xref: /titanic_51/usr/src/uts/sun4v/io/vdc.c (revision 949b855767d66dfa7821c669df554bea92973cc9)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 /*
30  * LDoms virtual disk client (vdc) device driver
31  *
32  * This driver runs on a guest logical domain and communicates with the virtual
33  * disk server (vds) driver running on the service domain which is exporting
34  * virtualized "disks" to the guest logical domain.
35  *
36  * The driver can be divided into four sections:
37  *
38  * 1) generic device driver housekeeping
39  *	_init, _fini, attach, detach, ops structures, etc.
40  *
41  * 2) communication channel setup
42  *	Setup the communications link over the LDC channel that vdc uses to
43  *	talk to the vDisk server. Initialise the descriptor ring which
44  *	allows the LDC clients to transfer data via memory mappings.
45  *
46  * 3) Support exported to upper layers (filesystems, etc)
47  *	The upper layers call into vdc via strategy(9E) and DKIO(7I)
48  *	ioctl calls. vdc will copy the data to be written to the descriptor
49  *	ring or maps the buffer to store the data read by the vDisk
50  *	server into the descriptor ring. It then sends a message to the
51  *	vDisk server requesting it to complete the operation.
52  *
53  * 4) Handling responses from vDisk server.
54  *	The vDisk server will ACK some or all of the messages vdc sends to it
55  *	(this is configured during the handshake). Upon receipt of an ACK
56  *	vdc will check the descriptor ring and signal to the upper layer
57  *	code waiting on the IO.
58  */
59 
60 #include <sys/atomic.h>
61 #include <sys/conf.h>
62 #include <sys/disp.h>
63 #include <sys/ddi.h>
64 #include <sys/dkio.h>
65 #include <sys/efi_partition.h>
66 #include <sys/fcntl.h>
67 #include <sys/file.h>
68 #include <sys/kstat.h>
69 #include <sys/mach_descrip.h>
70 #include <sys/modctl.h>
71 #include <sys/mdeg.h>
72 #include <sys/note.h>
73 #include <sys/open.h>
74 #include <sys/sdt.h>
75 #include <sys/stat.h>
76 #include <sys/sunddi.h>
77 #include <sys/types.h>
78 #include <sys/promif.h>
79 #include <sys/var.h>
80 #include <sys/vtoc.h>
81 #include <sys/archsystm.h>
82 #include <sys/sysmacros.h>
83 
84 #include <sys/cdio.h>
85 #include <sys/dktp/fdisk.h>
86 #include <sys/dktp/dadkio.h>
87 #include <sys/mhd.h>
88 #include <sys/scsi/generic/sense.h>
89 #include <sys/scsi/impl/uscsi.h>
90 #include <sys/scsi/impl/services.h>
91 #include <sys/scsi/targets/sddef.h>
92 
93 #include <sys/ldoms.h>
94 #include <sys/ldc.h>
95 #include <sys/vio_common.h>
96 #include <sys/vio_mailbox.h>
97 #include <sys/vio_util.h>
98 #include <sys/vdsk_common.h>
99 #include <sys/vdsk_mailbox.h>
100 #include <sys/vdc.h>
101 
102 /*
103  * function prototypes
104  */
105 
106 /* standard driver functions */
107 static int	vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred);
108 static int	vdc_close(dev_t dev, int flag, int otyp, cred_t *cred);
109 static int	vdc_strategy(struct buf *buf);
110 static int	vdc_print(dev_t dev, char *str);
111 static int	vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk);
112 static int	vdc_read(dev_t dev, struct uio *uio, cred_t *cred);
113 static int	vdc_write(dev_t dev, struct uio *uio, cred_t *cred);
114 static int	vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode,
115 			cred_t *credp, int *rvalp);
116 static int	vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred);
117 static int	vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred);
118 
119 static int	vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,
120 			void *arg, void **resultp);
121 static int	vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd);
122 static int	vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd);
123 
124 /* setup */
125 static void	vdc_min(struct buf *bufp);
126 static int	vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen);
127 static int	vdc_do_ldc_init(vdc_t *vdc, md_t *mdp, mde_cookie_t vd_node);
128 static int	vdc_start_ldc_connection(vdc_t *vdc);
129 static int	vdc_create_device_nodes(vdc_t *vdc);
130 static int	vdc_create_device_nodes_efi(vdc_t *vdc);
131 static int	vdc_create_device_nodes_vtoc(vdc_t *vdc);
132 static int	vdc_create_device_nodes_props(vdc_t *vdc);
133 static void	vdc_create_io_kstats(vdc_t *vdc);
134 static void	vdc_create_err_kstats(vdc_t *vdc);
135 static void	vdc_set_err_kstats(vdc_t *vdc);
136 static int	vdc_get_md_node(dev_info_t *dip, md_t **mdpp,
137 		    mde_cookie_t *vd_nodep, mde_cookie_t *vd_portp);
138 static int	vdc_get_ldc_id(md_t *, mde_cookie_t, uint64_t *);
139 static int	vdc_do_ldc_up(vdc_t *vdc);
140 static void	vdc_terminate_ldc(vdc_t *vdc);
141 static int	vdc_init_descriptor_ring(vdc_t *vdc);
142 static void	vdc_destroy_descriptor_ring(vdc_t *vdc);
143 static int	vdc_setup_devid(vdc_t *vdc);
144 static void	vdc_store_label_efi(vdc_t *, efi_gpt_t *, efi_gpe_t *);
145 static void	vdc_store_label_vtoc(vdc_t *, struct dk_geom *, struct vtoc *);
146 static void	vdc_store_label_unk(vdc_t *vdc);
147 static boolean_t vdc_is_opened(vdc_t *vdc);
148 
149 /* handshake with vds */
150 static int		vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver);
151 static int		vdc_ver_negotiation(vdc_t *vdcp);
152 static int		vdc_init_attr_negotiation(vdc_t *vdc);
153 static int		vdc_attr_negotiation(vdc_t *vdcp);
154 static int		vdc_init_dring_negotiate(vdc_t *vdc);
155 static int		vdc_dring_negotiation(vdc_t *vdcp);
156 static int		vdc_send_rdx(vdc_t *vdcp);
157 static int		vdc_rdx_exchange(vdc_t *vdcp);
158 static boolean_t	vdc_is_supported_version(vio_ver_msg_t *ver_msg);
159 
160 /* processing incoming messages from vDisk server */
161 static void	vdc_process_msg_thread(vdc_t *vdc);
162 static int	vdc_recv(vdc_t *vdc, vio_msg_t *msgp, size_t *nbytesp);
163 
164 static uint_t	vdc_handle_cb(uint64_t event, caddr_t arg);
165 static int	vdc_process_data_msg(vdc_t *vdc, vio_msg_t *msg);
166 static int	vdc_handle_ver_msg(vdc_t *vdc, vio_ver_msg_t *ver_msg);
167 static int	vdc_handle_attr_msg(vdc_t *vdc, vd_attr_msg_t *attr_msg);
168 static int	vdc_handle_dring_reg_msg(vdc_t *vdc, vio_dring_reg_msg_t *msg);
169 static int 	vdc_send_request(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_map_to_shared_dring(vdc_t *vdcp, int idx);
173 static int 	vdc_populate_descriptor(vdc_t *vdcp, int operation,
174 		    caddr_t addr, size_t nbytes, int slice, diskaddr_t offset,
175 		    int cb_type, void *cb_arg, vio_desc_direction_t dir);
176 static int 	vdc_do_sync_op(vdc_t *vdcp, int operation, caddr_t addr,
177 		    size_t nbytes, int slice, diskaddr_t offset, int cb_type,
178 		    void *cb_arg, vio_desc_direction_t dir, boolean_t);
179 
180 static int	vdc_wait_for_response(vdc_t *vdcp, vio_msg_t *msgp);
181 static int	vdc_drain_response(vdc_t *vdcp);
182 static int	vdc_depopulate_descriptor(vdc_t *vdc, uint_t idx);
183 static int	vdc_populate_mem_hdl(vdc_t *vdcp, vdc_local_desc_t *ldep);
184 static int	vdc_verify_seq_num(vdc_t *vdc, vio_dring_msg_t *dring_msg);
185 
186 /* dkio */
187 static int	vd_process_ioctl(dev_t dev, int cmd, caddr_t arg, int mode,
188 		    int *rvalp);
189 static int	vd_process_efi_ioctl(void *vdisk, int cmd, uintptr_t arg);
190 static void	vdc_create_fake_geometry(vdc_t *vdc);
191 static int	vdc_validate_geometry(vdc_t *vdc);
192 static void	vdc_validate(vdc_t *vdc);
193 static void	vdc_validate_task(void *arg);
194 static int	vdc_null_copy_func(vdc_t *vdc, void *from, void *to,
195 		    int mode, int dir);
196 static int	vdc_get_wce_convert(vdc_t *vdc, void *from, void *to,
197 		    int mode, int dir);
198 static int	vdc_set_wce_convert(vdc_t *vdc, void *from, void *to,
199 		    int mode, int dir);
200 static int	vdc_get_vtoc_convert(vdc_t *vdc, void *from, void *to,
201 		    int mode, int dir);
202 static int	vdc_set_vtoc_convert(vdc_t *vdc, void *from, void *to,
203 		    int mode, int dir);
204 static int	vdc_get_geom_convert(vdc_t *vdc, void *from, void *to,
205 		    int mode, int dir);
206 static int	vdc_set_geom_convert(vdc_t *vdc, void *from, void *to,
207 		    int mode, int dir);
208 static int	vdc_get_efi_convert(vdc_t *vdc, void *from, void *to,
209 		    int mode, int dir);
210 static int	vdc_set_efi_convert(vdc_t *vdc, void *from, void *to,
211 		    int mode, int dir);
212 
213 static void 	vdc_ownership_update(vdc_t *vdc, int ownership_flags);
214 static int	vdc_access_set(vdc_t *vdc, uint64_t flags, int mode);
215 static vdc_io_t	*vdc_failfast_io_queue(vdc_t *vdc, struct buf *buf);
216 static int	vdc_failfast_check_resv(vdc_t *vdc);
217 
218 /*
219  * Module variables
220  */
221 
222 /*
223  * Tunable variables to control how long vdc waits before timing out on
224  * various operations
225  */
226 static int	vdc_hshake_retries = 3;
227 
228 static int	vdc_timeout = 0; /* units: seconds */
229 
230 static uint64_t vdc_hz_min_ldc_delay;
231 static uint64_t vdc_min_timeout_ldc = 1 * MILLISEC;
232 static uint64_t vdc_hz_max_ldc_delay;
233 static uint64_t vdc_max_timeout_ldc = 100 * MILLISEC;
234 
235 static uint64_t vdc_ldc_read_init_delay = 1 * MILLISEC;
236 static uint64_t vdc_ldc_read_max_delay = 100 * MILLISEC;
237 
238 /* values for dumping - need to run in a tighter loop */
239 static uint64_t	vdc_usec_timeout_dump = 100 * MILLISEC;	/* 0.1s units: ns */
240 static int	vdc_dump_retries = 100;
241 
242 static uint16_t	vdc_scsi_timeout = 60;	/* 60s units: seconds  */
243 
244 static uint64_t vdc_ownership_delay = 6 * MICROSEC; /* 6s units: usec */
245 
246 /* Count of the number of vdc instances attached */
247 static volatile uint32_t	vdc_instance_count = 0;
248 
249 /* Tunable to log all SCSI errors */
250 static boolean_t vdc_scsi_log_error = B_FALSE;
251 
252 /* Soft state pointer */
253 static void	*vdc_state;
254 
255 /*
256  * Controlling the verbosity of the error/debug messages
257  *
258  * vdc_msglevel - controls level of messages
259  * vdc_matchinst - 64-bit variable where each bit corresponds
260  *                 to the vdc instance the vdc_msglevel applies.
261  */
262 int		vdc_msglevel = 0x0;
263 uint64_t	vdc_matchinst = 0ull;
264 
265 /*
266  * Supported vDisk protocol version pairs.
267  *
268  * The first array entry is the latest and preferred version.
