xref: /titanic_51/usr/src/uts/sun4v/io/vdc.c (revision d62bc4badc1c1f1549c961cfb8b420e650e1272b)
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/mach_descrip.h>
69 #include <sys/modctl.h>
70 #include <sys/mdeg.h>
71 #include <sys/note.h>
72 #include <sys/open.h>
73 #include <sys/sdt.h>
74 #include <sys/stat.h>
75 #include <sys/sunddi.h>
76 #include <sys/types.h>
77 #include <sys/promif.h>
78 #include <sys/var.h>
79 #include <sys/vtoc.h>
80 #include <sys/archsystm.h>
81 #include <sys/sysmacros.h>
82 
83 #include <sys/cdio.h>
84 #include <sys/dktp/fdisk.h>
85 #include <sys/dktp/dadkio.h>
86 #include <sys/mhd.h>
87 #include <sys/scsi/generic/sense.h>
88 #include <sys/scsi/impl/uscsi.h>
89 #include <sys/scsi/impl/services.h>
90 #include <sys/scsi/targets/sddef.h>
91 
92 #include <sys/ldoms.h>
93 #include <sys/ldc.h>
94 #include <sys/vio_common.h>
95 #include <sys/vio_mailbox.h>
96 #include <sys/vio_util.h>
97 #include <sys/vdsk_common.h>
98 #include <sys/vdsk_mailbox.h>
99 #include <sys/vdc.h>
100 
101 /*
102  * function prototypes
103  */
104 
105 /* standard driver functions */
106 static int	vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred);
107 static int	vdc_close(dev_t dev, int flag, int otyp, cred_t *cred);
108 static int	vdc_strategy(struct buf *buf);
109 static int	vdc_print(dev_t dev, char *str);
110 static int	vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk);
111 static int	vdc_read(dev_t dev, struct uio *uio, cred_t *cred);
112 static int	vdc_write(dev_t dev, struct uio *uio, cred_t *cred);
113 static int	vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode,
114 			cred_t *credp, int *rvalp);
115 static int	vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred);
116 static int	vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred);
117 
118 static int	vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,
119 			void *arg, void **resultp);
120 static int	vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd);
121 static int	vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd);
122 
123 /* setup */
124 static void	vdc_min(struct buf *bufp);
125 static int	vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen);
126 static int	vdc_do_ldc_init(vdc_t *vdc, md_t *mdp, mde_cookie_t vd_node);
127 static int	vdc_start_ldc_connection(vdc_t *vdc);
128 static int	vdc_create_device_nodes(vdc_t *vdc);
129 static int	vdc_create_device_nodes_efi(vdc_t *vdc);
130 static int	vdc_create_device_nodes_vtoc(vdc_t *vdc);
131 static int	vdc_create_device_nodes_props(vdc_t *vdc);
132 static int	vdc_get_md_node(dev_info_t *dip, md_t **mdpp,
133 		    mde_cookie_t *vd_nodep, mde_cookie_t *vd_portp);
134 static int	vdc_get_ldc_id(md_t *, mde_cookie_t, uint64_t *);
135 static int	vdc_do_ldc_up(vdc_t *vdc);
136 static void	vdc_terminate_ldc(vdc_t *vdc);
137 static int	vdc_init_descriptor_ring(vdc_t *vdc);
138 static void	vdc_destroy_descriptor_ring(vdc_t *vdc);
139 static int	vdc_setup_devid(vdc_t *vdc);
140 static void	vdc_store_label_efi(vdc_t *, efi_gpt_t *, efi_gpe_t *);
141 static void	vdc_store_label_vtoc(vdc_t *, struct dk_geom *, struct vtoc *);
142 static void	vdc_store_label_unk(vdc_t *vdc);
143 static boolean_t vdc_is_opened(vdc_t *vdc);
144 
145 /* handshake with vds */
146 static int		vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver);
147 static int		vdc_ver_negotiation(vdc_t *vdcp);
148 static int		vdc_init_attr_negotiation(vdc_t *vdc);
149 static int		vdc_attr_negotiation(vdc_t *vdcp);
150 static int		vdc_init_dring_negotiate(vdc_t *vdc);
151 static int		vdc_dring_negotiation(vdc_t *vdcp);
152 static int		vdc_send_rdx(vdc_t *vdcp);
153 static int		vdc_rdx_exchange(vdc_t *vdcp);
154 static boolean_t	vdc_is_supported_version(vio_ver_msg_t *ver_msg);
155 
156 /* processing incoming messages from vDisk server */
157 static void	vdc_process_msg_thread(vdc_t *vdc);
158 static int	vdc_recv(vdc_t *vdc, vio_msg_t *msgp, size_t *nbytesp);
159 
160 static uint_t	vdc_handle_cb(uint64_t event, caddr_t arg);
161 static int	vdc_process_data_msg(vdc_t *vdc, vio_msg_t *msg);
162 static int	vdc_handle_ver_msg(vdc_t *vdc, vio_ver_msg_t *ver_msg);
163 static int	vdc_handle_attr_msg(vdc_t *vdc, vd_attr_msg_t *attr_msg);
164 static int	vdc_handle_dring_reg_msg(vdc_t *vdc, vio_dring_reg_msg_t *msg);
165 static int 	vdc_send_request(vdc_t *vdcp, int operation,
166 		    caddr_t addr, size_t nbytes, int slice, diskaddr_t offset,
167 		    int cb_type, void *cb_arg, vio_desc_direction_t dir);
168 static int	vdc_map_to_shared_dring(vdc_t *vdcp, int idx);
169 static int 	vdc_populate_descriptor(vdc_t *vdcp, int operation,
170 		    caddr_t addr, size_t nbytes, int slice, diskaddr_t offset,
171 		    int cb_type, void *cb_arg, vio_desc_direction_t dir);
172 static int 	vdc_do_sync_op(vdc_t *vdcp, int operation, caddr_t addr,
173 		    size_t nbytes, int slice, diskaddr_t offset, int cb_type,
174 		    void *cb_arg, vio_desc_direction_t dir, boolean_t);
175 
176 static int	vdc_wait_for_response(vdc_t *vdcp, vio_msg_t *msgp);
177 static int	vdc_drain_response(vdc_t *vdcp);
178 static int	vdc_depopulate_descriptor(vdc_t *vdc, uint_t idx);
179 static int	vdc_populate_mem_hdl(vdc_t *vdcp, vdc_local_desc_t *ldep);
180 static int	vdc_verify_seq_num(vdc_t *vdc, vio_dring_msg_t *dring_msg);
181 
182 /* dkio */
183 static int	vd_process_ioctl(dev_t dev, int cmd, caddr_t arg, int mode,
184 		    int *rvalp);
185 static int	vd_process_efi_ioctl(void *vdisk, int cmd, uintptr_t arg);
186 static void	vdc_create_fake_geometry(vdc_t *vdc);
187 static int	vdc_validate_geometry(vdc_t *vdc);
188 static void	vdc_validate(vdc_t *vdc);
189 static void	vdc_validate_task(void *arg);
190 static int	vdc_null_copy_func(vdc_t *vdc, void *from, void *to,
191 		    int mode, int dir);
192 static int	vdc_get_wce_convert(vdc_t *vdc, void *from, void *to,
193 		    int mode, int dir);
194 static int	vdc_set_wce_convert(vdc_t *vdc, void *from, void *to,
195 		    int mode, int dir);
196 static int	vdc_get_vtoc_convert(vdc_t *vdc, void *from, void *to,
197 		    int mode, int dir);
198 static int	vdc_set_vtoc_convert(vdc_t *vdc, void *from, void *to,
199 		    int mode, int dir);
200 static int	vdc_get_geom_convert(vdc_t *vdc, void *from, void *to,
201 		    int mode, int dir);
202 static int	vdc_set_geom_convert(vdc_t *vdc, void *from, void *to,
203 		    int mode, int dir);
204 static int	vdc_get_efi_convert(vdc_t *vdc, void *from, void *to,
205 		    int mode, int dir);
206 static int	vdc_set_efi_convert(vdc_t *vdc, void *from, void *to,
207 		    int mode, int dir);
208 
209 static void 	vdc_ownership_update(vdc_t *vdc, int ownership_flags);
210 static int	vdc_access_set(vdc_t *vdc, uint64_t flags, int mode);
211 static vdc_io_t	*vdc_failfast_io_queue(vdc_t *vdc, struct buf *buf);
212 static int	vdc_failfast_check_resv(vdc_t *vdc);
213 
214 /*
215  * Module variables
216  */
217 
218 /*
219  * Tunable variables to control how long vdc waits before timing out on
220  * various operations
221  */
222 static int	vdc_hshake_retries = 3;
223 
224 static int	vdc_timeout = 0; /* units: seconds */
225 
226 static uint64_t vdc_hz_min_ldc_delay;
227 static uint64_t vdc_min_timeout_ldc = 1 * MILLISEC;
228 static uint64_t vdc_hz_max_ldc_delay;
229 static uint64_t vdc_max_timeout_ldc = 100 * MILLISEC;
230 
231 static uint64_t vdc_ldc_read_init_delay = 1 * MILLISEC;
232 static uint64_t vdc_ldc_read_max_delay = 100 * MILLISEC;
233 
234 /* values for dumping - need to run in a tighter loop */
235 static uint64_t	vdc_usec_timeout_dump = 100 * MILLISEC;	/* 0.1s units: ns */
236 static int	vdc_dump_retries = 100;
237 
238 static uint16_t	vdc_scsi_timeout = 60;	/* 60s units: seconds  */
239 
240 static uint64_t vdc_ownership_delay = 6 * MICROSEC; /* 6s units: usec */
241 
242 /* Count of the number of vdc instances attached */
243 static volatile uint32_t	vdc_instance_count = 0;
244 
245 /* Tunable to log all SCSI errors */
246 static boolean_t vdc_scsi_log_error = B_FALSE;
247 
248 /* Soft state pointer */
249 static void	*vdc_state;
250 
251 /*
252  * Controlling the verbosity of the error/debug messages
253  *
254  * vdc_msglevel - controls level of messages
255  * vdc_matchinst - 64-bit variable where each bit corresponds
256  *                 to the vdc instance the vdc_msglevel applies.
257  */
258 int		vdc_msglevel = 0x0;
259 uint64_t	vdc_matchinst = 0ull;
260 
261 /*
262  * Supported vDisk protocol version pairs.
263  *
264  * The first array entry is the latest and preferred version.
265  */
266 static const vio_ver_t	vdc_version[] = {{1, 1}};
267 
268 static struct cb_ops vdc_cb_ops = {
269 	vdc_open,	/* cb_open */
270 	vdc_close,	/* cb_close */
271 	vdc_strategy,	/* cb_strategy */
272 	vdc_print,	/* cb_print */
273 	vdc_dump,	/* cb_dump */
274 	vdc_read,	/* cb_read */
275 	vdc_write,	/* cb_write */
276 	vdc_ioctl,	/* cb_ioctl */
277 	nodev,		/* cb_devmap */
278 	nodev,		/* cb_mmap */
279 	nodev,		/* cb_segmap */
280 	nochpoll,	/* cb_chpoll */
281 	ddi_prop_op,	/* cb_prop_op */
282 	NULL,		/* cb_str */
283 	D_MP | D_64BIT,	/* cb_flag */
284 	CB_REV,		/* cb_rev */
285 	vdc_aread,	/* cb_aread */
286 	vdc_awrite	/* cb_awrite */
287 };
288 
289 static struct dev_ops vdc_ops = {
290 	DEVO_REV,	/* devo_rev */
291 	0,		/* devo_refcnt */
292 	vdc_getinfo,	/* devo_getinfo */
293 	nulldev,	/* devo_identify */
294 	nulldev,	/* devo_probe */
295 	vdc_attach,	/* devo_attach */
296 	vdc_detach,	/* devo_detach */
297 	nodev,		/* devo_reset */
298 	&vdc_cb_ops,	/* devo_cb_ops */
299 	NULL,		/* devo_bus_ops */
300 	nulldev		/* devo_power */
301 };
302 
303 static struct modldrv modldrv = {
304 	&mod_driverops,
305 	"virtual disk client",
306 	&vdc_ops,
307 };
308 
309 static struct modlinkage modlinkage = {
310 	MODREV_1,
311 	&modldrv,
312 	NULL
313 };
314 
315 /* -------------------------------------------------------------------------- */
316 
317 /*
318  * Device Driver housekeeping and setup
319  */
320 
321 int
322 _init(void)
323 {
324 	int	status;
325 
326 	if ((status = ddi_soft_state_init(&vdc_state, sizeof (vdc_t), 1)) != 0)
327 		return (status);
328 	if ((status = mod_install(&modlinkage)) != 0)
329 		ddi_soft_state_fini(&vdc_state);
330 	return (status);
331 }
332 
333 int
334 _info(struct modinfo *modinfop)
335 {
336 	return (mod_info(&modlinkage, modinfop));
337 }
338 
339 int
340 _fini(void)
341 {
342 	int	status;
343 
344 	if ((status = mod_remove(&modlinkage)) != 0)
345 		return (status);
346 	ddi_soft_state_fini(&vdc_state);
347 	return (0);
348 }
349 
350 static int
351 vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,  void *arg, void **resultp)
352 {
353 	_NOTE(ARGUNUSED(dip))
354 
355 	int	instance = VDCUNIT((dev_t)arg);
356 	vdc_t	*vdc = NULL;
357 
358 	switch (cmd) {
359 	case DDI_INFO_DEVT2DEVINFO:
360 		if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
361 			*resultp = NULL;
362 			return (DDI_FAILURE);
363 		}
364 		*resultp = vdc->dip;
365 		return (DDI_SUCCESS);
366 	case DDI_INFO_DEVT2INSTANCE:
367 		*resultp = (void *)(uintptr_t)instance;
368 		return (DDI_SUCCESS);
369 	default:
370 		*resultp = NULL;
371 		return (DDI_FAILURE);
372 	}
373 }
374 
375 static int
376 vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
377 {
378 	kt_did_t failfast_tid, ownership_tid;
379 	int	instance;
380 	int	rv;
381 	vdc_t	*vdc = NULL;
382 
383 	switch (cmd) {
384 	case DDI_DETACH:
385 		/* the real work happens below */
386 		break;
387 	case DDI_SUSPEND:
388 		/* nothing to do for this non-device */
389 		return (DDI_SUCCESS);
390 	default:
391 		return (DDI_FAILURE);
392 	}
393 
394 	ASSERT(cmd == DDI_DETACH);
395 	instance = ddi_get_instance(dip);
396 	DMSGX(1, "[%d] Entered\n", instance);
397 
398 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
399 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
400 		return (DDI_FAILURE);
401 	}
402 
403 	/*
404 	 * This function is called when vdc is detached or if it has failed to
405 	 * attach. In that case, the attach may have fail before the vdisk type
406 	 * has been set so we can't call vdc_is_opened(). However as the attach
407 	 * has failed, we know that the vdisk is not opened and we can safely
408 	 * detach.
409 	 */
410 	if (vdc->vdisk_type != VD_DISK_TYPE_UNK && vdc_is_opened(vdc)) {
411 		DMSG(vdc, 0, "[%d] Cannot detach: device is open", instance);
412 		return (DDI_FAILURE);
413 	}
414 
415 	if (vdc->dkio_flush_pending) {
416 		DMSG(vdc, 0,
417 		    "[%d] Cannot detach: %d outstanding DKIO flushes\n",
418 		    instance, vdc->dkio_flush_pending);
419 		return (DDI_FAILURE);
420 	}
421 
422 	if (vdc->validate_pending) {
423 		DMSG(vdc, 0,
424 		    "[%d] Cannot detach: %d outstanding validate request\n",
425 		    instance, vdc->validate_pending);
426 		return (DDI_FAILURE);
427 	}
428 
429 	DMSG(vdc, 0, "[%d] proceeding...\n", instance);
430 
431 	/* If we took ownership, release ownership */
432 	mutex_enter(&vdc->ownership_lock);
433 	if (vdc->ownership & VDC_OWNERSHIP_GRANTED) {
434 		rv = vdc_access_set(vdc, VD_ACCESS_SET_CLEAR, FKIOCTL);
435 		if (rv == 0) {
436 			vdc_ownership_update(vdc, VDC_OWNERSHIP_NONE);
437 		}
438 	}
439 	mutex_exit(&vdc->ownership_lock);
440 
441 	/* mark instance as detaching */
442 	vdc->lifecycle	= VDC_LC_DETACHING;
443 
444 	/*
445 	 * try and disable callbacks to prevent another handshake
446 	 */
447 	rv = ldc_set_cb_mode(vdc->ldc_handle, LDC_CB_DISABLE);
448 	DMSG(vdc, 0, "callback disabled (rv=%d)\n", rv);
449 
450 	if (vdc->initialized & VDC_THREAD) {
451 		mutex_enter(&vdc->read_lock);
452 		if ((vdc->read_state == VDC_READ_WAITING) ||
453 		    (vdc->read_state == VDC_READ_RESET)) {
454 			vdc->read_state = VDC_READ_RESET;
455 			cv_signal(&vdc->read_cv);
456 		}
457 
458 		mutex_exit(&vdc->read_lock);
459 
460 		/* wake up any thread waiting for connection to come online */
461 		mutex_enter(&vdc->lock);
462 		if (vdc->state == VDC_STATE_INIT_WAITING) {
463 			DMSG(vdc, 0,
464 			    "[%d] write reset - move to resetting state...\n",
465 			    instance);
466 			vdc->state = VDC_STATE_RESETTING;
467 			cv_signal(&vdc->initwait_cv);
468 		}
469 		mutex_exit(&vdc->lock);
470 
471 		/* now wait until state transitions to VDC_STATE_DETACH */
472 		thread_join(vdc->msg_proc_thr->t_did);
473 		ASSERT(vdc->state == VDC_STATE_DETACH);
474 		DMSG(vdc, 0, "[%d] Reset thread exit and join ..\n",
475 		    vdc->instance);
476 	}
477 
478 	mutex_enter(&vdc->lock);
479 
480 	if (vdc->initialized & VDC_DRING)
481 		vdc_destroy_descriptor_ring(vdc);
482 
483 	if (vdc->initialized & VDC_LDC)
484 		vdc_terminate_ldc(vdc);
485 
486 	if (vdc->failfast_thread) {
487 		failfast_tid = vdc->failfast_thread->t_did;
488 		vdc->failfast_interval = 0;
489 		cv_signal(&vdc->failfast_cv);
490 	} else {
491 		failfast_tid = 0;
492 	}
493 
494 	if (vdc->ownership & VDC_OWNERSHIP_WANTED) {
495 		ownership_tid = vdc->ownership_thread->t_did;
496 		vdc->ownership = VDC_OWNERSHIP_NONE;
497 		cv_signal(&vdc->ownership_cv);
498 	} else {
499 		ownership_tid = 0;
500 	}
501 
502 	mutex_exit(&vdc->lock);
503 
504 	if (failfast_tid != 0)
505 		thread_join(failfast_tid);
506 
507 	if (ownership_tid != 0)
508 		thread_join(ownership_tid);
509 
510 	if (vdc->initialized & VDC_MINOR) {
511 		ddi_prop_remove_all(dip);
512 		ddi_remove_minor_node(dip, NULL);
513 	}
514 
515 	if (vdc->initialized & VDC_LOCKS) {
516 		mutex_destroy(&vdc->lock);
517 		mutex_destroy(&vdc->read_lock);
518 		mutex_destroy(&vdc->ownership_lock);
519 		cv_destroy(&vdc->initwait_cv);
520 		cv_destroy(&vdc->dring_free_cv);
521 		cv_destroy(&vdc->membind_cv);
522 		cv_destroy(&vdc->sync_pending_cv);
523 		cv_destroy(&vdc->sync_blocked_cv);
524 		cv_destroy(&vdc->read_cv);
525 		cv_destroy(&vdc->running_cv);
526 		cv_destroy(&vdc->ownership_cv);
527 		cv_destroy(&vdc->failfast_cv);
528 		cv_destroy(&vdc->failfast_io_cv);
529 	}
530 
531 	if (vdc->minfo)
532 		kmem_free(vdc->minfo, sizeof (struct dk_minfo));
533 
534 	if (vdc->cinfo)
535 		kmem_free(vdc->cinfo, sizeof (struct dk_cinfo));
536 
537 	if (vdc->vtoc)
538 		kmem_free(vdc->vtoc, sizeof (struct vtoc));
539 
540 	if (vdc->geom)
541 		kmem_free(vdc->geom, sizeof (struct dk_geom));
542 
543 	if (vdc->devid) {
544 		ddi_devid_unregister(dip);
545 		ddi_devid_free(vdc->devid);
546 	}
547 
548 	if (vdc->initialized & VDC_SOFT_STATE)
549 		ddi_soft_state_free(vdc_state, instance);
550 
551 	DMSG(vdc, 0, "[%d] End %p\n", instance, (void *)vdc);
552 
553 	return (DDI_SUCCESS);
554 }
555 
556 
557 static int
558 vdc_do_attach(dev_info_t *dip)
559 {
560 	int		instance;
561 	vdc_t		*vdc = NULL;
562 	int		status;
563 	md_t		*mdp;
564 	mde_cookie_t	vd_node, vd_port;
565 
566 	ASSERT(dip != NULL);
567 
568 	instance = ddi_get_instance(dip);
569 	if (ddi_soft_state_zalloc(vdc_state, instance) != DDI_SUCCESS) {
570 		cmn_err(CE_NOTE, "[%d] Couldn't alloc state structure",
571 		    instance);
572 		return (DDI_FAILURE);
573 	}
574 
575 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
576 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
577 		return (DDI_FAILURE);
578 	}
579 
580 	/*
581 	 * We assign the value to initialized in this case to zero out the
582 	 * variable and then set bits in it to indicate what has been done
583 	 */
584 	vdc->initialized = VDC_SOFT_STATE;
585 
586 	vdc_hz_min_ldc_delay = drv_usectohz(vdc_min_timeout_ldc);
587 	vdc_hz_max_ldc_delay = drv_usectohz(vdc_max_timeout_ldc);
588 
589 	vdc->dip	= dip;
590 	vdc->instance	= instance;
591 	vdc->vdisk_type	= VD_DISK_TYPE_UNK;
592 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
593 	vdc->state	= VDC_STATE_INIT;
594 	vdc->lifecycle	= VDC_LC_ATTACHING;
595 	vdc->ldc_state	= 0;
596 	vdc->session_id = 0;
597 	vdc->block_size = DEV_BSIZE;
598 	vdc->max_xfer_sz = maxphys / DEV_BSIZE;
599 
600 	/*
601 	 * We assume, for now, that the vDisk server will export 'read'
602 	 * operations to us at a minimum (this is needed because of checks
603 	 * in vdc for supported operations early in the handshake process).
604 	 * The vDisk server will return ENOTSUP if this is not the case.
605 	 * The value will be overwritten during the attribute exchange with
606 	 * the bitmask of operations exported by server.
607 	 */
608 	vdc->operations = VD_OP_MASK_READ;
609 
610 	vdc->vtoc = NULL;
611 	vdc->geom = NULL;
612 	vdc->cinfo = NULL;
613 	vdc->minfo = NULL;
614 
615 	mutex_init(&vdc->lock, NULL, MUTEX_DRIVER, NULL);
616 	cv_init(&vdc->initwait_cv, NULL, CV_DRIVER, NULL);
617 	cv_init(&vdc->dring_free_cv, NULL, CV_DRIVER, NULL);
618 	cv_init(&vdc->membind_cv, NULL, CV_DRIVER, NULL);
619 	cv_init(&vdc->running_cv, NULL, CV_DRIVER, NULL);
620 
621 	vdc->threads_pending = 0;
622 	vdc->sync_op_pending = B_FALSE;
623 	vdc->sync_op_blocked = B_FALSE;
624 	cv_init(&vdc->sync_pending_cv, NULL, CV_DRIVER, NULL);
625 	cv_init(&vdc->sync_blocked_cv, NULL, CV_DRIVER, NULL);
626 
627 	mutex_init(&vdc->ownership_lock, NULL, MUTEX_DRIVER, NULL);
628 	cv_init(&vdc->ownership_cv, NULL, CV_DRIVER, NULL);
629 	cv_init(&vdc->failfast_cv, NULL, CV_DRIVER, NULL);
630 	cv_init(&vdc->failfast_io_cv, NULL, CV_DRIVER, NULL);
631 
632 	/* init blocking msg read functionality */
633 	mutex_init(&vdc->read_lock, NULL, MUTEX_DRIVER, NULL);
634 	cv_init(&vdc->read_cv, NULL, CV_DRIVER, NULL);
635 	vdc->read_state = VDC_READ_IDLE;
636 
637 	vdc->initialized |= VDC_LOCKS;
638 
639 	/* get device and port MD node for this disk instance */
640 	if (vdc_get_md_node(dip, &mdp, &vd_node, &vd_port) != 0) {
641 		cmn_err(CE_NOTE, "[%d] Could not get machine description node",
642 		    instance);
643 		return (DDI_FAILURE);
644 	}
645 
646 	/* set the connection timeout */
647 	if (vd_port == NULL || (md_get_prop_val(mdp, vd_port,
648 	    VDC_MD_TIMEOUT, &vdc->ctimeout) != 0)) {
649 		vdc->ctimeout = 0;
650 	}
651 
652 	/* initialise LDC channel which will be used to communicate with vds */
653 	status = vdc_do_ldc_init(vdc, mdp, vd_node);
654 
655 	(void) md_fini_handle(mdp);
656 
657 	if (status != 0) {
658 		cmn_err(CE_NOTE, "[%d] Couldn't initialize LDC", instance);
659 		goto return_status;
660 	}
661 
662 	/* initialize the thread responsible for managing state with server */
663 	vdc->msg_proc_thr = thread_create(NULL, 0, vdc_process_msg_thread,
664 	    vdc, 0, &p0, TS_RUN, minclsyspri);
665 	if (vdc->msg_proc_thr == NULL) {
666 		cmn_err(CE_NOTE, "[%d] Failed to create msg processing thread",
667 		    instance);
668 		return (DDI_FAILURE);
669 	}
670 
671 	vdc->initialized |= VDC_THREAD;
672 
673 	atomic_inc_32(&vdc_instance_count);
674 
675 	/*
676 	 * Check the disk label. This will send requests and do the handshake.
677 	 * We don't really care about the disk label now. What we really need is
678 	 * the handshake do be done so that we know the type of the disk (slice
679 	 * or full disk) and the appropriate device nodes can be created.
680 	 */
681 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
682 	vdc->vtoc = kmem_zalloc(sizeof (struct vtoc), KM_SLEEP);
683 	vdc->geom = kmem_zalloc(sizeof (struct dk_geom), KM_SLEEP);
684 	vdc->minfo = kmem_zalloc(sizeof (struct dk_minfo), KM_SLEEP);
685 
686 	mutex_enter(&vdc->lock);
687 	(void) vdc_validate_geometry(vdc);
688 	mutex_exit(&vdc->lock);
689 
690 	/*
691 	 * Now that we have the device info we can create the
692 	 * device nodes and properties
693 	 */
694 	status = vdc_create_device_nodes(vdc);
695 	if (status) {
696 		DMSG(vdc, 0, "[%d] Failed to create device nodes",
697 		    instance);
698 		goto return_status;
699 	}
700 	status = vdc_create_device_nodes_props(vdc);
701 	if (status) {
702 		DMSG(vdc, 0, "[%d] Failed to create device nodes"
703 		    " properties (%d)", instance, status);
704 		goto return_status;
705 	}
706 
707 	/*
708 	 * Setup devid
709 	 */
710 	if (vdc_setup_devid(vdc)) {
711 		DMSG(vdc, 0, "[%d] No device id available\n", instance);
712 	}
713 
714 	ddi_report_dev(dip);
715 	vdc->lifecycle	= VDC_LC_ONLINE;
716 	DMSG(vdc, 0, "[%d] Attach tasks successful\n", instance);
717 
718 return_status:
719 	DMSG(vdc, 0, "[%d] Attach completed\n", instance);
720 	return (status);
721 }
722 
723 static int
724 vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
725 {
726 	int	status;
727 
728 	switch (cmd) {
729 	case DDI_ATTACH:
730 		if ((status = vdc_do_attach(dip)) != 0)
731 			(void) vdc_detach(dip, DDI_DETACH);
732 		return (status);
733 	case DDI_RESUME:
734 		/* nothing to do for this non-device */
735 		return (DDI_SUCCESS);
736 	default:
737 		return (DDI_FAILURE);
738 	}
739 }
740 
741 static int
742 vdc_do_ldc_init(vdc_t *vdc, md_t *mdp, mde_cookie_t vd_node)
743 {
744 	int			status = 0;
745 	ldc_status_t		ldc_state;
746 	ldc_attr_t		ldc_attr;
747 	uint64_t		ldc_id = 0;
748 
749 	ASSERT(vdc != NULL);
750 
751 	vdc->initialized |= VDC_LDC;
752 
753 	if ((status = vdc_get_ldc_id(mdp, vd_node, &ldc_id)) != 0) {
754 		DMSG(vdc, 0, "[%d] Failed to get LDC channel ID property",
755 		    vdc->instance);
756 		return (EIO);
757 	}
758 
759 	DMSGX(0, "[%d] LDC id is 0x%lx\n", vdc->instance, ldc_id);
760 
761 	vdc->ldc_id = ldc_id;
762 
763 	ldc_attr.devclass = LDC_DEV_BLK;
764 	ldc_attr.instance = vdc->instance;
765 	ldc_attr.mode = LDC_MODE_UNRELIABLE;	/* unreliable transport */
766 	ldc_attr.mtu = VD_LDC_MTU;
767 
768 	if ((vdc->initialized & VDC_LDC_INIT) == 0) {
769 		status = ldc_init(ldc_id, &ldc_attr, &vdc->ldc_handle);
770 		if (status != 0) {
771 			DMSG(vdc, 0, "[%d] ldc_init(chan %ld) returned %d",
772 			    vdc->instance, ldc_id, status);
773 			return (status);
774 		}
775 		vdc->initialized |= VDC_LDC_INIT;
776 	}
777 	status = ldc_status(vdc->ldc_handle, &ldc_state);
778 	if (status != 0) {
779 		DMSG(vdc, 0, "[%d] Cannot discover LDC status [err=%d]",
780 		    vdc->instance, status);
781 		return (status);
782 	}
783 	vdc->ldc_state = ldc_state;
784 
785 	if ((vdc->initialized & VDC_LDC_CB) == 0) {
786 		status = ldc_reg_callback(vdc->ldc_handle, vdc_handle_cb,
787 		    (caddr_t)vdc);
788 		if (status != 0) {
789 			DMSG(vdc, 0, "[%d] LDC callback reg. failed (%d)",
790 			    vdc->instance, status);
791 			return (status);
792 		}
793 		vdc->initialized |= VDC_LDC_CB;
794 	}
795 
796 	vdc->initialized |= VDC_LDC;
797 
798 	/*
799 	 * At this stage we have initialised LDC, we will now try and open
800 	 * the connection.
801 	 */
802 	if (vdc->ldc_state == LDC_INIT) {
803 		status = ldc_open(vdc->ldc_handle);
804 		if (status != 0) {
805 			DMSG(vdc, 0, "[%d] ldc_open(chan %ld) returned %d",
806 			    vdc->instance, vdc->ldc_id, status);
807 			return (status);
808 		}
809 		vdc->initialized |= VDC_LDC_OPEN;
810 	}
811 
812 	return (status);
813 }
814 
815 static int
816 vdc_start_ldc_connection(vdc_t *vdc)
817 {
818 	int		status = 0;
819 
820 	ASSERT(vdc != NULL);
821 
822 	ASSERT(MUTEX_HELD(&vdc->lock));
823 
824 	status = vdc_do_ldc_up(vdc);
825 
826 	DMSG(vdc, 0, "[%d] Finished bringing up LDC\n", vdc->instance);
827 
828 	return (status);
829 }
830 
831 static int
832 vdc_stop_ldc_connection(vdc_t *vdcp)
833 {
834 	int	status;
835 
836 	DMSG(vdcp, 0, ": Resetting connection to vDisk server : state %d\n",
837 	    vdcp->state);
838 
839 	status = ldc_down(vdcp->ldc_handle);
840 	DMSG(vdcp, 0, "ldc_down() = %d\n", status);
841 
842 	vdcp->initialized &= ~VDC_HANDSHAKE;
843 	DMSG(vdcp, 0, "initialized=%x\n", vdcp->initialized);
844 
845 	return (status);
846 }
847 
848 static int
849 vdc_create_device_nodes_efi(vdc_t *vdc)
850 {
851 	ddi_remove_minor_node(vdc->dip, "h");
852 	ddi_remove_minor_node(vdc->dip, "h,raw");
853 
854 	if (ddi_create_minor_node(vdc->dip, "wd", S_IFBLK,
855 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
856 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
857 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'wd'",
858 		    vdc->instance);
859 		return (EIO);
860 	}
861 
862 	/* if any device node is created we set this flag */
863 	vdc->initialized |= VDC_MINOR;
864 
865 	if (ddi_create_minor_node(vdc->dip, "wd,raw", S_IFCHR,
866 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
867 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
868 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'wd,raw'",
869 		    vdc->instance);
870 		return (EIO);
871 	}
872 
873 	return (0);
874 }
875 
876 static int
877 vdc_create_device_nodes_vtoc(vdc_t *vdc)
878 {
879 	ddi_remove_minor_node(vdc->dip, "wd");
880 	ddi_remove_minor_node(vdc->dip, "wd,raw");
881 
882 	if (ddi_create_minor_node(vdc->dip, "h", S_IFBLK,
883 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
884 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
885 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'h'",
886 		    vdc->instance);
887 		return (EIO);
888 	}
889 
890 	/* if any device node is created we set this flag */
891 	vdc->initialized |= VDC_MINOR;
892 
893 	if (ddi_create_minor_node(vdc->dip, "h,raw", S_IFCHR,
894 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
895 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
896 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'h,raw'",
897 		    vdc->instance);
898 		return (EIO);
899 	}
900 
901 	return (0);
902 }
903 
904 /*
905  * Function:
906  *	vdc_create_device_nodes
907  *
908  * Description:
909  *	This function creates the block and character device nodes under
910  *	/devices along with the node properties. It is called as part of
911  *	the attach(9E) of the instance during the handshake with vds after
912  *	vds has sent the attributes to vdc.
913  *
914  *	If the device is of type VD_DISK_TYPE_SLICE then the minor node
915  *	of 2 is used in keeping with the Solaris convention that slice 2
916  *	refers to a whole disk. Slices start at 'a'
917  *
918  * Parameters:
919  *	vdc 		- soft state pointer
920  *
921  * Return Values
922  *	0		- Success
923  *	EIO		- Failed to create node
924  *	EINVAL		- Unknown type of disk exported
925  */
926 static int
927 vdc_create_device_nodes(vdc_t *vdc)
928 {
929 	char		name[sizeof ("s,raw")];
930 	dev_info_t	*dip = NULL;
931 	int		instance, status;
932 	int		num_slices = 1;
933 	int		i;
934 
935 	ASSERT(vdc != NULL);
936 
937 	instance = vdc->instance;
938 	dip = vdc->dip;
939 
940 	switch (vdc->vdisk_type) {
941 	case VD_DISK_TYPE_DISK:
942 		num_slices = V_NUMPAR;
943 		break;
944 	case VD_DISK_TYPE_SLICE:
945 		num_slices = 1;
946 		break;
947 	case VD_DISK_TYPE_UNK:
948 	default:
949 		return (EINVAL);
950 	}
951 
952 	/*
953 	 * Minor nodes are different for EFI disks: EFI disks do not have
954 	 * a minor node 'g' for the minor number corresponding to slice
955 	 * VD_EFI_WD_SLICE (slice 7) instead they have a minor node 'wd'
956 	 * representing the whole disk.
