xref: /titanic_41/usr/src/uts/sun4v/io/vdc.c (revision d29f5a711240f866521445b1656d114da090335e)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #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/kstat.h>
69 #include <sys/mach_descrip.h>
70 #include <sys/modctl.h>
71 #include <sys/mdeg.h>
72 #include <sys/note.h>
73 #include <sys/open.h>
74 #include <sys/sdt.h>
75 #include <sys/stat.h>
76 #include <sys/sunddi.h>
77 #include <sys/types.h>
78 #include <sys/promif.h>
79 #include <sys/var.h>
80 #include <sys/vtoc.h>
81 #include <sys/archsystm.h>
82 #include <sys/sysmacros.h>
83 
84 #include <sys/cdio.h>
85 #include <sys/dktp/fdisk.h>
86 #include <sys/dktp/dadkio.h>
87 #include <sys/mhd.h>
88 #include <sys/scsi/generic/sense.h>
89 #include <sys/scsi/impl/uscsi.h>
90 #include <sys/scsi/impl/services.h>
91 #include <sys/scsi/targets/sddef.h>
92 
93 #include <sys/ldoms.h>
94 #include <sys/ldc.h>
95 #include <sys/vio_common.h>
96 #include <sys/vio_mailbox.h>
97 #include <sys/vio_util.h>
98 #include <sys/vdsk_common.h>
99 #include <sys/vdsk_mailbox.h>
100 #include <sys/vdc.h>
101 
102 /*
103  * function prototypes
104  */
105 
106 /* standard driver functions */
107 static int	vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred);
108 static int	vdc_close(dev_t dev, int flag, int otyp, cred_t *cred);
109 static int	vdc_strategy(struct buf *buf);
110 static int	vdc_print(dev_t dev, char *str);
111 static int	vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk);
112 static int	vdc_read(dev_t dev, struct uio *uio, cred_t *cred);
113 static int	vdc_write(dev_t dev, struct uio *uio, cred_t *cred);
114 static int	vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode,
115 			cred_t *credp, int *rvalp);
116 static int	vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred);
117 static int	vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred);
118 
119 static int	vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,
120 			void *arg, void **resultp);
121 static int	vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd);
122 static int	vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd);
123 static int	vdc_prop_op(dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op,
124 		    int mod_flags, char *name, caddr_t valuep, int *lengthp);
125 
126 /* setup */
127 static void	vdc_min(struct buf *bufp);
128 static int	vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen);
129 static int	vdc_do_ldc_init(vdc_t *vdc, vdc_server_t *srvr);
130 static int	vdc_start_ldc_connection(vdc_t *vdc);
131 static int	vdc_create_device_nodes(vdc_t *vdc);
132 static int	vdc_create_device_nodes_efi(vdc_t *vdc);
133 static int	vdc_create_device_nodes_vtoc(vdc_t *vdc);
134 static void	vdc_create_io_kstats(vdc_t *vdc);
135 static void	vdc_create_err_kstats(vdc_t *vdc);
136 static void	vdc_set_err_kstats(vdc_t *vdc);
137 static int	vdc_get_md_node(dev_info_t *dip, md_t **mdpp,
138 		    mde_cookie_t *vd_nodep);
139 static int	vdc_init_ports(vdc_t *vdc, md_t *mdp, mde_cookie_t vd_nodep);
140 static void	vdc_fini_ports(vdc_t *vdc);
141 static void	vdc_switch_server(vdc_t *vdcp);
142 static int	vdc_do_ldc_up(vdc_t *vdc);
143 static void	vdc_terminate_ldc(vdc_t *vdc, vdc_server_t *srvr);
144 static int	vdc_init_descriptor_ring(vdc_t *vdc);
145 static void	vdc_destroy_descriptor_ring(vdc_t *vdc);
146 static int	vdc_setup_devid(vdc_t *vdc);
147 static void	vdc_store_label_efi(vdc_t *, efi_gpt_t *, efi_gpe_t *);
148 static void	vdc_store_label_vtoc(vdc_t *, struct dk_geom *, struct vtoc *);
149 static void	vdc_store_label_unk(vdc_t *vdc);
150 static boolean_t vdc_is_opened(vdc_t *vdc);
151 
152 /* handshake with vds */
153 static int		vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver);
154 static int		vdc_ver_negotiation(vdc_t *vdcp);
155 static int		vdc_init_attr_negotiation(vdc_t *vdc);
156 static int		vdc_attr_negotiation(vdc_t *vdcp);
157 static int		vdc_init_dring_negotiate(vdc_t *vdc);
158 static int		vdc_dring_negotiation(vdc_t *vdcp);
159 static int		vdc_send_rdx(vdc_t *vdcp);
160 static int		vdc_rdx_exchange(vdc_t *vdcp);
161 static boolean_t	vdc_is_supported_version(vio_ver_msg_t *ver_msg);
162 
163 /* processing incoming messages from vDisk server */
164 static void	vdc_process_msg_thread(vdc_t *vdc);
165 static int	vdc_recv(vdc_t *vdc, vio_msg_t *msgp, size_t *nbytesp);
166 
167 static uint_t	vdc_handle_cb(uint64_t event, caddr_t arg);
168 static int	vdc_process_data_msg(vdc_t *vdc, vio_msg_t *msg);
169 static int	vdc_handle_ver_msg(vdc_t *vdc, vio_ver_msg_t *ver_msg);
170 static int	vdc_handle_attr_msg(vdc_t *vdc, vd_attr_msg_t *attr_msg);
171 static int	vdc_handle_dring_reg_msg(vdc_t *vdc, vio_dring_reg_msg_t *msg);
172 static int 	vdc_send_request(vdc_t *vdcp, int operation,
173 		    caddr_t addr, size_t nbytes, int slice, diskaddr_t offset,
174 		    int cb_type, void *cb_arg, vio_desc_direction_t dir);
175 static int	vdc_map_to_shared_dring(vdc_t *vdcp, int idx);
176 static int 	vdc_populate_descriptor(vdc_t *vdcp, int operation,
177 		    caddr_t addr, size_t nbytes, int slice, diskaddr_t offset,
178 		    int cb_type, void *cb_arg, vio_desc_direction_t dir);
179 static int 	vdc_do_sync_op(vdc_t *vdcp, int operation, caddr_t addr,
180 		    size_t nbytes, int slice, diskaddr_t offset, int cb_type,
181 		    void *cb_arg, vio_desc_direction_t dir, boolean_t);
182 
183 static int	vdc_wait_for_response(vdc_t *vdcp, vio_msg_t *msgp);
184 static int	vdc_drain_response(vdc_t *vdcp, struct buf *buf);
185 static int	vdc_depopulate_descriptor(vdc_t *vdc, uint_t idx);
186 static int	vdc_populate_mem_hdl(vdc_t *vdcp, vdc_local_desc_t *ldep);
187 static int	vdc_verify_seq_num(vdc_t *vdc, vio_dring_msg_t *dring_msg);
188 
189 /* dkio */
190 static int	vd_process_ioctl(dev_t dev, int cmd, caddr_t arg, int mode,
191 		    int *rvalp);
192 static int	vd_process_efi_ioctl(void *vdisk, int cmd, uintptr_t arg);
193 static void	vdc_create_fake_geometry(vdc_t *vdc);
194 static int	vdc_validate_geometry(vdc_t *vdc);
195 static void	vdc_validate(vdc_t *vdc);
196 static void	vdc_validate_task(void *arg);
197 static int	vdc_null_copy_func(vdc_t *vdc, void *from, void *to,
198 		    int mode, int dir);
199 static int	vdc_get_wce_convert(vdc_t *vdc, void *from, void *to,
200 		    int mode, int dir);
201 static int	vdc_set_wce_convert(vdc_t *vdc, void *from, void *to,
202 		    int mode, int dir);
203 static int	vdc_get_vtoc_convert(vdc_t *vdc, void *from, void *to,
204 		    int mode, int dir);
205 static int	vdc_set_vtoc_convert(vdc_t *vdc, void *from, void *to,
206 		    int mode, int dir);
207 static int	vdc_get_geom_convert(vdc_t *vdc, void *from, void *to,
208 		    int mode, int dir);
209 static int	vdc_set_geom_convert(vdc_t *vdc, void *from, void *to,
210 		    int mode, int dir);
211 static int	vdc_get_efi_convert(vdc_t *vdc, void *from, void *to,
212 		    int mode, int dir);
213 static int	vdc_set_efi_convert(vdc_t *vdc, void *from, void *to,
214 		    int mode, int dir);
215 
216 static void 	vdc_ownership_update(vdc_t *vdc, int ownership_flags);
217 static int	vdc_access_set(vdc_t *vdc, uint64_t flags, int mode);
218 static vdc_io_t	*vdc_failfast_io_queue(vdc_t *vdc, struct buf *buf);
219 static int	vdc_failfast_check_resv(vdc_t *vdc);
220 
221 /*
222  * Module variables
223  */
224 
225 /*
226  * Tunable variables to control how long vdc waits before timing out on
227  * various operations
228  */
229 static int	vdc_hshake_retries = 3;
230 
231 static int	vdc_timeout = 0; /* units: seconds */
232 static int 	vdc_ldcup_timeout = 1; /* units: seconds */
233 
234 static uint64_t vdc_hz_min_ldc_delay;
235 static uint64_t vdc_min_timeout_ldc = 1 * MILLISEC;
236 static uint64_t vdc_hz_max_ldc_delay;
237 static uint64_t vdc_max_timeout_ldc = 100 * MILLISEC;
238 
239 static uint64_t vdc_ldc_read_init_delay = 1 * MILLISEC;
240 static uint64_t vdc_ldc_read_max_delay = 100 * MILLISEC;
241 
242 /* values for dumping - need to run in a tighter loop */
243 static uint64_t	vdc_usec_timeout_dump = 100 * MILLISEC;	/* 0.1s units: ns */
244 static int	vdc_dump_retries = 100;
245 
246 static uint16_t	vdc_scsi_timeout = 60;	/* 60s units: seconds  */
247 
248 static uint64_t vdc_ownership_delay = 6 * MICROSEC; /* 6s units: usec */
249 
250 /* Count of the number of vdc instances attached */
251 static volatile uint32_t	vdc_instance_count = 0;
252 
253 /* Tunable to log all SCSI errors */
254 static boolean_t vdc_scsi_log_error = B_FALSE;
255 
256 /* Soft state pointer */
257 static void	*vdc_state;
258 
259 /*
260  * Controlling the verbosity of the error/debug messages
261  *
262  * vdc_msglevel - controls level of messages
263  * vdc_matchinst - 64-bit variable where each bit corresponds
264  *                 to the vdc instance the vdc_msglevel applies.
265  */
266 int		vdc_msglevel = 0x0;
267 uint64_t	vdc_matchinst = 0ull;
268 
269 /*
270  * Supported vDisk protocol version pairs.
271  *
272  * The first array entry is the latest and preferred version.
273  */
274 static const vio_ver_t	vdc_version[] = {{1, 1}};
275 
276 static struct cb_ops vdc_cb_ops = {
277 	vdc_open,	/* cb_open */
278 	vdc_close,	/* cb_close */
279 	vdc_strategy,	/* cb_strategy */
280 	vdc_print,	/* cb_print */
281 	vdc_dump,	/* cb_dump */
282 	vdc_read,	/* cb_read */
283 	vdc_write,	/* cb_write */
284 	vdc_ioctl,	/* cb_ioctl */
285 	nodev,		/* cb_devmap */
286 	nodev,		/* cb_mmap */
287 	nodev,		/* cb_segmap */
288 	nochpoll,	/* cb_chpoll */
289 	vdc_prop_op,	/* cb_prop_op */
290 	NULL,		/* cb_str */
291 	D_MP | D_64BIT,	/* cb_flag */
292 	CB_REV,		/* cb_rev */
293 	vdc_aread,	/* cb_aread */
294 	vdc_awrite	/* cb_awrite */
295 };
296 
297 static struct dev_ops vdc_ops = {
298 	DEVO_REV,	/* devo_rev */
299 	0,		/* devo_refcnt */
300 	vdc_getinfo,	/* devo_getinfo */
301 	nulldev,	/* devo_identify */
302 	nulldev,	/* devo_probe */
303 	vdc_attach,	/* devo_attach */
304 	vdc_detach,	/* devo_detach */
305 	nodev,		/* devo_reset */
306 	&vdc_cb_ops,	/* devo_cb_ops */
307 	NULL,		/* devo_bus_ops */
308 	nulldev		/* devo_power */
309 };
310 
311 static struct modldrv modldrv = {
312 	&mod_driverops,
313 	"virtual disk client",
314 	&vdc_ops,
315 };
316 
317 static struct modlinkage modlinkage = {
318 	MODREV_1,
319 	&modldrv,
320 	NULL
321 };
322 
323 /* -------------------------------------------------------------------------- */
324 
325 /*
326  * Device Driver housekeeping and setup
327  */
328 
329 int
330 _init(void)
331 {
332 	int	status;
333 
334 	if ((status = ddi_soft_state_init(&vdc_state, sizeof (vdc_t), 1)) != 0)
335 		return (status);
336 	if ((status = mod_install(&modlinkage)) != 0)
337 		ddi_soft_state_fini(&vdc_state);
338 	return (status);
339 }
340 
341 int
342 _info(struct modinfo *modinfop)
343 {
344 	return (mod_info(&modlinkage, modinfop));
345 }
346 
347 int
348 _fini(void)
349 {
350 	int	status;
351 
352 	if ((status = mod_remove(&modlinkage)) != 0)
353 		return (status);
354 	ddi_soft_state_fini(&vdc_state);
355 	return (0);
356 }
357 
358 static int
359 vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,  void *arg, void **resultp)
360 {
361 	_NOTE(ARGUNUSED(dip))
362 
363 	int	instance = VDCUNIT((dev_t)arg);
364 	vdc_t	*vdc = NULL;
365 
366 	switch (cmd) {
367 	case DDI_INFO_DEVT2DEVINFO:
368 		if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
369 			*resultp = NULL;
370 			return (DDI_FAILURE);
371 		}
372 		*resultp = vdc->dip;
373 		return (DDI_SUCCESS);
374 	case DDI_INFO_DEVT2INSTANCE:
375 		*resultp = (void *)(uintptr_t)instance;
376 		return (DDI_SUCCESS);
377 	default:
378 		*resultp = NULL;
379 		return (DDI_FAILURE);
380 	}
381 }
382 
383 static int
384 vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
385 {
386 	kt_did_t failfast_tid, ownership_tid;
387 	int	instance;
388 	int	rv;
389 	vdc_server_t *srvr;
390 	vdc_t	*vdc = NULL;
391 
392 	switch (cmd) {
393 	case DDI_DETACH:
394 		/* the real work happens below */
395 		break;
396 	case DDI_SUSPEND:
397 		/* nothing to do for this non-device */
398 		return (DDI_SUCCESS);
399 	default:
400 		return (DDI_FAILURE);
401 	}
402 
403 	ASSERT(cmd == DDI_DETACH);
404 	instance = ddi_get_instance(dip);
405 	DMSGX(1, "[%d] Entered\n", instance);
406 
407 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
408 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
409 		return (DDI_FAILURE);
410 	}
411 
412 	/*
413 	 * This function is called when vdc is detached or if it has failed to
414 	 * attach. In that case, the attach may have fail before the vdisk type
415 	 * has been set so we can't call vdc_is_opened(). However as the attach
416 	 * has failed, we know that the vdisk is not opened and we can safely
417 	 * detach.
418 	 */
419 	if (vdc->vdisk_type != VD_DISK_TYPE_UNK && vdc_is_opened(vdc)) {
420 		DMSG(vdc, 0, "[%d] Cannot detach: device is open", instance);
421 		return (DDI_FAILURE);
422 	}
423 
424 	if (vdc->dkio_flush_pending) {
425 		DMSG(vdc, 0,
426 		    "[%d] Cannot detach: %d outstanding DKIO flushes\n",
427 		    instance, vdc->dkio_flush_pending);
428 		return (DDI_FAILURE);
429 	}
430 
431 	if (vdc->validate_pending) {
432 		DMSG(vdc, 0,
433 		    "[%d] Cannot detach: %d outstanding validate request\n",
434 		    instance, vdc->validate_pending);
435 		return (DDI_FAILURE);
436 	}
437 
438 	DMSG(vdc, 0, "[%d] proceeding...\n", instance);
439 
440 	/* If we took ownership, release ownership */
441 	mutex_enter(&vdc->ownership_lock);
442 	if (vdc->ownership & VDC_OWNERSHIP_GRANTED) {
443 		rv = vdc_access_set(vdc, VD_ACCESS_SET_CLEAR, FKIOCTL);
444 		if (rv == 0) {
445 			vdc_ownership_update(vdc, VDC_OWNERSHIP_NONE);
446 		}
447 	}
448 	mutex_exit(&vdc->ownership_lock);
449 
450 	/* mark instance as detaching */
451 	vdc->lifecycle	= VDC_LC_DETACHING;
452 
453 	/*
454 	 * Try and disable callbacks to prevent another handshake. We have to
455 	 * disable callbacks for all servers.
456 	 */
457 	for (srvr = vdc->server_list; srvr != NULL; srvr = srvr->next) {
458 		rv = ldc_set_cb_mode(srvr->ldc_handle, LDC_CB_DISABLE);
459 		DMSG(vdc, 0, "callback disabled (ldc=%lu, rv=%d)\n",
460 		    srvr->ldc_id, rv);
461 	}
462 
463 	if (vdc->initialized & VDC_THREAD) {
464 		mutex_enter(&vdc->read_lock);
465 		if ((vdc->read_state == VDC_READ_WAITING) ||
466 		    (vdc->read_state == VDC_READ_RESET)) {
467 			vdc->read_state = VDC_READ_RESET;
468 			cv_signal(&vdc->read_cv);
469 		}
470 
471 		mutex_exit(&vdc->read_lock);
472 
473 		/* wake up any thread waiting for connection to come online */
474 		mutex_enter(&vdc->lock);
475 		if (vdc->state == VDC_STATE_INIT_WAITING) {
476 			DMSG(vdc, 0,
477 			    "[%d] write reset - move to resetting state...\n",
478 			    instance);
479 			vdc->state = VDC_STATE_RESETTING;
480 			cv_signal(&vdc->initwait_cv);
481 		}
482 		mutex_exit(&vdc->lock);
483 
484 		/* now wait until state transitions to VDC_STATE_DETACH */
485 		thread_join(vdc->msg_proc_thr->t_did);
486 		ASSERT(vdc->state == VDC_STATE_DETACH);
487 		DMSG(vdc, 0, "[%d] Reset thread exit and join ..\n",
488 		    vdc->instance);
489 	}
490 
491 	mutex_enter(&vdc->lock);
492 
493 	if (vdc->initialized & VDC_DRING)
494 		vdc_destroy_descriptor_ring(vdc);
495 
496 	vdc_fini_ports(vdc);
497 
498 	if (vdc->failfast_thread) {
499 		failfast_tid = vdc->failfast_thread->t_did;
500 		vdc->failfast_interval = 0;
501 		cv_signal(&vdc->failfast_cv);
502 	} else {
503 		failfast_tid = 0;
504 	}
505 
506 	if (vdc->ownership & VDC_OWNERSHIP_WANTED) {
507 		ownership_tid = vdc->ownership_thread->t_did;
508 		vdc->ownership = VDC_OWNERSHIP_NONE;
509 		cv_signal(&vdc->ownership_cv);
510 	} else {
511 		ownership_tid = 0;
512 	}
513 
514 	mutex_exit(&vdc->lock);
515 
516 	if (failfast_tid != 0)
517 		thread_join(failfast_tid);
518 
519 	if (ownership_tid != 0)
520 		thread_join(ownership_tid);
521 
522 	if (vdc->initialized & VDC_MINOR)
523 		ddi_remove_minor_node(dip, NULL);
524 
525 	if (vdc->io_stats) {
526 		kstat_delete(vdc->io_stats);
527 		vdc->io_stats = NULL;
528 	}
529 
530 	if (vdc->err_stats) {
531 		kstat_delete(vdc->err_stats);
532 		vdc->err_stats = NULL;
533 	}
534 
535 	if (vdc->initialized & VDC_LOCKS) {
536 		mutex_destroy(&vdc->lock);
537 		mutex_destroy(&vdc->read_lock);
538 		mutex_destroy(&vdc->ownership_lock);
539 		cv_destroy(&vdc->initwait_cv);
540 		cv_destroy(&vdc->dring_free_cv);
541 		cv_destroy(&vdc->membind_cv);
542 		cv_destroy(&vdc->sync_pending_cv);
543 		cv_destroy(&vdc->sync_blocked_cv);
544 		cv_destroy(&vdc->read_cv);
545 		cv_destroy(&vdc->running_cv);
546 		cv_destroy(&vdc->ownership_cv);
547 		cv_destroy(&vdc->failfast_cv);
548 		cv_destroy(&vdc->failfast_io_cv);
549 	}
550 
551 	if (vdc->minfo)
552 		kmem_free(vdc->minfo, sizeof (struct dk_minfo));
553 
554 	if (vdc->cinfo)
555 		kmem_free(vdc->cinfo, sizeof (struct dk_cinfo));
556 
557 	if (vdc->vtoc)
558 		kmem_free(vdc->vtoc, sizeof (struct vtoc));
559 
560 	if (vdc->geom)
561 		kmem_free(vdc->geom, sizeof (struct dk_geom));
562 
563 	if (vdc->devid) {
564 		ddi_devid_unregister(dip);
565 		ddi_devid_free(vdc->devid);
566 	}
567 
568 	if (vdc->initialized & VDC_SOFT_STATE)
569 		ddi_soft_state_free(vdc_state, instance);
570 
571 	DMSG(vdc, 0, "[%d] End %p\n", instance, (void *)vdc);
572 
573 	return (DDI_SUCCESS);
574 }
575 
576 
577 static int
578 vdc_do_attach(dev_info_t *dip)
579 {
580 	int		instance;
581 	vdc_t		*vdc = NULL;
582 	int		status;
583 	md_t		*mdp;
584 	mde_cookie_t	vd_node;
585 
586 	ASSERT(dip != NULL);
587 
588 	instance = ddi_get_instance(dip);
589 	if (ddi_soft_state_zalloc(vdc_state, instance) != DDI_SUCCESS) {
590 		cmn_err(CE_NOTE, "[%d] Couldn't alloc state structure",
591 		    instance);
592 		return (DDI_FAILURE);
593 	}
594 
595 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
596 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
597 		return (DDI_FAILURE);
598 	}
599 
600 	/*
601 	 * We assign the value to initialized in this case to zero out the
602 	 * variable and then set bits in it to indicate what has been done
603 	 */
604 	vdc->initialized = VDC_SOFT_STATE;
605 
606 	vdc_hz_min_ldc_delay = drv_usectohz(vdc_min_timeout_ldc);
607 	vdc_hz_max_ldc_delay = drv_usectohz(vdc_max_timeout_ldc);
608 
609 	vdc->dip	= dip;
610 	vdc->instance	= instance;
611 	vdc->vdisk_type	= VD_DISK_TYPE_UNK;
612 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
613 	vdc->state	= VDC_STATE_INIT;
614 	vdc->lifecycle	= VDC_LC_ATTACHING;
615 	vdc->session_id = 0;
616 	vdc->block_size = DEV_BSIZE;
617 	vdc->max_xfer_sz = maxphys / DEV_BSIZE;
618 
619 	/*
620 	 * We assume, for now, that the vDisk server will export 'read'
621 	 * operations to us at a minimum (this is needed because of checks
622 	 * in vdc for supported operations early in the handshake process).
623 	 * The vDisk server will return ENOTSUP if this is not the case.
624 	 * The value will be overwritten during the attribute exchange with
625 	 * the bitmask of operations exported by server.
626 	 */
627 	vdc->operations = VD_OP_MASK_READ;
628 
629 	vdc->vtoc = NULL;
630 	vdc->geom = NULL;
631 	vdc->cinfo = NULL;
632 	vdc->minfo = NULL;
633 
634 	mutex_init(&vdc->lock, NULL, MUTEX_DRIVER, NULL);
635 	cv_init(&vdc->initwait_cv, NULL, CV_DRIVER, NULL);
636 	cv_init(&vdc->dring_free_cv, NULL, CV_DRIVER, NULL);
637 	cv_init(&vdc->membind_cv, NULL, CV_DRIVER, NULL);
638 	cv_init(&vdc->running_cv, NULL, CV_DRIVER, NULL);
639 
640 	vdc->threads_pending = 0;
641 	vdc->sync_op_pending = B_FALSE;
642 	vdc->sync_op_blocked = B_FALSE;
643 	cv_init(&vdc->sync_pending_cv, NULL, CV_DRIVER, NULL);
644 	cv_init(&vdc->sync_blocked_cv, NULL, CV_DRIVER, NULL);
645 
646 	mutex_init(&vdc->ownership_lock, NULL, MUTEX_DRIVER, NULL);
647 	cv_init(&vdc->ownership_cv, NULL, CV_DRIVER, NULL);
648 	cv_init(&vdc->failfast_cv, NULL, CV_DRIVER, NULL);
649 	cv_init(&vdc->failfast_io_cv, NULL, CV_DRIVER, NULL);
650 
651 	/* init blocking msg read functionality */
652 	mutex_init(&vdc->read_lock, NULL, MUTEX_DRIVER, NULL);
653 	cv_init(&vdc->read_cv, NULL, CV_DRIVER, NULL);
654 	vdc->read_state = VDC_READ_IDLE;
655 
656 	vdc->initialized |= VDC_LOCKS;
657 
658 	/* get device and port MD node for this disk instance */
659 	if (vdc_get_md_node(dip, &mdp, &vd_node) != 0) {
660 		cmn_err(CE_NOTE, "[%d] Could not get machine description node",
661 		    instance);
662 		return (DDI_FAILURE);
663 	}
664 
665 	if (vdc_init_ports(vdc, mdp, vd_node) != 0) {
666 		cmn_err(CE_NOTE, "[%d] Error initialising ports", instance);
667 		return (DDI_FAILURE);
668 	}
669 
670 	(void) md_fini_handle(mdp);
671 
672 	/* initialize the thread responsible for managing state with server */
673 	vdc->msg_proc_thr = thread_create(NULL, 0, vdc_process_msg_thread,
674 	    vdc, 0, &p0, TS_RUN, minclsyspri);
675 	if (vdc->msg_proc_thr == NULL) {
676 		cmn_err(CE_NOTE, "[%d] Failed to create msg processing thread",
677 		    instance);
678 		return (DDI_FAILURE);
679 	}
680 
681 	vdc->initialized |= VDC_THREAD;
682 
683 	/* Create the kstats for saving the I/O statistics used by iostat(1M) */
684 	vdc_create_io_kstats(vdc);
685 	vdc_create_err_kstats(vdc);
686 
687 	atomic_inc_32(&vdc_instance_count);
688 
689 	/*
690 	 * Check the disk label. This will send requests and do the handshake.
691 	 * We don't really care about the disk label now. What we really need is
692 	 * the handshake do be done so that we know the type of the disk (slice
693 	 * or full disk) and the appropriate device nodes can be created.
694 	 */
695 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
696 	vdc->vtoc = kmem_zalloc(sizeof (struct vtoc), KM_SLEEP);
697 	vdc->geom = kmem_zalloc(sizeof (struct dk_geom), KM_SLEEP);
698 	vdc->minfo = kmem_zalloc(sizeof (struct dk_minfo), KM_SLEEP);
699 
700 	mutex_enter(&vdc->lock);
701 	(void) vdc_validate_geometry(vdc);
702 	mutex_exit(&vdc->lock);
703 
704 	/*
705 	 * Now that we have the device info we can create the device nodes
706 	 */
707 	status = vdc_create_device_nodes(vdc);
708 	if (status) {
709 		DMSG(vdc, 0, "[%d] Failed to create device nodes",
710 		    instance);
711 		goto return_status;
712 	}
713 
714 	/*
715 	 * Setup devid
716 	 */
717 	if (vdc_setup_devid(vdc)) {
718 		DMSG(vdc, 0, "[%d] No device id available\n", instance);
719 	}
720 
721 	/*
722 	 * Fill in the fields of the error statistics kstat that were not
723 	 * available when creating the kstat
724 	 */
725 	vdc_set_err_kstats(vdc);
726 
727 	ddi_report_dev(dip);
728 	vdc->lifecycle	= VDC_LC_ONLINE;
729 	DMSG(vdc, 0, "[%d] Attach tasks successful\n", instance);
730 
731 return_status:
732 	DMSG(vdc, 0, "[%d] Attach completed\n", instance);
733 	return (status);
734 }
735 
736 static int
737 vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
738 {
739 	int	status;
740 
741 	switch (cmd) {
742 	case DDI_ATTACH:
743 		if ((status = vdc_do_attach(dip)) != 0)
744 			(void) vdc_detach(dip, DDI_DETACH);
745 		return (status);
746 	case DDI_RESUME:
747 		/* nothing to do for this non-device */
748 		return (DDI_SUCCESS);
749 	default:
750 		return (DDI_FAILURE);
751 	}
752 }
753 
754 static int
755 vdc_do_ldc_init(vdc_t *vdc, vdc_server_t *srvr)
756 {
757 	int			status = 0;
758 	ldc_status_t		ldc_state;
759 	ldc_attr_t		ldc_attr;
760 
761 	ASSERT(vdc != NULL);
762 	ASSERT(srvr != NULL);
763 
764 	ldc_attr.devclass = LDC_DEV_BLK;
765 	ldc_attr.instance = vdc->instance;
766 	ldc_attr.mode = LDC_MODE_UNRELIABLE;	/* unreliable transport */
767 	ldc_attr.mtu = VD_LDC_MTU;
768 
769 	if ((srvr->state & VDC_LDC_INIT) == 0) {
770 		status = ldc_init(srvr->ldc_id, &ldc_attr,
771 		    &srvr->ldc_handle);
772 		if (status != 0) {
773 			DMSG(vdc, 0, "[%d] ldc_init(chan %ld) returned %d",
774 			    vdc->instance, srvr->ldc_id, status);
775 			return (status);
776 		}
777 		srvr->state |= VDC_LDC_INIT;
778 	}
779 	status = ldc_status(srvr->ldc_handle, &ldc_state);
780 	if (status != 0) {
781 		DMSG(vdc, 0, "[%d] Cannot discover LDC status [err=%d]",
782 		    vdc->instance, status);
783 		goto init_exit;
784 	}
785 	srvr->ldc_state = ldc_state;
786 
787 	if ((srvr->state & VDC_LDC_CB) == 0) {
788 		status = ldc_reg_callback(srvr->ldc_handle, vdc_handle_cb,
789 		    (caddr_t)srvr);
790 		if (status != 0) {
791 			DMSG(vdc, 0, "[%d] LDC callback reg. failed (%d)",
792 			    vdc->instance, status);
793 			goto init_exit;
794 		}
795 		srvr->state |= VDC_LDC_CB;
796 	}
797 
798 	/*
799 	 * At this stage we have initialised LDC, we will now try and open
800 	 * the connection.
801 	 */
802 	if (srvr->ldc_state == LDC_INIT) {
803 		status = ldc_open(srvr->ldc_handle);
804 		if (status != 0) {
805 			DMSG(vdc, 0, "[%d] ldc_open(chan %ld) returned %d",
806 			    vdc->instance, srvr->ldc_id, status);
807 			goto init_exit;
808 		}
809 		srvr->state |= VDC_LDC_OPEN;
810 	}
811 
812 init_exit:
813 	if (status) {
814 		vdc_terminate_ldc(vdc, srvr);
815 	}
816 
817 	return (status);
818 }
819 
820 static int
821 vdc_start_ldc_connection(vdc_t *vdc)
822 {
823 	int		status = 0;
824 
825 	ASSERT(vdc != NULL);
826 
827 	ASSERT(MUTEX_HELD(&vdc->lock));
828 
829 	status = vdc_do_ldc_up(vdc);
830 
831 	DMSG(vdc, 0, "[%d] Finished bringing up LDC\n", vdc->instance);
832 
833 	return (status);
834 }
835 
836 static int
837 vdc_stop_ldc_connection(vdc_t *vdcp)
838 {
839 	int	status;
840 
841 	ASSERT(vdcp != NULL);
842 
843 	ASSERT(MUTEX_HELD(&vdcp->lock));
844 
845 	DMSG(vdcp, 0, ": Resetting connection to vDisk server : state %d\n",
846 	    vdcp->state);
847 
848 	status = ldc_down(vdcp->curr_server->ldc_handle);
849 	DMSG(vdcp, 0, "ldc_down() = %d\n", status);
850 
851 	vdcp->initialized &= ~VDC_HANDSHAKE;
852 	DMSG(vdcp, 0, "initialized=%x\n", vdcp->initialized);
853 
854 	return (status);
855 }
856 
857 static void
858 vdc_create_io_kstats(vdc_t *vdc)
859 {
860 	if (vdc->io_stats != NULL) {
861 		DMSG(vdc, 0, "[%d] I/O kstat already exists\n", vdc->instance);
862 		return;
863 	}
864 
865 	vdc->io_stats = kstat_create(VDC_DRIVER_NAME, vdc->instance, NULL,
866 	    "disk", KSTAT_TYPE_IO, 1, KSTAT_FLAG_PERSISTENT);
867 	if (vdc->io_stats != NULL) {
868 		vdc->io_stats->ks_lock = &vdc->lock;
869 		kstat_install(vdc->io_stats);
870 	} else {
871 		cmn_err(CE_NOTE, "[%d] Failed to create kstat: I/O statistics"
872 		    " will not be gathered", vdc->instance);
873 	}
874 }
875 
876 static void
877 vdc_create_err_kstats(vdc_t *vdc)
878 {
879 	vd_err_stats_t	*stp;
880 	char	kstatmodule_err[KSTAT_STRLEN];
881 	char	kstatname[KSTAT_STRLEN];
882 	int	ndata = (sizeof (vd_err_stats_t) / sizeof (kstat_named_t));
883 	int	instance = vdc->instance;
884 
885 	if (vdc->err_stats != NULL) {
886 		DMSG(vdc, 0, "[%d] ERR kstat already exists\n", vdc->instance);
887 		return;
888 	}
889 
890 	(void) snprintf(kstatmodule_err, sizeof (kstatmodule_err),
891 	    "%serr", VDC_DRIVER_NAME);
892 	(void) snprintf(kstatname, sizeof (kstatname),
893 	    "%s%d,err", VDC_DRIVER_NAME, instance);
894 
895 	vdc->err_stats = kstat_create(kstatmodule_err, instance, kstatname,
896 	    "device_error", KSTAT_TYPE_NAMED, ndata, KSTAT_FLAG_PERSISTENT);
897 
898 	if (vdc->err_stats == NULL) {
899 		cmn_err(CE_NOTE, "[%d] Failed to create kstat: Error statistics"
900 		    " will not be gathered", instance);
901 		return;
902 	}
903 
904 	stp = (vd_err_stats_t *)vdc->err_stats->ks_data;
905 	kstat_named_init(&stp->vd_softerrs,	"Soft Errors",
906 	    KSTAT_DATA_UINT32);
907 	kstat_named_init(&stp->vd_transerrs,	"Transport Errors",
908 	    KSTAT_DATA_UINT32);
909 	kstat_named_init(&stp->vd_protoerrs,	"Protocol Errors",
910 	    KSTAT_DATA_UINT32);
911 	kstat_named_init(&stp->vd_vid,		"Vendor",
912 	    KSTAT_DATA_CHAR);
913 	kstat_named_init(&stp->vd_pid,		"Product",
914 	    KSTAT_DATA_CHAR);
915 	kstat_named_init(&stp->vd_capacity,	"Size",
916 	    KSTAT_DATA_ULONGLONG);
917 
918 	vdc->err_stats->ks_update  = nulldev;
919 
920 	kstat_install(vdc->err_stats);
921 }
922 
923 static void
924 vdc_set_err_kstats(vdc_t *vdc)
925 {
926 	vd_err_stats_t  *stp;
927 
928 	if (vdc->err_stats == NULL)
929 		return;
930 
931 	mutex_enter(&vdc->lock);
932 
933 	stp = (vd_err_stats_t *)vdc->err_stats->ks_data;
934 	ASSERT(stp != NULL);
935 
936 	stp->vd_capacity.value.ui64 = vdc->vdisk_size * vdc->block_size;
937 	(void) strcpy(stp->vd_vid.value.c, "SUN");
938 	(void) strcpy(stp->vd_pid.value.c, "VDSK");
939 
940 	mutex_exit(&vdc->lock);
941 }
942 
943 static int
944 vdc_create_device_nodes_efi(vdc_t *vdc)
945 {
946 	ddi_remove_minor_node(vdc->dip, "h");
947 	ddi_remove_minor_node(vdc->dip, "h,raw");
948 
949 	if (ddi_create_minor_node(vdc->dip, "wd", S_IFBLK,
950 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
951 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
952 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'wd'",
953 		    vdc->instance);
954 		return (EIO);
955 	}
956 
957 	/* if any device node is created we set this flag */
958 	vdc->initialized |= VDC_MINOR;
959 
960 	if (ddi_create_minor_node(vdc->dip, "wd,raw", S_IFCHR,
961 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
962 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
963 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'wd,raw'",
964 		    vdc->instance);
965 		return (EIO);
966 	}
967 
968 	return (0);
969 }
970 
971 static int
972 vdc_create_device_nodes_vtoc(vdc_t *vdc)
973 {
974 	ddi_remove_minor_node(vdc->dip, "wd");
975 	ddi_remove_minor_node(vdc->dip, "wd,raw");
976 
977 	if (ddi_create_minor_node(vdc->dip, "h", S_IFBLK,
978 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
979 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
980 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'h'",
981 		    vdc->instance);
982 		return (EIO);
983 	}
984 
985 	/* if any device node is created we set this flag */
986 	vdc->initialized |= VDC_MINOR;
987 
988 	if (ddi_create_minor_node(vdc->dip, "h,raw", S_IFCHR,
989 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
990 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
991 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'h,raw'",
992 		    vdc->instance);
993 		return (EIO);
994 	}
995 
996 	return (0);
997 }
998 
999 /*
1000  * Function:
1001  *	vdc_create_device_nodes
1002  *
1003  * Description:
1004  *	This function creates the block and character device nodes under
1005  *	/devices. It is called as part of the attach(9E) of the instance
1006  *	during the handshake with vds after vds has sent the attributes
1007  *	to vdc.