269  */
270 static const vio_ver_t	vdc_version[] = {{1, 1}};
271 
272 static struct cb_ops vdc_cb_ops = {
273 	vdc_open,	/* cb_open */
274 	vdc_close,	/* cb_close */
275 	vdc_strategy,	/* cb_strategy */
276 	vdc_print,	/* cb_print */
277 	vdc_dump,	/* cb_dump */
278 	vdc_read,	/* cb_read */
279 	vdc_write,	/* cb_write */
280 	vdc_ioctl,	/* cb_ioctl */
281 	nodev,		/* cb_devmap */
282 	nodev,		/* cb_mmap */
283 	nodev,		/* cb_segmap */
284 	nochpoll,	/* cb_chpoll */
285 	ddi_prop_op,	/* cb_prop_op */
286 	NULL,		/* cb_str */
287 	D_MP | D_64BIT,	/* cb_flag */
288 	CB_REV,		/* cb_rev */
289 	vdc_aread,	/* cb_aread */
290 	vdc_awrite	/* cb_awrite */
291 };
292 
293 static struct dev_ops vdc_ops = {
294 	DEVO_REV,	/* devo_rev */
295 	0,		/* devo_refcnt */
296 	vdc_getinfo,	/* devo_getinfo */
297 	nulldev,	/* devo_identify */
298 	nulldev,	/* devo_probe */
299 	vdc_attach,	/* devo_attach */
300 	vdc_detach,	/* devo_detach */
301 	nodev,		/* devo_reset */
302 	&vdc_cb_ops,	/* devo_cb_ops */
303 	NULL,		/* devo_bus_ops */
304 	nulldev		/* devo_power */
305 };
306 
307 static struct modldrv modldrv = {
308 	&mod_driverops,
309 	"virtual disk client",
310 	&vdc_ops,
311 };
312 
313 static struct modlinkage modlinkage = {
314 	MODREV_1,
315 	&modldrv,
316 	NULL
317 };
318 
319 /* -------------------------------------------------------------------------- */
320 
321 /*
322  * Device Driver housekeeping and setup
323  */
324 
325 int
326 _init(void)
327 {
328 	int	status;
329 
330 	if ((status = ddi_soft_state_init(&vdc_state, sizeof (vdc_t), 1)) != 0)
331 		return (status);
332 	if ((status = mod_install(&modlinkage)) != 0)
333 		ddi_soft_state_fini(&vdc_state);
334 	return (status);
335 }
336 
337 int
338 _info(struct modinfo *modinfop)
339 {
340 	return (mod_info(&modlinkage, modinfop));
341 }
342 
343 int
344 _fini(void)
345 {
346 	int	status;
347 
348 	if ((status = mod_remove(&modlinkage)) != 0)
349 		return (status);
350 	ddi_soft_state_fini(&vdc_state);
351 	return (0);
352 }
353 
354 static int
355 vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,  void *arg, void **resultp)
356 {
357 	_NOTE(ARGUNUSED(dip))
358 
359 	int	instance = VDCUNIT((dev_t)arg);
360 	vdc_t	*vdc = NULL;
361 
362 	switch (cmd) {
363 	case DDI_INFO_DEVT2DEVINFO:
364 		if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
365 			*resultp = NULL;
366 			return (DDI_FAILURE);
367 		}
368 		*resultp = vdc->dip;
369 		return (DDI_SUCCESS);
370 	case DDI_INFO_DEVT2INSTANCE:
371 		*resultp = (void *)(uintptr_t)instance;
372 		return (DDI_SUCCESS);
373 	default:
374 		*resultp = NULL;
375 		return (DDI_FAILURE);
376 	}
377 }
378 
379 static int
380 vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
381 {
382 	kt_did_t failfast_tid, ownership_tid;
383 	int	instance;
384 	int	rv;
385 	vdc_t	*vdc = NULL;
386 
387 	switch (cmd) {
388 	case DDI_DETACH:
389 		/* the real work happens below */
390 		break;
391 	case DDI_SUSPEND:
392 		/* nothing to do for this non-device */
393 		return (DDI_SUCCESS);
394 	default:
395 		return (DDI_FAILURE);
396 	}
397 
398 	ASSERT(cmd == DDI_DETACH);
399 	instance = ddi_get_instance(dip);
400 	DMSGX(1, "[%d] Entered\n", instance);
401 
402 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
403 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
404 		return (DDI_FAILURE);
405 	}
406 
407 	/*
408 	 * This function is called when vdc is detached or if it has failed to
409 	 * attach. In that case, the attach may have fail before the vdisk type
410 	 * has been set so we can't call vdc_is_opened(). However as the attach
411 	 * has failed, we know that the vdisk is not opened and we can safely
412 	 * detach.
413 	 */
414 	if (vdc->vdisk_type != VD_DISK_TYPE_UNK && vdc_is_opened(vdc)) {
415 		DMSG(vdc, 0, "[%d] Cannot detach: device is open", instance);
416 		return (DDI_FAILURE);
417 	}
418 
419 	if (vdc->dkio_flush_pending) {
420 		DMSG(vdc, 0,
421 		    "[%d] Cannot detach: %d outstanding DKIO flushes\n",
422 		    instance, vdc->dkio_flush_pending);
423 		return (DDI_FAILURE);
424 	}
425 
426 	if (vdc->validate_pending) {
427 		DMSG(vdc, 0,
428 		    "[%d] Cannot detach: %d outstanding validate request\n",
429 		    instance, vdc->validate_pending);
430 		return (DDI_FAILURE);
431 	}
432 
433 	DMSG(vdc, 0, "[%d] proceeding...\n", instance);
434 
435 	/* If we took ownership, release ownership */
436 	mutex_enter(&vdc->ownership_lock);
437 	if (vdc->ownership & VDC_OWNERSHIP_GRANTED) {
438 		rv = vdc_access_set(vdc, VD_ACCESS_SET_CLEAR, FKIOCTL);
439 		if (rv == 0) {
440 			vdc_ownership_update(vdc, VDC_OWNERSHIP_NONE);
441 		}
442 	}
443 	mutex_exit(&vdc->ownership_lock);
444 
445 	/* mark instance as detaching */
446 	vdc->lifecycle	= VDC_LC_DETACHING;
447 
448 	/*
449 	 * try and disable callbacks to prevent another handshake
450 	 */
451 	rv = ldc_set_cb_mode(vdc->ldc_handle, LDC_CB_DISABLE);
452 	DMSG(vdc, 0, "callback disabled (rv=%d)\n", rv);
453 
454 	if (vdc->initialized & VDC_THREAD) {
455 		mutex_enter(&vdc->read_lock);
456 		if ((vdc->read_state == VDC_READ_WAITING) ||
457 		    (vdc->read_state == VDC_READ_RESET)) {
458 			vdc->read_state = VDC_READ_RESET;
459 			cv_signal(&vdc->read_cv);
460 		}
461 
462 		mutex_exit(&vdc->read_lock);
463 
464 		/* wake up any thread waiting for connection to come online */
465 		mutex_enter(&vdc->lock);
466 		if (vdc->state == VDC_STATE_INIT_WAITING) {
467 			DMSG(vdc, 0,
468 			    "[%d] write reset - move to resetting state...\n",
469 			    instance);
470 			vdc->state = VDC_STATE_RESETTING;
471 			cv_signal(&vdc->initwait_cv);
472 		}
473 		mutex_exit(&vdc->lock);
474 
475 		/* now wait until state transitions to VDC_STATE_DETACH */
476 		thread_join(vdc->msg_proc_thr->t_did);
477 		ASSERT(vdc->state == VDC_STATE_DETACH);
478 		DMSG(vdc, 0, "[%d] Reset thread exit and join ..\n",
479 		    vdc->instance);
480 	}
481 
482 	mutex_enter(&vdc->lock);
483 
484 	if (vdc->initialized & VDC_DRING)
485 		vdc_destroy_descriptor_ring(vdc);
486 
487 	if (vdc->initialized & VDC_LDC)
488 		vdc_terminate_ldc(vdc);
489 
490 	if (vdc->failfast_thread) {
491 		failfast_tid = vdc->failfast_thread->t_did;
492 		vdc->failfast_interval = 0;
493 		cv_signal(&vdc->failfast_cv);
494 	} else {
495 		failfast_tid = 0;
496 	}
497 
498 	if (vdc->ownership & VDC_OWNERSHIP_WANTED) {
499 		ownership_tid = vdc->ownership_thread->t_did;
500 		vdc->ownership = VDC_OWNERSHIP_NONE;
501 		cv_signal(&vdc->ownership_cv);
502 	} else {
503 		ownership_tid = 0;
504 	}
505 
506 	mutex_exit(&vdc->lock);
507 
508 	if (failfast_tid != 0)
509 		thread_join(failfast_tid);
510 
511 	if (ownership_tid != 0)
512 		thread_join(ownership_tid);
513 
514 	if (vdc->initialized & VDC_MINOR) {
515 		ddi_prop_remove_all(dip);
516 		ddi_remove_minor_node(dip, NULL);
517 	}
518 
519 	if (vdc->io_stats) {
520 		kstat_delete(vdc->io_stats);
521 		vdc->io_stats = NULL;
522 	}
523 
524 	if (vdc->err_stats) {
525 		kstat_delete(vdc->err_stats);
526 		vdc->err_stats = NULL;
527 	}
528 
529 	if (vdc->initialized & VDC_LOCKS) {
530 		mutex_destroy(&vdc->lock);
531 		mutex_destroy(&vdc->read_lock);
532 		mutex_destroy(&vdc->ownership_lock);
533 		cv_destroy(&vdc->initwait_cv);
534 		cv_destroy(&vdc->dring_free_cv);
535 		cv_destroy(&vdc->membind_cv);
536 		cv_destroy(&vdc->sync_pending_cv);
537 		cv_destroy(&vdc->sync_blocked_cv);
538 		cv_destroy(&vdc->read_cv);
539 		cv_destroy(&vdc->running_cv);
540 		cv_destroy(&vdc->ownership_cv);
541 		cv_destroy(&vdc->failfast_cv);
542 		cv_destroy(&vdc->failfast_io_cv);
543 	}
544 
545 	if (vdc->minfo)
546 		kmem_free(vdc->minfo, sizeof (struct dk_minfo));
547 
548 	if (vdc->cinfo)
549 		kmem_free(vdc->cinfo, sizeof (struct dk_cinfo));
550 
551 	if (vdc->vtoc)
552 		kmem_free(vdc->vtoc, sizeof (struct vtoc));
553 
554 	if (vdc->geom)
555 		kmem_free(vdc->geom, sizeof (struct dk_geom));
556 
557 	if (vdc->devid) {
558 		ddi_devid_unregister(dip);
559 		ddi_devid_free(vdc->devid);
560 	}
561 
562 	if (vdc->initialized & VDC_SOFT_STATE)
563 		ddi_soft_state_free(vdc_state, instance);
564 
565 	DMSG(vdc, 0, "[%d] End %p\n", instance, (void *)vdc);
566 
567 	return (DDI_SUCCESS);
568 }
569 
570 
571 static int
572 vdc_do_attach(dev_info_t *dip)
573 {
574 	int		instance;
575 	vdc_t		*vdc = NULL;
576 	int		status;
577 	md_t		*mdp;
578 	mde_cookie_t	vd_node, vd_port;
579 
580 	ASSERT(dip != NULL);
581 
582 	instance = ddi_get_instance(dip);
583 	if (ddi_soft_state_zalloc(vdc_state, instance) != DDI_SUCCESS) {
584 		cmn_err(CE_NOTE, "[%d] Couldn't alloc state structure",
585 		    instance);
586 		return (DDI_FAILURE);
587 	}
588 
589 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
590 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
591 		return (DDI_FAILURE);
592 	}
593 
594 	/*
595 	 * We assign the value to initialized in this case to zero out the
596 	 * variable and then set bits in it to indicate what has been done
597 	 */
598 	vdc->initialized = VDC_SOFT_STATE;
599 
600 	vdc_hz_min_ldc_delay = drv_usectohz(vdc_min_timeout_ldc);
601 	vdc_hz_max_ldc_delay = drv_usectohz(vdc_max_timeout_ldc);
602 
603 	vdc->dip	= dip;
604 	vdc->instance	= instance;
605 	vdc->vdisk_type	= VD_DISK_TYPE_UNK;
606 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
607 	vdc->state	= VDC_STATE_INIT;
608 	vdc->lifecycle	= VDC_LC_ATTACHING;
609 	vdc->ldc_state	= 0;
610 	vdc->session_id = 0;
611 	vdc->block_size = DEV_BSIZE;
612 	vdc->max_xfer_sz = maxphys / DEV_BSIZE;
613 
614 	/*
615 	 * We assume, for now, that the vDisk server will export 'read'
616 	 * operations to us at a minimum (this is needed because of checks
617 	 * in vdc for supported operations early in the handshake process).
