EVT_READ\n", vdc->instance);
3261 		mutex_enter(&vdc->read_lock);
3262 		cv_signal(&vdc->read_cv);
3263 		vdc->read_state = VDC_READ_PENDING;
3264 		mutex_exit(&vdc->read_lock);
3265 
3266 		/* that's all we have to do - no need to handle DOWN/RESET */
3267 		return (LDC_SUCCESS);
3268 	}
3269 
3270 	if (event & (LDC_EVT_RESET|LDC_EVT_DOWN)) {
3271 
3272 		DMSG(vdc, 0, "[%d] Received LDC RESET event\n", vdc->instance);
3273 
3274 		mutex_enter(&vdc->lock);
3275 		/*
3276 		 * Need to wake up any readers so they will
3277 		 * detect that a reset has occurred.
3278 		 */
3279 		mutex_enter(&vdc->read_lock);
3280 		if ((vdc->read_state == VDC_READ_WAITING) ||
3281 		    (vdc->read_state == VDC_READ_RESET))
3282 			cv_signal(&vdc->read_cv);
3283 		vdc->read_state = VDC_READ_RESET;
3284 		mutex_exit(&vdc->read_lock);
3285 
3286 		/* wake up any threads waiting for connection to come up */
3287 		if (vdc->state == VDC_STATE_INIT_WAITING) {
3288 			vdc->state = VDC_STATE_RESETTING;
3289 			cv_signal(&vdc->initwait_cv);
3290 		}
3291 
3292 		mutex_exit(&vdc->lock);
3293 	}
3294 
3295 	if (event & ~(LDC_EVT_UP | LDC_EVT_RESET | LDC_EVT_DOWN | LDC_EVT_READ))
3296 		DMSG(vdc, 0, "![%d] Unexpected LDC event (%lx) received",
3297 		    vdc->instance, event);
3298 
3299 	return (LDC_SUCCESS);
3300 }
3301 
3302 /*
3303  * Function:
3304  *	vdc_wait_for_response()
3305  *
3306  * Description:
3307  *	Block waiting for a response from the server. If there is
3308  *	no data the thread block on the read_cv that is signalled
3309  *	by the callback when an EVT_READ occurs.
3310  *
3311  * Arguments:
3312  *	vdcp	- soft state pointer for this instance of the device driver.
3313  *
3314  * Return Code:
3315  *	0	- Success
3316  */
3317 static int
3318 vdc_wait_for_response(vdc_t *vdcp, vio_msg_t *msgp)
3319 {
3320 	size_t		nbytes = sizeof (*msgp);
3321 	int		status;
3322 
3323 	ASSERT(vdcp != NULL);
3324 
3325 	DMSG(vdcp, 1, "[%d] Entered\n", vdcp->instance);
3326 
3327 	status = vdc_recv(vdcp, msgp, &nbytes);
3328 	DMSG(vdcp, 3, "vdc_read() done.. status=0x%x size=0x%x\n",
3329 	    status, (int)nbytes);
3330 	if (status) {
3331 		DMSG(vdcp, 0, "?[%d] Error %d reading LDC msg\n",
3332 		    vdcp->instance, status);
3333 		return (status);
3334 	}
3335 
3336 	if (nbytes < sizeof (vio_msg_tag_t)) {
3337 		DMSG(vdcp, 0, "?[%d] Expect %lu bytes; recv'd %lu\n",
3338 		    vdcp->instance, sizeof (vio_msg_tag_t), nbytes);
3339 		return (ENOMSG);
3340 	}
3341 
3342 	DMSG(vdcp, 2, "[%d] (%x/%x/%x)\n", vdcp->instance,
3343 	    msgp->tag.vio_msgtype,
3344 	    msgp->tag.vio_subtype,
3345 	    msgp->tag.vio_subtype_env);
3346 
3347 	/*
3348 	 * Verify the Session ID of the message
3349 	 *
3350 	 * Every message after the Version has been negotiated should
3351 	 * have the correct session ID set.
3352 	 */
3353 	if ((msgp->tag.vio_sid != vdcp->session_id) &&
3354 	    (msgp->tag.vio_subtype_env != VIO_VER_INFO)) {
3355 		DMSG(vdcp, 0, "[%d] Invalid SID: received 0x%x, "
3356 		    "expected 0x%lx [seq num %lx @ %d]",
3357 		    vdcp->instance, msgp->tag.vio_sid,
3358 		    vdcp->session_id,
3359 		    ((vio_dring_msg_t *)msgp)->seq_num,
3360 		    ((vio_dring_msg_t *)msgp)->start_idx);
3361 		return (ENOMSG);
3362 	}
3363 	return (0);
3364 }
3365 
3366 
3367 /*
3368  * Function:
3369  *	vdc_resubmit_backup_dring()
3370  *
3371  * Description:
3372  *	Resubmit each descriptor in the backed up dring to
3373  * 	vDisk server. The Dring was backed up during connection
3374  *	reset.
3375  *
3376  * Arguments:
3377  *	vdcp	- soft state pointer for this instance of the device driver.
3378  *
3379  * Return Code:
3380  *	0	- Success
3381  */
3382 static int
3383 vdc_resubmit_backup_dring(vdc_t *vdcp)
3384 {
3385 	int		count;
3386 	int		b_idx;
3387 	int		rv;
3388 	int		dring_size;
3389 	int		status;
3390 	vio_msg_t	vio_msg;
3391 	vdc_local_desc_t	*curr_ldep;
3392 
3393 	ASSERT(MUTEX_NOT_HELD(&vdcp->lock));
3394 	ASSERT(vdcp->state == VDC_STATE_HANDLE_PENDING);
3395 
3396 	if (vdcp->local_dring_backup == NULL) {
3397 		/* the pending requests have already been processed */
3398 		return (0);
3399 	}
3400 
3401 	DMSG(vdcp, 1, "restoring pending dring entries (len=%d, tail=%d)\n",
3402 	    vdcp->local_dring_backup_len, vdcp->local_dring_backup_tail);
3403 
3404 	/*
3405 	 * Walk the backup copy of the local descriptor ring and
3406 	 * resubmit all the outstanding transactions.
3407 	 */
3408 	b_idx = vdcp->local_dring_backup_tail;
3409 	for (count = 0; count < vdcp->local_dring_backup_len; count++) {
3410 
3411 		curr_ldep = &(vdcp->local_dring_backup[b_idx]);
3412 
3413 		/* only resubmit outstanding transactions */
3414 		if (!curr_ldep->is_free) {
3415 
3416 			DMSG(vdcp, 1, "resubmitting entry idx=%x\n", b_idx);
3417 			mutex_enter(&vdcp->lock);
3418 			rv = vdc_populate_descriptor(vdcp, curr_ldep->operation,
3419 			    curr_ldep->addr, curr_ldep->nbytes,
3420 			    curr_ldep->slice, curr_ldep->offset,
3421 			    curr_ldep->cb_type, curr_ldep->cb_arg,
3422 			    curr_ldep->dir);
3423 			mutex_exit(&vdcp->lock);
3424 			if (rv) {
3425 				DMSG(vdcp, 1, "[%d] cannot resubmit entry %d\n",
3426 				    vdcp->instance, b_idx);
3427 				return (rv);
3428 			}
3429 
3430 			/* Wait for the response message. */
3431 			DMSG(vdcp, 1, "waiting for response to idx=%x\n",
3432 			    b_idx);
3433 			status = vdc_wait_for_response(vdcp, &vio_msg);
3434 			if (status) {
3435 				DMSG(vdcp, 1, "[%d] wait_for_response "
3436 				    "returned err=%d\n", vdcp->instance,
3437 				    status);
3438 				return (status);
3439 			}
3440 
3441 			DMSG(vdcp, 1, "processing msg for idx=%x\n", b_idx);
3442 			status = vdc_process_data_msg(vdcp, &vio_msg);
3443 			if (status) {
3444 				DMSG(vdcp, 1, "[%d] process_data_msg "
3445 				    "returned err=%d\n", vdcp->instance,
3446 				    status);
3447 				return (status);
3448 			}
3449 		}
3450 
3451 		/* get the next element to submit */
3452 		if (++b_idx >= vdcp->local_dring_backup_len)
3453 			b_idx = 0;
3454 	}
3455 
3456 	/* all done - now clear up pending dring copy */
3457 	dring_size = vdcp->local_dring_backup_len *
3458 	    sizeof (vdcp->local_dring_backup[0]);
3459 
3460 	(void) kmem_free(vdcp->local_dring_backup, dring_size);
3461 
3462 	vdcp->local_dring_backup = NULL;
3463 
3464 	return (0);
3465 }
3466 
3467 /*
3468  * Function:
3469  *	vdc_cancel_backup_dring
3470  *
3471  * Description:
3472  *	Cancel each descriptor in the backed up dring to vDisk server.
3473  *	The Dring was backed up during connection reset.
3474  *
3475  * Arguments:
3476  *	vdcp	- soft state pointer for this instance of the device driver.
3477  *
3478  * Return Code:
3479  *	None
3480  */
3481 void
3482 vdc_cancel_backup_ring(vdc_t *vdcp)
3483 {
3484 	vdc_local_desc_t *ldep;
3485 	struct buf 	*bufp;
3486 	int		count;
3487 	int		b_idx;
3488 	int		dring_size;
3489 
3490 	ASSERT(MUTEX_HELD(&vdcp->lock));
3491 	ASSERT(vdcp->state == VDC_STATE_INIT ||
3492 	    vdcp->state == VDC_STATE_INIT_WAITING ||
3493 	    vdcp->state == VDC_STATE_NEGOTIATE ||
3494 	    vdcp->state == VDC_STATE_RESETTING);
3495 
3496 	if (vdcp->local_dring_backup == NULL) {
3497 		/* the pending requests have already been processed */
3498 		return;
3499 	}
3500 
3501 	DMSG(vdcp, 1, "cancelling pending dring entries (len=%d, tail=%d)\n",
3502 	    vdcp->local_dring_backup_len, vdcp->local_dring_backup_tail);
3503 
3504 	/*
3505 	 * Walk the backup copy of the local descriptor ring and
3506 	 * cancel all the outstanding transactions.
3507 	 */
3508 	b_idx = vdcp->local_dring_backup_tail;
3509 	for (count = 0; count < vdcp->local_dring_backup_len; count++) {
3510 
3511 		ldep = &(vdcp->local_dring_backup[b_idx]);
3512 
3513 		/* only cancel outstanding transactions */
3514 		if (!ldep->is_free) {
3515 
3516 			DMSG(vdcp, 1, "cancelling entry idx=%x\n", b_idx);
3517 
3518 			/*
3519 			 * All requests have already been cleared from the
3520 			 * local descriptor ring and the LDC channel has been
3521 			 * reset so we will never get any reply for these
3522 			 * requests. Now we just have to notify threads waiting
3523 			 * for replies that the request has failed.
3524 			 */
3525 			switch (ldep->cb_type) {
3526 			case CB_SYNC:
3527 				ASSERT(vdcp->sync_op_pending);
3528 				vdcp->sync_op_status = EIO;
3529 				vdcp->sync_op_pending = B_FALSE;
3530 				cv_signal(&vdcp->sync_pending_cv);
3531 				break;
3532 
3533 			case CB_STRATEGY:
3534 				bufp = ldep->cb_arg;
3535 				ASSERT(bufp != NULL);
3536 				bufp->b_resid = bufp->b_bcount;
3537 				bioerror(bufp, EIO);
3538 				biodone(bufp);
3539 				break;
3540 
3541 			default:
3542 				ASSERT(0);
3543 			}
3544 
3545 		}
3546 
3547 		/* get the next element to cancel */
3548 		if (++b_idx >= vdcp->local_dring_backup_len)
3549 			b_idx = 0;
3550 	}
3551 
3552 	/* all done - now clear up pending dring copy */
3553 	dring_size = vdcp->local_dring_backup_len *
3554 	    sizeof (vdcp->local_dring_backup[0]);
3555 
3556 	(void) kmem_free(vdcp->local_dring_backup, dring_size);
3557 
3558 	vdcp->local_dring_backup = NULL;
3559 
3560 	DTRACE_IO2(processed, int, count, vdc_t *, vdcp);
3561 }
3562 
3563 /*
3564  * Function:
3565  *	vdc_connection_timeout
3566  *
3567  * Description:
3568  *	This function is invoked if the timeout set to establish the connection
3569  *	with vds expires. This will happen if we spend too much time in the
3570  *	VDC_STATE_INIT_WAITING or VDC_STATE_NEGOTIATE states. Then we will
3571  *	cancel any pending request and mark them as failed.
3572  *
3573  *	If the timeout does not expire, it will be cancelled when we reach the
3574  *	VDC_STATE_HANDLE_PENDING or VDC_STATE_RESETTING state. This function can
3575  *	be invoked while we are in the VDC_STATE_HANDLE_PENDING or
3576  *	VDC_STATE_RESETTING state in which case we do nothing because the
3577  *	timeout is being cancelled.