1008  *
1009  *	If the device is of type VD_DISK_TYPE_SLICE then the minor node
1010  *	of 2 is used in keeping with the Solaris convention that slice 2
1011  *	refers to a whole disk. Slices start at 'a'
1012  *
1013  * Parameters:
1014  *	vdc 		- soft state pointer
1015  *
1016  * Return Values
1017  *	0		- Success
1018  *	EIO		- Failed to create node
1019  *	EINVAL		- Unknown type of disk exported
1020  */
1021 static int
1022 vdc_create_device_nodes(vdc_t *vdc)
1023 {
1024 	char		name[sizeof ("s,raw")];
1025 	dev_info_t	*dip = NULL;
1026 	int		instance, status;
1027 	int		num_slices = 1;
1028 	int		i;
1029 
1030 	ASSERT(vdc != NULL);
1031 
1032 	instance = vdc->instance;
1033 	dip = vdc->dip;
1034 
1035 	switch (vdc->vdisk_type) {
1036 	case VD_DISK_TYPE_DISK:
1037 		num_slices = V_NUMPAR;
1038 		break;
1039 	case VD_DISK_TYPE_SLICE:
1040 		num_slices = 1;
1041 		break;
1042 	case VD_DISK_TYPE_UNK:
1043 	default:
1044 		return (EINVAL);
1045 	}
1046 
1047 	/*
1048 	 * Minor nodes are different for EFI disks: EFI disks do not have
1049 	 * a minor node 'g' for the minor number corresponding to slice
1050 	 * VD_EFI_WD_SLICE (slice 7) instead they have a minor node 'wd'
1051 	 * representing the whole disk.
1052 	 */
1053 	for (i = 0; i < num_slices; i++) {
1054 
1055 		if (i == VD_EFI_WD_SLICE) {
1056 			if (vdc->vdisk_label == VD_DISK_LABEL_EFI)
1057 				status = vdc_create_device_nodes_efi(vdc);
1058 			else
1059 				status = vdc_create_device_nodes_vtoc(vdc);
1060 			if (status != 0)
1061 				return (status);
1062 			continue;
1063 		}
1064 
1065 		(void) snprintf(name, sizeof (name), "%c", 'a' + i);
1066 		if (ddi_create_minor_node(dip, name, S_IFBLK,
1067 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
1068 			cmn_err(CE_NOTE, "[%d] Couldn't add block node '%s'",
1069 			    instance, name);
1070 			return (EIO);
1071 		}
1072 
1073 		/* if any device node is created we set this flag */
1074 		vdc->initialized |= VDC_MINOR;
1075 
1076 		(void) snprintf(name, sizeof (name), "%c%s", 'a' + i, ",raw");
1077 
1078 		if (ddi_create_minor_node(dip, name, S_IFCHR,
1079 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
1080 			cmn_err(CE_NOTE, "[%d] Couldn't add raw node '%s'",
1081 			    instance, name);
1082 			return (EIO);
1083 		}
1084 	}
1085 
1086 	return (0);
1087 }
1088 
1089 /*
1090  * Driver prop_op(9e) entry point function. Return the number of blocks for
1091  * the partition in question or forward the request to the property facilities.
1092  */
1093 static int
1094 vdc_prop_op(dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op, int mod_flags,
1095     char *name, caddr_t valuep, int *lengthp)
1096 {
1097 	int instance = ddi_get_instance(dip);
1098 	vdc_t *vdc;
1099 	uint64_t nblocks;
1100 	uint_t blksize;
1101 
1102 	vdc = ddi_get_soft_state(vdc_state, instance);
1103 
1104 	if (dev == DDI_DEV_T_ANY || vdc == NULL) {
1105 		return (ddi_prop_op(dev, dip, prop_op, mod_flags,
1106 		    name, valuep, lengthp));
1107 	}
1108 
1109 	mutex_enter(&vdc->lock);
1110 	(void) vdc_validate_geometry(vdc);
1111 	if (vdc->vdisk_label == VD_DISK_LABEL_UNK) {
1112 		mutex_exit(&vdc->lock);
1113 		return (ddi_prop_op(dev, dip, prop_op, mod_flags,
1114 		    name, valuep, lengthp));
1115 	}
1116 	nblocks = vdc->slice[VDCPART(dev)].nblocks;
1117 	blksize = vdc->block_size;
1118 	mutex_exit(&vdc->lock);
1119 
1120 	return (ddi_prop_op_nblocks_blksize(dev, dip, prop_op, mod_flags,
1121 	    name, valuep, lengthp, nblocks, blksize));
1122 }
1123 
1124 /*
1125  * Function:
1126  *	vdc_is_opened
1127  *
1128  * Description:
1129  *	This function checks if any slice of a given virtual disk is
1130  *	currently opened.
1131  *
1132  * Parameters:
1133  *	vdc 		- soft state pointer
1134  *
1135  * Return Values
1136  *	B_TRUE		- at least one slice is opened.
1137  *	B_FALSE		- no slice is opened.
1138  */
1139 static boolean_t
1140 vdc_is_opened(vdc_t *vdc)
1141 {
1142 	int i, nslices;
1143 
1144 	switch (vdc->vdisk_type) {
1145 	case VD_DISK_TYPE_DISK:
1146 		nslices = V_NUMPAR;
1147 		break;
1148 	case VD_DISK_TYPE_SLICE:
1149 		nslices = 1;
1150 		break;
1151 	case VD_DISK_TYPE_UNK:
1152 	default:
1153 		ASSERT(0);
1154 	}
1155 
1156 	/* check if there's any layered open */
1157 	for (i = 0; i < nslices; i++) {
1158 		if (vdc->open_lyr[i] > 0)
1159 			return (B_TRUE);
1160 	}
1161 
1162 	/* check if there is any other kind of open */
1163 	for (i = 0; i < OTYPCNT; i++) {
1164 		if (vdc->open[i] != 0)
1165 			return (B_TRUE);
1166 	}
1167 
1168 	return (B_FALSE);
1169 }
1170 
1171 static int
1172 vdc_mark_opened(vdc_t *vdc, int slice, int flag, int otyp)
1173 {
1174 	uint8_t slicemask;
1175 	int i;
1176 
1177 	ASSERT(otyp < OTYPCNT);
1178 	ASSERT(slice < V_NUMPAR);
1179 	ASSERT(MUTEX_HELD(&vdc->lock));
1180 
1181 	slicemask = 1 << slice;
1182 
1183 	/* check if slice is already exclusively opened */
1184 	if (vdc->open_excl & slicemask)
1185 		return (EBUSY);
1186 
1187 	/* if open exclusive, check if slice is already opened */
1188 	if (flag & FEXCL) {
1189 		if (vdc->open_lyr[slice] > 0)
1190 			return (EBUSY);
1191 		for (i = 0; i < OTYPCNT; i++) {
1192 			if (vdc->open[i] & slicemask)
1193 				return (EBUSY);
1194 		}
1195 		vdc->open_excl |= slicemask;
1196 	}
1197 
1198 	/* mark slice as opened */
1199 	if (otyp == OTYP_LYR) {
1200 		vdc->open_lyr[slice]++;
1201 	} else {
1202 		vdc->open[otyp] |= slicemask;
1203 	}
1204 
1205 	return (0);
1206 }
1207 
1208 static void
1209 vdc_mark_closed(vdc_t *vdc, int slice, int flag, int otyp)
1210 {
1211 	uint8_t slicemask;
1212 
1213 	ASSERT(otyp < OTYPCNT);
1214 	ASSERT(slice < V_NUMPAR);
1215 	ASSERT(MUTEX_HELD(&vdc->lock));
1216 
1217 	slicemask = 1 << slice;
1218 
1219 	if (otyp == OTYP_LYR) {
1220 		ASSERT(vdc->open_lyr[slice] > 0);
1221 		vdc->open_lyr[slice]--;
1222 	} else {
1223 		vdc->open[otyp] &= ~slicemask;
1224 	}
1225 
1226 	if (flag & FEXCL)
1227 		vdc->open_excl &= ~slicemask;
1228 }
1229 
1230 static int
1231 vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred)
1232 {
1233 	_NOTE(ARGUNUSED(cred))
1234 
1235 	int	instance, nodelay;
1236 	int	slice, status = 0;
1237 	vdc_t	*vdc;
1238 
1239 	ASSERT(dev != NULL);
1240 	instance = VDCUNIT(*dev);
1241 
1242 	if (otyp >= OTYPCNT)
1243 		return (EINVAL);
1244 
1245 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1246 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1247 		return (ENXIO);
1248 	}
1249 
1250 	DMSG(vdc, 0, "minor = %d flag = %x, otyp = %x\n",
1251 	    getminor(*dev), flag, otyp);
1252 
1253 	slice = VDCPART(*dev);
1254 
1255 	nodelay = flag & (FNDELAY | FNONBLOCK);
1256 
1257 	if ((flag & FWRITE) && (!nodelay) &&
1258 	    !(VD_OP_SUPPORTED(vdc->operations, VD_OP_BWRITE))) {
1259 		return (EROFS);
1260 	}
1261 
1262 	mutex_enter(&vdc->lock);
1263 
1264 	status = vdc_mark_opened(vdc, slice, flag, otyp);
1265 
1266 	if (status != 0) {
1267 		mutex_exit(&vdc->lock);
1268 		return (status);
1269 	}
1270 
1271 	if (nodelay) {
1272 
1273 		/* don't resubmit a validate request if there's already one */
1274 		if (vdc->validate_pending > 0) {
1275 			mutex_exit(&vdc->lock);
1276 			return (0);
1277 		}
1278 
1279 		/* call vdc_validate() asynchronously to avoid blocking */
1280 		if (taskq_dispatch(system_taskq, vdc_validate_task,
1281 		    (void *)vdc, TQ_NOSLEEP) == NULL) {
1282 			vdc_mark_closed(vdc, slice, flag, otyp);
1283 			mutex_exit(&vdc->lock);
1284 			return (ENXIO);
1285 		}
1286 
1287 		vdc->validate_pending++;
1288 		mutex_exit(&vdc->lock);
1289 		return (0);
1290 	}
1291 
1292 	mutex_exit(&vdc->lock);
1293 
1294 	vdc_validate(vdc);
1295 
1296 	mutex_enter(&vdc->lock);
1297 
1298 	if (vdc->vdisk_label == VD_DISK_LABEL_UNK ||
1299 	    vdc->slice[slice].nblocks == 0) {
1300 		vdc_mark_closed(vdc, slice, flag, otyp);
1301 		status = EIO;
1302 	}
1303 
1304 	mutex_exit(&vdc->lock);
1305 
1306 	return (status);
1307 }
1308 
1309 static int
1310 vdc_close(dev_t dev, int flag, int otyp, cred_t *cred)
1311 {
1312 	_NOTE(ARGUNUSED(cred))
1313 
1314 	int	instance;
1315 	int	slice;
1316 	int	rv, rval;
1317 	vdc_t	*vdc;
1318 
1319 	instance = VDCUNIT(dev);
1320 
1321 	if (otyp >= OTYPCNT)
1322 		return (EINVAL);
1323 
1324 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1325 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1326 		return (ENXIO);
1327 	}
1328 
1329 	DMSG(vdc, 0, "[%d] flag = %x, otyp = %x\n", instance, flag, otyp);
1330 
1331 	slice = VDCPART(dev);
1332 
1333 	/*
1334 	 * Attempt to flush the W$ on a close operation. If this is
1335 	 * not a supported IOCTL command or the backing device is read-only
1336 	 * do not fail the close operation.
1337 	 */
1338 	rv = vd_process_ioctl(dev, DKIOCFLUSHWRITECACHE, NULL, FKIOCTL, &rval);
1339 
1340 	if (rv != 0 && rv != ENOTSUP && rv != ENOTTY && rv != EROFS) {
1341 		DMSG(vdc, 0, "[%d] flush failed with error %d on close\n",
1342 		    instance, rv);
1343 		return (EIO);
1344 	}
1345 
1346 	mutex_enter(&vdc->lock);
1347 	vdc_mark_closed(vdc, slice, flag, otyp);
1348 	mutex_exit(&vdc->lock);
1349 
1350 	return (0);
1351 }
1352 
1353 static int
1354 vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
1355 {
1356 	_NOTE(ARGUNUSED(credp))
1357 
1358 	return (vd_process_ioctl(dev, cmd, (caddr_t)arg, mode, rvalp));
1359 }
1360 
1361 static int
1362 vdc_print(dev_t dev, char *str)
1363 {
1364 	cmn_err(CE_NOTE, "vdc%d:  %s", VDCUNIT(dev), str);
1365 	return (0);
1366 }
1367 
1368 static int
1369 vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk)
1370 {
1371 	int	rv;
1372 	size_t	nbytes = nblk * DEV_BSIZE;
1373 	int	instance = VDCUNIT(dev);
1374 	vdc_t	*vdc = NULL;
1375 
1376 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1377 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1378 		return (ENXIO);
1379 	}
1380 
1381 	DMSG(vdc, 2, "[%d] dump %ld bytes at block 0x%lx : addr=0x%p\n",
1382 	    instance, nbytes, blkno, (void *)addr);
1383 	rv = vdc_send_request(vdc, VD_OP_BWRITE, addr, nbytes,
1384 	    VDCPART(dev), blkno, CB_STRATEGY, 0, VIO_write_dir);
1385 	if (rv) {
1386 		DMSG(vdc, 0, "Failed to do a disk dump (err=%d)\n", rv);
1387 		return (rv);
1388 	}
1389 
1390 	if (ddi_in_panic())
1391 		(void) vdc_drain_response(vdc, NULL);
1392 
1393 	DMSG(vdc, 0, "[%d] End\n", instance);
1394 
1395 	return (0);
1396 }
1397 
1398 /* -------------------------------------------------------------------------- */
1399 
1400 /*
1401  * Disk access routines
1402  *
1403  */
1404 
1405 /*
1406  * vdc_strategy()
1407  *
1408  * Return Value:
1409  *	0:	As per strategy(9E), the strategy() function must return 0
1410  *		[ bioerror(9f) sets b_flags to the proper error code ]
1411  */
1412 static int
1413 vdc_strategy(struct buf *buf)
1414 {
1415 	int	rv = -1;
1416 	vdc_t	*vdc = NULL;
1417 	int	instance = VDCUNIT(buf->b_edev);
1418 	int	op = (buf->b_flags & B_READ) ? VD_OP_BREAD : VD_OP_BWRITE;
1419 	int	slice;
1420 
1421 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1422 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1423 		bioerror(buf, ENXIO);
1424 		biodone(buf);
1425 		return (0);
1426 	}
1427 
1428 	DMSG(vdc, 2, "[%d] %s %ld bytes at block %llx : b_addr=0x%p\n",
1429 	    instance, (buf->b_flags & B_READ) ? "Read" : "Write",
1430 	    buf->b_bcount, buf->b_lblkno, (void *)buf->b_un.b_addr);
1431 
1432 	bp_mapin(buf);
1433 
1434 	if ((long)buf->b_private == VD_SLICE_NONE) {
1435 		/* I/O using an absolute disk offset */
1436 		slice = VD_SLICE_NONE;
1437 	} else {
1438 		slice = VDCPART(buf->b_edev);
1439 	}
1440 
1441 	rv = vdc_send_request(vdc, op, (caddr_t)buf->b_un.b_addr,
1442 	    buf->b_bcount, slice, buf->b_lblkno,
1443 	    CB_STRATEGY, buf, (op == VD_OP_BREAD) ? VIO_read_dir :
1444 	    VIO_write_dir);
1445 
1446 	/*
1447 	 * If the request was successfully sent, the strategy call returns and
1448 	 * the ACK handler calls the bioxxx functions when the vDisk server is
1449 	 * done otherwise we handle the error here.
1450 	 */
1451 	if (rv) {
1452 		DMSG(vdc, 0, "Failed to read/write (err=%d)\n", rv);
1453 		bioerror(buf, rv);
1454 		biodone(buf);
1455 	} else if (ddi_in_panic()) {
1456 		(void) vdc_drain_response(vdc, buf);
1457 	}
1458 
1459 	return (0);
1460 }
1461 
1462 /*
1463  * Function:
1464  *	vdc_min
1465  *
1466  * Description:
1467  *	Routine to limit the size of a data transfer. Used in
1468  *	conjunction with physio(9F).
1469  *
1470  * Arguments:
1471  *	bp - pointer to the indicated buf(9S) struct.
1472  *
1473  */
1474 static void
1475 vdc_min(struct buf *bufp)
1476 {
1477 	vdc_t	*vdc = NULL;
1478 	int	instance = VDCUNIT(bufp->b_edev);
1479 
1480 	vdc = ddi_get_soft_state(vdc_state, instance);
1481 	VERIFY(vdc != NULL);
1482 
1483 	if (bufp->b_bcount > (vdc->max_xfer_sz * vdc->block_size)) {
1484 		bufp->b_bcount = vdc->max_xfer_sz * vdc->block_size;
1485 	}
1486 }
1487 
1488 static int
1489 vdc_read(dev_t dev, struct uio *uio, cred_t *cred)
1490 {
1491 	_NOTE(ARGUNUSED(cred))
1492 
1493 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1494 	return (physio(vdc_strategy, NULL, dev, B_READ, vdc_min, uio));
1495 }
1496 
1497 static int
1498 vdc_write(dev_t dev, struct uio *uio, cred_t *cred)
1499 {
1500 	_NOTE(ARGUNUSED(cred))
1501 
1502 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1503 	return (physio(vdc_strategy, NULL, dev, B_WRITE, vdc_min, uio));
1504 }
1505 
1506 static int
1507 vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred)
1508 {
1509 	_NOTE(ARGUNUSED(cred))
1510 
1511 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1512 	return (aphysio(vdc_strategy, anocancel, dev, B_READ, vdc_min, aio));
1513 }
1514 
1515 static int
1516 vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred)
1517 {
1518 	_NOTE(ARGUNUSED(cred))
1519 
1520 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1521 	return (aphysio(vdc_strategy, anocancel, dev, B_WRITE, vdc_min, aio));
1522 }
1523 
1524 
1525 /* -------------------------------------------------------------------------- */
1526 
1527 /*
1528  * Handshake support
1529  */
1530 
1531 
1532 /*
1533  * Function:
1534  *	vdc_init_ver_negotiation()
1535  *
1536  * Description:
1537  *
1538  * Arguments:
1539  *	vdc	- soft state pointer for this instance of the device driver.
1540  *
1541  * Return Code:
1542  *	0	- Success
1543  */
1544 static int
1545 vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver)
1546 {
1547 	vio_ver_msg_t	pkt;
1548 	size_t		msglen = sizeof (pkt);
1549 	int		status = -1;
1550 
1551 	ASSERT(vdc != NULL);
1552 	ASSERT(mutex_owned(&vdc->lock));
1553 
1554 	DMSG(vdc, 0, "[%d] Entered.\n", vdc->instance);
1555 
1556 	/*
1557 	 * set the Session ID to a unique value
1558 	 * (the lower 32 bits of the clock tick)
1559 	 */
1560 	vdc->session_id = ((uint32_t)gettick() & 0xffffffff);
1561 	DMSG(vdc, 0, "[%d] Set SID to 0x%lx\n", vdc->instance, vdc->session_id);
1562 
1563 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1564 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1565 	pkt.tag.vio_subtype_env = VIO_VER_INFO;
1566 	pkt.tag.vio_sid = vdc->session_id;
1567 	pkt.dev_class = VDEV_DISK;
1568 	pkt.ver_major = ver.major;
1569 	pkt.ver_minor = ver.minor;
1570 
1571 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1572 	DMSG(vdc, 0, "[%d] Ver info sent (status = %d)\n",
1573 	    vdc->instance, status);
1574 	if ((status != 0) || (msglen != sizeof (vio_ver_msg_t))) {
1575 		DMSG(vdc, 0, "[%d] Failed to send Ver negotiation info: "
1576 		    "id(%lx) rv(%d) size(%ld)", vdc->instance,
1577 		    vdc->curr_server->ldc_handle, status, msglen);
1578 		if (msglen != sizeof (vio_ver_msg_t))
1579 			status = ENOMSG;
1580 	}
1581 
1582 	return (status);
1583 }
1584 
1585 /*
1586  * Function:
1587  *	vdc_ver_negotiation()
1588  *
1589  * Description:
1590  *
1591  * Arguments:
1592  *	vdcp	- soft state pointer for this instance of the device driver.
1593  *
1594  * Return Code:
1595  *	0	- Success
1596  */
1597 static int
1598 vdc_ver_negotiation(vdc_t *vdcp)
1599 {
1600 	vio_msg_t vio_msg;
1601 	int status;
1602 
1603 	if (status = vdc_init_ver_negotiation(vdcp, vdc_version[0]))
1604 		return (status);
1605 
1606 	/* release lock and wait for response */
1607 	mutex_exit(&vdcp->lock);
1608 	status = vdc_wait_for_response(vdcp, &vio_msg);
1609 	mutex_enter(&vdcp->lock);
1610 	if (status) {
1611 		DMSG(vdcp, 0,
1612 		    "[%d] Failed waiting for Ver negotiation response, rv(%d)",
1613 		    vdcp->instance, status);
1614 		return (status);
1615 	}
1616 
1617 	/* check type and sub_type ... */
1618 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1619 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1620 		DMSG(vdcp, 0, "[%d] Invalid ver negotiation response\n",
1621 		    vdcp->instance);
1622 		return (EPROTO);
1623 	}
1624 
1625 	return (vdc_handle_ver_msg(vdcp, (vio_ver_msg_t *)&vio_msg));
1626 }
1627 
1628 /*
1629  * Function:
1630  *	vdc_init_attr_negotiation()
1631  *
1632  * Description:
1633  *
1634  * Arguments:
1635  *	vdc	- soft state pointer for this instance of the device driver.
1636  *
1637  * Return Code:
1638  *	0	- Success
1639  */
1640 static int
1641 vdc_init_attr_negotiation(vdc_t *vdc)
1642 {
1643 	vd_attr_msg_t	pkt;
1644 	size_t		msglen = sizeof (pkt);
1645 	int		status;
1646 
1647 	ASSERT(vdc != NULL);
1648 	ASSERT(mutex_owned(&vdc->lock));
1649 
1650 	DMSG(vdc, 0, "[%d] entered\n", vdc->instance);
1651 
1652 	/* fill in tag */
1653 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1654 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1655 	pkt.tag.vio_subtype_env = VIO_ATTR_INFO;
1656 	pkt.tag.vio_sid = vdc->session_id;
1657 	/* fill in payload */
1658 	pkt.max_xfer_sz = vdc->max_xfer_sz;
1659 	pkt.vdisk_block_size = vdc->block_size;
1660 	pkt.xfer_mode = VIO_DRING_MODE_V1_0;
1661 	pkt.operations = 0;	/* server will set bits of valid operations */
1662 	pkt.vdisk_type = 0;	/* server will set to valid device type */
1663 	pkt.vdisk_media = 0;	/* server will set to valid media type */
1664 	pkt.vdisk_size = 0;	/* server will set to valid size */
1665 
1666 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1667 	DMSG(vdc, 0, "Attr info sent (status = %d)\n", status);
1668 
1669 	if ((status != 0) || (msglen != sizeof (vd_attr_msg_t))) {
1670 		DMSG(vdc, 0, "[%d] Failed to send Attr negotiation info: "
1671 		    "id(%lx) rv(%d) size(%ld)", vdc->instance,
1672 		    vdc->curr_server->ldc_handle, status, msglen);
1673 		if (msglen != sizeof (vd_attr_msg_t))
1674 			status = ENOMSG;
1675 	}
1676 
1677 	return (status);
1678 }
1679 
1680 /*
1681  * Function:
1682  *	vdc_attr_negotiation()
1683  *
1684  * Description:
1685  *
1686  * Arguments:
1687  *	vdc	- soft state pointer for this instance of the device driver.
1688  *
1689  * Return Code:
1690  *	0	- Success
1691  */
1692 static int
1693 vdc_attr_negotiation(vdc_t *vdcp)
1694 {
1695 	int status;
1696 	vio_msg_t vio_msg;
1697 
1698 	if (status = vdc_init_attr_negotiation(vdcp))
1699 		return (status);
1700 
1701 	/* release lock and wait for response */
1702 	mutex_exit(&vdcp->lock);
1703 	status = vdc_wait_for_response(vdcp, &vio_msg);
1704 	mutex_enter(&vdcp->lock);
1705 	if (status) {
1706 		DMSG(vdcp, 0,
1707 		    "[%d] Failed waiting for Attr negotiation response, rv(%d)",
1708 		    vdcp->instance, status);
1709 		return (status);
1710 	}
1711 
1712 	/* check type and sub_type ... */
1713 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1714 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1715 		DMSG(vdcp, 0, "[%d] Invalid attr negotiation response\n",
1716 		    vdcp->instance);
1717 		return (EPROTO);
1718 	}
1719 
1720 	return (vdc_handle_attr_msg(vdcp, (vd_attr_msg_t *)&vio_msg));
1721 }
1722 
1723 
1724 /*
1725  * Function:
1726  *	vdc_init_dring_negotiate()
1727  *
1728  * Description:
1729  *
1730  * Arguments:
1731  *	vdc	- soft state pointer for this instance of the device driver.
1732  *
1733  * Return Code:
1734  *	0	- Success
1735  */
1736 static int
1737 vdc_init_dring_negotiate(vdc_t *vdc)
1738 {
1739 	vio_dring_reg_msg_t	pkt;
1740 	size_t			msglen = sizeof (pkt);
1741 	int			status = -1;
1742 	int			retry;
1743 	int			nretries = 10;
1744 
1745 	ASSERT(vdc != NULL);
1746 	ASSERT(mutex_owned(&vdc->lock));
1747 
1748 	for (retry = 0; retry < nretries; retry++) {
1749 		status = vdc_init_descriptor_ring(vdc);
1750 		if (status != EAGAIN)
1751 			break;
1752 		drv_usecwait(vdc_min_timeout_ldc);
1753 	}
1754 
1755 	if (status != 0) {
1756 		DMSG(vdc, 0, "[%d] Failed to init DRing (status = %d)\n",
1757 		    vdc->instance, status);
1758 		return (status);
1759 	}
1760 
1761 	DMSG(vdc, 0, "[%d] Init of descriptor ring completed (status = %d)\n",
1762 	    vdc->instance, status);
1763 
1764 	/* fill in tag */
1765 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1766 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1767 	pkt.tag.vio_subtype_env = VIO_DRING_REG;
1768 	pkt.tag.vio_sid = vdc->session_id;
1769 	/* fill in payload */
1770 	pkt.dring_ident = 0;
1771 	pkt.num_descriptors = vdc->dring_len;
1772 	pkt.descriptor_size = vdc->dring_entry_size;
1773 	pkt.options = (VIO_TX_DRING | VIO_RX_DRING);
1774 	pkt.ncookies = vdc->dring_cookie_count;
1775 	pkt.cookie[0] = vdc->dring_cookie[0];	/* for now just one cookie */
1776 
1777 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1778 	if (status != 0) {
1779 		DMSG(vdc, 0, "[%d] Failed to register DRing (err = %d)",
1780 		    vdc->instance, status);
1781 	}
1782 
1783 	return (status);
1784 }
1785 
1786 
1787 /*
1788  * Function:
1789  *	vdc_dring_negotiation()
1790  *
1791  * Description:
1792  *
1793  * Arguments:
1794  *	vdc	- soft state pointer for this instance of the device driver.
1795  *
1796  * Return Code:
1797  *	0	- Success
1798  */
1799 static int
1800 vdc_dring_negotiation(vdc_t *vdcp)
1801 {
1802 	int status;
1803 	vio_msg_t vio_msg;
1804 
1805 	if (status = vdc_init_dring_negotiate(vdcp))
1806 		return (status);
1807 
1808 	/* release lock and wait for response */
1809 	mutex_exit(&vdcp->lock);
1810 	status = vdc_wait_for_response(vdcp, &vio_msg);
1811 	mutex_enter(&vdcp->lock);
1812 	if (status) {
1813 		DMSG(vdcp, 0,
1814 		    "[%d] Failed waiting for Dring negotiation response,"
1815 		    " rv(%d)", vdcp->instance, status);
1816 		return (status);
1817 	}
1818 
1819 	/* check type and sub_type ... */
1820 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1821 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1822 		DMSG(vdcp, 0, "[%d] Invalid Dring negotiation response\n",
1823 		    vdcp->instance);
1824 		return (EPROTO);
1825 	}
1826 
1827 	return (vdc_handle_dring_reg_msg(vdcp,
1828 	    (vio_dring_reg_msg_t *)&vio_msg));
1829 }
1830 
1831 
1832 /*
1833  * Function:
1834  *	vdc_send_rdx()
1835  *
1836  * Description:
1837  *
1838  * Arguments:
1839  *	vdc	- soft state pointer for this instance of the device driver.
1840  *
1841  * Return Code:
1842  *	0	- Success
1843  */
1844 static int
1845 vdc_send_rdx(vdc_t *vdcp)
1846 {
1847 	vio_msg_t	msg;
1848 	size_t		msglen = sizeof (vio_msg_t);
1849 	int		status;
1850 
1851 	/*
1852 	 * Send an RDX message to vds to indicate we are ready
1853 	 * to send data
1854 	 */
1855 	msg.tag.vio_msgtype = VIO_TYPE_CTRL;
1856 	msg.tag.vio_subtype = VIO_SUBTYPE_INFO;
1857 	msg.tag.vio_subtype_env = VIO_RDX;
1858 	msg.tag.vio_sid = vdcp->session_id;
1859 	status = vdc_send(vdcp, (caddr_t)&msg, &msglen);
1860 	if (status != 0) {
1861 		DMSG(vdcp, 0, "[%d] Failed to send RDX message (%d)",
1862 		    vdcp->instance, status);
1863 	}
1864 
1865 	return (status);
1866 }
1867 
1868 /*
1869  * Function:
1870  *	vdc_handle_rdx()
1871  *
1872  * Description:
1873  *
1874  * Arguments:
1875  *	vdc	- soft state pointer for this instance of the device driver.
1876  *	msgp	- received msg
1877  *
1878  * Return Code:
1879  *	0	- Success
1880  */
1881 static int
1882 vdc_handle_rdx(vdc_t *vdcp, vio_rdx_msg_t *msgp)
1883 {
1884 	_NOTE(ARGUNUSED(vdcp))
1885 	_NOTE(ARGUNUSED(msgp))
1886 
1887 	ASSERT(msgp->tag.vio_msgtype == VIO_TYPE_CTRL);
1888 	ASSERT(msgp->tag.vio_subtype == VIO_SUBTYPE_ACK);
1889 	ASSERT(msgp->tag.vio_subtype_env == VIO_RDX);
1890 
1891 	DMSG(vdcp, 1, "[%d] Got an RDX msg", vdcp->instance);
1892 
1893 	return (0);
1894 }
1895 
1896 /*
1897  * Function:
1898  *	vdc_rdx_exchange()
1899  *
1900  * Description:
1901  *
1902  * Arguments:
1903  *	vdc	- soft state pointer for this instance of the device driver.
1904  *
1905  * Return Code:
1906  *	0	- Success
1907  */
1908 static int
1909 vdc_rdx_exchange(vdc_t *vdcp)
1910 {
1911 	int status;
1912 	vio_msg_t vio_msg;
1913 
1914 	if (status = vdc_send_rdx(vdcp))
1915 		return (status);
1916 
1917 	/* release lock and wait for response */
1918 	mutex_exit(&vdcp->lock);
1919 	status = vdc_wait_for_response(vdcp, &vio_msg);
1920 	mutex_enter(&vdcp->lock);
1921 	if (status) {
1922 		DMSG(vdcp, 0, "[%d] Failed waiting for RDX response, rv(%d)",
1923 		    vdcp->instance, status);
1924 		return (status);
1925 	}
1926 
1927 	/* check type and sub_type ... */
1928 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1929 	    vio_msg.tag.vio_subtype != VIO_SUBTYPE_ACK) {
1930 		DMSG(vdcp, 0, "[%d] Invalid RDX response\n", vdcp->instance);
1931 		return (EPROTO);
1932 	}
1933 
1934 	return (vdc_handle_rdx(vdcp, (vio_rdx_msg_t *)&vio_msg));
1935 }
1936 
1937 
1938 /* -------------------------------------------------------------------------- */
1939 
1940 /*
1941  * LDC helper routines
1942  */
1943 
1944 static int
1945 vdc_recv(vdc_t *vdc, vio_msg_t *msgp, size_t *nbytesp)
1946 {
1947 	int		status;
1948 	boolean_t	q_has_pkts = B_FALSE;
1949 	uint64_t	delay_time;
1950 	size_t		len;
1951 
1952 	mutex_enter(&vdc->read_lock);
1953 
1954 	if (vdc->read_state == VDC_READ_IDLE)
1955 		vdc->read_state = VDC_READ_WAITING;
1956 
1957 	while (vdc->read_state != VDC_READ_PENDING) {
1958 
1959 		/* detect if the connection has been reset */
1960 		if (vdc->read_state == VDC_READ_RESET) {
1961 			status = ECONNRESET;
1962 			goto done;
1963 		}
1964 
1965 		cv_wait(&vdc->read_cv, &vdc->read_lock);
1966 	}
1967 
1968 	/*
1969 	 * Until we get a blocking ldc read we have to retry
1970 	 * until the entire LDC message has arrived before
1971 	 * ldc_read() will succeed. Note we also bail out if
1972 	 * the channel is reset or goes away.
1973 	 */
1974 	delay_time = vdc_ldc_read_init_delay;
1975 loop:
1976 	len = *nbytesp;
1977 	status = ldc_read(vdc->curr_server->ldc_handle, (caddr_t)msgp, &len);
1978 	switch (status) {
1979 	case EAGAIN:
1980 		delay_time *= 2;
1981 		if (delay_time >= vdc_ldc_read_max_delay)
1982 			delay_time = vdc_ldc_read_max_delay;
1983 		delay(delay_time);
1984 		goto loop;
1985 
1986 	case 0:
1987 		if (len == 0) {
1988 			DMSG(vdc, 1, "[%d] ldc_read returned 0 bytes with "
1989 			    "no error!\n", vdc->instance);
1990 			goto loop;
1991 		}
1992 
1993 		*nbytesp = len;
1994 
1995 		/*
1996 		 * If there are pending messages, leave the
1997 		 * read state as pending. Otherwise, set the state
1998 		 * back to idle.
1999 		 */
2000 		status = ldc_chkq(vdc->curr_server->ldc_handle, &q_has_pkts);
2001 		if (status == 0 && !q_has_pkts)
2002 			vdc->read_state = VDC_READ_IDLE;
2003 
2004 		break;
2005 	default:
2006 		DMSG(vdc, 0, "ldc_read returned %d\n", status);
2007 		break;
2008 	}
2009 
2010 done:
2011 	mutex_exit(&vdc->read_lock);
2012 
2013 	return (status);
2014 }
2015 
2016 
2017 
2018 #ifdef DEBUG
2019 void
2020 vdc_decode_tag(vdc_t *vdcp, vio_msg_t *msg)
2021 {
2022 	char *ms, *ss, *ses;
2023 	switch (msg->tag.vio_msgtype) {
2024 #define	Q(_s)	case _s : ms = #_s; break;
2025 	Q(VIO_TYPE_CTRL)
2026 	Q(VIO_TYPE_DATA)
2027 	Q(VIO_TYPE_ERR)
2028 #undef Q
2029 	default: ms = "unknown"; break;
2030 	}
2031 
2032 	switch (msg->tag.vio_subtype) {
2033 #define	Q(_s)	case _s : ss = #_s; break;
2034 	Q(VIO_SUBTYPE_INFO)
2035 	Q(VIO_SUBTYPE_ACK)
2036 	Q(VIO_SUBTYPE_NACK)
2037 #undef Q
2038 	default: ss = "unknown"; break;
2039 	}
2040 
2041 	switch (msg->tag.vio_subtype_env) {
2042 #define	Q(_s)	case _s : ses = #_s; break;
2043 	Q(VIO_VER_INFO)
2044 	Q(VIO_ATTR_INFO)
2045 	Q(VIO_DRING_REG)
2046 	Q(VIO_DRING_UNREG)
2047 	Q(VIO_RDX)
2048 	Q(VIO_PKT_DATA)
2049 	Q(VIO_DESC_DATA)
2050 	Q(VIO_DRING_DATA)
2051 #undef Q
2052 	default: ses = "unknown"; break;
2053 	}
2054 
2055 	DMSG(vdcp, 3, "(%x/%x/%x) message : (%s/%s/%s)\n",
2056 	    msg->tag.vio_msgtype, msg->tag.vio_subtype,
2057 	    msg->tag.vio_subtype_env, ms, ss, ses);
2058 }
2059 #endif
2060 
2061 /*
2062  * Function:
2063  *	vdc_send()
2064  *
2065  * Description:
2066  *	The function encapsulates the call to write a message using LDC.
2067  *	If LDC indicates that the call failed due to the queue being full,
2068  *	we retry the ldc_write(), otherwise we return the error returned by LDC.
2069  *
2070  * Arguments:
2071  *	ldc_handle	- LDC handle for the channel this instance of vdc uses
2072  *	pkt		- address of LDC message to be sent
2073  *	msglen		- the size of the message being sent. When the function
2074  *			  returns, this contains the number of bytes written.
2075  *
2076  * Return Code:
2077  *	0		- Success.
2078  *	EINVAL		- pkt or msglen were NULL
2079  *	ECONNRESET	- The connection was not up.
2080  *	EWOULDBLOCK	- LDC queue is full
2081  *	xxx		- other error codes returned by ldc_write
2082  */
2083 static int
2084 vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen)
2085 {
2086 	size_t	size = 0;
2087 	int	status = 0;
2088 	clock_t delay_ticks;
2089 
2090 	ASSERT(vdc != NULL);
2091 	ASSERT(mutex_owned(&vdc->lock));
2092 	ASSERT(msglen != NULL);
2093 	ASSERT(*msglen != 0);
2094 
2095 #ifdef DEBUG
2096 	vdc_decode_tag(vdc, (vio_msg_t *)(uintptr_t)pkt);
2097 #endif
2098 	/*
2099 	 * Wait indefinitely to send if channel
2100 	 * is busy, but bail out if we succeed or
2101 	 * if the channel closes or is reset.
2102 	 */
2103 	delay_ticks = vdc_hz_min_ldc_delay;
2104 	do {
2105 		size = *msglen;
2106 		status = ldc_write(vdc->curr_server->ldc_handle, pkt, &size);
2107 		if (status == EWOULDBLOCK) {
2108 			delay(delay_ticks);
2109 			/* geometric backoff */
2110 			delay_ticks *= 2;
2111 			if (delay_ticks > vdc_hz_max_ldc_delay)
2112 				delay_ticks = vdc_hz_max_ldc_delay;
2113 		}
2114 	} while (status == EWOULDBLOCK);
2115 
2116 	/* if LDC had serious issues --- reset vdc state */
2117 	if (status == EIO || status == ECONNRESET) {
2118 		/* LDC had serious issues --- reset vdc state */
2119 		mutex_enter(&vdc->read_lock);
2120 		if ((vdc->read_state == VDC_READ_WAITING) ||
2121 		    (vdc->read_state == VDC_READ_RESET))
2122 			cv_signal(&vdc->read_cv);
2123 		vdc->read_state = VDC_READ_RESET;
2124 		mutex_exit(&vdc->read_lock);
2125 
2126 		/* wake up any waiters in the reset thread */
2127 		if (vdc->state == VDC_STATE_INIT_WAITING) {
2128 			DMSG(vdc, 0, "[%d] write reset - "
2129 			    "vdc is resetting ..\n", vdc->instance);
2130 			vdc->state = VDC_STATE_RESETTING;
2131 			cv_signal(&vdc->initwait_cv);
2132 		}
2133 
2134 		return (ECONNRESET);
2135 	}
2136 
2137 	/* return the last size written */
2138 	*msglen = size;
2139 
2140 	return (status);
2141 }
2142 
2143 /*
2144  * Function:
2145  *	vdc_get_md_node
2146  *
2147  * Description:
2148  *	Get the MD, the device node for the given disk instance. The
2149  *	caller is responsible for cleaning up the reference to the
2150  *	returned MD (mdpp) by calling md_fini_handle().