618 	 * The vDisk server will return ENOTSUP if this is not the case.
619 	 * The value will be overwritten during the attribute exchange with
620 	 * the bitmask of operations exported by server.
621 	 */
622 	vdc->operations = VD_OP_MASK_READ;
623 
624 	vdc->vtoc = NULL;
625 	vdc->geom = NULL;
626 	vdc->cinfo = NULL;
627 	vdc->minfo = NULL;
628 
629 	mutex_init(&vdc->lock, NULL, MUTEX_DRIVER, NULL);
630 	cv_init(&vdc->initwait_cv, NULL, CV_DRIVER, NULL);
631 	cv_init(&vdc->dring_free_cv, NULL, CV_DRIVER, NULL);
632 	cv_init(&vdc->membind_cv, NULL, CV_DRIVER, NULL);
633 	cv_init(&vdc->running_cv, NULL, CV_DRIVER, NULL);
634 
635 	vdc->threads_pending = 0;
636 	vdc->sync_op_pending = B_FALSE;
637 	vdc->sync_op_blocked = B_FALSE;
638 	cv_init(&vdc->sync_pending_cv, NULL, CV_DRIVER, NULL);
639 	cv_init(&vdc->sync_blocked_cv, NULL, CV_DRIVER, NULL);
640 
641 	mutex_init(&vdc->ownership_lock, NULL, MUTEX_DRIVER, NULL);
642 	cv_init(&vdc->ownership_cv, NULL, CV_DRIVER, NULL);
643 	cv_init(&vdc->failfast_cv, NULL, CV_DRIVER, NULL);
644 	cv_init(&vdc->failfast_io_cv, NULL, CV_DRIVER, NULL);
645 
646 	/* init blocking msg read functionality */
647 	mutex_init(&vdc->read_lock, NULL, MUTEX_DRIVER, NULL);
648 	cv_init(&vdc->read_cv, NULL, CV_DRIVER, NULL);
649 	vdc->read_state = VDC_READ_IDLE;
650 
651 	vdc->initialized |= VDC_LOCKS;
652 
653 	/* get device and port MD node for this disk instance */
654 	if (vdc_get_md_node(dip, &mdp, &vd_node, &vd_port) != 0) {
655 		cmn_err(CE_NOTE, "[%d] Could not get machine description node",
656 		    instance);
657 		return (DDI_FAILURE);
658 	}
659 
660 	/* set the connection timeout */
661 	if (vd_port == NULL || (md_get_prop_val(mdp, vd_port,
662 	    VDC_MD_TIMEOUT, &vdc->ctimeout) != 0)) {
663 		vdc->ctimeout = 0;
664 	}
665 
666 	/* initialise LDC channel which will be used to communicate with vds */
667 	status = vdc_do_ldc_init(vdc, mdp, vd_node);
668 
669 	(void) md_fini_handle(mdp);
670 
671 	if (status != 0) {
672 		cmn_err(CE_NOTE, "[%d] Couldn't initialize LDC", instance);
673 		goto return_status;
674 	}
675 
676 	/* initialize the thread responsible for managing state with server */
677 	vdc->msg_proc_thr = thread_create(NULL, 0, vdc_process_msg_thread,
678 	    vdc, 0, &p0, TS_RUN, minclsyspri);
679 	if (vdc->msg_proc_thr == NULL) {
680 		cmn_err(CE_NOTE, "[%d] Failed to create msg processing thread",
681 		    instance);
682 		return (DDI_FAILURE);
683 	}
684 
685 	vdc->initialized |= VDC_THREAD;
686 
687 	/* Create the kstats for saving the I/O statistics used by iostat(1M) */
688 	vdc_create_io_kstats(vdc);
689 	vdc_create_err_kstats(vdc);
690 
691 	atomic_inc_32(&vdc_instance_count);
692 
693 	/*
694 	 * Check the disk label. This will send requests and do the handshake.
695 	 * We don't really care about the disk label now. What we really need is
696 	 * the handshake do be done so that we know the type of the disk (slice
697 	 * or full disk) and the appropriate device nodes can be created.
698 	 */
699 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
700 	vdc->vtoc = kmem_zalloc(sizeof (struct vtoc), KM_SLEEP);
701 	vdc->geom = kmem_zalloc(sizeof (struct dk_geom), KM_SLEEP);
702 	vdc->minfo = kmem_zalloc(sizeof (struct dk_minfo), KM_SLEEP);
703 
704 	mutex_enter(&vdc->lock);
705 	(void) vdc_validate_geometry(vdc);
706 	mutex_exit(&vdc->lock);
707 
708 	/*
709 	 * Now that we have the device info we can create the
710 	 * device nodes and properties
711 	 */
712 	status = vdc_create_device_nodes(vdc);
713 	if (status) {
714 		DMSG(vdc, 0, "[%d] Failed to create device nodes",
715 		    instance);
716 		goto return_status;
717 	}
718 	status = vdc_create_device_nodes_props(vdc);
719 	if (status) {
720 		DMSG(vdc, 0, "[%d] Failed to create device nodes"
721 		    " properties (%d)", instance, status);
722 		goto return_status;
723 	}
724 
725 	/*
726 	 * Setup devid
727 	 */
728 	if (vdc_setup_devid(vdc)) {
729 		DMSG(vdc, 0, "[%d] No device id available\n", instance);
730 	}
731 
732 	/*
733 	 * Fill in the fields of the error statistics kstat that were not
734 	 * available when creating the kstat
735 	 */
736 	vdc_set_err_kstats(vdc);
737 
738 	ddi_report_dev(dip);
739 	vdc->lifecycle	= VDC_LC_ONLINE;
740 	DMSG(vdc, 0, "[%d] Attach tasks successful\n", instance);
741 
742 return_status:
743 	DMSG(vdc, 0, "[%d] Attach completed\n", instance);
744 	return (status);
745 }
746 
747 static int
748 vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
749 {
750 	int	status;
751 
752 	switch (cmd) {
753 	case DDI_ATTACH:
754 		if ((status = vdc_do_attach(dip)) != 0)
755 			(void) vdc_detach(dip, DDI_DETACH);
756 		return (status);
757 	case DDI_RESUME:
758 		/* nothing to do for this non-device */
759 		return (DDI_SUCCESS);
760 	default:
761 		return (DDI_FAILURE);
762 	}
763 }
764 
765 static int
766 vdc_do_ldc_init(vdc_t *vdc, md_t *mdp, mde_cookie_t vd_node)
767 {
768 	int			status = 0;
769 	ldc_status_t		ldc_state;
770 	ldc_attr_t		ldc_attr;
771 	uint64_t		ldc_id = 0;
772 
773 	ASSERT(vdc != NULL);
774 
775 	vdc->initialized |= VDC_LDC;
776 
777 	if ((status = vdc_get_ldc_id(mdp, vd_node, &ldc_id)) != 0) {
778 		DMSG(vdc, 0, "[%d] Failed to get LDC channel ID property",
779 		    vdc->instance);
780 		return (EIO);
781 	}
782 
783 	DMSGX(0, "[%d] LDC id is 0x%lx\n", vdc->instance, ldc_id);
784 
785 	vdc->ldc_id = ldc_id;
786 
787 	ldc_attr.devclass = LDC_DEV_BLK;
788 	ldc_attr.instance = vdc->instance;
789 	ldc_attr.mode = LDC_MODE_UNRELIABLE;	/* unreliable transport */
790 	ldc_attr.mtu = VD_LDC_MTU;
791 
792 	if ((vdc->initialized & VDC_LDC_INIT) == 0) {
793 		status = ldc_init(ldc_id, &ldc_attr, &vdc->ldc_handle);
794 		if (status != 0) {
795 			DMSG(vdc, 0, "[%d] ldc_init(chan %ld) returned %d",
796 			    vdc->instance, ldc_id, status);
797 			return (status);
798 		}
799 		vdc->initialized |= VDC_LDC_INIT;
800 	}
801 	status = ldc_status(vdc->ldc_handle, &ldc_state);
802 	if (status != 0) {
803 		DMSG(vdc, 0, "[%d] Cannot discover LDC status [err=%d]",
804 		    vdc->instance, status);
805 		return (status);
806 	}
807 	vdc->ldc_state = ldc_state;
808 
809 	if ((vdc->initialized & VDC_LDC_CB) == 0) {
810 		status = ldc_reg_callback(vdc->ldc_handle, vdc_handle_cb,
811 		    (caddr_t)vdc);
812 		if (status != 0) {
813 			DMSG(vdc, 0, "[%d] LDC callback reg. failed (%d)",
814 			    vdc->instance, status);
815 			return (status);
816 		}
817 		vdc->initialized |= VDC_LDC_CB;
818 	}
819 
820 	vdc->initialized |= VDC_LDC;
821 
822 	/*
823 	 * At this stage we have initialised LDC, we will now try and open
824 	 * the connection.