3578  *
3579  * Arguments:
3580  *	arg	- argument of the timeout function actually a soft state
3581  *		  pointer for the instance of the device driver.
3582  *
3583  * Return Code:
3584  *	None
3585  */
3586 void
3587 vdc_connection_timeout(void *arg)
3588 {
3589 	vdc_t 		*vdcp = (vdc_t *)arg;
3590 
3591 	mutex_enter(&vdcp->lock);
3592 
3593 	if (vdcp->state == VDC_STATE_HANDLE_PENDING ||
3594 	    vdcp->state == VDC_STATE_DETACH) {
3595 		/*
3596 		 * The connection has just been re-established or
3597 		 * we are detaching.
3598 		 */
3599 		vdcp->ctimeout_reached = B_FALSE;
3600 		mutex_exit(&vdcp->lock);
3601 		return;
3602 	}
3603 
3604 	vdcp->ctimeout_reached = B_TRUE;
3605 
3606 	/* notify requests waiting for sending */
3607 	cv_broadcast(&vdcp->running_cv);
3608 
3609 	/* cancel requests waiting for a result */
3610 	vdc_cancel_backup_ring(vdcp);
3611 
3612 	mutex_exit(&vdcp->lock);
3613 
3614 	cmn_err(CE_NOTE, "[%d] connection to service domain timeout",
3615 	    vdcp->instance);
3616 }
3617 
3618 /*
3619  * Function:
3620  *	vdc_backup_local_dring()
3621  *
3622  * Description:
3623  *	Backup the current dring in the event of a reset. The Dring
3624  *	transactions will be resubmitted to the server when the
3625  *	connection is restored.
3626  *
3627  * Arguments:
3628  *	vdcp	- soft state pointer for this instance of the device driver.
3629  *
3630  * Return Code:
3631  *	NONE
3632  */
3633 static void
3634 vdc_backup_local_dring(vdc_t *vdcp)
3635 {
3636 	int dring_size;
3637 
3638 	ASSERT(MUTEX_HELD(&vdcp->lock));
3639 	ASSERT(vdcp->state == VDC_STATE_RESETTING);
3640 
3641 	/*
3642 	 * If the backup dring is stil around, it means
3643 	 * that the last restore did not complete. However,
3644 	 * since we never got back into the running state,
3645 	 * the backup copy we have is still valid.
3646 	 */
3647 	if (vdcp->local_dring_backup != NULL) {
3648 		DMSG(vdcp, 1, "reusing local descriptor ring backup "
3649 		    "(len=%d, tail=%d)\n", vdcp->local_dring_backup_len,
3650 		    vdcp->local_dring_backup_tail);
3651 		return;
3652 	}
3653 
3654 	/*
3655 	 * The backup dring can be NULL and the local dring may not be
3656 	 * initialized. This can happen if we had a reset while establishing
3657 	 * a new connection but after the connection has timed out. In that
3658 	 * case the backup dring is NULL because the requests have been
3659 	 * cancelled and the request occured before the local dring is
3660 	 * initialized.
3661 	 */
3662 	if (!(vdcp->initialized & VDC_DRING_LOCAL))
3663 		return;
3664 
3665 	DMSG(vdcp, 1, "backing up the local descriptor ring (len=%d, "
3666 	    "tail=%d)\n", vdcp->dring_len, vdcp->dring_curr_idx);
3667 
3668 	dring_size = vdcp->dring_len * sizeof (vdcp->local_dring[0]);
3669 
3670 	vdcp->local_dring_backup = kmem_alloc(dring_size, KM_SLEEP);
3671 	bcopy(vdcp->local_dring, vdcp->local_dring_backup, dring_size);
3672 
3673 	vdcp->local_dring_backup_tail = vdcp->dring_curr_idx;
3674 	vdcp->local_dring_backup_len = vdcp->dring_len;
3675 }
3676 
3677 /* -------------------------------------------------------------------------- */
3678 
3679 /*
3680  * The following functions process the incoming messages from vds
3681  */
3682 
3683 /*
3684  * Function:
3685  *      vdc_process_msg_thread()
3686  *
3687  * Description:
3688  *
3689  *	Main VDC message processing thread. Each vDisk instance
3690  * 	consists of a copy of this thread. This thread triggers
3691  * 	all the handshakes and data exchange with the server. It
3692  * 	also handles all channel resets
3693  *
3694  * Arguments:
3695  *      vdc     - soft state pointer for this instance of the device driver.
3696  *
3697  * Return Code:
3698  *      None
3699  */
3700 static void
3701 vdc_process_msg_thread(vdc_t *vdcp)
3702 {
3703 	int	status;
3704 	int	ctimeout;
3705 	timeout_id_t tmid = 0;
3706 
3707 	mutex_enter(&vdcp->lock);
3708 
3709 	for (;;) {
3710 
3711 #define	Q(_s)	(vdcp->state == _s) ? #_s :
3712 		DMSG(vdcp, 3, "state = %d (%s)\n", vdcp->state,
3713 		    Q(VDC_STATE_INIT)
3714 		    Q(VDC_STATE_INIT_WAITING)
3715 		    Q(VDC_STATE_NEGOTIATE)
3716 		    Q(VDC_STATE_HANDLE_PENDING)
3717 		    Q(VDC_STATE_RUNNING)
3718 		    Q(VDC_STATE_RESETTING)
3719 		    Q(VDC_STATE_DETACH)
3720 		    "UNKNOWN");
3721 
3722 		switch (vdcp->state) {
3723 		case VDC_STATE_INIT:
3724 
3725 			/*
3726 			 * If requested, start a timeout to check if the
3727 			 * connection with vds is established in the
3728 			 * specified delay. If the timeout expires, we
3729 			 * will cancel any pending request.
3730 			 *
3731 			 * If some reset have occurred while establishing
3732 			 * the connection, we already have a timeout armed
3733 			 * and in that case we don't need to arm a new one.
3734 			 */
3735 			ctimeout = (vdc_timeout != 0)?
3736 			    vdc_timeout : vdcp->ctimeout;
3737 
3738 			if (ctimeout != 0 && tmid == 0) {
3739 				tmid = timeout(vdc_connection_timeout, vdcp,
3740 				    ctimeout * drv_usectohz(1000000));
3741 			}
3742 
3743 			/* Check if have re-initializing repeatedly */
3744 			if (vdcp->hshake_cnt++ > vdc_hshake_retries &&
3745 			    vdcp->lifecycle != VDC_LC_ONLINE) {
3746 				cmn_err(CE_NOTE, "[%d] disk access failed.\n",
3747 				    vdcp->instance);
3748 				vdcp->state = VDC_STATE_DETACH;
3749 				break;
3750 			}
3751 
3752 			/* Bring up connection with vds via LDC */
3753 			status = vdc_start_ldc_connection(vdcp);
3754 			if (status == EINVAL) {
3755 				DMSG(vdcp, 0, "[%d] Could not start LDC",
3756 				    vdcp->instance);
3757 				vdcp->state = VDC_STATE_DETACH;
3758 			} else {
3759 				vdcp->state = VDC_STATE_INIT_WAITING;
3760 			}
3761 			break;
3762 
3763 		case VDC_STATE_INIT_WAITING:
3764 
3765 			/*
3766 			 * Let the callback event move us on
3767 			 * when channel is open to server
3768 			 */
3769 			while (vdcp->ldc_state != LDC_UP) {
3770 				cv_wait(&vdcp->initwait_cv, &vdcp->lock);
3771 				if (vdcp->state != VDC_STATE_INIT_WAITING) {
3772 					DMSG(vdcp, 0,
3773 				"state moved to %d out from under us...\n",
3774 					    vdcp->state);
3775 
3776 					break;
3777 				}
3778 			}
3779 			if (vdcp->state == VDC_STATE_INIT_WAITING &&
3780 			    vdcp->ldc_state == LDC_UP) {
3781 				vdcp->state = VDC_STATE_NEGOTIATE;
3782 			}
3783 			break;
3784 
3785 		case VDC_STATE_NEGOTIATE:
3786 			switch (status = vdc_ver_negotiation(vdcp)) {
3787 			case 0:
3788 				break;
3789 			default:
3790 				DMSG(vdcp, 0, "ver negotiate failed (%d)..\n",
3791 				    status);
3792 				goto reset;
3793 			}
3794 
3795 			switch (status = vdc_attr_negotiation(vdcp)) {
3796 			case 0:
3797 				break;
3798 			default:
3799 				DMSG(vdcp, 0, "attr negotiate failed (%d)..\n",
3800 				    status);
3801 				goto reset;
3802 			}
3803 
3804 			switch (status = vdc_dring_negotiation(vdcp)) {
3805 			case 0:
3806 				break;
3807 			default:
3808 				DMSG(vdcp, 0, "dring negotiate failed (%d)..\n",
3809 				    status);
3810 				goto reset;
3811 			}
3812 
3813 			switch (status = vdc_rdx_exchange(vdcp)) {
3814 			case 0:
3815 				vdcp->state = VDC_STATE_HANDLE_PENDING;
3816 				goto done;
3817 			default:
3818 				DMSG(vdcp, 0, "RDX xchg failed ..(%d)\n",
3819 				    status);
3820 				goto reset;
3821 			}
3822 reset:
3823 			DMSG(vdcp, 0, "negotiation failed: resetting (%d)\n",
3824 			    status);
3825 			vdcp->state = VDC_STATE_RESETTING;
3826 			vdcp->self_reset = B_TRUE;
3827 done:
3828 			DMSG(vdcp, 0, "negotiation complete (state=0x%x)...\n",
3829 			    vdcp->state);
3830 			break;
3831 
3832 		case VDC_STATE_HANDLE_PENDING:
3833 
3834 			if (vdcp->ctimeout_reached) {
3835 				/*
3836 				 * The connection timeout had been reached so
3837 				 * pending requests have been cancelled. Now
3838 				 * that the connection is back we can reset
3839 				 * the timeout.
3840 				 */
3841 				ASSERT(vdcp->local_dring_backup == NULL);
3842 				ASSERT(tmid != 0);
3843 				tmid = 0;
3844 				vdcp->ctimeout_reached = B_FALSE;
3845 				vdcp->state = VDC_STATE_RUNNING;
3846 				DMSG(vdcp, 0, "[%d] connection to service "
3847 				    "domain is up", vdcp->instance);
3848 				break;
3849 			}
3850 
3851 			mutex_exit(&vdcp->lock);
3852 			if (tmid != 0) {
3853 				(void) untimeout(tmid);
3854 				tmid = 0;
3855 			}
3856 			status = vdc_resubmit_backup_dring(vdcp);
3857 			mutex_enter(&vdcp->lock);
3858 
3859 			if (status)
3860 				vdcp->state = VDC_STATE_RESETTING;
3861 			else
3862 				vdcp->state = VDC_STATE_RUNNING;
3863 
3864 			break;
3865 
3866 		/* enter running state */
3867 		case VDC_STATE_RUNNING:
3868 			/*
3869 			 * Signal anyone waiting for the connection
3870 			 * to come on line.
3871 			 */
3872 			vdcp->hshake_cnt = 0;
3873 			cv_broadcast(&vdcp->running_cv);
3874 			mutex_exit(&vdcp->lock);
3875 
3876 			for (;;) {
3877 				vio_msg_t msg;
3878 				status = vdc_wait_for_response(vdcp, &msg);
3879 				if (status) break;
3880 
3881 				DMSG(vdcp, 1, "[%d] new pkt(s) available\n",
3882 				    vdcp->instance);
3883 				status = vdc_process_data_msg(vdcp, &msg);
3884 				if (status) {
3885 					DMSG(vdcp, 1, "[%d] process_data_msg "
3886 					    "returned err=%d\n", vdcp->instance,
3887 					    status);
3888 					break;
3889 				}
3890 
3891 			}
3892 
3893 			mutex_enter(&vdcp->lock);
3894 
3895 			vdcp->state = VDC_STATE_RESETTING;
3896 			vdcp->self_reset = B_TRUE;
3897 			break;
3898 
3899 		case VDC_STATE_RESETTING:
3900 			/*
3901 			 * When we reach this state, we either come from the
3902 			 * VDC_STATE_RUNNING state and we can have pending
3903 			 * request but no timeout is armed; or we come from
3904 			 * the VDC_STATE_INIT_WAITING, VDC_NEGOTIATE or
3905 			 * VDC_HANDLE_PENDING state and there is no pending
3906 			 * request or pending requests have already been copied
3907 			 * into the backup dring. So we can safely keep the
3908 			 * connection timeout armed while we are in this state.
3909 			 */
3910 
3911 			DMSG(vdcp, 0, "Initiating channel reset "
3912 			    "(pending = %d)\n", (int)vdcp->threads_pending);
3913 
3914 			if (vdcp->self_reset) {
3915 				DMSG(vdcp, 0,
3916 				    "[%d] calling stop_ldc_connection.\n",
3917 				    vdcp->instance);
3918 				status = vdc_stop_ldc_connection(vdcp);
3919 				vdcp->self_reset = B_FALSE;
3920 			}
3921 
3922 			/*
3923 			 * Wait for all threads currently waiting
3924 			 * for a free dring entry to use.