957 	 */
958 	for (i = 0; i < num_slices; i++) {
959 
960 		if (i == VD_EFI_WD_SLICE) {
961 			if (vdc->vdisk_label == VD_DISK_LABEL_EFI)
962 				status = vdc_create_device_nodes_efi(vdc);
963 			else
964 				status = vdc_create_device_nodes_vtoc(vdc);
965 			if (status != 0)
966 				return (status);
967 			continue;
968 		}
969 
970 		(void) snprintf(name, sizeof (name), "%c", 'a' + i);
971 		if (ddi_create_minor_node(dip, name, S_IFBLK,
972 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
973 			cmn_err(CE_NOTE, "[%d] Couldn't add block node '%s'",
974 			    instance, name);
975 			return (EIO);
976 		}
977 
978 		/* if any device node is created we set this flag */
979 		vdc->initialized |= VDC_MINOR;
980 
981 		(void) snprintf(name, sizeof (name), "%c%s", 'a' + i, ",raw");
982 
983 		if (ddi_create_minor_node(dip, name, S_IFCHR,
984 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
985 			cmn_err(CE_NOTE, "[%d] Couldn't add raw node '%s'",
986 			    instance, name);
987 			return (EIO);
988 		}
989 	}
990 
991 	return (0);
992 }
993 
994 /*
995  * Function:
996  *	vdc_create_device_nodes_props
997  *
998  * Description:
999  *	This function creates the block and character device nodes under
1000  *	/devices along with the node properties. It is called as part of
1001  *	the attach(9E) of the instance during the handshake with vds after
1002  *	vds has sent the attributes to vdc.
1003  *
1004  * Parameters:
1005  *	vdc 		- soft state pointer
1006  *
1007  * Return Values
1008  *	0		- Success
1009  *	EIO		- Failed to create device node property
1010  *	EINVAL		- Unknown type of disk exported
1011  */
1012 static int
1013 vdc_create_device_nodes_props(vdc_t *vdc)
1014 {
1015 	dev_info_t	*dip = NULL;
1016 	int		instance;
1017 	int		num_slices = 1;
1018 	int64_t		size = 0;
1019 	dev_t		dev;
1020 	int		rv;
1021 	int		i;
1022 
1023 	ASSERT(vdc != NULL);
1024 
1025 	instance = vdc->instance;
1026 	dip = vdc->dip;
1027 
1028 	switch (vdc->vdisk_type) {
1029 	case VD_DISK_TYPE_DISK:
1030 		num_slices = V_NUMPAR;
1031 		break;
1032 	case VD_DISK_TYPE_SLICE:
1033 		num_slices = 1;
1034 		break;
1035 	case VD_DISK_TYPE_UNK:
1036 	default:
1037 		return (EINVAL);
1038 	}
1039 
1040 	if (vdc->vdisk_label == VD_DISK_LABEL_UNK) {
1041 		/* remove all properties */
1042 		for (i = 0; i < num_slices; i++) {
1043 			dev = makedevice(ddi_driver_major(dip),
1044 			    VD_MAKE_DEV(instance, i));
1045 			(void) ddi_prop_remove(dev, dip, VDC_SIZE_PROP_NAME);
1046 			(void) ddi_prop_remove(dev, dip, VDC_NBLOCKS_PROP_NAME);
1047 		}
1048 		return (0);
1049 	}
1050 
1051 	for (i = 0; i < num_slices; i++) {
1052 		dev = makedevice(ddi_driver_major(dip),
1053 		    VD_MAKE_DEV(instance, i));
1054 
1055 		size = vdc->slice[i].nblocks * vdc->block_size;
1056 		DMSG(vdc, 0, "[%d] sz %ld (%ld Mb)  p_size %lx\n",
1057 		    instance, size, size / (1024 * 1024),
1058 		    vdc->slice[i].nblocks);
1059 
1060 		rv = ddi_prop_update_int64(dev, dip, VDC_SIZE_PROP_NAME, size);
1061 		if (rv != DDI_PROP_SUCCESS) {
1062 			cmn_err(CE_NOTE, "[%d] Couldn't add '%s' prop of [%ld]",
1063 			    instance, VDC_SIZE_PROP_NAME, size);
1064 			return (EIO);
1065 		}
1066 
1067 		rv = ddi_prop_update_int64(dev, dip, VDC_NBLOCKS_PROP_NAME,
1068 		    lbtodb(size));
1069 		if (rv != DDI_PROP_SUCCESS) {
1070 			cmn_err(CE_NOTE, "[%d] Couldn't add '%s' prop [%llu]",
1071 			    instance, VDC_NBLOCKS_PROP_NAME, lbtodb(size));
1072 			return (EIO);
1073 		}
1074 	}
1075 
1076 	return (0);
1077 }
1078 
1079 /*
1080  * Function:
1081  *	vdc_is_opened
1082  *
1083  * Description:
1084  *	This function checks if any slice of a given virtual disk is
1085  *	currently opened.
1086  *
1087  * Parameters:
1088  *	vdc 		- soft state pointer
1089  *
1090  * Return Values
1091  *	B_TRUE		- at least one slice is opened.
1092  *	B_FALSE		- no slice is opened.
1093  */
1094 static boolean_t
1095 vdc_is_opened(vdc_t *vdc)
1096 {
1097 	int i, nslices;
1098 
1099 	switch (vdc->vdisk_type) {
1100 	case VD_DISK_TYPE_DISK:
1101 		nslices = V_NUMPAR;
1102 		break;
1103 	case VD_DISK_TYPE_SLICE:
1104 		nslices = 1;
1105 		break;
1106 	case VD_DISK_TYPE_UNK:
1107 	default:
1108 		ASSERT(0);
1109 	}
1110 
1111 	/* check if there's any layered open */
1112 	for (i = 0; i < nslices; i++) {
1113 		if (vdc->open_lyr[i] > 0)
1114 			return (B_TRUE);
1115 	}
1116 
1117 	/* check if there is any other kind of open */
1118 	for (i = 0; i < OTYPCNT; i++) {
1119 		if (vdc->open[i] != 0)
1120 			return (B_TRUE);
1121 	}
1122 
1123 	return (B_FALSE);
1124 }
1125 
1126 static int
1127 vdc_mark_opened(vdc_t *vdc, int slice, int flag, int otyp)
1128 {
1129 	uint8_t slicemask;
1130 	int i;
1131 
1132 	ASSERT(otyp < OTYPCNT);
1133 	ASSERT(slice < V_NUMPAR);
1134 	ASSERT(MUTEX_HELD(&vdc->lock));
1135 
1136 	slicemask = 1 << slice;
1137 
1138 	/* check if slice is already exclusively opened */
1139 	if (vdc->open_excl & slicemask)
1140 		return (EBUSY);
1141 
1142 	/* if open exclusive, check if slice is already opened */
1143 	if (flag & FEXCL) {
1144 		if (vdc->open_lyr[slice] > 0)
1145 			return (EBUSY);
1146 		for (i = 0; i < OTYPCNT; i++) {
1147 			if (vdc->open[i] & slicemask)
1148 				return (EBUSY);
1149 		}
1150 		vdc->open_excl |= slicemask;
1151 	}
1152 
1153 	/* mark slice as opened */
1154 	if (otyp == OTYP_LYR) {
1155 		vdc->open_lyr[slice]++;
1156 	} else {
1157 		vdc->open[otyp] |= slicemask;
1158 	}
1159 
1160 	return (0);
1161 }
1162 
1163 static void
1164 vdc_mark_closed(vdc_t *vdc, int slice, int flag, int otyp)
1165 {
1166 	uint8_t slicemask;
1167 
1168 	ASSERT(otyp < OTYPCNT);
1169 	ASSERT(slice < V_NUMPAR);
1170 	ASSERT(MUTEX_HELD(&vdc->lock));
1171 
1172 	slicemask = 1 << slice;
1173 
1174 	if (otyp == OTYP_LYR) {
1175 		ASSERT(vdc->open_lyr[slice] > 0);
1176 		vdc->open_lyr[slice]--;
1177 	} else {
1178 		vdc->open[otyp] &= ~slicemask;
1179 	}
1180 
1181 	if (flag & FEXCL)
1182 		vdc->open_excl &= ~slicemask;
1183 }
1184 
1185 static int
1186 vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred)
1187 {
1188 	_NOTE(ARGUNUSED(cred))
1189 
1190 	int	instance;
1191 	int	slice, status = 0;
1192 	vdc_t	*vdc;
1193 
1194 	ASSERT(dev != NULL);
1195 	instance = VDCUNIT(*dev);
1196 
1197 	if (otyp >= OTYPCNT)
1198 		return (EINVAL);
1199 
1200 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1201 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1202 		return (ENXIO);
1203 	}
1204 
1205 	DMSG(vdc, 0, "minor = %d flag = %x, otyp = %x\n",
1206 	    getminor(*dev), flag, otyp);
1207 
1208 	slice = VDCPART(*dev);
1209 
1210 	mutex_enter(&vdc->lock);
1211 
1212 	status = vdc_mark_opened(vdc, slice, flag, otyp);
1213 
1214 	if (status != 0) {
1215 		mutex_exit(&vdc->lock);
1216 		return (status);
1217 	}
1218 
1219 	if (flag & (FNDELAY | FNONBLOCK)) {
1220 
1221 		/* don't resubmit a validate request if there's already one */
1222 		if (vdc->validate_pending > 0) {
1223 			mutex_exit(&vdc->lock);
1224 			return (0);
1225 		}
1226 
1227 		/* call vdc_validate() asynchronously to avoid blocking */
1228 		if (taskq_dispatch(system_taskq, vdc_validate_task,
1229 		    (void *)vdc, TQ_NOSLEEP) == NULL) {
1230 			vdc_mark_closed(vdc, slice, flag, otyp);
1231 			mutex_exit(&vdc->lock);
1232 			return (ENXIO);
1233 		}
1234 
1235 		vdc->validate_pending++;
1236 		mutex_exit(&vdc->lock);
1237 		return (0);
1238 	}
1239 
1240 	mutex_exit(&vdc->lock);
1241 
1242 	vdc_validate(vdc);
1243 
1244 	mutex_enter(&vdc->lock);
1245 
1246 	if (vdc->vdisk_label == VD_DISK_LABEL_UNK ||
1247 	    vdc->slice[slice].nblocks == 0) {
1248 		vdc_mark_closed(vdc, slice, flag, otyp);
1249 		status = EIO;
1250 	}
1251 
1252 	mutex_exit(&vdc->lock);
1253 
1254 	return (status);
1255 }
1256 
1257 static int
1258 vdc_close(dev_t dev, int flag, int otyp, cred_t *cred)
1259 {
1260 	_NOTE(ARGUNUSED(cred))
1261 
1262 	int	instance;
1263 	int	slice;
1264 	int	rv, rval;
1265 	vdc_t	*vdc;
1266 
1267 	instance = VDCUNIT(dev);
1268 
1269 	if (otyp >= OTYPCNT)
1270 		return (EINVAL);
1271 
1272 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1273 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1274 		return (ENXIO);
1275 	}
1276 
1277 	DMSG(vdc, 0, "[%d] flag = %x, otyp = %x\n", instance, flag, otyp);
1278 
1279 	slice = VDCPART(dev);
1280 
1281 	/*
1282 	 * Attempt to flush the W$ on a close operation. If this is
1283 	 * not a supported IOCTL command or the backing device is read-only
1284 	 * do not fail the close operation.
1285 	 */
1286 	rv = vd_process_ioctl(dev, DKIOCFLUSHWRITECACHE, NULL, FKIOCTL, &rval);
1287 
1288 	if (rv != 0 && rv != ENOTSUP && rv != ENOTTY && rv != EROFS) {
1289 		DMSG(vdc, 0, "[%d] flush failed with error %d on close\n",
1290 		    instance, rv);
1291 		return (EIO);
1292 	}
1293 
1294 	mutex_enter(&vdc->lock);
1295 	vdc_mark_closed(vdc, slice, flag, otyp);
1296 	mutex_exit(&vdc->lock);
1297 
1298 	return (0);
1299 }
1300 
1301 static int
1302 vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
1303 {
1304 	_NOTE(ARGUNUSED(credp))
1305 
1306 	return (vd_process_ioctl(dev, cmd, (caddr_t)arg, mode, rvalp));
1307 }
1308 
1309 static int
1310 vdc_print(dev_t dev, char *str)
1311 {
1312 	cmn_err(CE_NOTE, "vdc%d:  %s", VDCUNIT(dev), str);
1313 	return (0);
1314 }
1315 
1316 static int
1317 vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk)
1318 {
1319 	int	rv;
1320 	size_t	nbytes = nblk * DEV_BSIZE;
1321 	int	instance = VDCUNIT(dev);
1322 	vdc_t	*vdc = NULL;
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, 2, "[%d] dump %ld bytes at block 0x%lx : addr=0x%p\n",
1330 	    instance, nbytes, blkno, (void *)addr);
1331 	rv = vdc_send_request(vdc, VD_OP_BWRITE, addr, nbytes,
1332 	    VDCPART(dev), blkno, CB_STRATEGY, 0, VIO_write_dir);
1333 	if (rv) {
1334 		DMSG(vdc, 0, "Failed to do a disk dump (err=%d)\n", rv);
1335 		return (rv);
1336 	}
1337 
1338 	if (ddi_in_panic())
1339 		(void) vdc_drain_response(vdc);
1340 
1341 	DMSG(vdc, 0, "[%d] End\n", instance);
1342 
1343 	return (0);
1344 }
1345 
1346 /* -------------------------------------------------------------------------- */
1347 
1348 /*
1349  * Disk access routines
1350  *
1351  */
1352 
1353 /*
1354  * vdc_strategy()
1355  *
1356  * Return Value:
1357  *	0:	As per strategy(9E), the strategy() function must return 0
1358  *		[ bioerror(9f) sets b_flags to the proper error code ]
1359  */
1360 static int
1361 vdc_strategy(struct buf *buf)
1362 {
1363 	int	rv = -1;
1364 	vdc_t	*vdc = NULL;
1365 	int	instance = VDCUNIT(buf->b_edev);
1366 	int	op = (buf->b_flags & B_READ) ? VD_OP_BREAD : VD_OP_BWRITE;
1367 	int	slice;
1368 
1369 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1370 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1371 		bioerror(buf, ENXIO);
1372 		biodone(buf);
1373 		return (0);
1374 	}
1375 
1376 	DMSG(vdc, 2, "[%d] %s %ld bytes at block %llx : b_addr=0x%p\n",
1377 	    instance, (buf->b_flags & B_READ) ? "Read" : "Write",
1378 	    buf->b_bcount, buf->b_lblkno, (void *)buf->b_un.b_addr);
1379 	DTRACE_IO2(vstart, buf_t *, buf, vdc_t *, vdc);
1380 
1381 	bp_mapin(buf);
1382 
1383 	if ((long)buf->b_private == VD_SLICE_NONE) {
1384 		/* I/O using an absolute disk offset */
1385 		slice = VD_SLICE_NONE;
1386 	} else {
1387 		slice = VDCPART(buf->b_edev);
1388 	}
1389 
1390 	rv = vdc_send_request(vdc, op, (caddr_t)buf->b_un.b_addr,
1391 	    buf->b_bcount, slice, buf->b_lblkno,
1392 	    CB_STRATEGY, buf, (op == VD_OP_BREAD) ? VIO_read_dir :
1393 	    VIO_write_dir);
1394 
1395 	/*
1396 	 * If the request was successfully sent, the strategy call returns and
1397 	 * the ACK handler calls the bioxxx functions when the vDisk server is
1398 	 * done.
1399 	 */
1400 	if (rv) {
1401 		DMSG(vdc, 0, "Failed to read/write (err=%d)\n", rv);
1402 		bioerror(buf, rv);
1403 		biodone(buf);
1404 	}
1405 
1406 	return (0);
1407 }
1408 
1409 /*
1410  * Function:
1411  *	vdc_min
1412  *
1413  * Description:
1414  *	Routine to limit the size of a data transfer. Used in
1415  *	conjunction with physio(9F).
1416  *
1417  * Arguments:
1418  *	bp - pointer to the indicated buf(9S) struct.
1419  *
1420  */
1421 static void
1422 vdc_min(struct buf *bufp)
1423 {
1424 	vdc_t	*vdc = NULL;
1425 	int	instance = VDCUNIT(bufp->b_edev);
1426 
1427 	vdc = ddi_get_soft_state(vdc_state, instance);
1428 	VERIFY(vdc != NULL);
1429 
1430 	if (bufp->b_bcount > (vdc->max_xfer_sz * vdc->block_size)) {
1431 		bufp->b_bcount = vdc->max_xfer_sz * vdc->block_size;
1432 	}
1433 }
1434 
1435 static int
1436 vdc_read(dev_t dev, struct uio *uio, cred_t *cred)
1437 {
1438 	_NOTE(ARGUNUSED(cred))
1439 
1440 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1441 	return (physio(vdc_strategy, NULL, dev, B_READ, vdc_min, uio));
1442 }
1443 
1444 static int
1445 vdc_write(dev_t dev, struct uio *uio, cred_t *cred)
1446 {
1447 	_NOTE(ARGUNUSED(cred))
1448 
1449 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1450 	return (physio(vdc_strategy, NULL, dev, B_WRITE, vdc_min, uio));
1451 }
1452 
1453 static int
1454 vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred)
1455 {
1456 	_NOTE(ARGUNUSED(cred))
1457 
1458 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1459 	return (aphysio(vdc_strategy, anocancel, dev, B_READ, vdc_min, aio));
1460 }
1461 
1462 static int
1463 vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred)
1464 {
1465 	_NOTE(ARGUNUSED(cred))
1466 
1467 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1468 	return (aphysio(vdc_strategy, anocancel, dev, B_WRITE, vdc_min, aio));
1469 }
1470 
1471 
1472 /* -------------------------------------------------------------------------- */
1473 
1474 /*
1475  * Handshake support
1476  */
1477 
1478 
1479 /*
1480  * Function:
1481  *	vdc_init_ver_negotiation()
1482  *
1483  * Description:
1484  *
1485  * Arguments:
1486  *	vdc	- soft state pointer for this instance of the device driver.
1487  *
1488  * Return Code:
1489  *	0	- Success
1490  */
1491 static int
1492 vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver)
1493 {
1494 	vio_ver_msg_t	pkt;
1495 	size_t		msglen = sizeof (pkt);
1496 	int		status = -1;
1497 
1498 	ASSERT(vdc != NULL);
1499 	ASSERT(mutex_owned(&vdc->lock));
1500 
1501 	DMSG(vdc, 0, "[%d] Entered.\n", vdc->instance);
1502 
1503 	/*
1504 	 * set the Session ID to a unique value
1505 	 * (the lower 32 bits of the clock tick)
1506 	 */
1507 	vdc->session_id = ((uint32_t)gettick() & 0xffffffff);
1508 	DMSG(vdc, 0, "[%d] Set SID to 0x%lx\n", vdc->instance, vdc->session_id);
1509 
1510 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1511 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1512 	pkt.tag.vio_subtype_env = VIO_VER_INFO;
1513 	pkt.tag.vio_sid = vdc->session_id;
1514 	pkt.dev_class = VDEV_DISK;
1515 	pkt.ver_major = ver.major;
1516 	pkt.ver_minor = ver.minor;
1517 
1518 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1519 	DMSG(vdc, 0, "[%d] Ver info sent (status = %d)\n",
1520 	    vdc->instance, status);
1521 	if ((status != 0) || (msglen != sizeof (vio_ver_msg_t))) {
1522 		DMSG(vdc, 0, "[%d] Failed to send Ver negotiation info: "
1523 		    "id(%lx) rv(%d) size(%ld)", vdc->instance, vdc->ldc_handle,
1524 		    status, msglen);
1525 		if (msglen != sizeof (vio_ver_msg_t))
1526 			status = ENOMSG;
1527 	}
1528 
1529 	return (status);
1530 }
1531 
1532 /*
1533  * Function:
1534  *	vdc_ver_negotiation()
1535  *
1536  * Description:
1537  *
1538  * Arguments:
1539  *	vdcp	- soft state pointer for this instance of the device driver.
1540  *
1541  * Return Code:
1542  *	0	- Success
1543  */
1544 static int
1545 vdc_ver_negotiation(vdc_t *vdcp)
1546 {
1547 	vio_msg_t vio_msg;
1548 	int status;
1549 
1550 	if (status = vdc_init_ver_negotiation(vdcp, vdc_version[0]))
1551 		return (status);
1552 
1553 	/* release lock and wait for response */
1554 	mutex_exit(&vdcp->lock);
1555 	status = vdc_wait_for_response(vdcp, &vio_msg);
1556 	mutex_enter(&vdcp->lock);
1557 	if (status) {
1558 		DMSG(vdcp, 0,
1559 		    "[%d] Failed waiting for Ver negotiation response, rv(%d)",
1560 		    vdcp->instance, status);
1561 		return (status);
1562 	}
1563 
1564 	/* check type and sub_type ... */
1565 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1566 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1567 		DMSG(vdcp, 0, "[%d] Invalid ver negotiation response\n",
1568 		    vdcp->instance);
1569 		return (EPROTO);
1570 	}
1571 
1572 	return (vdc_handle_ver_msg(vdcp, (vio_ver_msg_t *)&vio_msg));
1573 }
1574 
1575 /*
1576  * Function:
1577  *	vdc_init_attr_negotiation()
1578  *
1579  * Description:
1580  *
1581  * Arguments:
1582  *	vdc	- soft state pointer for this instance of the device driver.
1583  *
1584  * Return Code:
1585  *	0	- Success
1586  */
1587 static int
1588 vdc_init_attr_negotiation(vdc_t *vdc)
1589 {
1590 	vd_attr_msg_t	pkt;
1591 	size_t		msglen = sizeof (pkt);
1592 	int		status;
1593 
1594 	ASSERT(vdc != NULL);
1595 	ASSERT(mutex_owned(&vdc->lock));
1596 
1597 	DMSG(vdc, 0, "[%d] entered\n", vdc->instance);
1598 
1599 	/* fill in tag */
1600 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1601 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1602 	pkt.tag.vio_subtype_env = VIO_ATTR_INFO;
1603 	pkt.tag.vio_sid = vdc->session_id;
1604 	/* fill in payload */
1605 	pkt.max_xfer_sz = vdc->max_xfer_sz;
1606 	pkt.vdisk_block_size = vdc->block_size;
1607 	pkt.xfer_mode = VIO_DRING_MODE;
1608 	pkt.operations = 0;	/* server will set bits of valid operations */
1609 	pkt.vdisk_type = 0;	/* server will set to valid device type */
1610 	pkt.vdisk_media = 0;	/* server will set to valid media type */
1611 	pkt.vdisk_size = 0;	/* server will set to valid size */
1612 
1613 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1614 	DMSG(vdc, 0, "Attr info sent (status = %d)\n", status);
1615 
1616 	if ((status != 0) || (msglen != sizeof (vio_ver_msg_t))) {
1617 		DMSG(vdc, 0, "[%d] Failed to send Attr negotiation info: "
1618 		    "id(%lx) rv(%d) size(%ld)", vdc->instance, vdc->ldc_handle,
1619 		    status, msglen);
1620 		if (msglen != sizeof (vio_ver_msg_t))
1621 			status = ENOMSG;
1622 	}
1623 
1624 	return (status);
1625 }
1626 
1627 /*
1628  * Function:
1629  *	vdc_attr_negotiation()
1630  *
1631  * Description:
1632  *
1633  * Arguments:
1634  *	vdc	- soft state pointer for this instance of the device driver.
1635  *
1636  * Return Code:
1637  *	0	- Success
1638  */
1639 static int
1640 vdc_attr_negotiation(vdc_t *vdcp)
1641 {
1642 	int status;
1643 	vio_msg_t vio_msg;
1644 
1645 	if (status = vdc_init_attr_negotiation(vdcp))
1646 		return (status);
1647 
1648 	/* release lock and wait for response */
1649 	mutex_exit(&vdcp->lock);
1650 	status = vdc_wait_for_response(vdcp, &vio_msg);
1651 	mutex_enter(&vdcp->lock);
1652 	if (status) {
1653 		DMSG(vdcp, 0,
1654 		    "[%d] Failed waiting for Attr negotiation response, rv(%d)",
1655 		    vdcp->instance, status);
1656 		return (status);
1657 	}
1658 
1659 	/* check type and sub_type ... */
1660 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1661 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1662 		DMSG(vdcp, 0, "[%d] Invalid attr negotiation response\n",
1663 		    vdcp->instance);
1664 		return (EPROTO);
1665 	}
1666 
1667 	return (vdc_handle_attr_msg(vdcp, (vd_attr_msg_t *)&vio_msg));
1668 }
1669 
1670 
1671 /*
1672  * Function:
1673  *	vdc_init_dring_negotiate()
1674  *
1675  * Description:
1676  *
1677  * Arguments:
1678  *	vdc	- soft state pointer for this instance of the device driver.
1679  *
1680  * Return Code:
1681  *	0	- Success
1682  */
1683 static int
1684 vdc_init_dring_negotiate(vdc_t *vdc)
1685 {
1686 	vio_dring_reg_msg_t	pkt;
1687 	size_t			msglen = sizeof (pkt);
1688 	int			status = -1;
1689 	int			retry;
1690 	int			nretries = 10;
1691 
1692 	ASSERT(vdc != NULL);
1693 	ASSERT(mutex_owned(&vdc->lock));
1694 
1695 	for (retry = 0; retry < nretries; retry++) {
1696 		status = vdc_init_descriptor_ring(vdc);
1697 		if (status != EAGAIN)
1698 			break;
1699 		drv_usecwait(vdc_min_timeout_ldc);
1700 	}
1701 
1702 	if (status != 0) {
1703 		DMSG(vdc, 0, "[%d] Failed to init DRing (status = %d)\n",
1704 		    vdc->instance, status);
1705 		return (status);
1706 	}
1707 
1708 	DMSG(vdc, 0, "[%d] Init of descriptor ring completed (status = %d)\n",
1709 	    vdc->instance, status);
1710 
1711 	/* fill in tag */
1712 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1713 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1714 	pkt.tag.vio_subtype_env = VIO_DRING_REG;
1715 	pkt.tag.vio_sid = vdc->session_id;
1716 	/* fill in payload */
1717 	pkt.dring_ident = 0;
1718 	pkt.num_descriptors = vdc->dring_len;
1719 	pkt.descriptor_size = vdc->dring_entry_size;
1720 	pkt.options = (VIO_TX_DRING | VIO_RX_DRING);
1721 	pkt.ncookies = vdc->dring_cookie_count;
1722 	pkt.cookie[0] = vdc->dring_cookie[0];	/* for now just one cookie */
1723 
1724 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1725 	if (status != 0) {
1726 		DMSG(vdc, 0, "[%d] Failed to register DRing (err = %d)",
1727 		    vdc->instance, status);
1728 	}
1729 
1730 	return (status);
1731 }
1732 
1733 
1734 /*
1735  * Function:
1736  *	vdc_dring_negotiation()
1737  *
1738  * Description:
1739  *
1740  * Arguments:
1741  *	vdc	- soft state pointer for this instance of the device driver.
1742  *
1743  * Return Code:
1744  *	0	- Success
1745  */
1746 static int
1747 vdc_dring_negotiation(vdc_t *vdcp)
1748 {
1749 	int status;
1750 	vio_msg_t vio_msg;
1751 
1752 	if (status = vdc_init_dring_negotiate(vdcp))
1753 		return (status);
1754 
1755 	/* release lock and wait for response */
1756 	mutex_exit(&vdcp->lock);
1757 	status = vdc_wait_for_response(vdcp, &vio_msg);
1758 	mutex_enter(&vdcp->lock);
1759 	if (status) {
1760 		DMSG(vdcp, 0,
1761 		    "[%d] Failed waiting for Dring negotiation response,"
1762 		    " rv(%d)", vdcp->instance, status);
1763 		return (status);
1764 	}
1765 
1766 	/* check type and sub_type ... */
1767 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1768 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1769 		DMSG(vdcp, 0, "[%d] Invalid Dring negotiation response\n",
1770 		    vdcp->instance);
1771 		return (EPROTO);
1772 	}
1773 
1774 	return (vdc_handle_dring_reg_msg(vdcp,
1775 	    (vio_dring_reg_msg_t *)&vio_msg));
1776 }
1777 
1778 
1779 /*
1780  * Function:
1781  *	vdc_send_rdx()
1782  *
1783  * Description:
1784  *
1785  * Arguments:
1786  *	vdc	- soft state pointer for this instance of the device driver.
1787  *
1788  * Return Code:
1789  *	0	- Success
1790  */
1791 static int
1792 vdc_send_rdx(vdc_t *vdcp)
1793 {
1794 	vio_msg_t	msg;
1795 	size_t		msglen = sizeof (vio_msg_t);
1796 	int		status;
1797 
1798 	/*
1799 	 * Send an RDX message to vds to indicate we are ready
1800 	 * to send data
1801 	 */
1802 	msg.tag.vio_msgtype = VIO_TYPE_CTRL;
1803 	msg.tag.vio_subtype = VIO_SUBTYPE_INFO;
1804 	msg.tag.vio_subtype_env = VIO_RDX;
1805 	msg.tag.vio_sid = vdcp->session_id;
1806 	status = vdc_send(vdcp, (caddr_t)&msg, &msglen);
1807 	if (status != 0) {
1808 		DMSG(vdcp, 0, "[%d] Failed to send RDX message (%d)",
1809 		    vdcp->instance, status);
1810 	}
1811 
1812 	return (status);
1813 }
1814 
1815 /*
1816  * Function:
1817  *	vdc_handle_rdx()
1818  *
1819  * Description:
1820  *
1821  * Arguments:
1822  *	vdc	- soft state pointer for this instance of the device driver.
1823  *	msgp	- received msg
1824  *
1825  * Return Code:
1826  *	0	- Success
1827  */
1828 static int
1829 vdc_handle_rdx(vdc_t *vdcp, vio_rdx_msg_t *msgp)
1830 {
1831 	_NOTE(ARGUNUSED(vdcp))
1832 	_NOTE(ARGUNUSED(msgp))
1833 
1834 	ASSERT(msgp->tag.vio_msgtype == VIO_TYPE_CTRL);
1835 	ASSERT(msgp->tag.vio_subtype == VIO_SUBTYPE_ACK);
1836 	ASSERT(msgp->tag.vio_subtype_env == VIO_RDX);
1837 
1838 	DMSG(vdcp, 1, "[%d] Got an RDX msg", vdcp->instance);
1839 
1840 	return (0);
1841 }
1842 
1843 /*
1844  * Function:
1845  *	vdc_rdx_exchange()
1846  *
1847  * Description:
1848  *
1849  * Arguments:
1850  *	vdc	- soft state pointer for this instance of the device driver.
1851  *
1852  * Return Code:
1853  *	0	- Success
1854  */
1855 static int
1856 vdc_rdx_exchange(vdc_t *vdcp)
1857 {
1858 	int status;
1859 	vio_msg_t vio_msg;
1860 
1861 	if (status = vdc_send_rdx(vdcp))
1862 		return (status);
1863 
1864 	/* release lock and wait for response */
1865 	mutex_exit(&vdcp->lock);
1866 	status = vdc_wait_for_response(vdcp, &vio_msg);
1867 	mutex_enter(&vdcp->lock);
1868 	if (status) {
1869 		DMSG(vdcp, 0, "[%d] Failed waiting for RDX response, rv(%d)",
1870 		    vdcp->instance, status);
1871 		return (status);
1872 	}
1873 
1874 	/* check type and sub_type ... */
1875 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1876 	    vio_msg.tag.vio_subtype != VIO_SUBTYPE_ACK) {
1877 		DMSG(vdcp, 0, "[%d] Invalid RDX response\n", vdcp->instance);
1878 		return (EPROTO);
1879 	}
1880 
1881 	return (vdc_handle_rdx(vdcp, (vio_rdx_msg_t *)&vio_msg));
1882 }
1883 
1884 
1885 /* -------------------------------------------------------------------------- */
1886 
1887 /*
1888  * LDC helper routines
1889  */
1890 
1891 static int
1892 vdc_recv(vdc_t *vdc, vio_msg_t *msgp, size_t *nbytesp)
1893 {
1894 	int		status;
1895 	boolean_t	q_has_pkts = B_FALSE;
1896 	uint64_t	delay_time;
1897 	size_t		len;
1898 
1899 	mutex_enter(&vdc->read_lock);
1900 
1901 	if (vdc->read_state == VDC_READ_IDLE)
1902 		vdc->read_state = VDC_READ_WAITING;
1903 
1904 	while (vdc->read_state != VDC_READ_PENDING) {
1905 
1906 		/* detect if the connection has been reset */
1907 		if (vdc->read_state == VDC_READ_RESET) {
1908 			status = ECONNRESET;
1909 			goto done;
1910 		}
1911 
1912 		cv_wait(&vdc->read_cv, &vdc->read_lock);
1913 	}
1914 
1915 	/*
1916 	 * Until we get a blocking ldc read we have to retry
1917 	 * until the entire LDC message has arrived before
1918 	 * ldc_read() will succeed. Note we also bail out if
1919 	 * the channel is reset or goes away.
1920 	 */
1921 	delay_time = vdc_ldc_read_init_delay;
1922 loop:
1923 	len = *nbytesp;
1924 	status = ldc_read(vdc->ldc_handle, (caddr_t)msgp, &len);
1925 	switch (status) {
1926 	case EAGAIN:
1927 		delay_time *= 2;
1928 		if (delay_time >= vdc_ldc_read_max_delay)
1929 			delay_time = vdc_ldc_read_max_delay;
1930 		delay(delay_time);
1931 		goto loop;
1932 
1933 	case 0:
1934 		if (len == 0) {
1935 			DMSG(vdc, 1, "[%d] ldc_read returned 0 bytes with "
1936 			    "no error!\n", vdc->instance);
1937 			goto loop;
1938 		}
1939 
1940 		*nbytesp = len;
1941 
1942 		/*
1943 		 * If there are pending messages, leave the
1944 		 * read state as pending. Otherwise, set the state
1945 		 * back to idle.
1946 		 */
1947 		status = ldc_chkq(vdc->ldc_handle, &q_has_pkts);
1948 		if (status == 0 && !q_has_pkts)
1949 			vdc->read_state = VDC_READ_IDLE;
1950 
1951 		break;
1952 	default:
1953 		DMSG(vdc, 0, "ldc_read returned %d\n", status);
1954 		break;
1955 	}
1956 
1957 done:
1958 	mutex_exit(&vdc->read_lock);
1959 
1960 	return (status);
1961 }
1962 
1963 
1964 
1965 #ifdef DEBUG
1966 void
1967 vdc_decode_tag(vdc_t *vdcp, vio_msg_t *msg)
1968 {
1969 	char *ms, *ss, *ses;
1970 	switch (msg->tag.vio_msgtype) {
1971 #define	Q(_s)	case _s : ms = #_s; break;
1972 	Q(VIO_TYPE_CTRL)
1973 	Q(VIO_TYPE_DATA)
1974 	Q(VIO_TYPE_ERR)
1975 #undef Q
1976 	default: ms = "unknown"; break;
1977 	}
1978 
1979 	switch (msg->tag.vio_subtype) {
1980 #define	Q(_s)	case _s : ss = #_s; break;
1981 	Q(VIO_SUBTYPE_INFO)
1982 	Q(VIO_SUBTYPE_ACK)
1983 	Q(VIO_SUBTYPE_NACK)
1984 #undef Q
1985 	default: ss = "unknown"; break;
1986 	}
1987 
1988 	switch (msg->tag.vio_subtype_env) {
1989 #define	Q(_s)	case _s : ses = #_s; break;
1990 	Q(VIO_VER_INFO)
1991 	Q(VIO_ATTR_INFO)
1992 	Q(VIO_DRING_REG)
1993 	Q(VIO_DRING_UNREG)
1994 	Q(VIO_RDX)
1995 	Q(VIO_PKT_DATA)
1996 	Q(VIO_DESC_DATA)
1997 	Q(VIO_DRING_DATA)
1998 #undef Q
1999 	default: ses = "unknown"; break;
2000 	}
2001 
2002 	DMSG(vdcp, 3, "(%x/%x/%x) message : (%s/%s/%s)\n",
2003 	    msg->tag.vio_msgtype, msg->tag.vio_subtype,
2004 	    msg->tag.vio_subtype_env, ms, ss, ses);
2005 }
2006 #endif
2007 
2008 /*
2009  * Function:
2010  *	vdc_send()
2011  *
2012  * Description:
2013  *	The function encapsulates the call to write a message using LDC.
2014  *	If LDC indicates that the call failed due to the queue being full,
2015  *	we retry the ldc_write(), otherwise we return the error returned by LDC.
2016  *
2017  * Arguments:
2018  *	ldc_handle	- LDC handle for the channel this instance of vdc uses
2019  *	pkt		- address of LDC message to be sent
2020  *	msglen		- the size of the message being sent. When the function
2021  *			  returns, this contains the number of bytes written.