825 	 */
826 	if (vdc->ldc_state == LDC_INIT) {
827 		status = ldc_open(vdc->ldc_handle);
828 		if (status != 0) {
829 			DMSG(vdc, 0, "[%d] ldc_open(chan %ld) returned %d",
830 			    vdc->instance, vdc->ldc_id, status);
831 			return (status);
832 		}
833 		vdc->initialized |= VDC_LDC_OPEN;
834 	}
835 
836 	return (status);
837 }
838 
839 static int
840 vdc_start_ldc_connection(vdc_t *vdc)
841 {
842 	int		status = 0;
843 
844 	ASSERT(vdc != NULL);
845 
846 	ASSERT(MUTEX_HELD(&vdc->lock));
847 
848 	status = vdc_do_ldc_up(vdc);
849 
850 	DMSG(vdc, 0, "[%d] Finished bringing up LDC\n", vdc->instance);
851 
852 	return (status);
853 }
854 
855 static int
856 vdc_stop_ldc_connection(vdc_t *vdcp)
857 {
858 	int	status;
859 
860 	DMSG(vdcp, 0, ": Resetting connection to vDisk server : state %d\n",
861 	    vdcp->state);
862 
863 	status = ldc_down(vdcp->ldc_handle);
864 	DMSG(vdcp, 0, "ldc_down() = %d\n", status);
865 
866 	vdcp->initialized &= ~VDC_HANDSHAKE;
867 	DMSG(vdcp, 0, "initialized=%x\n", vdcp->initialized);
868 
869 	return (status);
870 }
871 
872 static void
873 vdc_create_io_kstats(vdc_t *vdc)
874 {
875 	if (vdc->io_stats != NULL) {
876 		DMSG(vdc, 0, "[%d] I/O kstat already exists\n", vdc->instance);
877 		return;
878 	}
879 
880 	vdc->io_stats = kstat_create(VDC_DRIVER_NAME, vdc->instance, NULL,
881 	    "disk", KSTAT_TYPE_IO, 1, KSTAT_FLAG_PERSISTENT);
882 	if (vdc->io_stats != NULL) {
883 		vdc->io_stats->ks_lock = &vdc->lock;
884 		kstat_install(vdc->io_stats);
885 	} else {
886 		cmn_err(CE_NOTE, "[%d] Failed to create kstat: I/O statistics"
887 		    " will not be gathered", vdc->instance);
888 	}
889 }
890 
891 static void
892 vdc_create_err_kstats(vdc_t *vdc)
893 {
894 	vd_err_stats_t	*stp;
895 	char	kstatmodule_err[KSTAT_STRLEN];
896 	char	kstatname[KSTAT_STRLEN];
897 	int	ndata = (sizeof (vd_err_stats_t) / sizeof (kstat_named_t));
898 	int	instance = vdc->instance;
899 
900 	if (vdc->err_stats != NULL) {
901 		DMSG(vdc, 0, "[%d] ERR kstat already exists\n", vdc->instance);
902 		return;
903 	}
904 
905 	(void) snprintf(kstatmodule_err, sizeof (kstatmodule_err),
906 	    "%serr", VDC_DRIVER_NAME);
907 	(void) snprintf(kstatname, sizeof (kstatname),
908 	    "%s%d,err", VDC_DRIVER_NAME, instance);
909 
910 	vdc->err_stats = kstat_create(kstatmodule_err, instance, kstatname,
911 	    "device_error", KSTAT_TYPE_NAMED, ndata, KSTAT_FLAG_PERSISTENT);
912 
913 	if (vdc->err_stats == NULL) {
914 		cmn_err(CE_NOTE, "[%d] Failed to create kstat: Error statistics"
915 		    " will not be gathered", instance);
916 		return;
917 	}
918 
919 	stp = (vd_err_stats_t *)vdc->err_stats->ks_data;
920 	kstat_named_init(&stp->vd_softerrs,	"Soft Errors",
921 	    KSTAT_DATA_UINT32);
922 	kstat_named_init(&stp->vd_transerrs,	"Transport Errors",
923 	    KSTAT_DATA_UINT32);
924 	kstat_named_init(&stp->vd_protoerrs,	"Protocol Errors",
925 	    KSTAT_DATA_UINT32);
926 	kstat_named_init(&stp->vd_vid,		"Vendor",
927 	    KSTAT_DATA_CHAR);
928 	kstat_named_init(&stp->vd_pid,		"Product",
929 	    KSTAT_DATA_CHAR);
930 	kstat_named_init(&stp->vd_capacity,	"Size",
931 	    KSTAT_DATA_ULONGLONG);
932 
933 	vdc->err_stats->ks_update  = nulldev;
934 
935 	kstat_install(vdc->err_stats);
936 }
937 
938 static void
939 vdc_set_err_kstats(vdc_t *vdc)
940 {
941 	vd_err_stats_t  *stp;
942 
943 	if (vdc->err_stats == NULL)
944 		return;
945 
946 	mutex_enter(&vdc->lock);
947 
948 	stp = (vd_err_stats_t *)vdc->err_stats->ks_data;
949 	ASSERT(stp != NULL);
950 
951 	stp->vd_capacity.value.ui64 = vdc->vdisk_size * vdc->block_size;
952 	(void) strcpy(stp->vd_vid.value.c, "SUN");
953 	(void) strcpy(stp->vd_pid.value.c, "VDSK");
954 
955 	mutex_exit(&vdc->lock);
956 }
957 
958 static int
959 vdc_create_device_nodes_efi(vdc_t *vdc)
960 {
961 	ddi_remove_minor_node(vdc->dip, "h");
962 	ddi_remove_minor_node(vdc->dip, "h,raw");
963 
964 	if (ddi_create_minor_node(vdc->dip, "wd", S_IFBLK,
965 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
966 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
967 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'wd'",
968 		    vdc->instance);
969 		return (EIO);
970 	}
971 
972 	/* if any device node is created we set this flag */
973 	vdc->initialized |= VDC_MINOR;
974 
975 	if (ddi_create_minor_node(vdc->dip, "wd,raw", S_IFCHR,
976 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
977 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
978 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'wd,raw'",
979 		    vdc->instance);
980 		return (EIO);
981 	}
982 
983 	return (0);
984 }
985 
986 static int
987 vdc_create_device_nodes_vtoc(vdc_t *vdc)
988 {
989 	ddi_remove_minor_node(vdc->dip, "wd");
990 	ddi_remove_minor_node(vdc->dip, "wd,raw");
991 
992 	if (ddi_create_minor_node(vdc->dip, "h", S_IFBLK,
993 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
994 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
995 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'h'",
996 		    vdc->instance);
997 		return (EIO);
998 	}
999 
1000 	/* if any device node is created we set this flag */
1001 	vdc->initialized |= VDC_MINOR;
1002 
1003 	if (ddi_create_minor_node(vdc->dip, "h,raw", S_IFCHR,
1004 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
1005 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
1006 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'h,raw'",
1007 		    vdc->instance);
1008 		return (EIO);
1009 	}
1010 
1011 	return (0);
1012 }
1013 
1014 /*
1015  * Function:
1016  *	vdc_create_device_nodes
1017  *
1018  * Description:
1019  *	This function creates the block and character device nodes under
1020  *	/devices along with the node properties. It is called as part of
1021  *	the attach(9E) of the instance during the handshake with vds after
1022  *	vds has sent the attributes to vdc.
1023  *
1024  *	If the device is of type VD_DISK_TYPE_SLICE then the minor node
1025  *	of 2 is used in keeping with the Solaris convention that slice 2
1026  *	refers to a whole disk. Slices start at 'a'
1027  *
1028  * Parameters:
1029  *	vdc 		- soft state pointer
1030  *
1031  * Return Values
1032  *	0		- Success
1033  *	EIO		- Failed to create node
1034  *	EINVAL		- Unknown type of disk exported
1035  */
1036 static int
1037 vdc_create_device_nodes(vdc_t *vdc)
1038 {
1039 	char		name[sizeof ("s,raw")];
1040 	dev_info_t	*dip = NULL;
1041 	int		instance, status;
1042 	int		num_slices = 1;
1043 	int		i;
1044 
1045 	ASSERT(vdc != NULL);
1046 
1047 	instance = vdc->instance;
1048 	dip = vdc->dip;
1049 
1050 	switch (vdc->vdisk_type) {
1051 	case VD_DISK_TYPE_DISK:
1052 		num_slices = V_NUMPAR;
1053 		break;
1054 	case VD_DISK_TYPE_SLICE:
1055 		num_slices = 1;
1056 		break;
1057 	case VD_DISK_TYPE_UNK:
1058 	default:
1059 		return (EINVAL);
1060 	}
1061 
1062 	/*
1063 	 * Minor nodes are different for EFI disks: EFI disks do not have
1064 	 * a minor node 'g' for the minor number corresponding to slice
1065 	 * VD_EFI_WD_SLICE (slice 7) instead they have a minor node 'wd'
1066 	 * representing the whole disk.
1067 	 */
1068 	for (i = 0; i < num_slices; i++) {
1069 
1070 		if (i == VD_EFI_WD_SLICE) {
1071 			if (vdc->vdisk_label == VD_DISK_LABEL_EFI)
1072 				status = vdc_create_device_nodes_efi(vdc);
1073 			else
1074 				status = vdc_create_device_nodes_vtoc(vdc);
1075 			if (status != 0)
1076 				return (status);
1077 			continue;
1078 		}
1079 
1080 		(void) snprintf(name, sizeof (name), "%c", 'a' + i);
1081 		if (ddi_create_minor_node(dip, name, S_IFBLK,
1082 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
1083 			cmn_err(CE_NOTE, "[%d] Couldn't add block node '%s'",
1084 			    instance, name);
1085 			return (EIO);
1086 		}
1087 
1088 		/* if any device node is created we set this flag */
1089 		vdc->initialized |= VDC_MINOR;
1090 
1091 		(void) snprintf(name, sizeof (name), "%c%s", 'a' + i, ",raw");
1092 
1093 		if (ddi_create_minor_node(dip, name, S_IFCHR,
1094 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
1095 			cmn_err(CE_NOTE, "[%d] Couldn't add raw node '%s'",
1096 			    instance, name);
1097 			return (EIO);
1098 		}
1099 	}
1100 
1101 	return (0);
1102 }
1103 
1104 /*
1105  * Function:
1106  *	vdc_create_device_nodes_props
1107  *
1108  * Description:
1109  *	This function creates the block and character device nodes under
1110  *	/devices along with the node properties. It is called as part of
1111  *	the attach(9E) of the instance during the handshake with vds after
1112  *	vds has sent the attributes to vdc.
1113  *
1114  * Parameters:
1115  *	vdc 		- soft state pointer
1116  *
1117  * Return Values
1118  *	0		- Success
1119  *	EIO		- Failed to create device node property
1120  *	EINVAL		- Unknown type of disk exported
1121  */
1122 static int
1123 vdc_create_device_nodes_props(vdc_t *vdc)
1124 {
1125 	dev_info_t	*dip = NULL;
1126 	int		instance;
1127 	int		num_slices = 1;
1128 	int64_t		size = 0;
1129 	dev_t		dev;
1130 	int		rv;
1131 	int		i;
1132 
1133 	ASSERT(vdc != NULL);
1134 
1135 	instance = vdc->instance;
1136 	dip = vdc->dip;
1137 
1138 	switch (vdc->vdisk_type) {
1139 	case VD_DISK_TYPE_DISK:
1140 		num_slices = V_NUMPAR;
1141 		break;
1142 	case VD_DISK_TYPE_SLICE:
1143 		num_slices = 1;
1144 		break;
1145 	case VD_DISK_TYPE_UNK:
1146 	default:
1147 		return (EINVAL);
1148 	}
1149 
1150 	if (vdc->vdisk_label == VD_DISK_LABEL_UNK) {
1151 		/* remove all properties */
1152 		for (i = 0; i < num_slices; i++) {
1153 			dev = makedevice(ddi_driver_major(dip),
1154 			    VD_MAKE_DEV(instance, i));
1155 			(void) ddi_prop_remove(dev, dip, VDC_SIZE_PROP_NAME);
1156 			(void) ddi_prop_remove(dev, dip, VDC_NBLOCKS_PROP_NAME);
1157 		}
1158 		return (0);
1159 	}
1160 
1161 	for (i = 0; i < num_slices; i++) {
1162 		dev = makedevice(ddi_driver_major(dip),
1163 		    VD_MAKE_DEV(instance, i));
1164 
1165 		size = vdc->slice[i].nblocks * vdc->block_size;
1166 		DMSG(vdc, 0, "[%d] sz %ld (%ld Mb)  p_size %lx\n",
1167 		    instance, size, size / (1024 * 1024),
1168 		    vdc->slice[i].nblocks);
1169 
1170 		rv = ddi_prop_update_int64(dev, dip, VDC_SIZE_PROP_NAME, size);
1171 		if (rv != DDI_PROP_SUCCESS) {
1172 			cmn_err(CE_NOTE, "[%d] Couldn't add '%s' prop of [%ld]",
1173 			    instance, VDC_SIZE_PROP_NAME, size);
1174 			return (EIO);
1175 		}
1176 
1177 		rv = ddi_prop_update_int64(dev, dip, VDC_NBLOCKS_PROP_NAME,
1178 		    lbtodb(size));
1179 		if (rv != DDI_PROP_SUCCESS) {
1180 			cmn_err(CE_NOTE, "[%d] Couldn't add '%s' prop [%llu]",
1181 			    instance, VDC_NBLOCKS_PROP_NAME, lbtodb(size));
1182 			return (EIO);
1183 		}
1184 	}
1185 
1186 	return (0);
1187 }
1188 
1189 /*
1190  * Function:
1191  *	vdc_is_opened
1192  *
1193  * Description:
1194  *	This function checks if any slice of a given virtual disk is
1195  *	currently opened.
1196  *
1197  * Parameters:
1198  *	vdc 		- soft state pointer
1199  *
1200  * Return Values
1201  *	B_TRUE		- at least one slice is opened.
1202  *	B_FALSE		- no slice is opened.