2151  *
2152  * Arguments:
2153  *	dip	- dev info pointer for this instance of the device driver.
2154  *	mdpp	- the returned MD.
2155  *	vd_nodep - the returned device node.
2156  *
2157  * Return Code:
2158  *	0	- Success.
2159  *	ENOENT	- Expected node or property did not exist.
2160  *	ENXIO	- Unexpected error communicating with MD framework
2161  */
2162 static int
2163 vdc_get_md_node(dev_info_t *dip, md_t **mdpp, mde_cookie_t *vd_nodep)
2164 {
2165 	int		status = ENOENT;
2166 	char		*node_name = NULL;
2167 	md_t		*mdp = NULL;
2168 	int		num_nodes;
2169 	int		num_vdevs;
2170 	mde_cookie_t	rootnode;
2171 	mde_cookie_t	*listp = NULL;
2172 	boolean_t	found_inst = B_FALSE;
2173 	int		listsz;
2174 	int		idx;
2175 	uint64_t	md_inst;
2176 	int		obp_inst;
2177 	int		instance = ddi_get_instance(dip);
2178 
2179 	/*
2180 	 * Get the OBP instance number for comparison with the MD instance
2181 	 *
2182 	 * The "cfg-handle" property of a vdc node in an MD contains the MD's
2183 	 * notion of "instance", or unique identifier, for that node; OBP
2184 	 * stores the value of the "cfg-handle" MD property as the value of
2185 	 * the "reg" property on the node in the device tree it builds from
2186 	 * the MD and passes to Solaris.  Thus, we look up the devinfo node's
2187 	 * "reg" property value to uniquely identify this device instance.
2188 	 * If the "reg" property cannot be found, the device tree state is
2189 	 * presumably so broken that there is no point in continuing.
2190 	 */
2191 	if (!ddi_prop_exists(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS, OBP_REG)) {
2192 		cmn_err(CE_WARN, "'%s' property does not exist", OBP_REG);
2193 		return (ENOENT);
2194 	}
2195 	obp_inst = ddi_prop_get_int(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
2196 	    OBP_REG, -1);
2197 	DMSGX(1, "[%d] OBP inst=%d\n", instance, obp_inst);
2198 
2199 	/*
2200 	 * We now walk the MD nodes to find the node for this vdisk.
2201 	 */
2202 	if ((mdp = md_get_handle()) == NULL) {
2203 		cmn_err(CE_WARN, "unable to init machine description");
2204 		return (ENXIO);
2205 	}
2206 
2207 	num_nodes = md_node_count(mdp);
2208 	ASSERT(num_nodes > 0);
2209 
2210 	listsz = num_nodes * sizeof (mde_cookie_t);
2211 
2212 	/* allocate memory for nodes */
2213 	listp = kmem_zalloc(listsz, KM_SLEEP);
2214 
2215 	rootnode = md_root_node(mdp);
2216 	ASSERT(rootnode != MDE_INVAL_ELEM_COOKIE);
2217 
2218 	/*
2219 	 * Search for all the virtual devices, we will then check to see which
2220 	 * ones are disk nodes.
2221 	 */
2222 	num_vdevs = md_scan_dag(mdp, rootnode,
2223 	    md_find_name(mdp, VDC_MD_VDEV_NAME),
2224 	    md_find_name(mdp, "fwd"), listp);
2225 
2226 	if (num_vdevs <= 0) {
2227 		cmn_err(CE_NOTE, "No '%s' node found", VDC_MD_VDEV_NAME);
2228 		status = ENOENT;
2229 		goto done;
2230 	}
2231 
2232 	DMSGX(1, "[%d] num_vdevs=%d\n", instance, num_vdevs);
2233 	for (idx = 0; idx < num_vdevs; idx++) {
2234 		status = md_get_prop_str(mdp, listp[idx], "name", &node_name);
2235 		if ((status != 0) || (node_name == NULL)) {
2236 			cmn_err(CE_NOTE, "Unable to get name of node type '%s'"
2237 			    ": err %d", VDC_MD_VDEV_NAME, status);
2238 			continue;
2239 		}
2240 
2241 		DMSGX(1, "[%d] Found node '%s'\n", instance, node_name);
2242 		if (strcmp(VDC_MD_DISK_NAME, node_name) == 0) {
2243 			status = md_get_prop_val(mdp, listp[idx],
2244 			    VDC_MD_CFG_HDL, &md_inst);
2245 			DMSGX(1, "[%d] vdc inst in MD=%lx\n",
2246 			    instance, md_inst);
2247 			if ((status == 0) && (md_inst == obp_inst)) {
2248 				found_inst = B_TRUE;
2249 				break;
2250 			}
2251 		}
2252 	}
2253 
2254 	if (!found_inst) {
2255 		DMSGX(0, "Unable to find correct '%s' node", VDC_MD_DISK_NAME);
2256 		status = ENOENT;
2257 		goto done;
2258 	}
2259 	DMSGX(0, "[%d] MD inst=%lx\n", instance, md_inst);
2260 
2261 	*vd_nodep = listp[idx];
2262 	*mdpp = mdp;
2263 done:
2264 	kmem_free(listp, listsz);
2265 	return (status);
2266 }
2267 
2268 /*
2269  * Function:
2270  *	vdc_init_ports
2271  *
2272  * Description:
2273  *	Initialize all the ports for this vdisk instance.
2274  *
2275  * Arguments:
2276  *	vdc	- soft state pointer for this instance of the device driver.
2277  *	mdp	- md pointer
2278  *	vd_nodep - device md node.
2279  *
2280  * Return Code:
2281  *	0	- Success.
2282  *	ENOENT	- Expected node or property did not exist.
2283  */
2284 static int
2285 vdc_init_ports(vdc_t *vdc, md_t *mdp, mde_cookie_t vd_nodep)
2286 {
2287 	int		status = 0;
2288 	int		idx;
2289 	int		num_nodes;
2290 	int		num_vports;
2291 	int		num_chans;
2292 	int		listsz;
2293 	mde_cookie_t	vd_port;
2294 	mde_cookie_t	*chanp = NULL;
2295 	mde_cookie_t	*portp = NULL;
2296 	vdc_server_t	*srvr;
2297 	vdc_server_t	*prev_srvr = NULL;
2298 
2299 	/*
2300 	 * We now walk the MD nodes to find the port nodes for this vdisk.
2301 	 */
2302 	num_nodes = md_node_count(mdp);
2303 	ASSERT(num_nodes > 0);
2304 
2305 	listsz = num_nodes * sizeof (mde_cookie_t);
2306 
2307 	/* allocate memory for nodes */
2308 	portp = kmem_zalloc(listsz, KM_SLEEP);
2309 	chanp = kmem_zalloc(listsz, KM_SLEEP);
2310 
2311 	num_vports = md_scan_dag(mdp, vd_nodep,
2312 	    md_find_name(mdp, VDC_MD_PORT_NAME),
2313 	    md_find_name(mdp, "fwd"), portp);
2314 	if (num_vports == 0) {
2315 		DMSGX(0, "Found no '%s' node for '%s' port\n",
2316 		    VDC_MD_PORT_NAME, VDC_MD_VDEV_NAME);
2317 		status = ENOENT;
2318 		goto done;
2319 	}
2320 
2321 	DMSGX(1, "Found %d '%s' node(s) for '%s' port\n",
2322 	    num_vports, VDC_MD_PORT_NAME, VDC_MD_VDEV_NAME);
2323 
2324 	vdc->num_servers = 0;
2325 	for (idx = 0; idx < num_vports; idx++) {
2326 
2327 		/* initialize this port */
2328 		vd_port = portp[idx];
2329 		srvr = kmem_zalloc(sizeof (vdc_server_t), KM_SLEEP);
2330 		srvr->vdcp = vdc;
2331 
2332 		/* get port id */
2333 		if (md_get_prop_val(mdp, vd_port, VDC_MD_ID, &srvr->id) != 0) {
2334 			cmn_err(CE_NOTE, "vDisk port '%s' property not found",
2335 			    VDC_MD_ID);
2336 			kmem_free(srvr, sizeof (vdc_server_t));
2337 			continue;
2338 		}
2339 
2340 		/* set the connection timeout */
2341 		if (md_get_prop_val(mdp, vd_port, VDC_MD_TIMEOUT,
2342 		    &srvr->ctimeout) != 0) {
2343 			srvr->ctimeout = 0;
2344 		}
2345 
2346 		/* get the ldc id */
2347 		num_chans = md_scan_dag(mdp, vd_port,
2348 		    md_find_name(mdp, VDC_MD_CHAN_NAME),
2349 		    md_find_name(mdp, "fwd"), chanp);
2350 
2351 		/* expecting at least one channel */
2352 		if (num_chans <= 0) {
2353 			cmn_err(CE_NOTE, "No '%s' node for '%s' port",
2354 			    VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME);
2355 			kmem_free(srvr, sizeof (vdc_server_t));
2356 			continue;
2357 		} else if (num_chans != 1) {
2358 			DMSGX(0, "Expected 1 '%s' node for '%s' port, "
2359 			    "found %d\n", VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME,
2360 			    num_chans);
2361 		}
2362 
2363 		/*
2364 		 * We use the first channel found (index 0), irrespective of how
2365 		 * many are there in total.
2366 		 */
2367 		if (md_get_prop_val(mdp, chanp[0], VDC_MD_ID,
2368 		    &srvr->ldc_id) != 0) {
2369 			cmn_err(CE_NOTE, "Channel '%s' property not found",
2370 			    VDC_MD_ID);
2371 			kmem_free(srvr, sizeof (vdc_server_t));
2372 			continue;
2373 		}
2374 
2375 		/*
2376 		 * now initialise LDC channel which will be used to
2377 		 * communicate with this server
2378 		 */
2379 		if (vdc_do_ldc_init(vdc, srvr) != 0) {
2380 			kmem_free(srvr, sizeof (vdc_server_t));
2381 			continue;
2382 		}
2383 
2384 		/* add server to list */
2385 		if (prev_srvr)
2386 			prev_srvr->next = srvr;
2387 		else
2388 			vdc->server_list = srvr;
2389 
2390 		prev_srvr = srvr;
2391 
2392 		/* inc numbers of servers */
2393 		vdc->num_servers++;
2394 	}
2395 
2396 	/*
2397 	 * Adjust the max number of handshake retries to match
2398 	 * the number of vdisk servers.
2399 	 */
2400 	if (vdc_hshake_retries < vdc->num_servers)
2401 		vdc_hshake_retries = vdc->num_servers;
2402 
2403 	/* pick first server as current server */
2404 	if (vdc->server_list != NULL) {
2405 		vdc->curr_server = vdc->server_list;
2406 		status = 0;
2407 	} else {
2408 		status = ENOENT;
2409 	}
2410 
2411 done:
2412 	kmem_free(chanp, listsz);
2413 	kmem_free(portp, listsz);
2414 	return (status);
2415 }
2416 
2417 
2418 /*
2419  * Function:
2420  *	vdc_do_ldc_up
2421  *
2422  * Description:
2423  *	Bring the channel for the current server up.
2424  *
2425  * Arguments:
2426  *	vdc	- soft state pointer for this instance of the device driver.
2427  *
2428  * Return Code:
2429  *	0		- Success.
2430  *	EINVAL		- Driver is detaching / LDC error
2431  *	ECONNREFUSED	- Other end is not listening
2432  */
2433 static int
2434 vdc_do_ldc_up(vdc_t *vdc)
2435 {
2436 	int		status;
2437 	ldc_status_t	ldc_state;
2438 
2439 	ASSERT(MUTEX_HELD(&vdc->lock));
2440 
2441 	DMSG(vdc, 0, "[%d] Bringing up channel %lx\n",
2442 	    vdc->instance, vdc->curr_server->ldc_id);
2443 
2444 	if (vdc->lifecycle == VDC_LC_DETACHING)
2445 		return (EINVAL);
2446 
2447 	if ((status = ldc_up(vdc->curr_server->ldc_handle)) != 0) {
2448 		switch (status) {
2449 		case ECONNREFUSED:	/* listener not ready at other end */
2450 			DMSG(vdc, 0, "[%d] ldc_up(%lx,...) return %d\n",
2451 			    vdc->instance, vdc->curr_server->ldc_id, status);
2452 			status = 0;
2453 			break;
2454 		default:
2455 			DMSG(vdc, 0, "[%d] Failed to bring up LDC: "
2456 			    "channel=%ld, err=%d", vdc->instance,
2457 			    vdc->curr_server->ldc_id, status);
2458 			break;
2459 		}
2460 	}
2461 
2462 	if (ldc_status(vdc->curr_server->ldc_handle, &ldc_state) == 0) {
2463 		vdc->curr_server->ldc_state = ldc_state;
2464 		if (ldc_state == LDC_UP) {
2465 			DMSG(vdc, 0, "[%d] LDC channel already up\n",
2466 			    vdc->instance);
2467 			vdc->seq_num = 1;
2468 			vdc->seq_num_reply = 0;
2469 		}
2470 	}
2471 
2472 	return (status);
2473 }
2474 
2475 /*
2476  * Function:
2477  *	vdc_terminate_ldc()
2478  *
2479  * Description:
2480  *
2481  * Arguments:
2482  *	vdc	- soft state pointer for this instance of the device driver.
2483  *	srvr	- vdc per-server info structure
2484  *
2485  * Return Code:
2486  *	None
2487  */
2488 static void
2489 vdc_terminate_ldc(vdc_t *vdc, vdc_server_t *srvr)
2490 {
2491 	int	instance = ddi_get_instance(vdc->dip);
2492 
2493 	if (srvr->state & VDC_LDC_OPEN) {
2494 		DMSG(vdc, 0, "[%d] ldc_close()\n", instance);
2495 		(void) ldc_close(srvr->ldc_handle);
2496 	}
2497 	if (srvr->state & VDC_LDC_CB) {
2498 		DMSG(vdc, 0, "[%d] ldc_unreg_callback()\n", instance);
2499 		(void) ldc_unreg_callback(srvr->ldc_handle);
2500 	}
2501 	if (srvr->state & VDC_LDC_INIT) {
2502 		DMSG(vdc, 0, "[%d] ldc_fini()\n", instance);
2503 		(void) ldc_fini(srvr->ldc_handle);
2504 		srvr->ldc_handle = NULL;
2505 	}
2506 
2507 	srvr->state &= ~(VDC_LDC_INIT | VDC_LDC_CB | VDC_LDC_OPEN);
2508 }
2509 
2510 /*
2511  * Function:
2512  *	vdc_fini_ports()
2513  *
2514  * Description:
2515  *	Finalize all ports by closing the channel associated with each
2516  *	port and also freeing the server structure.
2517  *
2518  * Arguments:
2519  *	vdc	- soft state pointer for this instance of the device driver.
2520  *
2521  * Return Code:
2522  *	None
2523  */
2524 static void
2525 vdc_fini_ports(vdc_t *vdc)
2526 {
2527 	int		instance = ddi_get_instance(vdc->dip);
2528 	vdc_server_t	*srvr, *prev_srvr;
2529 
2530 	ASSERT(vdc != NULL);
2531 	ASSERT(mutex_owned(&vdc->lock));
2532 
2533 	DMSG(vdc, 0, "[%d] initialized=%x\n", instance, vdc->initialized);
2534 
2535 	srvr = vdc->server_list;
2536 
2537 	while (srvr) {
2538 
2539 		vdc_terminate_ldc(vdc, srvr);
2540 
2541 		/* next server */
2542 		prev_srvr = srvr;
2543 		srvr = srvr->next;
2544 
2545 		/* free server */
2546 		kmem_free(prev_srvr, sizeof (vdc_server_t));
2547 	}
2548 
2549 	vdc->server_list = NULL;
2550 }
2551 
2552 /* -------------------------------------------------------------------------- */
2553 
2554 /*
2555  * Descriptor Ring helper routines
2556  */
2557 
2558 /*
2559  * Function:
2560  *	vdc_init_descriptor_ring()
2561  *
2562  * Description:
2563  *
2564  * Arguments:
2565  *	vdc	- soft state pointer for this instance of the device driver.
2566  *
2567  * Return Code:
2568  *	0	- Success
2569  */
2570 static int
2571 vdc_init_descriptor_ring(vdc_t *vdc)
2572 {
2573 	vd_dring_entry_t	*dep = NULL;	/* DRing Entry pointer */
2574 	int	status = 0;
2575 	int	i;
2576 
2577 	DMSG(vdc, 0, "[%d] initialized=%x\n", vdc->instance, vdc->initialized);
2578 
2579 	ASSERT(vdc != NULL);
2580 	ASSERT(mutex_owned(&vdc->lock));
2581 
2582 	/* ensure we have enough room to store max sized block */
2583 	ASSERT(maxphys <= VD_MAX_BLOCK_SIZE);
2584 
2585 	if ((vdc->initialized & VDC_DRING_INIT) == 0) {
2586 		DMSG(vdc, 0, "[%d] ldc_mem_dring_create\n", vdc->instance);
2587 		/*
2588 		 * Calculate the maximum block size we can transmit using one
2589 		 * Descriptor Ring entry from the attributes returned by the
2590 		 * vDisk server. This is subject to a minimum of 'maxphys'
2591 		 * as we do not have the capability to split requests over
2592 		 * multiple DRing entries.
2593 		 */
2594 		if ((vdc->max_xfer_sz * vdc->block_size) < maxphys) {
2595 			DMSG(vdc, 0, "[%d] using minimum DRing size\n",
2596 			    vdc->instance);
2597 			vdc->dring_max_cookies = maxphys / PAGESIZE;
2598 		} else {
2599 			vdc->dring_max_cookies =
2600 			    (vdc->max_xfer_sz * vdc->block_size) / PAGESIZE;
2601 		}
2602 		vdc->dring_entry_size = (sizeof (vd_dring_entry_t) +
2603 		    (sizeof (ldc_mem_cookie_t) *
2604 		    (vdc->dring_max_cookies - 1)));
2605 		vdc->dring_len = VD_DRING_LEN;
2606 
2607 		status = ldc_mem_dring_create(vdc->dring_len,
2608 		    vdc->dring_entry_size, &vdc->dring_hdl);
2609 		if ((vdc->dring_hdl == NULL) || (status != 0)) {
2610 			DMSG(vdc, 0, "[%d] Descriptor ring creation failed",
2611 			    vdc->instance);
2612 			return (status);
2613 		}
2614 		vdc->initialized |= VDC_DRING_INIT;
2615 	}
2616 
2617 	if ((vdc->initialized & VDC_DRING_BOUND) == 0) {
2618 		DMSG(vdc, 0, "[%d] ldc_mem_dring_bind\n", vdc->instance);
2619 		vdc->dring_cookie =
2620 		    kmem_zalloc(sizeof (ldc_mem_cookie_t), KM_SLEEP);
2621 
2622 		status = ldc_mem_dring_bind(vdc->curr_server->ldc_handle,
2623 		    vdc->dring_hdl,
2624 		    LDC_SHADOW_MAP|LDC_DIRECT_MAP, LDC_MEM_RW,
2625 		    &vdc->dring_cookie[0],
2626 		    &vdc->dring_cookie_count);
2627 		if (status != 0) {
2628 			DMSG(vdc, 0, "[%d] Failed to bind descriptor ring "
2629 			    "(%lx) to channel (%lx) status=%d\n",
2630 			    vdc->instance, vdc->dring_hdl,
2631 			    vdc->curr_server->ldc_handle, status);
2632 			return (status);
2633 		}
2634 		ASSERT(vdc->dring_cookie_count == 1);
2635 		vdc->initialized |= VDC_DRING_BOUND;
2636 	}
2637 
2638 	status = ldc_mem_dring_info(vdc->dring_hdl, &vdc->dring_mem_info);
2639 	if (status != 0) {
2640 		DMSG(vdc, 0,
2641 		    "[%d] Failed to get info for descriptor ring (%lx)\n",
2642 		    vdc->instance, vdc->dring_hdl);
2643 		return (status);
2644 	}
2645 
2646 	if ((vdc->initialized & VDC_DRING_LOCAL) == 0) {
2647 		DMSG(vdc, 0, "[%d] local dring\n", vdc->instance);
2648 
2649 		/* Allocate the local copy of this dring */
2650 		vdc->local_dring =
2651 		    kmem_zalloc(vdc->dring_len * sizeof (vdc_local_desc_t),
2652 		    KM_SLEEP);
2653 		vdc->initialized |= VDC_DRING_LOCAL;
2654 	}
2655 
2656 	/*
2657 	 * Mark all DRing entries as free and initialize the private
2658 	 * descriptor's memory handles. If any entry is initialized,
2659 	 * we need to free it later so we set the bit in 'initialized'
2660 	 * at the start.
2661 	 */
2662 	vdc->initialized |= VDC_DRING_ENTRY;
2663 	for (i = 0; i < vdc->dring_len; i++) {
2664 		dep = VDC_GET_DRING_ENTRY_PTR(vdc, i);
2665 		dep->hdr.dstate = VIO_DESC_FREE;
2666 
2667 		status = ldc_mem_alloc_handle(vdc->curr_server->ldc_handle,
2668 		    &vdc->local_dring[i].desc_mhdl);
2669 		if (status != 0) {
2670 			DMSG(vdc, 0, "![%d] Failed to alloc mem handle for"
2671 			    " descriptor %d", vdc->instance, i);
2672 			return (status);
2673 		}
2674 		vdc->local_dring[i].is_free = B_TRUE;
2675 		vdc->local_dring[i].dep = dep;
2676 	}
2677 
2678 	/* Initialize the starting index */
2679 	vdc->dring_curr_idx = 0;
2680 
2681 	return (status);
2682 }
2683 
2684 /*
2685  * Function:
2686  *	vdc_destroy_descriptor_ring()
2687  *
2688  * Description:
2689  *
2690  * Arguments:
2691  *	vdc	- soft state pointer for this instance of the device driver.
2692  *
2693  * Return Code:
2694  *	None
2695  */
2696 static void
2697 vdc_destroy_descriptor_ring(vdc_t *vdc)
2698 {
2699 	vdc_local_desc_t	*ldep = NULL;	/* Local Dring Entry Pointer */
2700 	ldc_mem_handle_t	mhdl = NULL;
2701 	ldc_mem_info_t		minfo;
2702 	int			status = -1;
2703 	int			i;	/* loop */
2704 
2705 	ASSERT(vdc != NULL);
2706 	ASSERT(mutex_owned(&vdc->lock));
2707 
2708 	DMSG(vdc, 0, "[%d] Entered\n", vdc->instance);
2709 
2710 	if (vdc->initialized & VDC_DRING_ENTRY) {
2711 		DMSG(vdc, 0,
2712 		    "[%d] Removing Local DRing entries\n", vdc->instance);
2713 		for (i = 0; i < vdc->dring_len; i++) {
2714 			ldep = &vdc->local_dring[i];
2715 			mhdl = ldep->desc_mhdl;
2716 
2717 			if (mhdl == NULL)
2718 				continue;
2719 
2720 			if ((status = ldc_mem_info(mhdl, &minfo)) != 0) {
2721 				DMSG(vdc, 0,
2722 				    "ldc_mem_info returned an error: %d\n",
2723 				    status);
2724 
2725 				/*
2726 				 * This must mean that the mem handle
2727 				 * is not valid. Clear it out so that
2728 				 * no one tries to use it.
2729 				 */
2730 				ldep->desc_mhdl = NULL;
2731 				continue;
2732 			}
2733 
2734 			if (minfo.status == LDC_BOUND) {
2735 				(void) ldc_mem_unbind_handle(mhdl);
2736 			}
2737 
2738 			(void) ldc_mem_free_handle(mhdl);
2739 
2740 			ldep->desc_mhdl = NULL;
2741 		}
2742 		vdc->initialized &= ~VDC_DRING_ENTRY;
2743 	}
2744 
2745 	if (vdc->initialized & VDC_DRING_LOCAL) {
2746 		DMSG(vdc, 0, "[%d] Freeing Local DRing\n", vdc->instance);
2747 		kmem_free(vdc->local_dring,
2748 		    vdc->dring_len * sizeof (vdc_local_desc_t));
2749 		vdc->initialized &= ~VDC_DRING_LOCAL;
2750 	}
2751 
2752 	if (vdc->initialized & VDC_DRING_BOUND) {
2753 		DMSG(vdc, 0, "[%d] Unbinding DRing\n", vdc->instance);
2754 		status = ldc_mem_dring_unbind(vdc->dring_hdl);
2755 		if (status == 0) {
2756 			vdc->initialized &= ~VDC_DRING_BOUND;
2757 		} else {
2758 			DMSG(vdc, 0, "[%d] Error %d unbinding DRing %lx",
2759 			    vdc->instance, status, vdc->dring_hdl);
2760 		}
2761 		kmem_free(vdc->dring_cookie, sizeof (ldc_mem_cookie_t));
2762 	}
2763 
2764 	if (vdc->initialized & VDC_DRING_INIT) {
2765 		DMSG(vdc, 0, "[%d] Destroying DRing\n", vdc->instance);
2766 		status = ldc_mem_dring_destroy(vdc->dring_hdl);
2767 		if (status == 0) {
2768 			vdc->dring_hdl = NULL;
2769 			bzero(&vdc->dring_mem_info, sizeof (ldc_mem_info_t));
2770 			vdc->initialized &= ~VDC_DRING_INIT;
2771 		} else {
2772 			DMSG(vdc, 0, "[%d] Error %d destroying DRing (%lx)",
2773 			    vdc->instance, status, vdc->dring_hdl);
2774 		}
2775 	}
2776 }
2777 
2778 /*
2779  * Function:
2780  *	vdc_map_to_shared_dring()
2781  *
2782  * Description:
2783  *	Copy contents of the local descriptor to the shared
2784  *	memory descriptor.
2785  *
2786  * Arguments:
2787  *	vdcp	- soft state pointer for this instance of the device driver.
2788  *	idx	- descriptor ring index
2789  *
2790  * Return Code:
2791  *	None
2792  */
2793 static int
2794 vdc_map_to_shared_dring(vdc_t *vdcp, int idx)
2795 {
2796 	vdc_local_desc_t	*ldep;
2797 	vd_dring_entry_t	*dep;
2798 	int			rv;
2799 
2800 	ldep = &(vdcp->local_dring[idx]);
2801 
2802 	/* for now leave in the old pop_mem_hdl stuff */
2803 	if (ldep->nbytes > 0) {
2804 		rv = vdc_populate_mem_hdl(vdcp, ldep);
2805 		if (rv) {
2806 			DMSG(vdcp, 0, "[%d] Cannot populate mem handle\n",
2807 			    vdcp->instance);
2808 			return (rv);
2809 		}
2810 	}
2811 
2812 	/*
2813 	 * fill in the data details into the DRing
2814 	 */
2815 	dep = ldep->dep;
2816 	ASSERT(dep != NULL);
2817 
2818 	dep->payload.req_id = VDC_GET_NEXT_REQ_ID(vdcp);
2819 	dep->payload.operation = ldep->operation;
2820 	dep->payload.addr = ldep->offset;
2821 	dep->payload.nbytes = ldep->nbytes;
2822 	dep->payload.status = (uint32_t)-1;	/* vds will set valid value */
2823 	dep->payload.slice = ldep->slice;
2824 	dep->hdr.dstate = VIO_DESC_READY;
2825 	dep->hdr.ack = 1;		/* request an ACK for every message */
2826 
2827 	return (0);
2828 }
2829 
2830 /*
2831  * Function:
2832  *	vdc_send_request
2833  *
2834  * Description:
2835  *	This routine writes the data to be transmitted to vds into the
2836  *	descriptor, notifies vds that the ring has been updated and
2837  *	then waits for the request to be processed.
2838  *
2839  * Arguments:
2840  *	vdcp	  - the soft state pointer
2841  *	operation - operation we want vds to perform (VD_OP_XXX)
2842  *	addr	  - address of data buf to be read/written.
2843  *	nbytes	  - number of bytes to read/write
2844  *	slice	  - the disk slice this request is for
2845  *	offset	  - relative disk offset
2846  *	cb_type   - type of call - STRATEGY or SYNC
2847  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
2848  *			. mode for ioctl(9e)
2849  *			. LP64 diskaddr_t (block I/O)
2850  *	dir	  - direction of operation (READ/WRITE/BOTH)
2851  *
2852  * Return Codes:
2853  *	0
2854  *	ENXIO
2855  */
2856 static int
2857 vdc_send_request(vdc_t *vdcp, int operation, caddr_t addr,
2858     size_t nbytes, int slice, diskaddr_t offset, int cb_type,
2859     void *cb_arg, vio_desc_direction_t dir)
2860 {
2861 	int	rv = 0;
2862 
2863 	ASSERT(vdcp != NULL);
2864 	ASSERT(slice == VD_SLICE_NONE || slice < V_NUMPAR);
2865 
2866 	mutex_enter(&vdcp->lock);
2867 
2868 	/*
2869 	 * If this is a block read/write operation we update the I/O statistics
2870 	 * to indicate that the request is being put on the waitq to be
2871 	 * serviced.
2872 	 *
2873 	 * We do it here (a common routine for both synchronous and strategy
2874 	 * calls) for performance reasons - we are already holding vdc->lock
2875 	 * so there is no extra locking overhead. We would have to explicitly
2876 	 * grab the 'lock' mutex to update the stats if we were to do this
2877 	 * higher up the stack in vdc_strategy() et. al.
2878 	 */
2879 	if ((operation == VD_OP_BREAD) || (operation == VD_OP_BWRITE)) {
2880 		DTRACE_IO1(start, buf_t *, cb_arg);
2881 		VD_KSTAT_WAITQ_ENTER(vdcp);
2882 	}
2883 
2884 	do {
2885 		while (vdcp->state != VDC_STATE_RUNNING) {
2886 
2887 			/* return error if detaching */
2888 			if (vdcp->state == VDC_STATE_DETACH) {
2889 				rv = ENXIO;
2890 				goto done;
2891 			}
2892 
2893 			/* fail request if connection timeout is reached */
2894 			if (vdcp->ctimeout_reached) {
2895 				rv = EIO;
2896 				goto done;
2897 			}
2898 
2899 			/*
2900 			 * If we are panicking and the disk is not ready then
2901 			 * we can't send any request because we can't complete
2902 			 * the handshake now.
2903 			 */
2904 			if (ddi_in_panic()) {
2905 				rv = EIO;
2906 				goto done;
2907 			}
2908 
2909 			cv_wait(&vdcp->running_cv, &vdcp->lock);
2910 		}
2911 
2912 	} while (vdc_populate_descriptor(vdcp, operation, addr,
2913 	    nbytes, slice, offset, cb_type, cb_arg, dir));
2914 
2915 done:
2916 	/*
2917 	 * If this is a block read/write we update the I/O statistics kstat
2918 	 * to indicate that this request has been placed on the queue for
2919 	 * processing (i.e sent to the vDisk server) - iostat(1M) will
2920 	 * report the time waiting for the vDisk server under the %b column
2921 	 * In the case of an error we simply take it off the wait queue.
2922 	 */
2923 	if ((operation == VD_OP_BREAD) || (operation == VD_OP_BWRITE)) {
2924 		if (rv == 0) {
2925 			VD_KSTAT_WAITQ_TO_RUNQ(vdcp);
2926 			DTRACE_PROBE1(send, buf_t *, cb_arg);
2927 		} else {
2928 			VD_UPDATE_ERR_STATS(vdcp, vd_transerrs);
2929 			VD_KSTAT_WAITQ_EXIT(vdcp);
2930 			DTRACE_IO1(done, buf_t *, cb_arg);
2931 		}
2932 	}
2933 
2934 	mutex_exit(&vdcp->lock);
2935 
2936 	return (rv);
2937 }
2938 
2939 
2940 /*
2941  * Function:
2942  *	vdc_populate_descriptor
2943  *
2944  * Description:
2945  *	This routine writes the data to be transmitted to vds into the
2946  *	descriptor, notifies vds that the ring has been updated and
2947  *	then waits for the request to be processed.
2948  *
2949  * Arguments:
2950  *	vdcp	  - the soft state pointer
2951  *	operation - operation we want vds to perform (VD_OP_XXX)
2952  *	addr	  - address of data buf to be read/written.
2953  *	nbytes	  - number of bytes to read/write
2954  *	slice	  - the disk slice this request is for
2955  *	offset	  - relative disk offset
2956  *	cb_type   - type of call - STRATEGY or SYNC
2957  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
2958  *			. mode for ioctl(9e)
2959  *			. LP64 diskaddr_t (block I/O)
2960  *	dir	  - direction of operation (READ/WRITE/BOTH)
2961  *
2962  * Return Codes:
2963  *	0
2964  *	EAGAIN
2965  *	ECONNRESET
2966  *	ENXIO
2967  */
2968 static int
2969 vdc_populate_descriptor(vdc_t *vdcp, int operation, caddr_t addr,
2970     size_t nbytes, int slice, diskaddr_t offset, int cb_type,
2971     void *cb_arg, vio_desc_direction_t dir)
2972 {
2973 	vdc_local_desc_t	*local_dep = NULL; /* Local Dring Pointer */
2974 	int			idx;		/* Index of DRing entry used */
2975 	int			next_idx;
2976 	vio_dring_msg_t		dmsg;
2977 	size_t			msglen;
2978 	int			rv;
2979 
2980 	ASSERT(MUTEX_HELD(&vdcp->lock));
2981 	vdcp->threads_pending++;
2982 loop:
2983 	DMSG(vdcp, 2, ": dring_curr_idx = %d\n", vdcp->dring_curr_idx);
2984 
2985 	/* Get next available D-Ring entry */
2986 	idx = vdcp->dring_curr_idx;
2987 	local_dep = &(vdcp->local_dring[idx]);
2988 
2989 	if (!local_dep->is_free) {
2990 		DMSG(vdcp, 2, "[%d]: dring full - waiting for space\n",
2991 		    vdcp->instance);
2992 		cv_wait(&vdcp->dring_free_cv, &vdcp->lock);
2993 		if (vdcp->state == VDC_STATE_RUNNING ||
2994 		    vdcp->state == VDC_STATE_HANDLE_PENDING) {
2995 			goto loop;
2996 		}
2997 		vdcp->threads_pending--;
2998 		return (ECONNRESET);
2999 	}
3000 
3001 	next_idx = idx + 1;
3002 	if (next_idx >= vdcp->dring_len)
3003 		next_idx = 0;
3004 	vdcp->dring_curr_idx = next_idx;
3005 
3006 	ASSERT(local_dep->is_free);
3007 
3008 	local_dep->operation = operation;
3009 	local_dep->addr = addr;
3010 	local_dep->nbytes = nbytes;
3011 	local_dep->slice = slice;
3012 	local_dep->offset = offset;
3013 	local_dep->cb_type = cb_type;
3014 	local_dep->cb_arg = cb_arg;
3015 	local_dep->dir = dir;
3016 
3017 	local_dep->is_free = B_FALSE;
3018 
3019 	rv = vdc_map_to_shared_dring(vdcp, idx);
3020 	if (rv) {
3021 		DMSG(vdcp, 0, "[%d]: cannot bind memory - waiting ..\n",
3022 		    vdcp->instance);
3023 		/* free the descriptor */
3024 		local_dep->is_free = B_TRUE;
3025 		vdcp->dring_curr_idx = idx;
3026 		cv_wait(&vdcp->membind_cv, &vdcp->lock);
3027 		if (vdcp->state == VDC_STATE_RUNNING ||
3028 		    vdcp->state == VDC_STATE_HANDLE_PENDING) {
3029 			goto loop;
3030 		}
3031 		vdcp->threads_pending--;
3032 		return (ECONNRESET);
3033 	}
3034 
3035 	/*
3036 	 * Send a msg with the DRing details to vds
3037 	 */
3038 	VIO_INIT_DRING_DATA_TAG(dmsg);
3039 	VDC_INIT_DRING_DATA_MSG_IDS(dmsg, vdcp);
3040 	dmsg.dring_ident = vdcp->dring_ident;
3041 	dmsg.start_idx = idx;
3042 	dmsg.end_idx = idx;
3043 	vdcp->seq_num++;
3044 
3045 	DTRACE_PROBE2(populate, int, vdcp->instance,
3046 	    vdc_local_desc_t *, local_dep);
3047 	DMSG(vdcp, 2, "ident=0x%lx, st=%u, end=%u, seq=%ld\n",
3048 	    vdcp->dring_ident, dmsg.start_idx, dmsg.end_idx, dmsg.seq_num);
3049 
3050 	/*
3051 	 * note we're still holding the lock here to
3052 	 * make sure the message goes out in order !!!...
3053 	 */
3054 	msglen = sizeof (dmsg);
3055 	rv = vdc_send(vdcp, (caddr_t)&dmsg, &msglen);
3056 	switch (rv) {
3057 	case ECONNRESET:
3058 		/*
3059 		 * vdc_send initiates the reset on failure.
3060 		 * Since the transaction has already been put
3061 		 * on the local dring, it will automatically get
3062 		 * retried when the channel is reset. Given that,
3063 		 * it is ok to just return success even though the
3064 		 * send failed.
3065 		 */
3066 		rv = 0;
3067 		break;
3068 
3069 	case 0: /* EOK */
3070 		DMSG(vdcp, 1, "sent via LDC: rv=%d\n", rv);
3071 		break;
3072 
3073 	default:
3074 		goto cleanup_and_exit;
3075 	}
3076 
3077 	vdcp->threads_pending--;
3078 	return (rv);
3079 
3080 cleanup_and_exit:
3081 	DMSG(vdcp, 0, "unexpected error, rv=%d\n", rv);
3082 	return (ENXIO);
3083 }
3084 
3085 /*
3086  * Function:
3087  *	vdc_do_sync_op
3088  *
3089  * Description:
3090  * 	Wrapper around vdc_populate_descriptor that blocks until the
3091  * 	response to the message is available.
3092  *
3093  * Arguments:
3094  *	vdcp	  - the soft state pointer
3095  *	operation - operation we want vds to perform (VD_OP_XXX)
3096  *	addr	  - address of data buf to be read/written.