2022  *
2023  * Return Code:
2024  *	0		- Success.
2025  *	EINVAL		- pkt or msglen were NULL
2026  *	ECONNRESET	- The connection was not up.
2027  *	EWOULDBLOCK	- LDC queue is full
2028  *	xxx		- other error codes returned by ldc_write
2029  */
2030 static int
2031 vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen)
2032 {
2033 	size_t	size = 0;
2034 	int	status = 0;
2035 	clock_t delay_ticks;
2036 
2037 	ASSERT(vdc != NULL);
2038 	ASSERT(mutex_owned(&vdc->lock));
2039 	ASSERT(msglen != NULL);
2040 	ASSERT(*msglen != 0);
2041 
2042 #ifdef DEBUG
2043 	vdc_decode_tag(vdc, (vio_msg_t *)(uintptr_t)pkt);
2044 #endif
2045 	/*
2046 	 * Wait indefinitely to send if channel
2047 	 * is busy, but bail out if we succeed or
2048 	 * if the channel closes or is reset.
2049 	 */
2050 	delay_ticks = vdc_hz_min_ldc_delay;
2051 	do {
2052 		size = *msglen;
2053 		status = ldc_write(vdc->ldc_handle, pkt, &size);
2054 		if (status == EWOULDBLOCK) {
2055 			delay(delay_ticks);
2056 			/* geometric backoff */
2057 			delay_ticks *= 2;
2058 			if (delay_ticks > vdc_hz_max_ldc_delay)
2059 				delay_ticks = vdc_hz_max_ldc_delay;
2060 		}
2061 	} while (status == EWOULDBLOCK);
2062 
2063 	/* if LDC had serious issues --- reset vdc state */
2064 	if (status == EIO || status == ECONNRESET) {
2065 		/* LDC had serious issues --- reset vdc state */
2066 		mutex_enter(&vdc->read_lock);
2067 		if ((vdc->read_state == VDC_READ_WAITING) ||
2068 		    (vdc->read_state == VDC_READ_RESET))
2069 			cv_signal(&vdc->read_cv);
2070 		vdc->read_state = VDC_READ_RESET;
2071 		mutex_exit(&vdc->read_lock);
2072 
2073 		/* wake up any waiters in the reset thread */
2074 		if (vdc->state == VDC_STATE_INIT_WAITING) {
2075 			DMSG(vdc, 0, "[%d] write reset - "
2076 			    "vdc is resetting ..\n", vdc->instance);
2077 			vdc->state = VDC_STATE_RESETTING;
2078 			cv_signal(&vdc->initwait_cv);
2079 		}
2080 
2081 		return (ECONNRESET);
2082 	}
2083 
2084 	/* return the last size written */
2085 	*msglen = size;
2086 
2087 	return (status);
2088 }
2089 
2090 /*
2091  * Function:
2092  *	vdc_get_md_node
2093  *
2094  * Description:
2095  *	Get the MD, the device node and the port node for the given
2096  *	disk instance. The caller is responsible for cleaning up the
2097  *	reference to the returned MD (mdpp) by calling md_fini_handle().
2098  *
2099  * Arguments:
2100  *	dip	- dev info pointer for this instance of the device driver.
2101  *	mdpp	- the returned MD.
2102  *	vd_nodep - the returned device node.
2103  *	vd_portp - the returned port node. The returned port node is NULL
2104  *		   if no port node is found.
2105  *
2106  * Return Code:
2107  *	0	- Success.
2108  *	ENOENT	- Expected node or property did not exist.
2109  *	ENXIO	- Unexpected error communicating with MD framework
2110  */
2111 static int
2112 vdc_get_md_node(dev_info_t *dip, md_t **mdpp, mde_cookie_t *vd_nodep,
2113     mde_cookie_t *vd_portp)
2114 {
2115 	int		status = ENOENT;
2116 	char		*node_name = NULL;
2117 	md_t		*mdp = NULL;
2118 	int		num_nodes;
2119 	int		num_vdevs;
2120 	int		num_vports;
2121 	mde_cookie_t	rootnode;
2122 	mde_cookie_t	*listp = NULL;
2123 	boolean_t	found_inst = B_FALSE;
2124 	int		listsz;
2125 	int		idx;
2126 	uint64_t	md_inst;
2127 	int		obp_inst;
2128 	int		instance = ddi_get_instance(dip);
2129 
2130 	/*
2131 	 * Get the OBP instance number for comparison with the MD instance
2132 	 *
2133 	 * The "cfg-handle" property of a vdc node in an MD contains the MD's
2134 	 * notion of "instance", or unique identifier, for that node; OBP
2135 	 * stores the value of the "cfg-handle" MD property as the value of
2136 	 * the "reg" property on the node in the device tree it builds from
2137 	 * the MD and passes to Solaris.  Thus, we look up the devinfo node's
2138 	 * "reg" property value to uniquely identify this device instance.
2139 	 * If the "reg" property cannot be found, the device tree state is
2140 	 * presumably so broken that there is no point in continuing.
2141 	 */
2142 	if (!ddi_prop_exists(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS, OBP_REG)) {
2143 		cmn_err(CE_WARN, "'%s' property does not exist", OBP_REG);
2144 		return (ENOENT);
2145 	}
2146 	obp_inst = ddi_prop_get_int(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
2147 	    OBP_REG, -1);
2148 	DMSGX(1, "[%d] OBP inst=%d\n", instance, obp_inst);
2149 
2150 	/*
2151 	 * We now walk the MD nodes to find the node for this vdisk.
2152 	 */
2153 	if ((mdp = md_get_handle()) == NULL) {
2154 		cmn_err(CE_WARN, "unable to init machine description");
2155 		return (ENXIO);
2156 	}
2157 
2158 	num_nodes = md_node_count(mdp);
2159 	ASSERT(num_nodes > 0);
2160 
2161 	listsz = num_nodes * sizeof (mde_cookie_t);
2162 
2163 	/* allocate memory for nodes */
2164 	listp = kmem_zalloc(listsz, KM_SLEEP);
2165 
2166 	rootnode = md_root_node(mdp);
2167 	ASSERT(rootnode != MDE_INVAL_ELEM_COOKIE);
2168 
2169 	/*
2170 	 * Search for all the virtual devices, we will then check to see which
2171 	 * ones are disk nodes.
2172 	 */
2173 	num_vdevs = md_scan_dag(mdp, rootnode,
2174 	    md_find_name(mdp, VDC_MD_VDEV_NAME),
2175 	    md_find_name(mdp, "fwd"), listp);
2176 
2177 	if (num_vdevs <= 0) {
2178 		cmn_err(CE_NOTE, "No '%s' node found", VDC_MD_VDEV_NAME);
2179 		status = ENOENT;
2180 		goto done;
2181 	}
2182 
2183 	DMSGX(1, "[%d] num_vdevs=%d\n", instance, num_vdevs);
2184 	for (idx = 0; idx < num_vdevs; idx++) {
2185 		status = md_get_prop_str(mdp, listp[idx], "name", &node_name);
2186 		if ((status != 0) || (node_name == NULL)) {
2187 			cmn_err(CE_NOTE, "Unable to get name of node type '%s'"
2188 			    ": err %d", VDC_MD_VDEV_NAME, status);
2189 			continue;
2190 		}
2191 
2192 		DMSGX(1, "[%d] Found node '%s'\n", instance, node_name);
2193 		if (strcmp(VDC_MD_DISK_NAME, node_name) == 0) {
2194 			status = md_get_prop_val(mdp, listp[idx],
2195 			    VDC_MD_CFG_HDL, &md_inst);
2196 			DMSGX(1, "[%d] vdc inst in MD=%lx\n",
2197 			    instance, md_inst);
2198 			if ((status == 0) && (md_inst == obp_inst)) {
2199 				found_inst = B_TRUE;
2200 				break;
2201 			}
2202 		}
2203 	}
2204 
2205 	if (!found_inst) {
2206 		DMSGX(0, "Unable to find correct '%s' node", VDC_MD_DISK_NAME);
2207 		status = ENOENT;
2208 		goto done;
2209 	}
2210 	DMSGX(0, "[%d] MD inst=%lx\n", instance, md_inst);
2211 
2212 	*vd_nodep = listp[idx];
2213 	*mdpp = mdp;
2214 
2215 	num_vports = md_scan_dag(mdp, *vd_nodep,
2216 	    md_find_name(mdp, VDC_MD_PORT_NAME),
2217 	    md_find_name(mdp, "fwd"), listp);
2218 
2219 	if (num_vports != 1) {
2220 		DMSGX(0, "Expected 1 '%s' node for '%s' port, found %d\n",
2221 		    VDC_MD_PORT_NAME, VDC_MD_VDEV_NAME, num_vports);
2222 	}
2223 
2224 	*vd_portp = (num_vports == 0)? NULL: listp[0];
2225 
2226 done:
2227 	kmem_free(listp, listsz);
2228 	return (status);
2229 }
2230 
2231 /*
2232  * Function:
2233  *	vdc_get_ldc_id()
2234  *
2235  * Description:
2236  *	This function gets the 'ldc-id' for this particular instance of vdc.
2237  *	The id returned is the guest domain channel endpoint LDC uses for
2238  *	communication with vds.
2239  *
2240  * Arguments:
2241  *	mdp	- pointer to the machine description.
2242  *	vd_node	- the vdisk element from the MD.
2243  *	ldc_id	- pointer to variable used to return the 'ldc-id' found.
2244  *
2245  * Return Code:
2246  *	0	- Success.
2247  *	ENOENT	- Expected node or property did not exist.
2248  */
2249 static int
2250 vdc_get_ldc_id(md_t *mdp, mde_cookie_t vd_node, uint64_t *ldc_id)
2251 {
2252 	mde_cookie_t	*chanp = NULL;
2253 	int		listsz;
2254 	int		num_chans;
2255 	int		num_nodes;
2256 	int		status = 0;
2257 
2258 	num_nodes = md_node_count(mdp);
2259 	ASSERT(num_nodes > 0);
2260 
2261 	listsz = num_nodes * sizeof (mde_cookie_t);
2262 
2263 	/* allocate memory for nodes */
2264 	chanp = kmem_zalloc(listsz, KM_SLEEP);
2265 
2266 	/* get the channels for this node */
2267 	num_chans = md_scan_dag(mdp, vd_node,
2268 	    md_find_name(mdp, VDC_MD_CHAN_NAME),
2269 	    md_find_name(mdp, "fwd"), chanp);
2270 
2271 	/* expecting at least one channel */
2272 	if (num_chans <= 0) {
2273 		cmn_err(CE_NOTE, "No '%s' node for '%s' port",
2274 		    VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME);
2275 		status = ENOENT;
2276 		goto done;
2277 
2278 	} else if (num_chans != 1) {
2279 		DMSGX(0, "Expected 1 '%s' node for '%s' port, found %d\n",
2280 		    VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME, num_chans);
2281 	}
2282 
2283 	/*
2284 	 * We use the first channel found (index 0), irrespective of how
2285 	 * many are there in total.
2286 	 */
2287 	if (md_get_prop_val(mdp, chanp[0], VDC_MD_ID, ldc_id) != 0) {
2288 		cmn_err(CE_NOTE, "Channel '%s' property not found", VDC_MD_ID);
2289 		status = ENOENT;
2290 	}
2291 
2292 done:
2293 	kmem_free(chanp, listsz);
2294 	return (status);
2295 }
2296 
2297 static int
2298 vdc_do_ldc_up(vdc_t *vdc)
2299 {
2300 	int		status;
2301 	ldc_status_t	ldc_state;
2302 
2303 	DMSG(vdc, 0, "[%d] Bringing up channel %lx\n",
2304 	    vdc->instance, vdc->ldc_id);
2305 
2306 	if (vdc->lifecycle == VDC_LC_DETACHING)
2307 		return (EINVAL);
2308 
2309 	if ((status = ldc_up(vdc->ldc_handle)) != 0) {
2310 		switch (status) {
2311 		case ECONNREFUSED:	/* listener not ready at other end */
2312 			DMSG(vdc, 0, "[%d] ldc_up(%lx,...) return %d\n",
2313 			    vdc->instance, vdc->ldc_id, status);
2314 			status = 0;
2315 			break;
2316 		default:
2317 			DMSG(vdc, 0, "[%d] Failed to bring up LDC: "
2318 			    "channel=%ld, err=%d", vdc->instance, vdc->ldc_id,
2319 			    status);
2320 			break;
2321 		}
2322 	}
2323 
2324 	if (ldc_status(vdc->ldc_handle, &ldc_state) == 0) {
2325 		vdc->ldc_state = ldc_state;
2326 		if (ldc_state == LDC_UP) {
2327 			DMSG(vdc, 0, "[%d] LDC channel already up\n",
2328 			    vdc->instance);
2329 			vdc->seq_num = 1;
2330 			vdc->seq_num_reply = 0;
2331 		}
2332 	}
2333 
2334 	return (status);
2335 }
2336 
2337 /*
2338  * Function:
2339  *	vdc_terminate_ldc()
2340  *
2341  * Description:
2342  *
2343  * Arguments:
2344  *	vdc	- soft state pointer for this instance of the device driver.
2345  *
2346  * Return Code:
2347  *	None
2348  */
2349 static void
2350 vdc_terminate_ldc(vdc_t *vdc)
2351 {
2352 	int	instance = ddi_get_instance(vdc->dip);
2353 
2354 	ASSERT(vdc != NULL);
2355 	ASSERT(mutex_owned(&vdc->lock));
2356 
2357 	DMSG(vdc, 0, "[%d] initialized=%x\n", instance, vdc->initialized);
2358 
2359 	if (vdc->initialized & VDC_LDC_OPEN) {
2360 		DMSG(vdc, 0, "[%d] ldc_close()\n", instance);
2361 		(void) ldc_close(vdc->ldc_handle);
2362 	}
2363 	if (vdc->initialized & VDC_LDC_CB) {
2364 		DMSG(vdc, 0, "[%d] ldc_unreg_callback()\n", instance);
2365 		(void) ldc_unreg_callback(vdc->ldc_handle);
2366 	}
2367 	if (vdc->initialized & VDC_LDC) {
2368 		DMSG(vdc, 0, "[%d] ldc_fini()\n", instance);
2369 		(void) ldc_fini(vdc->ldc_handle);
2370 		vdc->ldc_handle = NULL;
2371 	}
2372 
2373 	vdc->initialized &= ~(VDC_LDC | VDC_LDC_CB | VDC_LDC_OPEN);
2374 }
2375 
2376 /* -------------------------------------------------------------------------- */
2377 
2378 /*
2379  * Descriptor Ring helper routines
2380  */
2381 
2382 /*
2383  * Function:
2384  *	vdc_init_descriptor_ring()
2385  *
2386  * Description:
2387  *
2388  * Arguments:
2389  *	vdc	- soft state pointer for this instance of the device driver.
2390  *
2391  * Return Code:
2392  *	0	- Success
2393  */
2394 static int
2395 vdc_init_descriptor_ring(vdc_t *vdc)
2396 {
2397 	vd_dring_entry_t	*dep = NULL;	/* DRing Entry pointer */
2398 	int	status = 0;
2399 	int	i;
2400 
2401 	DMSG(vdc, 0, "[%d] initialized=%x\n", vdc->instance, vdc->initialized);
2402 
2403 	ASSERT(vdc != NULL);
2404 	ASSERT(mutex_owned(&vdc->lock));
2405 	ASSERT(vdc->ldc_handle != NULL);
2406 
2407 	/* ensure we have enough room to store max sized block */
2408 	ASSERT(maxphys <= VD_MAX_BLOCK_SIZE);
2409 
2410 	if ((vdc->initialized & VDC_DRING_INIT) == 0) {
2411 		DMSG(vdc, 0, "[%d] ldc_mem_dring_create\n", vdc->instance);
2412 		/*
2413 		 * Calculate the maximum block size we can transmit using one
2414 		 * Descriptor Ring entry from the attributes returned by the
2415 		 * vDisk server. This is subject to a minimum of 'maxphys'
2416 		 * as we do not have the capability to split requests over
2417 		 * multiple DRing entries.
2418 		 */
2419 		if ((vdc->max_xfer_sz * vdc->block_size) < maxphys) {
2420 			DMSG(vdc, 0, "[%d] using minimum DRing size\n",
2421 			    vdc->instance);
2422 			vdc->dring_max_cookies = maxphys / PAGESIZE;
2423 		} else {
2424 			vdc->dring_max_cookies =
2425 			    (vdc->max_xfer_sz * vdc->block_size) / PAGESIZE;
2426 		}
2427 		vdc->dring_entry_size = (sizeof (vd_dring_entry_t) +
2428 		    (sizeof (ldc_mem_cookie_t) *
2429 		    (vdc->dring_max_cookies - 1)));
2430 		vdc->dring_len = VD_DRING_LEN;
2431 
2432 		status = ldc_mem_dring_create(vdc->dring_len,
2433 		    vdc->dring_entry_size, &vdc->ldc_dring_hdl);
2434 		if ((vdc->ldc_dring_hdl == NULL) || (status != 0)) {
2435 			DMSG(vdc, 0, "[%d] Descriptor ring creation failed",
2436 			    vdc->instance);
2437 			return (status);
2438 		}
2439 		vdc->initialized |= VDC_DRING_INIT;
2440 	}
2441 
2442 	if ((vdc->initialized & VDC_DRING_BOUND) == 0) {
2443 		DMSG(vdc, 0, "[%d] ldc_mem_dring_bind\n", vdc->instance);
2444 		vdc->dring_cookie =
2445 		    kmem_zalloc(sizeof (ldc_mem_cookie_t), KM_SLEEP);
2446 
2447 		status = ldc_mem_dring_bind(vdc->ldc_handle, vdc->ldc_dring_hdl,
2448 		    LDC_SHADOW_MAP|LDC_DIRECT_MAP, LDC_MEM_RW,
2449 		    &vdc->dring_cookie[0],
2450 		    &vdc->dring_cookie_count);
2451 		if (status != 0) {
2452 			DMSG(vdc, 0, "[%d] Failed to bind descriptor ring "
2453 			    "(%lx) to channel (%lx) status=%d\n",
2454 			    vdc->instance, vdc->ldc_dring_hdl,
2455 			    vdc->ldc_handle, status);
2456 			return (status);
2457 		}
2458 		ASSERT(vdc->dring_cookie_count == 1);
2459 		vdc->initialized |= VDC_DRING_BOUND;
2460 	}
2461 
2462 	status = ldc_mem_dring_info(vdc->ldc_dring_hdl, &vdc->dring_mem_info);
2463 	if (status != 0) {
2464 		DMSG(vdc, 0,
2465 		    "[%d] Failed to get info for descriptor ring (%lx)\n",
2466 		    vdc->instance, vdc->ldc_dring_hdl);
2467 		return (status);
2468 	}
2469 
2470 	if ((vdc->initialized & VDC_DRING_LOCAL) == 0) {
2471 		DMSG(vdc, 0, "[%d] local dring\n", vdc->instance);
2472 
2473 		/* Allocate the local copy of this dring */
2474 		vdc->local_dring =
2475 		    kmem_zalloc(vdc->dring_len * sizeof (vdc_local_desc_t),
2476 		    KM_SLEEP);
2477 		vdc->initialized |= VDC_DRING_LOCAL;
2478 	}
2479 
2480 	/*
2481 	 * Mark all DRing entries as free and initialize the private
2482 	 * descriptor's memory handles. If any entry is initialized,
2483 	 * we need to free it later so we set the bit in 'initialized'
2484 	 * at the start.
2485 	 */
2486 	vdc->initialized |= VDC_DRING_ENTRY;
2487 	for (i = 0; i < vdc->dring_len; i++) {
2488 		dep = VDC_GET_DRING_ENTRY_PTR(vdc, i);
2489 		dep->hdr.dstate = VIO_DESC_FREE;
2490 
2491 		status = ldc_mem_alloc_handle(vdc->ldc_handle,
2492 		    &vdc->local_dring[i].desc_mhdl);
2493 		if (status != 0) {
2494 			DMSG(vdc, 0, "![%d] Failed to alloc mem handle for"
2495 			    " descriptor %d", vdc->instance, i);
2496 			return (status);
2497 		}
2498 		vdc->local_dring[i].is_free = B_TRUE;
2499 		vdc->local_dring[i].dep = dep;
2500 	}
2501 
2502 	/* Initialize the starting index */
2503 	vdc->dring_curr_idx = 0;
2504 
2505 	return (status);
2506 }
2507 
2508 /*
2509  * Function:
2510  *	vdc_destroy_descriptor_ring()
2511  *
2512  * Description:
2513  *
2514  * Arguments:
2515  *	vdc	- soft state pointer for this instance of the device driver.
2516  *
2517  * Return Code:
2518  *	None
2519  */
2520 static void
2521 vdc_destroy_descriptor_ring(vdc_t *vdc)
2522 {
2523 	vdc_local_desc_t	*ldep = NULL;	/* Local Dring Entry Pointer */
2524 	ldc_mem_handle_t	mhdl = NULL;
2525 	ldc_mem_info_t		minfo;
2526 	int			status = -1;
2527 	int			i;	/* loop */
2528 
2529 	ASSERT(vdc != NULL);
2530 	ASSERT(mutex_owned(&vdc->lock));
2531 
2532 	DMSG(vdc, 0, "[%d] Entered\n", vdc->instance);
2533 
2534 	if (vdc->initialized & VDC_DRING_ENTRY) {
2535 		DMSG(vdc, 0,
2536 		    "[%d] Removing Local DRing entries\n", vdc->instance);
2537 		for (i = 0; i < vdc->dring_len; i++) {
2538 			ldep = &vdc->local_dring[i];
2539 			mhdl = ldep->desc_mhdl;
2540 
2541 			if (mhdl == NULL)
2542 				continue;
2543 
2544 			if ((status = ldc_mem_info(mhdl, &minfo)) != 0) {
2545 				DMSG(vdc, 0,
2546 				    "ldc_mem_info returned an error: %d\n",
2547 				    status);
2548 
2549 				/*
2550 				 * This must mean that the mem handle
2551 				 * is not valid. Clear it out so that
2552 				 * no one tries to use it.
2553 				 */
2554 				ldep->desc_mhdl = NULL;
2555 				continue;
2556 			}
2557 
2558 			if (minfo.status == LDC_BOUND) {
2559 				(void) ldc_mem_unbind_handle(mhdl);
2560 			}
2561 
2562 			(void) ldc_mem_free_handle(mhdl);
2563 
2564 			ldep->desc_mhdl = NULL;
2565 		}
2566 		vdc->initialized &= ~VDC_DRING_ENTRY;
2567 	}
2568 
2569 	if (vdc->initialized & VDC_DRING_LOCAL) {
2570 		DMSG(vdc, 0, "[%d] Freeing Local DRing\n", vdc->instance);
2571 		kmem_free(vdc->local_dring,
2572 		    vdc->dring_len * sizeof (vdc_local_desc_t));
2573 		vdc->initialized &= ~VDC_DRING_LOCAL;
2574 	}
2575 
2576 	if (vdc->initialized & VDC_DRING_BOUND) {
2577 		DMSG(vdc, 0, "[%d] Unbinding DRing\n", vdc->instance);
2578 		status = ldc_mem_dring_unbind(vdc->ldc_dring_hdl);
2579 		if (status == 0) {
2580 			vdc->initialized &= ~VDC_DRING_BOUND;
2581 		} else {
2582 			DMSG(vdc, 0, "[%d] Error %d unbinding DRing %lx",
2583 			    vdc->instance, status, vdc->ldc_dring_hdl);
2584 		}
2585 		kmem_free(vdc->dring_cookie, sizeof (ldc_mem_cookie_t));
2586 	}
2587 
2588 	if (vdc->initialized & VDC_DRING_INIT) {
2589 		DMSG(vdc, 0, "[%d] Destroying DRing\n", vdc->instance);
2590 		status = ldc_mem_dring_destroy(vdc->ldc_dring_hdl);
2591 		if (status == 0) {
2592 			vdc->ldc_dring_hdl = NULL;
2593 			bzero(&vdc->dring_mem_info, sizeof (ldc_mem_info_t));
2594 			vdc->initialized &= ~VDC_DRING_INIT;
2595 		} else {
2596 			DMSG(vdc, 0, "[%d] Error %d destroying DRing (%lx)",
2597 			    vdc->instance, status, vdc->ldc_dring_hdl);
2598 		}
2599 	}
2600 }
2601 
2602 /*
2603  * Function:
2604  *	vdc_map_to_shared_ring()
2605  *
2606  * Description:
2607  *	Copy contents of the local descriptor to the shared
2608  *	memory descriptor.
2609  *
2610  * Arguments:
2611  *	vdcp	- soft state pointer for this instance of the device driver.
2612  *	idx	- descriptor ring index
2613  *
2614  * Return Code:
2615  *	None
2616  */
2617 static int
2618 vdc_map_to_shared_dring(vdc_t *vdcp, int idx)
2619 {
2620 	vdc_local_desc_t	*ldep;
2621 	vd_dring_entry_t	*dep;
2622 	int			rv;
2623 
2624 	ldep = &(vdcp->local_dring[idx]);
2625 
2626 	/* for now leave in the old pop_mem_hdl stuff */
2627 	if (ldep->nbytes > 0) {
2628 		rv = vdc_populate_mem_hdl(vdcp, ldep);
2629 		if (rv) {
2630 			DMSG(vdcp, 0, "[%d] Cannot populate mem handle\n",
2631 			    vdcp->instance);
2632 			return (rv);
2633 		}
2634 	}
2635 
2636 	/*
2637 	 * fill in the data details into the DRing
2638 	 */
2639 	dep = ldep->dep;
2640 	ASSERT(dep != NULL);
2641 
2642 	dep->payload.req_id = VDC_GET_NEXT_REQ_ID(vdcp);
2643 	dep->payload.operation = ldep->operation;
2644 	dep->payload.addr = ldep->offset;
2645 	dep->payload.nbytes = ldep->nbytes;
2646 	dep->payload.status = (uint32_t)-1;	/* vds will set valid value */
2647 	dep->payload.slice = ldep->slice;
2648 	dep->hdr.dstate = VIO_DESC_READY;
2649 	dep->hdr.ack = 1;		/* request an ACK for every message */
2650 
2651 	return (0);
2652 }
2653 
2654 /*
2655  * Function:
2656  *	vdc_send_request
2657  *
2658  * Description:
2659  *	This routine writes the data to be transmitted to vds into the
2660  *	descriptor, notifies vds that the ring has been updated and
2661  *	then waits for the request to be processed.
2662  *
2663  * Arguments:
2664  *	vdcp	  - the soft state pointer
2665  *	operation - operation we want vds to perform (VD_OP_XXX)
2666  *	addr	  - address of data buf to be read/written.
2667  *	nbytes	  - number of bytes to read/write
2668  *	slice	  - the disk slice this request is for
2669  *	offset	  - relative disk offset
2670  *	cb_type   - type of call - STRATEGY or SYNC
2671  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
2672  *			. mode for ioctl(9e)
2673  *			. LP64 diskaddr_t (block I/O)
2674  *	dir	  - direction of operation (READ/WRITE/BOTH)
2675  *
2676  * Return Codes:
2677  *	0
2678  *	ENXIO
2679  */
2680 static int
2681 vdc_send_request(vdc_t *vdcp, int operation, caddr_t addr,
2682     size_t nbytes, int slice, diskaddr_t offset, int cb_type,
2683     void *cb_arg, vio_desc_direction_t dir)
2684 {
2685 	ASSERT(vdcp != NULL);
2686 	ASSERT(slice == VD_SLICE_NONE || slice < V_NUMPAR);
2687 
2688 	mutex_enter(&vdcp->lock);
2689 
2690 	do {
2691 		while (vdcp->state != VDC_STATE_RUNNING) {
2692 
2693 			/* return error if detaching */
2694 			if (vdcp->state == VDC_STATE_DETACH) {
2695 				mutex_exit(&vdcp->lock);
2696 				return (ENXIO);
2697 			}
2698 
2699 			/* fail request if connection timeout is reached */
2700 			if (vdcp->ctimeout_reached) {
2701 				mutex_exit(&vdcp->lock);
2702 				return (EIO);
2703 			}
2704 
2705 			/*
2706 			 * If we are panicking and the disk is not ready then
2707 			 * we can't send any request because we can't complete
2708 			 * the handshake now.
2709 			 */
2710 			if (ddi_in_panic()) {
2711 				mutex_exit(&vdcp->lock);
2712 				return (EIO);
2713 			}
2714 
2715 			cv_wait(&vdcp->running_cv, &vdcp->lock);
2716 		}
2717 
2718 	} while (vdc_populate_descriptor(vdcp, operation, addr,
2719 	    nbytes, slice, offset, cb_type, cb_arg, dir));
2720 
2721 	mutex_exit(&vdcp->lock);
2722 	return (0);
2723 }
2724 
2725 
2726 /*
2727  * Function:
2728  *	vdc_populate_descriptor
2729  *
2730  * Description:
2731  *	This routine writes the data to be transmitted to vds into the
2732  *	descriptor, notifies vds that the ring has been updated and
2733  *	then waits for the request to be processed.
2734  *
2735  * Arguments:
2736  *	vdcp	  - the soft state pointer
2737  *	operation - operation we want vds to perform (VD_OP_XXX)
2738  *	addr	  - address of data buf to be read/written.
2739  *	nbytes	  - number of bytes to read/write
2740  *	slice	  - the disk slice this request is for
2741  *	offset	  - relative disk offset
2742  *	cb_type   - type of call - STRATEGY or SYNC
2743  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
2744  *			. mode for ioctl(9e)
2745  *			. LP64 diskaddr_t (block I/O)
2746  *	dir	  - direction of operation (READ/WRITE/BOTH)
2747  *
2748  * Return Codes:
2749  *	0
2750  *	EAGAIN
2751  *	ECONNRESET
2752  *	ENXIO
2753  */
2754 static int
2755 vdc_populate_descriptor(vdc_t *vdcp, int operation, caddr_t addr,
2756     size_t nbytes, int slice, diskaddr_t offset, int cb_type,
2757     void *cb_arg, vio_desc_direction_t dir)
2758 {
2759 	vdc_local_desc_t	*local_dep = NULL; /* Local Dring Pointer */
2760 	int			idx;		/* Index of DRing entry used */
2761 	int			next_idx;
2762 	vio_dring_msg_t		dmsg;
2763 	size_t			msglen;
2764 	int			rv;
2765 
2766 	ASSERT(MUTEX_HELD(&vdcp->lock));
2767 	vdcp->threads_pending++;
2768 loop:
2769 	DMSG(vdcp, 2, ": dring_curr_idx = %d\n", vdcp->dring_curr_idx);
2770 
2771 	/* Get next available D-Ring entry */
2772 	idx = vdcp->dring_curr_idx;
2773 	local_dep = &(vdcp->local_dring[idx]);
2774 
2775 	if (!local_dep->is_free) {
2776 		DMSG(vdcp, 2, "[%d]: dring full - waiting for space\n",
2777 		    vdcp->instance);
2778 		cv_wait(&vdcp->dring_free_cv, &vdcp->lock);
2779 		if (vdcp->state == VDC_STATE_RUNNING ||
2780 		    vdcp->state == VDC_STATE_HANDLE_PENDING) {
2781 			goto loop;
2782 		}
2783 		vdcp->threads_pending--;
2784 		return (ECONNRESET);
2785 	}
2786 
2787 	next_idx = idx + 1;
2788 	if (next_idx >= vdcp->dring_len)
2789 		next_idx = 0;
2790 	vdcp->dring_curr_idx = next_idx;
2791 
2792 	ASSERT(local_dep->is_free);
2793 
2794 	local_dep->operation = operation;
2795 	local_dep->addr = addr;
2796 	local_dep->nbytes = nbytes;
2797 	local_dep->slice = slice;
2798 	local_dep->offset = offset;
2799 	local_dep->cb_type = cb_type;
2800 	local_dep->cb_arg = cb_arg;
2801 	local_dep->dir = dir;
2802 
2803 	local_dep->is_free = B_FALSE;
2804 
2805 	rv = vdc_map_to_shared_dring(vdcp, idx);
2806 	if (rv) {
2807 		DMSG(vdcp, 0, "[%d]: cannot bind memory - waiting ..\n",
2808 		    vdcp->instance);
2809 		/* free the descriptor */
2810 		local_dep->is_free = B_TRUE;
2811 		vdcp->dring_curr_idx = idx;
2812 		cv_wait(&vdcp->membind_cv, &vdcp->lock);
2813 		if (vdcp->state == VDC_STATE_RUNNING ||
2814 		    vdcp->state == VDC_STATE_HANDLE_PENDING) {
2815 			goto loop;
2816 		}
2817 		vdcp->threads_pending--;
2818 		return (ECONNRESET);
2819 	}
2820 
2821 	/*
2822 	 * Send a msg with the DRing details to vds
2823 	 */
2824 	VIO_INIT_DRING_DATA_TAG(dmsg);
2825 	VDC_INIT_DRING_DATA_MSG_IDS(dmsg, vdcp);
2826 	dmsg.dring_ident = vdcp->dring_ident;
2827 	dmsg.start_idx = idx;
2828 	dmsg.end_idx = idx;
2829 	vdcp->seq_num++;
2830 
2831 	DTRACE_IO2(send, vio_dring_msg_t *, &dmsg, vdc_t *, vdcp);
2832 
2833 	DMSG(vdcp, 2, "ident=0x%lx, st=%u, end=%u, seq=%ld\n",
2834 	    vdcp->dring_ident, dmsg.start_idx, dmsg.end_idx, dmsg.seq_num);
2835 
2836 	/*
2837 	 * note we're still holding the lock here to
2838 	 * make sure the message goes out in order !!!...
2839 	 */
2840 	msglen = sizeof (dmsg);
2841 	rv = vdc_send(vdcp, (caddr_t)&dmsg, &msglen);
2842 	switch (rv) {
2843 	case ECONNRESET:
2844 		/*
2845 		 * vdc_send initiates the reset on failure.
2846 		 * Since the transaction has already been put
2847 		 * on the local dring, it will automatically get
2848 		 * retried when the channel is reset. Given that,
2849 		 * it is ok to just return success even though the
2850 		 * send failed.
2851 		 */
2852 		rv = 0;
2853 		break;
2854 
2855 	case 0: /* EOK */
2856 		DMSG(vdcp, 1, "sent via LDC: rv=%d\n", rv);
2857 		break;
2858 
2859 	default:
2860 		goto cleanup_and_exit;
2861 	}
2862 
2863 	vdcp->threads_pending--;
2864 	return (rv);
2865 
2866 cleanup_and_exit:
2867 	DMSG(vdcp, 0, "unexpected error, rv=%d\n", rv);
2868 	return (ENXIO);
2869 }
2870 
2871 /*
2872  * Function:
2873  *	vdc_do_sync_op
2874  *
2875  * Description:
2876  * 	Wrapper around vdc_populate_descriptor that blocks until the
2877  * 	response to the message is available.
2878  *
2879  * Arguments:
2880  *	vdcp	  - the soft state pointer
2881  *	operation - operation we want vds to perform (VD_OP_XXX)
2882  *	addr	  - address of data buf to be read/written.
2883  *	nbytes	  - number of bytes to read/write
2884  *	slice	  - the disk slice this request is for
2885  *	offset	  - relative disk offset
2886  *	cb_type   - type of call - STRATEGY or SYNC
2887  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
2888  *			. mode for ioctl(9e)
2889  *			. LP64 diskaddr_t (block I/O)
2890  *	dir	  - direction of operation (READ/WRITE/BOTH)
2891  *	rconflict - check for reservation conflict in case of failure
2892  *
2893  * rconflict should be set to B_TRUE by most callers. Callers invoking the
2894  * VD_OP_SCSICMD operation can set rconflict to B_FALSE if they check the
2895  * result of a successful operation with vd_scsi_status().
2896  *
2897  * Return Codes:
2898  *	0
2899  *	EAGAIN
2900  *	EFAULT
2901  *	ENXIO
2902  *	EIO
2903  */
2904 static int
2905 vdc_do_sync_op(vdc_t *vdcp, int operation, caddr_t addr, size_t nbytes,
2906     int slice, diskaddr_t offset, int cb_type, void *cb_arg,
2907     vio_desc_direction_t dir, boolean_t rconflict)
2908 {
2909 	int status;
2910 	vdc_io_t *vio;
2911 	boolean_t check_resv_conflict = B_FALSE;
2912 
2913 	ASSERT(cb_type == CB_SYNC);
2914 
2915 	/*
2916 	 * Grab the lock, if blocked wait until the server
2917 	 * response causes us to wake up again.