3925 			 */
3926 			while (vdcp->threads_pending) {
3927 				cv_broadcast(&vdcp->membind_cv);
3928 				cv_broadcast(&vdcp->dring_free_cv);
3929 				mutex_exit(&vdcp->lock);
3930 				/* give the waiters enough time to wake up */
3931 				delay(vdc_hz_min_ldc_delay);
3932 				mutex_enter(&vdcp->lock);
3933 			}
3934 
3935 			ASSERT(vdcp->threads_pending == 0);
3936 
3937 			/* Sanity check that no thread is receiving */
3938 			ASSERT(vdcp->read_state != VDC_READ_WAITING);
3939 
3940 			vdcp->read_state = VDC_READ_IDLE;
3941 
3942 			vdc_backup_local_dring(vdcp);
3943 
3944 			/* cleanup the old d-ring */
3945 			vdc_destroy_descriptor_ring(vdcp);
3946 
3947 			/* go and start again */
3948 			vdcp->state = VDC_STATE_INIT;
3949 
3950 			break;
3951 
3952 		case VDC_STATE_DETACH:
3953 			DMSG(vdcp, 0, "[%d] Reset thread exit cleanup ..\n",
3954 			    vdcp->instance);
3955 
3956 			/* cancel any pending timeout */
3957 			mutex_exit(&vdcp->lock);
3958 			if (tmid != 0) {
3959 				(void) untimeout(tmid);
3960 				tmid = 0;
3961 			}
3962 			mutex_enter(&vdcp->lock);
3963 
3964 			/*
3965 			 * Signal anyone waiting for connection
3966 			 * to come online
3967 			 */
3968 			cv_broadcast(&vdcp->running_cv);
3969 
3970 			while (vdcp->sync_op_pending) {
3971 				cv_signal(&vdcp->sync_pending_cv);
3972 				cv_signal(&vdcp->sync_blocked_cv);
3973 				mutex_exit(&vdcp->lock);
3974 				/* give the waiters enough time to wake up */
3975 				delay(vdc_hz_min_ldc_delay);
3976 				mutex_enter(&vdcp->lock);
3977 			}
3978 
3979 			mutex_exit(&vdcp->lock);
3980 
3981 			DMSG(vdcp, 0, "[%d] Msg processing thread exiting ..\n",
3982 			    vdcp->instance);
3983 			thread_exit();
3984 			break;
3985 		}
3986 	}
3987 }
3988 
3989 
3990 /*
3991  * Function:
3992  *	vdc_process_data_msg()
3993  *
3994  * Description:
3995  *	This function is called by the message processing thread each time
3996  *	a message with a msgtype of VIO_TYPE_DATA is received. It will either
3997  *	be an ACK or NACK from vds[1] which vdc handles as follows.
3998  *		ACK	- wake up the waiting thread
3999  *		NACK	- resend any messages necessary
4000  *
4001  *	[1] Although the message format allows it, vds should not send a
4002  *	    VIO_SUBTYPE_INFO message to vdc asking it to read data; if for
4003  *	    some bizarre reason it does, vdc will reset the connection.
4004  *
4005  * Arguments:
4006  *	vdc	- soft state pointer for this instance of the device driver.
4007  *	msg	- the LDC message sent by vds
4008  *
4009  * Return Code:
4010  *	0	- Success.
4011  *	> 0	- error value returned by LDC
4012  */
4013 static int
4014 vdc_process_data_msg(vdc_t *vdcp, vio_msg_t *msg)
4015 {
4016 	int			status = 0;
4017 	vio_dring_msg_t		*dring_msg;
4018 	vdc_local_desc_t	*ldep = NULL;
4019 	int			start, end;
4020 	int			idx;
4021 
4022 	dring_msg = (vio_dring_msg_t *)msg;
4023 
4024 	ASSERT(msg->tag.vio_msgtype == VIO_TYPE_DATA);
4025 	ASSERT(vdcp != NULL);
4026 
4027 	mutex_enter(&vdcp->lock);
4028 
4029 	/*
4030 	 * Check to see if the message has bogus data
4031 	 */
4032 	idx = start = dring_msg->start_idx;
4033 	end = dring_msg->end_idx;
4034 	if ((start >= vdcp->dring_len) ||
4035 	    (end >= vdcp->dring_len) || (end < -1)) {
4036 		DMSG(vdcp, 0, "[%d] Bogus ACK data : start %d, end %d\n",
4037 		    vdcp->instance, start, end);
4038 		mutex_exit(&vdcp->lock);
4039 		return (EINVAL);
4040 	}
4041 
4042 	/*
4043 	 * Verify that the sequence number is what vdc expects.
4044 	 */
4045 	switch (vdc_verify_seq_num(vdcp, dring_msg)) {
4046 	case VDC_SEQ_NUM_TODO:
4047 		break;	/* keep processing this message */
4048 	case VDC_SEQ_NUM_SKIP:
4049 		mutex_exit(&vdcp->lock);
4050 		return (0);
4051 	case VDC_SEQ_NUM_INVALID:
4052 		mutex_exit(&vdcp->lock);
4053 		DMSG(vdcp, 0, "[%d] invalid seqno\n", vdcp->instance);
4054 		return (ENXIO);
4055 	}
4056 
4057 	if (msg->tag.vio_subtype == VIO_SUBTYPE_NACK) {
4058 		DMSG(vdcp, 0, "[%d] DATA NACK\n", vdcp->instance);
4059 		VDC_DUMP_DRING_MSG(dring_msg);
4060 		mutex_exit(&vdcp->lock);
4061 		return (EIO);
4062 
4063 	} else if (msg->tag.vio_subtype == VIO_SUBTYPE_INFO) {
4064 		mutex_exit(&vdcp->lock);
4065 		return (EPROTO);
4066 	}
4067 
4068 	DTRACE_IO2(recv, vio_dring_msg_t, dring_msg, vdc_t *, vdcp);
4069 	DMSG(vdcp, 1, ": start %d end %d\n", start, end);
4070 	ASSERT(start == end);
4071 
4072 	ldep = &vdcp->local_dring[idx];
4073 
4074 	DMSG(vdcp, 1, ": state 0x%x - cb_type 0x%x\n",
4075 	    ldep->dep->hdr.dstate, ldep->cb_type);
4076 
4077 	if (ldep->dep->hdr.dstate == VIO_DESC_DONE) {
4078 		struct buf *bufp;
4079 
4080 		switch (ldep->cb_type) {
4081 		case CB_SYNC:
4082 			ASSERT(vdcp->sync_op_pending);
4083 
4084 			status = vdc_depopulate_descriptor(vdcp, idx);
4085 			vdcp->sync_op_status = status;
4086 			vdcp->sync_op_pending = B_FALSE;
4087 			cv_signal(&vdcp->sync_pending_cv);
4088 			break;
4089 
4090 		case CB_STRATEGY:
4091 			bufp = ldep->cb_arg;
4092 			ASSERT(bufp != NULL);
4093 			bufp->b_resid =
4094 			    bufp->b_bcount - ldep->dep->payload.nbytes;
4095 			status = ldep->dep->payload.status; /* Future:ntoh */
4096 			if (status != 0) {
4097 				DMSG(vdcp, 1, "strategy status=%d\n", status);
4098 				bioerror(bufp, status);
4099 			}
4100 			status = vdc_depopulate_descriptor(vdcp, idx);
4101 			biodone(bufp);
4102 
4103 			DMSG(vdcp, 1,
4104 			    "strategy complete req=%ld bytes resp=%ld bytes\n",
4105 			    bufp->b_bcount, ldep->dep->payload.nbytes);
4106 			break;
4107 
4108 		default:
4109 			ASSERT(0);
4110 		}
4111 	}
4112 
4113 	/* let the arrival signal propogate */
4114 	mutex_exit(&vdcp->lock);
4115 
4116 	/* probe gives the count of how many entries were processed */
4117 	DTRACE_IO2(processed, int, 1, vdc_t *, vdcp);
4118 
4119 	return (0);
4120 }
4121 
4122 /*
4123  * Function:
4124  *	vdc_process_err_msg()
4125  *
4126  * NOTE: No error messages are used as part of the vDisk protocol
4127  */
4128 static int
4129 vdc_process_err_msg(vdc_t *vdc, vio_msg_t msg)
4130 {
4131 	_NOTE(ARGUNUSED(vdc))
4132 	_NOTE(ARGUNUSED(msg))
4133 
4134 	ASSERT(msg.tag.vio_msgtype == VIO_TYPE_ERR);
4135 	DMSG(vdc, 1, "[%d] Got an ERR msg", vdc->instance);
4136 
4137 	return (ENOTSUP);
4138 }
4139 
4140 /*
4141  * Function:
4142  *	vdc_handle_ver_msg()
4143  *
4144  * Description:
4145  *
4146  * Arguments:
4147  *	vdc	- soft state pointer for this instance of the device driver.
4148  *	ver_msg	- LDC message sent by vDisk server
4149  *
4150  * Return Code:
4151  *	0	- Success
4152  */
4153 static int
4154 vdc_handle_ver_msg(vdc_t *vdc, vio_ver_msg_t *ver_msg)
4155 {
4156 	int status = 0;
4157 
4158 	ASSERT(vdc != NULL);
4159 	ASSERT(mutex_owned(&vdc->lock));
4160 
4161 	if (ver_msg->tag.vio_subtype_env != VIO_VER_INFO) {
4162 		return (EPROTO);
4163 	}
4164 
4165 	if (ver_msg->dev_class != VDEV_DISK_SERVER) {
4166 		return (EINVAL);
4167 	}
4168 
4169 	switch (ver_msg->tag.vio_subtype) {
4170 	case VIO_SUBTYPE_ACK:
4171 		/*
4172 		 * We check to see if the version returned is indeed supported
4173 		 * (The server may have also adjusted the minor number downwards
4174 		 * and if so 'ver_msg' will contain the actual version agreed)
4175 		 */
4176 		if (vdc_is_supported_version(ver_msg)) {
4177 			vdc->ver.major = ver_msg->ver_major;
4178 			vdc->ver.minor = ver_msg->ver_minor;
4179 			ASSERT(vdc->ver.major > 0);
4180 		} else {
4181 			status = EPROTO;
4182 		}
4183 		break;
4184 
4185 	case VIO_SUBTYPE_NACK:
4186 		/*
4187 		 * call vdc_is_supported_version() which will return the next
4188 		 * supported version (if any) in 'ver_msg'
4189 		 */
4190 		(void) vdc_is_supported_version(ver_msg);
4191 		if (ver_msg->ver_major > 0) {
4192 			size_t len = sizeof (*ver_msg);
4193 
4194 			ASSERT(vdc->ver.major > 0);
4195 
4196 			/* reset the necessary fields and resend */
4197 			ver_msg->tag.vio_subtype = VIO_SUBTYPE_INFO;
4198 			ver_msg->dev_class = VDEV_DISK;
4199 
4200 			status = vdc_send(vdc, (caddr_t)ver_msg, &len);
4201 			DMSG(vdc, 0, "[%d] Resend VER info (LDC status = %d)\n",
4202 			    vdc->instance, status);
4203 			if (len != sizeof (*ver_msg))
4204 				status = EBADMSG;
4205 		} else {
4206 			DMSG(vdc, 0, "[%d] No common version with vDisk server",
4207 			    vdc->instance);
4208 			status = ENOTSUP;
4209 		}
4210 
4211 		break;
4212 	case VIO_SUBTYPE_INFO:
4213 		/*
4214 		 * Handle the case where vds starts handshake
4215 		 * (for now only vdc is the instigator)
4216 		 */
4217 		status = ENOTSUP;
4218 		break;
4219 
4220 	default:
4221 		status = EINVAL;
4222 		break;
4223 	}
4224 
4225 	return (status);
4226 }
4227 
4228 /*
4229  * Function:
4230  *	vdc_handle_attr_msg()
4231  *
4232  * Description:
4233  *
4234  * Arguments:
4235  *	vdc	- soft state pointer for this instance of the device driver.
4236  *	attr_msg	- LDC message sent by vDisk server
4237  *
4238  * Return Code:
4239  *	0	- Success
4240  */
4241 static int
4242 vdc_handle_attr_msg(vdc_t *vdc, vd_attr_msg_t *attr_msg)
4243 {
4244 	int status = 0;
4245 
4246 	ASSERT(vdc != NULL);
4247 	ASSERT(mutex_owned(&vdc->lock));
4248 
4249 	if (attr_msg->tag.vio_subtype_env != VIO_ATTR_INFO) {
4250 		return (EPROTO);
4251 	}
4252 
4253 	switch (attr_msg->tag.vio_subtype) {
4254 	case VIO_SUBTYPE_ACK:
4255 		/*
4256 		 * We now verify the attributes sent by vds.