3097  *	nbytes	  - number of bytes to read/write
3098  *	slice	  - the disk slice this request is for
3099  *	offset	  - relative disk offset
3100  *	cb_type   - type of call - STRATEGY or SYNC
3101  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
3102  *			. mode for ioctl(9e)
3103  *			. LP64 diskaddr_t (block I/O)
3104  *	dir	  - direction of operation (READ/WRITE/BOTH)
3105  *	rconflict - check for reservation conflict in case of failure
3106  *
3107  * rconflict should be set to B_TRUE by most callers. Callers invoking the
3108  * VD_OP_SCSICMD operation can set rconflict to B_FALSE if they check the
3109  * result of a successful operation with vd_scsi_status().
3110  *
3111  * Return Codes:
3112  *	0
3113  *	EAGAIN
3114  *	EFAULT
3115  *	ENXIO
3116  *	EIO
3117  */
3118 static int
3119 vdc_do_sync_op(vdc_t *vdcp, int operation, caddr_t addr, size_t nbytes,
3120     int slice, diskaddr_t offset, int cb_type, void *cb_arg,
3121     vio_desc_direction_t dir, boolean_t rconflict)
3122 {
3123 	int status;
3124 	vdc_io_t *vio;
3125 	boolean_t check_resv_conflict = B_FALSE;
3126 
3127 	ASSERT(cb_type == CB_SYNC);
3128 
3129 	/*
3130 	 * Grab the lock, if blocked wait until the server
3131 	 * response causes us to wake up again.
3132 	 */
3133 	mutex_enter(&vdcp->lock);
3134 	vdcp->sync_op_cnt++;
3135 	while (vdcp->sync_op_blocked && vdcp->state != VDC_STATE_DETACH)
3136 		cv_wait(&vdcp->sync_blocked_cv, &vdcp->lock);
3137 
3138 	if (vdcp->state == VDC_STATE_DETACH) {
3139 		cv_broadcast(&vdcp->sync_blocked_cv);
3140 		vdcp->sync_op_cnt--;
3141 		mutex_exit(&vdcp->lock);
3142 		return (ENXIO);
3143 	}
3144 
3145 	/* now block anyone other thread entering after us */
3146 	vdcp->sync_op_blocked = B_TRUE;
3147 	vdcp->sync_op_pending = B_TRUE;
3148 	mutex_exit(&vdcp->lock);
3149 
3150 	status = vdc_send_request(vdcp, operation, addr,
3151 	    nbytes, slice, offset, cb_type, cb_arg, dir);
3152 
3153 	mutex_enter(&vdcp->lock);
3154 
3155 	if (status != 0) {
3156 		vdcp->sync_op_pending = B_FALSE;
3157 	} else {
3158 		/*
3159 		 * block until our transaction completes.
3160 		 * Also anyone else waiting also gets to go next.
3161 		 */
3162 		while (vdcp->sync_op_pending && vdcp->state != VDC_STATE_DETACH)
3163 			cv_wait(&vdcp->sync_pending_cv, &vdcp->lock);
3164 
3165 		DMSG(vdcp, 2, ": operation returned %d\n",
3166 		    vdcp->sync_op_status);
3167 		if (vdcp->state == VDC_STATE_DETACH) {
3168 			vdcp->sync_op_pending = B_FALSE;
3169 			status = ENXIO;
3170 		} else {
3171 			status = vdcp->sync_op_status;
3172 			if (status != 0 && vdcp->failfast_interval != 0) {
3173 				/*
3174 				 * Operation has failed and failfast is enabled.
3175 				 * We need to check if the failure is due to a
3176 				 * reservation conflict if this was requested.
3177 				 */
3178 				check_resv_conflict = rconflict;
3179 			}
3180 
3181 		}
3182 	}
3183 
3184 	vdcp->sync_op_status = 0;
3185 	vdcp->sync_op_blocked = B_FALSE;
3186 	vdcp->sync_op_cnt--;
3187 
3188 	/* signal the next waiting thread */
3189 	cv_signal(&vdcp->sync_blocked_cv);
3190 
3191 	/*
3192 	 * We have to check for reservation conflict after unblocking sync
3193 	 * operations because some sync operations will be used to do this
3194 	 * check.
3195 	 */
3196 	if (check_resv_conflict) {
3197 		vio = vdc_failfast_io_queue(vdcp, NULL);
3198 		while (vio->vio_qtime != 0)
3199 			cv_wait(&vdcp->failfast_io_cv, &vdcp->lock);
3200 		kmem_free(vio, sizeof (vdc_io_t));
3201 	}
3202 
3203 	mutex_exit(&vdcp->lock);
3204 
3205 	return (status);
3206 }
3207 
3208 
3209 /*
3210  * Function:
3211  *	vdc_drain_response()
3212  *
3213  * Description:
3214  * 	When a guest is panicking, the completion of requests needs to be
3215  * 	handled differently because interrupts are disabled and vdc
3216  * 	will not get messages. We have to poll for the messages instead.
3217  *
3218  *	Note: since we are panicking we don't implement	the io:::done
3219  *	DTrace probe or update the I/O statistics kstats.
3220  *
3221  * Arguments:
3222  *	vdc	- soft state pointer for this instance of the device driver.
3223  *	buf	- if buf is NULL then we drain all responses, otherwise we
3224  *		  poll until we receive a ACK/NACK for the specific I/O
3225  *		  described by buf.
3226  *
3227  * Return Code:
3228  *	0	- Success
3229  */
3230 static int
3231 vdc_drain_response(vdc_t *vdc, struct buf *buf)
3232 {
3233 	int 			rv, idx, retries;
3234 	size_t			msglen;
3235 	vdc_local_desc_t 	*ldep = NULL;	/* Local Dring Entry Pointer */
3236 	vio_dring_msg_t		dmsg;
3237 	struct buf		*mbuf;
3238 
3239 	mutex_enter(&vdc->lock);
3240 
3241 	retries = 0;
3242 	for (;;) {
3243 		msglen = sizeof (dmsg);
3244 		rv = ldc_read(vdc->curr_server->ldc_handle, (caddr_t)&dmsg,
3245 		    &msglen);
3246 		if (rv) {
3247 			rv = EINVAL;
3248 			break;
3249 		}
3250 
3251 		/*
3252 		 * if there are no packets wait and check again
3253 		 */
3254 		if ((rv == 0) && (msglen == 0)) {
3255 			if (retries++ > vdc_dump_retries) {
3256 				rv = EAGAIN;
3257 				break;
3258 			}
3259 
3260 			drv_usecwait(vdc_usec_timeout_dump);
3261 			continue;
3262 		}
3263 
3264 		/*
3265 		 * Ignore all messages that are not ACKs/NACKs to
3266 		 * DRing requests.
3267 		 */
3268 		if ((dmsg.tag.vio_msgtype != VIO_TYPE_DATA) ||
3269 		    (dmsg.tag.vio_subtype_env != VIO_DRING_DATA)) {
3270 			DMSG(vdc, 0, "discard pkt: type=%d sub=%d env=%d\n",
3271 			    dmsg.tag.vio_msgtype,
3272 			    dmsg.tag.vio_subtype,
3273 			    dmsg.tag.vio_subtype_env);
3274 			continue;
3275 		}
3276 
3277 		/*
3278 		 * set the appropriate return value for the current request.
3279 		 */
3280 		switch (dmsg.tag.vio_subtype) {
3281 		case VIO_SUBTYPE_ACK:
3282 			rv = 0;
3283 			break;
3284 		case VIO_SUBTYPE_NACK:
3285 			rv = EAGAIN;
3286 			break;
3287 		default:
3288 			continue;
3289 		}
3290 
3291 		idx = dmsg.start_idx;
3292 		if (idx >= vdc->dring_len) {
3293 			DMSG(vdc, 0, "[%d] Bogus ack data : start %d\n",
3294 			    vdc->instance, idx);
3295 			continue;
3296 		}
3297 		ldep = &vdc->local_dring[idx];
3298 		if (ldep->dep->hdr.dstate != VIO_DESC_DONE) {
3299 			DMSG(vdc, 0, "[%d] Entry @ %d - state !DONE %d\n",
3300 			    vdc->instance, idx, ldep->dep->hdr.dstate);
3301 			continue;
3302 		}
3303 
3304 		if (buf != NULL && ldep->cb_type == CB_STRATEGY) {
3305 			mbuf = ldep->cb_arg;
3306 			mbuf->b_resid = mbuf->b_bcount -
3307 			    ldep->dep->payload.nbytes;
3308 			bioerror(mbuf, (rv == EAGAIN)? EIO:
3309 			    ldep->dep->payload.status);
3310 			biodone(mbuf);
3311 		} else {
3312 			mbuf = NULL;
3313 		}
3314 
3315 		DMSG(vdc, 1, "[%d] Depopulating idx=%d state=%d\n",
3316 		    vdc->instance, idx, ldep->dep->hdr.dstate);
3317 
3318 		rv = vdc_depopulate_descriptor(vdc, idx);
3319 		if (rv) {
3320 			DMSG(vdc, 0,
3321 			    "[%d] Entry @ %d - depopulate failed ..\n",
3322 			    vdc->instance, idx);
3323 		}
3324 
3325 		/* we have received an ACK/NACK for the specified buffer */
3326 		if (buf != NULL && buf == mbuf) {
3327 			rv = 0;
3328 			break;
3329 		}
3330 
3331 		/* if this is the last descriptor - break out of loop */
3332 		if ((idx + 1) % vdc->dring_len == vdc->dring_curr_idx) {
3333 			if (buf != NULL) {
3334 				/*
3335 				 * We never got a response for the specified
3336 				 * buffer so we fail the I/O.
3337 				 */
3338 				bioerror(buf, EIO);
3339 				biodone(buf);
3340 			}
3341 			break;
3342 		}
3343 	}
3344 
3345 	mutex_exit(&vdc->lock);
3346 	DMSG(vdc, 0, "End idx=%d\n", idx);
3347 
3348 	return (rv);
3349 }
3350 
3351 
3352 /*
3353  * Function:
3354  *	vdc_depopulate_descriptor()
3355  *
3356  * Description:
3357  *
3358  * Arguments:
3359  *	vdc	- soft state pointer for this instance of the device driver.
3360  *	idx	- Index of the Descriptor Ring entry being modified
3361  *
3362  * Return Code:
3363  *	0	- Success
3364  */
3365 static int
3366 vdc_depopulate_descriptor(vdc_t *vdc, uint_t idx)
3367 {
3368 	vd_dring_entry_t *dep = NULL;		/* Dring Entry Pointer */
3369 	vdc_local_desc_t *ldep = NULL;		/* Local Dring Entry Pointer */
3370 	int		status = ENXIO;
3371 	int		rv = 0;
3372 
3373 	ASSERT(vdc != NULL);
3374 	ASSERT(idx < vdc->dring_len);
3375 	ldep = &vdc->local_dring[idx];
3376 	ASSERT(ldep != NULL);
3377 	ASSERT(MUTEX_HELD(&vdc->lock));
3378 
3379 	DTRACE_PROBE2(depopulate, int, vdc->instance, vdc_local_desc_t *, ldep);
3380 	DMSG(vdc, 2, ": idx = %d\n", idx);
3381 
3382 	dep = ldep->dep;
3383 	ASSERT(dep != NULL);
3384 	ASSERT((dep->hdr.dstate == VIO_DESC_DONE) ||
3385 	    (dep->payload.status == ECANCELED));
3386 
3387 	VDC_MARK_DRING_ENTRY_FREE(vdc, idx);
3388 
3389 	ldep->is_free = B_TRUE;
3390 	status = dep->payload.status;
3391 	DMSG(vdc, 2, ": is_free = %d : status = %d\n", ldep->is_free, status);
3392 
3393 	/*
3394 	 * If no buffers were used to transfer information to the server when
3395 	 * populating the descriptor then no memory handles need to be unbound
3396 	 * and we can return now.
3397 	 */
3398 	if (ldep->nbytes == 0) {
3399 		cv_signal(&vdc->dring_free_cv);
3400 		return (status);
3401 	}
3402 
3403 	/*
3404 	 * If the upper layer passed in a misaligned address we copied the
3405 	 * data into an aligned buffer before sending it to LDC - we now
3406 	 * copy it back to the original buffer.
3407 	 */
3408 	if (ldep->align_addr) {
3409 		ASSERT(ldep->addr != NULL);
3410 
3411 		if (dep->payload.nbytes > 0)
3412 			bcopy(ldep->align_addr, ldep->addr,
3413 			    dep->payload.nbytes);
3414 		kmem_free(ldep->align_addr,
3415 		    sizeof (caddr_t) * P2ROUNDUP(ldep->nbytes, 8));
3416 		ldep->align_addr = NULL;
3417 	}
3418 
3419 	rv = ldc_mem_unbind_handle(ldep->desc_mhdl);
3420 	if (rv != 0) {
3421 		DMSG(vdc, 0, "?[%d] unbind mhdl 0x%lx @ idx %d failed (%d)",
3422 		    vdc->instance, ldep->desc_mhdl, idx, rv);
3423 		/*
3424 		 * The error returned by the vDisk server is more informative
3425 		 * and thus has a higher priority but if it isn't set we ensure
3426 		 * that this function returns an error.
3427 		 */
3428 		if (status == 0)
3429 			status = EINVAL;
3430 	}
3431 
3432 	cv_signal(&vdc->membind_cv);
3433 	cv_signal(&vdc->dring_free_cv);
3434 
3435 	return (status);
3436 }
3437 
3438 /*
3439  * Function:
3440  *	vdc_populate_mem_hdl()
3441  *
3442  * Description:
3443  *
3444  * Arguments:
3445  *	vdc	- soft state pointer for this instance of the device driver.
3446  *	idx	- Index of the Descriptor Ring entry being modified
3447  *	addr	- virtual address being mapped in
3448  *	nybtes	- number of bytes in 'addr'
3449  *	operation - the vDisk operation being performed (VD_OP_xxx)
3450  *
3451  * Return Code:
3452  *	0	- Success
3453  */
3454 static int
3455 vdc_populate_mem_hdl(vdc_t *vdcp, vdc_local_desc_t *ldep)
3456 {
3457 	vd_dring_entry_t	*dep = NULL;
3458 	ldc_mem_handle_t	mhdl;
3459 	caddr_t			vaddr;
3460 	size_t			nbytes;
3461 	uint8_t			perm = LDC_MEM_RW;
3462 	uint8_t			maptype;
3463 	int			rv = 0;
3464 	int			i;
3465 
3466 	ASSERT(vdcp != NULL);
3467 
3468 	dep = ldep->dep;
3469 	mhdl = ldep->desc_mhdl;
3470 
3471 	switch (ldep->dir) {
3472 	case VIO_read_dir:
3473 		perm = LDC_MEM_W;
3474 		break;
3475 
3476 	case VIO_write_dir:
3477 		perm = LDC_MEM_R;
3478 		break;
3479 
3480 	case VIO_both_dir:
3481 		perm = LDC_MEM_RW;
3482 		break;
3483 
3484 	default:
3485 		ASSERT(0);	/* catch bad programming in vdc */
3486 	}
3487 
3488 	/*
3489 	 * LDC expects any addresses passed in to be 8-byte aligned. We need
3490 	 * to copy the contents of any misaligned buffers to a newly allocated
3491 	 * buffer and bind it instead (and copy the the contents back to the
3492 	 * original buffer passed in when depopulating the descriptor)
3493 	 */
3494 	vaddr = ldep->addr;
3495 	nbytes = ldep->nbytes;
3496 	if (((uint64_t)vaddr & 0x7) != 0) {
3497 		ASSERT(ldep->align_addr == NULL);
3498 		ldep->align_addr =
3499 		    kmem_alloc(sizeof (caddr_t) *
3500 		    P2ROUNDUP(nbytes, 8), KM_SLEEP);
3501 		DMSG(vdcp, 0, "[%d] Misaligned address %p reallocating "
3502 		    "(buf=%p nb=%ld op=%d)\n",
3503 		    vdcp->instance, (void *)vaddr, (void *)ldep->align_addr,
3504 		    nbytes, ldep->operation);
3505 		if (perm != LDC_MEM_W)
3506 			bcopy(vaddr, ldep->align_addr, nbytes);
3507 		vaddr = ldep->align_addr;
3508 	}
3509 
3510 	maptype = LDC_IO_MAP|LDC_SHADOW_MAP|LDC_DIRECT_MAP;
3511 	rv = ldc_mem_bind_handle(mhdl, vaddr, P2ROUNDUP(nbytes, 8),
3512 	    maptype, perm, &dep->payload.cookie[0], &dep->payload.ncookies);
3513 	DMSG(vdcp, 2, "[%d] bound mem handle; ncookies=%d\n",
3514 	    vdcp->instance, dep->payload.ncookies);
3515 	if (rv != 0) {
3516 		DMSG(vdcp, 0, "[%d] Failed to bind LDC memory handle "
3517 		    "(mhdl=%p, buf=%p, err=%d)\n",
3518 		    vdcp->instance, (void *)mhdl, (void *)vaddr, rv);
3519 		if (ldep->align_addr) {
3520 			kmem_free(ldep->align_addr,
3521 			    sizeof (caddr_t) * P2ROUNDUP(nbytes, 8));
3522 			ldep->align_addr = NULL;
3523 		}
3524 		return (EAGAIN);
3525 	}
3526 
3527 	/*
3528 	 * Get the other cookies (if any).
3529 	 */
3530 	for (i = 1; i < dep->payload.ncookies; i++) {
3531 		rv = ldc_mem_nextcookie(mhdl, &dep->payload.cookie[i]);
3532 		if (rv != 0) {
3533 			(void) ldc_mem_unbind_handle(mhdl);
3534 			DMSG(vdcp, 0, "?[%d] Failed to get next cookie "
3535 			    "(mhdl=%lx cnum=%d), err=%d",
3536 			    vdcp->instance, mhdl, i, rv);
3537 			if (ldep->align_addr) {
3538 				kmem_free(ldep->align_addr,
3539 				    sizeof (caddr_t) * ldep->nbytes);
3540 				ldep->align_addr = NULL;
3541 			}
3542 			return (EAGAIN);
3543 		}
3544 	}
3545 
3546 	return (rv);
3547 }
3548 
3549 /*
3550  * Interrupt handlers for messages from LDC
3551  */
3552 
3553 /*
3554  * Function:
3555  *	vdc_handle_cb()
3556  *
3557  * Description:
3558  *
3559  * Arguments:
3560  *	event	- Type of event (LDC_EVT_xxx) that triggered the callback
3561  *	arg	- soft state pointer for this instance of the device driver.
3562  *
3563  * Return Code:
3564  *	0	- Success
3565  */
3566 static uint_t
3567 vdc_handle_cb(uint64_t event, caddr_t arg)
3568 {
3569 	ldc_status_t	ldc_state;
3570 	int		rv = 0;
3571 	vdc_server_t	*srvr = (vdc_server_t *)(void *)arg;
3572 	vdc_t		*vdc = srvr->vdcp;
3573 
3574 	ASSERT(vdc != NULL);
3575 
3576 	DMSG(vdc, 1, "evt=%lx seqID=%ld\n", event, vdc->seq_num);
3577 
3578 	/* If callback is not for the current server, ignore it */
3579 	mutex_enter(&vdc->lock);
3580 
3581 	if (vdc->curr_server != srvr) {
3582 		DMSG(vdc, 0, "[%d] Ignoring event 0x%lx for port@%ld\n",
3583 		    vdc->instance, event, srvr->id);
3584 		mutex_exit(&vdc->lock);
3585 		return (LDC_SUCCESS);
3586 	}
3587 
3588 	/*
3589 	 * Depending on the type of event that triggered this callback,
3590 	 * we modify the handshake state or read the data.
3591 	 *
3592 	 * NOTE: not done as a switch() as event could be triggered by
3593 	 * a state change and a read request. Also the ordering	of the
3594 	 * check for the event types is deliberate.
3595 	 */
3596 	if (event & LDC_EVT_UP) {
3597 		DMSG(vdc, 0, "[%d] Received LDC_EVT_UP\n", vdc->instance);
3598 
3599 		/* get LDC state */
3600 		rv = ldc_status(srvr->ldc_handle, &ldc_state);
3601 		if (rv != 0) {
3602 			DMSG(vdc, 0, "[%d] Couldn't get LDC status %d",
3603 			    vdc->instance, rv);
3604 			mutex_exit(&vdc->lock);
3605 			return (LDC_SUCCESS);
3606 		}
3607 		if (srvr->ldc_state != LDC_UP &&
3608 		    ldc_state == LDC_UP) {
3609 			/*
3610 			 * Reset the transaction sequence numbers when
3611 			 * LDC comes up. We then kick off the handshake
3612 			 * negotiation with the vDisk server.
3613 			 */
3614 			vdc->seq_num = 1;
3615 			vdc->seq_num_reply = 0;
3616 			srvr->ldc_state = ldc_state;
3617 			cv_signal(&vdc->initwait_cv);
3618 		}
3619 	}
3620 
3621 	if (event & LDC_EVT_READ) {
3622 		DMSG(vdc, 1, "[%d] Received LDC_EVT_READ\n", vdc->instance);
3623 		mutex_enter(&vdc->read_lock);
3624 		cv_signal(&vdc->read_cv);
3625 		vdc->read_state = VDC_READ_PENDING;
3626 		mutex_exit(&vdc->read_lock);
3627 		mutex_exit(&vdc->lock);
3628 
3629 		/* that's all we have to do - no need to handle DOWN/RESET */
3630 		return (LDC_SUCCESS);
3631 	}
3632 
3633 	if (event & (LDC_EVT_RESET|LDC_EVT_DOWN)) {
3634 
3635 		DMSG(vdc, 0, "[%d] Received LDC RESET event\n", vdc->instance);
3636 
3637 		/*
3638 		 * Need to wake up any readers so they will
3639 		 * detect that a reset has occurred.
3640 		 */
3641 		mutex_enter(&vdc->read_lock);
3642 		if ((vdc->read_state == VDC_READ_WAITING) ||
3643 		    (vdc->read_state == VDC_READ_RESET))
3644 			cv_signal(&vdc->read_cv);
3645 		vdc->read_state = VDC_READ_RESET;
3646 		mutex_exit(&vdc->read_lock);
3647 
3648 		/* wake up any threads waiting for connection to come up */
3649 		if (vdc->state == VDC_STATE_INIT_WAITING) {
3650 			vdc->state = VDC_STATE_RESETTING;
3651 			cv_signal(&vdc->initwait_cv);
3652 		}
3653 
3654 	}
3655 
3656 	mutex_exit(&vdc->lock);
3657 
3658 	if (event & ~(LDC_EVT_UP | LDC_EVT_RESET | LDC_EVT_DOWN | LDC_EVT_READ))
3659 		DMSG(vdc, 0, "![%d] Unexpected LDC event (%lx) received",
3660 		    vdc->instance, event);
3661 
3662 	return (LDC_SUCCESS);
3663 }
3664 
3665 /*
3666  * Function:
3667  *	vdc_wait_for_response()
3668  *
3669  * Description:
3670  *	Block waiting for a response from the server. If there is
3671  *	no data the thread block on the read_cv that is signalled
3672  *	by the callback when an EVT_READ occurs.
3673  *
3674  * Arguments:
3675  *	vdcp	- soft state pointer for this instance of the device driver.
3676  *
3677  * Return Code:
3678  *	0	- Success
3679  */
3680 static int
3681 vdc_wait_for_response(vdc_t *vdcp, vio_msg_t *msgp)
3682 {
3683 	size_t		nbytes = sizeof (*msgp);
3684 	int		status;
3685 
3686 	ASSERT(vdcp != NULL);
3687 
3688 	DMSG(vdcp, 1, "[%d] Entered\n", vdcp->instance);
3689 
3690 	status = vdc_recv(vdcp, msgp, &nbytes);
3691 	DMSG(vdcp, 3, "vdc_read() done.. status=0x%x size=0x%x\n",
3692 	    status, (int)nbytes);
3693 	if (status) {
3694 		DMSG(vdcp, 0, "?[%d] Error %d reading LDC msg\n",
3695 		    vdcp->instance, status);
3696 		return (status);
3697 	}
3698 
3699 	if (nbytes < sizeof (vio_msg_tag_t)) {
3700 		DMSG(vdcp, 0, "?[%d] Expect %lu bytes; recv'd %lu\n",
3701 		    vdcp->instance, sizeof (vio_msg_tag_t), nbytes);
3702 		return (ENOMSG);
3703 	}
3704 
3705 	DMSG(vdcp, 2, "[%d] (%x/%x/%x)\n", vdcp->instance,
3706 	    msgp->tag.vio_msgtype,
3707 	    msgp->tag.vio_subtype,
3708 	    msgp->tag.vio_subtype_env);
3709 
3710 	/*
3711 	 * Verify the Session ID of the message
3712 	 *
3713 	 * Every message after the Version has been negotiated should
3714 	 * have the correct session ID set.
3715 	 */
3716 	if ((msgp->tag.vio_sid != vdcp->session_id) &&
3717 	    (msgp->tag.vio_subtype_env != VIO_VER_INFO)) {
3718 		DMSG(vdcp, 0, "[%d] Invalid SID: received 0x%x, "
3719 		    "expected 0x%lx [seq num %lx @ %d]",
3720 		    vdcp->instance, msgp->tag.vio_sid,
3721 		    vdcp->session_id,
3722 		    ((vio_dring_msg_t *)msgp)->seq_num,
3723 		    ((vio_dring_msg_t *)msgp)->start_idx);
3724 		return (ENOMSG);
3725 	}
3726 	return (0);
3727 }
3728 
3729 
3730 /*
3731  * Function:
3732  *	vdc_resubmit_backup_dring()
3733  *
3734  * Description:
3735  *	Resubmit each descriptor in the backed up dring to
3736  * 	vDisk server. The Dring was backed up during connection
3737  *	reset.
3738  *
3739  * Arguments:
3740  *	vdcp	- soft state pointer for this instance of the device driver.
3741  *
3742  * Return Code:
3743  *	0	- Success
3744  */
3745 static int
3746 vdc_resubmit_backup_dring(vdc_t *vdcp)
3747 {
3748 	int		processed = 0;
3749 	int		count;
3750 	int		b_idx;
3751 	int		rv = 0;
3752 	int		dring_size;
3753 	int		op;
3754 	vio_msg_t	vio_msg;
3755 	vdc_local_desc_t	*curr_ldep;
3756 
3757 	ASSERT(MUTEX_NOT_HELD(&vdcp->lock));
3758 	ASSERT(vdcp->state == VDC_STATE_HANDLE_PENDING);
3759 
3760 	if (vdcp->local_dring_backup == NULL) {
3761 		/* the pending requests have already been processed */
3762 		return (0);
3763 	}
3764 
3765 	DMSG(vdcp, 1, "restoring pending dring entries (len=%d, tail=%d)\n",
3766 	    vdcp->local_dring_backup_len, vdcp->local_dring_backup_tail);
3767 
3768 	/*
3769 	 * Walk the backup copy of the local descriptor ring and
3770 	 * resubmit all the outstanding transactions.
3771 	 */
3772 	b_idx = vdcp->local_dring_backup_tail;
3773 	for (count = 0; count < vdcp->local_dring_backup_len; count++) {
3774 
3775 		curr_ldep = &(vdcp->local_dring_backup[b_idx]);
3776 
3777 		/* only resubmit outstanding transactions */
3778 		if (!curr_ldep->is_free) {
3779 			/*
3780 			 * If we are retrying a block read/write operation we
3781 			 * need to update the I/O statistics to indicate that
3782 			 * the request is being put back on the waitq to be
3783 			 * serviced (it will have been taken off after the
3784 			 * error was reported).
3785 			 */
3786 			mutex_enter(&vdcp->lock);
3787 			op = curr_ldep->operation;
3788 			if ((op == VD_OP_BREAD) || (op == VD_OP_BWRITE)) {
3789 				DTRACE_IO1(start, buf_t *, curr_ldep->cb_arg);
3790 				VD_KSTAT_WAITQ_ENTER(vdcp);
3791 			}
3792 
3793 			DMSG(vdcp, 1, "resubmitting entry idx=%x\n", b_idx);
3794 			rv = vdc_populate_descriptor(vdcp, op,
3795 			    curr_ldep->addr, curr_ldep->nbytes,
3796 			    curr_ldep->slice, curr_ldep->offset,
3797 			    curr_ldep->cb_type, curr_ldep->cb_arg,
3798 			    curr_ldep->dir);
3799 
3800 			if (rv) {
3801 				if (op == VD_OP_BREAD || op == VD_OP_BWRITE) {
3802 					VD_UPDATE_ERR_STATS(vdcp, vd_transerrs);
3803 					VD_KSTAT_WAITQ_EXIT(vdcp);
3804 					DTRACE_IO1(done, buf_t *,
3805 					    curr_ldep->cb_arg);
3806 				}
3807 				DMSG(vdcp, 1, "[%d] cannot resubmit entry %d\n",
3808 				    vdcp->instance, b_idx);
3809 				mutex_exit(&vdcp->lock);
3810 				goto done;
3811 			}
3812 
3813 			/*
3814 			 * If this is a block read/write we update the I/O
3815 			 * statistics kstat to indicate that the request
3816 			 * has been sent back to the vDisk server and should
3817 			 * now be put on the run queue.
3818 			 */
3819 			if ((op == VD_OP_BREAD) || (op == VD_OP_BWRITE)) {
3820 				DTRACE_PROBE1(send, buf_t *, curr_ldep->cb_arg);
3821 				VD_KSTAT_WAITQ_TO_RUNQ(vdcp);
3822 			}
3823 			mutex_exit(&vdcp->lock);
3824 
3825 			/* Wait for the response message. */
3826 			DMSG(vdcp, 1, "waiting for response to idx=%x\n",
3827 			    b_idx);
3828 			rv = vdc_wait_for_response(vdcp, &vio_msg);
3829 			if (rv) {
3830 				/*
3831 				 * If this is a block read/write we update
3832 				 * the I/O statistics kstat to take it
3833 				 * off the run queue.
3834 				 */
3835 				mutex_enter(&vdcp->lock);
3836 				if (op == VD_OP_BREAD || op == VD_OP_BWRITE) {
3837 					VD_UPDATE_ERR_STATS(vdcp, vd_transerrs);
3838 					VD_KSTAT_RUNQ_EXIT(vdcp);
3839 					DTRACE_IO1(done, buf_t *,
3840 					    curr_ldep->cb_arg);
3841 				}
3842 				DMSG(vdcp, 1, "[%d] wait_for_response "
3843 				    "returned err=%d\n", vdcp->instance,
3844 				    rv);
3845 				mutex_exit(&vdcp->lock);
3846 				goto done;
3847 			}
3848 
3849 			DMSG(vdcp, 1, "processing msg for idx=%x\n", b_idx);
3850 			rv = vdc_process_data_msg(vdcp, &vio_msg);
3851 			if (rv) {
3852 				DMSG(vdcp, 1, "[%d] process_data_msg "
3853 				    "returned err=%d\n", vdcp->instance,
3854 				    rv);
3855 				goto done;
3856 			}
3857 			/*
3858 			 * Mark this entry as free so that we will not resubmit
3859 			 * this "done" request again, if we were to use the same
3860 			 * backup_dring again in future. This could happen when
3861 			 * a reset happens while processing the backup_dring.
3862 			 */
3863 			curr_ldep->is_free = B_TRUE;
3864 			processed++;
3865 		}
3866 
3867 		/* get the next element to submit */
3868 		if (++b_idx >= vdcp->local_dring_backup_len)
3869 			b_idx = 0;
3870 	}
3871 
3872 	/* all done - now clear up pending dring copy */
3873 	dring_size = vdcp->local_dring_backup_len *
3874 	    sizeof (vdcp->local_dring_backup[0]);
3875 
3876 	(void) kmem_free(vdcp->local_dring_backup, dring_size);
3877 
3878 	vdcp->local_dring_backup = NULL;
3879 
3880 done:
3881 	DTRACE_PROBE2(processed, int, processed, vdc_t *, vdcp);
3882 
3883 	return (rv);
3884 }
3885 
3886 /*
3887  * Function:
3888  *	vdc_cancel_backup_dring
3889  *
3890  * Description:
3891  *	Cancel each descriptor in the backed up dring to vDisk server.
3892  *	The Dring was backed up during connection reset.
3893  *
3894  * Arguments:
3895  *	vdcp	- soft state pointer for this instance of the device driver.
3896  *
3897  * Return Code:
3898  *	None
3899  */
3900 void
3901 vdc_cancel_backup_dring(vdc_t *vdcp)
3902 {
3903 	vdc_local_desc_t *ldep;
3904 	struct buf 	*bufp;
3905 	int		count;
3906 	int		b_idx;
3907 	int		dring_size;
3908 	int		cancelled = 0;
3909 
3910 	ASSERT(MUTEX_HELD(&vdcp->lock));
3911 	ASSERT(vdcp->state == VDC_STATE_INIT ||
3912 	    vdcp->state == VDC_STATE_INIT_WAITING ||
3913 	    vdcp->state == VDC_STATE_NEGOTIATE ||
3914 	    vdcp->state == VDC_STATE_RESETTING);
3915 
3916 	if (vdcp->local_dring_backup == NULL) {
3917 		/* the pending requests have already been processed */
3918 		return;
3919 	}
3920 
3921 	DMSG(vdcp, 1, "cancelling pending dring entries (len=%d, tail=%d)\n",
3922 	    vdcp->local_dring_backup_len, vdcp->local_dring_backup_tail);
3923 
3924 	/*
3925 	 * Walk the backup copy of the local descriptor ring and
3926 	 * cancel all the outstanding transactions.
3927 	 */
3928 	b_idx = vdcp->local_dring_backup_tail;
3929 	for (count = 0; count < vdcp->local_dring_backup_len; count++) {
3930 
3931 		ldep = &(vdcp->local_dring_backup[b_idx]);
3932 
3933 		/* only cancel outstanding transactions */
3934 		if (!ldep->is_free) {
3935 
3936 			DMSG(vdcp, 1, "cancelling entry idx=%x\n", b_idx);
3937 			cancelled++;
3938 
3939 			/*
3940 			 * All requests have already been cleared from the
3941 			 * local descriptor ring and the LDC channel has been
3942 			 * reset so we will never get any reply for these
3943 			 * requests. Now we just have to notify threads waiting
3944 			 * for replies that the request has failed.
3945 			 */
3946 			switch (ldep->cb_type) {
3947 			case CB_SYNC:
3948 				ASSERT(vdcp->sync_op_pending);
3949 				vdcp->sync_op_status = EIO;
3950 				vdcp->sync_op_pending = B_FALSE;
3951 				cv_signal(&vdcp->sync_pending_cv);
3952 				break;
3953 
3954 			case CB_STRATEGY:
3955 				bufp = ldep->cb_arg;
3956 				ASSERT(bufp != NULL);
3957 				bufp->b_resid = bufp->b_bcount;
3958 				VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
3959 				VD_KSTAT_RUNQ_EXIT(vdcp);
3960 				DTRACE_IO1(done, buf_t *, bufp);
3961 				bioerror(bufp, EIO);
3962 				biodone(bufp);
3963 				break;
3964 
3965 			default:
3966 				ASSERT(0);
3967 			}
3968 
3969 		}
3970 
3971 		/* get the next element to cancel */
3972 		if (++b_idx >= vdcp->local_dring_backup_len)
3973 			b_idx = 0;
3974 	}
3975 
3976 	/* all done - now clear up pending dring copy */
3977 	dring_size = vdcp->local_dring_backup_len *
3978 	    sizeof (vdcp->local_dring_backup[0]);
3979 
3980 	(void) kmem_free(vdcp->local_dring_backup, dring_size);
3981 
3982 	vdcp->local_dring_backup = NULL;
3983 
3984 	DTRACE_PROBE2(cancelled, int, cancelled, vdc_t *, vdcp);
3985 }
3986 
3987 /*
3988  * Function:
3989  *	vdc_connection_timeout
3990  *
3991  * Description:
3992  *	This function is invoked if the timeout set to establish the connection
3993  *	with vds expires. This will happen if we spend too much time in the
3994  *	VDC_STATE_INIT_WAITING or VDC_STATE_NEGOTIATE states. Then we will
3995  *	cancel any pending request and mark them as failed.
3996  *
3997  *	If the timeout does not expire, it will be cancelled when we reach the
3998  *	VDC_STATE_HANDLE_PENDING or VDC_STATE_RESETTING state. This function can
3999  *	be invoked while we are in the VDC_STATE_HANDLE_PENDING or
4000  *	VDC_STATE_RESETTING state in which case we do nothing because the
4001  *	timeout is being cancelled.
4002  *
4003  * Arguments:
4004  *	arg	- argument of the timeout function actually a soft state
4005  *		  pointer for the instance of the device driver.
4006  *
4007  * Return Code:
4008  *	None
4009  */
4010 void
4011 vdc_connection_timeout(void *arg)
4012 {
4013 	vdc_t 		*vdcp = (vdc_t *)arg;
4014 
4015 	mutex_enter(&vdcp->lock);
4016 
4017 	if (vdcp->state == VDC_STATE_HANDLE_PENDING ||
4018 	    vdcp->state == VDC_STATE_DETACH) {
4019 		/*
4020 		 * The connection has just been re-established or
4021 		 * we are detaching.
4022 		 */
4023 		vdcp->ctimeout_reached = B_FALSE;
4024 		mutex_exit(&vdcp->lock);
4025 		return;
4026 	}
4027 
4028 	vdcp->ctimeout_reached = B_TRUE;
4029 
4030 	/* notify requests waiting for sending */
4031 	cv_broadcast(&vdcp->running_cv);
4032 
4033 	/* cancel requests waiting for a result */
4034 	vdc_cancel_backup_dring(vdcp);
4035 
4036 	mutex_exit(&vdcp->lock);
4037 
4038 	cmn_err(CE_NOTE, "[%d] connection to service domain timeout",
4039 	    vdcp->instance);
4040 }
4041 
4042 /*
4043  * Function:
4044  *	vdc_backup_local_dring()
4045  *
4046  * Description:
4047  *	Backup the current dring in the event of a reset. The Dring
4048  *	transactions will be resubmitted to the server when the
4049  *	connection is restored.
4050  *
4051  * Arguments:
4052  *	vdcp	- soft state pointer for this instance of the device driver.
4053  *
4054  * Return Code:
4055  *	NONE
4056  */
4057 static void
4058 vdc_backup_local_dring(vdc_t *vdcp)
4059 {
4060 	int dring_size;
4061 
4062 	ASSERT(MUTEX_HELD(&vdcp->lock));
4063 	ASSERT(vdcp->state == VDC_STATE_RESETTING);
4064 
4065 	/*
4066 	 * If the backup dring is stil around, it means
4067 	 * that the last restore did not complete. However,
4068 	 * since we never got back into the running state,
4069 	 * the backup copy we have is still valid.