2918 	 */
2919 	mutex_enter(&vdcp->lock);
2920 	vdcp->sync_op_cnt++;
2921 	while (vdcp->sync_op_blocked && vdcp->state != VDC_STATE_DETACH)
2922 		cv_wait(&vdcp->sync_blocked_cv, &vdcp->lock);
2923 
2924 	if (vdcp->state == VDC_STATE_DETACH) {
2925 		cv_broadcast(&vdcp->sync_blocked_cv);
2926 		vdcp->sync_op_cnt--;
2927 		mutex_exit(&vdcp->lock);
2928 		return (ENXIO);
2929 	}
2930 
2931 	/* now block anyone other thread entering after us */
2932 	vdcp->sync_op_blocked = B_TRUE;
2933 	vdcp->sync_op_pending = B_TRUE;
2934 	mutex_exit(&vdcp->lock);
2935 
2936 	status = vdc_send_request(vdcp, operation, addr,
2937 	    nbytes, slice, offset, cb_type, cb_arg, dir);
2938 
2939 	mutex_enter(&vdcp->lock);
2940 
2941 	if (status != 0) {
2942 		vdcp->sync_op_pending = B_FALSE;
2943 	} else {
2944 		/*
2945 		 * block until our transaction completes.
2946 		 * Also anyone else waiting also gets to go next.
2947 		 */
2948 		while (vdcp->sync_op_pending && vdcp->state != VDC_STATE_DETACH)
2949 			cv_wait(&vdcp->sync_pending_cv, &vdcp->lock);
2950 
2951 		DMSG(vdcp, 2, ": operation returned %d\n",
2952 		    vdcp->sync_op_status);
2953 		if (vdcp->state == VDC_STATE_DETACH) {
2954 			vdcp->sync_op_pending = B_FALSE;
2955 			status = ENXIO;
2956 		} else {
2957 			status = vdcp->sync_op_status;
2958 			if (status != 0 && vdcp->failfast_interval != 0) {
2959 				/*
2960 				 * Operation has failed and failfast is enabled.
2961 				 * We need to check if the failure is due to a
2962 				 * reservation conflict if this was requested.
2963 				 */
2964 				check_resv_conflict = rconflict;
2965 			}
2966 
2967 		}
2968 	}
2969 
2970 	vdcp->sync_op_status = 0;
2971 	vdcp->sync_op_blocked = B_FALSE;
2972 	vdcp->sync_op_cnt--;
2973 
2974 	/* signal the next waiting thread */
2975 	cv_signal(&vdcp->sync_blocked_cv);
2976 
2977 	/*
2978 	 * We have to check for reservation conflict after unblocking sync
2979 	 * operations because some sync operations will be used to do this
2980 	 * check.
2981 	 */
2982 	if (check_resv_conflict) {
2983 		vio = vdc_failfast_io_queue(vdcp, NULL);
2984 		while (vio->vio_qtime != 0)
2985 			cv_wait(&vdcp->failfast_io_cv, &vdcp->lock);
2986 		kmem_free(vio, sizeof (vdc_io_t));
2987 	}
2988 
2989 	mutex_exit(&vdcp->lock);
2990 
2991 	return (status);
2992 }
2993 
2994 
2995 /*
2996  * Function:
2997  *	vdc_drain_response()
2998  *
2999  * Description:
3000  * 	When a guest is panicking, the completion of requests needs to be
3001  * 	handled differently because interrupts are disabled and vdc
3002  * 	will not get messages. We have to poll for the messages instead.
3003  *
3004  * Arguments:
3005  *	vdc	- soft state pointer for this instance of the device driver.
3006  *
3007  * Return Code:
3008  *	0	- Success
3009  */
3010 static int
3011 vdc_drain_response(vdc_t *vdc)
3012 {
3013 	int 			rv, idx, retries;
3014 	size_t			msglen;
3015 	vdc_local_desc_t 	*ldep = NULL;	/* Local Dring Entry Pointer */
3016 	vio_dring_msg_t		dmsg;
3017 
3018 	mutex_enter(&vdc->lock);
3019 
3020 	retries = 0;
3021 	for (;;) {
3022 		msglen = sizeof (dmsg);
3023 		rv = ldc_read(vdc->ldc_handle, (caddr_t)&dmsg, &msglen);
3024 		if (rv) {
3025 			rv = EINVAL;
3026 			break;
3027 		}
3028 
3029 		/*
3030 		 * if there are no packets wait and check again
3031 		 */
3032 		if ((rv == 0) && (msglen == 0)) {
3033 			if (retries++ > vdc_dump_retries) {
3034 				rv = EAGAIN;
3035 				break;
3036 			}
3037 
3038 			drv_usecwait(vdc_usec_timeout_dump);
3039 			continue;
3040 		}
3041 
3042 		/*
3043 		 * Ignore all messages that are not ACKs/NACKs to
3044 		 * DRing requests.
3045 		 */
3046 		if ((dmsg.tag.vio_msgtype != VIO_TYPE_DATA) ||
3047 		    (dmsg.tag.vio_subtype_env != VIO_DRING_DATA)) {
3048 			DMSG(vdc, 0, "discard pkt: type=%d sub=%d env=%d\n",
3049 			    dmsg.tag.vio_msgtype,
3050 			    dmsg.tag.vio_subtype,
3051 			    dmsg.tag.vio_subtype_env);
3052 			continue;
3053 		}
3054 
3055 		/*
3056 		 * set the appropriate return value for the current request.
3057 		 */
3058 		switch (dmsg.tag.vio_subtype) {
3059 		case VIO_SUBTYPE_ACK:
3060 			rv = 0;
3061 			break;
3062 		case VIO_SUBTYPE_NACK:
3063 			rv = EAGAIN;
3064 			break;
3065 		default:
3066 			continue;
3067 		}
3068 
3069 		idx = dmsg.start_idx;
3070 		if (idx >= vdc->dring_len) {
3071 			DMSG(vdc, 0, "[%d] Bogus ack data : start %d\n",
3072 			    vdc->instance, idx);
3073 			continue;
3074 		}
3075 		ldep = &vdc->local_dring[idx];
3076 		if (ldep->dep->hdr.dstate != VIO_DESC_DONE) {
3077 			DMSG(vdc, 0, "[%d] Entry @ %d - state !DONE %d\n",
3078 			    vdc->instance, idx, ldep->dep->hdr.dstate);
3079 			continue;
3080 		}
3081 
3082 		DMSG(vdc, 1, "[%d] Depopulating idx=%d state=%d\n",
3083 		    vdc->instance, idx, ldep->dep->hdr.dstate);
3084 		rv = vdc_depopulate_descriptor(vdc, idx);
3085 		if (rv) {
3086 			DMSG(vdc, 0,
3087 			    "[%d] Entry @ %d - depopulate failed ..\n",
3088 			    vdc->instance, idx);
3089 		}
3090 
3091 		/* if this is the last descriptor - break out of loop */
3092 		if ((idx + 1) % vdc->dring_len == vdc->dring_curr_idx)
3093 			break;
3094 	}
3095 
3096 	mutex_exit(&vdc->lock);
3097 	DMSG(vdc, 0, "End idx=%d\n", idx);
3098 
3099 	return (rv);
3100 }
3101 
3102 
3103 /*
3104  * Function:
3105  *	vdc_depopulate_descriptor()
3106  *
3107  * Description:
3108  *
3109  * Arguments:
3110  *	vdc	- soft state pointer for this instance of the device driver.
3111  *	idx	- Index of the Descriptor Ring entry being modified
3112  *
3113  * Return Code:
3114  *	0	- Success
3115  */
3116 static int
3117 vdc_depopulate_descriptor(vdc_t *vdc, uint_t idx)
3118 {
3119 	vd_dring_entry_t *dep = NULL;		/* Dring Entry Pointer */
3120 	vdc_local_desc_t *ldep = NULL;		/* Local Dring Entry Pointer */
3121 	int		status = ENXIO;
3122 	int		rv = 0;
3123 
3124 	ASSERT(vdc != NULL);
3125 	ASSERT(idx < vdc->dring_len);
3126 	ldep = &vdc->local_dring[idx];
3127 	ASSERT(ldep != NULL);
3128 	ASSERT(MUTEX_HELD(&vdc->lock));
3129 
3130 	DMSG(vdc, 2, ": idx = %d\n", idx);
3131 	dep = ldep->dep;
3132 	ASSERT(dep != NULL);
3133 	ASSERT((dep->hdr.dstate == VIO_DESC_DONE) ||
3134 	    (dep->payload.status == ECANCELED));
3135 
3136 	VDC_MARK_DRING_ENTRY_FREE(vdc, idx);
3137 
3138 	ldep->is_free = B_TRUE;
3139 	status = dep->payload.status;
3140 	DMSG(vdc, 2, ": is_free = %d : status = %d\n", ldep->is_free, status);
3141 
3142 	/*
3143 	 * If no buffers were used to transfer information to the server when
3144 	 * populating the descriptor then no memory handles need to be unbound
3145 	 * and we can return now.
3146 	 */
3147 	if (ldep->nbytes == 0) {
3148 		cv_signal(&vdc->dring_free_cv);
3149 		return (status);
3150 	}
3151 
3152 	/*
3153 	 * If the upper layer passed in a misaligned address we copied the
3154 	 * data into an aligned buffer before sending it to LDC - we now
3155 	 * copy it back to the original buffer.
3156 	 */
3157 	if (ldep->align_addr) {
3158 		ASSERT(ldep->addr != NULL);
3159 
3160 		if (dep->payload.nbytes > 0)
3161 			bcopy(ldep->align_addr, ldep->addr,
3162 			    dep->payload.nbytes);
3163 		kmem_free(ldep->align_addr,
3164 		    sizeof (caddr_t) * P2ROUNDUP(ldep->nbytes, 8));
3165 		ldep->align_addr = NULL;
3166 	}
3167 
3168 	rv = ldc_mem_unbind_handle(ldep->desc_mhdl);
3169 	if (rv != 0) {
3170 		DMSG(vdc, 0, "?[%d] unbind mhdl 0x%lx @ idx %d failed (%d)",
3171 		    vdc->instance, ldep->desc_mhdl, idx, rv);
3172 		/*
3173 		 * The error returned by the vDisk server is more informative
3174 		 * and thus has a higher priority but if it isn't set we ensure
3175 		 * that this function returns an error.
3176 		 */
3177 		if (status == 0)
3178 			status = EINVAL;
3179 	}
3180 
3181 	cv_signal(&vdc->membind_cv);
3182 	cv_signal(&vdc->dring_free_cv);
3183 
3184 	return (status);
3185 }
3186 
3187 /*
3188  * Function:
3189  *	vdc_populate_mem_hdl()
3190  *
3191  * Description:
3192  *
3193  * Arguments:
3194  *	vdc	- soft state pointer for this instance of the device driver.
3195  *	idx	- Index of the Descriptor Ring entry being modified
3196  *	addr	- virtual address being mapped in
3197  *	nybtes	- number of bytes in 'addr'
3198  *	operation - the vDisk operation being performed (VD_OP_xxx)
3199  *
3200  * Return Code:
3201  *	0	- Success
3202  */
3203 static int
3204 vdc_populate_mem_hdl(vdc_t *vdcp, vdc_local_desc_t *ldep)
3205 {
3206 	vd_dring_entry_t	*dep = NULL;
3207 	ldc_mem_handle_t	mhdl;
3208 	caddr_t			vaddr;
3209 	size_t			nbytes;
3210 	uint8_t			perm = LDC_MEM_RW;
3211 	uint8_t			maptype;
3212 	int			rv = 0;
3213 	int			i;
3214 
3215 	ASSERT(vdcp != NULL);
3216 
3217 	dep = ldep->dep;
3218 	mhdl = ldep->desc_mhdl;
3219 
3220 	switch (ldep->dir) {
3221 	case VIO_read_dir:
3222 		perm = LDC_MEM_W;
3223 		break;
3224 
3225 	case VIO_write_dir:
3226 		perm = LDC_MEM_R;
3227 		break;
3228 
3229 	case VIO_both_dir:
3230 		perm = LDC_MEM_RW;
3231 		break;
3232 
3233 	default:
3234 		ASSERT(0);	/* catch bad programming in vdc */
3235 	}
3236 
3237 	/*
3238 	 * LDC expects any addresses passed in to be 8-byte aligned. We need
3239 	 * to copy the contents of any misaligned buffers to a newly allocated
3240 	 * buffer and bind it instead (and copy the the contents back to the
3241 	 * original buffer passed in when depopulating the descriptor)
3242 	 */
3243 	vaddr = ldep->addr;
3244 	nbytes = ldep->nbytes;
3245 	if (((uint64_t)vaddr & 0x7) != 0) {
3246 		ASSERT(ldep->align_addr == NULL);
3247 		ldep->align_addr =
3248 		    kmem_alloc(sizeof (caddr_t) *
3249 		    P2ROUNDUP(nbytes, 8), KM_SLEEP);
3250 		DMSG(vdcp, 0, "[%d] Misaligned address %p reallocating "
3251 		    "(buf=%p nb=%ld op=%d)\n",
3252 		    vdcp->instance, (void *)vaddr, (void *)ldep->align_addr,
3253 		    nbytes, ldep->operation);
3254 		if (perm != LDC_MEM_W)
3255 			bcopy(vaddr, ldep->align_addr, nbytes);
3256 		vaddr = ldep->align_addr;
3257 	}
3258 
3259 	maptype = LDC_IO_MAP|LDC_SHADOW_MAP|LDC_DIRECT_MAP;
3260 	rv = ldc_mem_bind_handle(mhdl, vaddr, P2ROUNDUP(nbytes, 8),
3261 	    maptype, perm, &dep->payload.cookie[0], &dep->payload.ncookies);
3262 	DMSG(vdcp, 2, "[%d] bound mem handle; ncookies=%d\n",
3263 	    vdcp->instance, dep->payload.ncookies);
3264 	if (rv != 0) {
3265 		DMSG(vdcp, 0, "[%d] Failed to bind LDC memory handle "
3266 		    "(mhdl=%p, buf=%p, err=%d)\n",
3267 		    vdcp->instance, (void *)mhdl, (void *)vaddr, rv);
3268 		if (ldep->align_addr) {
3269 			kmem_free(ldep->align_addr,
3270 			    sizeof (caddr_t) * P2ROUNDUP(nbytes, 8));
3271 			ldep->align_addr = NULL;
3272 		}
3273 		return (EAGAIN);
3274 	}
3275 
3276 	/*
3277 	 * Get the other cookies (if any).
3278 	 */
3279 	for (i = 1; i < dep->payload.ncookies; i++) {
3280 		rv = ldc_mem_nextcookie(mhdl, &dep->payload.cookie[i]);
3281 		if (rv != 0) {
3282 			(void) ldc_mem_unbind_handle(mhdl);
3283 			DMSG(vdcp, 0, "?[%d] Failed to get next cookie "
3284 			    "(mhdl=%lx cnum=%d), err=%d",
3285 			    vdcp->instance, mhdl, i, rv);
3286 			if (ldep->align_addr) {
3287 				kmem_free(ldep->align_addr,
3288 				    sizeof (caddr_t) * ldep->nbytes);
3289 				ldep->align_addr = NULL;
3290 			}
3291 			return (EAGAIN);
3292 		}
3293 	}
3294 
3295 	return (rv);
3296 }
3297 
3298 /*
3299  * Interrupt handlers for messages from LDC
3300  */
3301 
3302 /*
3303  * Function:
3304  *	vdc_handle_cb()
3305  *
3306  * Description:
3307  *
3308  * Arguments:
3309  *	event	- Type of event (LDC_EVT_xxx) that triggered the callback
3310  *	arg	- soft state pointer for this instance of the device driver.
3311  *
3312  * Return Code:
3313  *	0	- Success
3314  */
3315 static uint_t
3316 vdc_handle_cb(uint64_t event, caddr_t arg)
3317 {
3318 	ldc_status_t	ldc_state;
3319 	int		rv = 0;
3320 
3321 	vdc_t	*vdc = (vdc_t *)(void *)arg;
3322 
3323 	ASSERT(vdc != NULL);
3324 
3325 	DMSG(vdc, 1, "evt=%lx seqID=%ld\n", event, vdc->seq_num);
3326 
3327 	/*
3328 	 * Depending on the type of event that triggered this callback,
3329 	 * we modify the handshake state or read the data.
3330 	 *
3331 	 * NOTE: not done as a switch() as event could be triggered by
3332 	 * a state change and a read request. Also the ordering	of the
3333 	 * check for the event types is deliberate.
3334 	 */
3335 	if (event & LDC_EVT_UP) {
3336 		DMSG(vdc, 0, "[%d] Received LDC_EVT_UP\n", vdc->instance);
3337 
3338 		mutex_enter(&vdc->lock);
3339 
3340 		/* get LDC state */
3341 		rv = ldc_status(vdc->ldc_handle, &ldc_state);
3342 		if (rv != 0) {
3343 			DMSG(vdc, 0, "[%d] Couldn't get LDC status %d",
3344 			    vdc->instance, rv);
3345 			return (LDC_SUCCESS);
3346 		}
3347 		if (vdc->ldc_state != LDC_UP && ldc_state == LDC_UP) {
3348 			/*
3349 			 * Reset the transaction sequence numbers when
3350 			 * LDC comes up. We then kick off the handshake
3351 			 * negotiation with the vDisk server.
3352 			 */
3353 			vdc->seq_num = 1;
3354 			vdc->seq_num_reply = 0;
3355 			vdc->ldc_state = ldc_state;
3356 			cv_signal(&vdc->initwait_cv);
3357 		}
3358 
3359 		mutex_exit(&vdc->lock);
3360 	}
3361 
3362 	if (event & LDC_EVT_READ) {
3363 		DMSG(vdc, 1, "[%d] Received LDC_EVT_READ\n", vdc->instance);
3364 		mutex_enter(&vdc->read_lock);
3365 		cv_signal(&vdc->read_cv);
3366 		vdc->read_state = VDC_READ_PENDING;
3367 		mutex_exit(&vdc->read_lock);
3368 
3369 		/* that's all we have to do - no need to handle DOWN/RESET */
3370 		return (LDC_SUCCESS);
3371 	}
3372 
3373 	if (event & (LDC_EVT_RESET|LDC_EVT_DOWN)) {
3374 
3375 		DMSG(vdc, 0, "[%d] Received LDC RESET event\n", vdc->instance);
3376 
3377 		mutex_enter(&vdc->lock);
3378 		/*
3379 		 * Need to wake up any readers so they will
3380 		 * detect that a reset has occurred.
3381 		 */
3382 		mutex_enter(&vdc->read_lock);
3383 		if ((vdc->read_state == VDC_READ_WAITING) ||
3384 		    (vdc->read_state == VDC_READ_RESET))
3385 			cv_signal(&vdc->read_cv);
3386 		vdc->read_state = VDC_READ_RESET;
3387 		mutex_exit(&vdc->read_lock);
3388 
3389 		/* wake up any threads waiting for connection to come up */
3390 		if (vdc->state == VDC_STATE_INIT_WAITING) {
3391 			vdc->state = VDC_STATE_RESETTING;
3392 			cv_signal(&vdc->initwait_cv);
3393 		}
3394 
3395 		mutex_exit(&vdc->lock);
3396 	}
3397 
3398 	if (event & ~(LDC_EVT_UP | LDC_EVT_RESET | LDC_EVT_DOWN | LDC_EVT_READ))
3399 		DMSG(vdc, 0, "![%d] Unexpected LDC event (%lx) received",
3400 		    vdc->instance, event);
3401 
3402 	return (LDC_SUCCESS);
3403 }
3404 
3405 /*
3406  * Function:
3407  *	vdc_wait_for_response()
3408  *
3409  * Description:
3410  *	Block waiting for a response from the server. If there is
3411  *	no data the thread block on the read_cv that is signalled
3412  *	by the callback when an EVT_READ occurs.
3413  *
3414  * Arguments:
3415  *	vdcp	- soft state pointer for this instance of the device driver.
3416  *
3417  * Return Code:
3418  *	0	- Success
3419  */
3420 static int
3421 vdc_wait_for_response(vdc_t *vdcp, vio_msg_t *msgp)
3422 {
3423 	size_t		nbytes = sizeof (*msgp);
3424 	int		status;
3425 
3426 	ASSERT(vdcp != NULL);
3427 
3428 	DMSG(vdcp, 1, "[%d] Entered\n", vdcp->instance);
3429 
3430 	status = vdc_recv(vdcp, msgp, &nbytes);
3431 	DMSG(vdcp, 3, "vdc_read() done.. status=0x%x size=0x%x\n",
3432 	    status, (int)nbytes);
3433 	if (status) {
3434 		DMSG(vdcp, 0, "?[%d] Error %d reading LDC msg\n",
3435 		    vdcp->instance, status);
3436 		return (status);
3437 	}
3438 
3439 	if (nbytes < sizeof (vio_msg_tag_t)) {
3440 		DMSG(vdcp, 0, "?[%d] Expect %lu bytes; recv'd %lu\n",
3441 		    vdcp->instance, sizeof (vio_msg_tag_t), nbytes);
3442 		return (ENOMSG);
3443 	}
3444 
3445 	DMSG(vdcp, 2, "[%d] (%x/%x/%x)\n", vdcp->instance,
3446 	    msgp->tag.vio_msgtype,
3447 	    msgp->tag.vio_subtype,
3448 	    msgp->tag.vio_subtype_env);
3449 
3450 	/*
3451 	 * Verify the Session ID of the message
3452 	 *
3453 	 * Every message after the Version has been negotiated should
3454 	 * have the correct session ID set.
3455 	 */
3456 	if ((msgp->tag.vio_sid != vdcp->session_id) &&
3457 	    (msgp->tag.vio_subtype_env != VIO_VER_INFO)) {
3458 		DMSG(vdcp, 0, "[%d] Invalid SID: received 0x%x, "
3459 		    "expected 0x%lx [seq num %lx @ %d]",
3460 		    vdcp->instance, msgp->tag.vio_sid,
3461 		    vdcp->session_id,
3462 		    ((vio_dring_msg_t *)msgp)->seq_num,
3463 		    ((vio_dring_msg_t *)msgp)->start_idx);
3464 		return (ENOMSG);
3465 	}
3466 	return (0);
3467 }
3468 
3469 
3470 /*
3471  * Function:
3472  *	vdc_resubmit_backup_dring()
3473  *
3474  * Description:
3475  *	Resubmit each descriptor in the backed up dring to
3476  * 	vDisk server. The Dring was backed up during connection
3477  *	reset.
3478  *
3479  * Arguments:
3480  *	vdcp	- soft state pointer for this instance of the device driver.
3481  *
3482  * Return Code:
3483  *	0	- Success
3484  */
3485 static int
3486 vdc_resubmit_backup_dring(vdc_t *vdcp)
3487 {
3488 	int		count;
3489 	int		b_idx;
3490 	int		rv;
3491 	int		dring_size;
3492 	int		status;
3493 	vio_msg_t	vio_msg;
3494 	vdc_local_desc_t	*curr_ldep;
3495 
3496 	ASSERT(MUTEX_NOT_HELD(&vdcp->lock));
3497 	ASSERT(vdcp->state == VDC_STATE_HANDLE_PENDING);
3498 
3499 	if (vdcp->local_dring_backup == NULL) {
3500 		/* the pending requests have already been processed */
3501 		return (0);
3502 	}
3503 
3504 	DMSG(vdcp, 1, "restoring pending dring entries (len=%d, tail=%d)\n",
3505 	    vdcp->local_dring_backup_len, vdcp->local_dring_backup_tail);
3506 
3507 	/*
3508 	 * Walk the backup copy of the local descriptor ring and
3509 	 * resubmit all the outstanding transactions.
3510 	 */
3511 	b_idx = vdcp->local_dring_backup_tail;
3512 	for (count = 0; count < vdcp->local_dring_backup_len; count++) {
3513 
3514 		curr_ldep = &(vdcp->local_dring_backup[b_idx]);
3515 
3516 		/* only resubmit outstanding transactions */
3517 		if (!curr_ldep->is_free) {
3518 
3519 			DMSG(vdcp, 1, "resubmitting entry idx=%x\n", b_idx);
3520 			mutex_enter(&vdcp->lock);
3521 			rv = vdc_populate_descriptor(vdcp, curr_ldep->operation,
3522 			    curr_ldep->addr, curr_ldep->nbytes,
3523 			    curr_ldep->slice, curr_ldep->offset,
3524 			    curr_ldep->cb_type, curr_ldep->cb_arg,
3525 			    curr_ldep->dir);
3526 			mutex_exit(&vdcp->lock);
3527 			if (rv) {
3528 				DMSG(vdcp, 1, "[%d] cannot resubmit entry %d\n",
3529 				    vdcp->instance, b_idx);
3530 				return (rv);
3531 			}
3532 
3533 			/* Wait for the response message. */
3534 			DMSG(vdcp, 1, "waiting for response to idx=%x\n",
3535 			    b_idx);
3536 			status = vdc_wait_for_response(vdcp, &vio_msg);
3537 			if (status) {
3538 				DMSG(vdcp, 1, "[%d] wait_for_response "
3539 				    "returned err=%d\n", vdcp->instance,
3540 				    status);
3541 				return (status);
3542 			}
3543 
3544 			DMSG(vdcp, 1, "processing msg for idx=%x\n", b_idx);
3545 			status = vdc_process_data_msg(vdcp, &vio_msg);
3546 			if (status) {
3547 				DMSG(vdcp, 1, "[%d] process_data_msg "
3548 				    "returned err=%d\n", vdcp->instance,
3549 				    status);
3550 				return (status);
3551 			}
3552 		}
3553 
3554 		/* get the next element to submit */
3555 		if (++b_idx >= vdcp->local_dring_backup_len)
3556 			b_idx = 0;
3557 	}
3558 
3559 	/* all done - now clear up pending dring copy */
3560 	dring_size = vdcp->local_dring_backup_len *
3561 	    sizeof (vdcp->local_dring_backup[0]);
3562 
3563 	(void) kmem_free(vdcp->local_dring_backup, dring_size);
3564 
3565 	vdcp->local_dring_backup = NULL;
3566 
3567 	return (0);
3568 }
3569 
3570 /*
3571  * Function:
3572  *	vdc_cancel_backup_dring
3573  *
3574  * Description:
3575  *	Cancel each descriptor in the backed up dring to vDisk server.
3576  *	The Dring was backed up during connection reset.
3577  *
3578  * Arguments:
3579  *	vdcp	- soft state pointer for this instance of the device driver.
3580  *
3581  * Return Code:
3582  *	None
3583  */
3584 void
3585 vdc_cancel_backup_ring(vdc_t *vdcp)
3586 {
3587 	vdc_local_desc_t *ldep;
3588 	struct buf 	*bufp;
3589 	int		count;
3590 	int		b_idx;
3591 	int		dring_size;
3592 
3593 	ASSERT(MUTEX_HELD(&vdcp->lock));
3594 	ASSERT(vdcp->state == VDC_STATE_INIT ||
3595 	    vdcp->state == VDC_STATE_INIT_WAITING ||
3596 	    vdcp->state == VDC_STATE_NEGOTIATE ||
3597 	    vdcp->state == VDC_STATE_RESETTING);
3598 
3599 	if (vdcp->local_dring_backup == NULL) {
3600 		/* the pending requests have already been processed */
3601 		return;
3602 	}
3603 
3604 	DMSG(vdcp, 1, "cancelling pending dring entries (len=%d, tail=%d)\n",
3605 	    vdcp->local_dring_backup_len, vdcp->local_dring_backup_tail);
3606 
3607 	/*
3608 	 * Walk the backup copy of the local descriptor ring and
3609 	 * cancel all the outstanding transactions.
3610 	 */
3611 	b_idx = vdcp->local_dring_backup_tail;
3612 	for (count = 0; count < vdcp->local_dring_backup_len; count++) {
3613 
3614 		ldep = &(vdcp->local_dring_backup[b_idx]);
3615 
3616 		/* only cancel outstanding transactions */
3617 		if (!ldep->is_free) {
3618 
3619 			DMSG(vdcp, 1, "cancelling entry idx=%x\n", b_idx);
3620 
3621 			/*
3622 			 * All requests have already been cleared from the
3623 			 * local descriptor ring and the LDC channel has been
3624 			 * reset so we will never get any reply for these
3625 			 * requests. Now we just have to notify threads waiting
3626 			 * for replies that the request has failed.
3627 			 */
3628 			switch (ldep->cb_type) {
3629 			case CB_SYNC:
3630 				ASSERT(vdcp->sync_op_pending);
3631 				vdcp->sync_op_status = EIO;
3632 				vdcp->sync_op_pending = B_FALSE;
3633 				cv_signal(&vdcp->sync_pending_cv);
3634 				break;
3635 
3636 			case CB_STRATEGY:
3637 				bufp = ldep->cb_arg;
3638 				ASSERT(bufp != NULL);
3639 				bufp->b_resid = bufp->b_bcount;
3640 				bioerror(bufp, EIO);
3641 				biodone(bufp);
3642 				break;
3643 
3644 			default:
3645 				ASSERT(0);
3646 			}
3647 
3648 		}
3649 
3650 		/* get the next element to cancel */
3651 		if (++b_idx >= vdcp->local_dring_backup_len)
3652 			b_idx = 0;
3653 	}
3654 
3655 	/* all done - now clear up pending dring copy */
3656 	dring_size = vdcp->local_dring_backup_len *
3657 	    sizeof (vdcp->local_dring_backup[0]);
3658 
3659 	(void) kmem_free(vdcp->local_dring_backup, dring_size);
3660 
3661 	vdcp->local_dring_backup = NULL;
3662 
3663 	DTRACE_IO2(processed, int, count, vdc_t *, vdcp);
3664 }
3665 
3666 /*
3667  * Function:
3668  *	vdc_connection_timeout
3669  *
3670  * Description:
3671  *	This function is invoked if the timeout set to establish the connection
3672  *	with vds expires. This will happen if we spend too much time in the
3673  *	VDC_STATE_INIT_WAITING or VDC_STATE_NEGOTIATE states. Then we will
3674  *	cancel any pending request and mark them as failed.
3675  *
3676  *	If the timeout does not expire, it will be cancelled when we reach the
3677  *	VDC_STATE_HANDLE_PENDING or VDC_STATE_RESETTING state. This function can
3678  *	be invoked while we are in the VDC_STATE_HANDLE_PENDING or
3679  *	VDC_STATE_RESETTING state in which case we do nothing because the
3680  *	timeout is being cancelled.
3681  *
3682  * Arguments:
3683  *	arg	- argument of the timeout function actually a soft state
3684  *		  pointer for the instance of the device driver.
3685  *
3686  * Return Code:
3687  *	None
3688  */
3689 void
3690 vdc_connection_timeout(void *arg)
3691 {
3692 	vdc_t 		*vdcp = (vdc_t *)arg;
3693 
3694 	mutex_enter(&vdcp->lock);
3695 
3696 	if (vdcp->state == VDC_STATE_HANDLE_PENDING ||
3697 	    vdcp->state == VDC_STATE_DETACH) {
3698 		/*
3699 		 * The connection has just been re-established or
3700 		 * we are detaching.
3701 		 */
3702 		vdcp->ctimeout_reached = B_FALSE;
3703 		mutex_exit(&vdcp->lock);
3704 		return;
3705 	}
3706 
3707 	vdcp->ctimeout_reached = B_TRUE;
3708 
3709 	/* notify requests waiting for sending */
3710 	cv_broadcast(&vdcp->running_cv);
3711 
3712 	/* cancel requests waiting for a result */
3713 	vdc_cancel_backup_ring(vdcp);
3714 
3715 	mutex_exit(&vdcp->lock);
3716 
3717 	cmn_err(CE_NOTE, "[%d] connection to service domain timeout",
3718 	    vdcp->instance);
3719 }
3720 
3721 /*
3722  * Function:
3723  *	vdc_backup_local_dring()
3724  *
3725  * Description:
3726  *	Backup the current dring in the event of a reset. The Dring
3727  *	transactions will be resubmitted to the server when the
3728  *	connection is restored.
3729  *
3730  * Arguments:
3731  *	vdcp	- soft state pointer for this instance of the device driver.
3732  *
3733  * Return Code:
3734  *	NONE
3735  */
3736 static void
3737 vdc_backup_local_dring(vdc_t *vdcp)
3738 {
3739 	int dring_size;
3740 
3741 	ASSERT(MUTEX_HELD(&vdcp->lock));
3742 	ASSERT(vdcp->state == VDC_STATE_RESETTING);
3743 
3744 	/*
3745 	 * If the backup dring is stil around, it means
3746 	 * that the last restore did not complete. However,
3747 	 * since we never got back into the running state,
3748 	 * the backup copy we have is still valid.
3749 	 */
3750 	if (vdcp->local_dring_backup != NULL) {
3751 		DMSG(vdcp, 1, "reusing local descriptor ring backup "
3752 		    "(len=%d, tail=%d)\n", vdcp->local_dring_backup_len,
3753 		    vdcp->local_dring_backup_tail);
3754 		return;
3755 	}
3756 
3757 	/*
3758 	 * The backup dring can be NULL and the local dring may not be
3759 	 * initialized. This can happen if we had a reset while establishing
3760 	 * a new connection but after the connection has timed out. In that
3761 	 * case the backup dring is NULL because the requests have been
3762 	 * cancelled and the request occured before the local dring is
3763 	 * initialized.
3764 	 */
3765 	if (!(vdcp->initialized & VDC_DRING_LOCAL))
3766 		return;
3767 
3768 	DMSG(vdcp, 1, "backing up the local descriptor ring (len=%d, "
3769 	    "tail=%d)\n", vdcp->dring_len, vdcp->dring_curr_idx);
3770 
3771 	dring_size = vdcp->dring_len * sizeof (vdcp->local_dring[0]);
3772 
3773 	vdcp->local_dring_backup = kmem_alloc(dring_size, KM_SLEEP);
3774 	bcopy(vdcp->local_dring, vdcp->local_dring_backup, dring_size);
3775 
3776 	vdcp->local_dring_backup_tail = vdcp->dring_curr_idx;
3777 	vdcp->local_dring_backup_len = vdcp->dring_len;
3778 }
3779 
3780 /* -------------------------------------------------------------------------- */
3781 
3782 /*
3783  * The following functions process the incoming messages from vds
3784  */
3785 
3786 /*
3787  * Function:
3788  *      vdc_process_msg_thread()
3789  *
3790  * Description:
3791  *
3792  *	Main VDC message processing thread. Each vDisk instance
3793  * 	consists of a copy of this thread. This thread triggers
3794  * 	all the handshakes and data exchange with the server. It
3795  * 	also handles all channel resets
3796  *
3797  * Arguments:
3798  *      vdc     - soft state pointer for this instance of the device driver.
3799  *
3800  * Return Code:
3801  *      None
3802  */
3803 static void
3804 vdc_process_msg_thread(vdc_t *vdcp)
3805 {
3806 	int	status;
3807 	int	ctimeout;
3808 	timeout_id_t tmid = 0;
3809 
3810 	mutex_enter(&vdcp->lock);
3811 
3812 	for (;;) {
3813 
3814 #define	Q(_s)	(vdcp->state == _s) ? #_s :
3815 		DMSG(vdcp, 3, "state = %d (%s)\n", vdcp->state,
3816 		    Q(VDC_STATE_INIT)
3817 		    Q(VDC_STATE_INIT_WAITING)
3818 		    Q(VDC_STATE_NEGOTIATE)
3819 		    Q(VDC_STATE_HANDLE_PENDING)
3820 		    Q(VDC_STATE_RUNNING)
3821 		    Q(VDC_STATE_RESETTING)
3822 		    Q(VDC_STATE_DETACH)
3823 		    "UNKNOWN");
3824 
3825 		switch (vdcp->state) {
3826 		case VDC_STATE_INIT:
3827 
3828 			/*
3829 			 * If requested, start a timeout to check if the
3830 			 * connection with vds is established in the
3831 			 * specified delay. If the timeout expires, we
3832 			 * will cancel any pending request.
3833 			 *
3834 			 * If some reset have occurred while establishing
3835 			 * the connection, we already have a timeout armed
3836 			 * and in that case we don't need to arm a new one.