1203  */
1204 static boolean_t
1205 vdc_is_opened(vdc_t *vdc)
1206 {
1207 	int i, nslices;
1208 
1209 	switch (vdc->vdisk_type) {
1210 	case VD_DISK_TYPE_DISK:
1211 		nslices = V_NUMPAR;
1212 		break;
1213 	case VD_DISK_TYPE_SLICE:
1214 		nslices = 1;
1215 		break;
1216 	case VD_DISK_TYPE_UNK:
1217 	default:
1218 		ASSERT(0);
1219 	}
1220 
1221 	/* check if there's any layered open */
1222 	for (i = 0; i < nslices; i++) {
1223 		if (vdc->open_lyr[i] > 0)
1224 			return (B_TRUE);
1225 	}
1226 
1227 	/* check if there is any other kind of open */
1228 	for (i = 0; i < OTYPCNT; i++) {
1229 		if (vdc->open[i] != 0)
1230 			return (B_TRUE);
1231 	}
1232 
1233 	return (B_FALSE);
1234 }
1235 
1236 static int
1237 vdc_mark_opened(vdc_t *vdc, int slice, int flag, int otyp)
1238 {
1239 	uint8_t slicemask;
1240 	int i;
1241 
1242 	ASSERT(otyp < OTYPCNT);
1243 	ASSERT(slice < V_NUMPAR);
1244 	ASSERT(MUTEX_HELD(&vdc->lock));
1245 
1246 	slicemask = 1 << slice;
1247 
1248 	/* check if slice is already exclusively opened */
1249 	if (vdc->open_excl & slicemask)
1250 		return (EBUSY);
1251 
1252 	/* if open exclusive, check if slice is already opened */
1253 	if (flag & FEXCL) {
1254 		if (vdc->open_lyr[slice] > 0)
1255 			return (EBUSY);
1256 		for (i = 0; i < OTYPCNT; i++) {
1257 			if (vdc->open[i] & slicemask)
1258 				return (EBUSY);
1259 		}
1260 		vdc->open_excl |= slicemask;
1261 	}
1262 
1263 	/* mark slice as opened */
1264 	if (otyp == OTYP_LYR) {
1265 		vdc->open_lyr[slice]++;
1266 	} else {
1267 		vdc->open[otyp] |= slicemask;
1268 	}
1269 
1270 	return (0);
1271 }
1272 
1273 static void
1274 vdc_mark_closed(vdc_t *vdc, int slice, int flag, int otyp)
1275 {
1276 	uint8_t slicemask;
1277 
1278 	ASSERT(otyp < OTYPCNT);
1279 	ASSERT(slice < V_NUMPAR);
1280 	ASSERT(MUTEX_HELD(&vdc->lock));
1281 
1282 	slicemask = 1 << slice;
1283 
1284 	if (otyp == OTYP_LYR) {
1285 		ASSERT(vdc->open_lyr[slice] > 0);
1286 		vdc->open_lyr[slice]--;
1287 	} else {
1288 		vdc->open[otyp] &= ~slicemask;
1289 	}
1290 
1291 	if (flag & FEXCL)
1292 		vdc->open_excl &= ~slicemask;
1293 }
1294 
1295 static int
1296 vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred)
1297 {
1298 	_NOTE(ARGUNUSED(cred))
1299 
1300 	int	instance;
1301 	int	slice, status = 0;
1302 	vdc_t	*vdc;
1303 
1304 	ASSERT(dev != NULL);
1305 	instance = VDCUNIT(*dev);
1306 
1307 	if (otyp >= OTYPCNT)
1308 		return (EINVAL);
1309 
1310 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1311 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1312 		return (ENXIO);
1313 	}
1314 
1315 	DMSG(vdc, 0, "minor = %d flag = %x, otyp = %x\n",
1316 	    getminor(*dev), flag, otyp);
1317 
1318 	slice = VDCPART(*dev);
1319 
1320 	mutex_enter(&vdc->lock);
1321 
1322 	status = vdc_mark_opened(vdc, slice, flag, otyp);
1323 
1324 	if (status != 0) {
1325 		mutex_exit(&vdc->lock);
1326 		return (status);
1327 	}
1328 
1329 	if (flag & (FNDELAY | FNONBLOCK)) {
1330 
1331 		/* don't resubmit a validate request if there's already one */
1332 		if (vdc->validate_pending > 0) {
1333 			mutex_exit(&vdc->lock);
1334 			return (0);
1335 		}
1336 
1337 		/* call vdc_validate() asynchronously to avoid blocking */
1338 		if (taskq_dispatch(system_taskq, vdc_validate_task,
1339 		    (void *)vdc, TQ_NOSLEEP) == NULL) {
1340 			vdc_mark_closed(vdc, slice, flag, otyp);
1341 			mutex_exit(&vdc->lock);
1342 			return (ENXIO);
1343 		}
1344 
1345 		vdc->validate_pending++;
1346 		mutex_exit(&vdc->lock);
1347 		return (0);
1348 	}
1349 
1350 	mutex_exit(&vdc->lock);
1351 
1352 	vdc_validate(vdc);
1353 
1354 	mutex_enter(&vdc->lock);
1355 
1356 	if (vdc->vdisk_label == VD_DISK_LABEL_UNK ||
1357 	    vdc->slice[slice].nblocks == 0) {
1358 		vdc_mark_closed(vdc, slice, flag, otyp);
1359 		status = EIO;
1360 	}
1361 
1362 	mutex_exit(&vdc->lock);
1363 
1364 	return (status);
1365 }
1366 
1367 static int
1368 vdc_close(dev_t dev, int flag, int otyp, cred_t *cred)
1369 {
1370 	_NOTE(ARGUNUSED(cred))
1371 
1372 	int	instance;
1373 	int	slice;
1374 	int	rv, rval;
1375 	vdc_t	*vdc;
1376 
1377 	instance = VDCUNIT(dev);
1378 
1379 	if (otyp >= OTYPCNT)
1380 		return (EINVAL);
1381 
1382 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1383 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1384 		return (ENXIO);
1385 	}
1386 
1387 	DMSG(vdc, 0, "[%d] flag = %x, otyp = %x\n", instance, flag, otyp);
1388 
1389 	slice = VDCPART(dev);
1390 
1391 	/*
1392 	 * Attempt to flush the W$ on a close operation. If this is
1393 	 * not a supported IOCTL command or the backing device is read-only
1394 	 * do not fail the close operation.
1395 	 */
1396 	rv = vd_process_ioctl(dev, DKIOCFLUSHWRITECACHE, NULL, FKIOCTL, &rval);
1397 
1398 	if (rv != 0 && rv != ENOTSUP && rv != ENOTTY && rv != EROFS) {
1399 		DMSG(vdc, 0, "[%d] flush failed with error %d on close\n",
1400 		    instance, rv);
1401 		return (EIO);
1402 	}
1403 
1404 	mutex_enter(&vdc->lock);
1405 	vdc_mark_closed(vdc, slice, flag, otyp);
1406 	mutex_exit(&vdc->lock);
1407 
1408 	return (0);
1409 }
1410 
1411 static int
1412 vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
1413 {
1414 	_NOTE(ARGUNUSED(credp))
1415 
1416 	return (vd_process_ioctl(dev, cmd, (caddr_t)arg, mode, rvalp));
1417 }
1418 
1419 static int
1420 vdc_print(dev_t dev, char *str)
1421 {
1422 	cmn_err(CE_NOTE, "vdc%d:  %s", VDCUNIT(dev), str);
1423 	return (0);
1424 }
1425 
1426 static int
1427 vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk)
1428 {
1429 	int	rv;
1430 	size_t	nbytes = nblk * DEV_BSIZE;
1431 	int	instance = VDCUNIT(dev);
1432 	vdc_t	*vdc = NULL;
1433 
1434 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1435 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1436 		return (ENXIO);
1437 	}
1438 
1439 	DMSG(vdc, 2, "[%d] dump %ld bytes at block 0x%lx : addr=0x%p\n",
1440 	    instance, nbytes, blkno, (void *)addr);
1441 	rv = vdc_send_request(vdc, VD_OP_BWRITE, addr, nbytes,
1442 	    VDCPART(dev), blkno, CB_STRATEGY, 0, VIO_write_dir);
1443 	if (rv) {
1444 		DMSG(vdc, 0, "Failed to do a disk dump (err=%d)\n", rv);
1445 		return (rv);
1446 	}
1447 
1448 	if (ddi_in_panic())
1449 		(void) vdc_drain_response(vdc);
1450 
1451 	DMSG(vdc, 0, "[%d] End\n", instance);
1452 
1453 	return (0);
1454 }
1455 
1456 /* -------------------------------------------------------------------------- */
1457 
1458 /*
1459  * Disk access routines
1460  *
1461  */
1462 
1463 /*
1464  * vdc_strategy()
1465  *
1466  * Return Value:
1467  *	0:	As per strategy(9E), the strategy() function must return 0
1468  *		[ bioerror(9f) sets b_flags to the proper error code ]
1469  */
1470 static int
1471 vdc_strategy(struct buf *buf)
1472 {
1473 	int	rv = -1;
1474 	vdc_t	*vdc = NULL;
1475 	int	instance = VDCUNIT(buf->b_edev);
1476 	int	op = (buf->b_flags & B_READ) ? VD_OP_BREAD : VD_OP_BWRITE;
1477 	int	slice;
1478 
1479 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1480 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1481 		bioerror(buf, ENXIO);
1482 		biodone(buf);
1483 		return (0);
1484 	}
1485 
1486 	DMSG(vdc, 2, "[%d] %s %ld bytes at block %llx : b_addr=0x%p\n",
1487 	    instance, (buf->b_flags & B_READ) ? "Read" : "Write",
1488 	    buf->b_bcount, buf->b_lblkno, (void *)buf->b_un.b_addr);
1489 
1490 	bp_mapin(buf);
1491 
1492 	if ((long)buf->b_private == VD_SLICE_NONE) {
1493 		/* I/O using an absolute disk offset */
1494 		slice = VD_SLICE_NONE;
1495 	} else {
1496 		slice = VDCPART(buf->b_edev);
1497 	}
1498 
1499 	rv = vdc_send_request(vdc, op, (caddr_t)buf->b_un.b_addr,
1500 	    buf->b_bcount, slice, buf->b_lblkno,
1501 	    CB_STRATEGY, buf, (op == VD_OP_BREAD) ? VIO_read_dir :
1502 	    VIO_write_dir);
1503 
1504 	/*
1505 	 * If the request was successfully sent, the strategy call returns and
1506 	 * the ACK handler calls the bioxxx functions when the vDisk server is
1507 	 * done otherwise we handle the error here.
1508 	 */
1509 	if (rv) {
1510 		DMSG(vdc, 0, "Failed to read/write (err=%d)\n", rv);
1511 		bioerror(buf, rv);
1512 		biodone(buf);
1513 	}
1514 
1515 	return (0);
1516 }
1517 
1518 /*
1519  * Function:
1520  *	vdc_min
1521  *
1522  * Description:
1523  *	Routine to limit the size of a data transfer. Used in
1524  *	conjunction with physio(9F).
1525  *
1526  * Arguments:
1527  *	bp - pointer to the indicated buf(9S) struct.
1528  *
1529  */
1530 static void
1531 vdc_min(struct buf *bufp)
1532 {
1533 	vdc_t	*vdc = NULL;
1534 	int	instance = VDCUNIT(bufp->b_edev);
1535 
1536 	vdc = ddi_get_soft_state(vdc_state, instance);
1537 	VERIFY(vdc != NULL);
1538 
1539 	if (bufp->b_bcount > (vdc->max_xfer_sz * vdc->block_size)) {
1540 		bufp->b_bcount = vdc->max_xfer_sz * vdc->block_size;
1541 	}
1542 }
1543 
1544 static int
1545 vdc_read(dev_t dev, struct uio *uio, cred_t *cred)
1546 {
1547 	_NOTE(ARGUNUSED(cred))
1548 
1549 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1550 	return (physio(vdc_strategy, NULL, dev, B_READ, vdc_min, uio));
1551 }
1552 
1553 static int
1554 vdc_write(dev_t dev, struct uio *uio, cred_t *cred)
1555 {
1556 	_NOTE(ARGUNUSED(cred))
1557 
1558 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1559 	return (physio(vdc_strategy, NULL, dev, B_WRITE, vdc_min, uio));
1560 }
1561 
1562 static int
1563 vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred)
1564 {
1565 	_NOTE(ARGUNUSED(cred))
1566 
1567 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1568 	return (aphysio(vdc_strategy, anocancel, dev, B_READ, vdc_min, aio));
1569 }
1570 
1571 static int
1572 vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred)
1573 {
1574 	_NOTE(ARGUNUSED(cred))
1575 
1576 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1577 	return (aphysio(vdc_strategy, anocancel, dev, B_WRITE, vdc_min, aio));
1578 }
1579 
1580 
1581 /* -------------------------------------------------------------------------- */
1582 
1583 /*
1584  * Handshake support
1585  */
1586 
1587 
1588 /*
1589  * Function:
1590  *	vdc_init_ver_negotiation()
1591  *
1592  * Description:
1593  *
1594  * Arguments:
1595  *	vdc	- soft state pointer for this instance of the device driver.
1596  *
1597  * Return Code:
1598  *	0	- Success
1599  */
1600 static int
1601 vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver)
1602 {
1603 	vio_ver_msg_t	pkt;
1604 	size_t		msglen = sizeof (pkt);
1605 	int		status = -1;
1606 
1607 	ASSERT(vdc != NULL);
1608 	ASSERT(mutex_owned(&vdc->lock));
1609 
1610 	DMSG(vdc, 0, "[%d] Entered.\n", vdc->instance);
1611 
1612 	/*
1613 	 * set the Session ID to a unique value
1614 	 * (the lower 32 bits of the clock tick)
1615 	 */
1616 	vdc->session_id = ((uint32_t)gettick() & 0xffffffff);
1617 	DMSG(vdc, 0, "[%d] Set SID to 0x%lx\n", vdc->instance, vdc->session_id);
1618 
1619 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1620 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1621 	pkt.tag.vio_subtype_env = VIO_VER_INFO;
1622 	pkt.tag.vio_sid = vdc->session_id;
1623 	pkt.dev_class = VDEV_DISK;
1624 	pkt.ver_major = ver.major;
1625 	pkt.ver_minor = ver.minor;
1626 
1627 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1628 	DMSG(vdc, 0, "[%d] Ver info sent (status = %d)\n",
1629 	    vdc->instance, status);
1630 	if ((status != 0) || (msglen != sizeof (vio_ver_msg_t))) {
1631 		DMSG(vdc, 0, "[%d] Failed to send Ver negotiation info: "
1632 		    "id(%lx) rv(%d) size(%ld)", vdc->instance, vdc->ldc_handle,
1633 		    status, msglen);
1634 		if (msglen != sizeof (vio_ver_msg_t))
1635 			status = ENOMSG;
1636 	}
1637 
1638 	return (status);
1639 }
1640 
1641 /*
1642  * Function:
1643  *	vdc_ver_negotiation()
1644  *
1645  * Description:
1646  *
1647  * Arguments:
1648  *	vdcp	- soft state pointer for this instance of the device driver.