4070 	 */
4071 	if (vdcp->local_dring_backup != NULL) {
4072 		DMSG(vdcp, 1, "reusing local descriptor ring backup "
4073 		    "(len=%d, tail=%d)\n", vdcp->local_dring_backup_len,
4074 		    vdcp->local_dring_backup_tail);
4075 		return;
4076 	}
4077 
4078 	/*
4079 	 * The backup dring can be NULL and the local dring may not be
4080 	 * initialized. This can happen if we had a reset while establishing
4081 	 * a new connection but after the connection has timed out. In that
4082 	 * case the backup dring is NULL because the requests have been
4083 	 * cancelled and the request occured before the local dring is
4084 	 * initialized.
4085 	 */
4086 	if (!(vdcp->initialized & VDC_DRING_LOCAL))
4087 		return;
4088 
4089 	DMSG(vdcp, 1, "backing up the local descriptor ring (len=%d, "
4090 	    "tail=%d)\n", vdcp->dring_len, vdcp->dring_curr_idx);
4091 
4092 	dring_size = vdcp->dring_len * sizeof (vdcp->local_dring[0]);
4093 
4094 	vdcp->local_dring_backup = kmem_alloc(dring_size, KM_SLEEP);
4095 	bcopy(vdcp->local_dring, vdcp->local_dring_backup, dring_size);
4096 
4097 	vdcp->local_dring_backup_tail = vdcp->dring_curr_idx;
4098 	vdcp->local_dring_backup_len = vdcp->dring_len;
4099 }
4100 
4101 static void
4102 vdc_switch_server(vdc_t *vdcp)
4103 {
4104 	int		rv;
4105 	vdc_server_t 	*curr_server, *new_server;
4106 
4107 	ASSERT(MUTEX_HELD(&vdcp->lock));
4108 
4109 	/* if there is only one server return back */
4110 	if (vdcp->num_servers == 1) {
4111 		return;
4112 	}
4113 
4114 	/* Get current and next server */
4115 	curr_server = vdcp->curr_server;
4116 	new_server =
4117 	    (curr_server->next) ? curr_server->next : vdcp->server_list;
4118 	ASSERT(curr_server != new_server);
4119 
4120 	/* bring current server's channel down */
4121 	rv = ldc_down(curr_server->ldc_handle);
4122 	if (rv) {
4123 		DMSG(vdcp, 0, "[%d] Cannot bring channel down, port %ld\n",
4124 		    vdcp->instance, curr_server->id);
4125 		return;
4126 	}
4127 
4128 	/* switch the server */
4129 	vdcp->curr_server = new_server;
4130 
4131 	DMSG(vdcp, 0, "[%d] Switched to next vdisk server, port@%ld, ldc@%ld\n",
4132 	    vdcp->instance, vdcp->curr_server->id, vdcp->curr_server->ldc_id);
4133 }
4134 
4135 /* -------------------------------------------------------------------------- */
4136 
4137 /*
4138  * The following functions process the incoming messages from vds
4139  */
4140 
4141 /*
4142  * Function:
4143  *      vdc_process_msg_thread()
4144  *
4145  * Description:
4146  *
4147  *	Main VDC message processing thread. Each vDisk instance
4148  * 	consists of a copy of this thread. This thread triggers
4149  * 	all the handshakes and data exchange with the server. It
4150  * 	also handles all channel resets
4151  *
4152  * Arguments:
4153  *      vdc     - soft state pointer for this instance of the device driver.
4154  *
4155  * Return Code:
4156  *      None
4157  */
4158 static void
4159 vdc_process_msg_thread(vdc_t *vdcp)
4160 {
4161 	int		status;
4162 	int		ctimeout;
4163 	timeout_id_t	tmid = 0;
4164 	clock_t		ldcup_timeout = 0;
4165 
4166 	mutex_enter(&vdcp->lock);
4167 
4168 	for (;;) {
4169 
4170 #define	Q(_s)	(vdcp->state == _s) ? #_s :
4171 		DMSG(vdcp, 3, "state = %d (%s)\n", vdcp->state,
4172 		    Q(VDC_STATE_INIT)
4173 		    Q(VDC_STATE_INIT_WAITING)
4174 		    Q(VDC_STATE_NEGOTIATE)
4175 		    Q(VDC_STATE_HANDLE_PENDING)
4176 		    Q(VDC_STATE_RUNNING)
4177 		    Q(VDC_STATE_RESETTING)
4178 		    Q(VDC_STATE_DETACH)
4179 		    "UNKNOWN");
4180 
4181 		switch (vdcp->state) {
4182 		case VDC_STATE_INIT:
4183 
4184 			/*
4185 			 * If requested, start a timeout to check if the
4186 			 * connection with vds is established in the
4187 			 * specified delay. If the timeout expires, we
4188 			 * will cancel any pending request.
4189 			 *
4190 			 * If some reset have occurred while establishing
4191 			 * the connection, we already have a timeout armed
4192 			 * and in that case we don't need to arm a new one.
4193 			 *
4194 			 * The same rule applies when there are multiple vds'.
4195 			 * If either a connection cannot be established or
4196 			 * the handshake times out, the connection thread will
4197 			 * try another server. The 'ctimeout' will report
4198 			 * back an error after it expires irrespective of
4199 			 * whether the vdisk is trying to connect to just
4200 			 * one or multiple servers.
4201 			 */
4202 			ctimeout = (vdc_timeout != 0)?
4203 			    vdc_timeout : vdcp->curr_server->ctimeout;
4204 
4205 			if (ctimeout != 0 && tmid == 0) {
4206 				tmid = timeout(vdc_connection_timeout, vdcp,
4207 				    ctimeout * drv_usectohz(MICROSEC));
4208 			}
4209 
4210 			/* Check if we are re-initializing repeatedly */
4211 			if (vdcp->hshake_cnt > vdc_hshake_retries &&
4212 			    vdcp->lifecycle != VDC_LC_ONLINE) {
4213 
4214 				DMSG(vdcp, 0, "[%d] too many handshakes,cnt=%d",
4215 				    vdcp->instance, vdcp->hshake_cnt);
4216 				cmn_err(CE_NOTE, "[%d] disk access failed.\n",
4217 				    vdcp->instance);
4218 				vdcp->state = VDC_STATE_DETACH;
4219 				break;
4220 			}
4221 
4222 			/* Switch to STATE_DETACH if drv is detaching */
4223 			if (vdcp->lifecycle == VDC_LC_DETACHING) {
4224 				vdcp->state = VDC_STATE_DETACH;
4225 				break;
4226 			}
4227 
4228 			/* Switch server */
4229 			if (vdcp->hshake_cnt > 0)
4230 				vdc_switch_server(vdcp);
4231 			vdcp->hshake_cnt++;
4232 
4233 			/* Bring up connection with vds via LDC */
4234 			status = vdc_start_ldc_connection(vdcp);
4235 			if (status != EINVAL) {
4236 				vdcp->state = VDC_STATE_INIT_WAITING;
4237 			}
4238 			break;
4239 
4240 		case VDC_STATE_INIT_WAITING:
4241 
4242 			/* if channel is UP, start negotiation */
4243 			if (vdcp->curr_server->ldc_state == LDC_UP) {
4244 				vdcp->state = VDC_STATE_NEGOTIATE;
4245 				break;
4246 			}
4247 
4248 			/* check if only one server exists */
4249 			if (vdcp->num_servers == 1) {
4250 				cv_wait(&vdcp->initwait_cv, &vdcp->lock);
4251 			} else {
4252 				/*
4253 				 * wait for LDC_UP, if it times out, switch
4254 				 * to another server.
4255 				 */
4256 				ldcup_timeout = ddi_get_lbolt() +
4257 				    (vdc_ldcup_timeout *
4258 				    drv_usectohz(MICROSEC));
4259 				status = cv_timedwait(&vdcp->initwait_cv,
4260 				    &vdcp->lock, ldcup_timeout);
4261 				if (status == -1 &&
4262 				    vdcp->state == VDC_STATE_INIT_WAITING &&
4263 				    vdcp->curr_server->ldc_state != LDC_UP) {
4264 					/* timed out & still waiting */
4265 					vdcp->state = VDC_STATE_INIT;
4266 					break;
4267 				}
4268 			}
4269 
4270 			if (vdcp->state != VDC_STATE_INIT_WAITING) {
4271 				DMSG(vdcp, 0,
4272 				    "state moved to %d out from under us...\n",
4273 				    vdcp->state);
4274 			}
4275 			break;
4276 
4277 		case VDC_STATE_NEGOTIATE:
4278 			switch (status = vdc_ver_negotiation(vdcp)) {
4279 			case 0:
4280 				break;
4281 			default:
4282 				DMSG(vdcp, 0, "ver negotiate failed (%d)..\n",
4283 				    status);
4284 				goto reset;
4285 			}
4286 
4287 			switch (status = vdc_attr_negotiation(vdcp)) {
4288 			case 0:
4289 				break;
4290 			default:
4291 				DMSG(vdcp, 0, "attr negotiate failed (%d)..\n",
4292 				    status);
4293 				goto reset;
4294 			}
4295 
4296 			switch (status = vdc_dring_negotiation(vdcp)) {
4297 			case 0:
4298 				break;
4299 			default:
4300 				DMSG(vdcp, 0, "dring negotiate failed (%d)..\n",
4301 				    status);
4302 				goto reset;
4303 			}
4304 
4305 			switch (status = vdc_rdx_exchange(vdcp)) {
4306 			case 0:
4307 				vdcp->state = VDC_STATE_HANDLE_PENDING;
4308 				goto done;
4309 			default:
4310 				DMSG(vdcp, 0, "RDX xchg failed ..(%d)\n",
4311 				    status);
4312 				goto reset;
4313 			}
4314 reset:
4315 			DMSG(vdcp, 0, "negotiation failed: resetting (%d)\n",
4316 			    status);
4317 			vdcp->state = VDC_STATE_RESETTING;
4318 			vdcp->self_reset = B_TRUE;
4319 done:
4320 			DMSG(vdcp, 0, "negotiation complete (state=0x%x)...\n",
4321 			    vdcp->state);
4322 			break;
4323 
4324 		case VDC_STATE_HANDLE_PENDING:
4325 
4326 			if (vdcp->ctimeout_reached) {
4327 				/*
4328 				 * The connection timeout had been reached so
4329 				 * pending requests have been cancelled. Now
4330 				 * that the connection is back we can reset
4331 				 * the timeout.
4332 				 */
4333 				ASSERT(vdcp->local_dring_backup == NULL);
4334 				ASSERT(tmid != 0);
4335 				tmid = 0;
4336 				vdcp->ctimeout_reached = B_FALSE;
4337 				vdcp->state = VDC_STATE_RUNNING;
4338 				DMSG(vdcp, 0, "[%d] connection to service "
4339 				    "domain is up", vdcp->instance);
4340 				break;
4341 			}
4342 
4343 			mutex_exit(&vdcp->lock);
4344 			if (tmid != 0) {
4345 				(void) untimeout(tmid);
4346 				tmid = 0;
4347 			}
4348 			status = vdc_resubmit_backup_dring(vdcp);
4349 			mutex_enter(&vdcp->lock);
4350 
4351 			if (status)
4352 				vdcp->state = VDC_STATE_RESETTING;
4353 			else
4354 				vdcp->state = VDC_STATE_RUNNING;
4355 
4356 			break;
4357 
4358 		/* enter running state */
4359 		case VDC_STATE_RUNNING:
4360 			/*
4361 			 * Signal anyone waiting for the connection
4362 			 * to come on line.
4363 			 */
4364 			vdcp->hshake_cnt = 0;
4365 			cv_broadcast(&vdcp->running_cv);
4366 
4367 			/* failfast has to been checked after reset */
4368 			cv_signal(&vdcp->failfast_cv);
4369 
4370 			/* ownership is lost during reset */
4371 			if (vdcp->ownership & VDC_OWNERSHIP_WANTED)
4372 				vdcp->ownership |= VDC_OWNERSHIP_RESET;
4373 			cv_signal(&vdcp->ownership_cv);
4374 
4375 			cmn_err(CE_CONT, "?vdisk@%d is online using "
4376 			    "ldc@%ld,%ld\n", vdcp->instance,
4377 			    vdcp->curr_server->ldc_id, vdcp->curr_server->id);
4378 
4379 			mutex_exit(&vdcp->lock);
4380 
4381 			for (;;) {
4382 				vio_msg_t msg;
4383 				status = vdc_wait_for_response(vdcp, &msg);
4384 				if (status) break;
4385 
4386 				DMSG(vdcp, 1, "[%d] new pkt(s) available\n",
4387 				    vdcp->instance);
4388 				status = vdc_process_data_msg(vdcp, &msg);
4389 				if (status) {
4390 					DMSG(vdcp, 1, "[%d] process_data_msg "
4391 					    "returned err=%d\n", vdcp->instance,
4392 					    status);
4393 					break;
4394 				}
4395 
4396 			}
4397 
4398 			mutex_enter(&vdcp->lock);
4399 
4400 			cmn_err(CE_CONT, "?vdisk@%d is offline\n",
4401 			    vdcp->instance);
4402 
4403 			vdcp->state = VDC_STATE_RESETTING;
4404 			vdcp->self_reset = B_TRUE;
4405 			break;
4406 
4407 		case VDC_STATE_RESETTING:
4408 			/*
4409 			 * When we reach this state, we either come from the
4410 			 * VDC_STATE_RUNNING state and we can have pending
4411 			 * request but no timeout is armed; or we come from
4412 			 * the VDC_STATE_INIT_WAITING, VDC_NEGOTIATE or
4413 			 * VDC_HANDLE_PENDING state and there is no pending
4414 			 * request or pending requests have already been copied
4415 			 * into the backup dring. So we can safely keep the
4416 			 * connection timeout armed while we are in this state.
4417 			 */
4418 
4419 			DMSG(vdcp, 0, "Initiating channel reset "
4420 			    "(pending = %d)\n", (int)vdcp->threads_pending);
4421 
4422 			if (vdcp->self_reset) {
4423 				DMSG(vdcp, 0,
4424 				    "[%d] calling stop_ldc_connection.\n",
4425 				    vdcp->instance);
4426 				status = vdc_stop_ldc_connection(vdcp);
4427 				vdcp->self_reset = B_FALSE;
4428 			}
4429 
4430 			/*
4431 			 * Wait for all threads currently waiting
4432 			 * for a free dring entry to use.
4433 			 */
4434 			while (vdcp->threads_pending) {
4435 				cv_broadcast(&vdcp->membind_cv);
4436 				cv_broadcast(&vdcp->dring_free_cv);
4437 				mutex_exit(&vdcp->lock);
4438 				/* give the waiters enough time to wake up */
4439 				delay(vdc_hz_min_ldc_delay);
4440 				mutex_enter(&vdcp->lock);
4441 			}
4442 
4443 			ASSERT(vdcp->threads_pending == 0);
4444 
4445 			/* Sanity check that no thread is receiving */
4446 			ASSERT(vdcp->read_state != VDC_READ_WAITING);
4447 
4448 			vdcp->read_state = VDC_READ_IDLE;
4449 
4450 			vdc_backup_local_dring(vdcp);
4451 
4452 			/* cleanup the old d-ring */
4453 			vdc_destroy_descriptor_ring(vdcp);
4454 
4455 			/* go and start again */
4456 			vdcp->state = VDC_STATE_INIT;
4457 
4458 			break;
4459 
4460 		case VDC_STATE_DETACH:
4461 			DMSG(vdcp, 0, "[%d] Reset thread exit cleanup ..\n",
4462 			    vdcp->instance);
4463 
4464 			/* cancel any pending timeout */
4465 			mutex_exit(&vdcp->lock);
4466 			if (tmid != 0) {
4467 				(void) untimeout(tmid);
4468 				tmid = 0;
4469 			}
4470 			mutex_enter(&vdcp->lock);
4471 
4472 			/*
4473 			 * Signal anyone waiting for connection
4474 			 * to come online
4475 			 */
4476 			cv_broadcast(&vdcp->running_cv);
4477 
4478 			while (vdcp->sync_op_pending) {
4479 				cv_signal(&vdcp->sync_pending_cv);
4480 				cv_signal(&vdcp->sync_blocked_cv);
4481 				mutex_exit(&vdcp->lock);
4482 				/* give the waiters enough time to wake up */
4483 				delay(vdc_hz_min_ldc_delay);
4484 				mutex_enter(&vdcp->lock);
4485 			}
4486 
4487 			mutex_exit(&vdcp->lock);
4488 
4489 			DMSG(vdcp, 0, "[%d] Msg processing thread exiting ..\n",
4490 			    vdcp->instance);
4491 			thread_exit();
4492 			break;
4493 		}
4494 	}
4495 }
4496 
4497 
4498 /*
4499  * Function:
4500  *	vdc_process_data_msg()
4501  *
4502  * Description:
4503  *	This function is called by the message processing thread each time
4504  *	a message with a msgtype of VIO_TYPE_DATA is received. It will either
4505  *	be an ACK or NACK from vds[1] which vdc handles as follows.
4506  *		ACK	- wake up the waiting thread
4507  *		NACK	- resend any messages necessary
4508  *
4509  *	[1] Although the message format allows it, vds should not send a
4510  *	    VIO_SUBTYPE_INFO message to vdc asking it to read data; if for
4511  *	    some bizarre reason it does, vdc will reset the connection.
4512  *
4513  * Arguments:
4514  *	vdc	- soft state pointer for this instance of the device driver.
4515  *	msg	- the LDC message sent by vds
4516  *
4517  * Return Code:
4518  *	0	- Success.
4519  *	> 0	- error value returned by LDC
4520  */
4521 static int
4522 vdc_process_data_msg(vdc_t *vdcp, vio_msg_t *msg)
4523 {
4524 	int			status = 0;
4525 	vio_dring_msg_t		*dring_msg;
4526 	vdc_local_desc_t	*ldep = NULL;
4527 	int			start, end;
4528 	int			idx;
4529 	int			op;
4530 
4531 	dring_msg = (vio_dring_msg_t *)msg;
4532 
4533 	ASSERT(msg->tag.vio_msgtype == VIO_TYPE_DATA);
4534 	ASSERT(vdcp != NULL);
4535 
4536 	mutex_enter(&vdcp->lock);
4537 
4538 	/*
4539 	 * Check to see if the message has bogus data
4540 	 */
4541 	idx = start = dring_msg->start_idx;
4542 	end = dring_msg->end_idx;
4543 	if ((start >= vdcp->dring_len) ||
4544 	    (end >= vdcp->dring_len) || (end < -1)) {
4545 		/*
4546 		 * Update the I/O statistics to indicate that an error ocurred.
4547 		 * No need to update the wait/run queues as no specific read or
4548 		 * write request is being completed in response to this 'msg'.
4549 		 */
4550 		VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4551 		DMSG(vdcp, 0, "[%d] Bogus ACK data : start %d, end %d\n",
4552 		    vdcp->instance, start, end);
4553 		mutex_exit(&vdcp->lock);
4554 		return (EINVAL);
4555 	}
4556 
4557 	/*
4558 	 * Verify that the sequence number is what vdc expects.
4559 	 */
4560 	switch (vdc_verify_seq_num(vdcp, dring_msg)) {
4561 	case VDC_SEQ_NUM_TODO:
4562 		break;	/* keep processing this message */
4563 	case VDC_SEQ_NUM_SKIP:
4564 		mutex_exit(&vdcp->lock);
4565 		return (0);
4566 	case VDC_SEQ_NUM_INVALID:
4567 		/*
4568 		 * Update the I/O statistics to indicate that an error ocurred.
4569 		 * No need to update the wait/run queues as no specific read or
4570 		 * write request is being completed in response to this 'msg'.
4571 		 */
4572 		VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4573 		DMSG(vdcp, 0, "[%d] invalid seqno\n", vdcp->instance);
4574 		mutex_exit(&vdcp->lock);
4575 		return (ENXIO);
4576 	}
4577 
4578 	if (msg->tag.vio_subtype == VIO_SUBTYPE_NACK) {
4579 		/*
4580 		 * Update the I/O statistics to indicate that an error ocurred.
4581 		 *
4582 		 * We need to update the run queue if a read or write request
4583 		 * is being NACKed - otherwise there will appear to be an
4584 		 * indefinite outstanding request and statistics reported by
4585 		 * iostat(1M) will be incorrect. The transaction will be
4586 		 * resubmitted from the backup DRing following the reset
4587 		 * and the wait/run queues will be entered again.
4588 		 */
4589 		ldep = &vdcp->local_dring[idx];
4590 		op = ldep->operation;
4591 		if ((op == VD_OP_BREAD) || (op == VD_OP_BWRITE)) {
4592 			DTRACE_IO1(done, buf_t *, ldep->cb_arg);
4593 			VD_KSTAT_RUNQ_EXIT(vdcp);
4594 		}
4595 		VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4596 		VDC_DUMP_DRING_MSG(dring_msg);
4597 		DMSG(vdcp, 0, "[%d] DATA NACK\n", vdcp->instance);
4598 		mutex_exit(&vdcp->lock);
4599 		return (EIO);
4600 
4601 	} else if (msg->tag.vio_subtype == VIO_SUBTYPE_INFO) {
4602 		/*
4603 		 * Update the I/O statistics to indicate that an error occurred.
4604 		 * No need to update the wait/run queues as no specific read or
4605 		 * write request is being completed in response to this 'msg'.
4606 		 */
4607 		VD_UPDATE_ERR_STATS(vdcp, vd_protoerrs);
4608 		mutex_exit(&vdcp->lock);
4609 		return (EPROTO);
4610 	}
4611 
4612 	DMSG(vdcp, 1, ": start %d end %d\n", start, end);
4613 	ASSERT(start == end);
4614 
4615 	ldep = &vdcp->local_dring[idx];
4616 
4617 	DMSG(vdcp, 1, ": state 0x%x - cb_type 0x%x\n",
4618 	    ldep->dep->hdr.dstate, ldep->cb_type);
4619 
4620 	if (ldep->dep->hdr.dstate == VIO_DESC_DONE) {
4621 		struct buf *bufp;
4622 
4623 		switch (ldep->cb_type) {
4624 		case CB_SYNC:
4625 			ASSERT(vdcp->sync_op_pending);
4626 
4627 			status = vdc_depopulate_descriptor(vdcp, idx);
4628 			vdcp->sync_op_status = status;
4629 			vdcp->sync_op_pending = B_FALSE;
4630 			cv_signal(&vdcp->sync_pending_cv);
4631 			break;
4632 
4633 		case CB_STRATEGY:
4634 			bufp = ldep->cb_arg;
4635 			ASSERT(bufp != NULL);
4636 			bufp->b_resid =
4637 			    bufp->b_bcount - ldep->dep->payload.nbytes;
4638 			status = ldep->dep->payload.status; /* Future:ntoh */
4639 			if (status != 0) {
4640 				DMSG(vdcp, 1, "strategy status=%d\n", status);
4641 				VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4642 				bioerror(bufp, status);
4643 			}
4644 
4645 			(void) vdc_depopulate_descriptor(vdcp, idx);
4646 
4647 			DMSG(vdcp, 1,
4648 			    "strategy complete req=%ld bytes resp=%ld bytes\n",
4649 			    bufp->b_bcount, ldep->dep->payload.nbytes);
4650 
4651 			if (status != 0 && vdcp->failfast_interval != 0) {
4652 				/*
4653 				 * The I/O has failed and failfast is enabled.
4654 				 * We need the failfast thread to check if the
4655 				 * failure is due to a reservation conflict.
4656 				 */
4657 				(void) vdc_failfast_io_queue(vdcp, bufp);
4658 			} else {
4659 				if (status == 0) {
4660 					op = (bufp->b_flags & B_READ) ?
4661 					    VD_OP_BREAD : VD_OP_BWRITE;
4662 					VD_UPDATE_IO_STATS(vdcp, op,
4663 					    ldep->dep->payload.nbytes);
4664 				}
4665 				VD_KSTAT_RUNQ_EXIT(vdcp);
4666 				DTRACE_IO1(done, buf_t *, bufp);
4667 				biodone(bufp);
4668 			}
4669 			break;
4670 
4671 		default:
4672 			ASSERT(0);
4673 		}
4674 	}
4675 
4676 	/* let the arrival signal propogate */
4677 	mutex_exit(&vdcp->lock);
4678 
4679 	/* probe gives the count of how many entries were processed */
4680 	DTRACE_PROBE2(processed, int, 1, vdc_t *, vdcp);
4681 
4682 	return (0);
4683 }
4684 
4685 
4686 /*
4687  * Function:
4688  *	vdc_handle_ver_msg()
4689  *
4690  * Description:
4691  *
4692  * Arguments:
4693  *	vdc	- soft state pointer for this instance of the device driver.
4694  *	ver_msg	- LDC message sent by vDisk server
4695  *
4696  * Return Code:
4697  *	0	- Success
4698  */
4699 static int
4700 vdc_handle_ver_msg(vdc_t *vdc, vio_ver_msg_t *ver_msg)
4701 {
4702 	int status = 0;
4703 
4704 	ASSERT(vdc != NULL);
4705 	ASSERT(mutex_owned(&vdc->lock));
4706 
4707 	if (ver_msg->tag.vio_subtype_env != VIO_VER_INFO) {
4708 		return (EPROTO);
4709 	}
4710 
4711 	if (ver_msg->dev_class != VDEV_DISK_SERVER) {
4712 		return (EINVAL);
4713 	}
4714 
4715 	switch (ver_msg->tag.vio_subtype) {
4716 	case VIO_SUBTYPE_ACK:
4717 		/*
4718 		 * We check to see if the version returned is indeed supported
4719 		 * (The server may have also adjusted the minor number downwards
4720 		 * and if so 'ver_msg' will contain the actual version agreed)
4721 		 */
4722 		if (vdc_is_supported_version(ver_msg)) {
4723 			vdc->ver.major = ver_msg->ver_major;
4724 			vdc->ver.minor = ver_msg->ver_minor;
4725 			ASSERT(vdc->ver.major > 0);
4726 		} else {
4727 			status = EPROTO;
4728 		}
4729 		break;
4730 
4731 	case VIO_SUBTYPE_NACK:
4732 		/*
4733 		 * call vdc_is_supported_version() which will return the next
4734 		 * supported version (if any) in 'ver_msg'
4735 		 */
4736 		(void) vdc_is_supported_version(ver_msg);
4737 		if (ver_msg->ver_major > 0) {
4738 			size_t len = sizeof (*ver_msg);
4739 
4740 			ASSERT(vdc->ver.major > 0);
4741 
4742 			/* reset the necessary fields and resend */
4743 			ver_msg->tag.vio_subtype = VIO_SUBTYPE_INFO;
4744 			ver_msg->dev_class = VDEV_DISK;
4745 
4746 			status = vdc_send(vdc, (caddr_t)ver_msg, &len);
4747 			DMSG(vdc, 0, "[%d] Resend VER info (LDC status = %d)\n",
4748 			    vdc->instance, status);
4749 			if (len != sizeof (*ver_msg))
4750 				status = EBADMSG;
4751 		} else {
4752 			DMSG(vdc, 0, "[%d] No common version with vDisk server",
4753 			    vdc->instance);
4754 			status = ENOTSUP;
4755 		}
4756 
4757 		break;
4758 	case VIO_SUBTYPE_INFO:
4759 		/*
4760 		 * Handle the case where vds starts handshake
4761 		 * (for now only vdc is the instigator)
4762 		 */
4763 		status = ENOTSUP;
4764 		break;
4765 
4766 	default:
4767 		status = EINVAL;
4768 		break;
4769 	}
4770 
4771 	return (status);
4772 }
4773 
4774 /*
4775  * Function:
4776  *	vdc_handle_attr_msg()
4777  *
4778  * Description:
4779  *
4780  * Arguments:
4781  *	vdc	- soft state pointer for this instance of the device driver.
4782  *	attr_msg	- LDC message sent by vDisk server
4783  *
4784  * Return Code:
4785  *	0	- Success
4786  */
4787 static int
4788 vdc_handle_attr_msg(vdc_t *vdc, vd_attr_msg_t *attr_msg)
4789 {
4790 	int status = 0;
4791 
4792 	ASSERT(vdc != NULL);
4793 	ASSERT(mutex_owned(&vdc->lock));
4794 
4795 	if (attr_msg->tag.vio_subtype_env != VIO_ATTR_INFO) {
4796 		return (EPROTO);
4797 	}
4798 
4799 	switch (attr_msg->tag.vio_subtype) {
4800 	case VIO_SUBTYPE_ACK:
4801 		/*
4802 		 * We now verify the attributes sent by vds.
4803 		 */
4804 		if (attr_msg->vdisk_size == 0) {
4805 			DMSG(vdc, 0, "[%d] Invalid disk size from vds",
4806 			    vdc->instance);
4807 			status = EINVAL;
4808 			break;
4809 		}
4810 
4811 		if (attr_msg->max_xfer_sz == 0) {
4812 			DMSG(vdc, 0, "[%d] Invalid transfer size from vds",
4813 			    vdc->instance);
4814 			status = EINVAL;
4815 			break;
4816 		}
4817 
4818 		if (attr_msg->vdisk_size == VD_SIZE_UNKNOWN) {
4819 			DMSG(vdc, 0, "[%d] Unknown disk size from vds",
4820 			    vdc->instance);
4821 			attr_msg->vdisk_size = 0;
4822 		}
4823 
4824 		/*
4825 		 * If the disk size is already set check that it hasn't changed.
4826 		 */
4827 		if ((vdc->vdisk_size != 0) && (attr_msg->vdisk_size != 0) &&
4828 		    (vdc->vdisk_size != attr_msg->vdisk_size)) {
4829 			DMSG(vdc, 0, "[%d] Different disk size from vds "
4830 			    "(old=0x%lx - new=0x%lx", vdc->instance,
4831 			    vdc->vdisk_size, attr_msg->vdisk_size)
4832 			status = EINVAL;
4833 			break;
4834 		}
4835 
4836 		vdc->vdisk_size = attr_msg->vdisk_size;
4837 		vdc->vdisk_type = attr_msg->vdisk_type;
4838 		vdc->operations = attr_msg->operations;
4839 		if (vio_ver_is_supported(vdc->ver, 1, 1))
4840 			vdc->vdisk_media = attr_msg->vdisk_media;
4841 		else
4842 			vdc->vdisk_media = 0;
4843 
4844 		DMSG(vdc, 0, "[%d] max_xfer_sz: sent %lx acked %lx\n",
4845 		    vdc->instance, vdc->max_xfer_sz, attr_msg->max_xfer_sz);
4846 		DMSG(vdc, 0, "[%d] vdisk_block_size: sent %lx acked %x\n",
4847 		    vdc->instance, vdc->block_size,
4848 		    attr_msg->vdisk_block_size);
4849 
4850 		/*
4851 		 * We don't know at compile time what the vDisk server will
4852 		 * think are good values but we apply a large (arbitrary)
4853 		 * upper bound to prevent memory exhaustion in vdc if it was
4854 		 * allocating a DRing based of huge values sent by the server.
4855 		 * We probably will never exceed this except if the message
4856 		 * was garbage.
4857 		 */
4858 		if ((attr_msg->max_xfer_sz * attr_msg->vdisk_block_size) <=
4859 		    (PAGESIZE * DEV_BSIZE)) {
4860 			vdc->max_xfer_sz = attr_msg->max_xfer_sz;
4861 			vdc->block_size = attr_msg->vdisk_block_size;
4862 		} else {
4863 			DMSG(vdc, 0, "[%d] vds block transfer size too big;"
4864 			    " using max supported by vdc", vdc->instance);
4865 		}
4866 
4867 		if ((attr_msg->xfer_mode != VIO_DRING_MODE_V1_0) ||
4868 		    (attr_msg->vdisk_size > INT64_MAX) ||
4869 		    (attr_msg->operations == 0) ||
4870 		    (attr_msg->vdisk_type > VD_DISK_TYPE_DISK)) {
4871 			DMSG(vdc, 0, "[%d] Invalid attributes from vds",
4872 			    vdc->instance);
4873 			status = EINVAL;
4874 			break;
4875 		}
4876 
4877 		/*
4878 		 * Now that we have received all attributes we can create a
4879 		 * fake geometry for the disk.
4880 		 */
4881 		vdc_create_fake_geometry(vdc);
4882 		break;
4883 
4884 	case VIO_SUBTYPE_NACK:
4885 		/*
4886 		 * vds could not handle the attributes we sent so we
4887 		 * stop negotiating.
4888 		 */
4889 		status = EPROTO;
4890 		break;
4891 
4892 	case VIO_SUBTYPE_INFO:
4893 		/*
4894 		 * Handle the case where vds starts the handshake
4895 		 * (for now; vdc is the only supported instigatior)
4896 		 */
4897 		status = ENOTSUP;
4898 		break;
4899 
4900 	default:
4901 		status = ENOTSUP;
4902 		break;
4903 	}
4904 
4905 	return (status);
4906 }
4907 
4908 /*
4909  * Function:
4910  *	vdc_handle_dring_reg_msg()
4911  *
4912  * Description:
4913  *
4914  * Arguments:
4915  *	vdc		- soft state pointer for this instance of the driver.
4916  *	dring_msg	- LDC message sent by vDisk server
4917  *
4918  * Return Code:
4919  *	0	- Success
4920  */
4921 static int
4922 vdc_handle_dring_reg_msg(vdc_t *vdc, vio_dring_reg_msg_t *dring_msg)
4923 {
4924 	int		status = 0;
4925 
4926 	ASSERT(vdc != NULL);
4927 	ASSERT(mutex_owned(&vdc->lock));
4928 
4929 	if (dring_msg->tag.vio_subtype_env != VIO_DRING_REG) {
4930 		return (EPROTO);
4931 	}
4932 
4933 	switch (dring_msg->tag.vio_subtype) {
4934 	case VIO_SUBTYPE_ACK:
4935 		/* save the received dring_ident */
4936 		vdc->dring_ident = dring_msg->dring_ident;
4937 		DMSG(vdc, 0, "[%d] Received dring ident=0x%lx\n",
4938 		    vdc->instance, vdc->dring_ident);
4939 		break;
4940 
4941 	case VIO_SUBTYPE_NACK:
4942 		/*
4943 		 * vds could not handle the DRing info we sent so we
4944 		 * stop negotiating.
4945 		 */
4946 		DMSG(vdc, 0, "[%d] server could not register DRing\n",
4947 		    vdc->instance);
4948 		status = EPROTO;
4949 		break;
4950 
4951 	case VIO_SUBTYPE_INFO:
4952 		/*
4953 		 * Handle the case where vds starts handshake
4954 		 * (for now only vdc is the instigatior)
4955 		 */
4956 		status = ENOTSUP;
4957 		break;
4958 	default:
4959 		status = ENOTSUP;
4960 	}
4961 
4962 	return (status);
4963 }
4964 
4965 /*
4966  * Function:
4967  *	vdc_verify_seq_num()
4968  *
4969  * Description:
4970  *	This functions verifies that the sequence number sent back by the vDisk
4971  *	server with the latest message is what is expected (i.e. it is greater
4972  *	than the last seq num sent by the vDisk server and less than or equal
4973  *	to the last seq num generated by vdc).
4974  *
4975  *	It then checks the request ID to see if any requests need processing
4976  *	in the DRing.
4977  *
4978  * Arguments:
4979  *	vdc		- soft state pointer for this instance of the driver.
4980  *	dring_msg	- pointer to the LDC message sent by vds
4981  *
4982  * Return Code:
4983  *	VDC_SEQ_NUM_TODO	- Message needs to be processed
4984  *	VDC_SEQ_NUM_SKIP	- Message has already been processed
4985  *	VDC_SEQ_NUM_INVALID	- The seq numbers are so out of sync,
4986  *				  vdc cannot deal with them
4987  */
4988 static int
4989 vdc_verify_seq_num(vdc_t *vdc, vio_dring_msg_t *dring_msg)
4990 {
4991 	ASSERT(vdc != NULL);
4992 	ASSERT(dring_msg != NULL);
4993 	ASSERT(mutex_owned(&vdc->lock));
4994 
4995 	/*
4996 	 * Check to see if the messages were responded to in the correct
4997 	 * order by vds.
4998 	 */
4999 	if ((dring_msg->seq_num <= vdc->seq_num_reply) ||
5000 	    (dring_msg->seq_num > vdc->seq_num)) {
5001 		DMSG(vdc, 0, "?[%d] Bogus sequence_number %lu: "
5002 		    "%lu > expected <= %lu (last proc req %lu sent %lu)\n",
5003 		    vdc->instance, dring_msg->seq_num,
5004 		    vdc->seq_num_reply, vdc->seq_num,
5005 		    vdc->req_id_proc, vdc->req_id);
5006 		return (VDC_SEQ_NUM_INVALID);
5007 	}
5008 	vdc->seq_num_reply = dring_msg->seq_num;
5009 
5010 	if (vdc->req_id_proc < vdc->req_id)
5011 		return (VDC_SEQ_NUM_TODO);
5012 	else
5013 		return (VDC_SEQ_NUM_SKIP);
5014 }
5015 
5016 
5017 /*
5018  * Function:
5019  *	vdc_is_supported_version()
5020  *
5021  * Description:
5022  *	This routine checks if the major/minor version numbers specified in
5023  *	'ver_msg' are supported. If not it finds the next version that is
5024  *	in the supported version list 'vdc_version[]' and sets the fields in
5025  *	'ver_msg' to those values
5026  *
5027  * Arguments:
5028  *	ver_msg	- LDC message sent by vDisk server
5029  *
5030  * Return Code:
5031  *	B_TRUE	- Success
5032  *	B_FALSE	- Version not supported
5033  */
5034 static boolean_t
5035 vdc_is_supported_version(vio_ver_msg_t *ver_msg)
5036 {
5037 	int vdc_num_versions = sizeof (vdc_version) / sizeof (vdc_version[0]);
5038 
5039 	for (int i = 0; i < vdc_num_versions; i++) {
5040 		ASSERT(vdc_version[i].major > 0);
5041 		ASSERT((i == 0) ||
5042 		    (vdc_version[i].major < vdc_version[i-1].major));
5043 
5044 		/*
5045 		 * If the major versions match, adjust the minor version, if
5046 		 * necessary, down to the highest value supported by this
5047 		 * client. The server should support all minor versions lower
5048 		 * than the value it sent
5049 		 */
5050 		if (ver_msg->ver_major == vdc_version[i].major) {
5051 			if (ver_msg->ver_minor > vdc_version[i].minor) {
5052 				DMSGX(0,
5053 				    "Adjusting minor version from %u to %u",
5054 				    ver_msg->ver_minor, vdc_version[i].minor);
5055 				ver_msg->ver_minor = vdc_version[i].minor;
5056 			}
5057 			return (B_TRUE);
5058 		}
5059 
5060 		/*
5061 		 * If the message contains a higher major version number, set
5062 		 * the message's major/minor versions to the current values
5063 		 * and return false, so this message will get resent with
5064 		 * these values, and the server will potentially try again
5065 		 * with the same or a lower version
5066 		 */
5067 		if (ver_msg->ver_major > vdc_version[i].major) {
5068 			ver_msg->ver_major = vdc_version[i].major;
5069 			ver_msg->ver_minor = vdc_version[i].minor;
5070 			DMSGX(0, "Suggesting major/minor (0x%x/0x%x)\n",
5071 			    ver_msg->ver_major, ver_msg->ver_minor);
5072 
5073 			return (B_FALSE);
5074 		}
5075 
5076 		/*
5077 		 * Otherwise, the message's major version is less than the
5078 		 * current major version, so continue the loop to the next
5079 		 * (lower) supported version
5080 		 */
5081 	}
5082 
5083 	/*
5084 	 * No common version was found; "ground" the version pair in the
5085 	 * message to terminate negotiation
5086 	 */
5087 	ver_msg->ver_major = 0;
5088 	ver_msg->ver_minor = 0;
5089 
5090 	return (B_FALSE);
5091 }
5092 /* -------------------------------------------------------------------------- */
5093 
5094 /*
5095  * DKIO(7) support
5096  */
5097 
5098 typedef struct vdc_dk_arg {
5099 	struct dk_callback	dkc;
5100 	int			mode;
5101 	dev_t			dev;
5102 	vdc_t			*vdc;
5103 } vdc_dk_arg_t;
5104 
5105 /*
5106  * Function:
5107  * 	vdc_dkio_flush_cb()
5108  *
5109  * Description:
5110  *	This routine is a callback for DKIOCFLUSHWRITECACHE which can be called
5111  *	by kernel code.