3837 			 */
3838 			ctimeout = (vdc_timeout != 0)?
3839 			    vdc_timeout : vdcp->ctimeout;
3840 
3841 			if (ctimeout != 0 && tmid == 0) {
3842 				tmid = timeout(vdc_connection_timeout, vdcp,
3843 				    ctimeout * drv_usectohz(1000000));
3844 			}
3845 
3846 			/* Check if have re-initializing repeatedly */
3847 			if (vdcp->hshake_cnt++ > vdc_hshake_retries &&
3848 			    vdcp->lifecycle != VDC_LC_ONLINE) {
3849 				cmn_err(CE_NOTE, "[%d] disk access failed.\n",
3850 				    vdcp->instance);
3851 				vdcp->state = VDC_STATE_DETACH;
3852 				break;
3853 			}
3854 
3855 			/* Bring up connection with vds via LDC */
3856 			status = vdc_start_ldc_connection(vdcp);
3857 			if (status == EINVAL) {
3858 				DMSG(vdcp, 0, "[%d] Could not start LDC",
3859 				    vdcp->instance);
3860 				vdcp->state = VDC_STATE_DETACH;
3861 			} else {
3862 				vdcp->state = VDC_STATE_INIT_WAITING;
3863 			}
3864 			break;
3865 
3866 		case VDC_STATE_INIT_WAITING:
3867 
3868 			/*
3869 			 * Let the callback event move us on
3870 			 * when channel is open to server
3871 			 */
3872 			while (vdcp->ldc_state != LDC_UP) {
3873 				cv_wait(&vdcp->initwait_cv, &vdcp->lock);
3874 				if (vdcp->state != VDC_STATE_INIT_WAITING) {
3875 					DMSG(vdcp, 0,
3876 				"state moved to %d out from under us...\n",
3877 					    vdcp->state);
3878 
3879 					break;
3880 				}
3881 			}
3882 			if (vdcp->state == VDC_STATE_INIT_WAITING &&
3883 			    vdcp->ldc_state == LDC_UP) {
3884 				vdcp->state = VDC_STATE_NEGOTIATE;
3885 			}
3886 			break;
3887 
3888 		case VDC_STATE_NEGOTIATE:
3889 			switch (status = vdc_ver_negotiation(vdcp)) {
3890 			case 0:
3891 				break;
3892 			default:
3893 				DMSG(vdcp, 0, "ver negotiate failed (%d)..\n",
3894 				    status);
3895 				goto reset;
3896 			}
3897 
3898 			switch (status = vdc_attr_negotiation(vdcp)) {
3899 			case 0:
3900 				break;
3901 			default:
3902 				DMSG(vdcp, 0, "attr negotiate failed (%d)..\n",
3903 				    status);
3904 				goto reset;
3905 			}
3906 
3907 			switch (status = vdc_dring_negotiation(vdcp)) {
3908 			case 0:
3909 				break;
3910 			default:
3911 				DMSG(vdcp, 0, "dring negotiate failed (%d)..\n",
3912 				    status);
3913 				goto reset;
3914 			}
3915 
3916 			switch (status = vdc_rdx_exchange(vdcp)) {
3917 			case 0:
3918 				vdcp->state = VDC_STATE_HANDLE_PENDING;
3919 				goto done;
3920 			default:
3921 				DMSG(vdcp, 0, "RDX xchg failed ..(%d)\n",
3922 				    status);
3923 				goto reset;
3924 			}
3925 reset:
3926 			DMSG(vdcp, 0, "negotiation failed: resetting (%d)\n",
3927 			    status);
3928 			vdcp->state = VDC_STATE_RESETTING;
3929 			vdcp->self_reset = B_TRUE;
3930 done:
3931 			DMSG(vdcp, 0, "negotiation complete (state=0x%x)...\n",
3932 			    vdcp->state);
3933 			break;
3934 
3935 		case VDC_STATE_HANDLE_PENDING:
3936 
3937 			if (vdcp->ctimeout_reached) {
3938 				/*
3939 				 * The connection timeout had been reached so
3940 				 * pending requests have been cancelled. Now
3941 				 * that the connection is back we can reset
3942 				 * the timeout.
3943 				 */
3944 				ASSERT(vdcp->local_dring_backup == NULL);
3945 				ASSERT(tmid != 0);
3946 				tmid = 0;
3947 				vdcp->ctimeout_reached = B_FALSE;
3948 				vdcp->state = VDC_STATE_RUNNING;
3949 				DMSG(vdcp, 0, "[%d] connection to service "
3950 				    "domain is up", vdcp->instance);
3951 				break;
3952 			}
3953 
3954 			mutex_exit(&vdcp->lock);
3955 			if (tmid != 0) {
3956 				(void) untimeout(tmid);
3957 				tmid = 0;
3958 			}
3959 			status = vdc_resubmit_backup_dring(vdcp);
3960 			mutex_enter(&vdcp->lock);
3961 
3962 			if (status)
3963 				vdcp->state = VDC_STATE_RESETTING;
3964 			else
3965 				vdcp->state = VDC_STATE_RUNNING;
3966 
3967 			break;
3968 
3969 		/* enter running state */
3970 		case VDC_STATE_RUNNING:
3971 			/*
3972 			 * Signal anyone waiting for the connection
3973 			 * to come on line.
3974 			 */
3975 			vdcp->hshake_cnt = 0;
3976 			cv_broadcast(&vdcp->running_cv);
3977 
3978 			/* failfast has to been checked after reset */
3979 			cv_signal(&vdcp->failfast_cv);
3980 
3981 			/* ownership is lost during reset */
3982 			if (vdcp->ownership & VDC_OWNERSHIP_WANTED)
3983 				vdcp->ownership |= VDC_OWNERSHIP_RESET;
3984 			cv_signal(&vdcp->ownership_cv);
3985 
3986 			mutex_exit(&vdcp->lock);
3987 
3988 			for (;;) {
3989 				vio_msg_t msg;
3990 				status = vdc_wait_for_response(vdcp, &msg);
3991 				if (status) break;
3992 
3993 				DMSG(vdcp, 1, "[%d] new pkt(s) available\n",
3994 				    vdcp->instance);
3995 				status = vdc_process_data_msg(vdcp, &msg);
3996 				if (status) {
3997 					DMSG(vdcp, 1, "[%d] process_data_msg "
3998 					    "returned err=%d\n", vdcp->instance,
3999 					    status);
4000 					break;
4001 				}
4002 
4003 			}
4004 
4005 			mutex_enter(&vdcp->lock);
4006 
4007 			vdcp->state = VDC_STATE_RESETTING;
4008 			vdcp->self_reset = B_TRUE;
4009 			break;
4010 
4011 		case VDC_STATE_RESETTING:
4012 			/*
4013 			 * When we reach this state, we either come from the
4014 			 * VDC_STATE_RUNNING state and we can have pending
4015 			 * request but no timeout is armed; or we come from
4016 			 * the VDC_STATE_INIT_WAITING, VDC_NEGOTIATE or
4017 			 * VDC_HANDLE_PENDING state and there is no pending
4018 			 * request or pending requests have already been copied
4019 			 * into the backup dring. So we can safely keep the
4020 			 * connection timeout armed while we are in this state.
4021 			 */
4022 
4023 			DMSG(vdcp, 0, "Initiating channel reset "
4024 			    "(pending = %d)\n", (int)vdcp->threads_pending);
4025 
4026 			if (vdcp->self_reset) {
4027 				DMSG(vdcp, 0,
4028 				    "[%d] calling stop_ldc_connection.\n",
4029 				    vdcp->instance);
4030 				status = vdc_stop_ldc_connection(vdcp);
4031 				vdcp->self_reset = B_FALSE;
4032 			}
4033 
4034 			/*
4035 			 * Wait for all threads currently waiting
4036 			 * for a free dring entry to use.
4037 			 */
4038 			while (vdcp->threads_pending) {
4039 				cv_broadcast(&vdcp->membind_cv);
4040 				cv_broadcast(&vdcp->dring_free_cv);
4041 				mutex_exit(&vdcp->lock);
4042 				/* give the waiters enough time to wake up */
4043 				delay(vdc_hz_min_ldc_delay);
4044 				mutex_enter(&vdcp->lock);
4045 			}
4046 
4047 			ASSERT(vdcp->threads_pending == 0);
4048 
4049 			/* Sanity check that no thread is receiving */
4050 			ASSERT(vdcp->read_state != VDC_READ_WAITING);
4051 
4052 			vdcp->read_state = VDC_READ_IDLE;
4053 
4054 			vdc_backup_local_dring(vdcp);
4055 
4056 			/* cleanup the old d-ring */
4057 			vdc_destroy_descriptor_ring(vdcp);
4058 
4059 			/* go and start again */
4060 			vdcp->state = VDC_STATE_INIT;
4061 
4062 			break;
4063 
4064 		case VDC_STATE_DETACH:
4065 			DMSG(vdcp, 0, "[%d] Reset thread exit cleanup ..\n",
4066 			    vdcp->instance);
4067 
4068 			/* cancel any pending timeout */
4069 			mutex_exit(&vdcp->lock);
4070 			if (tmid != 0) {
4071 				(void) untimeout(tmid);
4072 				tmid = 0;
4073 			}
4074 			mutex_enter(&vdcp->lock);
4075 
4076 			/*
4077 			 * Signal anyone waiting for connection
4078 			 * to come online
4079 			 */
4080 			cv_broadcast(&vdcp->running_cv);
4081 
4082 			while (vdcp->sync_op_pending) {
4083 				cv_signal(&vdcp->sync_pending_cv);
4084 				cv_signal(&vdcp->sync_blocked_cv);
4085 				mutex_exit(&vdcp->lock);
4086 				/* give the waiters enough time to wake up */
4087 				delay(vdc_hz_min_ldc_delay);
4088 				mutex_enter(&vdcp->lock);
4089 			}
4090 
4091 			mutex_exit(&vdcp->lock);
4092 
4093 			DMSG(vdcp, 0, "[%d] Msg processing thread exiting ..\n",
4094 			    vdcp->instance);
4095 			thread_exit();
4096 			break;
4097 		}
4098 	}
4099 }
4100 
4101 
4102 /*
4103  * Function:
4104  *	vdc_process_data_msg()
4105  *
4106  * Description:
4107  *	This function is called by the message processing thread each time
4108  *	a message with a msgtype of VIO_TYPE_DATA is received. It will either
4109  *	be an ACK or NACK from vds[1] which vdc handles as follows.
4110  *		ACK	- wake up the waiting thread
4111  *		NACK	- resend any messages necessary
4112  *
4113  *	[1] Although the message format allows it, vds should not send a
4114  *	    VIO_SUBTYPE_INFO message to vdc asking it to read data; if for
4115  *	    some bizarre reason it does, vdc will reset the connection.
4116  *
4117  * Arguments:
4118  *	vdc	- soft state pointer for this instance of the device driver.
4119  *	msg	- the LDC message sent by vds
4120  *
4121  * Return Code:
4122  *	0	- Success.
4123  *	> 0	- error value returned by LDC
4124  */
4125 static int
4126 vdc_process_data_msg(vdc_t *vdcp, vio_msg_t *msg)
4127 {
4128 	int			status = 0;
4129 	vio_dring_msg_t		*dring_msg;
4130 	vdc_local_desc_t	*ldep = NULL;
4131 	int			start, end;
4132 	int			idx;
4133 
4134 	dring_msg = (vio_dring_msg_t *)msg;
4135 
4136 	ASSERT(msg->tag.vio_msgtype == VIO_TYPE_DATA);
4137 	ASSERT(vdcp != NULL);
4138 
4139 	mutex_enter(&vdcp->lock);
4140 
4141 	/*
4142 	 * Check to see if the message has bogus data
4143 	 */
4144 	idx = start = dring_msg->start_idx;
4145 	end = dring_msg->end_idx;
4146 	if ((start >= vdcp->dring_len) ||
4147 	    (end >= vdcp->dring_len) || (end < -1)) {
4148 		DMSG(vdcp, 0, "[%d] Bogus ACK data : start %d, end %d\n",
4149 		    vdcp->instance, start, end);
4150 		mutex_exit(&vdcp->lock);
4151 		return (EINVAL);
4152 	}
4153 
4154 	/*
4155 	 * Verify that the sequence number is what vdc expects.
4156 	 */
4157 	switch (vdc_verify_seq_num(vdcp, dring_msg)) {
4158 	case VDC_SEQ_NUM_TODO:
4159 		break;	/* keep processing this message */
4160 	case VDC_SEQ_NUM_SKIP:
4161 		mutex_exit(&vdcp->lock);
4162 		return (0);
4163 	case VDC_SEQ_NUM_INVALID:
4164 		mutex_exit(&vdcp->lock);
4165 		DMSG(vdcp, 0, "[%d] invalid seqno\n", vdcp->instance);
4166 		return (ENXIO);
4167 	}
4168 
4169 	if (msg->tag.vio_subtype == VIO_SUBTYPE_NACK) {
4170 		DMSG(vdcp, 0, "[%d] DATA NACK\n", vdcp->instance);
4171 		VDC_DUMP_DRING_MSG(dring_msg);
4172 		mutex_exit(&vdcp->lock);
4173 		return (EIO);
4174 
4175 	} else if (msg->tag.vio_subtype == VIO_SUBTYPE_INFO) {
4176 		mutex_exit(&vdcp->lock);
4177 		return (EPROTO);
4178 	}
4179 
4180 	DTRACE_IO2(recv, vio_dring_msg_t, dring_msg, vdc_t *, vdcp);
4181 	DMSG(vdcp, 1, ": start %d end %d\n", start, end);
4182 	ASSERT(start == end);
4183 
4184 	ldep = &vdcp->local_dring[idx];
4185 
4186 	DMSG(vdcp, 1, ": state 0x%x - cb_type 0x%x\n",
4187 	    ldep->dep->hdr.dstate, ldep->cb_type);
4188 
4189 	if (ldep->dep->hdr.dstate == VIO_DESC_DONE) {
4190 		struct buf *bufp;
4191 
4192 		switch (ldep->cb_type) {
4193 		case CB_SYNC:
4194 			ASSERT(vdcp->sync_op_pending);
4195 
4196 			status = vdc_depopulate_descriptor(vdcp, idx);
4197 			vdcp->sync_op_status = status;
4198 			vdcp->sync_op_pending = B_FALSE;
4199 			cv_signal(&vdcp->sync_pending_cv);
4200 			break;
4201 
4202 		case CB_STRATEGY:
4203 			bufp = ldep->cb_arg;
4204 			ASSERT(bufp != NULL);
4205 			bufp->b_resid =
4206 			    bufp->b_bcount - ldep->dep->payload.nbytes;
4207 			status = ldep->dep->payload.status; /* Future:ntoh */
4208 			if (status != 0) {
4209 				DMSG(vdcp, 1, "strategy status=%d\n", status);
4210 				bioerror(bufp, status);
4211 			}
4212 
4213 			(void) vdc_depopulate_descriptor(vdcp, idx);
4214 
4215 			DMSG(vdcp, 1,
4216 			    "strategy complete req=%ld bytes resp=%ld bytes\n",
4217 			    bufp->b_bcount, ldep->dep->payload.nbytes);
4218 
4219 			if (status != 0 && vdcp->failfast_interval != 0) {
4220 				/*
4221 				 * The I/O has failed and failfast is enabled.
4222 				 * We need the failfast thread to check if the
4223 				 * failure is due to a reservation conflict.
4224 				 */
4225 				(void) vdc_failfast_io_queue(vdcp, bufp);
4226 			} else {
4227 				biodone(bufp);
4228 			}
4229 			break;
4230 
4231 		default:
4232 			ASSERT(0);
4233 		}
4234 	}
4235 
4236 	/* let the arrival signal propogate */
4237 	mutex_exit(&vdcp->lock);
4238 
4239 	/* probe gives the count of how many entries were processed */
4240 	DTRACE_IO2(processed, int, 1, vdc_t *, vdcp);
4241 
4242 	return (0);
4243 }
4244 
4245 
4246 /*
4247  * Function:
4248  *	vdc_handle_ver_msg()
4249  *
4250  * Description:
4251  *
4252  * Arguments:
4253  *	vdc	- soft state pointer for this instance of the device driver.
4254  *	ver_msg	- LDC message sent by vDisk server
4255  *
4256  * Return Code:
4257  *	0	- Success
4258  */
4259 static int
4260 vdc_handle_ver_msg(vdc_t *vdc, vio_ver_msg_t *ver_msg)
4261 {
4262 	int status = 0;
4263 
4264 	ASSERT(vdc != NULL);
4265 	ASSERT(mutex_owned(&vdc->lock));
4266 
4267 	if (ver_msg->tag.vio_subtype_env != VIO_VER_INFO) {
4268 		return (EPROTO);
4269 	}
4270 
4271 	if (ver_msg->dev_class != VDEV_DISK_SERVER) {
4272 		return (EINVAL);
4273 	}
4274 
4275 	switch (ver_msg->tag.vio_subtype) {
4276 	case VIO_SUBTYPE_ACK:
4277 		/*
4278 		 * We check to see if the version returned is indeed supported
4279 		 * (The server may have also adjusted the minor number downwards
4280 		 * and if so 'ver_msg' will contain the actual version agreed)
4281 		 */
4282 		if (vdc_is_supported_version(ver_msg)) {
4283 			vdc->ver.major = ver_msg->ver_major;
4284 			vdc->ver.minor = ver_msg->ver_minor;
4285 			ASSERT(vdc->ver.major > 0);
4286 		} else {
4287 			status = EPROTO;
4288 		}
4289 		break;
4290 
4291 	case VIO_SUBTYPE_NACK:
4292 		/*
4293 		 * call vdc_is_supported_version() which will return the next
4294 		 * supported version (if any) in 'ver_msg'
4295 		 */
4296 		(void) vdc_is_supported_version(ver_msg);
4297 		if (ver_msg->ver_major > 0) {
4298 			size_t len = sizeof (*ver_msg);
4299 
4300 			ASSERT(vdc->ver.major > 0);
4301 
4302 			/* reset the necessary fields and resend */
4303 			ver_msg->tag.vio_subtype = VIO_SUBTYPE_INFO;
4304 			ver_msg->dev_class = VDEV_DISK;
4305 
4306 			status = vdc_send(vdc, (caddr_t)ver_msg, &len);
4307 			DMSG(vdc, 0, "[%d] Resend VER info (LDC status = %d)\n",
4308 			    vdc->instance, status);
4309 			if (len != sizeof (*ver_msg))
4310 				status = EBADMSG;
4311 		} else {
4312 			DMSG(vdc, 0, "[%d] No common version with vDisk server",
4313 			    vdc->instance);
4314 			status = ENOTSUP;
4315 		}
4316 
4317 		break;
4318 	case VIO_SUBTYPE_INFO:
4319 		/*
4320 		 * Handle the case where vds starts handshake
4321 		 * (for now only vdc is the instigator)
4322 		 */
4323 		status = ENOTSUP;
4324 		break;
4325 
4326 	default:
4327 		status = EINVAL;
4328 		break;
4329 	}
4330 
4331 	return (status);
4332 }
4333 
4334 /*
4335  * Function:
4336  *	vdc_handle_attr_msg()
4337  *
4338  * Description:
4339  *
4340  * Arguments:
4341  *	vdc	- soft state pointer for this instance of the device driver.
4342  *	attr_msg	- LDC message sent by vDisk server
4343  *
4344  * Return Code:
4345  *	0	- Success
4346  */
4347 static int
4348 vdc_handle_attr_msg(vdc_t *vdc, vd_attr_msg_t *attr_msg)
4349 {
4350 	int status = 0;
4351 
4352 	ASSERT(vdc != NULL);
4353 	ASSERT(mutex_owned(&vdc->lock));
4354 
4355 	if (attr_msg->tag.vio_subtype_env != VIO_ATTR_INFO) {
4356 		return (EPROTO);
4357 	}
4358 
4359 	switch (attr_msg->tag.vio_subtype) {
4360 	case VIO_SUBTYPE_ACK:
4361 		/*
4362 		 * We now verify the attributes sent by vds.
4363 		 */
4364 		if (attr_msg->vdisk_size == 0) {
4365 			DMSG(vdc, 0, "[%d] Invalid disk size from vds",
4366 			    vdc->instance);
4367 			status = EINVAL;
4368 			break;
4369 		}
4370 
4371 		if (attr_msg->max_xfer_sz == 0) {
4372 			DMSG(vdc, 0, "[%d] Invalid transfer size from vds",
4373 			    vdc->instance);
4374 			status = EINVAL;
4375 			break;
4376 		}
4377 
4378 		if (attr_msg->vdisk_size == VD_SIZE_UNKNOWN) {
4379 			DMSG(vdc, 0, "[%d] Unknown disk size from vds",
4380 			    vdc->instance);
4381 			attr_msg->vdisk_size = 0;
4382 		}
4383 
4384 		/*
4385 		 * If the disk size is already set check that it hasn't changed.
4386 		 */
4387 		if ((vdc->vdisk_size != 0) && (attr_msg->vdisk_size != 0) &&
4388 		    (vdc->vdisk_size != attr_msg->vdisk_size)) {
4389 			DMSG(vdc, 0, "[%d] Different disk size from vds "
4390 			    "(old=0x%lx - new=0x%lx", vdc->instance,
4391 			    vdc->vdisk_size, attr_msg->vdisk_size)
4392 			status = EINVAL;
4393 			break;
4394 		}
4395 
4396 		vdc->vdisk_size = attr_msg->vdisk_size;
4397 		vdc->vdisk_type = attr_msg->vdisk_type;
4398 		vdc->operations = attr_msg->operations;
4399 		if (vio_ver_is_supported(vdc->ver, 1, 1))
4400 			vdc->vdisk_media = attr_msg->vdisk_media;
4401 		else
4402 			vdc->vdisk_media = 0;
4403 
4404 		DMSG(vdc, 0, "[%d] max_xfer_sz: sent %lx acked %lx\n",
4405 		    vdc->instance, vdc->max_xfer_sz, attr_msg->max_xfer_sz);
4406 		DMSG(vdc, 0, "[%d] vdisk_block_size: sent %lx acked %x\n",
4407 		    vdc->instance, vdc->block_size,
4408 		    attr_msg->vdisk_block_size);
4409 
4410 		/*
4411 		 * We don't know at compile time what the vDisk server will
4412 		 * think are good values but we apply a large (arbitrary)
4413 		 * upper bound to prevent memory exhaustion in vdc if it was
4414 		 * allocating a DRing based of huge values sent by the server.
4415 		 * We probably will never exceed this except if the message
4416 		 * was garbage.
4417 		 */
4418 		if ((attr_msg->max_xfer_sz * attr_msg->vdisk_block_size) <=
4419 		    (PAGESIZE * DEV_BSIZE)) {
4420 			vdc->max_xfer_sz = attr_msg->max_xfer_sz;
4421 			vdc->block_size = attr_msg->vdisk_block_size;
4422 		} else {
4423 			DMSG(vdc, 0, "[%d] vds block transfer size too big;"
4424 			    " using max supported by vdc", vdc->instance);
4425 		}
4426 
4427 		if ((attr_msg->xfer_mode != VIO_DRING_MODE) ||
4428 		    (attr_msg->vdisk_size > INT64_MAX) ||
4429 		    (attr_msg->operations == 0) ||
4430 		    (attr_msg->vdisk_type > VD_DISK_TYPE_DISK)) {
4431 			DMSG(vdc, 0, "[%d] Invalid attributes from vds",
4432 			    vdc->instance);
4433 			status = EINVAL;
4434 			break;
4435 		}
4436 
4437 		/*
4438 		 * Now that we have received all attributes we can create a
4439 		 * fake geometry for the disk.
4440 		 */
4441 		vdc_create_fake_geometry(vdc);
4442 		break;
4443 
4444 	case VIO_SUBTYPE_NACK:
4445 		/*
4446 		 * vds could not handle the attributes we sent so we
4447 		 * stop negotiating.
4448 		 */
4449 		status = EPROTO;
4450 		break;
4451 
4452 	case VIO_SUBTYPE_INFO:
4453 		/*
4454 		 * Handle the case where vds starts the handshake
4455 		 * (for now; vdc is the only supported instigatior)
4456 		 */
4457 		status = ENOTSUP;
4458 		break;
4459 
4460 	default:
4461 		status = ENOTSUP;
4462 		break;
4463 	}
4464 
4465 	return (status);
4466 }
4467 
4468 /*
4469  * Function:
4470  *	vdc_handle_dring_reg_msg()
4471  *
4472  * Description:
4473  *
4474  * Arguments:
4475  *	vdc		- soft state pointer for this instance of the driver.
4476  *	dring_msg	- LDC message sent by vDisk server
4477  *
4478  * Return Code:
4479  *	0	- Success
4480  */
4481 static int
4482 vdc_handle_dring_reg_msg(vdc_t *vdc, vio_dring_reg_msg_t *dring_msg)
4483 {
4484 	int		status = 0;
4485 
4486 	ASSERT(vdc != NULL);
4487 	ASSERT(mutex_owned(&vdc->lock));
4488 
4489 	if (dring_msg->tag.vio_subtype_env != VIO_DRING_REG) {
4490 		return (EPROTO);
4491 	}
4492 
4493 	switch (dring_msg->tag.vio_subtype) {
4494 	case VIO_SUBTYPE_ACK:
4495 		/* save the received dring_ident */
4496 		vdc->dring_ident = dring_msg->dring_ident;
4497 		DMSG(vdc, 0, "[%d] Received dring ident=0x%lx\n",
4498 		    vdc->instance, vdc->dring_ident);
4499 		break;
4500 
4501 	case VIO_SUBTYPE_NACK:
4502 		/*
4503 		 * vds could not handle the DRing info we sent so we
4504 		 * stop negotiating.
4505 		 */
4506 		DMSG(vdc, 0, "[%d] server could not register DRing\n",
4507 		    vdc->instance);
4508 		status = EPROTO;
4509 		break;
4510 
4511 	case VIO_SUBTYPE_INFO:
4512 		/*
4513 		 * Handle the case where vds starts handshake
4514 		 * (for now only vdc is the instigatior)
4515 		 */
4516 		status = ENOTSUP;
4517 		break;
4518 	default:
4519 		status = ENOTSUP;
4520 	}
4521 
4522 	return (status);
4523 }
4524 
4525 /*
4526  * Function:
4527  *	vdc_verify_seq_num()
4528  *
4529  * Description:
4530  *	This functions verifies that the sequence number sent back by the vDisk
4531  *	server with the latest message is what is expected (i.e. it is greater
4532  *	than the last seq num sent by the vDisk server and less than or equal
4533  *	to the last seq num generated by vdc).
4534  *
4535  *	It then checks the request ID to see if any requests need processing
4536  *	in the DRing.
4537  *
4538  * Arguments:
4539  *	vdc		- soft state pointer for this instance of the driver.
4540  *	dring_msg	- pointer to the LDC message sent by vds
4541  *
4542  * Return Code:
4543  *	VDC_SEQ_NUM_TODO	- Message needs to be processed
4544  *	VDC_SEQ_NUM_SKIP	- Message has already been processed
4545  *	VDC_SEQ_NUM_INVALID	- The seq numbers are so out of sync,
4546  *				  vdc cannot deal with them
4547  */
4548 static int
4549 vdc_verify_seq_num(vdc_t *vdc, vio_dring_msg_t *dring_msg)
4550 {
4551 	ASSERT(vdc != NULL);
4552 	ASSERT(dring_msg != NULL);
4553 	ASSERT(mutex_owned(&vdc->lock));
4554 
4555 	/*
4556 	 * Check to see if the messages were responded to in the correct
4557 	 * order by vds.
4558 	 */
4559 	if ((dring_msg->seq_num <= vdc->seq_num_reply) ||
4560 	    (dring_msg->seq_num > vdc->seq_num)) {
4561 		DMSG(vdc, 0, "?[%d] Bogus sequence_number %lu: "
4562 		    "%lu > expected <= %lu (last proc req %lu sent %lu)\n",
4563 		    vdc->instance, dring_msg->seq_num,
4564 		    vdc->seq_num_reply, vdc->seq_num,
4565 		    vdc->req_id_proc, vdc->req_id);
4566 		return (VDC_SEQ_NUM_INVALID);
4567 	}
4568 	vdc->seq_num_reply = dring_msg->seq_num;
4569 
4570 	if (vdc->req_id_proc < vdc->req_id)
4571 		return (VDC_SEQ_NUM_TODO);
4572 	else
4573 		return (VDC_SEQ_NUM_SKIP);
4574 }
4575 
4576 
4577 /*
4578  * Function:
4579  *	vdc_is_supported_version()
4580  *
4581  * Description:
4582  *	This routine checks if the major/minor version numbers specified in
4583  *	'ver_msg' are supported. If not it finds the next version that is
4584  *	in the supported version list 'vdc_version[]' and sets the fields in
4585  *	'ver_msg' to those values
4586  *
4587  * Arguments:
4588  *	ver_msg	- LDC message sent by vDisk server
4589  *
4590  * Return Code:
4591  *	B_TRUE	- Success
4592  *	B_FALSE	- Version not supported
4593  */
4594 static boolean_t
4595 vdc_is_supported_version(vio_ver_msg_t *ver_msg)
4596 {
4597 	int vdc_num_versions = sizeof (vdc_version) / sizeof (vdc_version[0]);
4598 
4599 	for (int i = 0; i < vdc_num_versions; i++) {
4600 		ASSERT(vdc_version[i].major > 0);
4601 		ASSERT((i == 0) ||
4602 		    (vdc_version[i].major < vdc_version[i-1].major));
4603 
4604 		/*
4605 		 * If the major versions match, adjust the minor version, if
4606 		 * necessary, down to the highest value supported by this
4607 		 * client. The server should support all minor versions lower
4608 		 * than the value it sent
4609 		 */
4610 		if (ver_msg->ver_major == vdc_version[i].major) {
4611 			if (ver_msg->ver_minor > vdc_version[i].minor) {
4612 				DMSGX(0,
4613 				    "Adjusting minor version from %u to %u",
4614 				    ver_msg->ver_minor, vdc_version[i].minor);
4615 				ver_msg->ver_minor = vdc_version[i].minor;
4616 			}
4617 			return (B_TRUE);
4618 		}
4619 
4620 		/*
4621 		 * If the message contains a higher major version number, set
4622 		 * the message's major/minor versions to the current values
4623 		 * and return false, so this message will get resent with
4624 		 * these values, and the server will potentially try again
4625 		 * with the same or a lower version
4626 		 */
4627 		if (ver_msg->ver_major > vdc_version[i].major) {
4628 			ver_msg->ver_major = vdc_version[i].major;
4629 			ver_msg->ver_minor = vdc_version[i].minor;
4630 			DMSGX(0, "Suggesting major/minor (0x%x/0x%x)\n",
4631 			    ver_msg->ver_major, ver_msg->ver_minor);
4632 
4633 			return (B_FALSE);
4634 		}
4635 
4636 		/*
4637 		 * Otherwise, the message's major version is less than the
4638 		 * current major version, so continue the loop to the next
4639 		 * (lower) supported version
4640 		 */
4641 	}
4642 
4643 	/*
4644 	 * No common version was found; "ground" the version pair in the
4645 	 * message to terminate negotiation
4646 	 */
4647 	ver_msg->ver_major = 0;
4648 	ver_msg->ver_minor = 0;
4649 
4650 	return (B_FALSE);
4651 }
4652 /* -------------------------------------------------------------------------- */
4653 
4654 /*
4655  * DKIO(7) support
4656  */
4657 
4658 typedef struct vdc_dk_arg {
4659 	struct dk_callback	dkc;
4660 	int			mode;
4661 	dev_t			dev;
4662 	vdc_t			*vdc;
4663 } vdc_dk_arg_t;
4664 
4665 /*
4666  * Function:
4667  * 	vdc_dkio_flush_cb()
4668  *
4669  * Description:
4670  *	This routine is a callback for DKIOCFLUSHWRITECACHE which can be called
4671  *	by kernel code.
4672  *
4673  * Arguments:
4674  *	arg	- a pointer to a vdc_dk_arg_t structure.
4675  */
4676 void
4677 vdc_dkio_flush_cb(void *arg)
4678 {
4679 	struct vdc_dk_arg	*dk_arg = (struct vdc_dk_arg *)arg;
4680 	struct dk_callback	*dkc = NULL;
4681 	vdc_t			*vdc = NULL;
4682 	int			rv;
4683 
4684 	if (dk_arg == NULL) {
4685 		cmn_err(CE_NOTE, "?[Unk] DKIOCFLUSHWRITECACHE arg is NULL\n");
4686 		return;
4687 	}
4688 	dkc = &dk_arg->dkc;
4689 	vdc = dk_arg->vdc;
4690 	ASSERT(vdc != NULL);
4691 
4692 	rv = vdc_do_sync_op(vdc, VD_OP_FLUSH, NULL, 0,
4693 	    VDCPART(dk_arg->dev), 0, CB_SYNC, 0, VIO_both_dir, B_TRUE);
4694 	if (rv != 0) {
4695 		DMSG(vdc, 0, "[%d] DKIOCFLUSHWRITECACHE failed %d : model %x\n",
4696 		    vdc->instance, rv,
4697 		    ddi_model_convert_from(dk_arg->mode & FMODELS));
4698 	}
4699 
4700 	/*
4701 	 * Trigger the call back to notify the caller the the ioctl call has
4702 	 * been completed.
4703 	 */
4704 	if ((dk_arg->mode & FKIOCTL) &&
4705 	    (dkc != NULL) &&
4706 	    (dkc->dkc_callback != NULL)) {
4707 		ASSERT(dkc->dkc_cookie != NULL);
4708 		(*dkc->dkc_callback)(dkc->dkc_cookie, rv);
4709 	}
4710 
4711 	/* Indicate that one less DKIO write flush is outstanding */
4712 	mutex_enter(&vdc->lock);
4713 	vdc->dkio_flush_pending--;
4714 	ASSERT(vdc->dkio_flush_pending >= 0);
4715 	mutex_exit(&vdc->lock);
4716 
4717 	/* free the mem that was allocated when the callback was dispatched */
4718 	kmem_free(arg, sizeof (vdc_dk_arg_t));
4719 }
4720 
4721 /*
4722  * Function:
4723  * 	vdc_dkio_get_partition()
4724  *
4725  * Description:
4726  *	This function implements the DKIOCGAPART ioctl.
4727  *
4728  * Arguments:
4729  *	vdc	- soft state pointer
4730  *	arg	- a pointer to a dk_map[NDKMAP] or dk_map32[NDKMAP] structure
4731  *	flag	- ioctl flags
4732  */
4733 static int
4734 vdc_dkio_get_partition(vdc_t *vdc, caddr_t arg, int flag)
4735 {
4736 	struct dk_geom *geom;
4737 	struct vtoc *vtoc;
4738 	union {
4739 		struct dk_map map[NDKMAP];
4740 		struct dk_map32 map32[NDKMAP];
4741 	} data;
4742 	int i, rv, size;
4743 
4744 	mutex_enter(&vdc->lock);
4745 
4746 	if ((rv = vdc_validate_geometry(vdc)) != 0) {
4747 		mutex_exit(&vdc->lock);
4748 		return (rv);
4749 	}
4750 
4751 	vtoc = vdc->vtoc;
4752 	geom = vdc->geom;
4753 
4754 	if (ddi_model_convert_from(flag & FMODELS) == DDI_MODEL_ILP32) {
4755 
4756 		for (i = 0; i < vtoc->v_nparts; i++) {
4757 			data.map32[i].dkl_cylno = vtoc->v_part[i].p_start /
4758 			    (geom->dkg_nhead * geom->dkg_nsect);
4759 			data.map32[i].dkl_nblk = vtoc->v_part[i].p_size;
4760 		}
4761 		size = NDKMAP * sizeof (struct dk_map32);
4762 
4763 	} else {
4764 
4765 		for (i = 0; i < vtoc->v_nparts; i++) {
4766 			data.map[i].dkl_cylno = vtoc->v_part[i].p_start /
4767 			    (geom->dkg_nhead * geom->dkg_nsect);
4768 			data.map[i].dkl_nblk = vtoc->v_part[i].p_size;
4769 		}
4770 		size = NDKMAP * sizeof (struct dk_map);
4771 
4772 	}
4773 
4774 	mutex_exit(&vdc->lock);
4775 
4776 	if (ddi_copyout(&data, arg, size, flag) != 0)
4777 		return (EFAULT);
4778 
4779 	return (0);
4780 }
4781 
4782 /*
4783  * Function:
4784  * 	vdc_dioctl_rwcmd()
4785  *
4786  * Description:
4787  *	This function implements the DIOCTL_RWCMD ioctl. This ioctl is used
4788  *	for DKC_DIRECT disks to read or write at an absolute disk offset.