4257 		 */
4258 		if (attr_msg->vdisk_size == 0) {
4259 			DMSG(vdc, 0, "[%d] Invalid disk size from vds",
4260 			    vdc->instance);
4261 			status = EINVAL;
4262 			break;
4263 		}
4264 
4265 		if (attr_msg->max_xfer_sz == 0) {
4266 			DMSG(vdc, 0, "[%d] Invalid transfer size from vds",
4267 			    vdc->instance);
4268 			status = EINVAL;
4269 			break;
4270 		}
4271 
4272 		/*
4273 		 * If the disk size is already set check that it hasn't changed.
4274 		 */
4275 		if ((vdc->vdisk_size != 0) &&
4276 		    (vdc->vdisk_size != attr_msg->vdisk_size)) {
4277 			DMSG(vdc, 0, "[%d] Different disk size from vds "
4278 			    "(old=0x%lx - new=0x%lx", vdc->instance,
4279 			    vdc->vdisk_size, attr_msg->vdisk_size)
4280 			status = EINVAL;
4281 			break;
4282 		}
4283 
4284 		vdc->vdisk_size = attr_msg->vdisk_size;
4285 		vdc->vdisk_type = attr_msg->vdisk_type;
4286 
4287 		DMSG(vdc, 0, "[%d] max_xfer_sz: sent %lx acked %lx\n",
4288 		    vdc->instance, vdc->max_xfer_sz, attr_msg->max_xfer_sz);
4289 		DMSG(vdc, 0, "[%d] vdisk_block_size: sent %lx acked %x\n",
4290 		    vdc->instance, vdc->block_size,
4291 		    attr_msg->vdisk_block_size);
4292 
4293 		/*
4294 		 * We don't know at compile time what the vDisk server will
4295 		 * think are good values but we apply an large (arbitrary)
4296 		 * upper bound to prevent memory exhaustion in vdc if it was
4297 		 * allocating a DRing based of huge values sent by the server.
4298 		 * We probably will never exceed this except if the message
4299 		 * was garbage.
4300 		 */
4301 		if ((attr_msg->max_xfer_sz * attr_msg->vdisk_block_size) <=
4302 		    (PAGESIZE * DEV_BSIZE)) {
4303 			vdc->max_xfer_sz = attr_msg->max_xfer_sz;
4304 			vdc->block_size = attr_msg->vdisk_block_size;
4305 		} else {
4306 			DMSG(vdc, 0, "[%d] vds block transfer size too big;"
4307 			    " using max supported by vdc", vdc->instance);
4308 		}
4309 
4310 		if ((attr_msg->xfer_mode != VIO_DRING_MODE) ||
4311 		    (attr_msg->vdisk_size > INT64_MAX) ||
4312 		    (attr_msg->vdisk_type > VD_DISK_TYPE_DISK)) {
4313 			DMSG(vdc, 0, "[%d] Invalid attributes from vds",
4314 			    vdc->instance);
4315 			status = EINVAL;
4316 			break;
4317 		}
4318 
4319 		/*
4320 		 * Now that we have received all attributes we can create a
4321 		 * fake geometry for the disk.
4322 		 */
4323 		vdc_create_fake_geometry(vdc);
4324 		break;
4325 
4326 	case VIO_SUBTYPE_NACK:
4327 		/*
4328 		 * vds could not handle the attributes we sent so we
4329 		 * stop negotiating.
4330 		 */
4331 		status = EPROTO;
4332 		break;
4333 
4334 	case VIO_SUBTYPE_INFO:
4335 		/*
4336 		 * Handle the case where vds starts the handshake
4337 		 * (for now; vdc is the only supported instigatior)
4338 		 */
4339 		status = ENOTSUP;
4340 		break;
4341 
4342 	default:
4343 		status = ENOTSUP;
4344 		break;
4345 	}
4346 
4347 	return (status);
4348 }
4349 
4350 /*
4351  * Function:
4352  *	vdc_handle_dring_reg_msg()
4353  *
4354  * Description:
4355  *
4356  * Arguments:
4357  *	vdc		- soft state pointer for this instance of the driver.
4358  *	dring_msg	- LDC message sent by vDisk server
4359  *
4360  * Return Code:
4361  *	0	- Success
4362  */
4363 static int
4364 vdc_handle_dring_reg_msg(vdc_t *vdc, vio_dring_reg_msg_t *dring_msg)
4365 {
4366 	int		status = 0;
4367 
4368 	ASSERT(vdc != NULL);
4369 	ASSERT(mutex_owned(&vdc->lock));
4370 
4371 	if (dring_msg->tag.vio_subtype_env != VIO_DRING_REG) {
4372 		return (EPROTO);
4373 	}
4374 
4375 	switch (dring_msg->tag.vio_subtype) {
4376 	case VIO_SUBTYPE_ACK:
4377 		/* save the received dring_ident */
4378 		vdc->dring_ident = dring_msg->dring_ident;
4379 		DMSG(vdc, 0, "[%d] Received dring ident=0x%lx\n",
4380 		    vdc->instance, vdc->dring_ident);
4381 		break;
4382 
4383 	case VIO_SUBTYPE_NACK:
4384 		/*
4385 		 * vds could not handle the DRing info we sent so we
4386 		 * stop negotiating.
4387 		 */
4388 		DMSG(vdc, 0, "[%d] server could not register DRing\n",
4389 		    vdc->instance);
4390 		status = EPROTO;
4391 		break;
4392 
4393 	case VIO_SUBTYPE_INFO:
4394 		/*
4395 		 * Handle the case where vds starts handshake
4396 		 * (for now only vdc is the instigatior)
4397 		 */
4398 		status = ENOTSUP;
4399 		break;
4400 	default:
4401 		status = ENOTSUP;
4402 	}
4403 
4404 	return (status);
4405 }
4406 
4407 /*
4408  * Function:
4409  *	vdc_verify_seq_num()
4410  *
4411  * Description:
4412  *	This functions verifies that the sequence number sent back by the vDisk
4413  *	server with the latest message is what is expected (i.e. it is greater
4414  *	than the last seq num sent by the vDisk server and less than or equal
4415  *	to the last seq num generated by vdc).
4416  *
4417  *	It then checks the request ID to see if any requests need processing
4418  *	in the DRing.
4419  *
4420  * Arguments:
4421  *	vdc		- soft state pointer for this instance of the driver.
4422  *	dring_msg	- pointer to the LDC message sent by vds
4423  *
4424  * Return Code:
4425  *	VDC_SEQ_NUM_TODO	- Message needs to be processed
4426  *	VDC_SEQ_NUM_SKIP	- Message has already been processed
4427  *	VDC_SEQ_NUM_INVALID	- The seq numbers are so out of sync,
4428  *				  vdc cannot deal with them
4429  */
4430 static int
4431 vdc_verify_seq_num(vdc_t *vdc, vio_dring_msg_t *dring_msg)
4432 {
4433 	ASSERT(vdc != NULL);
4434 	ASSERT(dring_msg != NULL);
4435 	ASSERT(mutex_owned(&vdc->lock));
4436 
4437 	/*
4438 	 * Check to see if the messages were responded to in the correct
4439 	 * order by vds.
4440 	 */
4441 	if ((dring_msg->seq_num <= vdc->seq_num_reply) ||
4442 	    (dring_msg->seq_num > vdc->seq_num)) {
4443 		DMSG(vdc, 0, "?[%d] Bogus sequence_number %lu: "
4444 		    "%lu > expected <= %lu (last proc req %lu sent %lu)\n",
4445 		    vdc->instance, dring_msg->seq_num,
4446 		    vdc->seq_num_reply, vdc->seq_num,
4447 		    vdc->req_id_proc, vdc->req_id);
4448 		return (VDC_SEQ_NUM_INVALID);
4449 	}
4450 	vdc->seq_num_reply = dring_msg->seq_num;
4451 
4452 	if (vdc->req_id_proc < vdc->req_id)
4453 		return (VDC_SEQ_NUM_TODO);
4454 	else
4455 		return (VDC_SEQ_NUM_SKIP);
4456 }
4457 
4458 
4459 /*
4460  * Function:
4461  *	vdc_is_supported_version()
4462  *
4463  * Description:
4464  *	This routine checks if the major/minor version numbers specified in
4465  *	'ver_msg' are supported. If not it finds the next version that is
4466  *	in the supported version list 'vdc_version[]' and sets the fields in
4467  *	'ver_msg' to those values
4468  *
4469  * Arguments:
4470  *	ver_msg	- LDC message sent by vDisk server
4471  *
4472  * Return Code:
4473  *	B_TRUE	- Success
4474  *	B_FALSE	- Version not supported
4475  */
4476 static boolean_t
4477 vdc_is_supported_version(vio_ver_msg_t *ver_msg)
4478 {
4479 	int vdc_num_versions = sizeof (vdc_version) / sizeof (vdc_version[0]);
4480 
4481 	for (int i = 0; i < vdc_num_versions; i++) {
4482 		ASSERT(vdc_version[i].major > 0);
4483 		ASSERT((i == 0) ||
4484 		    (vdc_version[i].major < vdc_version[i-1].major));
4485 
4486 		/*
4487 		 * If the major versions match, adjust the minor version, if
4488 		 * necessary, down to the highest value supported by this
4489 		 * client. The server should support all minor versions lower
4490 		 * than the value it sent
4491 		 */
4492 		if (ver_msg->ver_major == vdc_version[i].major) {
4493 			if (ver_msg->ver_minor > vdc_version[i].minor) {
4494 				DMSGX(0,
4495 				    "Adjusting minor version from %u to %u",
4496 				    ver_msg->ver_minor, vdc_version[i].minor);
4497 				ver_msg->ver_minor = vdc_version[i].minor;
4498 			}
4499 			return (B_TRUE);
4500 		}
4501 
4502 		/*
4503 		 * If the message contains a higher major version number, set
4504 		 * the message's major/minor versions to the current values
4505 		 * and return false, so this message will get resent with
4506 		 * these values, and the server will potentially try again
4507 		 * with the same or a lower version
4508 		 */
4509 		if (ver_msg->ver_major > vdc_version[i].major) {
4510 			ver_msg->ver_major = vdc_version[i].major;
4511 			ver_msg->ver_minor = vdc_version[i].minor;
4512 			DMSGX(0, "Suggesting major/minor (0x%x/0x%x)\n",
4513 			    ver_msg->ver_major, ver_msg->ver_minor);
4514 
4515 			return (B_FALSE);
4516 		}
4517 
4518 		/*
4519 		 * Otherwise, the message's major version is less than the
4520 		 * current major version, so continue the loop to the next
4521 		 * (lower) supported version
4522 		 */
4523 	}
4524 
4525 	/*
4526 	 * No common version was found; "ground" the version pair in the
4527 	 * message to terminate negotiation
4528 	 */
4529 	ver_msg->ver_major = 0;
4530 	ver_msg->ver_minor = 0;
4531 
4532 	return (B_FALSE);
4533 }
4534 /* -------------------------------------------------------------------------- */
4535 
4536 /*
4537  * DKIO(7) support
4538  */
4539 
4540 typedef struct vdc_dk_arg {
4541 	struct dk_callback	dkc;
4542 	int			mode;
4543 	dev_t			dev;
4544 	vdc_t			*vdc;
4545 } vdc_dk_arg_t;
4546 
4547 /*
4548  * Function:
4549  * 	vdc_dkio_flush_cb()
4550  *
4551  * Description:
4552  *	This routine is a callback for DKIOCFLUSHWRITECACHE which can be called
4553  *	by kernel code.
4554  *
4555  * Arguments:
4556  *	arg	- a pointer to a vdc_dk_arg_t structure.
4557  */
4558 void
4559 vdc_dkio_flush_cb(void *arg)
4560 {
4561 	struct vdc_dk_arg	*dk_arg = (struct vdc_dk_arg *)arg;
4562 	struct dk_callback	*dkc = NULL;
4563 	vdc_t			*vdc = NULL;
4564 	int			rv;
4565 
4566 	if (dk_arg == NULL) {
4567 		cmn_err(CE_NOTE, "?[Unk] DKIOCFLUSHWRITECACHE arg is NULL\n");
4568 		return;
4569 	}
4570 	dkc = &dk_arg->dkc;
4571 	vdc = dk_arg->vdc;
4572 	ASSERT(vdc != NULL);
4573 
4574 	rv = vdc_do_sync_op(vdc, VD_OP_FLUSH, NULL, 0,
4575 	    VDCPART(dk_arg->dev), 0, CB_SYNC, 0, VIO_both_dir);
4576 	if (rv != 0) {
4577 		DMSG(vdc, 0, "[%d] DKIOCFLUSHWRITECACHE failed %d : model %x\n",
4578 		    vdc->instance, rv,
4579 		    ddi_model_convert_from(dk_arg->mode & FMODELS));
4580 	}
4581 
4582 	/*
4583 	 * Trigger the call back to notify the caller the the ioctl call has
4584 	 * been completed.