5112  *
5113  * Arguments:
5114  *	arg	- a pointer to a vdc_dk_arg_t structure.
5115  */
5116 void
5117 vdc_dkio_flush_cb(void *arg)
5118 {
5119 	struct vdc_dk_arg	*dk_arg = (struct vdc_dk_arg *)arg;
5120 	struct dk_callback	*dkc = NULL;
5121 	vdc_t			*vdc = NULL;
5122 	int			rv;
5123 
5124 	if (dk_arg == NULL) {
5125 		cmn_err(CE_NOTE, "?[Unk] DKIOCFLUSHWRITECACHE arg is NULL\n");
5126 		return;
5127 	}
5128 	dkc = &dk_arg->dkc;
5129 	vdc = dk_arg->vdc;
5130 	ASSERT(vdc != NULL);
5131 
5132 	rv = vdc_do_sync_op(vdc, VD_OP_FLUSH, NULL, 0,
5133 	    VDCPART(dk_arg->dev), 0, CB_SYNC, 0, VIO_both_dir, B_TRUE);
5134 	if (rv != 0) {
5135 		DMSG(vdc, 0, "[%d] DKIOCFLUSHWRITECACHE failed %d : model %x\n",
5136 		    vdc->instance, rv,
5137 		    ddi_model_convert_from(dk_arg->mode & FMODELS));
5138 	}
5139 
5140 	/*
5141 	 * Trigger the call back to notify the caller the the ioctl call has
5142 	 * been completed.
5143 	 */
5144 	if ((dk_arg->mode & FKIOCTL) &&
5145 	    (dkc != NULL) &&
5146 	    (dkc->dkc_callback != NULL)) {
5147 		ASSERT(dkc->dkc_cookie != NULL);
5148 		(*dkc->dkc_callback)(dkc->dkc_cookie, rv);
5149 	}
5150 
5151 	/* Indicate that one less DKIO write flush is outstanding */
5152 	mutex_enter(&vdc->lock);
5153 	vdc->dkio_flush_pending--;
5154 	ASSERT(vdc->dkio_flush_pending >= 0);
5155 	mutex_exit(&vdc->lock);
5156 
5157 	/* free the mem that was allocated when the callback was dispatched */
5158 	kmem_free(arg, sizeof (vdc_dk_arg_t));
5159 }
5160 
5161 /*
5162  * Function:
5163  * 	vdc_dkio_gapart()
5164  *
5165  * Description:
5166  *	This function implements the DKIOCGAPART ioctl.
5167  *
5168  * Arguments:
5169  *	vdc	- soft state pointer
5170  *	arg	- a pointer to a dk_map[NDKMAP] or dk_map32[NDKMAP] structure
5171  *	flag	- ioctl flags
5172  */
5173 static int
5174 vdc_dkio_gapart(vdc_t *vdc, caddr_t arg, int flag)
5175 {
5176 	struct dk_geom *geom;
5177 	struct vtoc *vtoc;
5178 	union {
5179 		struct dk_map map[NDKMAP];
5180 		struct dk_map32 map32[NDKMAP];
5181 	} data;
5182 	int i, rv, size;
5183 
5184 	mutex_enter(&vdc->lock);
5185 
5186 	if ((rv = vdc_validate_geometry(vdc)) != 0) {
5187 		mutex_exit(&vdc->lock);
5188 		return (rv);
5189 	}
5190 
5191 	vtoc = vdc->vtoc;
5192 	geom = vdc->geom;
5193 
5194 	if (ddi_model_convert_from(flag & FMODELS) == DDI_MODEL_ILP32) {
5195 
5196 		for (i = 0; i < vtoc->v_nparts; i++) {
5197 			data.map32[i].dkl_cylno = vtoc->v_part[i].p_start /
5198 			    (geom->dkg_nhead * geom->dkg_nsect);
5199 			data.map32[i].dkl_nblk = vtoc->v_part[i].p_size;
5200 		}
5201 		size = NDKMAP * sizeof (struct dk_map32);
5202 
5203 	} else {
5204 
5205 		for (i = 0; i < vtoc->v_nparts; i++) {
5206 			data.map[i].dkl_cylno = vtoc->v_part[i].p_start /
5207 			    (geom->dkg_nhead * geom->dkg_nsect);
5208 			data.map[i].dkl_nblk = vtoc->v_part[i].p_size;
5209 		}
5210 		size = NDKMAP * sizeof (struct dk_map);
5211 
5212 	}
5213 
5214 	mutex_exit(&vdc->lock);
5215 
5216 	if (ddi_copyout(&data, arg, size, flag) != 0)
5217 		return (EFAULT);
5218 
5219 	return (0);
5220 }
5221 
5222 /*
5223  * Function:
5224  * 	vdc_dkio_partition()
5225  *
5226  * Description:
5227  *	This function implements the DKIOCPARTITION ioctl.
5228  *
5229  * Arguments:
5230  *	vdc	- soft state pointer
5231  *	arg	- a pointer to a struct partition64 structure
5232  *	flag	- ioctl flags
5233  */
5234 static int
5235 vdc_dkio_partition(vdc_t *vdc, caddr_t arg, int flag)
5236 {
5237 	struct partition64 p64;
5238 	efi_gpt_t *gpt;
5239 	efi_gpe_t *gpe;
5240 	vd_efi_dev_t edev;
5241 	uint_t partno;
5242 	int rv;
5243 
5244 	if (ddi_copyin(arg, &p64, sizeof (struct partition64), flag)) {
5245 		return (EFAULT);
5246 	}
5247 
5248 	VD_EFI_DEV_SET(edev, vdc, vd_process_efi_ioctl);
5249 
5250 	if ((rv = vd_efi_alloc_and_read(&edev, &gpt, &gpe)) != 0) {
5251 		return (rv);
5252 	}
5253 
5254 	partno = p64.p_partno;
5255 
5256 	if (partno >= gpt->efi_gpt_NumberOfPartitionEntries) {
5257 		vd_efi_free(&edev, gpt, gpe);
5258 		return (ESRCH);
5259 	}
5260 
5261 	bcopy(&gpe[partno].efi_gpe_PartitionTypeGUID, &p64.p_type,
5262 	    sizeof (struct uuid));
5263 	p64.p_start = gpe[partno].efi_gpe_StartingLBA;
5264 	p64.p_size = gpe[partno].efi_gpe_EndingLBA - p64.p_start + 1;
5265 
5266 	if (ddi_copyout(&p64, arg, sizeof (struct partition64), flag)) {
5267 		vd_efi_free(&edev, gpt, gpe);
5268 		return (EFAULT);
5269 	}
5270 
5271 	vd_efi_free(&edev, gpt, gpe);
5272 	return (0);
5273 }
5274 
5275 /*
5276  * Function:
5277  * 	vdc_dioctl_rwcmd()
5278  *
5279  * Description:
5280  *	This function implements the DIOCTL_RWCMD ioctl. This ioctl is used
5281  *	for DKC_DIRECT disks to read or write at an absolute disk offset.
5282  *
5283  * Arguments:
5284  *	dev	- device
5285  *	arg	- a pointer to a dadkio_rwcmd or dadkio_rwcmd32 structure
5286  *	flag	- ioctl flags
5287  */
5288 static int
5289 vdc_dioctl_rwcmd(dev_t dev, caddr_t arg, int flag)
5290 {
5291 	struct dadkio_rwcmd32 rwcmd32;
5292 	struct dadkio_rwcmd rwcmd;
5293 	struct iovec aiov;
5294 	struct uio auio;
5295 	int rw, status;
5296 	struct buf *buf;
5297 
5298 	if (ddi_model_convert_from(flag & FMODELS) == DDI_MODEL_ILP32) {
5299 		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd32,
5300 		    sizeof (struct dadkio_rwcmd32), flag)) {
5301 			return (EFAULT);
5302 		}
5303 		rwcmd.cmd = rwcmd32.cmd;
5304 		rwcmd.flags = rwcmd32.flags;
5305 		rwcmd.blkaddr = (daddr_t)rwcmd32.blkaddr;
5306 		rwcmd.buflen = rwcmd32.buflen;
5307 		rwcmd.bufaddr = (caddr_t)(uintptr_t)rwcmd32.bufaddr;
5308 	} else {
5309 		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd,
5310 		    sizeof (struct dadkio_rwcmd), flag)) {
5311 			return (EFAULT);
5312 		}
5313 	}
5314 
5315 	switch (rwcmd.cmd) {
5316 	case DADKIO_RWCMD_READ:
5317 		rw = B_READ;
5318 		break;
5319 	case DADKIO_RWCMD_WRITE:
5320 		rw = B_WRITE;
5321 		break;
5322 	default:
5323 		return (EINVAL);
5324 	}
5325 
5326 	bzero((caddr_t)&aiov, sizeof (struct iovec));
5327 	aiov.iov_base   = rwcmd.bufaddr;
5328 	aiov.iov_len    = rwcmd.buflen;
5329 
5330 	bzero((caddr_t)&auio, sizeof (struct uio));
5331 	auio.uio_iov    = &aiov;
5332 	auio.uio_iovcnt = 1;
5333 	auio.uio_loffset = rwcmd.blkaddr * DEV_BSIZE;
5334 	auio.uio_resid  = rwcmd.buflen;
5335 	auio.uio_segflg = flag & FKIOCTL ? UIO_SYSSPACE : UIO_USERSPACE;
5336 
5337 	buf = kmem_alloc(sizeof (buf_t), KM_SLEEP);
5338 	bioinit(buf);
5339 	/*
5340 	 * We use the private field of buf to specify that this is an
5341 	 * I/O using an absolute offset.
5342 	 */
5343 	buf->b_private = (void *)VD_SLICE_NONE;
5344 
5345 	status = physio(vdc_strategy, buf, dev, rw, vdc_min, &auio);
5346 
5347 	biofini(buf);
5348 	kmem_free(buf, sizeof (buf_t));
5349 
5350 	return (status);
5351 }
5352 
5353 /*
5354  * Allocate a buffer for a VD_OP_SCSICMD operation. The size of the allocated
5355  * buffer is returned in alloc_len.
5356  */
5357 static vd_scsi_t *
5358 vdc_scsi_alloc(int cdb_len, int sense_len, int datain_len, int dataout_len,
5359     int *alloc_len)
5360 {
5361 	vd_scsi_t *vd_scsi;
5362 	int vd_scsi_len = VD_SCSI_SIZE;
5363 
5364 	vd_scsi_len += P2ROUNDUP(cdb_len, sizeof (uint64_t));
5365 	vd_scsi_len += P2ROUNDUP(sense_len, sizeof (uint64_t));
5366 	vd_scsi_len += P2ROUNDUP(datain_len, sizeof (uint64_t));
5367 	vd_scsi_len += P2ROUNDUP(dataout_len, sizeof (uint64_t));
5368 
5369 	ASSERT(vd_scsi_len % sizeof (uint64_t) == 0);
5370 
5371 	vd_scsi = kmem_zalloc(vd_scsi_len, KM_SLEEP);
5372 
5373 	vd_scsi->cdb_len = cdb_len;
5374 	vd_scsi->sense_len = sense_len;
5375 	vd_scsi->datain_len = datain_len;
5376 	vd_scsi->dataout_len = dataout_len;
5377 
5378 	*alloc_len = vd_scsi_len;
5379 
5380 	return (vd_scsi);
5381 }
5382 
5383 /*
5384  * Convert the status of a SCSI command to a Solaris return code.
5385  *
5386  * Arguments:
5387  *	vd_scsi		- The SCSI operation buffer.
5388  *	log_error	- indicate if an error message should be logged.
5389  *
5390  * Note that our SCSI error messages are rather primitive for the moment
5391  * and could be improved by decoding some data like the SCSI command and
5392  * the sense key.
5393  *
5394  * Return value:
5395  *	0		- Status is good.
5396  *	EACCES		- Status reports a reservation conflict.
5397  *	ENOTSUP		- Status reports a check condition and sense key
5398  *			  reports an illegal request.
5399  *	EIO		- Any other status.
5400  */
5401 static int
5402 vdc_scsi_status(vdc_t *vdc, vd_scsi_t *vd_scsi, boolean_t log_error)
5403 {
5404 	int rv;
5405 	char path_str[MAXPATHLEN];
5406 	char panic_str[VDC_RESV_CONFLICT_FMT_LEN + MAXPATHLEN];
5407 	union scsi_cdb *cdb;
5408 	struct scsi_extended_sense *sense;
5409 
5410 	if (vd_scsi->cmd_status == STATUS_GOOD)
5411 		/* no error */
5412 		return (0);
5413 
5414 	/* when the tunable vdc_scsi_log_error is true we log all errors */
5415 	if (vdc_scsi_log_error)
5416 		log_error = B_TRUE;
5417 
5418 	if (log_error) {
5419 		cmn_err(CE_WARN, "%s (vdc%d):\tError for Command: 0x%x)\n",
5420 		    ddi_pathname(vdc->dip, path_str), vdc->instance,
5421 		    GETCMD(VD_SCSI_DATA_CDB(vd_scsi)));
5422 	}
5423 
5424 	/* default returned value */
5425 	rv = EIO;
5426 
5427 	switch (vd_scsi->cmd_status) {
5428 
5429 	case STATUS_CHECK:
5430 	case STATUS_TERMINATED:
5431 		if (log_error)
5432 			cmn_err(CE_CONT, "\tCheck Condition Error\n");
5433 
5434 		/* check sense buffer */
5435 		if (vd_scsi->sense_len == 0 ||
5436 		    vd_scsi->sense_status != STATUS_GOOD) {
5437 			if (log_error)
5438 				cmn_err(CE_CONT, "\tNo Sense Data Available\n");
5439 			break;
5440 		}
5441 
5442 		sense = VD_SCSI_DATA_SENSE(vd_scsi);
5443 
5444 		if (log_error) {
5445 			cmn_err(CE_CONT, "\tSense Key:  0x%x\n"
5446 			    "\tASC: 0x%x, ASCQ: 0x%x\n",
5447 			    scsi_sense_key((uint8_t *)sense),
5448 			    scsi_sense_asc((uint8_t *)sense),
5449 			    scsi_sense_ascq((uint8_t *)sense));
5450 		}
5451 
5452 		if (scsi_sense_key((uint8_t *)sense) == KEY_ILLEGAL_REQUEST)
5453 			rv = ENOTSUP;
5454 		break;
5455 
5456 	case STATUS_BUSY:
5457 		if (log_error)
5458 			cmn_err(CE_NOTE, "\tDevice Busy\n");
5459 		break;
5460 
5461 	case STATUS_RESERVATION_CONFLICT:
5462 		/*
5463 		 * If the command was PERSISTENT_RESERVATION_[IN|OUT] then
5464 		 * reservation conflict could be due to various reasons like
5465 		 * incorrect keys, not registered or not reserved etc. So,
5466 		 * we should not panic in that case.
5467 		 */
5468 		cdb = VD_SCSI_DATA_CDB(vd_scsi);
5469 		if (vdc->failfast_interval != 0 &&
5470 		    cdb->scc_cmd != SCMD_PERSISTENT_RESERVE_IN &&
5471 		    cdb->scc_cmd != SCMD_PERSISTENT_RESERVE_OUT) {
5472 			/* failfast is enabled so we have to panic */
5473 			(void) snprintf(panic_str, sizeof (panic_str),
5474 			    VDC_RESV_CONFLICT_FMT_STR "%s",
5475 			    ddi_pathname(vdc->dip, path_str));
5476 			panic(panic_str);
5477 		}
5478 		if (log_error)
5479 			cmn_err(CE_NOTE, "\tReservation Conflict\n");
5480 		rv = EACCES;
5481 		break;
5482 
5483 	case STATUS_QFULL:
5484 		if (log_error)
5485 			cmn_err(CE_NOTE, "\tQueue Full\n");
5486 		break;
5487 
5488 	case STATUS_MET:
5489 	case STATUS_INTERMEDIATE:
5490 	case STATUS_SCSI2:
5491 	case STATUS_INTERMEDIATE_MET:
5492 	case STATUS_ACA_ACTIVE:
5493 		if (log_error)
5494 			cmn_err(CE_CONT,
5495 			    "\tUnexpected SCSI status received: 0x%x\n",
5496 			    vd_scsi->cmd_status);
5497 		break;
5498 
5499 	default:
5500 		if (log_error)
5501 			cmn_err(CE_CONT,
5502 			    "\tInvalid SCSI status received: 0x%x\n",
5503 			    vd_scsi->cmd_status);
5504 		break;
5505 	}
5506 
5507 	return (rv);
5508 }
5509 
5510 /*
5511  * Implemented the USCSICMD uscsi(7I) ioctl. This ioctl is converted to
5512  * a VD_OP_SCSICMD operation which is sent to the vdisk server. If a SCSI
5513  * reset is requested (i.e. a flag USCSI_RESET* is set) then the ioctl is
5514  * converted to a VD_OP_RESET operation.
5515  */
5516 static int
5517 vdc_uscsi_cmd(vdc_t *vdc, caddr_t arg, int mode)
5518 {
5519 	struct uscsi_cmd 	uscsi;
5520 	struct uscsi_cmd32	uscsi32;
5521 	vd_scsi_t 		*vd_scsi;
5522 	int 			vd_scsi_len;
5523 	union scsi_cdb		*cdb;
5524 	struct scsi_extended_sense *sense;
5525 	char 			*datain, *dataout;
5526 	size_t			cdb_len, datain_len, dataout_len, sense_len;
5527 	int 			rv;
5528 
5529 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5530 		if (ddi_copyin(arg, &uscsi32, sizeof (struct uscsi_cmd32),
5531 		    mode) != 0)
5532 			return (EFAULT);
5533 		uscsi_cmd32touscsi_cmd((&uscsi32), (&uscsi));
5534 	} else {
5535 		if (ddi_copyin(arg, &uscsi, sizeof (struct uscsi_cmd),
5536 		    mode) != 0)
5537 			return (EFAULT);
5538 	}
5539 
5540 	/* a uscsi reset is converted to a VD_OP_RESET operation */
5541 	if (uscsi.uscsi_flags & (USCSI_RESET | USCSI_RESET_LUN |
5542 	    USCSI_RESET_ALL)) {
5543 		rv = vdc_do_sync_op(vdc, VD_OP_RESET, NULL, 0, 0, 0, CB_SYNC,
5544 		    (void *)(uint64_t)mode, VIO_both_dir, B_TRUE);
5545 		return (rv);
5546 	}
5547 
5548 	/* cdb buffer length */
5549 	cdb_len = uscsi.uscsi_cdblen;
5550 
5551 	/* data in and out buffers length */
5552 	if (uscsi.uscsi_flags & USCSI_READ) {
5553 		datain_len = uscsi.uscsi_buflen;
5554 		dataout_len = 0;
5555 	} else {
5556 		datain_len = 0;
5557 		dataout_len = uscsi.uscsi_buflen;
5558 	}
5559 
5560 	/* sense buffer length */
5561 	if (uscsi.uscsi_flags & USCSI_RQENABLE)
5562 		sense_len = uscsi.uscsi_rqlen;
5563 	else
5564 		sense_len = 0;
5565 
5566 	/* allocate buffer for the VD_SCSICMD_OP operation */
5567 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, datain_len, dataout_len,
5568 	    &vd_scsi_len);
5569 
5570 	/*
5571 	 * The documentation of USCSI_ISOLATE and USCSI_DIAGNOSE is very vague,
5572 	 * but basically they prevent a SCSI command from being retried in case
5573 	 * of an error.
5574 	 */
5575 	if ((uscsi.uscsi_flags & USCSI_ISOLATE) ||
5576 	    (uscsi.uscsi_flags & USCSI_DIAGNOSE))
5577 		vd_scsi->options |= VD_SCSI_OPT_NORETRY;
5578 
5579 	/* set task attribute */
5580 	if (uscsi.uscsi_flags & USCSI_NOTAG) {
5581 		vd_scsi->task_attribute = 0;
5582 	} else {
5583 		if (uscsi.uscsi_flags & USCSI_HEAD)
5584 			vd_scsi->task_attribute = VD_SCSI_TASK_ACA;
5585 		else if (uscsi.uscsi_flags & USCSI_HTAG)
5586 			vd_scsi->task_attribute = VD_SCSI_TASK_HQUEUE;
5587 		else if (uscsi.uscsi_flags & USCSI_OTAG)
5588 			vd_scsi->task_attribute = VD_SCSI_TASK_ORDERED;
5589 		else
5590 			vd_scsi->task_attribute = 0;
5591 	}
5592 
5593 	/* set timeout */
5594 	vd_scsi->timeout = uscsi.uscsi_timeout;
5595 
5596 	/* copy-in cdb data */
5597 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5598 	if (ddi_copyin(uscsi.uscsi_cdb, cdb, cdb_len, mode) != 0) {
5599 		rv = EFAULT;
5600 		goto done;
5601 	}
5602 
5603 	/* keep a pointer to the sense buffer */
5604 	sense = VD_SCSI_DATA_SENSE(vd_scsi);
5605 
5606 	/* keep a pointer to the data-in buffer */
5607 	datain = (char *)VD_SCSI_DATA_IN(vd_scsi);
5608 
5609 	/* copy-in request data to the data-out buffer */
5610 	dataout = (char *)VD_SCSI_DATA_OUT(vd_scsi);
5611 	if (!(uscsi.uscsi_flags & USCSI_READ)) {
5612 		if (ddi_copyin(uscsi.uscsi_bufaddr, dataout, dataout_len,
5613 		    mode)) {
5614 			rv = EFAULT;
5615 			goto done;
5616 		}
5617 	}
5618 
5619 	/* submit the request */
5620 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5621 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5622 
5623 	if (rv != 0)
5624 		goto done;
5625 
5626 	/* update scsi status */
5627 	uscsi.uscsi_status = vd_scsi->cmd_status;
5628 
5629 	/* update sense data */
5630 	if ((uscsi.uscsi_flags & USCSI_RQENABLE) &&
5631 	    (uscsi.uscsi_status == STATUS_CHECK ||
5632 	    uscsi.uscsi_status == STATUS_TERMINATED)) {
5633 
5634 		uscsi.uscsi_rqstatus = vd_scsi->sense_status;
5635 
5636 		if (uscsi.uscsi_rqstatus == STATUS_GOOD) {
5637 			uscsi.uscsi_rqresid = uscsi.uscsi_rqlen -
5638 			    vd_scsi->sense_len;
5639 			if (ddi_copyout(sense, uscsi.uscsi_rqbuf,
5640 			    vd_scsi->sense_len, mode) != 0) {
5641 				rv = EFAULT;
5642 				goto done;
5643 			}
5644 		}
5645 	}
5646 
5647 	/* update request data */
5648 	if (uscsi.uscsi_status == STATUS_GOOD) {
5649 		if (uscsi.uscsi_flags & USCSI_READ) {
5650 			uscsi.uscsi_resid = uscsi.uscsi_buflen -
5651 			    vd_scsi->datain_len;
5652 			if (ddi_copyout(datain, uscsi.uscsi_bufaddr,
5653 			    vd_scsi->datain_len, mode) != 0) {
5654 				rv = EFAULT;
5655 				goto done;
5656 			}
5657 		} else {
5658 			uscsi.uscsi_resid = uscsi.uscsi_buflen -
5659 			    vd_scsi->dataout_len;
5660 		}
5661 	}
5662 
5663 	/* copy-out result */
5664 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5665 		uscsi_cmdtouscsi_cmd32((&uscsi), (&uscsi32));
5666 		if (ddi_copyout(&uscsi32, arg, sizeof (struct uscsi_cmd32),
5667 		    mode) != 0) {
5668 			rv = EFAULT;
5669 			goto done;
5670 		}
5671 	} else {
5672 		if (ddi_copyout(&uscsi, arg, sizeof (struct uscsi_cmd),
5673 		    mode) != 0) {
5674 			rv = EFAULT;
5675 			goto done;
5676 		}
5677 	}
5678 
5679 	/* get the return code from the SCSI command status */
5680 	rv = vdc_scsi_status(vdc, vd_scsi,
5681 	    !(uscsi.uscsi_flags & USCSI_SILENT));
5682 
5683 done:
5684 	kmem_free(vd_scsi, vd_scsi_len);
5685 	return (rv);
5686 }
5687 
5688 /*
5689  * Create a VD_OP_SCSICMD buffer for a SCSI PERSISTENT IN command.
5690  *
5691  * Arguments:
5692  *	cmd		- SCSI PERSISTENT IN command
5693  *	len		- length of the SCSI input buffer
5694  *	vd_scsi_len	- return the length of the allocated buffer
5695  *
5696  * Returned Value:
5697  *	a pointer to the allocated VD_OP_SCSICMD buffer.
5698  */
5699 static vd_scsi_t *
5700 vdc_scsi_alloc_persistent_in(uchar_t cmd, int len, int *vd_scsi_len)
5701 {
5702 	int cdb_len, sense_len, datain_len, dataout_len;
5703 	vd_scsi_t *vd_scsi;
5704 	union scsi_cdb *cdb;
5705 
5706 	cdb_len = CDB_GROUP1;
5707 	sense_len = sizeof (struct scsi_extended_sense);
5708 	datain_len = len;
5709 	dataout_len = 0;
5710 
5711 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, datain_len, dataout_len,
5712 	    vd_scsi_len);
5713 
5714 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5715 
5716 	/* set cdb */
5717 	cdb->scc_cmd = SCMD_PERSISTENT_RESERVE_IN;
5718 	cdb->cdb_opaque[1] = cmd;
5719 	FORMG1COUNT(cdb, datain_len);
5720 
5721 	vd_scsi->timeout = vdc_scsi_timeout;
5722 
5723 	return (vd_scsi);
5724 }
5725 
5726 /*
5727  * Create a VD_OP_SCSICMD buffer for a SCSI PERSISTENT OUT command.
5728  *
5729  * Arguments:
5730  *	cmd		- SCSI PERSISTENT OUT command
5731  *	len		- length of the SCSI output buffer
5732  *	vd_scsi_len	- return the length of the allocated buffer
5733  *
5734  * Returned Code:
5735  *	a pointer to the allocated VD_OP_SCSICMD buffer.
5736  */
5737 static vd_scsi_t *
5738 vdc_scsi_alloc_persistent_out(uchar_t cmd, int len, int *vd_scsi_len)
5739 {
5740 	int cdb_len, sense_len, datain_len, dataout_len;
5741 	vd_scsi_t *vd_scsi;
5742 	union scsi_cdb *cdb;
5743 
5744 	cdb_len = CDB_GROUP1;
5745 	sense_len = sizeof (struct scsi_extended_sense);
5746 	datain_len = 0;
5747 	dataout_len = len;
5748 
5749 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, datain_len, dataout_len,
5750 	    vd_scsi_len);
5751 
5752 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5753 
5754 	/* set cdb */
5755 	cdb->scc_cmd = SCMD_PERSISTENT_RESERVE_OUT;
5756 	cdb->cdb_opaque[1] = cmd;
5757 	FORMG1COUNT(cdb, dataout_len);
5758 
5759 	vd_scsi->timeout = vdc_scsi_timeout;
5760 
5761 	return (vd_scsi);
5762 }
5763 
5764 /*
5765  * Implement the MHIOCGRP_INKEYS mhd(7i) ioctl. The ioctl is converted
5766  * to a SCSI PERSISTENT IN READ KEYS command which is sent to the vdisk
5767  * server with a VD_OP_SCSICMD operation.
5768  */
5769 static int
5770 vdc_mhd_inkeys(vdc_t *vdc, caddr_t arg, int mode)
5771 {
5772 	vd_scsi_t *vd_scsi;
5773 	mhioc_inkeys_t inkeys;
5774 	mhioc_key_list_t klist;
5775 	struct mhioc_inkeys32 inkeys32;
5776 	struct mhioc_key_list32 klist32;
5777 	sd_prin_readkeys_t *scsi_keys;
5778 	void *user_keys;
5779 	int vd_scsi_len;
5780 	int listsize, listlen, rv;
5781 
5782 	/* copyin arguments */
5783 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5784 		rv = ddi_copyin(arg, &inkeys32, sizeof (inkeys32), mode);
5785 		if (rv != 0)
5786 			return (EFAULT);
5787 
5788 		rv = ddi_copyin((caddr_t)(uintptr_t)inkeys32.li, &klist32,
5789 		    sizeof (klist32), mode);
5790 		if (rv != 0)
5791 			return (EFAULT);
5792 
5793 		listsize = klist32.listsize;
5794 	} else {
5795 		rv = ddi_copyin(arg, &inkeys, sizeof (inkeys), mode);
5796 		if (rv != 0)
5797 			return (EFAULT);
5798 
5799 		rv = ddi_copyin(inkeys.li, &klist, sizeof (klist), mode);
5800 		if (rv != 0)
5801 			return (EFAULT);
5802 
5803 		listsize = klist.listsize;
5804 	}
5805 
5806 	/* build SCSI VD_OP request */
5807 	vd_scsi = vdc_scsi_alloc_persistent_in(SD_READ_KEYS,
5808 	    sizeof (sd_prin_readkeys_t) - sizeof (caddr_t) +
5809 	    (sizeof (mhioc_resv_key_t) * listsize), &vd_scsi_len);
5810 
5811 	scsi_keys = (sd_prin_readkeys_t *)VD_SCSI_DATA_IN(vd_scsi);
5812 
5813 	/* submit the request */
5814 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5815 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5816 
5817 	if (rv != 0)
5818 		goto done;
5819 
5820 	listlen = scsi_keys->len / MHIOC_RESV_KEY_SIZE;
5821 
5822 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5823 		inkeys32.generation = scsi_keys->generation;
5824 		rv = ddi_copyout(&inkeys32, arg, sizeof (inkeys32), mode);
5825 		if (rv != 0) {
5826 			rv = EFAULT;
5827 			goto done;
5828 		}
5829 
5830 		klist32.listlen = listlen;
5831 		rv = ddi_copyout(&klist32, (caddr_t)(uintptr_t)inkeys32.li,
5832 		    sizeof (klist32), mode);
5833 		if (rv != 0) {
5834 			rv = EFAULT;
5835 			goto done;
5836 		}
5837 
5838 		user_keys = (caddr_t)(uintptr_t)klist32.list;
5839 	} else {
5840 		inkeys.generation = scsi_keys->generation;
5841 		rv = ddi_copyout(&inkeys, arg, sizeof (inkeys), mode);
5842 		if (rv != 0) {
5843 			rv = EFAULT;
5844 			goto done;
5845 		}
5846 
5847 		klist.listlen = listlen;
5848 		rv = ddi_copyout(&klist, inkeys.li, sizeof (klist), mode);
5849 		if (rv != 0) {
5850 			rv = EFAULT;
5851 			goto done;
5852 		}
5853 
5854 		user_keys = klist.list;
5855 	}
5856 
5857 	/* copy out keys */
5858 	if (listlen > 0 && listsize > 0) {
5859 		if (listsize < listlen)
5860 			listlen = listsize;
5861 		rv = ddi_copyout(&scsi_keys->keylist, user_keys,
5862 		    listlen * MHIOC_RESV_KEY_SIZE, mode);
5863 		if (rv != 0)
5864 			rv = EFAULT;
5865 	}
5866 
5867 	if (rv == 0)
5868 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
5869 
5870 done:
5871 	kmem_free(vd_scsi, vd_scsi_len);
5872 
5873 	return (rv);
5874 }
5875 
5876 /*
5877  * Implement the MHIOCGRP_INRESV mhd(7i) ioctl. The ioctl is converted
5878  * to a SCSI PERSISTENT IN READ RESERVATION command which is sent to
5879  * the vdisk server with a VD_OP_SCSICMD operation.
5880  */
5881 static int
5882 vdc_mhd_inresv(vdc_t *vdc, caddr_t arg, int mode)
5883 {
5884 	vd_scsi_t *vd_scsi;
5885 	mhioc_inresvs_t inresv;
5886 	mhioc_resv_desc_list_t rlist;
5887 	struct mhioc_inresvs32 inresv32;
5888 	struct mhioc_resv_desc_list32 rlist32;
5889 	mhioc_resv_desc_t mhd_resv;
5890 	sd_prin_readresv_t *scsi_resv;
5891 	sd_readresv_desc_t *resv;
5892 	mhioc_resv_desc_t *user_resv;
5893 	int vd_scsi_len;
5894 	int listsize, listlen, i, rv;
5895 
5896 	/* copyin arguments */
5897 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5898 		rv = ddi_copyin(arg, &inresv32, sizeof (inresv32), mode);
5899 		if (rv != 0)
5900 			return (EFAULT);
5901 
5902 		rv = ddi_copyin((caddr_t)(uintptr_t)inresv32.li, &rlist32,
5903 		    sizeof (rlist32), mode);
5904 		if (rv != 0)
5905 			return (EFAULT);
5906 
5907 		listsize = rlist32.listsize;
5908 	} else {
5909 		rv = ddi_copyin(arg, &inresv, sizeof (inresv), mode);
5910 		if (rv != 0)
5911 			return (EFAULT);
5912 
5913 		rv = ddi_copyin(inresv.li, &rlist, sizeof (rlist), mode);
5914 		if (rv != 0)
5915 			return (EFAULT);
5916 
5917 		listsize = rlist.listsize;
5918 	}
5919 
5920 	/* build SCSI VD_OP request */
5921 	vd_scsi = vdc_scsi_alloc_persistent_in(SD_READ_RESV,
5922 	    sizeof (sd_prin_readresv_t) - sizeof (caddr_t) +
5923 	    (SCSI3_RESV_DESC_LEN * listsize), &vd_scsi_len);
5924 
5925 	scsi_resv = (sd_prin_readresv_t *)VD_SCSI_DATA_IN(vd_scsi);
5926 
5927 	/* submit the request */
5928 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5929 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5930 
5931 	if (rv != 0)
5932 		goto done;
5933 
5934 	listlen = scsi_resv->len / SCSI3_RESV_DESC_LEN;
5935 
5936 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5937 		inresv32.generation = scsi_resv->generation;
5938 		rv = ddi_copyout(&inresv32, arg, sizeof (inresv32), mode);
5939 		if (rv != 0) {
5940 			rv = EFAULT;
5941 			goto done;
5942 		}
5943 
5944 		rlist32.listlen = listlen;
5945 		rv = ddi_copyout(&rlist32, (caddr_t)(uintptr_t)inresv32.li,
5946 		    sizeof (rlist32), mode);
5947 		if (rv != 0) {
5948 			rv = EFAULT;
5949 			goto done;
5950 		}
5951 
5952 		user_resv = (mhioc_resv_desc_t *)(uintptr_t)rlist32.list;
5953 	} else {
5954 		inresv.generation = scsi_resv->generation;
5955 		rv = ddi_copyout(&inresv, arg, sizeof (inresv), mode);
5956 		if (rv != 0) {
5957 			rv = EFAULT;
5958 			goto done;
5959 		}
5960 
5961 		rlist.listlen = listlen;
5962 		rv = ddi_copyout(&rlist, inresv.li, sizeof (rlist), mode);
5963 		if (rv != 0) {
5964 			rv = EFAULT;
5965 			goto done;
5966 		}
5967 
5968 		user_resv = rlist.list;
5969 	}
5970 
5971 	/* copy out reservations */
5972 	if (listsize > 0 && listlen > 0) {
5973 		if (listsize < listlen)
5974 			listlen = listsize;
5975 		resv = (sd_readresv_desc_t *)&scsi_resv->readresv_desc;
5976 
5977 		for (i = 0; i < listlen; i++) {
5978 			mhd_resv.type = resv->type;
5979 			mhd_resv.scope = resv->scope;
5980 			mhd_resv.scope_specific_addr =
5981 			    BE_32(resv->scope_specific_addr);
5982 			bcopy(&resv->resvkey, &mhd_resv.key,
5983 			    MHIOC_RESV_KEY_SIZE);
5984 
5985 			rv = ddi_copyout(&mhd_resv, user_resv,
5986 			    sizeof (mhd_resv), mode);
5987 			if (rv != 0) {
5988 				rv = EFAULT;
5989 				goto done;
5990 			}
5991 			resv++;
5992 			user_resv++;
5993 		}
5994 	}
5995 
5996 	if (rv == 0)
5997 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
5998 
5999 done:
6000 	kmem_free(vd_scsi, vd_scsi_len);
6001 	return (rv);
6002 }
6003 
6004 /*
6005  * Implement the MHIOCGRP_REGISTER mhd(7i) ioctl. The ioctl is converted
6006  * to a SCSI PERSISTENT OUT REGISTER command which is sent to the vdisk
6007  * server with a VD_OP_SCSICMD operation.
6008  */
6009 static int
6010 vdc_mhd_register(vdc_t *vdc, caddr_t arg, int mode)
6011 {
6012 	vd_scsi_t *vd_scsi;
6013 	sd_prout_t *scsi_prout;
6014 	mhioc_register_t mhd_reg;
6015 	int vd_scsi_len, rv;
6016 
6017 	/* copyin arguments */
6018 	rv = ddi_copyin(arg, &mhd_reg, sizeof (mhd_reg), mode);
6019 	if (rv != 0)
6020 		return (EFAULT);
6021 
6022 	/* build SCSI VD_OP request */
6023 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_REGISTER,
6024 	    sizeof (sd_prout_t), &vd_scsi_len);
6025 
6026 	/* set parameters */
6027 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
6028 	bcopy(mhd_reg.oldkey.key, scsi_prout->res_key, MHIOC_RESV_KEY_SIZE);
6029 	bcopy(mhd_reg.newkey.key, scsi_prout->service_key, MHIOC_RESV_KEY_SIZE);
6030 	scsi_prout->aptpl = (uchar_t)mhd_reg.aptpl;
6031 
6032 	/* submit the request */
6033 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6034 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
6035 
6036 	if (rv == 0)
6037 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6038 
6039 	kmem_free(vd_scsi, vd_scsi_len);
6040 	return (rv);
6041 }
6042 
6043 /*
6044  * Implement the MHIOCGRP_RESERVE mhd(7i) ioctl. The ioctl is converted
6045  * to a SCSI PERSISTENT OUT RESERVE command which is sent to the vdisk
6046  * server with a VD_OP_SCSICMD operation.