4789  *
4790  * Arguments:
4791  *	dev	- device
4792  *	arg	- a pointer to a dadkio_rwcmd or dadkio_rwcmd32 structure
4793  *	flag	- ioctl flags
4794  */
4795 static int
4796 vdc_dioctl_rwcmd(dev_t dev, caddr_t arg, int flag)
4797 {
4798 	struct dadkio_rwcmd32 rwcmd32;
4799 	struct dadkio_rwcmd rwcmd;
4800 	struct iovec aiov;
4801 	struct uio auio;
4802 	int rw, status;
4803 	struct buf *buf;
4804 
4805 	if (ddi_model_convert_from(flag & FMODELS) == DDI_MODEL_ILP32) {
4806 		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd32,
4807 		    sizeof (struct dadkio_rwcmd32), flag)) {
4808 			return (EFAULT);
4809 		}
4810 		rwcmd.cmd = rwcmd32.cmd;
4811 		rwcmd.flags = rwcmd32.flags;
4812 		rwcmd.blkaddr = (daddr_t)rwcmd32.blkaddr;
4813 		rwcmd.buflen = rwcmd32.buflen;
4814 		rwcmd.bufaddr = (caddr_t)(uintptr_t)rwcmd32.bufaddr;
4815 	} else {
4816 		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd,
4817 		    sizeof (struct dadkio_rwcmd), flag)) {
4818 			return (EFAULT);
4819 		}
4820 	}
4821 
4822 	switch (rwcmd.cmd) {
4823 	case DADKIO_RWCMD_READ:
4824 		rw = B_READ;
4825 		break;
4826 	case DADKIO_RWCMD_WRITE:
4827 		rw = B_WRITE;
4828 		break;
4829 	default:
4830 		return (EINVAL);
4831 	}
4832 
4833 	bzero((caddr_t)&aiov, sizeof (struct iovec));
4834 	aiov.iov_base   = rwcmd.bufaddr;
4835 	aiov.iov_len    = rwcmd.buflen;
4836 
4837 	bzero((caddr_t)&auio, sizeof (struct uio));
4838 	auio.uio_iov    = &aiov;
4839 	auio.uio_iovcnt = 1;
4840 	auio.uio_loffset = rwcmd.blkaddr * DEV_BSIZE;
4841 	auio.uio_resid  = rwcmd.buflen;
4842 	auio.uio_segflg = flag & FKIOCTL ? UIO_SYSSPACE : UIO_USERSPACE;
4843 
4844 	buf = kmem_alloc(sizeof (buf_t), KM_SLEEP);
4845 	bioinit(buf);
4846 	/*
4847 	 * We use the private field of buf to specify that this is an
4848 	 * I/O using an absolute offset.
4849 	 */
4850 	buf->b_private = (void *)VD_SLICE_NONE;
4851 
4852 	status = physio(vdc_strategy, buf, dev, rw, vdc_min, &auio);
4853 
4854 	biofini(buf);
4855 	kmem_free(buf, sizeof (buf_t));
4856 
4857 	return (status);
4858 }
4859 
4860 /*
4861  * Allocate a buffer for a VD_OP_SCSICMD operation. The size of the allocated
4862  * buffer is returned in alloc_len.
4863  */
4864 static vd_scsi_t *
4865 vdc_scsi_alloc(int cdb_len, int sense_len, int datain_len, int dataout_len,
4866     int *alloc_len)
4867 {
4868 	vd_scsi_t *vd_scsi;
4869 	int vd_scsi_len = VD_SCSI_SIZE;
4870 
4871 	vd_scsi_len += P2ROUNDUP(cdb_len, sizeof (uint64_t));
4872 	vd_scsi_len += P2ROUNDUP(sense_len, sizeof (uint64_t));
4873 	vd_scsi_len += P2ROUNDUP(datain_len, sizeof (uint64_t));
4874 	vd_scsi_len += P2ROUNDUP(dataout_len, sizeof (uint64_t));
4875 
4876 	ASSERT(vd_scsi_len % sizeof (uint64_t) == 0);
4877 
4878 	vd_scsi = kmem_zalloc(vd_scsi_len, KM_SLEEP);
4879 
4880 	vd_scsi->cdb_len = cdb_len;
4881 	vd_scsi->sense_len = sense_len;
4882 	vd_scsi->datain_len = datain_len;
4883 	vd_scsi->dataout_len = dataout_len;
4884 
4885 	*alloc_len = vd_scsi_len;
4886 
4887 	return (vd_scsi);
4888 }
4889 
4890 /*
4891  * Convert the status of a SCSI command to a Solaris return code.
4892  *
4893  * Arguments:
4894  *	vd_scsi		- The SCSI operation buffer.
4895  *	log_error	- indicate if an error message should be logged.
4896  *
4897  * Note that our SCSI error messages are rather primitive for the moment
4898  * and could be improved by decoding some data like the SCSI command and
4899  * the sense key.
4900  *
4901  * Return value:
4902  *	0		- Status is good.
4903  *	EACCES		- Status reports a reservation conflict.
4904  *	ENOTSUP		- Status reports a check condition and sense key
4905  *			  reports an illegal request.
4906  *	EIO		- Any other status.
4907  */
4908 static int
4909 vdc_scsi_status(vdc_t *vdc, vd_scsi_t *vd_scsi, boolean_t log_error)
4910 {
4911 	int rv;
4912 	char path_str[MAXPATHLEN];
4913 	char panic_str[VDC_RESV_CONFLICT_FMT_LEN + MAXPATHLEN];
4914 	union scsi_cdb *cdb;
4915 	struct scsi_extended_sense *sense;
4916 
4917 	if (vd_scsi->cmd_status == STATUS_GOOD)
4918 		/* no error */
4919 		return (0);
4920 
4921 	/* when the tunable vdc_scsi_log_error is true we log all errors */
4922 	if (vdc_scsi_log_error)
4923 		log_error = B_TRUE;
4924 
4925 	if (log_error) {
4926 		cmn_err(CE_WARN, "%s (vdc%d):\tError for Command: 0x%x)\n",
4927 		    ddi_pathname(vdc->dip, path_str), vdc->instance,
4928 		    GETCMD(VD_SCSI_DATA_CDB(vd_scsi)));
4929 	}
4930 
4931 	/* default returned value */
4932 	rv = EIO;
4933 
4934 	switch (vd_scsi->cmd_status) {
4935 
4936 	case STATUS_CHECK:
4937 	case STATUS_TERMINATED:
4938 		if (log_error)
4939 			cmn_err(CE_CONT, "\tCheck Condition Error\n");
4940 
4941 		/* check sense buffer */
4942 		if (vd_scsi->sense_len == 0 ||
4943 		    vd_scsi->sense_status != STATUS_GOOD) {
4944 			if (log_error)
4945 				cmn_err(CE_CONT, "\tNo Sense Data Available\n");
4946 			break;
4947 		}
4948 
4949 		sense = VD_SCSI_DATA_SENSE(vd_scsi);
4950 
4951 		if (log_error) {
4952 			cmn_err(CE_CONT, "\tSense Key:  0x%x\n"
4953 			    "\tASC: 0x%x, ASCQ: 0x%x\n",
4954 			    scsi_sense_key((uint8_t *)sense),
4955 			    scsi_sense_asc((uint8_t *)sense),
4956 			    scsi_sense_ascq((uint8_t *)sense));
4957 		}
4958 
4959 		if (scsi_sense_key((uint8_t *)sense) == KEY_ILLEGAL_REQUEST)
4960 			rv = ENOTSUP;
4961 		break;
4962 
4963 	case STATUS_BUSY:
4964 		if (log_error)
4965 			cmn_err(CE_NOTE, "\tDevice Busy\n");
4966 		break;
4967 
4968 	case STATUS_RESERVATION_CONFLICT:
4969 		/*
4970 		 * If the command was PERSISTENT_RESERVATION_[IN|OUT] then
4971 		 * reservation conflict could be due to various reasons like
4972 		 * incorrect keys, not registered or not reserved etc. So,
4973 		 * we should not panic in that case.
4974 		 */
4975 		cdb = VD_SCSI_DATA_CDB(vd_scsi);
4976 		if (vdc->failfast_interval != 0 &&
4977 		    cdb->scc_cmd != SCMD_PERSISTENT_RESERVE_IN &&
4978 		    cdb->scc_cmd != SCMD_PERSISTENT_RESERVE_OUT) {
4979 			/* failfast is enabled so we have to panic */
4980 			(void) snprintf(panic_str, sizeof (panic_str),
4981 			    VDC_RESV_CONFLICT_FMT_STR "%s",
4982 			    ddi_pathname(vdc->dip, path_str));
4983 			panic(panic_str);
4984 		}
4985 		if (log_error)
4986 			cmn_err(CE_NOTE, "\tReservation Conflict\n");
4987 		rv = EACCES;
4988 		break;
4989 
4990 	case STATUS_QFULL:
4991 		if (log_error)
4992 			cmn_err(CE_NOTE, "\tQueue Full\n");
4993 		break;
4994 
4995 	case STATUS_MET:
4996 	case STATUS_INTERMEDIATE:
4997 	case STATUS_SCSI2:
4998 	case STATUS_INTERMEDIATE_MET:
4999 	case STATUS_ACA_ACTIVE:
5000 		if (log_error)
5001 			cmn_err(CE_CONT,
5002 			    "\tUnexpected SCSI status received: 0x%x\n",
5003 			    vd_scsi->cmd_status);
5004 		break;
5005 
5006 	default:
5007 		if (log_error)
5008 			cmn_err(CE_CONT,
5009 			    "\tInvalid SCSI status received: 0x%x\n",
5010 			    vd_scsi->cmd_status);
5011 		break;
5012 	}
5013 
5014 	return (rv);
5015 }
5016 
5017 /*
5018  * Implemented the USCSICMD uscsi(7I) ioctl. This ioctl is converted to
5019  * a VD_OP_SCSICMD operation which is sent to the vdisk server. If a SCSI
5020  * reset is requested (i.e. a flag USCSI_RESET* is set) then the ioctl is
5021  * converted to a VD_OP_RESET operation.
5022  */
5023 static int
5024 vdc_uscsi_cmd(vdc_t *vdc, caddr_t arg, int mode)
5025 {
5026 	struct uscsi_cmd 	uscsi;
5027 	struct uscsi_cmd32	uscsi32;
5028 	vd_scsi_t 		*vd_scsi;
5029 	int 			vd_scsi_len;
5030 	union scsi_cdb		*cdb;
5031 	struct scsi_extended_sense *sense;
5032 	char 			*datain, *dataout;
5033 	size_t			cdb_len, datain_len, dataout_len, sense_len;
5034 	int 			rv;
5035 
5036 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5037 		if (ddi_copyin(arg, &uscsi32, sizeof (struct uscsi_cmd32),
5038 		    mode) != 0)
5039 			return (EFAULT);
5040 		uscsi_cmd32touscsi_cmd((&uscsi32), (&uscsi));
5041 	} else {
5042 		if (ddi_copyin(arg, &uscsi, sizeof (struct uscsi_cmd),
5043 		    mode) != 0)
5044 			return (EFAULT);
5045 	}
5046 
5047 	/* a uscsi reset is converted to a VD_OP_RESET operation */
5048 	if (uscsi.uscsi_flags & (USCSI_RESET | USCSI_RESET_LUN |
5049 	    USCSI_RESET_ALL)) {
5050 		rv = vdc_do_sync_op(vdc, VD_OP_RESET, NULL, 0, 0, 0, CB_SYNC,
5051 		    (void *)(uint64_t)mode, VIO_both_dir, B_TRUE);
5052 		return (rv);
5053 	}
5054 
5055 	/* cdb buffer length */
5056 	cdb_len = uscsi.uscsi_cdblen;
5057 
5058 	/* data in and out buffers length */
5059 	if (uscsi.uscsi_flags & USCSI_READ) {
5060 		datain_len = uscsi.uscsi_buflen;
5061 		dataout_len = 0;
5062 	} else {
5063 		datain_len = 0;
5064 		dataout_len = uscsi.uscsi_buflen;
5065 	}
5066 
5067 	/* sense buffer length */
5068 	if (uscsi.uscsi_flags & USCSI_RQENABLE)
5069 		sense_len = uscsi.uscsi_rqlen;
5070 	else
5071 		sense_len = 0;
5072 
5073 	/* allocate buffer for the VD_SCSICMD_OP operation */
5074 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, datain_len, dataout_len,
5075 	    &vd_scsi_len);
5076 
5077 	/*
5078 	 * The documentation of USCSI_ISOLATE and USCSI_DIAGNOSE is very vague,
5079 	 * but basically they prevent a SCSI command from being retried in case
5080 	 * of an error.
5081 	 */
5082 	if ((uscsi.uscsi_flags & USCSI_ISOLATE) ||
5083 	    (uscsi.uscsi_flags & USCSI_DIAGNOSE))
5084 		vd_scsi->options |= VD_SCSI_OPT_NORETRY;
5085 
5086 	/* set task attribute */
5087 	if (uscsi.uscsi_flags & USCSI_NOTAG) {
5088 		vd_scsi->task_attribute = 0;
5089 	} else {
5090 		if (uscsi.uscsi_flags & USCSI_HEAD)
5091 			vd_scsi->task_attribute = VD_SCSI_TASK_ACA;
5092 		else if (uscsi.uscsi_flags & USCSI_HTAG)
5093 			vd_scsi->task_attribute = VD_SCSI_TASK_HQUEUE;
5094 		else if (uscsi.uscsi_flags & USCSI_OTAG)
5095 			vd_scsi->task_attribute = VD_SCSI_TASK_ORDERED;
5096 		else
5097 			vd_scsi->task_attribute = 0;
5098 	}
5099 
5100 	/* set timeout */
5101 	vd_scsi->timeout = uscsi.uscsi_timeout;
5102 
5103 	/* copy-in cdb data */
5104 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5105 	if (ddi_copyin(uscsi.uscsi_cdb, cdb, cdb_len, mode) != 0) {
5106 		rv = EFAULT;
5107 		goto done;
5108 	}
5109 
5110 	/* keep a pointer to the sense buffer */
5111 	sense = VD_SCSI_DATA_SENSE(vd_scsi);
5112 
5113 	/* keep a pointer to the data-in buffer */
5114 	datain = (char *)VD_SCSI_DATA_IN(vd_scsi);
5115 
5116 	/* copy-in request data to the data-out buffer */
5117 	dataout = (char *)VD_SCSI_DATA_OUT(vd_scsi);
5118 	if (!(uscsi.uscsi_flags & USCSI_READ)) {
5119 		if (ddi_copyin(uscsi.uscsi_bufaddr, dataout, dataout_len,
5120 		    mode)) {
5121 			rv = EFAULT;
5122 			goto done;
5123 		}
5124 	}
5125 
5126 	/* submit the request */
5127 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5128 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5129 
5130 	if (rv != 0)
5131 		goto done;
5132 
5133 	/* update scsi status */
5134 	uscsi.uscsi_status = vd_scsi->cmd_status;
5135 
5136 	/* update sense data */
5137 	if ((uscsi.uscsi_flags & USCSI_RQENABLE) &&
5138 	    (uscsi.uscsi_status == STATUS_CHECK ||
5139 	    uscsi.uscsi_status == STATUS_TERMINATED)) {
5140 
5141 		uscsi.uscsi_rqstatus = vd_scsi->sense_status;
5142 
5143 		if (uscsi.uscsi_rqstatus == STATUS_GOOD) {
5144 			uscsi.uscsi_rqresid = uscsi.uscsi_rqlen -
5145 			    vd_scsi->sense_len;
5146 			if (ddi_copyout(sense, uscsi.uscsi_rqbuf,
5147 			    vd_scsi->sense_len, mode) != 0) {
5148 				rv = EFAULT;
5149 				goto done;
5150 			}
5151 		}
5152 	}
5153 
5154 	/* update request data */
5155 	if (uscsi.uscsi_status == STATUS_GOOD) {
5156 		if (uscsi.uscsi_flags & USCSI_READ) {
5157 			uscsi.uscsi_resid = uscsi.uscsi_buflen -
5158 			    vd_scsi->datain_len;
5159 			if (ddi_copyout(datain, uscsi.uscsi_bufaddr,
5160 			    vd_scsi->datain_len, mode) != 0) {
5161 				rv = EFAULT;
5162 				goto done;
5163 			}
5164 		} else {
5165 			uscsi.uscsi_resid = uscsi.uscsi_buflen -
5166 			    vd_scsi->dataout_len;
5167 		}
5168 	}
5169 
5170 	/* copy-out result */
5171 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5172 		uscsi_cmdtouscsi_cmd32((&uscsi), (&uscsi32));
5173 		if (ddi_copyout(&uscsi32, arg, sizeof (struct uscsi_cmd32),
5174 		    mode) != 0) {
5175 			rv = EFAULT;
5176 			goto done;
5177 		}
5178 	} else {
5179 		if (ddi_copyout(&uscsi, arg, sizeof (struct uscsi_cmd),
5180 		    mode) != 0) {
5181 			rv = EFAULT;
5182 			goto done;
5183 		}
5184 	}
5185 
5186 	/* get the return code from the SCSI command status */
5187 	rv = vdc_scsi_status(vdc, vd_scsi,
5188 	    !(uscsi.uscsi_flags & USCSI_SILENT));
5189 
5190 done:
5191 	kmem_free(vd_scsi, vd_scsi_len);
5192 	return (rv);
5193 }
5194 
5195 /*
5196  * Create a VD_OP_SCSICMD buffer for a SCSI PERSISTENT IN command.
5197  *
5198  * Arguments:
5199  *	cmd		- SCSI PERSISTENT IN command
5200  *	len		- length of the SCSI input buffer
5201  *	vd_scsi_len	- return the length of the allocated buffer
5202  *
5203  * Returned Value:
5204  *	a pointer to the allocated VD_OP_SCSICMD buffer.
5205  */
5206 static vd_scsi_t *
5207 vdc_scsi_alloc_persistent_in(uchar_t cmd, int len, int *vd_scsi_len)
5208 {
5209 	int cdb_len, sense_len, datain_len, dataout_len;
5210 	vd_scsi_t *vd_scsi;
5211 	union scsi_cdb *cdb;
5212 
5213 	cdb_len = CDB_GROUP1;
5214 	sense_len = sizeof (struct scsi_extended_sense);
5215 	datain_len = len;
5216 	dataout_len = 0;
5217 
5218 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, datain_len, dataout_len,
5219 	    vd_scsi_len);
5220 
5221 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5222 
5223 	/* set cdb */
5224 	cdb->scc_cmd = SCMD_PERSISTENT_RESERVE_IN;
5225 	cdb->cdb_opaque[1] = cmd;
5226 	FORMG1COUNT(cdb, datain_len);
5227 
5228 	vd_scsi->timeout = vdc_scsi_timeout;
5229 
5230 	return (vd_scsi);
5231 }
5232 
5233 /*
5234  * Create a VD_OP_SCSICMD buffer for a SCSI PERSISTENT OUT command.
5235  *
5236  * Arguments:
5237  *	cmd		- SCSI PERSISTENT OUT command
5238  *	len		- length of the SCSI output buffer
5239  *	vd_scsi_len	- return the length of the allocated buffer
5240  *
5241  * Returned Code:
5242  *	a pointer to the allocated VD_OP_SCSICMD buffer.
5243  */
5244 static vd_scsi_t *
5245 vdc_scsi_alloc_persistent_out(uchar_t cmd, int len, int *vd_scsi_len)
5246 {
5247 	int cdb_len, sense_len, datain_len, dataout_len;
5248 	vd_scsi_t *vd_scsi;
5249 	union scsi_cdb *cdb;
5250 
5251 	cdb_len = CDB_GROUP1;
5252 	sense_len = sizeof (struct scsi_extended_sense);
5253 	datain_len = 0;
5254 	dataout_len = len;
5255 
5256 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, datain_len, dataout_len,
5257 	    vd_scsi_len);
5258 
5259 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5260 
5261 	/* set cdb */
5262 	cdb->scc_cmd = SCMD_PERSISTENT_RESERVE_OUT;
5263 	cdb->cdb_opaque[1] = cmd;
5264 	FORMG1COUNT(cdb, dataout_len);
5265 
5266 	vd_scsi->timeout = vdc_scsi_timeout;
5267 
5268 	return (vd_scsi);
5269 }
5270 
5271 /*
5272  * Implement the MHIOCGRP_INKEYS mhd(7i) ioctl. The ioctl is converted
5273  * to a SCSI PERSISTENT IN READ KEYS command which is sent to the vdisk
5274  * server with a VD_OP_SCSICMD operation.
5275  */
5276 static int
5277 vdc_mhd_inkeys(vdc_t *vdc, caddr_t arg, int mode)
5278 {
5279 	vd_scsi_t *vd_scsi;
5280 	mhioc_inkeys_t inkeys;
5281 	mhioc_key_list_t klist;
5282 	struct mhioc_inkeys32 inkeys32;
5283 	struct mhioc_key_list32 klist32;
5284 	sd_prin_readkeys_t *scsi_keys;
5285 	void *user_keys;
5286 	int vd_scsi_len;
5287 	int listsize, listlen, rv;
5288 
5289 	/* copyin arguments */
5290 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5291 		rv = ddi_copyin(arg, &inkeys32, sizeof (inkeys32), mode);
5292 		if (rv != 0)
5293 			return (EFAULT);
5294 
5295 		rv = ddi_copyin((caddr_t)(uintptr_t)inkeys32.li, &klist32,
5296 		    sizeof (klist32), mode);
5297 		if (rv != 0)
5298 			return (EFAULT);
5299 
5300 		listsize = klist32.listsize;
5301 	} else {
5302 		rv = ddi_copyin(arg, &inkeys, sizeof (inkeys), mode);
5303 		if (rv != 0)
5304 			return (EFAULT);
5305 
5306 		rv = ddi_copyin(inkeys.li, &klist, sizeof (klist), mode);
5307 		if (rv != 0)
5308 			return (EFAULT);
5309 
5310 		listsize = klist.listsize;
5311 	}
5312 
5313 	/* build SCSI VD_OP request */
5314 	vd_scsi = vdc_scsi_alloc_persistent_in(SD_READ_KEYS,
5315 	    sizeof (sd_prin_readkeys_t) - sizeof (caddr_t) +
5316 	    (sizeof (mhioc_resv_key_t) * listsize), &vd_scsi_len);
5317 
5318 	scsi_keys = (sd_prin_readkeys_t *)VD_SCSI_DATA_IN(vd_scsi);
5319 
5320 	/* submit the request */
5321 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5322 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5323 
5324 	if (rv != 0)
5325 		goto done;
5326 
5327 	listlen = scsi_keys->len / MHIOC_RESV_KEY_SIZE;
5328 
5329 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5330 		inkeys32.generation = scsi_keys->generation;
5331 		rv = ddi_copyout(&inkeys32, arg, sizeof (inkeys32), mode);
5332 		if (rv != 0) {
5333 			rv = EFAULT;
5334 			goto done;
5335 		}
5336 
5337 		klist32.listlen = listlen;
5338 		rv = ddi_copyout(&klist32, (caddr_t)(uintptr_t)inkeys32.li,
5339 		    sizeof (klist32), mode);
5340 		if (rv != 0) {
5341 			rv = EFAULT;
5342 			goto done;
5343 		}
5344 
5345 		user_keys = (caddr_t)(uintptr_t)klist32.list;
5346 	} else {
5347 		inkeys.generation = scsi_keys->generation;
5348 		rv = ddi_copyout(&inkeys, arg, sizeof (inkeys), mode);
5349 		if (rv != 0) {
5350 			rv = EFAULT;
5351 			goto done;
5352 		}
5353 
5354 		klist.listlen = listlen;
5355 		rv = ddi_copyout(&klist, inkeys.li, sizeof (klist), mode);
5356 		if (rv != 0) {
5357 			rv = EFAULT;
5358 			goto done;
5359 		}
5360 
5361 		user_keys = klist.list;
5362 	}
5363 
5364 	/* copy out keys */
5365 	if (listlen > 0 && listsize > 0) {
5366 		if (listsize < listlen)
5367 			listlen = listsize;
5368 		rv = ddi_copyout(&scsi_keys->keylist, user_keys,
5369 		    listlen * MHIOC_RESV_KEY_SIZE, mode);
5370 		if (rv != 0)
5371 			rv = EFAULT;
5372 	}
5373 
5374 	if (rv == 0)
5375 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
5376 
5377 done:
5378 	kmem_free(vd_scsi, vd_scsi_len);
5379 
5380 	return (rv);
5381 }
5382 
5383 /*
5384  * Implement the MHIOCGRP_INRESV mhd(7i) ioctl. The ioctl is converted
5385  * to a SCSI PERSISTENT IN READ RESERVATION command which is sent to
5386  * the vdisk server with a VD_OP_SCSICMD operation.
5387  */
5388 static int
5389 vdc_mhd_inresv(vdc_t *vdc, caddr_t arg, int mode)
5390 {
5391 	vd_scsi_t *vd_scsi;
5392 	mhioc_inresvs_t inresv;
5393 	mhioc_resv_desc_list_t rlist;
5394 	struct mhioc_inresvs32 inresv32;
5395 	struct mhioc_resv_desc_list32 rlist32;
5396 	mhioc_resv_desc_t mhd_resv;
5397 	sd_prin_readresv_t *scsi_resv;
5398 	sd_readresv_desc_t *resv;
5399 	mhioc_resv_desc_t *user_resv;
5400 	int vd_scsi_len;
5401 	int listsize, listlen, i, rv;
5402 
5403 	/* copyin arguments */
5404 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5405 		rv = ddi_copyin(arg, &inresv32, sizeof (inresv32), mode);
5406 		if (rv != 0)
5407 			return (EFAULT);
5408 
5409 		rv = ddi_copyin((caddr_t)(uintptr_t)inresv32.li, &rlist32,
5410 		    sizeof (rlist32), mode);
5411 		if (rv != 0)
5412 			return (EFAULT);
5413 
5414 		listsize = rlist32.listsize;
5415 	} else {
5416 		rv = ddi_copyin(arg, &inresv, sizeof (inresv), mode);
5417 		if (rv != 0)
5418 			return (EFAULT);
5419 
5420 		rv = ddi_copyin(inresv.li, &rlist, sizeof (rlist), mode);
5421 		if (rv != 0)
5422 			return (EFAULT);
5423 
5424 		listsize = rlist.listsize;
5425 	}
5426 
5427 	/* build SCSI VD_OP request */
5428 	vd_scsi = vdc_scsi_alloc_persistent_in(SD_READ_RESV,
5429 	    sizeof (sd_prin_readresv_t) - sizeof (caddr_t) +
5430 	    (SCSI3_RESV_DESC_LEN * listsize), &vd_scsi_len);
5431 
5432 	scsi_resv = (sd_prin_readresv_t *)VD_SCSI_DATA_IN(vd_scsi);
5433 
5434 	/* submit the request */
5435 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5436 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5437 
5438 	if (rv != 0)
5439 		goto done;
5440 
5441 	listlen = scsi_resv->len / SCSI3_RESV_DESC_LEN;
5442 
5443 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5444 		inresv32.generation = scsi_resv->generation;
5445 		rv = ddi_copyout(&inresv32, arg, sizeof (inresv32), mode);
5446 		if (rv != 0) {
5447 			rv = EFAULT;
5448 			goto done;
5449 		}
5450 
5451 		rlist32.listlen = listlen;
5452 		rv = ddi_copyout(&rlist32, (caddr_t)(uintptr_t)inresv32.li,
5453 		    sizeof (rlist32), mode);
5454 		if (rv != 0) {
5455 			rv = EFAULT;
5456 			goto done;
5457 		}
5458 
5459 		user_resv = (mhioc_resv_desc_t *)(uintptr_t)rlist32.list;
5460 	} else {
5461 		inresv.generation = scsi_resv->generation;
5462 		rv = ddi_copyout(&inresv, arg, sizeof (inresv), mode);
5463 		if (rv != 0) {
5464 			rv = EFAULT;
5465 			goto done;
5466 		}
5467 
5468 		rlist.listlen = listlen;
5469 		rv = ddi_copyout(&rlist, inresv.li, sizeof (rlist), mode);
5470 		if (rv != 0) {
5471 			rv = EFAULT;
5472 			goto done;
5473 		}
5474 
5475 		user_resv = rlist.list;
5476 	}
5477 
5478 	/* copy out reservations */
5479 	if (listsize > 0 && listlen > 0) {
5480 		if (listsize < listlen)
5481 			listlen = listsize;
5482 		resv = (sd_readresv_desc_t *)&scsi_resv->readresv_desc;
5483 
5484 		for (i = 0; i < listlen; i++) {
5485 			mhd_resv.type = resv->type;
5486 			mhd_resv.scope = resv->scope;
5487 			mhd_resv.scope_specific_addr =
5488 			    BE_32(resv->scope_specific_addr);
5489 			bcopy(&resv->resvkey, &mhd_resv.key,
5490 			    MHIOC_RESV_KEY_SIZE);
5491 
5492 			rv = ddi_copyout(&mhd_resv, user_resv,
5493 			    sizeof (mhd_resv), mode);
5494 			if (rv != 0) {
5495 				rv = EFAULT;
5496 				goto done;
5497 			}
5498 			resv++;
5499 			user_resv++;
5500 		}
5501 	}
5502 
5503 	if (rv == 0)
5504 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
5505 
5506 done:
5507 	kmem_free(vd_scsi, vd_scsi_len);
5508 	return (rv);
5509 }
5510 
5511 /*
5512  * Implement the MHIOCGRP_REGISTER mhd(7i) ioctl. The ioctl is converted
5513  * to a SCSI PERSISTENT OUT REGISTER command which is sent to the vdisk
5514  * server with a VD_OP_SCSICMD operation.
5515  */
5516 static int
5517 vdc_mhd_register(vdc_t *vdc, caddr_t arg, int mode)
5518 {
5519 	vd_scsi_t *vd_scsi;
5520 	sd_prout_t *scsi_prout;
5521 	mhioc_register_t mhd_reg;
5522 	int vd_scsi_len, rv;
5523 
5524 	/* copyin arguments */
5525 	rv = ddi_copyin(arg, &mhd_reg, sizeof (mhd_reg), mode);
5526 	if (rv != 0)
5527 		return (EFAULT);
5528 
5529 	/* build SCSI VD_OP request */
5530 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_REGISTER,
5531 	    sizeof (sd_prout_t), &vd_scsi_len);
5532 
5533 	/* set parameters */
5534 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
5535 	bcopy(mhd_reg.oldkey.key, scsi_prout->res_key, MHIOC_RESV_KEY_SIZE);
5536 	bcopy(mhd_reg.newkey.key, scsi_prout->service_key, MHIOC_RESV_KEY_SIZE);
5537 	scsi_prout->aptpl = (uchar_t)mhd_reg.aptpl;
5538 
5539 	/* submit the request */
5540 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5541 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5542 
5543 	if (rv == 0)
5544 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
5545 
5546 	kmem_free(vd_scsi, vd_scsi_len);
5547 	return (rv);
5548 }
5549 
5550 /*
5551  * Implement the MHIOCGRP_RESERVE mhd(7i) ioctl. The ioctl is converted
5552  * to a SCSI PERSISTENT OUT RESERVE command which is sent to the vdisk
5553  * server with a VD_OP_SCSICMD operation.
5554  */
5555 static int
5556 vdc_mhd_reserve(vdc_t *vdc, caddr_t arg, int mode)
5557 {
5558 	union scsi_cdb *cdb;
5559 	vd_scsi_t *vd_scsi;
5560 	sd_prout_t *scsi_prout;
5561 	mhioc_resv_desc_t mhd_resv;
5562 	int vd_scsi_len, rv;
5563 
5564 	/* copyin arguments */
5565 	rv = ddi_copyin(arg, &mhd_resv, sizeof (mhd_resv), mode);
5566 	if (rv != 0)
5567 		return (EFAULT);
5568 
5569 	/* build SCSI VD_OP request */
5570 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_RESERVE,
5571 	    sizeof (sd_prout_t), &vd_scsi_len);
5572 
5573 	/* set parameters */
5574 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5575 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
5576 	bcopy(mhd_resv.key.key, scsi_prout->res_key, MHIOC_RESV_KEY_SIZE);
5577 	scsi_prout->scope_address = mhd_resv.scope_specific_addr;
5578 	cdb->cdb_opaque[2] = mhd_resv.type;
5579 
5580 	/* submit the request */
5581 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5582 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5583 
5584 	if (rv == 0)
5585 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
5586 
5587 	kmem_free(vd_scsi, vd_scsi_len);
5588 	return (rv);
5589 }
5590 
5591 /*
5592  * Implement the MHIOCGRP_PREEMPTANDABORT mhd(7i) ioctl. The ioctl is
5593  * converted to a SCSI PERSISTENT OUT PREEMPT AND ABORT command which
5594  * is sent to the vdisk server with a VD_OP_SCSICMD operation.
5595  */
5596 static int
5597 vdc_mhd_preemptabort(vdc_t *vdc, caddr_t arg, int mode)
5598 {
5599 	union scsi_cdb *cdb;
5600 	vd_scsi_t *vd_scsi;
5601 	sd_prout_t *scsi_prout;
5602 	mhioc_preemptandabort_t mhd_preempt;
5603 	int vd_scsi_len, rv;
5604 
5605 	/* copyin arguments */
5606 	rv = ddi_copyin(arg, &mhd_preempt, sizeof (mhd_preempt), mode);
5607 	if (rv != 0)
5608 		return (EFAULT);
5609 
5610 	/* build SCSI VD_OP request */
5611 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_PREEMPTANDABORT,
5612 	    sizeof (sd_prout_t), &vd_scsi_len);
5613 
5614 	/* set parameters */
5615 	vd_scsi->task_attribute = VD_SCSI_TASK_ACA;
5616 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5617 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
5618 	bcopy(mhd_preempt.resvdesc.key.key, scsi_prout->res_key,
5619 	    MHIOC_RESV_KEY_SIZE);
5620 	bcopy(mhd_preempt.victim_key.key, scsi_prout->service_key,
5621 	    MHIOC_RESV_KEY_SIZE);
5622 	scsi_prout->scope_address = mhd_preempt.resvdesc.scope_specific_addr;
5623 	cdb->cdb_opaque[2] = mhd_preempt.resvdesc.type;
5624 
5625 	/* submit the request */
5626 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5627 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5628 
5629 	if (rv == 0)
5630 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
5631 
5632 	kmem_free(vd_scsi, vd_scsi_len);
5633 	return (rv);
5634 }
5635 
5636 /*
5637  * Implement the MHIOCGRP_REGISTERANDIGNOREKEY mhd(7i) ioctl. The ioctl
5638  * is converted to a SCSI PERSISTENT OUT REGISTER AND IGNORE EXISTING KEY
5639  * command which is sent to the vdisk server with a VD_OP_SCSICMD operation.
5640  */
5641 static int
5642 vdc_mhd_registerignore(vdc_t *vdc, caddr_t arg, int mode)
5643 {
5644 	vd_scsi_t *vd_scsi;
5645 	sd_prout_t *scsi_prout;
5646 	mhioc_registerandignorekey_t mhd_regi;
5647 	int vd_scsi_len, rv;
5648 
5649 	/* copyin arguments */
5650 	rv = ddi_copyin(arg, &mhd_regi, sizeof (mhd_regi), mode);
5651 	if (rv != 0)
5652 		return (EFAULT);
5653 
5654 	/* build SCSI VD_OP request */
5655 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_REGISTERANDIGNOREKEY,
5656 	    sizeof (sd_prout_t), &vd_scsi_len);
5657 
5658 	/* set parameters */
5659 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
5660 	bcopy(mhd_regi.newkey.key, scsi_prout->service_key,
5661 	    MHIOC_RESV_KEY_SIZE);
5662 	scsi_prout->aptpl = (uchar_t)mhd_regi.aptpl;
5663 
5664 	/* submit the request */
5665 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5666 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5667 
5668 	if (rv == 0)
5669 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
5670 
5671 	kmem_free(vd_scsi, vd_scsi_len);
5672 	return (rv);
5673 }
5674 
5675 /*
5676  * This function is used by the failfast mechanism to send a SCSI command
5677  * to check for reservation conflict.