4585 	 */
4586 	if ((dk_arg->mode & FKIOCTL) &&
4587 	    (dkc != NULL) &&
4588 	    (dkc->dkc_callback != NULL)) {
4589 		ASSERT(dkc->dkc_cookie != NULL);
4590 		(*dkc->dkc_callback)(dkc->dkc_cookie, rv);
4591 	}
4592 
4593 	/* Indicate that one less DKIO write flush is outstanding */
4594 	mutex_enter(&vdc->lock);
4595 	vdc->dkio_flush_pending--;
4596 	ASSERT(vdc->dkio_flush_pending >= 0);
4597 	mutex_exit(&vdc->lock);
4598 
4599 	/* free the mem that was allocated when the callback was dispatched */
4600 	kmem_free(arg, sizeof (vdc_dk_arg_t));
4601 }
4602 
4603 /*
4604  * Function:
4605  * 	vdc_dkio_get_partition()
4606  *
4607  * Description:
4608  *	This function implements the DKIOCGAPART ioctl.
4609  *
4610  * Arguments:
4611  *	vdc	- soft state pointer
4612  *	arg	- a pointer to a dk_map[NDKMAP] or dk_map32[NDKMAP] structure
4613  *	flag	- ioctl flags
4614  */
4615 static int
4616 vdc_dkio_get_partition(vdc_t *vdc, caddr_t arg, int flag)
4617 {
4618 	struct dk_geom *geom;
4619 	struct vtoc *vtoc;
4620 	union {
4621 		struct dk_map map[NDKMAP];
4622 		struct dk_map32 map32[NDKMAP];
4623 	} data;
4624 	int i, rv, size;
4625 
4626 	mutex_enter(&vdc->lock);
4627 
4628 	if ((rv = vdc_validate_geometry(vdc)) != 0) {
4629 		mutex_exit(&vdc->lock);
4630 		return (rv);
4631 	}
4632 
4633 	vtoc = vdc->vtoc;
4634 	geom = vdc->geom;
4635 
4636 	if (ddi_model_convert_from(flag & FMODELS) == DDI_MODEL_ILP32) {
4637 
4638 		for (i = 0; i < vtoc->v_nparts; i++) {
4639 			data.map32[i].dkl_cylno = vtoc->v_part[i].p_start /
4640 			    (geom->dkg_nhead * geom->dkg_nsect);
4641 			data.map32[i].dkl_nblk = vtoc->v_part[i].p_size;
4642 		}
4643 		size = NDKMAP * sizeof (struct dk_map32);
4644 
4645 	} else {
4646 
4647 		for (i = 0; i < vtoc->v_nparts; i++) {
4648 			data.map[i].dkl_cylno = vtoc->v_part[i].p_start /
4649 			    (geom->dkg_nhead * geom->dkg_nsect);
4650 			data.map[i].dkl_nblk = vtoc->v_part[i].p_size;
4651 		}
4652 		size = NDKMAP * sizeof (struct dk_map);
4653 
4654 	}
4655 
4656 	mutex_exit(&vdc->lock);
4657 
4658 	if (ddi_copyout(&data, arg, size, flag) != 0)
4659 		return (EFAULT);
4660 
4661 	return (0);
4662 }
4663 
4664 /*
4665  * Function:
4666  * 	vdc_dioctl_rwcmd()
4667  *
4668  * Description:
4669  *	This function implements the DIOCTL_RWCMD ioctl. This ioctl is used
4670  *	for DKC_DIRECT disks to read or write at an absolute disk offset.
4671  *
4672  * Arguments:
4673  *	dev	- device
4674  *	arg	- a pointer to a dadkio_rwcmd or dadkio_rwcmd32 structure
4675  *	flag	- ioctl flags
4676  */
4677 static int
4678 vdc_dioctl_rwcmd(dev_t dev, caddr_t arg, int flag)
4679 {
4680 	struct dadkio_rwcmd32 rwcmd32;
4681 	struct dadkio_rwcmd rwcmd;
4682 	struct iovec aiov;
4683 	struct uio auio;
4684 	int rw, status;
4685 	struct buf *buf;
4686 
4687 	if (ddi_model_convert_from(flag & FMODELS) == DDI_MODEL_ILP32) {
4688 		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd32,
4689 		    sizeof (struct dadkio_rwcmd32), flag)) {
4690 			return (EFAULT);
4691 		}
4692 		rwcmd.cmd = rwcmd32.cmd;
4693 		rwcmd.flags = rwcmd32.flags;
4694 		rwcmd.blkaddr = (daddr_t)rwcmd32.blkaddr;
4695 		rwcmd.buflen = rwcmd32.buflen;
4696 		rwcmd.bufaddr = (caddr_t)(uintptr_t)rwcmd32.bufaddr;
4697 	} else {
4698 		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd,
4699 		    sizeof (struct dadkio_rwcmd), flag)) {
4700 			return (EFAULT);
4701 		}
4702 	}
4703 
4704 	switch (rwcmd.cmd) {
4705 	case DADKIO_RWCMD_READ:
4706 		rw = B_READ;
4707 		break;
4708 	case DADKIO_RWCMD_WRITE:
4709 		rw = B_WRITE;
4710 		break;
4711 	default:
4712 		return (EINVAL);
4713 	}
4714 
4715 	bzero((caddr_t)&aiov, sizeof (struct iovec));
4716 	aiov.iov_base   = rwcmd.bufaddr;
4717 	aiov.iov_len    = rwcmd.buflen;
4718 
4719 	bzero((caddr_t)&auio, sizeof (struct uio));
4720 	auio.uio_iov    = &aiov;
4721 	auio.uio_iovcnt = 1;
4722 	auio.uio_loffset = rwcmd.blkaddr * DEV_BSIZE;
4723 	auio.uio_resid  = rwcmd.buflen;
4724 	auio.uio_segflg = flag & FKIOCTL ? UIO_SYSSPACE : UIO_USERSPACE;
4725 
4726 	buf = kmem_alloc(sizeof (buf_t), KM_SLEEP);
4727 	bioinit(buf);
4728 	/*
4729 	 * We use the private field of buf to specify that this is an
4730 	 * I/O using an absolute offset.
4731 	 */
4732 	buf->b_private = (void *)VD_SLICE_NONE;
4733 
4734 	status = physio(vdc_strategy, buf, dev, rw, vdc_min, &auio);
4735 
4736 	biofini(buf);
4737 	kmem_free(buf, sizeof (buf_t));
4738 
4739 	return (status);
4740 }
4741 
4742 /*
4743  * This structure is used in the DKIO(7I) array below.
4744  */
4745 typedef struct vdc_dk_ioctl {
4746 	uint8_t		op;		/* VD_OP_XXX value */
4747 	int		cmd;		/* Solaris ioctl operation number */
4748 	size_t		nbytes;		/* size of structure to be copied */
4749 
4750 	/* function to convert between vDisk and Solaris structure formats */
4751 	int	(*convert)(vdc_t *vdc, void *vd_buf, void *ioctl_arg,
4752 	    int mode, int dir);
4753 } vdc_dk_ioctl_t;
4754 
4755 /*
4756  * Subset of DKIO(7I) operations currently supported
4757  */
4758 static vdc_dk_ioctl_t	dk_ioctl[] = {
4759 	{VD_OP_FLUSH,		DKIOCFLUSHWRITECACHE,	0,
4760 		vdc_null_copy_func},
4761 	{VD_OP_GET_WCE,		DKIOCGETWCE,		sizeof (int),
4762 		vdc_get_wce_convert},
4763 	{VD_OP_SET_WCE,		DKIOCSETWCE,		sizeof (int),
4764 		vdc_set_wce_convert},
4765 	{VD_OP_GET_VTOC,	DKIOCGVTOC,		sizeof (vd_vtoc_t),
4766 		vdc_get_vtoc_convert},
4767 	{VD_OP_SET_VTOC,	DKIOCSVTOC,		sizeof (vd_vtoc_t),
4768 		vdc_set_vtoc_convert},
4769 	{VD_OP_GET_DISKGEOM,	DKIOCGGEOM,		sizeof (vd_geom_t),
4770 		vdc_get_geom_convert},
4771 	{VD_OP_GET_DISKGEOM,	DKIOCG_PHYGEOM,		sizeof (vd_geom_t),
4772 		vdc_get_geom_convert},
4773 	{VD_OP_GET_DISKGEOM, 	DKIOCG_VIRTGEOM,	sizeof (vd_geom_t),
4774 		vdc_get_geom_convert},
4775 	{VD_OP_SET_DISKGEOM,	DKIOCSGEOM,		sizeof (vd_geom_t),
4776 		vdc_set_geom_convert},
4777 	{VD_OP_GET_EFI,		DKIOCGETEFI,		0,
4778 		vdc_get_efi_convert},
4779 	{VD_OP_SET_EFI,		DKIOCSETEFI,		0,
4780 		vdc_set_efi_convert},
4781 
4782 	/* DIOCTL_RWCMD is converted to a read or a write */
4783 	{0, DIOCTL_RWCMD,  sizeof (struct dadkio_rwcmd), NULL},
4784 
4785 	/*
4786 	 * These particular ioctls are not sent to the server - vdc fakes up
4787 	 * the necessary info.
4788 	 */
4789 	{0, DKIOCINFO, sizeof (struct dk_cinfo), vdc_null_copy_func},
4790 	{0, DKIOCGMEDIAINFO, sizeof (struct dk_minfo), vdc_null_copy_func},
4791 	{0, USCSICMD,	sizeof (struct uscsi_cmd), vdc_null_copy_func},
4792 	{0, DKIOCGAPART, 0, vdc_null_copy_func },
4793 	{0, DKIOCREMOVABLE, 0, vdc_null_copy_func},
4794 	{0, CDROMREADOFFSET, 0, vdc_null_copy_func}
4795 };
4796 
4797 /*
4798  * Function:
4799  *	vd_process_ioctl()
4800  *
4801  * Description:
4802  *	This routine processes disk specific ioctl calls
4803  *
4804  * Arguments:
4805  *	dev	- the device number
4806  *	cmd	- the operation [dkio(7I)] to be processed
4807  *	arg	- pointer to user provided structure
4808  *		  (contains data to be set or reference parameter for get)
4809  *	mode	- bit flag, indicating open settings, 32/64 bit type, etc
4810  *
4811  * Return Code:
4812  *	0
4813  *	EFAULT
4814  *	ENXIO
4815  *	EIO
4816  *	ENOTSUP
4817  */
4818 static int
4819 vd_process_ioctl(dev_t dev, int cmd, caddr_t arg, int mode)
4820 {
4821 	int		instance = VDCUNIT(dev);
4822 	vdc_t		*vdc = NULL;
4823 	int		rv = -1;
4824 	int		idx = 0;		/* index into dk_ioctl[] */
4825 	size_t		len = 0;		/* #bytes to send to vds */
4826 	size_t		alloc_len = 0;		/* #bytes to allocate mem for */
4827 	caddr_t		mem_p = NULL;
4828 	size_t		nioctls = (sizeof (dk_ioctl)) / (sizeof (dk_ioctl[0]));
4829 	vdc_dk_ioctl_t	*iop;
4830 
4831 	vdc = ddi_get_soft_state(vdc_state, instance);
4832 	if (vdc == NULL) {
4833 		cmn_err(CE_NOTE, "![%d] Could not get soft state structure",
4834 		    instance);
4835 		return (ENXIO);
4836 	}
4837 
4838 	DMSG(vdc, 0, "[%d] Processing ioctl(%x) for dev %lx : model %x\n",
4839 	    instance, cmd, dev, ddi_model_convert_from(mode & FMODELS));
4840 
4841 	/*
4842 	 * Validate the ioctl operation to be performed.
4843 	 *
4844 	 * If we have looped through the array without finding a match then we
4845 	 * don't support this ioctl.