6047  */
6048 static int
6049 vdc_mhd_reserve(vdc_t *vdc, caddr_t arg, int mode)
6050 {
6051 	union scsi_cdb *cdb;
6052 	vd_scsi_t *vd_scsi;
6053 	sd_prout_t *scsi_prout;
6054 	mhioc_resv_desc_t mhd_resv;
6055 	int vd_scsi_len, rv;
6056 
6057 	/* copyin arguments */
6058 	rv = ddi_copyin(arg, &mhd_resv, sizeof (mhd_resv), mode);
6059 	if (rv != 0)
6060 		return (EFAULT);
6061 
6062 	/* build SCSI VD_OP request */
6063 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_RESERVE,
6064 	    sizeof (sd_prout_t), &vd_scsi_len);
6065 
6066 	/* set parameters */
6067 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
6068 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
6069 	bcopy(mhd_resv.key.key, scsi_prout->res_key, MHIOC_RESV_KEY_SIZE);
6070 	scsi_prout->scope_address = mhd_resv.scope_specific_addr;
6071 	cdb->cdb_opaque[2] = mhd_resv.type;
6072 
6073 	/* submit the request */
6074 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6075 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
6076 
6077 	if (rv == 0)
6078 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6079 
6080 	kmem_free(vd_scsi, vd_scsi_len);
6081 	return (rv);
6082 }
6083 
6084 /*
6085  * Implement the MHIOCGRP_PREEMPTANDABORT mhd(7i) ioctl. The ioctl is
6086  * converted to a SCSI PERSISTENT OUT PREEMPT AND ABORT command which
6087  * is sent to the vdisk server with a VD_OP_SCSICMD operation.
6088  */
6089 static int
6090 vdc_mhd_preemptabort(vdc_t *vdc, caddr_t arg, int mode)
6091 {
6092 	union scsi_cdb *cdb;
6093 	vd_scsi_t *vd_scsi;
6094 	sd_prout_t *scsi_prout;
6095 	mhioc_preemptandabort_t mhd_preempt;
6096 	int vd_scsi_len, rv;
6097 
6098 	/* copyin arguments */
6099 	rv = ddi_copyin(arg, &mhd_preempt, sizeof (mhd_preempt), mode);
6100 	if (rv != 0)
6101 		return (EFAULT);
6102 
6103 	/* build SCSI VD_OP request */
6104 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_PREEMPTANDABORT,
6105 	    sizeof (sd_prout_t), &vd_scsi_len);
6106 
6107 	/* set parameters */
6108 	vd_scsi->task_attribute = VD_SCSI_TASK_ACA;
6109 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
6110 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
6111 	bcopy(mhd_preempt.resvdesc.key.key, scsi_prout->res_key,
6112 	    MHIOC_RESV_KEY_SIZE);
6113 	bcopy(mhd_preempt.victim_key.key, scsi_prout->service_key,
6114 	    MHIOC_RESV_KEY_SIZE);
6115 	scsi_prout->scope_address = mhd_preempt.resvdesc.scope_specific_addr;
6116 	cdb->cdb_opaque[2] = mhd_preempt.resvdesc.type;
6117 
6118 	/* submit the request */
6119 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6120 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
6121 
6122 	if (rv == 0)
6123 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6124 
6125 	kmem_free(vd_scsi, vd_scsi_len);
6126 	return (rv);
6127 }
6128 
6129 /*
6130  * Implement the MHIOCGRP_REGISTERANDIGNOREKEY mhd(7i) ioctl. The ioctl
6131  * is converted to a SCSI PERSISTENT OUT REGISTER AND IGNORE EXISTING KEY
6132  * command which is sent to the vdisk server with a VD_OP_SCSICMD operation.
6133  */
6134 static int
6135 vdc_mhd_registerignore(vdc_t *vdc, caddr_t arg, int mode)
6136 {
6137 	vd_scsi_t *vd_scsi;
6138 	sd_prout_t *scsi_prout;
6139 	mhioc_registerandignorekey_t mhd_regi;
6140 	int vd_scsi_len, rv;
6141 
6142 	/* copyin arguments */
6143 	rv = ddi_copyin(arg, &mhd_regi, sizeof (mhd_regi), mode);
6144 	if (rv != 0)
6145 		return (EFAULT);
6146 
6147 	/* build SCSI VD_OP request */
6148 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_REGISTERANDIGNOREKEY,
6149 	    sizeof (sd_prout_t), &vd_scsi_len);
6150 
6151 	/* set parameters */
6152 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
6153 	bcopy(mhd_regi.newkey.key, scsi_prout->service_key,
6154 	    MHIOC_RESV_KEY_SIZE);
6155 	scsi_prout->aptpl = (uchar_t)mhd_regi.aptpl;
6156 
6157 	/* submit the request */
6158 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6159 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
6160 
6161 	if (rv == 0)
6162 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6163 
6164 	kmem_free(vd_scsi, vd_scsi_len);
6165 	return (rv);
6166 }
6167 
6168 /*
6169  * This function is used by the failfast mechanism to send a SCSI command
6170  * to check for reservation conflict.
6171  */
6172 static int
6173 vdc_failfast_scsi_cmd(vdc_t *vdc, uchar_t scmd)
6174 {
6175 	int cdb_len, sense_len, vd_scsi_len;
6176 	vd_scsi_t *vd_scsi;
6177 	union scsi_cdb *cdb;
6178 	int rv;
6179 
6180 	ASSERT(scmd == SCMD_TEST_UNIT_READY || scmd == SCMD_WRITE_G1);
6181 
6182 	if (scmd == SCMD_WRITE_G1)
6183 		cdb_len = CDB_GROUP1;
6184 	else
6185 		cdb_len = CDB_GROUP0;
6186 
6187 	sense_len = sizeof (struct scsi_extended_sense);
6188 
6189 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, 0, 0, &vd_scsi_len);
6190 
6191 	/* set cdb */
6192 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
6193 	cdb->scc_cmd = scmd;
6194 
6195 	vd_scsi->timeout = vdc_scsi_timeout;
6196 
6197 	/*
6198 	 * Submit the request. The last argument has to be B_FALSE so that
6199 	 * vdc_do_sync_op does not loop checking for reservation conflict if
6200 	 * the operation returns an error.
6201 	 */
6202 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6203 	    0, 0, CB_SYNC, (void *)(uint64_t)FKIOCTL, VIO_both_dir, B_FALSE);
6204 
6205 	if (rv == 0)
6206 		(void) vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6207 
6208 	kmem_free(vd_scsi, vd_scsi_len);
6209 	return (rv);
6210 }
6211 
6212 /*
6213  * This function is used by the failfast mechanism to check for reservation
6214  * conflict. It sends some SCSI commands which will fail with a reservation
6215  * conflict error if the system does not have access to the disk and this
6216  * will panic the system.
6217  *
6218  * Returned Code:
6219  *	0	- disk is accessible without reservation conflict error
6220  *	!= 0	- unable to check if disk is accessible
6221  */
6222 int
6223 vdc_failfast_check_resv(vdc_t *vdc)
6224 {
6225 	int failure = 0;
6226 
6227 	/*
6228 	 * Send a TEST UNIT READY command. The command will panic
6229 	 * the system if it fails with a reservation conflict.
6230 	 */
6231 	if (vdc_failfast_scsi_cmd(vdc, SCMD_TEST_UNIT_READY) != 0)
6232 		failure++;
6233 
6234 	/*
6235 	 * With SPC-3 compliant devices TEST UNIT READY will succeed on
6236 	 * a reserved device, so we also do a WRITE(10) of zero byte in
6237 	 * order to provoke a Reservation Conflict status on those newer
6238 	 * devices.
6239 	 */
6240 	if (vdc_failfast_scsi_cmd(vdc, SCMD_WRITE_G1) != 0)
6241 		failure++;
6242 
6243 	return (failure);
6244 }
6245 
6246 /*
6247  * Add a pending I/O to the failfast I/O queue. An I/O is added to this
6248  * queue when it has failed and failfast is enabled. Then we have to check
6249  * if it has failed because of a reservation conflict in which case we have
6250  * to panic the system.
6251  *
6252  * Async I/O should be queued with their block I/O data transfer structure
6253  * (buf). Sync I/O should be queued with buf = NULL.
6254  */
6255 static vdc_io_t *
6256 vdc_failfast_io_queue(vdc_t *vdc, struct buf *buf)
6257 {
6258 	vdc_io_t *vio;
6259 
6260 	ASSERT(MUTEX_HELD(&vdc->lock));
6261 
6262 	vio = kmem_alloc(sizeof (vdc_io_t), KM_SLEEP);
6263 	vio->vio_next = vdc->failfast_io_queue;
6264 	vio->vio_buf = buf;
6265 	vio->vio_qtime = ddi_get_lbolt();
6266 
6267 	vdc->failfast_io_queue = vio;
6268 
6269 	/* notify the failfast thread that a new I/O is queued */
6270 	cv_signal(&vdc->failfast_cv);
6271 
6272 	return (vio);
6273 }
6274 
6275 /*
6276  * Remove and complete I/O in the failfast I/O queue which have been
6277  * added after the indicated deadline. A deadline of 0 means that all
6278  * I/O have to be unqueued and marked as completed.
6279  */
6280 static void
6281 vdc_failfast_io_unqueue(vdc_t *vdc, clock_t deadline)
6282 {
6283 	vdc_io_t *vio, *vio_tmp;
6284 
6285 	ASSERT(MUTEX_HELD(&vdc->lock));
6286 
6287 	vio_tmp = NULL;
6288 	vio = vdc->failfast_io_queue;
6289 
6290 	if (deadline != 0) {
6291 		/*
6292 		 * Skip any io queued after the deadline. The failfast
6293 		 * I/O queue is ordered starting with the last I/O added
6294 		 * to the queue.
6295 		 */
6296 		while (vio != NULL && vio->vio_qtime > deadline) {
6297 			vio_tmp = vio;
6298 			vio = vio->vio_next;
6299 		}
6300 	}
6301 
6302 	if (vio == NULL)
6303 		/* nothing to unqueue */
6304 		return;
6305 
6306 	/* update the queue */
6307 	if (vio_tmp == NULL)
6308 		vdc->failfast_io_queue = NULL;
6309 	else
6310 		vio_tmp->vio_next = NULL;
6311 
6312 	/*
6313 	 * Complete unqueued I/O. Async I/O have a block I/O data transfer
6314 	 * structure (buf) and they are completed by calling biodone(). Sync
6315 	 * I/O do not have a buf and they are completed by setting the
6316 	 * vio_qtime to zero and signaling failfast_io_cv. In that case, the
6317 	 * thread waiting for the I/O to complete is responsible for freeing
6318 	 * the vio structure.
6319 	 */
6320 	while (vio != NULL) {
6321 		vio_tmp = vio->vio_next;
6322 		if (vio->vio_buf != NULL) {
6323 			VD_KSTAT_RUNQ_EXIT(vdc);
6324 			DTRACE_IO1(done, buf_t *, vio->vio_buf);
6325 			biodone(vio->vio_buf);
6326 			kmem_free(vio, sizeof (vdc_io_t));
6327 		} else {
6328 			vio->vio_qtime = 0;
6329 		}
6330 		vio = vio_tmp;
6331 	}
6332 
6333 	cv_broadcast(&vdc->failfast_io_cv);
6334 }
6335 
6336 /*
6337  * Failfast Thread.
6338  *
6339  * While failfast is enabled, the failfast thread sends a TEST UNIT READY
6340  * and a zero size WRITE(10) SCSI commands on a regular basis to check that
6341  * we still have access to the disk. If a command fails with a RESERVATION
6342  * CONFLICT error then the system will immediatly panic.
6343  *
6344  * The failfast thread is also woken up when an I/O has failed. It then check
6345  * the access to the disk to ensure that the I/O failure was not due to a
6346  * reservation conflict.
6347  *
6348  * There is one failfast thread for each virtual disk for which failfast is
6349  * enabled. We could have only one thread sending requests for all disks but
6350  * this would need vdc to send asynchronous requests and to have callbacks to
6351  * process replies.
6352  */
6353 static void
6354 vdc_failfast_thread(void *arg)
6355 {
6356 	int status;
6357 	vdc_t *vdc = (vdc_t *)arg;
6358 	clock_t timeout, starttime;
6359 
6360 	mutex_enter(&vdc->lock);
6361 
6362 	while (vdc->failfast_interval != 0) {
6363 
6364 		starttime = ddi_get_lbolt();
6365 
6366 		mutex_exit(&vdc->lock);
6367 
6368 		/* check for reservation conflict */
6369 		status = vdc_failfast_check_resv(vdc);
6370 
6371 		mutex_enter(&vdc->lock);
6372 		/*
6373 		 * We have dropped the lock to send the SCSI command so we have
6374 		 * to check that failfast is still enabled.
6375 		 */
6376 		if (vdc->failfast_interval == 0)
6377 			break;
6378 
6379 		/*
6380 		 * If we have successfully check the disk access and there was
6381 		 * no reservation conflict then we can complete any I/O queued
6382 		 * before the last check.
6383 		 */
6384 		if (status == 0)
6385 			vdc_failfast_io_unqueue(vdc, starttime);
6386 
6387 		/* proceed again if some I/O are still in the queue */
6388 		if (vdc->failfast_io_queue != NULL)
6389 			continue;
6390 
6391 		timeout = ddi_get_lbolt() +
6392 		    drv_usectohz(vdc->failfast_interval);
6393 		(void) cv_timedwait(&vdc->failfast_cv, &vdc->lock, timeout);
6394 	}
6395 
6396 	/*
6397 	 * Failfast is being stop so we can complete any queued I/O.
6398 	 */
6399 	vdc_failfast_io_unqueue(vdc, 0);
6400 	vdc->failfast_thread = NULL;
6401 	mutex_exit(&vdc->lock);
6402 	thread_exit();
6403 }
6404 
6405 /*
6406  * Implement the MHIOCENFAILFAST mhd(7i) ioctl.
6407  */
6408 static int
6409 vdc_failfast(vdc_t *vdc, caddr_t arg, int mode)
6410 {
6411 	unsigned int mh_time;
6412 
6413 	if (ddi_copyin((void *)arg, &mh_time, sizeof (int), mode))
6414 		return (EFAULT);
6415 
6416 	mutex_enter(&vdc->lock);
6417 	if (mh_time != 0 && vdc->failfast_thread == NULL) {
6418 		vdc->failfast_thread = thread_create(NULL, 0,
6419 		    vdc_failfast_thread, vdc, 0, &p0, TS_RUN,
6420 		    v.v_maxsyspri - 2);
6421 	}
6422 
6423 	vdc->failfast_interval = mh_time * 1000;
6424 	cv_signal(&vdc->failfast_cv);
6425 	mutex_exit(&vdc->lock);
6426 
6427 	return (0);
6428 }
6429 
6430 /*
6431  * Implement the MHIOCTKOWN and MHIOCRELEASE mhd(7i) ioctls. These ioctls are
6432  * converted to VD_OP_SET_ACCESS operations.
6433  */
6434 static int
6435 vdc_access_set(vdc_t *vdc, uint64_t flags, int mode)
6436 {
6437 	int rv;
6438 
6439 	/* submit owership command request */
6440 	rv = vdc_do_sync_op(vdc, VD_OP_SET_ACCESS, (caddr_t)&flags,
6441 	    sizeof (uint64_t), 0, 0, CB_SYNC, (void *)(uint64_t)mode,
6442 	    VIO_both_dir, B_TRUE);
6443 
6444 	return (rv);
6445 }
6446 
6447 /*
6448  * Implement the MHIOCSTATUS mhd(7i) ioctl. This ioctl is converted to a
6449  * VD_OP_GET_ACCESS operation.
6450  */
6451 static int
6452 vdc_access_get(vdc_t *vdc, uint64_t *status, int mode)
6453 {
6454 	int rv;
6455 
6456 	/* submit owership command request */
6457 	rv = vdc_do_sync_op(vdc, VD_OP_GET_ACCESS, (caddr_t)status,
6458 	    sizeof (uint64_t), 0, 0, CB_SYNC, (void *)(uint64_t)mode,
6459 	    VIO_both_dir, B_TRUE);
6460 
6461 	return (rv);
6462 }
6463 
6464 /*
6465  * Disk Ownership Thread.
6466  *
6467  * When we have taken the ownership of a disk, this thread waits to be
6468  * notified when the LDC channel is reset so that it can recover the
6469  * ownership.
6470  *
6471  * Note that the thread handling the LDC reset (vdc_process_msg_thread())
6472  * can not be used to do the ownership recovery because it has to be
6473  * running to handle the reply message to the ownership operation.
6474  */
6475 static void
6476 vdc_ownership_thread(void *arg)
6477 {
6478 	vdc_t *vdc = (vdc_t *)arg;
6479 	clock_t timeout;
6480 	uint64_t status;
6481 
6482 	mutex_enter(&vdc->ownership_lock);
6483 	mutex_enter(&vdc->lock);
6484 
6485 	while (vdc->ownership & VDC_OWNERSHIP_WANTED) {
6486 
6487 		if ((vdc->ownership & VDC_OWNERSHIP_RESET) ||
6488 		    !(vdc->ownership & VDC_OWNERSHIP_GRANTED)) {
6489 			/*
6490 			 * There was a reset so the ownership has been lost,
6491 			 * try to recover. We do this without using the preempt
6492 			 * option so that we don't steal the ownership from
6493 			 * someone who has preempted us.
6494 			 */
6495 			DMSG(vdc, 0, "[%d] Ownership lost, recovering",
6496 			    vdc->instance);
6497 
6498 			vdc->ownership &= ~(VDC_OWNERSHIP_RESET |
6499 			    VDC_OWNERSHIP_GRANTED);
6500 
6501 			mutex_exit(&vdc->lock);
6502 
6503 			status = vdc_access_set(vdc, VD_ACCESS_SET_EXCLUSIVE |
6504 			    VD_ACCESS_SET_PRESERVE, FKIOCTL);
6505 
6506 			mutex_enter(&vdc->lock);
6507 
6508 			if (status == 0) {
6509 				DMSG(vdc, 0, "[%d] Ownership recovered",
6510 				    vdc->instance);
6511 				vdc->ownership |= VDC_OWNERSHIP_GRANTED;
6512 			} else {
6513 				DMSG(vdc, 0, "[%d] Fail to recover ownership",
6514 				    vdc->instance);
6515 			}
6516 
6517 		}
6518 
6519 		/*
6520 		 * If we have the ownership then we just wait for an event
6521 		 * to happen (LDC reset), otherwise we will retry to recover
6522 		 * after a delay.
6523 		 */
6524 		if (vdc->ownership & VDC_OWNERSHIP_GRANTED)
6525 			timeout = 0;
6526 		else
6527 			timeout = ddi_get_lbolt() +
6528 			    drv_usectohz(vdc_ownership_delay);
6529 
6530 		/* Release the ownership_lock and wait on the vdc lock */
6531 		mutex_exit(&vdc->ownership_lock);
6532 
6533 		if (timeout == 0)
6534 			(void) cv_wait(&vdc->ownership_cv, &vdc->lock);
6535 		else
6536 			(void) cv_timedwait(&vdc->ownership_cv,
6537 			    &vdc->lock, timeout);
6538 
6539 		mutex_exit(&vdc->lock);
6540 
6541 		mutex_enter(&vdc->ownership_lock);
6542 		mutex_enter(&vdc->lock);
6543 	}
6544 
6545 	vdc->ownership_thread = NULL;
6546 	mutex_exit(&vdc->lock);
6547 	mutex_exit(&vdc->ownership_lock);
6548 
6549 	thread_exit();
6550 }
6551 
6552 static void
6553 vdc_ownership_update(vdc_t *vdc, int ownership_flags)
6554 {
6555 	ASSERT(MUTEX_HELD(&vdc->ownership_lock));
6556 
6557 	mutex_enter(&vdc->lock);
6558 	vdc->ownership = ownership_flags;
6559 	if ((vdc->ownership & VDC_OWNERSHIP_WANTED) &&
6560 	    vdc->ownership_thread == NULL) {
6561 		/* start ownership thread */
6562 		vdc->ownership_thread = thread_create(NULL, 0,
6563 		    vdc_ownership_thread, vdc, 0, &p0, TS_RUN,
6564 		    v.v_maxsyspri - 2);
6565 	} else {
6566 		/* notify the ownership thread */
6567 		cv_signal(&vdc->ownership_cv);
6568 	}
6569 	mutex_exit(&vdc->lock);
6570 }
6571 
6572 /*
6573  * Get the size and the block size of a virtual disk from the vdisk server.
6574  * We need to use this operation when the vdisk_size attribute was not
6575  * available during the handshake with the vdisk server.
6576  */
6577 static int
6578 vdc_check_capacity(vdc_t *vdc)
6579 {
6580 	int rv = 0;
6581 	size_t alloc_len;
6582 	vd_capacity_t *vd_cap;
6583 
6584 	if (vdc->vdisk_size != 0)
6585 		return (0);
6586 
6587 	alloc_len = P2ROUNDUP(sizeof (vd_capacity_t), sizeof (uint64_t));
6588 
6589 	vd_cap = kmem_zalloc(alloc_len, KM_SLEEP);
6590 
6591 	rv = vdc_do_sync_op(vdc, VD_OP_GET_CAPACITY, (caddr_t)vd_cap, alloc_len,
6592 	    0, 0, CB_SYNC, (void *)(uint64_t)FKIOCTL, VIO_both_dir, B_TRUE);
6593 
6594 	if (rv == 0) {
6595 		if (vd_cap->vdisk_block_size != vdc->block_size ||
6596 		    vd_cap->vdisk_size == VD_SIZE_UNKNOWN ||
6597 		    vd_cap->vdisk_size == 0)
6598 			rv = EINVAL;
6599 		else
6600 			vdc->vdisk_size = vd_cap->vdisk_size;
6601 	}
6602 
6603 	kmem_free(vd_cap, alloc_len);
6604 	return (rv);
6605 }
6606 
6607 /*
6608  * This structure is used in the DKIO(7I) array below.
6609  */
6610 typedef struct vdc_dk_ioctl {
6611 	uint8_t		op;		/* VD_OP_XXX value */
6612 	int		cmd;		/* Solaris ioctl operation number */
6613 	size_t		nbytes;		/* size of structure to be copied */
6614 
6615 	/* function to convert between vDisk and Solaris structure formats */
6616 	int	(*convert)(vdc_t *vdc, void *vd_buf, void *ioctl_arg,
6617 	    int mode, int dir);
6618 } vdc_dk_ioctl_t;
6619 
6620 /*
6621  * Subset of DKIO(7I) operations currently supported
6622  */
6623 static vdc_dk_ioctl_t	dk_ioctl[] = {
6624 	{VD_OP_FLUSH,		DKIOCFLUSHWRITECACHE,	0,
6625 		vdc_null_copy_func},
6626 	{VD_OP_GET_WCE,		DKIOCGETWCE,		sizeof (int),
6627 		vdc_get_wce_convert},
6628 	{VD_OP_SET_WCE,		DKIOCSETWCE,		sizeof (int),
6629 		vdc_set_wce_convert},
6630 	{VD_OP_GET_VTOC,	DKIOCGVTOC,		sizeof (vd_vtoc_t),
6631 		vdc_get_vtoc_convert},
6632 	{VD_OP_SET_VTOC,	DKIOCSVTOC,		sizeof (vd_vtoc_t),
6633 		vdc_set_vtoc_convert},
6634 	{VD_OP_GET_DISKGEOM,	DKIOCGGEOM,		sizeof (vd_geom_t),
6635 		vdc_get_geom_convert},
6636 	{VD_OP_GET_DISKGEOM,	DKIOCG_PHYGEOM,		sizeof (vd_geom_t),
6637 		vdc_get_geom_convert},
6638 	{VD_OP_GET_DISKGEOM, 	DKIOCG_VIRTGEOM,	sizeof (vd_geom_t),
6639 		vdc_get_geom_convert},
6640 	{VD_OP_SET_DISKGEOM,	DKIOCSGEOM,		sizeof (vd_geom_t),
6641 		vdc_set_geom_convert},
6642 	{VD_OP_GET_EFI,		DKIOCGETEFI,		0,
6643 		vdc_get_efi_convert},
6644 	{VD_OP_SET_EFI,		DKIOCSETEFI,		0,
6645 		vdc_set_efi_convert},
6646 
6647 	/* DIOCTL_RWCMD is converted to a read or a write */
6648 	{0, DIOCTL_RWCMD,  sizeof (struct dadkio_rwcmd), NULL},
6649 
6650 	/* mhd(7I) non-shared multihost disks ioctls */
6651 	{0, MHIOCTKOWN,				0, vdc_null_copy_func},
6652 	{0, MHIOCRELEASE,			0, vdc_null_copy_func},
6653 	{0, MHIOCSTATUS,			0, vdc_null_copy_func},
6654 	{0, MHIOCQRESERVE,			0, vdc_null_copy_func},
6655 
6656 	/* mhd(7I) shared multihost disks ioctls */
6657 	{0, MHIOCGRP_INKEYS,			0, vdc_null_copy_func},
6658 	{0, MHIOCGRP_INRESV,			0, vdc_null_copy_func},
6659 	{0, MHIOCGRP_REGISTER,			0, vdc_null_copy_func},
6660 	{0, MHIOCGRP_RESERVE, 			0, vdc_null_copy_func},
6661 	{0, MHIOCGRP_PREEMPTANDABORT,		0, vdc_null_copy_func},
6662 	{0, MHIOCGRP_REGISTERANDIGNOREKEY,	0, vdc_null_copy_func},
6663 
6664 	/* mhd(7I) failfast ioctl */
6665 	{0, MHIOCENFAILFAST,			0, vdc_null_copy_func},
6666 
6667 	/*
6668 	 * These particular ioctls are not sent to the server - vdc fakes up
6669 	 * the necessary info.
6670 	 */
6671 	{0, DKIOCINFO, sizeof (struct dk_cinfo), vdc_null_copy_func},
6672 	{0, DKIOCGMEDIAINFO, sizeof (struct dk_minfo), vdc_null_copy_func},
6673 	{0, USCSICMD,	sizeof (struct uscsi_cmd), vdc_null_copy_func},
6674 	{0, DKIOCPARTITION, 0, vdc_null_copy_func },
6675 	{0, DKIOCGAPART, 0, vdc_null_copy_func },
6676 	{0, DKIOCREMOVABLE, 0, vdc_null_copy_func},
6677 	{0, CDROMREADOFFSET, 0, vdc_null_copy_func}
6678 };
6679 
6680 /*
6681  * This function handles ioctl requests from the vd_efi_alloc_and_read()
6682  * function and forward them to the vdisk.
6683  */
6684 static int
6685 vd_process_efi_ioctl(void *vdisk, int cmd, uintptr_t arg)
6686 {
6687 	vdc_t *vdc = (vdc_t *)vdisk;
6688 	dev_t dev;
6689 	int rval;
6690 
6691 	dev = makedevice(ddi_driver_major(vdc->dip),
6692 	    VD_MAKE_DEV(vdc->instance, 0));
6693 
6694 	return (vd_process_ioctl(dev, cmd, (caddr_t)arg, FKIOCTL, &rval));
6695 }
6696 
6697 /*
6698  * Function:
6699  *	vd_process_ioctl()
6700  *
6701  * Description:
6702  *	This routine processes disk specific ioctl calls
6703  *
6704  * Arguments:
6705  *	dev	- the device number
6706  *	cmd	- the operation [dkio(7I)] to be processed
6707  *	arg	- pointer to user provided structure
6708  *		  (contains data to be set or reference parameter for get)
6709  *	mode	- bit flag, indicating open settings, 32/64 bit type, etc
6710  *	rvalp	- pointer to return value for calling process.
6711  *
6712  * Return Code:
6713  *	0
6714  *	EFAULT
6715  *	ENXIO
6716  *	EIO
6717  *	ENOTSUP
6718  */
6719 static int
6720 vd_process_ioctl(dev_t dev, int cmd, caddr_t arg, int mode, int *rvalp)
6721 {
6722 	int		instance = VDCUNIT(dev);
6723 	vdc_t		*vdc = NULL;
6724 	int		rv = -1;
6725 	int		idx = 0;		/* index into dk_ioctl[] */
6726 	size_t		len = 0;		/* #bytes to send to vds */
6727 	size_t		alloc_len = 0;		/* #bytes to allocate mem for */
6728 	caddr_t		mem_p = NULL;
6729 	size_t		nioctls = (sizeof (dk_ioctl)) / (sizeof (dk_ioctl[0]));
6730 	vdc_dk_ioctl_t	*iop;
6731 
6732 	vdc = ddi_get_soft_state(vdc_state, instance);
6733 	if (vdc == NULL) {
6734 		cmn_err(CE_NOTE, "![%d] Could not get soft state structure",
6735 		    instance);
6736 		return (ENXIO);
6737 	}
6738 
6739 	DMSG(vdc, 0, "[%d] Processing ioctl(%x) for dev %lx : model %x\n",
6740 	    instance, cmd, dev, ddi_model_convert_from(mode & FMODELS));
6741 
6742 	if (rvalp != NULL) {
6743 		/* the return value of the ioctl is 0 by default */
6744 		*rvalp = 0;
6745 	}
6746 
6747 	/*
6748 	 * Validate the ioctl operation to be performed.
6749 	 *
6750 	 * If we have looped through the array without finding a match then we
6751 	 * don't support this ioctl.
6752 	 */
6753 	for (idx = 0; idx < nioctls; idx++) {
6754 		if (cmd == dk_ioctl[idx].cmd)
6755 			break;
6756 	}
6757 
6758 	if (idx >= nioctls) {
6759 		DMSG(vdc, 0, "[%d] Unsupported ioctl (0x%x)\n",
6760 		    vdc->instance, cmd);
6761 		return (ENOTSUP);
6762 	}
6763 
6764 	iop = &(dk_ioctl[idx]);
6765 
6766 	if (cmd == DKIOCGETEFI || cmd == DKIOCSETEFI) {
6767 		/* size is not fixed for EFI ioctls, it depends on ioctl arg */
6768 		dk_efi_t	dk_efi;
6769 
6770 		rv = ddi_copyin(arg, &dk_efi, sizeof (dk_efi_t), mode);
6771 		if (rv != 0)
6772 			return (EFAULT);
6773 
6774 		len = sizeof (vd_efi_t) - 1 + dk_efi.dki_length;
6775 	} else {
6776 		len = iop->nbytes;
6777 	}
6778 
6779 	/* check if the ioctl is applicable */
6780 	switch (cmd) {
6781 	case CDROMREADOFFSET:
6782 	case DKIOCREMOVABLE:
6783 		return (ENOTTY);
6784 
6785 	case USCSICMD:
6786 	case MHIOCTKOWN:
6787 	case MHIOCSTATUS:
6788 	case MHIOCQRESERVE:
6789 	case MHIOCRELEASE:
6790 	case MHIOCGRP_INKEYS:
6791 	case MHIOCGRP_INRESV:
6792 	case MHIOCGRP_REGISTER:
6793 	case MHIOCGRP_RESERVE:
6794 	case MHIOCGRP_PREEMPTANDABORT:
6795 	case MHIOCGRP_REGISTERANDIGNOREKEY:
6796 	case MHIOCENFAILFAST:
6797 		if (vdc->cinfo == NULL)
6798 			return (ENXIO);
6799 		if (vdc->cinfo->dki_ctype != DKC_SCSI_CCS)
6800 			return (ENOTTY);
6801 		break;
6802 
6803 	case DIOCTL_RWCMD:
6804 		if (vdc->cinfo == NULL)
6805 			return (ENXIO);
6806 		if (vdc->cinfo->dki_ctype != DKC_DIRECT)
6807 			return (ENOTTY);
6808 		break;
6809 
6810 	case DKIOCINFO:
6811 		if (vdc->cinfo == NULL)
6812 			return (ENXIO);
6813 		break;
6814 
6815 	case DKIOCGMEDIAINFO:
6816 		if (vdc->minfo == NULL)
6817 			return (ENXIO);
6818 		if (vdc_check_capacity(vdc) != 0)
6819 			/* disk capacity is not available */
6820 			return (EIO);
6821 		break;
6822 	}
6823 
6824 	/*
6825 	 * Deal with ioctls which require a processing different than
6826 	 * converting ioctl arguments and sending a corresponding
6827 	 * VD operation.
6828 	 */
6829 	switch (cmd) {
6830 
6831 	case USCSICMD:
6832 	{
6833 		return (vdc_uscsi_cmd(vdc, arg, mode));
6834 	}
6835 
6836 	case MHIOCTKOWN:
6837 	{
6838 		mutex_enter(&vdc->ownership_lock);
6839 		/*
6840 		 * We have to set VDC_OWNERSHIP_WANTED now so that the ownership
6841 		 * can be flagged with VDC_OWNERSHIP_RESET if the LDC is reset
6842 		 * while we are processing the ioctl.
6843 		 */
6844 		vdc_ownership_update(vdc, VDC_OWNERSHIP_WANTED);
6845 
6846 		rv = vdc_access_set(vdc, VD_ACCESS_SET_EXCLUSIVE |
6847 		    VD_ACCESS_SET_PREEMPT | VD_ACCESS_SET_PRESERVE, mode);
6848 		if (rv == 0) {
6849 			vdc_ownership_update(vdc, VDC_OWNERSHIP_WANTED |
6850 			    VDC_OWNERSHIP_GRANTED);
6851 		} else {
6852 			vdc_ownership_update(vdc, VDC_OWNERSHIP_NONE);
6853 		}
6854 		mutex_exit(&vdc->ownership_lock);
6855 		return (rv);
6856 	}
6857 
6858 	case MHIOCRELEASE:
6859 	{
6860 		mutex_enter(&vdc->ownership_lock);
6861 		rv = vdc_access_set(vdc, VD_ACCESS_SET_CLEAR, mode);
6862 		if (rv == 0) {
6863 			vdc_ownership_update(vdc, VDC_OWNERSHIP_NONE);
6864 		}
6865 		mutex_exit(&vdc->ownership_lock);
6866 		return (rv);
6867 	}
6868 
6869 	case MHIOCSTATUS:
6870 	{
6871 		uint64_t status;
6872 
6873 		rv = vdc_access_get(vdc, &status, mode);
6874 		if (rv == 0 && rvalp != NULL)
6875 			*rvalp = (status & VD_ACCESS_ALLOWED)? 0 : 1;
6876 		return (rv);
6877 	}
6878 
6879 	case MHIOCQRESERVE:
6880 	{
6881 		rv = vdc_access_set(vdc, VD_ACCESS_SET_EXCLUSIVE, mode);
6882 		return (rv);
6883 	}
6884 
6885 	case MHIOCGRP_INKEYS:
6886 	{
6887 		return (vdc_mhd_inkeys(vdc, arg, mode));
6888 	}
6889 
6890 	case MHIOCGRP_INRESV:
6891 	{
6892 		return (vdc_mhd_inresv(vdc, arg, mode));
6893 	}
6894 
6895 	case MHIOCGRP_REGISTER:
6896 	{
6897 		return (vdc_mhd_register(vdc, arg, mode));
6898 	}
6899 
6900 	case MHIOCGRP_RESERVE:
6901 	{
6902 		return (vdc_mhd_reserve(vdc, arg, mode));
6903 	}
6904 
6905 	case MHIOCGRP_PREEMPTANDABORT:
6906 	{
6907 		return (vdc_mhd_preemptabort(vdc, arg, mode));
6908 	}
6909 
6910 	case MHIOCGRP_REGISTERANDIGNOREKEY:
6911 	{
6912 		return (vdc_mhd_registerignore(vdc, arg, mode));
6913 	}
6914 
6915 	case MHIOCENFAILFAST:
6916 	{
6917 		rv = vdc_failfast(vdc, arg, mode);
6918 		return (rv);
6919 	}
6920 
6921 	case DIOCTL_RWCMD:
6922 	{
6923 		return (vdc_dioctl_rwcmd(dev, arg, mode));
6924 	}
6925 
6926 	case DKIOCGAPART:
6927 	{
6928 		return (vdc_dkio_gapart(vdc, arg, mode));
6929 	}
6930 
6931 	case DKIOCPARTITION:
6932 	{
6933 		return (vdc_dkio_partition(vdc, arg, mode));
6934 	}
6935 
6936 	case DKIOCINFO:
6937 	{
6938 		struct dk_cinfo	cinfo;
6939 
6940 		bcopy(vdc->cinfo, &cinfo, sizeof (struct dk_cinfo));
6941 		cinfo.dki_partition = VDCPART(dev);
6942 
6943 		rv = ddi_copyout(&cinfo, (void *)arg,
6944 		    sizeof (struct dk_cinfo), mode);
6945 		if (rv != 0)
6946 			return (EFAULT);
6947 
6948 		return (0);
6949 	}
6950 
6951 	case DKIOCGMEDIAINFO:
6952 	{
6953 		ASSERT(vdc->vdisk_size != 0);
6954 		if (vdc->minfo->dki_capacity == 0)
6955 			vdc->minfo->dki_capacity = vdc->vdisk_size;
6956 		rv = ddi_copyout(vdc->minfo, (void *)arg,
6957 		    sizeof (struct dk_minfo), mode);
6958 		if (rv != 0)
6959 			return (EFAULT);
6960 
6961 		return (0);
6962 	}
6963 
6964 	case DKIOCFLUSHWRITECACHE:
6965 		{
6966 			struct dk_callback *dkc =
6967 			    (struct dk_callback *)(uintptr_t)arg;
6968 			vdc_dk_arg_t	*dkarg = NULL;
6969 
6970 			DMSG(vdc, 1, "[%d] Flush W$: mode %x\n",
6971 			    instance, mode);
6972 
6973 			/*
6974 			 * If arg is NULL, then there is no callback function
6975 			 * registered and the call operates synchronously; we
6976 			 * break and continue with the rest of the function and
6977 			 * wait for vds to return (i.e. after the request to
6978 			 * vds returns successfully, all writes completed prior
6979 			 * to the ioctl will have been flushed from the disk
6980 			 * write cache to persistent media.
6981 			 *
6982 			 * If a callback function is registered, we dispatch
6983 			 * the request on a task queue and return immediately.
6984 			 * The callback will deal with informing the calling
6985 			 * thread that the flush request is completed.