5678  */
5679 static int
5680 vdc_failfast_scsi_cmd(vdc_t *vdc, uchar_t scmd)
5681 {
5682 	int cdb_len, sense_len, vd_scsi_len;
5683 	vd_scsi_t *vd_scsi;
5684 	union scsi_cdb *cdb;
5685 	int rv;
5686 
5687 	ASSERT(scmd == SCMD_TEST_UNIT_READY || scmd == SCMD_WRITE_G1);
5688 
5689 	if (scmd == SCMD_WRITE_G1)
5690 		cdb_len = CDB_GROUP1;
5691 	else
5692 		cdb_len = CDB_GROUP0;
5693 
5694 	sense_len = sizeof (struct scsi_extended_sense);
5695 
5696 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, 0, 0, &vd_scsi_len);
5697 
5698 	/* set cdb */
5699 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5700 	cdb->scc_cmd = scmd;
5701 
5702 	vd_scsi->timeout = vdc_scsi_timeout;
5703 
5704 	/*
5705 	 * Submit the request. The last argument has to be B_FALSE so that
5706 	 * vdc_do_sync_op does not loop checking for reservation conflict if
5707 	 * the operation returns an error.
5708 	 */
5709 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5710 	    0, 0, CB_SYNC, (void *)(uint64_t)FKIOCTL, VIO_both_dir, B_FALSE);
5711 
5712 	if (rv == 0)
5713 		(void) vdc_scsi_status(vdc, vd_scsi, B_FALSE);
5714 
5715 	kmem_free(vd_scsi, vd_scsi_len);
5716 	return (rv);
5717 }
5718 
5719 /*
5720  * This function is used by the failfast mechanism to check for reservation
5721  * conflict. It sends some SCSI commands which will fail with a reservation
5722  * conflict error if the system does not have access to the disk and this
5723  * will panic the system.
5724  *
5725  * Returned Code:
5726  *	0	- disk is accessible without reservation conflict error
5727  *	!= 0	- unable to check if disk is accessible
5728  */
5729 int
5730 vdc_failfast_check_resv(vdc_t *vdc)
5731 {
5732 	int failure = 0;
5733 
5734 	/*
5735 	 * Send a TEST UNIT READY command. The command will panic
5736 	 * the system if it fails with a reservation conflict.
5737 	 */
5738 	if (vdc_failfast_scsi_cmd(vdc, SCMD_TEST_UNIT_READY) != 0)
5739 		failure++;
5740 
5741 	/*
5742 	 * With SPC-3 compliant devices TEST UNIT READY will succeed on
5743 	 * a reserved device, so we also do a WRITE(10) of zero byte in
5744 	 * order to provoke a Reservation Conflict status on those newer
5745 	 * devices.
5746 	 */
5747 	if (vdc_failfast_scsi_cmd(vdc, SCMD_WRITE_G1) != 0)
5748 		failure++;
5749 
5750 	return (failure);
5751 }
5752 
5753 /*
5754  * Add a pending I/O to the failfast I/O queue. An I/O is added to this
5755  * queue when it has failed and failfast is enabled. Then we have to check
5756  * if it has failed because of a reservation conflict in which case we have
5757  * to panic the system.
5758  *
5759  * Async I/O should be queued with their block I/O data transfer structure
5760  * (buf). Sync I/O should be queued with buf = NULL.
5761  */
5762 static vdc_io_t *
5763 vdc_failfast_io_queue(vdc_t *vdc, struct buf *buf)
5764 {
5765 	vdc_io_t *vio;
5766 
5767 	ASSERT(MUTEX_HELD(&vdc->lock));
5768 
5769 	vio = kmem_alloc(sizeof (vdc_io_t), KM_SLEEP);
5770 	vio->vio_next = vdc->failfast_io_queue;
5771 	vio->vio_buf = buf;
5772 	vio->vio_qtime = ddi_get_lbolt();
5773 
5774 	vdc->failfast_io_queue = vio;
5775 
5776 	/* notify the failfast thread that a new I/O is queued */
5777 	cv_signal(&vdc->failfast_cv);
5778 
5779 	return (vio);
5780 }
5781 
5782 /*
5783  * Remove and complete I/O in the failfast I/O queue which have been
5784  * added after the indicated deadline. A deadline of 0 means that all
5785  * I/O have to be unqueued and marked as completed.
5786  */
5787 static void
5788 vdc_failfast_io_unqueue(vdc_t *vdc, clock_t deadline)
5789 {
5790 	vdc_io_t *vio, *vio_tmp;
5791 
5792 	ASSERT(MUTEX_HELD(&vdc->lock));
5793 
5794 	vio_tmp = NULL;
5795 	vio = vdc->failfast_io_queue;
5796 
5797 	if (deadline != 0) {
5798 		/*
5799 		 * Skip any io queued after the deadline. The failfast
5800 		 * I/O queue is ordered starting with the last I/O added
5801 		 * to the queue.
5802 		 */
5803 		while (vio != NULL && vio->vio_qtime > deadline) {
5804 			vio_tmp = vio;
5805 			vio = vio->vio_next;
5806 		}
5807 	}
5808 
5809 	if (vio == NULL)
5810 		/* nothing to unqueue */
5811 		return;
5812 
5813 	/* update the queue */
5814 	if (vio_tmp == NULL)
5815 		vdc->failfast_io_queue = NULL;
5816 	else
5817 		vio_tmp->vio_next = NULL;
5818 
5819 	/*
5820 	 * Complete unqueued I/O. Async I/O have a block I/O data transfer
5821 	 * structure (buf) and they are completed by calling biodone(). Sync
5822 	 * I/O do not have a buf and they are completed by setting the
5823 	 * vio_qtime to zero and signaling failfast_io_cv. In that case, the
5824 	 * thread waiting for the I/O to complete is responsible for freeing
5825 	 * the vio structure.
5826 	 */
5827 	while (vio != NULL) {
5828 		vio_tmp = vio->vio_next;
5829 		if (vio->vio_buf != NULL) {
5830 			biodone(vio->vio_buf);
5831 			kmem_free(vio, sizeof (vdc_io_t));
5832 		} else {
5833 			vio->vio_qtime = 0;
5834 		}
5835 		vio = vio_tmp;
5836 	}
5837 
5838 	cv_broadcast(&vdc->failfast_io_cv);
5839 }
5840 
5841 /*
5842  * Failfast Thread.
5843  *
5844  * While failfast is enabled, the failfast thread sends a TEST UNIT READY
5845  * and a zero size WRITE(10) SCSI commands on a regular basis to check that
5846  * we still have access to the disk. If a command fails with a RESERVATION
5847  * CONFLICT error then the system will immediatly panic.
5848  *
5849  * The failfast thread is also woken up when an I/O has failed. It then check
5850  * the access to the disk to ensure that the I/O failure was not due to a
5851  * reservation conflict.
5852  *
5853  * There is one failfast thread for each virtual disk for which failfast is
5854  * enabled. We could have only one thread sending requests for all disks but
5855  * this would need vdc to send asynchronous requests and to have callbacks to
5856  * process replies.
5857  */
5858 static void
5859 vdc_failfast_thread(void *arg)
5860 {
5861 	int status;
5862 	vdc_t *vdc = (vdc_t *)arg;
5863 	clock_t timeout, starttime;
5864 
5865 	mutex_enter(&vdc->lock);
5866 
5867 	while (vdc->failfast_interval != 0) {
5868 
5869 		starttime = ddi_get_lbolt();
5870 
5871 		mutex_exit(&vdc->lock);
5872 
5873 		/* check for reservation conflict */
5874 		status = vdc_failfast_check_resv(vdc);
5875 
5876 		mutex_enter(&vdc->lock);
5877 		/*
5878 		 * We have dropped the lock to send the SCSI command so we have
5879 		 * to check that failfast is still enabled.
5880 		 */
5881 		if (vdc->failfast_interval == 0)
5882 			break;
5883 
5884 		/*
5885 		 * If we have successfully check the disk access and there was
5886 		 * no reservation conflict then we can complete any I/O queued
5887 		 * before the last check.
5888 		 */
5889 		if (status == 0)
5890 			vdc_failfast_io_unqueue(vdc, starttime);
5891 
5892 		/* proceed again if some I/O are still in the queue */
5893 		if (vdc->failfast_io_queue != NULL)
5894 			continue;
5895 
5896 		timeout = ddi_get_lbolt() +
5897 		    drv_usectohz(vdc->failfast_interval);
5898 		(void) cv_timedwait(&vdc->failfast_cv, &vdc->lock, timeout);
5899 	}
5900 
5901 	/*
5902 	 * Failfast is being stop so we can complete any queued I/O.
5903 	 */
5904 	vdc_failfast_io_unqueue(vdc, 0);
5905 	vdc->failfast_thread = NULL;
5906 	mutex_exit(&vdc->lock);
5907 	thread_exit();
5908 }
5909 
5910 /*
5911  * Implement the MHIOCENFAILFAST mhd(7i) ioctl.
5912  */
5913 static int
5914 vdc_failfast(vdc_t *vdc, caddr_t arg, int mode)
5915 {
5916 	unsigned int mh_time;
5917 
5918 	if (ddi_copyin((void *)arg, &mh_time, sizeof (int), mode))
5919 		return (EFAULT);
5920 
5921 	mutex_enter(&vdc->lock);
5922 	if (mh_time != 0 && vdc->failfast_thread == NULL) {
5923 		vdc->failfast_thread = thread_create(NULL, 0,
5924 		    vdc_failfast_thread, vdc, 0, &p0, TS_RUN,
5925 		    v.v_maxsyspri - 2);
5926 	}
5927 
5928 	vdc->failfast_interval = mh_time * 1000;
5929 	cv_signal(&vdc->failfast_cv);
5930 	mutex_exit(&vdc->lock);
5931 
5932 	return (0);
5933 }
5934 
5935 /*
5936  * Implement the MHIOCTKOWN and MHIOCRELEASE mhd(7i) ioctls. These ioctls are
5937  * converted to VD_OP_SET_ACCESS operations.
5938  */
5939 static int
5940 vdc_access_set(vdc_t *vdc, uint64_t flags, int mode)
5941 {
5942 	int rv;
5943 
5944 	/* submit owership command request */
5945 	rv = vdc_do_sync_op(vdc, VD_OP_SET_ACCESS, (caddr_t)&flags,
5946 	    sizeof (uint64_t), 0, 0, CB_SYNC, (void *)(uint64_t)mode,
5947 	    VIO_both_dir, B_TRUE);
5948 
5949 	return (rv);
5950 }
5951 
5952 /*
5953  * Implement the MHIOCSTATUS mhd(7i) ioctl. This ioctl is converted to a
5954  * VD_OP_GET_ACCESS operation.
5955  */
5956 static int
5957 vdc_access_get(vdc_t *vdc, uint64_t *status, int mode)
5958 {
5959 	int rv;
5960 
5961 	/* submit owership command request */
5962 	rv = vdc_do_sync_op(vdc, VD_OP_GET_ACCESS, (caddr_t)status,
5963 	    sizeof (uint64_t), 0, 0, CB_SYNC, (void *)(uint64_t)mode,
5964 	    VIO_both_dir, B_TRUE);
5965 
5966 	return (rv);
5967 }
5968 
5969 /*
5970  * Disk Ownership Thread.
5971  *
5972  * When we have taken the ownership of a disk, this thread waits to be
5973  * notified when the LDC channel is reset so that it can recover the
5974  * ownership.
5975  *
5976  * Note that the thread handling the LDC reset (vdc_process_msg_thread())
5977  * can not be used to do the ownership recovery because it has to be
5978  * running to handle the reply message to the ownership operation.
5979  */
5980 static void
5981 vdc_ownership_thread(void *arg)
5982 {
5983 	vdc_t *vdc = (vdc_t *)arg;
5984 	clock_t timeout;
5985 	uint64_t status;
5986 
5987 	mutex_enter(&vdc->ownership_lock);
5988 	mutex_enter(&vdc->lock);
5989 
5990 	while (vdc->ownership & VDC_OWNERSHIP_WANTED) {
5991 
5992 		if ((vdc->ownership & VDC_OWNERSHIP_RESET) ||
5993 		    !(vdc->ownership & VDC_OWNERSHIP_GRANTED)) {
5994 			/*
5995 			 * There was a reset so the ownership has been lost,
5996 			 * try to recover. We do this without using the preempt
5997 			 * option so that we don't steal the ownership from
5998 			 * someone who has preempted us.
5999 			 */
6000 			DMSG(vdc, 0, "[%d] Ownership lost, recovering",
6001 			    vdc->instance);
6002 
6003 			vdc->ownership &= ~(VDC_OWNERSHIP_RESET |
6004 			    VDC_OWNERSHIP_GRANTED);
6005 
6006 			mutex_exit(&vdc->lock);
6007 
6008 			status = vdc_access_set(vdc, VD_ACCESS_SET_EXCLUSIVE |
6009 			    VD_ACCESS_SET_PRESERVE, FKIOCTL);
6010 
6011 			mutex_enter(&vdc->lock);
6012 
6013 			if (status == 0) {
6014 				DMSG(vdc, 0, "[%d] Ownership recovered",
6015 				    vdc->instance);
6016 				vdc->ownership |= VDC_OWNERSHIP_GRANTED;
6017 			} else {
6018 				DMSG(vdc, 0, "[%d] Fail to recover ownership",
6019 				    vdc->instance);
6020 			}
6021 
6022 		}
6023 
6024 		/*
6025 		 * If we have the ownership then we just wait for an event
6026 		 * to happen (LDC reset), otherwise we will retry to recover
6027 		 * after a delay.
6028 		 */
6029 		if (vdc->ownership & VDC_OWNERSHIP_GRANTED)
6030 			timeout = 0;
6031 		else
6032 			timeout = ddi_get_lbolt() +
6033 			    drv_usectohz(vdc_ownership_delay);
6034 
6035 		/* Release the ownership_lock and wait on the vdc lock */
6036 		mutex_exit(&vdc->ownership_lock);
6037 
6038 		if (timeout == 0)
6039 			(void) cv_wait(&vdc->ownership_cv, &vdc->lock);
6040 		else
6041 			(void) cv_timedwait(&vdc->ownership_cv,
6042 			    &vdc->lock, timeout);
6043 
6044 		mutex_exit(&vdc->lock);
6045 
6046 		mutex_enter(&vdc->ownership_lock);
6047 		mutex_enter(&vdc->lock);
6048 	}
6049 
6050 	vdc->ownership_thread = NULL;
6051 	mutex_exit(&vdc->lock);
6052 	mutex_exit(&vdc->ownership_lock);
6053 
6054 	thread_exit();
6055 }
6056 
6057 static void
6058 vdc_ownership_update(vdc_t *vdc, int ownership_flags)
6059 {
6060 	ASSERT(MUTEX_HELD(&vdc->ownership_lock));
6061 
6062 	mutex_enter(&vdc->lock);
6063 	vdc->ownership = ownership_flags;
6064 	if ((vdc->ownership & VDC_OWNERSHIP_WANTED) &&
6065 	    vdc->ownership_thread == NULL) {
6066 		/* start ownership thread */
6067 		vdc->ownership_thread = thread_create(NULL, 0,
6068 		    vdc_ownership_thread, vdc, 0, &p0, TS_RUN,
6069 		    v.v_maxsyspri - 2);
6070 	} else {
6071 		/* notify the ownership thread */
6072 		cv_signal(&vdc->ownership_cv);
6073 	}
6074 	mutex_exit(&vdc->lock);
6075 }
6076 
6077 /*
6078  * Get the size and the block size of a virtual disk from the vdisk server.
6079  * We need to use this operation when the vdisk_size attribute was not
6080  * available during the handshake with the vdisk server.
6081  */
6082 static int
6083 vdc_check_capacity(vdc_t *vdc)
6084 {
6085 	int rv = 0;
6086 	size_t alloc_len;
6087 	vd_capacity_t *vd_cap;
6088 
6089 	if (vdc->vdisk_size != 0)
6090 		return (0);
6091 
6092 	alloc_len = P2ROUNDUP(sizeof (vd_capacity_t), sizeof (uint64_t));
6093 
6094 	vd_cap = kmem_zalloc(alloc_len, KM_SLEEP);
6095 
6096 	rv = vdc_do_sync_op(vdc, VD_OP_GET_CAPACITY, (caddr_t)vd_cap, alloc_len,
6097 	    0, 0, CB_SYNC, (void *)(uint64_t)FKIOCTL, VIO_both_dir, B_TRUE);
6098 
6099 	if (rv == 0) {
6100 		if (vd_cap->vdisk_block_size != vdc->block_size ||
6101 		    vd_cap->vdisk_size == VD_SIZE_UNKNOWN ||
6102 		    vd_cap->vdisk_size == 0)
6103 			rv = EINVAL;
6104 		else
6105 			vdc->vdisk_size = vd_cap->vdisk_size;
6106 	}
6107 
6108 	kmem_free(vd_cap, alloc_len);
6109 	return (rv);
6110 }
6111 
6112 /*
6113  * This structure is used in the DKIO(7I) array below.
6114  */
6115 typedef struct vdc_dk_ioctl {
6116 	uint8_t		op;		/* VD_OP_XXX value */
6117 	int		cmd;		/* Solaris ioctl operation number */
6118 	size_t		nbytes;		/* size of structure to be copied */
6119 
6120 	/* function to convert between vDisk and Solaris structure formats */
6121 	int	(*convert)(vdc_t *vdc, void *vd_buf, void *ioctl_arg,
6122 	    int mode, int dir);
6123 } vdc_dk_ioctl_t;
6124 
6125 /*
6126  * Subset of DKIO(7I) operations currently supported
6127  */
6128 static vdc_dk_ioctl_t	dk_ioctl[] = {
6129 	{VD_OP_FLUSH,		DKIOCFLUSHWRITECACHE,	0,
6130 		vdc_null_copy_func},
6131 	{VD_OP_GET_WCE,		DKIOCGETWCE,		sizeof (int),
6132 		vdc_get_wce_convert},
6133 	{VD_OP_SET_WCE,		DKIOCSETWCE,		sizeof (int),
6134 		vdc_set_wce_convert},
6135 	{VD_OP_GET_VTOC,	DKIOCGVTOC,		sizeof (vd_vtoc_t),
6136 		vdc_get_vtoc_convert},
6137 	{VD_OP_SET_VTOC,	DKIOCSVTOC,		sizeof (vd_vtoc_t),
6138 		vdc_set_vtoc_convert},
6139 	{VD_OP_GET_DISKGEOM,	DKIOCGGEOM,		sizeof (vd_geom_t),
6140 		vdc_get_geom_convert},
6141 	{VD_OP_GET_DISKGEOM,	DKIOCG_PHYGEOM,		sizeof (vd_geom_t),
6142 		vdc_get_geom_convert},
6143 	{VD_OP_GET_DISKGEOM, 	DKIOCG_VIRTGEOM,	sizeof (vd_geom_t),
6144 		vdc_get_geom_convert},
6145 	{VD_OP_SET_DISKGEOM,	DKIOCSGEOM,		sizeof (vd_geom_t),
6146 		vdc_set_geom_convert},
6147 	{VD_OP_GET_EFI,		DKIOCGETEFI,		0,
6148 		vdc_get_efi_convert},
6149 	{VD_OP_SET_EFI,		DKIOCSETEFI,		0,
6150 		vdc_set_efi_convert},
6151 
6152 	/* DIOCTL_RWCMD is converted to a read or a write */
6153 	{0, DIOCTL_RWCMD,  sizeof (struct dadkio_rwcmd), NULL},
6154 
6155 	/* mhd(7I) non-shared multihost disks ioctls */
6156 	{0, MHIOCTKOWN,				0, vdc_null_copy_func},
6157 	{0, MHIOCRELEASE,			0, vdc_null_copy_func},
6158 	{0, MHIOCSTATUS,			0, vdc_null_copy_func},
6159 	{0, MHIOCQRESERVE,			0, vdc_null_copy_func},
6160 
6161 	/* mhd(7I) shared multihost disks ioctls */
6162 	{0, MHIOCGRP_INKEYS,			0, vdc_null_copy_func},
6163 	{0, MHIOCGRP_INRESV,			0, vdc_null_copy_func},
6164 	{0, MHIOCGRP_REGISTER,			0, vdc_null_copy_func},
6165 	{0, MHIOCGRP_RESERVE, 			0, vdc_null_copy_func},
6166 	{0, MHIOCGRP_PREEMPTANDABORT,		0, vdc_null_copy_func},
6167 	{0, MHIOCGRP_REGISTERANDIGNOREKEY,	0, vdc_null_copy_func},
6168 
6169 	/* mhd(7I) failfast ioctl */
6170 	{0, MHIOCENFAILFAST,			0, vdc_null_copy_func},
6171 
6172 	/*
6173 	 * These particular ioctls are not sent to the server - vdc fakes up
6174 	 * the necessary info.
6175 	 */
6176 	{0, DKIOCINFO, sizeof (struct dk_cinfo), vdc_null_copy_func},
6177 	{0, DKIOCGMEDIAINFO, sizeof (struct dk_minfo), vdc_null_copy_func},
6178 	{0, USCSICMD,	sizeof (struct uscsi_cmd), vdc_null_copy_func},
6179 	{0, DKIOCGAPART, 0, vdc_null_copy_func },
6180 	{0, DKIOCREMOVABLE, 0, vdc_null_copy_func},
6181 	{0, CDROMREADOFFSET, 0, vdc_null_copy_func}
6182 };
6183 
6184 /*
6185  * This function handles ioctl requests from the vd_efi_alloc_and_read()
6186  * function and forward them to the vdisk.
6187  */
6188 static int
6189 vd_process_efi_ioctl(void *vdisk, int cmd, uintptr_t arg)
6190 {
6191 	vdc_t *vdc = (vdc_t *)vdisk;
6192 	dev_t dev;
6193 	int rval;
6194 
6195 	dev = makedevice(ddi_driver_major(vdc->dip),
6196 	    VD_MAKE_DEV(vdc->instance, 0));
6197 
6198 	return (vd_process_ioctl(dev, cmd, (caddr_t)arg, FKIOCTL, &rval));
6199 }
6200 
6201 /*
6202  * Function:
6203  *	vd_process_ioctl()
6204  *
6205  * Description:
6206  *	This routine processes disk specific ioctl calls
6207  *
6208  * Arguments:
6209  *	dev	- the device number
6210  *	cmd	- the operation [dkio(7I)] to be processed
6211  *	arg	- pointer to user provided structure
6212  *		  (contains data to be set or reference parameter for get)
6213  *	mode	- bit flag, indicating open settings, 32/64 bit type, etc
6214  *	rvalp	- pointer to return value for calling process.
6215  *
6216  * Return Code:
6217  *	0
6218  *	EFAULT
6219  *	ENXIO
6220  *	EIO
6221  *	ENOTSUP
6222  */
6223 static int
6224 vd_process_ioctl(dev_t dev, int cmd, caddr_t arg, int mode, int *rvalp)
6225 {
6226 	int		instance = VDCUNIT(dev);
6227 	vdc_t		*vdc = NULL;
6228 	int		rv = -1;
6229 	int		idx = 0;		/* index into dk_ioctl[] */
6230 	size_t		len = 0;		/* #bytes to send to vds */
6231 	size_t		alloc_len = 0;		/* #bytes to allocate mem for */
6232 	caddr_t		mem_p = NULL;
6233 	size_t		nioctls = (sizeof (dk_ioctl)) / (sizeof (dk_ioctl[0]));
6234 	vdc_dk_ioctl_t	*iop;
6235 
6236 	vdc = ddi_get_soft_state(vdc_state, instance);
6237 	if (vdc == NULL) {
6238 		cmn_err(CE_NOTE, "![%d] Could not get soft state structure",
6239 		    instance);
6240 		return (ENXIO);
6241 	}
6242 
6243 	DMSG(vdc, 0, "[%d] Processing ioctl(%x) for dev %lx : model %x\n",
6244 	    instance, cmd, dev, ddi_model_convert_from(mode & FMODELS));
6245 
6246 	if (rvalp != NULL) {
6247 		/* the return value of the ioctl is 0 by default */
6248 		*rvalp = 0;
6249 	}
6250 
6251 	/*
6252 	 * Validate the ioctl operation to be performed.
6253 	 *
6254 	 * If we have looped through the array without finding a match then we
6255 	 * don't support this ioctl.
6256 	 */
6257 	for (idx = 0; idx < nioctls; idx++) {
6258 		if (cmd == dk_ioctl[idx].cmd)
6259 			break;
6260 	}
6261 
6262 	if (idx >= nioctls) {
6263 		DMSG(vdc, 0, "[%d] Unsupported ioctl (0x%x)\n",
6264 		    vdc->instance, cmd);
6265 		return (ENOTSUP);
6266 	}
6267 
6268 	iop = &(dk_ioctl[idx]);
6269 
6270 	if (cmd == DKIOCGETEFI || cmd == DKIOCSETEFI) {
6271 		/* size is not fixed for EFI ioctls, it depends on ioctl arg */
6272 		dk_efi_t	dk_efi;
6273 
6274 		rv = ddi_copyin(arg, &dk_efi, sizeof (dk_efi_t), mode);
6275 		if (rv != 0)
6276 			return (EFAULT);
6277 
6278 		len = sizeof (vd_efi_t) - 1 + dk_efi.dki_length;
6279 	} else {
6280 		len = iop->nbytes;
6281 	}
6282 
6283 	/* check if the ioctl is applicable */
6284 	switch (cmd) {
6285 	case CDROMREADOFFSET:
6286 	case DKIOCREMOVABLE:
6287 		return (ENOTTY);
6288 
6289 	case USCSICMD:
6290 	case MHIOCTKOWN:
6291 	case MHIOCSTATUS:
6292 	case MHIOCQRESERVE:
6293 	case MHIOCRELEASE:
6294 	case MHIOCGRP_INKEYS:
6295 	case MHIOCGRP_INRESV:
6296 	case MHIOCGRP_REGISTER:
6297 	case MHIOCGRP_RESERVE:
6298 	case MHIOCGRP_PREEMPTANDABORT:
6299 	case MHIOCGRP_REGISTERANDIGNOREKEY:
6300 	case MHIOCENFAILFAST:
6301 		if (vdc->cinfo == NULL)
6302 			return (ENXIO);
6303 		if (vdc->cinfo->dki_ctype != DKC_SCSI_CCS)
6304 			return (ENOTTY);
6305 		break;
6306 
6307 	case DIOCTL_RWCMD:
6308 		if (vdc->cinfo == NULL)
6309 			return (ENXIO);
6310 		if (vdc->cinfo->dki_ctype != DKC_DIRECT)
6311 			return (ENOTTY);
6312 		break;
6313 
6314 	case DKIOCINFO:
6315 		if (vdc->cinfo == NULL)
6316 			return (ENXIO);
6317 		break;
6318 
6319 	case DKIOCGMEDIAINFO:
6320 		if (vdc->minfo == NULL)
6321 			return (ENXIO);
6322 		if (vdc_check_capacity(vdc) != 0)
6323 			/* disk capacity is not available */
6324 			return (EIO);
6325 		break;
6326 	}
6327 
6328 	/*
6329 	 * Deal with ioctls which require a processing different than
6330 	 * converting ioctl arguments and sending a corresponding
6331 	 * VD operation.
6332 	 */
6333 	switch (cmd) {
6334 
6335 	case USCSICMD:
6336 	{
6337 		return (vdc_uscsi_cmd(vdc, arg, mode));
6338 	}
6339 
6340 	case MHIOCTKOWN:
6341 	{
6342 		mutex_enter(&vdc->ownership_lock);
6343 		/*
6344 		 * We have to set VDC_OWNERSHIP_WANTED now so that the ownership
6345 		 * can be flagged with VDC_OWNERSHIP_RESET if the LDC is reset
6346 		 * while we are processing the ioctl.
6347 		 */
6348 		vdc_ownership_update(vdc, VDC_OWNERSHIP_WANTED);
6349 
6350 		rv = vdc_access_set(vdc, VD_ACCESS_SET_EXCLUSIVE |
6351 		    VD_ACCESS_SET_PREEMPT | VD_ACCESS_SET_PRESERVE, mode);
6352 		if (rv == 0) {
6353 			vdc_ownership_update(vdc, VDC_OWNERSHIP_WANTED |
6354 			    VDC_OWNERSHIP_GRANTED);
6355 		} else {
6356 			vdc_ownership_update(vdc, VDC_OWNERSHIP_NONE);
6357 		}
6358 		mutex_exit(&vdc->ownership_lock);
6359 		return (rv);
6360 	}
6361 
6362 	case MHIOCRELEASE:
6363 	{
6364 		mutex_enter(&vdc->ownership_lock);
6365 		rv = vdc_access_set(vdc, VD_ACCESS_SET_CLEAR, mode);
6366 		if (rv == 0) {
6367 			vdc_ownership_update(vdc, VDC_OWNERSHIP_NONE);
6368 		}
6369 		mutex_exit(&vdc->ownership_lock);
6370 		return (rv);
6371 	}
6372 
6373 	case MHIOCSTATUS:
6374 	{
6375 		uint64_t status;
6376 
6377 		rv = vdc_access_get(vdc, &status, mode);
6378 		if (rv == 0 && rvalp != NULL)
6379 			*rvalp = (status & VD_ACCESS_ALLOWED)? 0 : 1;
6380 		return (rv);
6381 	}
6382 
6383 	case MHIOCQRESERVE:
6384 	{
6385 		rv = vdc_access_set(vdc, VD_ACCESS_SET_EXCLUSIVE, mode);
6386 		return (rv);
6387 	}
6388 
6389 	case MHIOCGRP_INKEYS:
6390 	{
6391 		return (vdc_mhd_inkeys(vdc, arg, mode));
6392 	}
6393 
6394 	case MHIOCGRP_INRESV:
6395 	{
6396 		return (vdc_mhd_inresv(vdc, arg, mode));
6397 	}
6398 
6399 	case MHIOCGRP_REGISTER:
6400 	{
6401 		return (vdc_mhd_register(vdc, arg, mode));
6402 	}
6403 
6404 	case MHIOCGRP_RESERVE:
6405 	{
6406 		return (vdc_mhd_reserve(vdc, arg, mode));
6407 	}
6408 
6409 	case MHIOCGRP_PREEMPTANDABORT:
6410 	{
6411 		return (vdc_mhd_preemptabort(vdc, arg, mode));
6412 	}
6413 
6414 	case MHIOCGRP_REGISTERANDIGNOREKEY:
6415 	{
6416 		return (vdc_mhd_registerignore(vdc, arg, mode));
6417 	}
6418 
6419 	case MHIOCENFAILFAST:
6420 	{
6421 		rv = vdc_failfast(vdc, arg, mode);
6422 		return (rv);
6423 	}
6424 
6425 	case DIOCTL_RWCMD:
6426 	{
6427 		return (vdc_dioctl_rwcmd(dev, arg, mode));
6428 	}
6429 
6430 	case DKIOCGAPART:
6431 	{
6432 		return (vdc_dkio_get_partition(vdc, arg, mode));
6433 	}
6434 
6435 	case DKIOCINFO:
6436 	{
6437 		struct dk_cinfo	cinfo;
6438 
6439 		bcopy(vdc->cinfo, &cinfo, sizeof (struct dk_cinfo));
6440 		cinfo.dki_partition = VDCPART(dev);
6441 
6442 		rv = ddi_copyout(&cinfo, (void *)arg,
6443 		    sizeof (struct dk_cinfo), mode);
6444 		if (rv != 0)
6445 			return (EFAULT);
6446 
6447 		return (0);
6448 	}
6449 
6450 	case DKIOCGMEDIAINFO:
6451 	{
6452 		ASSERT(vdc->vdisk_size != 0);
6453 		if (vdc->minfo->dki_capacity == 0)
6454 			vdc->minfo->dki_capacity = vdc->vdisk_size;
6455 		rv = ddi_copyout(vdc->minfo, (void *)arg,
6456 		    sizeof (struct dk_minfo), mode);
6457 		if (rv != 0)
6458 			return (EFAULT);
6459 
6460 		return (0);
6461 	}
6462 
6463 	case DKIOCFLUSHWRITECACHE:
6464 		{
6465 			struct dk_callback *dkc =
6466 			    (struct dk_callback *)(uintptr_t)arg;
6467 			vdc_dk_arg_t	*dkarg = NULL;
6468 
6469 			DMSG(vdc, 1, "[%d] Flush W$: mode %x\n",
6470 			    instance, mode);
6471 
6472 			/*
6473 			 * If arg is NULL, then there is no callback function
6474 			 * registered and the call operates synchronously; we
6475 			 * break and continue with the rest of the function and
6476 			 * wait for vds to return (i.e. after the request to
6477 			 * vds returns successfully, all writes completed prior
6478 			 * to the ioctl will have been flushed from the disk
6479 			 * write cache to persistent media.
6480 			 *
6481 			 * If a callback function is registered, we dispatch
6482 			 * the request on a task queue and return immediately.
6483 			 * The callback will deal with informing the calling
6484 			 * thread that the flush request is completed.
6485 			 */
6486 			if (dkc == NULL)
6487 				break;
6488 
6489 			/*
6490 			 * the asynchronous callback is only supported if
6491 			 * invoked from within the kernel
6492 			 */
6493 			if ((mode & FKIOCTL) == 0)
6494 				return (ENOTSUP);
6495 
6496 			dkarg = kmem_zalloc(sizeof (vdc_dk_arg_t), KM_SLEEP);
6497 
6498 			dkarg->mode = mode;
6499 			dkarg->dev = dev;
6500 			bcopy(dkc, &dkarg->dkc, sizeof (*dkc));
6501 
6502 			mutex_enter(&vdc->lock);
6503 			vdc->dkio_flush_pending++;
6504 			dkarg->vdc = vdc;
6505 			mutex_exit(&vdc->lock);
6506 
6507 			/* put the request on a task queue */
6508 			rv = taskq_dispatch(system_taskq, vdc_dkio_flush_cb,
6509 			    (void *)dkarg, DDI_SLEEP);
6510 			if (rv == NULL) {
6511 				/* clean up if dispatch fails */
6512 				mutex_enter(&vdc->lock);
6513 				vdc->dkio_flush_pending--;
6514 				mutex_exit(&vdc->lock);
6515 				kmem_free(dkarg, sizeof (vdc_dk_arg_t));
6516 			}
6517 
6518 			return (rv == NULL ? ENOMEM : 0);
6519 		}
6520 	}
6521 
6522 	/* catch programming error in vdc - should be a VD_OP_XXX ioctl */
6523 	ASSERT(iop->op != 0);
6524 
6525 	/* check if the vDisk server handles the operation for this vDisk */
6526 	if (VD_OP_SUPPORTED(vdc->operations, iop->op) == B_FALSE) {
6527 		DMSG(vdc, 0, "[%d] Unsupported VD_OP operation (0x%x)\n",
6528 		    vdc->instance, iop->op);
6529 		return (ENOTSUP);
6530 	}
6531 
6532 	/* LDC requires that the memory being mapped is 8-byte aligned */
6533 	alloc_len = P2ROUNDUP(len, sizeof (uint64_t));
6534 	DMSG(vdc, 1, "[%d] struct size %ld alloc %ld\n",
6535 	    instance, len, alloc_len);
6536 
6537 	if (alloc_len > 0)
6538 		mem_p = kmem_zalloc(alloc_len, KM_SLEEP);
6539 
6540 	/*
6541 	 * Call the conversion function for this ioctl which, if necessary,
6542 	 * converts from the Solaris format to the format ARC'ed
6543 	 * as part of the vDisk protocol (FWARC 2006/195)
6544 	 */
6545 	ASSERT(iop->convert != NULL);
6546 	rv = (iop->convert)(vdc, arg, mem_p, mode, VD_COPYIN);
6547 	if (rv != 0) {
6548 		DMSG(vdc, 0, "[%d] convert func returned %d for ioctl 0x%x\n",
6549 		    instance, rv, cmd);
6550 		if (mem_p != NULL)
6551 			kmem_free(mem_p, alloc_len);
6552 		return (rv);
6553 	}
6554 
6555 	/*
6556 	 * send request to vds to service the ioctl.