1649  *
1650  * Return Code:
1651  *	0	- Success
1652  */
1653 static int
1654 vdc_ver_negotiation(vdc_t *vdcp)
1655 {
1656 	vio_msg_t vio_msg;
1657 	int status;
1658 
1659 	if (status = vdc_init_ver_negotiation(vdcp, vdc_version[0]))
1660 		return (status);
1661 
1662 	/* release lock and wait for response */
1663 	mutex_exit(&vdcp->lock);
1664 	status = vdc_wait_for_response(vdcp, &vio_msg);
1665 	mutex_enter(&vdcp->lock);
1666 	if (status) {
1667 		DMSG(vdcp, 0,
1668 		    "[%d] Failed waiting for Ver negotiation response, rv(%d)",
1669 		    vdcp->instance, status);
1670 		return (status);
1671 	}
1672 
1673 	/* check type and sub_type ... */
1674 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1675 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1676 		DMSG(vdcp, 0, "[%d] Invalid ver negotiation response\n",
1677 		    vdcp->instance);
1678 		return (EPROTO);
1679 	}
1680 
1681 	return (vdc_handle_ver_msg(vdcp, (vio_ver_msg_t *)&vio_msg));
1682 }
1683 
1684 /*
1685  * Function:
1686  *	vdc_init_attr_negotiation()
1687  *
1688  * Description:
1689  *
1690  * Arguments:
1691  *	vdc	- soft state pointer for this instance of the device driver.
1692  *
1693  * Return Code:
1694  *	0	- Success
1695  */
1696 static int
1697 vdc_init_attr_negotiation(vdc_t *vdc)
1698 {
1699 	vd_attr_msg_t	pkt;
1700 	size_t		msglen = sizeof (pkt);
1701 	int		status;
1702 
1703 	ASSERT(vdc != NULL);
1704 	ASSERT(mutex_owned(&vdc->lock));
1705 
1706 	DMSG(vdc, 0, "[%d] entered\n", vdc->instance);
1707 
1708 	/* fill in tag */
1709 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1710 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1711 	pkt.tag.vio_subtype_env = VIO_ATTR_INFO;
1712 	pkt.tag.vio_sid = vdc->session_id;
1713 	/* fill in payload */
1714 	pkt.max_xfer_sz = vdc->max_xfer_sz;
1715 	pkt.vdisk_block_size = vdc->block_size;
1716 	pkt.xfer_mode = VIO_DRING_MODE_V1_0;
1717 	pkt.operations = 0;	/* server will set bits of valid operations */
1718 	pkt.vdisk_type = 0;	/* server will set to valid device type */
1719 	pkt.vdisk_media = 0;	/* server will set to valid media type */
1720 	pkt.vdisk_size = 0;	/* server will set to valid size */
1721 
1722 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1723 	DMSG(vdc, 0, "Attr info sent (status = %d)\n", status);
1724 
1725 	if ((status != 0) || (msglen != sizeof (vio_ver_msg_t))) {
1726 		DMSG(vdc, 0, "[%d] Failed to send Attr negotiation info: "
1727 		    "id(%lx) rv(%d) size(%ld)", vdc->instance, vdc->ldc_handle,
1728 		    status, msglen);
1729 		if (msglen != sizeof (vio_ver_msg_t))
1730 			status = ENOMSG;
1731 	}
1732 
1733 	return (status);
1734 }
1735 
1736 /*
1737  * Function:
1738  *	vdc_attr_negotiation()
1739  *
1740  * Description:
1741  *
1742  * Arguments:
1743  *	vdc	- soft state pointer for this instance of the device driver.
1744  *
1745  * Return Code:
1746  *	0	- Success
1747  */
1748 static int
1749 vdc_attr_negotiation(vdc_t *vdcp)
1750 {
1751 	int status;
1752 	vio_msg_t vio_msg;
1753 
1754 	if (status = vdc_init_attr_negotiation(vdcp))
1755 		return (status);
1756 
1757 	/* release lock and wait for response */
1758 	mutex_exit(&vdcp->lock);
1759 	status = vdc_wait_for_response(vdcp, &vio_msg);
1760 	mutex_enter(&vdcp->lock);
1761 	if (status) {
1762 		DMSG(vdcp, 0,
1763 		    "[%d] Failed waiting for Attr negotiation response, rv(%d)",
1764 		    vdcp->instance, status);
1765 		return (status);
1766 	}
1767 
1768 	/* check type and sub_type ... */
1769 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1770 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1771 		DMSG(vdcp, 0, "[%d] Invalid attr negotiation response\n",
1772 		    vdcp->instance);
1773 		return (EPROTO);
1774 	}
1775 
1776 	return (vdc_handle_attr_msg(vdcp, (vd_attr_msg_t *)&vio_msg));
1777 }
1778 
1779 
1780 /*
1781  * Function:
1782  *	vdc_init_dring_negotiate()
1783  *
1784  * Description:
1785  *
1786  * Arguments:
1787  *	vdc	- soft state pointer for this instance of the device driver.
1788  *
1789  * Return Code:
1790  *	0	- Success
1791  */
1792 static int
1793 vdc_init_dring_negotiate(vdc_t *vdc)
1794 {
1795 	vio_dring_reg_msg_t	pkt;
1796 	size_t			msglen = sizeof (pkt);
1797 	int			status = -1;
1798 	int			retry;
1799 	int			nretries = 10;
1800 
1801 	ASSERT(vdc != NULL);
1802 	ASSERT(mutex_owned(&vdc->lock));
1803 
1804 	for (retry = 0; retry < nretries; retry++) {
1805 		status = vdc_init_descriptor_ring(vdc);
1806 		if (status != EAGAIN)
1807 			break;
1808 		drv_usecwait(vdc_min_timeout_ldc);
1809 	}
1810 
1811 	if (status != 0) {
1812 		DMSG(vdc, 0, "[%d] Failed to init DRing (status = %d)\n",
1813 		    vdc->instance, status);
1814 		return (status);
1815 	}
1816 
1817 	DMSG(vdc, 0, "[%d] Init of descriptor ring completed (status = %d)\n",
1818 	    vdc->instance, status);
1819 
1820 	/* fill in tag */
1821 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1822 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1823 	pkt.tag.vio_subtype_env = VIO_DRING_REG;
1824 	pkt.tag.vio_sid = vdc->session_id;
1825 	/* fill in payload */
1826 	pkt.dring_ident = 0;
1827 	pkt.num_descriptors = vdc->dring_len;
1828 	pkt.descriptor_size = vdc->dring_entry_size;
1829 	pkt.options = (VIO_TX_DRING | VIO_RX_DRING);
1830 	pkt.ncookies = vdc->dring_cookie_count;
1831 	pkt.cookie[0] = vdc->dring_cookie[0];	/* for now just one cookie */
1832 
1833 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1834 	if (status != 0) {
1835 		DMSG(vdc, 0, "[%d] Failed to register DRing (err = %d)",
1836 		    vdc->instance, status);
1837 	}
1838 
1839 	return (status);
1840 }
1841 
1842 
1843 /*
1844  * Function:
1845  *	vdc_dring_negotiation()
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_dring_negotiation(vdc_t *vdcp)
1857 {
1858 	int status;
1859 	vio_msg_t vio_msg;
1860 
1861 	if (status = vdc_init_dring_negotiate(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,
1870 		    "[%d] Failed waiting for Dring negotiation response,"
1871 		    " rv(%d)", vdcp->instance, status);
1872 		return (status);
1873 	}
1874 
1875 	/* check type and sub_type ... */
1876 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1877 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1878 		DMSG(vdcp, 0, "[%d] Invalid Dring negotiation response\n",
1879 		    vdcp->instance);
1880 		return (EPROTO);
1881 	}
1882 
1883 	return (vdc_handle_dring_reg_msg(vdcp,
1884 	    (vio_dring_reg_msg_t *)&vio_msg));
1885 }
1886 
1887 
1888 /*
1889  * Function:
1890  *	vdc_send_rdx()
1891  *
1892  * Description:
1893  *
1894  * Arguments:
1895  *	vdc	- soft state pointer for this instance of the device driver.
1896  *
1897  * Return Code:
1898  *	0	- Success
1899  */
1900 static int
1901 vdc_send_rdx(vdc_t *vdcp)
1902 {
1903 	vio_msg_t	msg;
1904 	size_t		msglen = sizeof (vio_msg_t);
1905 	int		status;
1906 
1907 	/*
1908 	 * Send an RDX message to vds to indicate we are ready
1909 	 * to send data
1910 	 */
1911 	msg.tag.vio_msgtype = VIO_TYPE_CTRL;
1912 	msg.tag.vio_subtype = VIO_SUBTYPE_INFO;
1913 	msg.tag.vio_subtype_env = VIO_RDX;
1914 	msg.tag.vio_sid = vdcp->session_id;
1915 	status = vdc_send(vdcp, (caddr_t)&msg, &msglen);
1916 	if (status != 0) {
1917 		DMSG(vdcp, 0, "[%d] Failed to send RDX message (%d)",
1918 		    vdcp->instance, status);
1919 	}
1920 
1921 	return (status);
1922 }
1923 
1924 /*
1925  * Function:
1926  *	vdc_handle_rdx()
1927  *
1928  * Description:
1929  *
1930  * Arguments:
1931  *	vdc	- soft state pointer for this instance of the device driver.
1932  *	msgp	- received msg
1933  *
1934  * Return Code:
1935  *	0	- Success
1936  */
1937 static int
1938 vdc_handle_rdx(vdc_t *vdcp, vio_rdx_msg_t *msgp)
1939 {
1940 	_NOTE(ARGUNUSED(vdcp))
1941 	_NOTE(ARGUNUSED(msgp))
1942 
1943 	ASSERT(msgp->tag.vio_msgtype == VIO_TYPE_CTRL);
1944 	ASSERT(msgp->tag.vio_subtype == VIO_SUBTYPE_ACK);
1945 	ASSERT(msgp->tag.vio_subtype_env == VIO_RDX);
1946 
1947 	DMSG(vdcp, 1, "[%d] Got an RDX msg", vdcp->instance);
1948 
1949 	return (0);
1950 }
1951 
1952 /*
1953  * Function:
1954  *	vdc_rdx_exchange()
1955  *
1956  * Description:
1957  *
1958  * Arguments:
1959  *	vdc	- soft state pointer for this instance of the device driver.