4846 	 */
4847 	for (idx = 0; idx < nioctls; idx++) {
4848 		if (cmd == dk_ioctl[idx].cmd)
4849 			break;
4850 	}
4851 
4852 	if (idx >= nioctls) {
4853 		DMSG(vdc, 0, "[%d] Unsupported ioctl (0x%x)\n",
4854 		    vdc->instance, cmd);
4855 		return (ENOTSUP);
4856 	}
4857 
4858 	iop = &(dk_ioctl[idx]);
4859 
4860 	if (cmd == DKIOCGETEFI || cmd == DKIOCSETEFI) {
4861 		/* size is not fixed for EFI ioctls, it depends on ioctl arg */
4862 		dk_efi_t	dk_efi;
4863 
4864 		rv = ddi_copyin(arg, &dk_efi, sizeof (dk_efi_t), mode);
4865 		if (rv != 0)
4866 			return (EFAULT);
4867 
4868 		len = sizeof (vd_efi_t) - 1 + dk_efi.dki_length;
4869 	} else {
4870 		len = iop->nbytes;
4871 	}
4872 
4873 	/*
4874 	 * Deal with the ioctls which the server does not provide. vdc can
4875 	 * fake these up and return immediately
4876 	 */
4877 	switch (cmd) {
4878 	case CDROMREADOFFSET:
4879 	case DKIOCREMOVABLE:
4880 	case USCSICMD:
4881 		return (ENOTTY);
4882 
4883 	case DIOCTL_RWCMD:
4884 		{
4885 			if (vdc->cinfo == NULL)
4886 				return (ENXIO);
4887 
4888 			if (vdc->cinfo->dki_ctype != DKC_DIRECT)
4889 				return (ENOTTY);
4890 
4891 			return (vdc_dioctl_rwcmd(dev, arg, mode));
4892 		}
4893 
4894 	case DKIOCGAPART:
4895 		{
4896 			return (vdc_dkio_get_partition(vdc, arg, mode));
4897 		}
4898 
4899 	case DKIOCINFO:
4900 		{
4901 			struct dk_cinfo	cinfo;
4902 			if (vdc->cinfo == NULL)
4903 				return (ENXIO);
4904 
4905 			bcopy(vdc->cinfo, &cinfo, sizeof (struct dk_cinfo));
4906 			cinfo.dki_partition = VDCPART(dev);
4907 
4908 			rv = ddi_copyout(&cinfo, (void *)arg,
4909 			    sizeof (struct dk_cinfo), mode);
4910 			if (rv != 0)
4911 				return (EFAULT);
4912 
4913 			return (0);
4914 		}
4915 
4916 	case DKIOCGMEDIAINFO:
4917 		{
4918 			if (vdc->minfo == NULL)
4919 				return (ENXIO);
4920 
4921 			rv = ddi_copyout(vdc->minfo, (void *)arg,
4922 			    sizeof (struct dk_minfo), mode);
4923 			if (rv != 0)
4924 				return (EFAULT);
4925 
4926 			return (0);
4927 		}
4928 
4929 	case DKIOCFLUSHWRITECACHE:
4930 		{
4931 			struct dk_callback *dkc = (struct dk_callback *)arg;
4932 			vdc_dk_arg_t	*dkarg = NULL;
4933 
4934 			DMSG(vdc, 1, "[%d] Flush W$: mode %x\n",
4935 			    instance, mode);
4936 
4937 			/*
4938 			 * If arg is NULL, then there is no callback function
4939 			 * registered and the call operates synchronously; we
4940 			 * break and continue with the rest of the function and
4941 			 * wait for vds to return (i.e. after the request to
4942 			 * vds returns successfully, all writes completed prior
4943 			 * to the ioctl will have been flushed from the disk
4944 			 * write cache to persistent media.
4945 			 *
4946 			 * If a callback function is registered, we dispatch
4947 			 * the request on a task queue and return immediately.
4948 			 * The callback will deal with informing the calling
4949 			 * thread that the flush request is completed.
4950 			 */
4951 			if (dkc == NULL)
4952 				break;
4953 
4954 			/*
4955 			 * the asynchronous callback is only supported if
4956 			 * invoked from within the kernel
4957 			 */
4958 			if ((mode & FKIOCTL) == 0)
4959 				return (ENOTSUP);
4960 
4961 			dkarg = kmem_zalloc(sizeof (vdc_dk_arg_t), KM_SLEEP);
4962 
4963 			dkarg->mode = mode;
4964 			dkarg->dev = dev;
4965 			bcopy(dkc, &dkarg->dkc, sizeof (*dkc));
4966 
4967 			mutex_enter(&vdc->lock);
4968 			vdc->dkio_flush_pending++;
4969 			dkarg->vdc = vdc;
4970 			mutex_exit(&vdc->lock);
4971 
4972 			/* put the request on a task queue */
4973 			rv = taskq_dispatch(system_taskq, vdc_dkio_flush_cb,
4974 			    (void *)dkarg, DDI_SLEEP);
4975 			if (rv == NULL) {
4976 				/* clean up if dispatch fails */
4977 				mutex_enter(&vdc->lock);
4978 				vdc->dkio_flush_pending--;
4979 				mutex_exit(&vdc->lock);
4980 				kmem_free(dkarg, sizeof (vdc_dk_arg_t));
4981 			}
4982 
4983 			return (rv == NULL ? ENOMEM : 0);
4984 		}
4985 	}
4986 
4987 	/* catch programming error in vdc - should be a VD_OP_XXX ioctl */
4988 	ASSERT(iop->op != 0);
4989 
4990 	/* LDC requires that the memory being mapped is 8-byte aligned */
4991 	alloc_len = P2ROUNDUP(len, sizeof (uint64_t));
4992 	DMSG(vdc, 1, "[%d] struct size %ld alloc %ld\n",
4993 	    instance, len, alloc_len);
4994 
4995 	ASSERT(alloc_len >= 0); /* sanity check */
4996 	if (alloc_len > 0)
4997 		mem_p = kmem_zalloc(alloc_len, KM_SLEEP);
4998 
4999 	/*
5000 	 * Call the conversion function for this ioctl which, if necessary,
5001 	 * converts from the Solaris format to the format ARC'ed
5002 	 * as part of the vDisk protocol (FWARC 2006/195)
5003 	 */
5004 	ASSERT(iop->convert != NULL);
5005 	rv = (iop->convert)(vdc, arg, mem_p, mode, VD_COPYIN);
5006 	if (rv != 0) {
5007 		DMSG(vdc, 0, "[%d] convert func returned %d for ioctl 0x%x\n",
5008 		    instance, rv, cmd);
5009 		if (mem_p != NULL)
5010 			kmem_free(mem_p, alloc_len);
5011 		return (rv);
5012 	}
5013 
5014 	/*
5015 	 * send request to vds to service the ioctl.
5016 	 */
5017 	rv = vdc_do_sync_op(vdc, iop->op, mem_p, alloc_len,
5018 	    VDCPART(dev), 0, CB_SYNC, (void *)(uint64_t)mode,
5019 	    VIO_both_dir);
5020 
5021 	if (cmd == DKIOCSVTOC || cmd == DKIOCSETEFI) {
5022 		/*
5023 		 * The disk label may have changed. Revalidate the disk
5024 		 * geometry. This will also update the device nodes and
5025 		 * properties.
5026 		 */
5027 		vdc_validate(vdc);
5028 	}
5029 
5030 	if (rv != 0) {
5031 		/*
5032 		 * This is not necessarily an error. The ioctl could
5033 		 * be returning a value such as ENOTTY to indicate
5034 		 * that the ioctl is not applicable.
5035 		 */
5036 		DMSG(vdc, 0, "[%d] vds returned %d for ioctl 0x%x\n",
5037 		    instance, rv, cmd);
5038 		if (mem_p != NULL)
5039 			kmem_free(mem_p, alloc_len);
5040 
5041 		return (rv);
5042 	}
5043 
5044 	/*
5045 	 * Call the conversion function (if it exists) for this ioctl
5046 	 * which converts from the format ARC'ed as part of the vDisk
5047 	 * protocol (FWARC 2006/195) back to a format understood by
5048 	 * the rest of Solaris.
5049 	 */
5050 	rv = (iop->convert)(vdc, mem_p, arg, mode, VD_COPYOUT);
5051 	if (rv != 0) {
5052 		DMSG(vdc, 0, "[%d] convert func returned %d for ioctl 0x%x\n",
5053 		    instance, rv, cmd);
5054 		if (mem_p != NULL)
5055 			kmem_free(mem_p, alloc_len);
5056 		return (rv);
5057 	}
5058 
5059 	if (mem_p != NULL)
5060 		kmem_free(mem_p, alloc_len);
5061 
5062 	return (rv);
5063 }
5064 
5065 /*
5066  * Function:
5067  *
5068  * Description:
5069  *	This is an empty conversion function used by ioctl calls which
5070  *	do not need to convert the data being passed in/out to userland
5071  */
5072 static int
5073 vdc_null_copy_func(vdc_t *vdc, void *from, void *to, int mode, int dir)
5074 {
5075 	_NOTE(ARGUNUSED(vdc))
5076 	_NOTE(ARGUNUSED(from))
5077 	_NOTE(ARGUNUSED(to))
5078 	_NOTE(ARGUNUSED(mode))
5079 	_NOTE(ARGUNUSED(dir))
5080 
5081 	return (0);
5082 }
5083 
5084 static int
5085 vdc_get_wce_convert(vdc_t *vdc, void *from, void *to,
5086     int mode, int dir)
5087 {
5088 	_NOTE(ARGUNUSED(vdc))
5089 
5090 	if (dir == VD_COPYIN)
5091 		return (0);		/* nothing to do */
5092 
5093 	if (ddi_copyout(from, to, sizeof (int), mode) != 0)
5094 		return (EFAULT);
5095 
5096 	return (0);
5097 }
5098 
5099 static int
5100 vdc_set_wce_convert(vdc_t *vdc, void *from, void *to,
5101     int mode, int dir)
5102 {
5103 	_NOTE(ARGUNUSED(vdc))
5104 
5105 	if (dir == VD_COPYOUT)
5106 		return (0);		/* nothing to do */
5107 
5108 	if (ddi_copyin(from, to, sizeof (int), mode) != 0)
5109 		return (EFAULT);
5110 
5111 	return (0);
5112 }
5113 
5114 /*
5115  * Function:
5116  *	vdc_get_vtoc_convert()
5117  *
5118  * Description:
5119  *	This routine performs the necessary convertions from the DKIOCGVTOC
5120  *	Solaris structure to the format defined in FWARC 2006/195.
5121  *
5122  *	In the struct vtoc definition, the timestamp field is marked as not
5123  *	supported so it is not part of vDisk protocol (FWARC 2006/195).
5124  *	However SVM uses that field to check it can write into the VTOC,
5125  *	so we fake up the info of that field.
5126  *
5127  * Arguments:
5128  *	vdc	- the vDisk client
5129  *	from	- the buffer containing the data to be copied from
5130  *	to	- the buffer to be copied to
5131  *	mode	- flags passed to ioctl() call
5132  *	dir	- the "direction" of the copy - VD_COPYIN or VD_COPYOUT
5133  *
5134  * Return Code:
5135  *	0	- Success
5136  *	ENXIO	- incorrect buffer passed in.
5137  *	EFAULT	- ddi_copyout routine encountered an error.
5138  */
5139 static int
5140 vdc_get_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
5141 {
5142 	int		i;
5143 	void		*tmp_mem = NULL;
5144 	void		*tmp_memp;
5145 	struct vtoc	vt;
5146 	struct vtoc32	vt32;
5147 	int		copy_len = 0;
5148 	int		rv = 0;
5149 
5150 	if (dir != VD_COPYOUT)
5151 		return (0);	/* nothing to do */
5152 
5153 	if ((from == NULL) || (to == NULL))
5154 		return (ENXIO);
5155 
5156 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32)
5157 		copy_len = sizeof (struct vtoc32);
5158 	else
5159 		copy_len = sizeof (struct vtoc);
5160 
5161 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
5162 
5163 	VD_VTOC2VTOC((vd_vtoc_t *)from, &vt);
5164 
5165 	/* fake the VTOC timestamp field */
5166 	for (i = 0; i < V_NUMPAR; i++) {
5167 		vt.timestamp[i] = vdc->vtoc->timestamp[i];
5168 	}
5169 
5170 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5171 		vtoctovtoc32(vt, vt32);
5172 		tmp_memp = &vt32;
5173 	} else {
5174 		tmp_memp = &vt;
5175 	}
5176 	rv = ddi_copyout(tmp_memp, to, copy_len, mode);
5177 	if (rv != 0)
5178 		rv = EFAULT;
5179 
5180 	kmem_free(tmp_mem, copy_len);
5181 	return (rv);
5182 }
5183 
5184 /*
5185  * Function:
5186  *	vdc_set_vtoc_convert()
5187  *
5188  * Description:
5189  *	This routine performs the necessary convertions from the DKIOCSVTOC
5190  *	Solaris structure to the format defined in FWARC 2006/195.