6557 	 */
6558 	rv = vdc_do_sync_op(vdc, iop->op, mem_p, alloc_len,
6559 	    VDCPART(dev), 0, CB_SYNC, (void *)(uint64_t)mode,
6560 	    VIO_both_dir, B_TRUE);
6561 
6562 	if (rv != 0) {
6563 		/*
6564 		 * This is not necessarily an error. The ioctl could
6565 		 * be returning a value such as ENOTTY to indicate
6566 		 * that the ioctl is not applicable.
6567 		 */
6568 		DMSG(vdc, 0, "[%d] vds returned %d for ioctl 0x%x\n",
6569 		    instance, rv, cmd);
6570 		if (mem_p != NULL)
6571 			kmem_free(mem_p, alloc_len);
6572 
6573 		return (rv);
6574 	}
6575 
6576 	/*
6577 	 * Call the conversion function (if it exists) for this ioctl
6578 	 * which converts from the format ARC'ed as part of the vDisk
6579 	 * protocol (FWARC 2006/195) back to a format understood by
6580 	 * the rest of Solaris.
6581 	 */
6582 	rv = (iop->convert)(vdc, mem_p, arg, mode, VD_COPYOUT);
6583 	if (rv != 0) {
6584 		DMSG(vdc, 0, "[%d] convert func returned %d for ioctl 0x%x\n",
6585 		    instance, rv, cmd);
6586 		if (mem_p != NULL)
6587 			kmem_free(mem_p, alloc_len);
6588 		return (rv);
6589 	}
6590 
6591 	if (mem_p != NULL)
6592 		kmem_free(mem_p, alloc_len);
6593 
6594 	return (rv);
6595 }
6596 
6597 /*
6598  * Function:
6599  *
6600  * Description:
6601  *	This is an empty conversion function used by ioctl calls which
6602  *	do not need to convert the data being passed in/out to userland
6603  */
6604 static int
6605 vdc_null_copy_func(vdc_t *vdc, void *from, void *to, int mode, int dir)
6606 {
6607 	_NOTE(ARGUNUSED(vdc))
6608 	_NOTE(ARGUNUSED(from))
6609 	_NOTE(ARGUNUSED(to))
6610 	_NOTE(ARGUNUSED(mode))
6611 	_NOTE(ARGUNUSED(dir))
6612 
6613 	return (0);
6614 }
6615 
6616 static int
6617 vdc_get_wce_convert(vdc_t *vdc, void *from, void *to,
6618     int mode, int dir)
6619 {
6620 	_NOTE(ARGUNUSED(vdc))
6621 
6622 	if (dir == VD_COPYIN)
6623 		return (0);		/* nothing to do */
6624 
6625 	if (ddi_copyout(from, to, sizeof (int), mode) != 0)
6626 		return (EFAULT);
6627 
6628 	return (0);
6629 }
6630 
6631 static int
6632 vdc_set_wce_convert(vdc_t *vdc, void *from, void *to,
6633     int mode, int dir)
6634 {
6635 	_NOTE(ARGUNUSED(vdc))
6636 
6637 	if (dir == VD_COPYOUT)
6638 		return (0);		/* nothing to do */
6639 
6640 	if (ddi_copyin(from, to, sizeof (int), mode) != 0)
6641 		return (EFAULT);
6642 
6643 	return (0);
6644 }
6645 
6646 /*
6647  * Function:
6648  *	vdc_get_vtoc_convert()
6649  *
6650  * Description:
6651  *	This routine performs the necessary convertions from the DKIOCGVTOC
6652  *	Solaris structure to the format defined in FWARC 2006/195.
6653  *
6654  *	In the struct vtoc definition, the timestamp field is marked as not
6655  *	supported so it is not part of vDisk protocol (FWARC 2006/195).
6656  *	However SVM uses that field to check it can write into the VTOC,
6657  *	so we fake up the info of that field.
6658  *
6659  * Arguments:
6660  *	vdc	- the vDisk client
6661  *	from	- the buffer containing the data to be copied from
6662  *	to	- the buffer to be copied to
6663  *	mode	- flags passed to ioctl() call
6664  *	dir	- the "direction" of the copy - VD_COPYIN or VD_COPYOUT
6665  *
6666  * Return Code:
6667  *	0	- Success
6668  *	ENXIO	- incorrect buffer passed in.
6669  *	EFAULT	- ddi_copyout routine encountered an error.
6670  */
6671 static int
6672 vdc_get_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
6673 {
6674 	int		i;
6675 	void		*tmp_mem = NULL;
6676 	void		*tmp_memp;
6677 	struct vtoc	vt;
6678 	struct vtoc32	vt32;
6679 	int		copy_len = 0;
6680 	int		rv = 0;
6681 
6682 	if (dir != VD_COPYOUT)
6683 		return (0);	/* nothing to do */
6684 
6685 	if ((from == NULL) || (to == NULL))
6686 		return (ENXIO);
6687 
6688 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32)
6689 		copy_len = sizeof (struct vtoc32);
6690 	else
6691 		copy_len = sizeof (struct vtoc);
6692 
6693 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
6694 
6695 	VD_VTOC2VTOC((vd_vtoc_t *)from, &vt);
6696 
6697 	/* fake the VTOC timestamp field */
6698 	for (i = 0; i < V_NUMPAR; i++) {
6699 		vt.timestamp[i] = vdc->vtoc->timestamp[i];
6700 	}
6701 
6702 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
6703 		/* LINTED E_ASSIGN_NARROW_CONV */
6704 		vtoctovtoc32(vt, vt32);
6705 		tmp_memp = &vt32;
6706 	} else {
6707 		tmp_memp = &vt;
6708 	}
6709 	rv = ddi_copyout(tmp_memp, to, copy_len, mode);
6710 	if (rv != 0)
6711 		rv = EFAULT;
6712 
6713 	kmem_free(tmp_mem, copy_len);
6714 	return (rv);
6715 }
6716 
6717 /*
6718  * Function:
6719  *	vdc_set_vtoc_convert()
6720  *
6721  * Description:
6722  *	This routine performs the necessary convertions from the DKIOCSVTOC
6723  *	Solaris structure to the format defined in FWARC 2006/195.
6724  *
6725  * Arguments:
6726  *	vdc	- the vDisk client
6727  *	from	- Buffer with data
6728  *	to	- Buffer where data is to be copied to
6729  *	mode	- flags passed to ioctl
6730  *	dir	- direction of copy (in or out)
6731  *
6732  * Return Code:
6733  *	0	- Success
6734  *	ENXIO	- Invalid buffer passed in
6735  *	EFAULT	- ddi_copyin of data failed
6736  */
6737 static int
6738 vdc_set_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
6739 {
6740 	_NOTE(ARGUNUSED(vdc))
6741 
6742 	void		*tmp_mem = NULL, *uvtoc;
6743 	struct vtoc	vt;
6744 	struct vtoc	*vtp = &vt;
6745 	vd_vtoc_t	vtvd;
6746 	int		copy_len = 0;
6747 	int		i, rv = 0;
6748 
6749 	if ((from == NULL) || (to == NULL))
6750 		return (ENXIO);
6751 
6752 	if (dir == VD_COPYIN)
6753 		uvtoc = from;
6754 	else
6755 		uvtoc = to;
6756 
6757 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32)
6758 		copy_len = sizeof (struct vtoc32);
6759 	else
6760 		copy_len = sizeof (struct vtoc);
6761 
6762 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
6763 
6764 	rv = ddi_copyin(uvtoc, tmp_mem, copy_len, mode);
6765 	if (rv != 0) {
6766 		kmem_free(tmp_mem, copy_len);
6767 		return (EFAULT);
6768 	}
6769 
6770 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
6771 		vtoc32tovtoc((*(struct vtoc32 *)tmp_mem), vt);
6772 	} else {
6773 		vtp = tmp_mem;
6774 	}
6775 
6776 	if (dir == VD_COPYOUT) {
6777 		/*
6778 		 * The disk label may have changed. Revalidate the disk
6779 		 * geometry. This will also update the device nodes and
6780 		 * properties.
6781 		 */
6782 		vdc_validate(vdc);
6783 
6784 		/*
6785 		 * We also need to keep track of the timestamp fields.
6786 		 */
6787 		for (i = 0; i < V_NUMPAR; i++) {
6788 			vdc->vtoc->timestamp[i] = vtp->timestamp[i];
6789 		}
6790 
6791 		return (0);
6792 	}
6793 
6794 	VTOC2VD_VTOC(vtp, &vtvd);
6795 	bcopy(&vtvd, to, sizeof (vd_vtoc_t));
6796 	kmem_free(tmp_mem, copy_len);
6797 
6798 	return (0);
6799 }
6800 
6801 /*
6802  * Function:
6803  *	vdc_get_geom_convert()
6804  *
6805  * Description:
6806  *	This routine performs the necessary convertions from the DKIOCGGEOM,
6807  *	DKIOCG_PHYSGEOM and DKIOG_VIRTGEOM Solaris structures to the format
6808  *	defined in FWARC 2006/195
6809  *
6810  * Arguments:
6811  *	vdc	- the vDisk client
6812  *	from	- Buffer with data
6813  *	to	- Buffer where data is to be copied to
6814  *	mode	- flags passed to ioctl
6815  *	dir	- direction of copy (in or out)
6816  *
6817  * Return Code:
6818  *	0	- Success
6819  *	ENXIO	- Invalid buffer passed in
6820  *	EFAULT	- ddi_copyout of data failed
6821  */
6822 static int
6823 vdc_get_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
6824 {
6825 	_NOTE(ARGUNUSED(vdc))
6826 
6827 	struct dk_geom	geom;
6828 	int	copy_len = sizeof (struct dk_geom);
6829 	int	rv = 0;
6830 
6831 	if (dir != VD_COPYOUT)
6832 		return (0);	/* nothing to do */
6833 
6834 	if ((from == NULL) || (to == NULL))
6835 		return (ENXIO);
6836 
6837 	VD_GEOM2DK_GEOM((vd_geom_t *)from, &geom);
6838 	rv = ddi_copyout(&geom, to, copy_len, mode);
6839 	if (rv != 0)
6840 		rv = EFAULT;
6841 
6842 	return (rv);
6843 }
6844 
6845 /*
6846  * Function:
6847  *	vdc_set_geom_convert()
6848  *
6849  * Description:
6850  *	This routine performs the necessary convertions from the DKIOCSGEOM
6851  *	Solaris structure to the format defined in FWARC 2006/195.
6852  *
6853  * Arguments:
6854  *	vdc	- the vDisk client
6855  *	from	- Buffer with data
6856  *	to	- Buffer where data is to be copied to
6857  *	mode	- flags passed to ioctl
6858  *	dir	- direction of copy (in or out)
6859  *
6860  * Return Code:
6861  *	0	- Success
6862  *	ENXIO	- Invalid buffer passed in
6863  *	EFAULT	- ddi_copyin of data failed
6864  */
6865 static int
6866 vdc_set_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
6867 {
6868 	_NOTE(ARGUNUSED(vdc))
6869 
6870 	vd_geom_t	vdgeom;
6871 	void		*tmp_mem = NULL;
6872 	int		copy_len = sizeof (struct dk_geom);
6873 	int		rv = 0;
6874 
6875 	if (dir != VD_COPYIN)
6876 		return (0);	/* nothing to do */
6877 
6878 	if ((from == NULL) || (to == NULL))
6879 		return (ENXIO);
6880 
6881 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
6882 
6883 	rv = ddi_copyin(from, tmp_mem, copy_len, mode);
6884 	if (rv != 0) {
6885 		kmem_free(tmp_mem, copy_len);
6886 		return (EFAULT);
6887 	}
6888 	DK_GEOM2VD_GEOM((struct dk_geom *)tmp_mem, &vdgeom);
6889 	bcopy(&vdgeom, to, sizeof (vdgeom));
6890 	kmem_free(tmp_mem, copy_len);
6891 
6892 	return (0);
6893 }
6894 
6895 static int
6896 vdc_get_efi_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
6897 {
6898 	_NOTE(ARGUNUSED(vdc))
6899 
6900 	vd_efi_t	*vd_efi;
6901 	dk_efi_t	dk_efi;
6902 	int		rv = 0;
6903 	void		*uaddr;
6904 
6905 	if ((from == NULL) || (to == NULL))
6906 		return (ENXIO);
6907 
6908 	if (dir == VD_COPYIN) {
6909 
6910 		vd_efi = (vd_efi_t *)to;
6911 
6912 		rv = ddi_copyin(from, &dk_efi, sizeof (dk_efi_t), mode);
6913 		if (rv != 0)
6914 			return (EFAULT);
6915 
6916 		vd_efi->lba = dk_efi.dki_lba;
6917 		vd_efi->length = dk_efi.dki_length;
6918 		bzero(vd_efi->data, vd_efi->length);
6919 
6920 	} else {
6921 
6922 		rv = ddi_copyin(to, &dk_efi, sizeof (dk_efi_t), mode);
6923 		if (rv != 0)
6924 			return (EFAULT);
6925 
6926 		uaddr = dk_efi.dki_data;
6927 
6928 		dk_efi.dki_data = kmem_alloc(dk_efi.dki_length, KM_SLEEP);
6929 
6930 		VD_EFI2DK_EFI((vd_efi_t *)from, &dk_efi);
6931 
6932 		rv = ddi_copyout(dk_efi.dki_data, uaddr, dk_efi.dki_length,
6933 		    mode);
6934 		if (rv != 0)
6935 			return (EFAULT);
6936 
6937 		kmem_free(dk_efi.dki_data, dk_efi.dki_length);
6938 	}
6939 
6940 	return (0);
6941 }
6942 
6943 static int
6944 vdc_set_efi_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
6945 {
6946 	_NOTE(ARGUNUSED(vdc))
6947 
6948 	dk_efi_t	dk_efi;
6949 	void		*uaddr;
6950 
6951 	if (dir == VD_COPYOUT) {
6952 		/*
6953 		 * The disk label may have changed. Revalidate the disk
6954 		 * geometry. This will also update the device nodes and
6955 		 * properties.
6956 		 */
6957 		vdc_validate(vdc);
6958 		return (0);
6959 	}
6960 
6961 	if ((from == NULL) || (to == NULL))
6962 		return (ENXIO);
6963 
6964 	if (ddi_copyin(from, &dk_efi, sizeof (dk_efi_t), mode) != 0)
6965 		return (EFAULT);
6966 
6967 	uaddr = dk_efi.dki_data;
6968 
6969 	dk_efi.dki_data = kmem_alloc(dk_efi.dki_length, KM_SLEEP);
6970 
6971 	if (ddi_copyin(uaddr, dk_efi.dki_data, dk_efi.dki_length, mode) != 0)
6972 		return (EFAULT);
6973 
6974 	DK_EFI2VD_EFI(&dk_efi, (vd_efi_t *)to);
6975 
6976 	kmem_free(dk_efi.dki_data, dk_efi.dki_length);
6977 
6978 	return (0);
6979 }
6980 
6981 
6982 /* -------------------------------------------------------------------------- */
6983 
6984 /*
6985  * Function:
6986  *	vdc_create_fake_geometry()
6987  *
6988  * Description:
6989  *	This routine fakes up the disk info needed for some DKIO ioctls such
6990  *	as DKIOCINFO and DKIOCGMEDIAINFO [just like lofi(7D) and ramdisk(7D) do]
6991  *
6992  *	Note: This function must not be called until the vDisk attributes have
6993  *	been exchanged as part of the handshake with the vDisk server.
6994  *
6995  * Arguments:
6996  *	vdc	- soft state pointer for this instance of the device driver.
6997  *
6998  * Return Code:
6999  *	none.
7000  */
7001 static void
7002 vdc_create_fake_geometry(vdc_t *vdc)
7003 {
7004 	ASSERT(vdc != NULL);
7005 	ASSERT(vdc->max_xfer_sz != 0);
7006 
7007 	/*
7008 	 * DKIOCINFO support
7009 	 */
7010 	if (vdc->cinfo == NULL)
7011 		vdc->cinfo = kmem_zalloc(sizeof (struct dk_cinfo), KM_SLEEP);
7012 
7013 	(void) strcpy(vdc->cinfo->dki_cname, VDC_DRIVER_NAME);
7014 	(void) strcpy(vdc->cinfo->dki_dname, VDC_DRIVER_NAME);
7015 	/* max_xfer_sz is #blocks so we don't need to divide by DEV_BSIZE */
7016 	vdc->cinfo->dki_maxtransfer = vdc->max_xfer_sz;
7017 
7018 	/*
7019 	 * We set the controller type to DKC_SCSI_CCS only if the VD_OP_SCSICMD
7020 	 * operation is supported, otherwise the controller type is DKC_DIRECT.
7021 	 * Version 1.0 does not support the VD_OP_SCSICMD operation, so the
7022 	 * controller type is always DKC_DIRECT in that case.
7023 	 *
7024 	 * If the virtual disk is backed by a physical CD/DVD device or
7025 	 * an ISO image, modify the controller type to indicate this
7026 	 */
7027 	switch (vdc->vdisk_media) {
7028 	case VD_MEDIA_CD:
7029 	case VD_MEDIA_DVD:
7030 		vdc->cinfo->dki_ctype = DKC_CDROM;
7031 		break;
7032 	case VD_MEDIA_FIXED:
7033 		if (VD_OP_SUPPORTED(vdc->operations, VD_OP_SCSICMD))
7034 			vdc->cinfo->dki_ctype = DKC_SCSI_CCS;
7035 		else
7036 			vdc->cinfo->dki_ctype = DKC_DIRECT;
7037 		break;
7038 	default:
7039 		/* in the case of v1.0 we default to a fixed disk */
7040 		vdc->cinfo->dki_ctype = DKC_DIRECT;
7041 		break;
7042 	}
7043 	vdc->cinfo->dki_flags = DKI_FMTVOL;
7044 	vdc->cinfo->dki_cnum = 0;
7045 	vdc->cinfo->dki_addr = 0;
7046 	vdc->cinfo->dki_space = 0;
7047 	vdc->cinfo->dki_prio = 0;
7048 	vdc->cinfo->dki_vec = 0;
7049 	vdc->cinfo->dki_unit = vdc->instance;
7050 	vdc->cinfo->dki_slave = 0;
7051 	/*
7052 	 * The partition number will be created on the fly depending on the
7053 	 * actual slice (i.e. minor node) that is used to request the data.
7054 	 */
7055 	vdc->cinfo->dki_partition = 0;
7056 
7057 	/*
7058 	 * DKIOCGMEDIAINFO support
7059 	 */
7060 	if (vdc->minfo == NULL)
7061 		vdc->minfo = kmem_zalloc(sizeof (struct dk_minfo), KM_SLEEP);
7062 
7063 	if (vio_ver_is_supported(vdc->ver, 1, 1)) {
7064 		vdc->minfo->dki_media_type =
7065 		    VD_MEDIATYPE2DK_MEDIATYPE(vdc->vdisk_media);
7066 	} else {
7067 		vdc->minfo->dki_media_type = DK_FIXED_DISK;
7068 	}
7069 
7070 	vdc->minfo->dki_capacity = vdc->vdisk_size;
7071 	vdc->minfo->dki_lbsize = vdc->block_size;
7072 }
7073 
7074 static ushort_t
7075 vdc_lbl2cksum(struct dk_label *label)
7076 {
7077 	int	count;
7078 	ushort_t sum, *sp;
7079 
7080 	count =	(sizeof (struct dk_label)) / (sizeof (short)) - 1;
7081 	sp = (ushort_t *)label;
7082 	sum = 0;
7083 	while (count--) {
7084 		sum ^= *sp++;
7085 	}
7086 
7087 	return (sum);
7088 }
7089 
7090 /*
7091  * Function:
7092  *	vdc_validate_geometry
7093  *
7094  * Description:
7095  *	This routine discovers the label and geometry of the disk. It stores
7096  *	the disk label and related information in the vdc structure. If it
7097  *	fails to validate the geometry or to discover the disk label then
7098  *	the label is marked as unknown (VD_DISK_LABEL_UNK).
7099  *
7100  * Arguments:
7101  *	vdc	- soft state pointer for this instance of the device driver.
7102  *
7103  * Return Code:
7104  *	0	- success.
7105  *	EINVAL	- unknown disk label.
7106  *	ENOTSUP	- geometry not applicable (EFI label).
7107  *	EIO	- error accessing the disk.
7108  */
7109 static int
7110 vdc_validate_geometry(vdc_t *vdc)
7111 {
7112 	buf_t	*buf;	/* BREAD requests need to be in a buf_t structure */
7113 	dev_t	dev;
7114 	int	rv, rval;
7115 	struct dk_label label;
7116 	struct dk_geom geom;
7117 	struct vtoc vtoc;
7118 	efi_gpt_t *gpt;
7119 	efi_gpe_t *gpe;
7120 	vd_efi_dev_t edev;
7121 
7122 	ASSERT(vdc != NULL);
7123 	ASSERT(vdc->vtoc != NULL && vdc->geom != NULL);
7124 	ASSERT(MUTEX_HELD(&vdc->lock));
7125 
7126 	mutex_exit(&vdc->lock);
7127 
7128 	dev = makedevice(ddi_driver_major(vdc->dip),
7129 	    VD_MAKE_DEV(vdc->instance, 0));
7130 
7131 	rv = vd_process_ioctl(dev, DKIOCGGEOM, (caddr_t)&geom, FKIOCTL, &rval);
7132 	if (rv == 0)
7133 		rv = vd_process_ioctl(dev, DKIOCGVTOC, (caddr_t)&vtoc,
7134 		    FKIOCTL, &rval);
7135 
7136 	if (rv == ENOTSUP) {
7137 		/*
7138 		 * If the device does not support VTOC then we try
7139 		 * to read an EFI label.
7140 		 *
7141 		 * We need to know the block size and the disk size to
7142 		 * be able to read an EFI label.
7143 		 */
7144 		if (vdc->vdisk_size == 0) {
7145 			if ((rv = vdc_check_capacity(vdc)) != 0) {
7146 				mutex_enter(&vdc->lock);
7147 				vdc_store_label_unk(vdc);
7148 				return (rv);
7149 			}
7150 		}
7151 
7152 		VD_EFI_DEV_SET(edev, vdc, vd_process_efi_ioctl);
7153 
7154 		rv = vd_efi_alloc_and_read(&edev, &gpt, &gpe);
7155 
7156 		if (rv) {
7157 			DMSG(vdc, 0, "[%d] Failed to get EFI (err=%d)",
7158 			    vdc->instance, rv);
7159 			mutex_enter(&vdc->lock);
7160 			vdc_store_label_unk(vdc);
7161 			return (EIO);
7162 		}
7163 
7164 		mutex_enter(&vdc->lock);
7165 		vdc_store_label_efi(vdc, gpt, gpe);
7166 		vd_efi_free(&edev, gpt, gpe);
7167 		return (ENOTSUP);
7168 	}
7169 
7170 	if (rv != 0) {
7171 		DMSG(vdc, 0, "[%d] Failed to get VTOC (err=%d)",
7172 		    vdc->instance, rv);
7173 		mutex_enter(&vdc->lock);
7174 		vdc_store_label_unk(vdc);
7175 		if (rv != EINVAL)
7176 			rv = EIO;
7177 		return (rv);
7178 	}
7179 
7180 	/* check that geometry and vtoc are valid */
7181 	if (geom.dkg_nhead == 0 || geom.dkg_nsect == 0 ||
7182 	    vtoc.v_sanity != VTOC_SANE) {
7183 		mutex_enter(&vdc->lock);
7184 		vdc_store_label_unk(vdc);
7185 		return (EINVAL);
7186 	}
7187 
7188 	/*
7189 	 * We have a disk and a valid VTOC. However this does not mean
7190 	 * that the disk currently have a VTOC label. The returned VTOC may
7191 	 * be a default VTOC to be used for configuring the disk (this is
7192 	 * what is done for disk image). So we read the label from the
7193 	 * beginning of the disk to ensure we really have a VTOC label.
7194 	 *
7195 	 * FUTURE: This could be the default way for reading the VTOC
7196 	 * from the disk as opposed to sending the VD_OP_GET_VTOC
7197 	 * to the server. This will be the default if vdc is implemented
7198 	 * ontop of cmlb.
7199 	 */
7200 
7201 	/*
7202 	 * Single slice disk does not support read using an absolute disk
7203 	 * offset so we just rely on the DKIOCGVTOC ioctl in that case.
7204 	 */
7205 	if (vdc->vdisk_type == VD_DISK_TYPE_SLICE) {
7206 		mutex_enter(&vdc->lock);
7207 		if (vtoc.v_nparts != 1) {
7208 			vdc_store_label_unk(vdc);
7209 			return (EINVAL);
7210 		}
7211 		vdc_store_label_vtoc(vdc, &geom, &vtoc);
7212 		return (0);
7213 	}
7214 
7215 	if (vtoc.v_nparts != V_NUMPAR) {
7216 		mutex_enter(&vdc->lock);
7217 		vdc_store_label_unk(vdc);
7218 		return (EINVAL);
7219 	}
7220 
7221 	/*
7222 	 * Read disk label from start of disk
7223 	 */
7224 	buf = kmem_alloc(sizeof (buf_t), KM_SLEEP);
7225 	bioinit(buf);
7226 	buf->b_un.b_addr = (caddr_t)&label;
7227 	buf->b_bcount = DK_LABEL_SIZE;
7228 	buf->b_flags = B_BUSY | B_READ;
7229 	buf->b_dev = cmpdev(dev);
7230 	rv = vdc_send_request(vdc, VD_OP_BREAD, (caddr_t)&label,
7231 	    DK_LABEL_SIZE, VD_SLICE_NONE, 0, CB_STRATEGY, buf, VIO_read_dir);
7232 	if (rv) {
7233 		DMSG(vdc, 1, "[%d] Failed to read disk block 0\n",
7234 		    vdc->instance);
7235 	} else {
7236 		rv = biowait(buf);
7237 		biofini(buf);
7238 	}
7239 	kmem_free(buf, sizeof (buf_t));
7240 
7241 	if (rv != 0 || label.dkl_magic != DKL_MAGIC ||
7242 	    label.dkl_cksum != vdc_lbl2cksum(&label)) {
7243 		DMSG(vdc, 1, "[%d] Got VTOC with invalid label\n",
7244 		    vdc->instance);
7245 		mutex_enter(&vdc->lock);
7246 		vdc_store_label_unk(vdc);
7247 		return (EINVAL);
7248 	}
7249 
7250 	mutex_enter(&vdc->lock);
7251 	vdc_store_label_vtoc(vdc, &geom, &vtoc);
7252 	return (0);
7253 }
7254 
7255 /*
7256  * Function:
7257  *	vdc_validate
7258  *
7259  * Description:
7260  *	This routine discovers the label of the disk and create the
7261  *	appropriate device nodes if the label has changed.
7262  *
7263  * Arguments:
7264  *	vdc	- soft state pointer for this instance of the device driver.
7265  *
7266  * Return Code:
7267  *	none.
7268  */
7269 static void
7270 vdc_validate(vdc_t *vdc)
7271 {
7272 	vd_disk_label_t old_label;
7273 	vd_slice_t old_slice[V_NUMPAR];
7274 	int rv;
7275 
7276 	ASSERT(!MUTEX_HELD(&vdc->lock));
7277 
7278 	mutex_enter(&vdc->lock);
7279 
7280 	/* save the current label and vtoc */
7281 	old_label = vdc->vdisk_label;
7282 	bcopy(vdc->slice, &old_slice, sizeof (vd_slice_t) * V_NUMPAR);
7283 
7284 	/* check the geometry */
7285 	(void) vdc_validate_geometry(vdc);
7286 
7287 	/* if the disk label has changed, update device nodes */
7288 	if (vdc->vdisk_label != old_label) {
7289 
7290 		if (vdc->vdisk_label == VD_DISK_LABEL_EFI)
7291 			rv = vdc_create_device_nodes_efi(vdc);
7292 		else
7293 			rv = vdc_create_device_nodes_vtoc(vdc);
7294 
7295 		if (rv != 0) {
7296 			DMSG(vdc, 0, "![%d] Failed to update device nodes",
7297 			    vdc->instance);
7298 		}
7299 	}
7300 
7301 	/* if the vtoc has changed, update device nodes properties */
7302 	if (bcmp(vdc->slice, &old_slice, sizeof (vd_slice_t) * V_NUMPAR) != 0) {
7303 
7304 		if (vdc_create_device_nodes_props(vdc) != 0) {
7305 			DMSG(vdc, 0, "![%d] Failed to update device nodes"
7306 			    " properties", vdc->instance);
7307 		}
7308 	}
7309 
7310 	mutex_exit(&vdc->lock);
7311 }
7312 
7313 static void
7314 vdc_validate_task(void *arg)
7315 {
7316 	vdc_t *vdc = (vdc_t *)arg;
7317 
7318 	vdc_validate(vdc);
7319 
7320 	mutex_enter(&vdc->lock);
7321 	ASSERT(vdc->validate_pending > 0);
7322 	vdc->validate_pending--;
7323 	mutex_exit(&vdc->lock);
7324 }
7325 
7326 /*
7327  * Function:
7328  *	vdc_setup_devid()
7329  *
7330  * Description:
7331  *	This routine discovers the devid of a vDisk. It requests the devid of
7332  *	the underlying device from the vDisk server, builds an encapsulated
7333  *	devid based on the retrieved devid and registers that new devid to
7334  *	the vDisk.
7335  *
7336  * Arguments:
7337  *	vdc	- soft state pointer for this instance of the device driver.
7338  *
7339  * Return Code:
7340  *	0	- A devid was succesfully registered for the vDisk
7341  */
7342 static int
7343 vdc_setup_devid(vdc_t *vdc)
7344 {
7345 	int rv;
7346 	vd_devid_t *vd_devid;
7347 	size_t bufsize, bufid_len;
7348 
7349 	/*
7350 	 * At first sight, we don't know the size of the devid that the
7351 	 * server will return but this size will be encoded into the
7352 	 * reply. So we do a first request using a default size then we
7353 	 * check if this size was large enough. If not then we do a second
7354 	 * request with the correct size returned by the server. Note that
7355 	 * ldc requires size to be 8-byte aligned.
7356 	 */
7357 	bufsize = P2ROUNDUP(VD_DEVID_SIZE(VD_DEVID_DEFAULT_LEN),
7358 	    sizeof (uint64_t));
7359 	vd_devid = kmem_zalloc(bufsize, KM_SLEEP);
7360 	bufid_len = bufsize - sizeof (vd_efi_t) - 1;
7361 
7362 	rv = vdc_do_sync_op(vdc, VD_OP_GET_DEVID, (caddr_t)vd_devid,
7363 	    bufsize, 0, 0, CB_SYNC, 0, VIO_both_dir, B_TRUE);
7364 
7365 	DMSG(vdc, 2, "sync_op returned %d\n", rv);
7366 
7367 	if (rv) {
7368 		kmem_free(vd_devid, bufsize);
7369 		return (rv);
7370 	}
7371 
7372 	if (vd_devid->length > bufid_len) {
7373 		/*
7374 		 * The returned devid is larger than the buffer used. Try again
7375 		 * with a buffer with the right size.
7376 		 */
7377 		kmem_free(vd_devid, bufsize);
7378 		bufsize = P2ROUNDUP(VD_DEVID_SIZE(vd_devid->length),
7379 		    sizeof (uint64_t));
7380 		vd_devid = kmem_zalloc(bufsize, KM_SLEEP);
7381 		bufid_len = bufsize - sizeof (vd_efi_t) - 1;
7382 
7383 		rv = vdc_do_sync_op(vdc, VD_OP_GET_DEVID,
7384 		    (caddr_t)vd_devid, bufsize, 0, 0, CB_SYNC, 0,
7385 		    VIO_both_dir, B_TRUE);
7386 
7387 		if (rv) {
7388 			kmem_free(vd_devid, bufsize);
7389 			return (rv);
7390 		}
7391 	}
7392 
7393 	/*
7394 	 * The virtual disk should have the same device id as the one associated
7395 	 * with the physical disk it is mapped on, otherwise sharing a disk
7396 	 * between a LDom and a non-LDom may not work (for example for a shared
7397 	 * SVM disk set).
7398 	 *
7399 	 * The DDI framework does not allow creating a device id with any
7400 	 * type so we first create a device id of type DEVID_ENCAP and then
7401 	 * we restore the orignal type of the physical device.
7402 	 */
7403 
7404 	DMSG(vdc, 2, ": devid length = %d\n", vd_devid->length);
7405 
7406 	/* build an encapsulated devid based on the returned devid */
7407 	if (ddi_devid_init(vdc->dip, DEVID_ENCAP, vd_devid->length,
7408 	    vd_devid->id, &vdc->devid) != DDI_SUCCESS) {
7409 		DMSG(vdc, 1, "[%d] Fail to created devid\n", vdc->instance);
7410 		kmem_free(vd_devid, bufsize);
7411 		return (1);
7412 	}
7413 
7414 	DEVID_FORMTYPE((impl_devid_t *)vdc->devid, vd_devid->type);
7415 
7416 	ASSERT(ddi_devid_valid(vdc->devid) == DDI_SUCCESS);
7417 
7418 	kmem_free(vd_devid, bufsize);
7419 
7420 	if (ddi_devid_register(vdc->dip, vdc->devid) != DDI_SUCCESS) {
7421 		DMSG(vdc, 1, "[%d] Fail to register devid\n", vdc->instance);
7422 		return (1);
7423 	}
7424 
7425 	return (0);
7426 }
7427 
7428 static void
7429 vdc_store_label_efi(vdc_t *vdc, efi_gpt_t *gpt, efi_gpe_t *gpe)
7430 {
7431 	int i, nparts;
7432 
7433 	ASSERT(MUTEX_HELD(&vdc->lock));
7434 
7435 	vdc->vdisk_label = VD_DISK_LABEL_EFI;
7436 	bzero(vdc->vtoc, sizeof (struct vtoc));
7437 	bzero(vdc->geom, sizeof (struct dk_geom));
7438 	bzero(vdc->slice, sizeof (vd_slice_t) * V_NUMPAR);
7439 
7440 	nparts = gpt->efi_gpt_NumberOfPartitionEntries;
7441 
7442 	for (i = 0; i < nparts && i < VD_EFI_WD_SLICE; i++) {
7443 
7444 		if (gpe[i].efi_gpe_StartingLBA == 0 ||
7445 		    gpe[i].efi_gpe_EndingLBA == 0) {
7446 			continue;
7447 		}
7448 
7449 		vdc->slice[i].start = gpe[i].efi_gpe_StartingLBA;
7450 		vdc->slice[i].nblocks = gpe[i].efi_gpe_EndingLBA -
7451 		    gpe[i].efi_gpe_StartingLBA + 1;
7452 	}
7453 
7454 	ASSERT(vdc->vdisk_size != 0);
7455 	vdc->slice[VD_EFI_WD_SLICE].start = 0;
7456 	vdc->slice[VD_EFI_WD_SLICE].nblocks = vdc->vdisk_size;
7457 
7458 }
7459 
7460 static void
7461 vdc_store_label_vtoc(vdc_t *vdc, struct dk_geom *geom, struct vtoc *vtoc)
7462 {
7463 	int i;
7464 
7465 	ASSERT(MUTEX_HELD(&vdc->lock));
7466 	ASSERT(vdc->block_size == vtoc->v_sectorsz);
7467 
7468 	vdc->vdisk_label = VD_DISK_LABEL_VTOC;
7469 	bcopy(vtoc, vdc->vtoc, sizeof (struct vtoc));
7470 	bcopy(geom, vdc->geom, sizeof (struct dk_geom));
7471 	bzero(vdc->slice, sizeof (vd_slice_t) * V_NUMPAR);
7472 
7473 	for (i = 0; i < vtoc->v_nparts; i++) {
7474 		vdc->slice[i].start = vtoc->v_part[i].p_start;
7475 		vdc->slice[i].nblocks = vtoc->v_part[i].p_size;
7476 	}
7477 }
7478 
7479 static void
7480 vdc_store_label_unk(vdc_t *vdc)
7481 {
7482 	ASSERT(MUTEX_HELD(&vdc->lock));
7483 
7484 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
7485 	bzero(vdc->vtoc, sizeof (struct vtoc));
7486 	bzero(vdc->geom, sizeof (struct dk_geom));
7487 	bzero(vdc->slice, sizeof (vd_slice_t) * V_NUMPAR);
7488 }
7489