1960  *
1961  * Return Code:
1962  *	0	- Success
1963  */
1964 static int
1965 vdc_rdx_exchange(vdc_t *vdcp)
1966 {
1967 	int status;
1968 	vio_msg_t vio_msg;
1969 
1970 	if (status = vdc_send_rdx(vdcp))
1971 		return (status);
1972 
1973 	/* release lock and wait for response */
1974 	mutex_exit(&vdcp->lock);
1975 	status = vdc_wait_for_response(vdcp, &vio_msg);
1976 	mutex_enter(&vdcp->lock);
1977 	if (status) {
1978 		DMSG(vdcp, 0, "[%d] Failed waiting for RDX response, rv(%d)",
1979 		    vdcp->instance, status);
1980 		return (status);
1981 	}
1982 
1983 	/* check type and sub_type ... */
1984 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1985 	    vio_msg.tag.vio_subtype != VIO_SUBTYPE_ACK) {
1986 		DMSG(vdcp, 0, "[%d] Invalid RDX response\n", vdcp->instance);
1987 		return (EPROTO);
1988 	}
1989 
1990 	return (vdc_handle_rdx(vdcp, (vio_rdx_msg_t *)&vio_msg));
1991 }
1992 
1993 
1994 /* -------------------------------------------------------------------------- */
1995 
1996 /*
1997  * LDC helper routines
1998  */
1999 
2000 static int
2001 vdc_recv(vdc_t *vdc, vio_msg_t *msgp, size_t *nbytesp)
2002 {
2003 	int		status;
2004 	boolean_t	q_has_pkts = B_FALSE;
2005 	uint64_t	delay_time;
2006 	size_t		len;
2007 
2008 	mutex_enter(&vdc->read_lock);
2009 
2010 	if (vdc->read_state == VDC_READ_IDLE)
2011 		vdc->read_state = VDC_READ_WAITING;
2012 
2013 	while (vdc->read_state != VDC_READ_PENDING) {
2014 
2015 		/* detect if the connection has been reset */
2016 		if (vdc->read_state == VDC_READ_RESET) {
2017 			status = ECONNRESET;
2018 			goto done;
2019 		}
2020 
2021 		cv_wait(&vdc->read_cv, &vdc->read_lock);
2022 	}
2023 
2024 	/*
2025 	 * Until we get a blocking ldc read we have to retry
2026 	 * until the entire LDC message has arrived before
2027 	 * ldc_read() will succeed. Note we also bail out if
2028 	 * the channel is reset or goes away.
2029 	 */
2030 	delay_time = vdc_ldc_read_init_delay;
2031 loop:
2032 	len = *nbytesp;
2033 	status = ldc_read(vdc->ldc_handle, (caddr_t)msgp, &len);
2034 	switch (status) {
2035 	case EAGAIN:
2036 		delay_time *= 2;
2037 		if (delay_time >= vdc_ldc_read_max_delay)
2038 			delay_time = vdc_ldc_read_max_delay;
2039 		delay(delay_time);
2040 		goto loop;
2041 
2042 	case 0:
2043 		if (len == 0) {
2044 			DMSG(vdc, 1, "[%d] ldc_read returned 0 bytes with "
2045 			    "no error!\n", vdc->instance);
2046 			goto loop;
2047 		}
2048 
2049 		*nbytesp = len;
2050 
2051 		/*
2052 		 * If there are pending messages, leave the
2053 		 * read state as pending. Otherwise, set the state
2054 		 * back to idle.
2055 		 */
2056 		status = ldc_chkq(vdc->ldc_handle, &q_has_pkts);
2057 		if (status == 0 && !q_has_pkts)
2058 			vdc->read_state = VDC_READ_IDLE;
2059 
2060 		break;
2061 	default:
2062 		DMSG(vdc, 0, "ldc_read returned %d\n", status);
2063 		break;
2064 	}
2065 
2066 done:
2067 	mutex_exit(&vdc->read_lock);
2068 
2069 	return (status);
2070 }
2071 
2072 
2073 
2074 #ifdef DEBUG
2075 void
2076 vdc_decode_tag(vdc_t *vdcp, vio_msg_t *msg)
2077 {
2078 	char *ms, *ss, *ses;
2079 	switch (msg->tag.vio_msgtype) {
2080 #define	Q(_s)	case _s : ms = #_s; break;
2081 	Q(VIO_TYPE_CTRL)
2082 	Q(VIO_TYPE_DATA)
2083 	Q(VIO_TYPE_ERR)
2084 #undef Q
2085 	default: ms = "unknown"; break;
2086 	}
2087 
2088 	switch (msg->tag.vio_subtype) {
2089 #define	Q(_s)	case _s : ss = #_s; break;
2090 	Q(VIO_SUBTYPE_INFO)
2091 	Q(VIO_SUBTYPE_ACK)
2092 	Q(VIO_SUBTYPE_NACK)
2093 #undef Q
2094 	default: ss = "unknown"; break;
2095 	}
2096 
2097 	switch (msg->tag.vio_subtype_env) {
2098 #define	Q(_s)	case _s : ses = #_s; break;
2099 	Q(VIO_VER_INFO)
2100 	Q(VIO_ATTR_INFO)
2101 	Q(VIO_DRING_REG)
2102 	Q(VIO_DRING_UNREG)
2103 	Q(VIO_RDX)
2104 	Q(VIO_PKT_DATA)
2105 	Q(VIO_DESC_DATA)
2106 	Q(VIO_DRING_DATA)
2107 #undef Q
2108 	default: ses = "unknown"; break;
2109 	}
2110 
2111 	DMSG(vdcp, 3, "(%x/%x/%x) message : (%s/%s/%s)\n",
2112 	    msg->tag.vio_msgtype, msg->tag.vio_subtype,
2113 	    msg->tag.vio_subtype_env, ms, ss, ses);
2114 }
2115 #endif
2116 
2117 /*
2118  * Function:
2119  *	vdc_send()
2120  *
2121  * Description:
2122  *	The function encapsulates the call to write a message using LDC.
2123  *	If LDC indicates that the call failed due to the queue being full,
2124  *	we retry the ldc_write(), otherwise we return the error returned by LDC.
2125  *
2126  * Arguments:
2127  *	ldc_handle	- LDC handle for the channel this instance of vdc uses
2128  *	pkt		- address of LDC message to be sent
2129  *	msglen		- the size of the message being sent. When the function
2130  *			  returns, this contains the number of bytes written.
2131  *
2132  * Return Code:
2133  *	0		- Success.
2134  *	EINVAL		- pkt or msglen were NULL
2135  *	ECONNRESET	- The connection was not up.
2136  *	EWOULDBLOCK	- LDC queue is full
2137  *	xxx		- other error codes returned by ldc_write
2138  */
2139 static int
2140 vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen)
2141 {
2142 	size_t	size = 0;
2143 	int	status = 0;
2144 	clock_t delay_ticks;
2145 
2146 	ASSERT(vdc != NULL);
2147 	ASSERT(mutex_owned(&vdc->lock));
2148 	ASSERT(msglen != NULL);
2149 	ASSERT(*msglen != 0);
2150 
2151 #ifdef DEBUG
2152 	vdc_decode_tag(vdc, (vio_msg_t *)(uintptr_t)pkt);
2153 #endif
2154 	/*
2155 	 * Wait indefinitely to send if channel
2156 	 * is busy, but bail out if we succeed or
2157 	 * if the channel closes or is reset.
2158 	 */
2159 	delay_ticks = vdc_hz_min_ldc_delay;
2160 	do {
2161 		size = *msglen;
2162 		status = ldc_write(vdc->ldc_handle, pkt, &size);
2163 		if (status == EWOULDBLOCK) {
2164 			delay(delay_ticks);
2165 			/* geometric backoff */
2166 			delay_ticks *= 2;
2167 			if (delay_ticks > vdc_hz_max_ldc_delay)
2168 				delay_ticks = vdc_hz_max_ldc_delay;
2169 		}
2170 	} while (status == EWOULDBLOCK);
2171 
2172 	/* if LDC had serious issues --- reset vdc state */
2173 	if (status == EIO || status == ECONNRESET) {
2174 		/* LDC had serious issues --- reset vdc state */
2175 		mutex_enter(&vdc->read_lock);
2176 		if ((vdc->read_state == VDC_READ_WAITING) ||
2177 		    (vdc->read_state == VDC_READ_RESET))
2178 			cv_signal(&vdc->read_cv);
2179 		vdc->read_state = VDC_READ_RESET;
2180 		mutex_exit(&vdc->read_lock);
2181 
2182 		/* wake up any waiters in the reset thread */
2183 		if (vdc->state == VDC_STATE_INIT_WAITING) {
2184 			DMSG(vdc, 0, "[%d] write reset - "
2185 			    "vdc is resetting ..\n", vdc->instance);
2186 			vdc->state = VDC_STATE_RESETTING;
2187 			cv_signal(&vdc->initwait_cv);
2188 		}
2189 
2190 		return (ECONNRESET);
2191 	}
2192 
2193 	/* return the last size written */
2194 	*msglen = size;
2195 
2196 	return (status);
2197 }
2198 
2199 /*
2200  * Function:
2201  *	vdc_get_md_node
2202  *
2203  * Description:
2204  *	Get the MD, the device node and the port node for the given
2205  *	disk instance. The caller is responsible for cleaning up the
2206  *	reference to the returned MD (mdpp) by calling md_fini_handle().
2207  *
2208  * Arguments:
2209  *	dip	- dev info pointer for this instance of the device driver.
2210  *	mdpp	- the returned MD.
2211  *	vd_nodep - the returned device node.
2212  *	vd_portp - the returned port node. The returned port node is NULL
2213  *		   if no port node is found.
2214  *
2215  * Return Code:
2216  *	0	- Success.
2217  *	ENOENT	- Expected node or property did not exist.
2218  *	ENXIO	- Unexpected error communicating with MD framework
2219  */
2220 static int
2221 vdc_get_md_node(dev_info_t *dip, md_t **mdpp, mde_cookie_t *vd_nodep,
2222     mde_cookie_t *vd_portp)
2223 {
2224 	int		status = ENOENT;
2225 	char		*node_name = NULL;
2226 	md_t		*mdp = NULL;
2227 	int		num_nodes;
2228 	int		num_vdevs;
2229 	int		num_vports;
2230 	mde_cookie_t	rootnode;
2231 	mde_cookie_t	*listp = NULL;
2232 	boolean_t	found_inst = B_FALSE;
2233 	int		listsz;
2234 	int		idx;
2235 	uint64_t	md_inst;
2236 	int		obp_inst;
2237 	int		instance = ddi_get_instance(dip);
2238 
2239 	/*
2240 	 * Get the OBP instance number for comparison with the MD instance
2241 	 *
2242 	 * The "cfg-handle" property of a vdc node in an MD contains the MD's
2243 	 * notion of "instance", or unique identifier, for that node; OBP
2244 	 * stores the value of the "cfg-handle" MD property as the value of
2245 	 * the "reg" property on the node in the device tree it builds from
2246 	 * the MD and passes to Solaris.  Thus, we look up the devinfo node's
2247 	 * "reg" property value to uniquely identify this device instance.
2248 	 * If the "reg" property cannot be found, the device tree state is
2249 	 * presumably so broken that there is no point in continuing.
2250 	 */
2251 	if (!ddi_prop_exists(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS, OBP_REG)) {
2252 		cmn_err(CE_WARN, "'%s' property does not exist", OBP_REG);
2253 		return (ENOENT);
2254 	}
2255 	obp_inst = ddi_prop_get_int(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
2256 	    OBP_REG, -1);
2257 	DMSGX(1, "[%d] OBP inst=%d\n", instance, obp_inst);
2258 
2259 	/*
2260 	 * We now walk the MD nodes to find the node for this vdisk.
2261 	 */
2262 	if ((mdp = md_get_handle()) == NULL) {
2263 		cmn_err(CE_WARN, "unable to init machine description");
2264 		return (ENXIO);
2265 	}
2266 
2267 	num_nodes = md_node_count(mdp);
2268 	ASSERT(num_nodes > 0);
2269 
2270 	listsz = num_nodes * sizeof (mde_cookie_t);
2271 
2272 	/* allocate memory for nodes */
2273 	listp = kmem_zalloc(listsz, KM_SLEEP);
2274 
2275 	rootnode = md_root_node(mdp);
2276 	ASSERT(rootnode != MDE_INVAL_ELEM_COOKIE);
2277 
2278 	/*
2279 	 * Search for all the virtual devices, we will then check to see which
2280 	 * ones are disk nodes.