5191  *
5192  * Arguments:
5193  *	vdc	- the vDisk client
5194  *	from	- Buffer with data
5195  *	to	- Buffer where data is to be copied to
5196  *	mode	- flags passed to ioctl
5197  *	dir	- direction of copy (in or out)
5198  *
5199  * Return Code:
5200  *	0	- Success
5201  *	ENXIO	- Invalid buffer passed in
5202  *	EFAULT	- ddi_copyin of data failed
5203  */
5204 static int
5205 vdc_set_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
5206 {
5207 	_NOTE(ARGUNUSED(vdc))
5208 
5209 	void		*tmp_mem = NULL;
5210 	struct vtoc	vt;
5211 	struct vtoc	*vtp = &vt;
5212 	vd_vtoc_t	vtvd;
5213 	int		copy_len = 0;
5214 	int		rv = 0;
5215 
5216 	if (dir != VD_COPYIN)
5217 		return (0);	/* nothing to do */
5218 
5219 	if ((from == NULL) || (to == NULL))
5220 		return (ENXIO);
5221 
5222 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32)
5223 		copy_len = sizeof (struct vtoc32);
5224 	else
5225 		copy_len = sizeof (struct vtoc);
5226 
5227 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
5228 
5229 	rv = ddi_copyin(from, tmp_mem, copy_len, mode);
5230 	if (rv != 0) {
5231 		kmem_free(tmp_mem, copy_len);
5232 		return (EFAULT);
5233 	}
5234 
5235 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5236 		vtoc32tovtoc((*(struct vtoc32 *)tmp_mem), vt);
5237 	} else {
5238 		vtp = tmp_mem;
5239 	}
5240 
5241 	VTOC2VD_VTOC(vtp, &vtvd);
5242 	bcopy(&vtvd, to, sizeof (vd_vtoc_t));
5243 	kmem_free(tmp_mem, copy_len);
5244 
5245 	return (0);
5246 }
5247 
5248 /*
5249  * Function:
5250  *	vdc_get_geom_convert()
5251  *
5252  * Description:
5253  *	This routine performs the necessary convertions from the DKIOCGGEOM,
5254  *	DKIOCG_PHYSGEOM and DKIOG_VIRTGEOM Solaris structures to the format
5255  *	defined in FWARC 2006/195
5256  *
5257  * Arguments:
5258  *	vdc	- the vDisk client
5259  *	from	- Buffer with data
5260  *	to	- Buffer where data is to be copied to
5261  *	mode	- flags passed to ioctl
5262  *	dir	- direction of copy (in or out)
5263  *
5264  * Return Code:
5265  *	0	- Success
5266  *	ENXIO	- Invalid buffer passed in
5267  *	EFAULT	- ddi_copyout of data failed
5268  */
5269 static int
5270 vdc_get_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
5271 {
5272 	_NOTE(ARGUNUSED(vdc))
5273 
5274 	struct dk_geom	geom;
5275 	int	copy_len = sizeof (struct dk_geom);
5276 	int	rv = 0;
5277 
5278 	if (dir != VD_COPYOUT)
5279 		return (0);	/* nothing to do */
5280 
5281 	if ((from == NULL) || (to == NULL))
5282 		return (ENXIO);
5283 
5284 	VD_GEOM2DK_GEOM((vd_geom_t *)from, &geom);
5285 	rv = ddi_copyout(&geom, to, copy_len, mode);
5286 	if (rv != 0)
5287 		rv = EFAULT;
5288 
5289 	return (rv);
5290 }
5291 
5292 /*
5293  * Function:
5294  *	vdc_set_geom_convert()
5295  *
5296  * Description:
5297  *	This routine performs the necessary convertions from the DKIOCSGEOM
5298  *	Solaris structure to the format defined in FWARC 2006/195.
5299  *
5300  * Arguments:
5301  *	vdc	- the vDisk client
5302  *	from	- Buffer with data
5303  *	to	- Buffer where data is to be copied to
5304  *	mode	- flags passed to ioctl
5305  *	dir	- direction of copy (in or out)
5306  *
5307  * Return Code:
5308  *	0	- Success
5309  *	ENXIO	- Invalid buffer passed in
5310  *	EFAULT	- ddi_copyin of data failed
5311  */
5312 static int
5313 vdc_set_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
5314 {
5315 	_NOTE(ARGUNUSED(vdc))
5316 
5317 	vd_geom_t	vdgeom;
5318 	void		*tmp_mem = NULL;
5319 	int		copy_len = sizeof (struct dk_geom);
5320 	int		rv = 0;
5321 
5322 	if (dir != VD_COPYIN)
5323 		return (0);	/* nothing to do */
5324 
5325 	if ((from == NULL) || (to == NULL))
5326 		return (ENXIO);
5327 
5328 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
5329 
5330 	rv = ddi_copyin(from, tmp_mem, copy_len, mode);
5331 	if (rv != 0) {
5332 		kmem_free(tmp_mem, copy_len);
5333 		return (EFAULT);
5334 	}
5335 	DK_GEOM2VD_GEOM((struct dk_geom *)tmp_mem, &vdgeom);
5336 	bcopy(&vdgeom, to, sizeof (vdgeom));
5337 	kmem_free(tmp_mem, copy_len);
5338 
5339 	return (0);
5340 }
5341 
5342 static int
5343 vdc_get_efi_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
5344 {
5345 	_NOTE(ARGUNUSED(vdc))
5346 
5347 	vd_efi_t	*vd_efi;
5348 	dk_efi_t	dk_efi;
5349 	int		rv = 0;
5350 	void		*uaddr;
5351 
5352 	if ((from == NULL) || (to == NULL))
5353 		return (ENXIO);
5354 
5355 	if (dir == VD_COPYIN) {
5356 
5357 		vd_efi = (vd_efi_t *)to;
5358 
5359 		rv = ddi_copyin(from, &dk_efi, sizeof (dk_efi_t), mode);
5360 		if (rv != 0)
5361 			return (EFAULT);
5362 
5363 		vd_efi->lba = dk_efi.dki_lba;
5364 		vd_efi->length = dk_efi.dki_length;
5365 		bzero(vd_efi->data, vd_efi->length);
5366 
5367 	} else {
5368 
5369 		rv = ddi_copyin(to, &dk_efi, sizeof (dk_efi_t), mode);
5370 		if (rv != 0)
5371 			return (EFAULT);
5372 
5373 		uaddr = dk_efi.dki_data;
5374 
5375 		dk_efi.dki_data = kmem_alloc(dk_efi.dki_length, KM_SLEEP);
5376 
5377 		VD_EFI2DK_EFI((vd_efi_t *)from, &dk_efi);
5378 
5379 		rv = ddi_copyout(dk_efi.dki_data, uaddr, dk_efi.dki_length,
5380 		    mode);
5381 		if (rv != 0)
5382 			return (EFAULT);
5383 
5384 		kmem_free(dk_efi.dki_data, dk_efi.dki_length);
5385 	}
5386 
5387 	return (0);
5388 }
5389 
5390 static int
5391 vdc_set_efi_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
5392 {
5393 	_NOTE(ARGUNUSED(vdc))
5394 
5395 	dk_efi_t	dk_efi;
5396 	void		*uaddr;
5397 
5398 	if (dir == VD_COPYOUT)
5399 		return (0);	/* nothing to do */
5400 
5401 	if ((from == NULL) || (to == NULL))
5402 		return (ENXIO);
5403 
5404 	if (ddi_copyin(from, &dk_efi, sizeof (dk_efi_t), mode) != 0)
5405 		return (EFAULT);
5406 
5407 	uaddr = dk_efi.dki_data;
5408 
5409 	dk_efi.dki_data = kmem_alloc(dk_efi.dki_length, KM_SLEEP);
5410 
5411 	if (ddi_copyin(uaddr, dk_efi.dki_data, dk_efi.dki_length, mode) != 0)
5412 		return (EFAULT);
5413 
5414 	DK_EFI2VD_EFI(&dk_efi, (vd_efi_t *)to);
5415 
5416 	kmem_free(dk_efi.dki_data, dk_efi.dki_length);
5417 
5418 	return (0);
5419 }
5420 
5421 /*
5422  * Function:
5423  *	vdc_create_fake_geometry()
5424  *
5425  * Description:
5426  *	This routine fakes up the disk info needed for some DKIO ioctls.
5427  *		- DKIOCINFO
5428  *		- DKIOCGMEDIAINFO
5429  *
5430  *	[ just like lofi(7D) and ramdisk(7D) ]
5431  *
5432  * Arguments:
5433  *	vdc	- soft state pointer for this instance of the device driver.
5434  *
5435  * Return Code:
5436  *	none.
5437  */
5438 static void
5439 vdc_create_fake_geometry(vdc_t *vdc)
5440 {
5441 	ASSERT(vdc != NULL);
5442 	ASSERT(vdc->vdisk_size != 0);
5443 	ASSERT(vdc->max_xfer_sz != 0);
5444 
5445 	/*
5446 	 * DKIOCINFO support
5447 	 */
5448 	if (vdc->cinfo == NULL)
5449 		vdc->cinfo = kmem_zalloc(sizeof (struct dk_cinfo), KM_SLEEP);
5450 
5451 	(void) strcpy(vdc->cinfo->dki_cname, VDC_DRIVER_NAME);
5452 	(void) strcpy(vdc->cinfo->dki_dname, VDC_DRIVER_NAME);
5453 	/* max_xfer_sz is #blocks so we don't need to divide by DEV_BSIZE */
5454 	vdc->cinfo->dki_maxtransfer = vdc->max_xfer_sz;
5455 	/*
5456 	 * We currently set the controller type to DKC_DIRECT for any disk.
5457 	 * When SCSI support is implemented, we will eventually change this
5458 	 * type to DKC_SCSI_CCS for disks supporting the SCSI protocol.
5459 	 */
5460 	vdc->cinfo->dki_ctype = DKC_DIRECT;
5461 	vdc->cinfo->dki_flags = DKI_FMTVOL;
5462 	vdc->cinfo->dki_cnum = 0;
5463 	vdc->cinfo->dki_addr = 0;
5464 	vdc->cinfo->dki_space = 0;
5465 	vdc->cinfo->dki_prio = 0;
5466 	vdc->cinfo->dki_vec = 0;
5467 	vdc->cinfo->dki_unit = vdc->instance;
5468 	vdc->cinfo->dki_slave = 0;
5469 	/*
5470 	 * The partition number will be created on the fly depending on the
5471 	 * actual slice (i.e. minor node) that is used to request the data.
5472 	 */
5473 	vdc->cinfo->dki_partition = 0;
5474 
5475 	/*
5476 	 * DKIOCGMEDIAINFO support
5477 	 */
5478 	if (vdc->minfo == NULL)
5479 		vdc->minfo = kmem_zalloc(sizeof (struct dk_minfo), KM_SLEEP);
5480 	vdc->minfo->dki_media_type = DK_FIXED_DISK;
5481 	vdc->minfo->dki_capacity = vdc->vdisk_size;
5482 	vdc->minfo->dki_lbsize = DEV_BSIZE;
5483 }
5484 
5485 static ushort_t
5486 vdc_lbl2cksum(struct dk_label *label)
5487 {
5488 	int	count;
5489 	ushort_t sum, *sp;
5490 
5491 	count =	(sizeof (struct dk_label)) / (sizeof (short)) - 1;
5492 	sp = (ushort_t *)label;
5493 	sum = 0;
5494 	while (count--) {
5495 		sum ^= *sp++;
5496 	}
5497 
5498 	return (sum);
5499 }
5500 
5501 /*
5502  * Function:
5503  *	vdc_validate_geometry
5504  *
5505  * Description:
5506  *	This routine discovers the label and geometry of the disk. It stores
5507  *	the disk label and related information in the vdc structure. If it
5508  *	fails to validate the geometry or to discover the disk label then
5509  *	the label is marked as unknown (VD_DISK_LABEL_UNK).
5510  *
5511  * Arguments:
5512  *	vdc	- soft state pointer for this instance of the device driver.
5513  *
5514  * Return Code:
5515  *	0	- success.
5516  *	EINVAL	- unknown disk label.
5517  *	ENOTSUP	- geometry not applicable (EFI label).
5518  *	EIO	- error accessing the disk.
5519  */
5520 static int
5521 vdc_validate_geometry(vdc_t *vdc)
5522 {
5523 	buf_t	*buf;	/* BREAD requests need to be in a buf_t structure */
5524 	dev_t	dev;
5525 	int	rv;
5526 	struct dk_label label;
5527 	struct dk_geom geom;
5528 	struct vtoc vtoc;
5529 
5530 	ASSERT(vdc != NULL);
5531 	ASSERT(vdc->vtoc != NULL && vdc->geom != NULL);
5532 	ASSERT(MUTEX_HELD(&vdc->lock));
5533 
5534 	mutex_exit(&vdc->lock);
5535 
5536 	dev = makedevice(ddi_driver_major(vdc->dip),
5537 	    VD_MAKE_DEV(vdc->instance, 0));
5538 
5539 	rv = vd_process_ioctl(dev, DKIOCGGEOM, (caddr_t)&geom, FKIOCTL);
5540 	if (rv == 0)
5541 		rv = vd_process_ioctl(dev, DKIOCGVTOC, (caddr_t)&vtoc, FKIOCTL);
5542 
5543 	if (rv == ENOTSUP) {
5544 		/*
5545 		 * If the device does not support VTOC then we try
5546 		 * to read an EFI label.