6986 			 */
6987 			if (dkc == NULL)
6988 				break;
6989 
6990 			/*
6991 			 * the asynchronous callback is only supported if
6992 			 * invoked from within the kernel
6993 			 */
6994 			if ((mode & FKIOCTL) == 0)
6995 				return (ENOTSUP);
6996 
6997 			dkarg = kmem_zalloc(sizeof (vdc_dk_arg_t), KM_SLEEP);
6998 
6999 			dkarg->mode = mode;
7000 			dkarg->dev = dev;
7001 			bcopy(dkc, &dkarg->dkc, sizeof (*dkc));
7002 
7003 			mutex_enter(&vdc->lock);
7004 			vdc->dkio_flush_pending++;
7005 			dkarg->vdc = vdc;
7006 			mutex_exit(&vdc->lock);
7007 
7008 			/* put the request on a task queue */
7009 			rv = taskq_dispatch(system_taskq, vdc_dkio_flush_cb,
7010 			    (void *)dkarg, DDI_SLEEP);
7011 			if (rv == NULL) {
7012 				/* clean up if dispatch fails */
7013 				mutex_enter(&vdc->lock);
7014 				vdc->dkio_flush_pending--;
7015 				mutex_exit(&vdc->lock);
7016 				kmem_free(dkarg, sizeof (vdc_dk_arg_t));
7017 			}
7018 
7019 			return (rv == NULL ? ENOMEM : 0);
7020 		}
7021 	}
7022 
7023 	/* catch programming error in vdc - should be a VD_OP_XXX ioctl */
7024 	ASSERT(iop->op != 0);
7025 
7026 	/* check if the vDisk server handles the operation for this vDisk */
7027 	if (VD_OP_SUPPORTED(vdc->operations, iop->op) == B_FALSE) {
7028 		DMSG(vdc, 0, "[%d] Unsupported VD_OP operation (0x%x)\n",
7029 		    vdc->instance, iop->op);
7030 		return (ENOTSUP);
7031 	}
7032 
7033 	/* LDC requires that the memory being mapped is 8-byte aligned */
7034 	alloc_len = P2ROUNDUP(len, sizeof (uint64_t));
7035 	DMSG(vdc, 1, "[%d] struct size %ld alloc %ld\n",
7036 	    instance, len, alloc_len);
7037 
7038 	if (alloc_len > 0)
7039 		mem_p = kmem_zalloc(alloc_len, KM_SLEEP);
7040 
7041 	/*
7042 	 * Call the conversion function for this ioctl which, if necessary,
7043 	 * converts from the Solaris format to the format ARC'ed
7044 	 * as part of the vDisk protocol (FWARC 2006/195)
7045 	 */
7046 	ASSERT(iop->convert != NULL);
7047 	rv = (iop->convert)(vdc, arg, mem_p, mode, VD_COPYIN);
7048 	if (rv != 0) {
7049 		DMSG(vdc, 0, "[%d] convert func returned %d for ioctl 0x%x\n",
7050 		    instance, rv, cmd);
7051 		if (mem_p != NULL)
7052 			kmem_free(mem_p, alloc_len);
7053 		return (rv);
7054 	}
7055 
7056 	/*
7057 	 * send request to vds to service the ioctl.
7058 	 */
7059 	rv = vdc_do_sync_op(vdc, iop->op, mem_p, alloc_len,
7060 	    VDCPART(dev), 0, CB_SYNC, (void *)(uint64_t)mode,
7061 	    VIO_both_dir, B_TRUE);
7062 
7063 	if (rv != 0) {
7064 		/*
7065 		 * This is not necessarily an error. The ioctl could
7066 		 * be returning a value such as ENOTTY to indicate
7067 		 * that the ioctl is not applicable.
7068 		 */
7069 		DMSG(vdc, 0, "[%d] vds returned %d for ioctl 0x%x\n",
7070 		    instance, rv, cmd);
7071 		if (mem_p != NULL)
7072 			kmem_free(mem_p, alloc_len);
7073 
7074 		return (rv);
7075 	}
7076 
7077 	/*
7078 	 * Call the conversion function (if it exists) for this ioctl
7079 	 * which converts from the format ARC'ed as part of the vDisk
7080 	 * protocol (FWARC 2006/195) back to a format understood by
7081 	 * the rest of Solaris.
7082 	 */
7083 	rv = (iop->convert)(vdc, mem_p, arg, mode, VD_COPYOUT);
7084 	if (rv != 0) {
7085 		DMSG(vdc, 0, "[%d] convert func returned %d for ioctl 0x%x\n",
7086 		    instance, rv, cmd);
7087 		if (mem_p != NULL)
7088 			kmem_free(mem_p, alloc_len);
7089 		return (rv);
7090 	}
7091 
7092 	if (mem_p != NULL)
7093 		kmem_free(mem_p, alloc_len);
7094 
7095 	return (rv);
7096 }
7097 
7098 /*
7099  * Function:
7100  *
7101  * Description:
7102  *	This is an empty conversion function used by ioctl calls which
7103  *	do not need to convert the data being passed in/out to userland
7104  */
7105 static int
7106 vdc_null_copy_func(vdc_t *vdc, void *from, void *to, int mode, int dir)
7107 {
7108 	_NOTE(ARGUNUSED(vdc))
7109 	_NOTE(ARGUNUSED(from))
7110 	_NOTE(ARGUNUSED(to))
7111 	_NOTE(ARGUNUSED(mode))
7112 	_NOTE(ARGUNUSED(dir))
7113 
7114 	return (0);
7115 }
7116 
7117 static int
7118 vdc_get_wce_convert(vdc_t *vdc, void *from, void *to,
7119     int mode, int dir)
7120 {
7121 	_NOTE(ARGUNUSED(vdc))
7122 
7123 	if (dir == VD_COPYIN)
7124 		return (0);		/* nothing to do */
7125 
7126 	if (ddi_copyout(from, to, sizeof (int), mode) != 0)
7127 		return (EFAULT);
7128 
7129 	return (0);
7130 }
7131 
7132 static int
7133 vdc_set_wce_convert(vdc_t *vdc, void *from, void *to,
7134     int mode, int dir)
7135 {
7136 	_NOTE(ARGUNUSED(vdc))
7137 
7138 	if (dir == VD_COPYOUT)
7139 		return (0);		/* nothing to do */
7140 
7141 	if (ddi_copyin(from, to, sizeof (int), mode) != 0)
7142 		return (EFAULT);
7143 
7144 	return (0);
7145 }
7146 
7147 /*
7148  * Function:
7149  *	vdc_get_vtoc_convert()
7150  *
7151  * Description:
7152  *	This routine performs the necessary convertions from the DKIOCGVTOC
7153  *	Solaris structure to the format defined in FWARC 2006/195.
7154  *
7155  *	In the struct vtoc definition, the timestamp field is marked as not
7156  *	supported so it is not part of vDisk protocol (FWARC 2006/195).
7157  *	However SVM uses that field to check it can write into the VTOC,
7158  *	so we fake up the info of that field.
7159  *
7160  * Arguments:
7161  *	vdc	- the vDisk client
7162  *	from	- the buffer containing the data to be copied from
7163  *	to	- the buffer to be copied to
7164  *	mode	- flags passed to ioctl() call
7165  *	dir	- the "direction" of the copy - VD_COPYIN or VD_COPYOUT
7166  *
7167  * Return Code:
7168  *	0	- Success
7169  *	ENXIO	- incorrect buffer passed in.
7170  *	EFAULT	- ddi_copyout routine encountered an error.
7171  */
7172 static int
7173 vdc_get_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7174 {
7175 	int		i;
7176 	void		*tmp_mem = NULL;
7177 	void		*tmp_memp;
7178 	struct vtoc	vt;
7179 	struct vtoc32	vt32;
7180 	int		copy_len = 0;
7181 	int		rv = 0;
7182 
7183 	if (dir != VD_COPYOUT)
7184 		return (0);	/* nothing to do */
7185 
7186 	if ((from == NULL) || (to == NULL))
7187 		return (ENXIO);
7188 
7189 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32)
7190 		copy_len = sizeof (struct vtoc32);
7191 	else
7192 		copy_len = sizeof (struct vtoc);
7193 
7194 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
7195 
7196 	VD_VTOC2VTOC((vd_vtoc_t *)from, &vt);
7197 
7198 	/* fake the VTOC timestamp field */
7199 	for (i = 0; i < V_NUMPAR; i++) {
7200 		vt.timestamp[i] = vdc->vtoc->timestamp[i];
7201 	}
7202 
7203 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
7204 		/* LINTED E_ASSIGN_NARROW_CONV */
7205 		vtoctovtoc32(vt, vt32);
7206 		tmp_memp = &vt32;
7207 	} else {
7208 		tmp_memp = &vt;
7209 	}
7210 	rv = ddi_copyout(tmp_memp, to, copy_len, mode);
7211 	if (rv != 0)
7212 		rv = EFAULT;
7213 
7214 	kmem_free(tmp_mem, copy_len);
7215 	return (rv);
7216 }
7217 
7218 /*
7219  * Function:
7220  *	vdc_set_vtoc_convert()
7221  *
7222  * Description:
7223  *	This routine performs the necessary convertions from the DKIOCSVTOC
7224  *	Solaris structure to the format defined in FWARC 2006/195.
7225  *
7226  * Arguments:
7227  *	vdc	- the vDisk client
7228  *	from	- Buffer with data
7229  *	to	- Buffer where data is to be copied to
7230  *	mode	- flags passed to ioctl
7231  *	dir	- direction of copy (in or out)
7232  *
7233  * Return Code:
7234  *	0	- Success
7235  *	ENXIO	- Invalid buffer passed in
7236  *	EFAULT	- ddi_copyin of data failed
7237  */
7238 static int
7239 vdc_set_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7240 {
7241 	_NOTE(ARGUNUSED(vdc))
7242 
7243 	void		*tmp_mem = NULL, *uvtoc;
7244 	struct vtoc	vt;
7245 	struct vtoc	*vtp = &vt;
7246 	vd_vtoc_t	vtvd;
7247 	int		copy_len = 0;
7248 	int		i, rv = 0;
7249 
7250 	if ((from == NULL) || (to == NULL))
7251 		return (ENXIO);
7252 
7253 	if (dir == VD_COPYIN)
7254 		uvtoc = from;
7255 	else
7256 		uvtoc = to;
7257 
7258 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32)
7259 		copy_len = sizeof (struct vtoc32);
7260 	else
7261 		copy_len = sizeof (struct vtoc);
7262 
7263 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
7264 
7265 	rv = ddi_copyin(uvtoc, tmp_mem, copy_len, mode);
7266 	if (rv != 0) {
7267 		kmem_free(tmp_mem, copy_len);
7268 		return (EFAULT);
7269 	}
7270 
7271 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
7272 		vtoc32tovtoc((*(struct vtoc32 *)tmp_mem), vt);
7273 	} else {
7274 		vtp = tmp_mem;
7275 	}
7276 
7277 	if (dir == VD_COPYOUT) {
7278 		/*
7279 		 * The disk label may have changed. Revalidate the disk
7280 		 * geometry. This will also update the device nodes.
7281 		 */
7282 		vdc_validate(vdc);
7283 
7284 		/*
7285 		 * We also need to keep track of the timestamp fields.
7286 		 */
7287 		for (i = 0; i < V_NUMPAR; i++) {
7288 			vdc->vtoc->timestamp[i] = vtp->timestamp[i];
7289 		}
7290 
7291 		return (0);
7292 	}
7293 
7294 	VTOC2VD_VTOC(vtp, &vtvd);
7295 	bcopy(&vtvd, to, sizeof (vd_vtoc_t));
7296 	kmem_free(tmp_mem, copy_len);
7297 
7298 	return (0);
7299 }
7300 
7301 /*
7302  * Function:
7303  *	vdc_get_geom_convert()
7304  *
7305  * Description:
7306  *	This routine performs the necessary convertions from the DKIOCGGEOM,
7307  *	DKIOCG_PHYSGEOM and DKIOG_VIRTGEOM Solaris structures to the format
7308  *	defined in FWARC 2006/195
7309  *
7310  * Arguments:
7311  *	vdc	- the vDisk client
7312  *	from	- Buffer with data
7313  *	to	- Buffer where data is to be copied to
7314  *	mode	- flags passed to ioctl
7315  *	dir	- direction of copy (in or out)
7316  *
7317  * Return Code:
7318  *	0	- Success
7319  *	ENXIO	- Invalid buffer passed in
7320  *	EFAULT	- ddi_copyout of data failed
7321  */
7322 static int
7323 vdc_get_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7324 {
7325 	_NOTE(ARGUNUSED(vdc))
7326 
7327 	struct dk_geom	geom;
7328 	int	copy_len = sizeof (struct dk_geom);
7329 	int	rv = 0;
7330 
7331 	if (dir != VD_COPYOUT)
7332 		return (0);	/* nothing to do */
7333 
7334 	if ((from == NULL) || (to == NULL))
7335 		return (ENXIO);
7336 
7337 	VD_GEOM2DK_GEOM((vd_geom_t *)from, &geom);
7338 	rv = ddi_copyout(&geom, to, copy_len, mode);
7339 	if (rv != 0)
7340 		rv = EFAULT;
7341 
7342 	return (rv);
7343 }
7344 
7345 /*
7346  * Function:
7347  *	vdc_set_geom_convert()
7348  *
7349  * Description:
7350  *	This routine performs the necessary convertions from the DKIOCSGEOM
7351  *	Solaris structure to the format defined in FWARC 2006/195.
7352  *
7353  * Arguments:
7354  *	vdc	- the vDisk client
7355  *	from	- Buffer with data
7356  *	to	- Buffer where data is to be copied to
7357  *	mode	- flags passed to ioctl
7358  *	dir	- direction of copy (in or out)
7359  *
7360  * Return Code:
7361  *	0	- Success
7362  *	ENXIO	- Invalid buffer passed in
7363  *	EFAULT	- ddi_copyin of data failed
7364  */
7365 static int
7366 vdc_set_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7367 {
7368 	_NOTE(ARGUNUSED(vdc))
7369 
7370 	vd_geom_t	vdgeom;
7371 	void		*tmp_mem = NULL;
7372 	int		copy_len = sizeof (struct dk_geom);
7373 	int		rv = 0;
7374 
7375 	if (dir != VD_COPYIN)
7376 		return (0);	/* nothing to do */
7377 
7378 	if ((from == NULL) || (to == NULL))
7379 		return (ENXIO);
7380 
7381 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
7382 
7383 	rv = ddi_copyin(from, tmp_mem, copy_len, mode);
7384 	if (rv != 0) {
7385 		kmem_free(tmp_mem, copy_len);
7386 		return (EFAULT);
7387 	}
7388 	DK_GEOM2VD_GEOM((struct dk_geom *)tmp_mem, &vdgeom);
7389 	bcopy(&vdgeom, to, sizeof (vdgeom));
7390 	kmem_free(tmp_mem, copy_len);
7391 
7392 	return (0);
7393 }
7394 
7395 static int
7396 vdc_get_efi_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7397 {
7398 	_NOTE(ARGUNUSED(vdc))
7399 
7400 	vd_efi_t	*vd_efi;
7401 	dk_efi_t	dk_efi;
7402 	int		rv = 0;
7403 	void		*uaddr;
7404 
7405 	if ((from == NULL) || (to == NULL))
7406 		return (ENXIO);
7407 
7408 	if (dir == VD_COPYIN) {
7409 
7410 		vd_efi = (vd_efi_t *)to;
7411 
7412 		rv = ddi_copyin(from, &dk_efi, sizeof (dk_efi_t), mode);
7413 		if (rv != 0)
7414 			return (EFAULT);
7415 
7416 		vd_efi->lba = dk_efi.dki_lba;
7417 		vd_efi->length = dk_efi.dki_length;
7418 		bzero(vd_efi->data, vd_efi->length);
7419 
7420 	} else {
7421 
7422 		rv = ddi_copyin(to, &dk_efi, sizeof (dk_efi_t), mode);
7423 		if (rv != 0)
7424 			return (EFAULT);
7425 
7426 		uaddr = dk_efi.dki_data;
7427 
7428 		dk_efi.dki_data = kmem_alloc(dk_efi.dki_length, KM_SLEEP);
7429 
7430 		VD_EFI2DK_EFI((vd_efi_t *)from, &dk_efi);
7431 
7432 		rv = ddi_copyout(dk_efi.dki_data, uaddr, dk_efi.dki_length,
7433 		    mode);
7434 		if (rv != 0)
7435 			return (EFAULT);
7436 
7437 		kmem_free(dk_efi.dki_data, dk_efi.dki_length);
7438 	}
7439 
7440 	return (0);
7441 }
7442 
7443 static int
7444 vdc_set_efi_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7445 {
7446 	_NOTE(ARGUNUSED(vdc))
7447 
7448 	dk_efi_t	dk_efi;
7449 	void		*uaddr;
7450 
7451 	if (dir == VD_COPYOUT) {
7452 		/*
7453 		 * The disk label may have changed. Revalidate the disk
7454 		 * geometry. This will also update the device nodes.
7455 		 */
7456 		vdc_validate(vdc);
7457 		return (0);
7458 	}
7459 
7460 	if ((from == NULL) || (to == NULL))
7461 		return (ENXIO);
7462 
7463 	if (ddi_copyin(from, &dk_efi, sizeof (dk_efi_t), mode) != 0)
7464 		return (EFAULT);
7465 
7466 	uaddr = dk_efi.dki_data;
7467 
7468 	dk_efi.dki_data = kmem_alloc(dk_efi.dki_length, KM_SLEEP);
7469 
7470 	if (ddi_copyin(uaddr, dk_efi.dki_data, dk_efi.dki_length, mode) != 0)
7471 		return (EFAULT);
7472 
7473 	DK_EFI2VD_EFI(&dk_efi, (vd_efi_t *)to);
7474 
7475 	kmem_free(dk_efi.dki_data, dk_efi.dki_length);
7476 
7477 	return (0);
7478 }
7479 
7480 
7481 /* -------------------------------------------------------------------------- */
7482 
7483 /*
7484  * Function:
7485  *	vdc_create_fake_geometry()
7486  *
7487  * Description:
7488  *	This routine fakes up the disk info needed for some DKIO ioctls such
7489  *	as DKIOCINFO and DKIOCGMEDIAINFO [just like lofi(7D) and ramdisk(7D) do]
7490  *
7491  *	Note: This function must not be called until the vDisk attributes have
7492  *	been exchanged as part of the handshake with the vDisk server.
7493  *
7494  * Arguments:
7495  *	vdc	- soft state pointer for this instance of the device driver.
7496  *
7497  * Return Code:
7498  *	none.
7499  */
7500 static void
7501 vdc_create_fake_geometry(vdc_t *vdc)
7502 {
7503 	ASSERT(vdc != NULL);
7504 	ASSERT(vdc->max_xfer_sz != 0);
7505 
7506 	/*
7507 	 * DKIOCINFO support
7508 	 */
7509 	if (vdc->cinfo == NULL)
7510 		vdc->cinfo = kmem_zalloc(sizeof (struct dk_cinfo), KM_SLEEP);
7511 
7512 	(void) strcpy(vdc->cinfo->dki_cname, VDC_DRIVER_NAME);
7513 	(void) strcpy(vdc->cinfo->dki_dname, VDC_DRIVER_NAME);
7514 	/* max_xfer_sz is #blocks so we don't need to divide by DEV_BSIZE */
7515 	vdc->cinfo->dki_maxtransfer = vdc->max_xfer_sz;
7516 
7517 	/*
7518 	 * We set the controller type to DKC_SCSI_CCS only if the VD_OP_SCSICMD
7519 	 * operation is supported, otherwise the controller type is DKC_DIRECT.
7520 	 * Version 1.0 does not support the VD_OP_SCSICMD operation, so the
7521 	 * controller type is always DKC_DIRECT in that case.
7522 	 *
7523 	 * If the virtual disk is backed by a physical CD/DVD device or
7524 	 * an ISO image, modify the controller type to indicate this
7525 	 */
7526 	switch (vdc->vdisk_media) {
7527 	case VD_MEDIA_CD:
7528 	case VD_MEDIA_DVD:
7529 		vdc->cinfo->dki_ctype = DKC_CDROM;
7530 		break;
7531 	case VD_MEDIA_FIXED:
7532 		if (VD_OP_SUPPORTED(vdc->operations, VD_OP_SCSICMD))
7533 			vdc->cinfo->dki_ctype = DKC_SCSI_CCS;
7534 		else
7535 			vdc->cinfo->dki_ctype = DKC_DIRECT;
7536 		break;
7537 	default:
7538 		/* in the case of v1.0 we default to a fixed disk */
7539 		vdc->cinfo->dki_ctype = DKC_DIRECT;
7540 		break;
7541 	}
7542 	vdc->cinfo->dki_flags = DKI_FMTVOL;
7543 	vdc->cinfo->dki_cnum = 0;
7544 	vdc->cinfo->dki_addr = 0;
7545 	vdc->cinfo->dki_space = 0;
7546 	vdc->cinfo->dki_prio = 0;
7547 	vdc->cinfo->dki_vec = 0;
7548 	vdc->cinfo->dki_unit = vdc->instance;
7549 	vdc->cinfo->dki_slave = 0;
7550 	/*
7551 	 * The partition number will be created on the fly depending on the
7552 	 * actual slice (i.e. minor node) that is used to request the data.
7553 	 */
7554 	vdc->cinfo->dki_partition = 0;
7555 
7556 	/*
7557 	 * DKIOCGMEDIAINFO support
7558 	 */
7559 	if (vdc->minfo == NULL)
7560 		vdc->minfo = kmem_zalloc(sizeof (struct dk_minfo), KM_SLEEP);
7561 
7562 	if (vio_ver_is_supported(vdc->ver, 1, 1)) {
7563 		vdc->minfo->dki_media_type =
7564 		    VD_MEDIATYPE2DK_MEDIATYPE(vdc->vdisk_media);
7565 	} else {
7566 		vdc->minfo->dki_media_type = DK_FIXED_DISK;
7567 	}
7568 
7569 	vdc->minfo->dki_capacity = vdc->vdisk_size;
7570 	vdc->minfo->dki_lbsize = vdc->block_size;
7571 }
7572 
7573 static ushort_t
7574 vdc_lbl2cksum(struct dk_label *label)
7575 {
7576 	int	count;
7577 	ushort_t sum, *sp;
7578 
7579 	count =	(sizeof (struct dk_label)) / (sizeof (short)) - 1;
7580 	sp = (ushort_t *)label;
7581 	sum = 0;
7582 	while (count--) {
7583 		sum ^= *sp++;
7584 	}
7585 
7586 	return (sum);
7587 }
7588 
7589 /*
7590  * Function:
7591  *	vdc_validate_geometry
7592  *
7593  * Description:
7594  *	This routine discovers the label and geometry of the disk. It stores
7595  *	the disk label and related information in the vdc structure. If it
7596  *	fails to validate the geometry or to discover the disk label then
7597  *	the label is marked as unknown (VD_DISK_LABEL_UNK).
7598  *
7599  * Arguments:
7600  *	vdc	- soft state pointer for this instance of the device driver.
7601  *
7602  * Return Code:
7603  *	0	- success.
7604  *	EINVAL	- unknown disk label.
7605  *	ENOTSUP	- geometry not applicable (EFI label).
7606  *	EIO	- error accessing the disk.
7607  */
7608 static int
7609 vdc_validate_geometry(vdc_t *vdc)
7610 {
7611 	buf_t	*buf;	/* BREAD requests need to be in a buf_t structure */
7612 	dev_t	dev;
7613 	int	rv, rval;
7614 	struct dk_label label;
7615 	struct dk_geom geom;
7616 	struct vtoc vtoc;
7617 	efi_gpt_t *gpt;
7618 	efi_gpe_t *gpe;
7619 	vd_efi_dev_t edev;
7620 
7621 	ASSERT(vdc != NULL);
7622 	ASSERT(vdc->vtoc != NULL && vdc->geom != NULL);
7623 	ASSERT(MUTEX_HELD(&vdc->lock));
7624 
7625 	mutex_exit(&vdc->lock);
7626 
7627 	dev = makedevice(ddi_driver_major(vdc->dip),
7628 	    VD_MAKE_DEV(vdc->instance, 0));
7629 
7630 	rv = vd_process_ioctl(dev, DKIOCGGEOM, (caddr_t)&geom, FKIOCTL, &rval);
7631 	if (rv == 0)
7632 		rv = vd_process_ioctl(dev, DKIOCGVTOC, (caddr_t)&vtoc,
7633 		    FKIOCTL, &rval);
7634 
7635 	if (rv == ENOTSUP) {
7636 		/*
7637 		 * If the device does not support VTOC then we try
7638 		 * to read an EFI label.
7639 		 *
7640 		 * We need to know the block size and the disk size to
7641 		 * be able to read an EFI label.
7642 		 */
7643 		if (vdc->vdisk_size == 0) {
7644 			if ((rv = vdc_check_capacity(vdc)) != 0) {
7645 				mutex_enter(&vdc->lock);
7646 				vdc_store_label_unk(vdc);
7647 				return (rv);
7648 			}
7649 		}
7650 
7651 		VD_EFI_DEV_SET(edev, vdc, vd_process_efi_ioctl);
7652 
7653 		rv = vd_efi_alloc_and_read(&edev, &gpt, &gpe);
7654 
7655 		if (rv) {
7656 			DMSG(vdc, 0, "[%d] Failed to get EFI (err=%d)",
7657 			    vdc->instance, rv);
7658 			mutex_enter(&vdc->lock);
7659 			vdc_store_label_unk(vdc);
7660 			return (EIO);
7661 		}
7662 
7663 		mutex_enter(&vdc->lock);
7664 		vdc_store_label_efi(vdc, gpt, gpe);
7665 		vd_efi_free(&edev, gpt, gpe);
7666 		return (ENOTSUP);
7667 	}
7668 
7669 	if (rv != 0) {
7670 		DMSG(vdc, 0, "[%d] Failed to get VTOC (err=%d)",
7671 		    vdc->instance, rv);
7672 		mutex_enter(&vdc->lock);
7673 		vdc_store_label_unk(vdc);
7674 		if (rv != EINVAL)
7675 			rv = EIO;
7676 		return (rv);
7677 	}
7678 
7679 	/* check that geometry and vtoc are valid */
7680 	if (geom.dkg_nhead == 0 || geom.dkg_nsect == 0 ||
7681 	    vtoc.v_sanity != VTOC_SANE) {
7682 		mutex_enter(&vdc->lock);
7683 		vdc_store_label_unk(vdc);
7684 		return (EINVAL);
7685 	}
7686 
7687 	/*
7688 	 * We have a disk and a valid VTOC. However this does not mean
7689 	 * that the disk currently have a VTOC label. The returned VTOC may
7690 	 * be a default VTOC to be used for configuring the disk (this is
7691 	 * what is done for disk image). So we read the label from the
7692 	 * beginning of the disk to ensure we really have a VTOC label.
7693 	 *
7694 	 * FUTURE: This could be the default way for reading the VTOC
7695 	 * from the disk as opposed to sending the VD_OP_GET_VTOC
7696 	 * to the server. This will be the default if vdc is implemented
7697 	 * ontop of cmlb.
7698 	 */
7699 
7700 	/*
7701 	 * Single slice disk does not support read using an absolute disk
7702 	 * offset so we just rely on the DKIOCGVTOC ioctl in that case.
7703 	 */
7704 	if (vdc->vdisk_type == VD_DISK_TYPE_SLICE) {
7705 		mutex_enter(&vdc->lock);
7706 		if (vtoc.v_nparts != 1) {
7707 			vdc_store_label_unk(vdc);
7708 			return (EINVAL);
7709 		}
7710 		vdc_store_label_vtoc(vdc, &geom, &vtoc);
7711 		return (0);
7712 	}
7713 
7714 	if (vtoc.v_nparts != V_NUMPAR) {
7715 		mutex_enter(&vdc->lock);
7716 		vdc_store_label_unk(vdc);
7717 		return (EINVAL);
7718 	}
7719 
7720 	/*
7721 	 * Read disk label from start of disk
7722 	 */
7723 	buf = kmem_alloc(sizeof (buf_t), KM_SLEEP);
7724 	bioinit(buf);
7725 	buf->b_un.b_addr = (caddr_t)&label;
7726 	buf->b_bcount = DK_LABEL_SIZE;
7727 	buf->b_flags = B_BUSY | B_READ;
7728 	buf->b_dev = cmpdev(dev);
7729 	rv = vdc_send_request(vdc, VD_OP_BREAD, (caddr_t)&label,
7730 	    DK_LABEL_SIZE, VD_SLICE_NONE, 0, CB_STRATEGY, buf, VIO_read_dir);
7731 	if (rv) {
7732 		DMSG(vdc, 1, "[%d] Failed to read disk block 0\n",
7733 		    vdc->instance);
7734 	} else {
7735 		rv = biowait(buf);
7736 		biofini(buf);
7737 	}
7738 	kmem_free(buf, sizeof (buf_t));
7739 
7740 	if (rv != 0 || label.dkl_magic != DKL_MAGIC ||
7741 	    label.dkl_cksum != vdc_lbl2cksum(&label)) {
7742 		DMSG(vdc, 1, "[%d] Got VTOC with invalid label\n",
7743 		    vdc->instance);
7744 		mutex_enter(&vdc->lock);
7745 		vdc_store_label_unk(vdc);
7746 		return (EINVAL);
7747 	}
7748 
7749 	mutex_enter(&vdc->lock);
7750 	vdc_store_label_vtoc(vdc, &geom, &vtoc);
7751 	return (0);
7752 }
7753 
7754 /*
7755  * Function:
7756  *	vdc_validate
7757  *
7758  * Description:
7759  *	This routine discovers the label of the disk and create the
7760  *	appropriate device nodes if the label has changed.
7761  *
7762  * Arguments:
7763  *	vdc	- soft state pointer for this instance of the device driver.
7764  *
7765  * Return Code:
7766  *	none.
7767  */
7768 static void
7769 vdc_validate(vdc_t *vdc)
7770 {
7771 	vd_disk_label_t old_label;
7772 	vd_slice_t old_slice[V_NUMPAR];
7773 	int rv;
7774 
7775 	ASSERT(!MUTEX_HELD(&vdc->lock));
7776 
7777 	mutex_enter(&vdc->lock);
7778 
7779 	/* save the current label and vtoc */
7780 	old_label = vdc->vdisk_label;
7781 	bcopy(vdc->slice, &old_slice, sizeof (vd_slice_t) * V_NUMPAR);
7782 
7783 	/* check the geometry */
7784 	(void) vdc_validate_geometry(vdc);
7785 
7786 	/* if the disk label has changed, update device nodes */
7787 	if (vdc->vdisk_label != old_label) {
7788 
7789 		if (vdc->vdisk_label == VD_DISK_LABEL_EFI)
7790 			rv = vdc_create_device_nodes_efi(vdc);
7791 		else
7792 			rv = vdc_create_device_nodes_vtoc(vdc);
7793 
7794 		if (rv != 0) {
7795 			DMSG(vdc, 0, "![%d] Failed to update device nodes",
7796 			    vdc->instance);
7797 		}
7798 	}
7799 
7800 	mutex_exit(&vdc->lock);
7801 }
7802 
7803 static void
7804 vdc_validate_task(void *arg)
7805 {
7806 	vdc_t *vdc = (vdc_t *)arg;
7807 
7808 	vdc_validate(vdc);
7809 
7810 	mutex_enter(&vdc->lock);
7811 	ASSERT(vdc->validate_pending > 0);
7812 	vdc->validate_pending--;
7813 	mutex_exit(&vdc->lock);
7814 }
7815 
7816 /*
7817  * Function:
7818  *	vdc_setup_devid()
7819  *
7820  * Description:
7821  *	This routine discovers the devid of a vDisk. It requests the devid of
7822  *	the underlying device from the vDisk server, builds an encapsulated
7823  *	devid based on the retrieved devid and registers that new devid to
7824  *	the vDisk.
7825  *
7826  * Arguments:
7827  *	vdc	- soft state pointer for this instance of the device driver.
7828  *
7829  * Return Code:
7830  *	0	- A devid was succesfully registered for the vDisk
7831  */
7832 static int
7833 vdc_setup_devid(vdc_t *vdc)
7834 {
7835 	int rv;
7836 	vd_devid_t *vd_devid;
7837 	size_t bufsize, bufid_len;
7838 
7839 	/*
7840 	 * At first sight, we don't know the size of the devid that the
7841 	 * server will return but this size will be encoded into the
7842 	 * reply. So we do a first request using a default size then we
7843 	 * check if this size was large enough. If not then we do a second
7844 	 * request with the correct size returned by the server. Note that
7845 	 * ldc requires size to be 8-byte aligned.
7846 	 */
7847 	bufsize = P2ROUNDUP(VD_DEVID_SIZE(VD_DEVID_DEFAULT_LEN),
7848 	    sizeof (uint64_t));
7849 	vd_devid = kmem_zalloc(bufsize, KM_SLEEP);
7850 	bufid_len = bufsize - sizeof (vd_efi_t) - 1;
7851 
7852 	rv = vdc_do_sync_op(vdc, VD_OP_GET_DEVID, (caddr_t)vd_devid,
7853 	    bufsize, 0, 0, CB_SYNC, 0, VIO_both_dir, B_TRUE);
7854 
7855 	DMSG(vdc, 2, "sync_op returned %d\n", rv);
7856 
7857 	if (rv) {
7858 		kmem_free(vd_devid, bufsize);
7859 		return (rv);
7860 	}
7861 
7862 	if (vd_devid->length > bufid_len) {
7863 		/*
7864 		 * The returned devid is larger than the buffer used. Try again
7865 		 * with a buffer with the right size.
7866 		 */
7867 		kmem_free(vd_devid, bufsize);
7868 		bufsize = P2ROUNDUP(VD_DEVID_SIZE(vd_devid->length),
7869 		    sizeof (uint64_t));
7870 		vd_devid = kmem_zalloc(bufsize, KM_SLEEP);
7871 		bufid_len = bufsize - sizeof (vd_efi_t) - 1;
7872 
7873 		rv = vdc_do_sync_op(vdc, VD_OP_GET_DEVID,
7874 		    (caddr_t)vd_devid, bufsize, 0, 0, CB_SYNC, 0,
7875 		    VIO_both_dir, B_TRUE);
7876 
7877 		if (rv) {
7878 			kmem_free(vd_devid, bufsize);
7879 			return (rv);
7880 		}
7881 	}
7882 
7883 	/*
7884 	 * The virtual disk should have the same device id as the one associated
7885 	 * with the physical disk it is mapped on, otherwise sharing a disk
7886 	 * between a LDom and a non-LDom may not work (for example for a shared
7887 	 * SVM disk set).
7888 	 *
7889 	 * The DDI framework does not allow creating a device id with any
7890 	 * type so we first create a device id of type DEVID_ENCAP and then
7891 	 * we restore the orignal type of the physical device.
7892 	 */
7893 
7894 	DMSG(vdc, 2, ": devid length = %d\n", vd_devid->length);
7895 
7896 	/* build an encapsulated devid based on the returned devid */
7897 	if (ddi_devid_init(vdc->dip, DEVID_ENCAP, vd_devid->length,
7898 	    vd_devid->id, &vdc->devid) != DDI_SUCCESS) {
7899 		DMSG(vdc, 1, "[%d] Fail to created devid\n", vdc->instance);
7900 		kmem_free(vd_devid, bufsize);
7901 		return (1);
7902 	}
7903 
7904 	DEVID_FORMTYPE((impl_devid_t *)vdc->devid, vd_devid->type);
7905 
7906 	ASSERT(ddi_devid_valid(vdc->devid) == DDI_SUCCESS);
7907 
7908 	kmem_free(vd_devid, bufsize);
7909 
7910 	if (ddi_devid_register(vdc->dip, vdc->devid) != DDI_SUCCESS) {
7911 		DMSG(vdc, 1, "[%d] Fail to register devid\n", vdc->instance);
7912 		return (1);
7913 	}
7914 
7915 	return (0);
7916 }
7917 
7918 static void
7919 vdc_store_label_efi(vdc_t *vdc, efi_gpt_t *gpt, efi_gpe_t *gpe)
7920 {
7921 	int i, nparts;
7922 
7923 	ASSERT(MUTEX_HELD(&vdc->lock));
7924 
7925 	vdc->vdisk_label = VD_DISK_LABEL_EFI;
7926 	bzero(vdc->vtoc, sizeof (struct vtoc));
7927 	bzero(vdc->geom, sizeof (struct dk_geom));
7928 	bzero(vdc->slice, sizeof (vd_slice_t) * V_NUMPAR);
7929 
7930 	nparts = gpt->efi_gpt_NumberOfPartitionEntries;
7931 
7932 	for (i = 0; i < nparts && i < VD_EFI_WD_SLICE; i++) {
7933 
7934 		if (gpe[i].efi_gpe_StartingLBA == 0 ||
7935 		    gpe[i].efi_gpe_EndingLBA == 0) {
7936 			continue;
7937 		}
7938 
7939 		vdc->slice[i].start = gpe[i].efi_gpe_StartingLBA;
7940 		vdc->slice[i].nblocks = gpe[i].efi_gpe_EndingLBA -
7941 		    gpe[i].efi_gpe_StartingLBA + 1;
7942 	}
7943 
7944 	ASSERT(vdc->vdisk_size != 0);
7945 	vdc->slice[VD_EFI_WD_SLICE].start = 0;
7946 	vdc->slice[VD_EFI_WD_SLICE].nblocks = vdc->vdisk_size;
7947 
7948 }
7949 
7950 static void
7951 vdc_store_label_vtoc(vdc_t *vdc, struct dk_geom *geom, struct vtoc *vtoc)
7952 {
7953 	int i;
7954 
7955 	ASSERT(MUTEX_HELD(&vdc->lock));
7956 	ASSERT(vdc->block_size == vtoc->v_sectorsz);
7957 
7958 	vdc->vdisk_label = VD_DISK_LABEL_VTOC;
7959 	bcopy(vtoc, vdc->vtoc, sizeof (struct vtoc));
7960 	bcopy(geom, vdc->geom, sizeof (struct dk_geom));
7961 	bzero(vdc->slice, sizeof (vd_slice_t) * V_NUMPAR);
7962 
7963 	for (i = 0; i < vtoc->v_nparts; i++) {
7964 		vdc->slice[i].start = vtoc->v_part[i].p_start;
7965 		vdc->slice[i].nblocks = vtoc->v_part[i].p_size;
7966 	}
7967 }
7968 
7969 static void
7970 vdc_store_label_unk(vdc_t *vdc)
7971 {
7972 	ASSERT(MUTEX_HELD(&vdc->lock));
7973 
7974 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
7975 	bzero(vdc->vtoc, sizeof (struct vtoc));
7976 	bzero(vdc->geom, sizeof (struct dk_geom));
7977 	bzero(vdc->slice, sizeof (vd_slice_t) * V_NUMPAR);
7978 }
7979