5547 		 */
5548 		struct dk_gpt *efi;
5549 		size_t efi_len;
5550 
5551 		rv = vdc_efi_alloc_and_read(dev, &efi, &efi_len);
5552 
5553 		if (rv) {
5554 			DMSG(vdc, 0, "[%d] Failed to get EFI (err=%d)",
5555 			    vdc->instance, rv);
5556 			mutex_enter(&vdc->lock);
5557 			vdc_store_label_unk(vdc);
5558 			return (EIO);
5559 		}
5560 
5561 		mutex_enter(&vdc->lock);
5562 		vdc_store_label_efi(vdc, efi);
5563 		vd_efi_free(efi, efi_len);
5564 		return (ENOTSUP);
5565 	}
5566 
5567 	if (rv != 0) {
5568 		DMSG(vdc, 0, "[%d] Failed to get VTOC (err=%d)",
5569 		    vdc->instance, rv);
5570 		mutex_enter(&vdc->lock);
5571 		vdc_store_label_unk(vdc);
5572 		if (rv != EINVAL)
5573 			rv = EIO;
5574 		return (rv);
5575 	}
5576 
5577 	/* check that geometry and vtoc are valid */
5578 	if (geom.dkg_nhead == 0 || geom.dkg_nsect == 0 ||
5579 	    vtoc.v_sanity != VTOC_SANE) {
5580 		mutex_enter(&vdc->lock);
5581 		vdc_store_label_unk(vdc);
5582 		return (EINVAL);
5583 	}
5584 
5585 	/*
5586 	 * We have a disk and a valid VTOC. However this does not mean
5587 	 * that the disk currently have a VTOC label. The returned VTOC may
5588 	 * be a default VTOC to be used for configuring the disk (this is
5589 	 * what is done for disk image). So we read the label from the
5590 	 * beginning of the disk to ensure we really have a VTOC label.
5591 	 *
5592 	 * FUTURE: This could be the default way for reading the VTOC
5593 	 * from the disk as opposed to sending the VD_OP_GET_VTOC
5594 	 * to the server. This will be the default if vdc is implemented
5595 	 * ontop of cmlb.
5596 	 */
5597 
5598 	/*
5599 	 * Single slice disk does not support read using an absolute disk
5600 	 * offset so we just rely on the DKIOCGVTOC ioctl in that case.
5601 	 */
5602 	if (vdc->vdisk_type == VD_DISK_TYPE_SLICE) {
5603 		mutex_enter(&vdc->lock);
5604 		if (vtoc.v_nparts != 1) {
5605 			vdc_store_label_unk(vdc);
5606 			return (EINVAL);
5607 		}
5608 		vdc_store_label_vtoc(vdc, &geom, &vtoc);
5609 		return (0);
5610 	}
5611 
5612 	if (vtoc.v_nparts != V_NUMPAR) {
5613 		mutex_enter(&vdc->lock);
5614 		vdc_store_label_unk(vdc);
5615 		return (EINVAL);
5616 	}
5617 
5618 	/*
5619 	 * Read disk label from start of disk
5620 	 */
5621 	buf = kmem_alloc(sizeof (buf_t), KM_SLEEP);
5622 	bioinit(buf);
5623 	buf->b_un.b_addr = (caddr_t)&label;
5624 	buf->b_bcount = DK_LABEL_SIZE;
5625 	buf->b_flags = B_BUSY | B_READ;
5626 	buf->b_dev = dev;
5627 	rv = vdc_send_request(vdc, VD_OP_BREAD, (caddr_t)&label,
5628 	    DK_LABEL_SIZE, VD_SLICE_NONE, 0, CB_STRATEGY, buf, VIO_read_dir);
5629 	if (rv) {
5630 		DMSG(vdc, 1, "[%d] Failed to read disk block 0\n",
5631 		    vdc->instance);
5632 	} else {
5633 		rv = biowait(buf);
5634 		biofini(buf);
5635 	}
5636 	kmem_free(buf, sizeof (buf_t));
5637 
5638 	if (rv != 0 || label.dkl_magic != DKL_MAGIC ||
5639 	    label.dkl_cksum != vdc_lbl2cksum(&label)) {
5640 		DMSG(vdc, 1, "[%d] Got VTOC with invalid label\n",
5641 		    vdc->instance);
5642 		mutex_enter(&vdc->lock);
5643 		vdc_store_label_unk(vdc);
5644 		return (EINVAL);
5645 	}
5646 
5647 	mutex_enter(&vdc->lock);
5648 	vdc_store_label_vtoc(vdc, &geom, &vtoc);
5649 	return (0);
5650 }
5651 
5652 /*
5653  * Function:
5654  *	vdc_validate
5655  *
5656  * Description:
5657  *	This routine discovers the label of the disk and create the
5658  *	appropriate device nodes if the label has changed.
5659  *
5660  * Arguments:
5661  *	vdc	- soft state pointer for this instance of the device driver.
5662  *
5663  * Return Code:
5664  *	none.
5665  */
5666 static void
5667 vdc_validate(vdc_t *vdc)
5668 {
5669 	vd_disk_label_t old_label;
5670 	struct vtoc old_vtoc;
5671 	int rv;
5672 
5673 	ASSERT(!MUTEX_HELD(&vdc->lock));
5674 
5675 	mutex_enter(&vdc->lock);
5676 
5677 	/* save the current label and vtoc */
5678 	old_label = vdc->vdisk_label;
5679 	bcopy(vdc->vtoc, &old_vtoc, sizeof (struct vtoc));
5680 
5681 	/* check the geometry */
5682 	(void) vdc_validate_geometry(vdc);
5683 
5684 	/* if the disk label has changed, update device nodes */
5685 	if (vdc->vdisk_label != old_label) {
5686 
5687 		if (vdc->vdisk_label == VD_DISK_LABEL_EFI)
5688 			rv = vdc_create_device_nodes_efi(vdc);
5689 		else
5690 			rv = vdc_create_device_nodes_vtoc(vdc);
5691 
5692 		if (rv != 0) {
5693 			DMSG(vdc, 0, "![%d] Failed to update device nodes",
5694 			    vdc->instance);
5695 		}
5696 	}
5697 
5698 	/* if the vtoc has changed, update device nodes properties */
5699 	if (bcmp(vdc->vtoc, &old_vtoc, sizeof (struct vtoc)) != 0) {
5700 
5701 		if (vdc_create_device_nodes_props(vdc) != 0) {
5702 			DMSG(vdc, 0, "![%d] Failed to update device nodes"
5703 			    " properties", vdc->instance);
5704 		}
5705 	}
5706 
5707 	mutex_exit(&vdc->lock);
5708 }
5709 
5710 static void
5711 vdc_validate_task(void *arg)
5712 {
5713 	vdc_t *vdc = (vdc_t *)arg;
5714 
5715 	vdc_validate(vdc);
5716 
5717 	mutex_enter(&vdc->lock);
5718 	ASSERT(vdc->validate_pending > 0);
5719 	vdc->validate_pending--;
5720 	mutex_exit(&vdc->lock);
5721 }
5722 
5723 /*
5724  * Function:
5725  *	vdc_setup_devid()
5726  *
5727  * Description:
5728  *	This routine discovers the devid of a vDisk. It requests the devid of
5729  *	the underlying device from the vDisk server, builds an encapsulated
5730  *	devid based on the retrieved devid and registers that new devid to
5731  *	the vDisk.
5732  *
5733  * Arguments:
5734  *	vdc	- soft state pointer for this instance of the device driver.
5735  *
5736  * Return Code:
5737  *	0	- A devid was succesfully registered for the vDisk
5738  */
5739 static int
5740 vdc_setup_devid(vdc_t *vdc)
5741 {
5742 	int rv;
5743 	vd_devid_t *vd_devid;
5744 	size_t bufsize, bufid_len;
5745 
5746 	/*
5747 	 * At first sight, we don't know the size of the devid that the
5748 	 * server will return but this size will be encoded into the
5749 	 * reply. So we do a first request using a default size then we
5750 	 * check if this size was large enough. If not then we do a second
5751 	 * request with the correct size returned by the server. Note that
5752 	 * ldc requires size to be 8-byte aligned.
5753 	 */
5754 	bufsize = P2ROUNDUP(VD_DEVID_SIZE(VD_DEVID_DEFAULT_LEN),
5755 	    sizeof (uint64_t));
5756 	vd_devid = kmem_zalloc(bufsize, KM_SLEEP);
5757 	bufid_len = bufsize - sizeof (vd_efi_t) - 1;
5758 
5759 	rv = vdc_do_sync_op(vdc, VD_OP_GET_DEVID, (caddr_t)vd_devid,
5760 	    bufsize, 0, 0, CB_SYNC, 0, VIO_both_dir);
5761 
5762 	DMSG(vdc, 2, "sync_op returned %d\n", rv);
5763 
5764 	if (rv) {
5765 		kmem_free(vd_devid, bufsize);
5766 		return (rv);
5767 	}
5768 
5769 	if (vd_devid->length > bufid_len) {
5770 		/*
5771 		 * The returned devid is larger than the buffer used. Try again
5772 		 * with a buffer with the right size.
5773 		 */
5774 		kmem_free(vd_devid, bufsize);
5775 		bufsize = P2ROUNDUP(VD_DEVID_SIZE(vd_devid->length),
5776 		    sizeof (uint64_t));
5777 		vd_devid = kmem_zalloc(bufsize, KM_SLEEP);
5778 		bufid_len = bufsize - sizeof (vd_efi_t) - 1;
5779 
5780 		rv = vdc_do_sync_op(vdc, VD_OP_GET_DEVID,
5781 		    (caddr_t)vd_devid, bufsize, 0, 0, CB_SYNC, 0,
5782 		    VIO_both_dir);
5783 
5784 		if (rv) {
5785 			kmem_free(vd_devid, bufsize);
5786 			return (rv);
5787 		}
5788 	}
5789 
5790 	/*
5791 	 * The virtual disk should have the same device id as the one associated
5792 	 * with the physical disk it is mapped on, otherwise sharing a disk
5793 	 * between a LDom and a non-LDom may not work (for example for a shared
5794 	 * SVM disk set).
5795 	 *
5796 	 * The DDI framework does not allow creating a device id with any
5797 	 * type so we first create a device id of type DEVID_ENCAP and then
5798 	 * we restore the orignal type of the physical device.
5799 	 */
5800 
5801 	DMSG(vdc, 2, ": devid length = %d\n", vd_devid->length);
5802 
5803 	/* build an encapsulated devid based on the returned devid */
5804 	if (ddi_devid_init(vdc->dip, DEVID_ENCAP, vd_devid->length,
5805 	    vd_devid->id, &vdc->devid) != DDI_SUCCESS) {
5806 		DMSG(vdc, 1, "[%d] Fail to created devid\n", vdc->instance);
5807 		kmem_free(vd_devid, bufsize);
5808 		return (1);
5809 	}
5810 
5811 	DEVID_FORMTYPE((impl_devid_t *)vdc->devid, vd_devid->type);
5812 
5813 	ASSERT(ddi_devid_valid(vdc->devid) == DDI_SUCCESS);
5814 
5815 	kmem_free(vd_devid, bufsize);
5816 
5817 	if (ddi_devid_register(vdc->dip, vdc->devid) != DDI_SUCCESS) {
5818 		DMSG(vdc, 1, "[%d] Fail to register devid\n", vdc->instance);
5819 		return (1);
5820 	}
5821 
5822 	return (0);
5823 }
5824 
5825 static void
5826 vdc_store_label_efi(vdc_t *vdc, struct dk_gpt *efi)
5827 {
5828 	struct vtoc *vtoc = vdc->vtoc;
5829 
5830 	ASSERT(MUTEX_HELD(&vdc->lock));
5831 
5832 	vdc->vdisk_label = VD_DISK_LABEL_EFI;
5833 	bzero(vdc->geom, sizeof (struct dk_geom));
5834 	vd_efi_to_vtoc(efi, vtoc);
5835 	if (vdc->vdisk_type == VD_DISK_TYPE_SLICE) {
5836 		/*
5837 		 * vd_efi_to_vtoc() will store information about the EFI Sun
5838 		 * reserved partition (representing the entire disk) into
5839 		 * partition 7. However single-slice device will only have
5840 		 * that single partition and the vdc driver expects to find
5841 		 * information about that partition in slice 0. So we need
5842 		 * to copy information from slice 7 to slice 0.
5843 		 */
5844 		vtoc->v_part[0].p_tag = vtoc->v_part[VD_EFI_WD_SLICE].p_tag;
5845 		vtoc->v_part[0].p_flag = vtoc->v_part[VD_EFI_WD_SLICE].p_flag;
5846 		vtoc->v_part[0].p_start = vtoc->v_part[VD_EFI_WD_SLICE].p_start;
5847 		vtoc->v_part[0].p_size =  vtoc->v_part[VD_EFI_WD_SLICE].p_size;
5848 	}
5849 }
5850 
5851 static void
5852 vdc_store_label_vtoc(vdc_t *vdc, struct dk_geom *geom, struct vtoc *vtoc)
5853 {
5854 	ASSERT(MUTEX_HELD(&vdc->lock));
5855 
5856 	vdc->vdisk_label = VD_DISK_LABEL_VTOC;
5857 	bcopy(vtoc, vdc->vtoc, sizeof (struct vtoc));
5858 	bcopy(geom, vdc->geom, sizeof (struct dk_geom));
5859 }
5860 
5861 static void
5862 vdc_store_label_unk(vdc_t *vdc)
5863 {
5864 	ASSERT(MUTEX_HELD(&vdc->lock));
5865 
5866 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
5867 	bzero(vdc->vtoc, sizeof (struct vtoc));
5868 	bzero(vdc->geom, sizeof (struct dk_geom));
5869 }
5870