2281 	 */
2282 	num_vdevs = md_scan_dag(mdp, rootnode,
2283 	    md_find_name(mdp, VDC_MD_VDEV_NAME),
2284 	    md_find_name(mdp, "fwd"), listp);
2285 
2286 	if (num_vdevs <= 0) {
2287 		cmn_err(CE_NOTE, "No '%s' node found", VDC_MD_VDEV_NAME);
2288 		status = ENOENT;
2289 		goto done;
2290 	}
2291 
2292 	DMSGX(1, "[%d] num_vdevs=%d\n", instance, num_vdevs);
2293 	for (idx = 0; idx < num_vdevs; idx++) {
2294 		status = md_get_prop_str(mdp, listp[idx], "name", &node_name);
2295 		if ((status != 0) || (node_name == NULL)) {
2296 			cmn_err(CE_NOTE, "Unable to get name of node type '%s'"
2297 			    ": err %d", VDC_MD_VDEV_NAME, status);
2298 			continue;
2299 		}
2300 
2301 		DMSGX(1, "[%d] Found node '%s'\n", instance, node_name);
2302 		if (strcmp(VDC_MD_DISK_NAME, node_name) == 0) {
2303 			status = md_get_prop_val(mdp, listp[idx],
2304 			    VDC_MD_CFG_HDL, &md_inst);
2305 			DMSGX(1, "[%d] vdc inst in MD=%lx\n",
2306 			    instance, md_inst);
2307 			if ((status == 0) && (md_inst == obp_inst)) {
2308 				found_inst = B_TRUE;
2309 				break;
2310 			}
2311 		}
2312 	}
2313 
2314 	if (!found_inst) {
2315 		DMSGX(0, "Unable to find correct '%s' node", VDC_MD_DISK_NAME);
2316 		status = ENOENT;
2317 		goto done;
2318 	}
2319 	DMSGX(0, "[%d] MD inst=%lx\n", instance, md_inst);
2320 
2321 	*vd_nodep = listp[idx];
2322 	*mdpp = mdp;
2323 
2324 	num_vports = md_scan_dag(mdp, *vd_nodep,
2325 	    md_find_name(mdp, VDC_MD_PORT_NAME),
2326 	    md_find_name(mdp, "fwd"), listp);
2327 
2328 	if (num_vports != 1) {
2329 		DMSGX(0, "Expected 1 '%s' node for '%s' port, found %d\n",
2330 		    VDC_MD_PORT_NAME, VDC_MD_VDEV_NAME, num_vports);
2331 	}
2332 
2333 	*vd_portp = (num_vports == 0)? NULL: listp[0];
2334 
2335 done:
2336 	kmem_free(listp, listsz);
2337 	return (status);
2338 }
2339 
2340 /*
2341  * Function:
2342  *	vdc_get_ldc_id()
2343  *
2344  * Description:
2345  *	This function gets the 'ldc-id' for this particular instance of vdc.
2346  *	The id returned is the guest domain channel endpoint LDC uses for
2347  *	communication with vds.
2348  *
2349  * Arguments:
2350  *	mdp	- pointer to the machine description.
2351  *	vd_node	- the vdisk element from the MD.
2352  *	ldc_id	- pointer to variable used to return the 'ldc-id' found.
2353  *
2354  * Return Code:
2355  *	0	- Success.
2356  *	ENOENT	- Expected node or property did not exist.
2357  */
2358 static int
2359 vdc_get_ldc_id(md_t *mdp, mde_cookie_t vd_node, uint64_t *ldc_id)
2360 {
2361 	mde_cookie_t	*chanp = NULL;
2362 	int		listsz;
2363 	int		num_chans;
2364 	int		num_nodes;
2365 	int		status = 0;
2366 
2367 	num_nodes = md_node_count(mdp);
2368 	ASSERT(num_nodes > 0);
2369 
2370 	listsz = num_nodes * sizeof (mde_cookie_t);
2371 
2372 	/* allocate memory for nodes */
2373 	chanp = kmem_zalloc(listsz, KM_SLEEP);
2374 
2375 	/* get the channels for this node */
2376 	num_chans = md_scan_dag(mdp, vd_node,
2377 	    md_find_name(mdp, VDC_MD_CHAN_NAME),
2378 	    md_find_name(mdp, "fwd"), chanp);
2379 
2380 	/* expecting at least one channel */
2381 	if (num_chans <= 0) {
2382 		cmn_err(CE_NOTE, "No '%s' node for '%s' port",
2383 		    VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME);
2384 		status = ENOENT;
2385 		goto done;
2386 
2387 	} else if (num_chans != 1) {
2388 		DMSGX(0, "Expected 1 '%s' node for '%s' port, found %d\n",
2389 		    VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME, num_chans);
2390 	}
2391 
2392 	/*
2393 	 * We use the first channel found (index 0), irrespective of how
2394 	 * many are there in total.
2395 	 */
2396 	if (md_get_prop_val(mdp, chanp[0], VDC_MD_ID, ldc_id) != 0) {
2397 		cmn_err(CE_NOTE, "Channel '%s' property not found", VDC_MD_ID);
2398 		status = ENOENT;
2399 	}
2400 
2401 done:
2402 	kmem_free(chanp, listsz);
2403 	return (status);
2404 }
2405 
2406 static int
2407 vdc_do_ldc_up(vdc_t *vdc)
2408 {
2409 	int		status;
2410 	ldc_status_t	ldc_state;
2411 
2412 	DMSG(vdc, 0, "[%d] Bringing up channel %lx\n",
2413 	    vdc->instance, vdc->ldc_id);
2414 
2415 	if (vdc->lifecycle == VDC_LC_DETACHING)
2416 		return (EINVAL);
2417 
2418 	if ((status = ldc_up(vdc->ldc_handle)) != 0) {
2419 		switch (status) {
2420 		case ECONNREFUSED:	/* listener not ready at other end */
2421 			DMSG(vdc, 0, "[%d] ldc_up(%lx,...) return %d\n",
2422 			    vdc->instance, vdc->ldc_id, status);
2423 			status = 0;
2424 			break;
2425 		default:
2426 			DMSG(vdc, 0, "[%d] Failed to bring up LDC: "
2427 			    "channel=%ld, err=%d", vdc->instance, vdc->ldc_id,
2428 			    status);
2429 			break;
2430 		}
2431 	}
2432 
2433 	if (ldc_status(vdc->ldc_handle, &ldc_state) == 0) {
2434 		vdc->ldc_state = ldc_state;
2435 		if (ldc_state == LDC_UP) {
2436 			DMSG(vdc, 0, "[%d] LDC channel already up\n",
2437 			    vdc->instance);
2438 			vdc->seq_num = 1;
2439 			vdc->seq_num_reply = 0;
2440 		}
2441 	}
2442 
2443 	return (status);
2444 }
2445 
2446 /*
2447  * Function:
2448  *	vdc_terminate_ldc()
2449  *
2450  * Description:
2451  *
2452  * Arguments:
2453  *	vdc	- soft state pointer for this instance of the device driver.
2454  *
2455  * Return Code:
2456  *	None
2457  */
2458 static void
2459 vdc_terminate_ldc(vdc_t *vdc)
2460 {
2461 	int	instance = ddi_get_instance(vdc->dip);
2462 
2463 	ASSERT(vdc != NULL);
2464 	ASSERT(mutex_owned(&vdc->lock));
2465 
2466 	DMSG(vdc, 0, "[%d] initialized=%x\n", instance, vdc->initialized);
2467 
2468 	if (vdc->initialized & VDC_LDC_OPEN) {
2469 		DMSG(vdc, 0, "[%d] ldc_close()\n", instance);
2470 		(void) ldc_close(vdc->ldc_handle);
2471 	}
2472 	if (vdc->initialized & VDC_LDC_CB) {
2473 		DMSG(vdc, 0, "[%d] ldc_unreg_callback()\n", instance);
2474 		(void) ldc_unreg_callback(vdc->ldc_handle);
2475 	}
2476 	if (vdc->initialized & VDC_LDC) {
2477 		DMSG(vdc, 0, "[%d] ldc_fini()\n", instance);
2478 		(void) ldc_fini(vdc->ldc_handle);
2479 		vdc->ldc_handle = NULL;
2480 	}
2481 
2482 	vdc->initialized &= ~(VDC_LDC | VDC_LDC_CB | VDC_LDC_OPEN);
2483 }
2484 
2485 /* -------------------------------------------------------------------------- */
2486 
2487 /*
2488  * Descriptor Ring helper routines
2489  */
2490 
2491 /*
2492  * Function:
2493  *	vdc_init_descriptor_ring()
2494  *
2495  * Description:
2496  *
2497  * Arguments:
2498  *	vdc	- soft state pointer for this instance of the device driver.
2499  *
2500  * Return Code:
2501  *	0	- Success
2502  */
2503 static int
2504 vdc_init_descriptor_ring(vdc_t *vdc)
2505 {
2506 	vd_dring_entry_t	*dep = NULL;	/* DRing Entry pointer */
2507 	int	status = 0;
2508 	int	i;
2509 
2510 	DMSG(vdc, 0, "[%d] initialized=%x\n", vdc->instance, vdc->initialized);
2511 
2512 	ASSERT(vdc != NULL);
2513 	ASSERT(mutex_owned(&vdc->lock));
2514 	ASSERT(vdc->ldc_handle != NULL);
2515 
2516 	/* ensure we have enough room to store max sized block */
2517 	ASSERT(maxphys <= VD_MAX_BLOCK_SIZE);
2518 
2519 	if ((vdc->initialized & VDC_DRING_INIT) == 0) {
2520 		DMSG(vdc, 0, "[%d] ldc_mem_dring_create\n", vdc->instance);
2521 		/*
2522 		 * Calculate the maximum block size we can transmit using one
2523 		 * Descriptor Ring entry from the attributes returned by the
2524 		 * vDisk server. This is subject to a minimum of 'maxphys'
2525 		 * as we do not have the capability to split requests over
2526 		 * multiple DRing entries.
2527 		 */
2528 		if ((vdc->max_xfer_sz * vdc->block_size) < maxphys) {
2529 			DMSG(vdc, 0, "[%d] using minimum DRing size\n",
2530 			    vdc->instance);
2531 			vdc->dring_max_cookies = maxphys / PAGESIZE;
2532 		} else {
2533 			vdc->dring_max_cookies =
2534 			    (vdc->max_xfer_sz * vdc->block_size) / PAGESIZE;
2535 		}
2536 		vdc->dring_entry_size = (sizeof (vd_dring_entry_t) +
2537 		    (sizeof (ldc_mem_cookie_t) *
2538 		    (vdc->dring_max_cookies - 1)));
2539 		vdc->dring_len = VD_DRING_LEN;
2540 
2541 		status = ldc_mem_dring_create(vdc->dring_len,
2542 		    vdc->dring_entry_size, &vdc->ldc_dring_hdl);
2543 		if ((vdc->ldc_dring_hdl == NULL) || (status != 0)) {
2544 			DMSG(vdc, 0, "[%d] Descriptor ring creation failed",
2545 			    vdc->instance);
2546 			return (status);
2547 		}
2548 		vdc->initialized |= VDC_DRING_INIT;
2549 	}
2550 
2551 	if ((vdc->initialized & VDC_DRING_BOUND) == 0) {
2552 		DMSG(vdc, 0, "[%d] ldc_mem_dring_bind\n", vdc->instance);
2553 		vdc->dring_cookie =
2554 		    kmem_zalloc(sizeof (ldc_mem_cookie_t), KM_SLEEP);
2555 
2556 		status = ldc_mem_dring_bind(vdc->ldc_handle, vdc->ldc_dring_hdl,
2557 		    LDC_SHADOW_MAP|LDC_DIRECT_MAP, LDC_MEM_RW,
2558 		    &vdc->dring_cookie[0],
2559 		    &vdc->dring_cookie_count);
2560 		if (status != 0) {
2561 			DMSG(vdc, 0, "[%d] Failed to bind descriptor ring "
2562 			    "(%lx) to channel (%lx) status=%d\n",
2563 			    vdc->instance, vdc->ldc_dring_hdl,
2564 			    vdc->ldc_handle, status);
2565 			return (status);
2566 		}
2567 		ASSERT(vdc->dring_cookie_count == 1);
2568 		vdc->initialized |= VDC_DRING_BOUND;
2569 	}
2570 
2571 	status = ldc_mem_dring_info(vdc->ldc_dring_hdl, &vdc->dring_mem_info);
2572 	if (status != 0) {
2573 		DMSG(vdc, 0,
2574 		    "[%d] Failed to get info for descriptor ring (%lx)\n",
2575 		    vdc->instance, vdc->ldc_dring_hdl);
2576 		return (status);
2577 	}
2578 
2579 	if ((vdc->initialized & VDC_DRING_LOCAL) == 0) {
2580 		DMSG(vdc, 0, "[%d] local dring\n", vdc->instance);
2581 
2582 		/* Allocate the local copy of this dring */
2583 		vdc->local_dring =
2584 		    kmem_zalloc(vdc->dring_len * sizeof (vdc_local_desc_t),
2585 		    KM_SLEEP);
2586 		vdc->initialized |= VDC_DRING_LOCAL;
2587 	}
2588 
2589 	/*
2590 	 * Mark all DRing entries as free and initialize the private
2591 	 * descriptor's memory handles. If any entry is initialized,
2592 	 * we need to free it later so we set the bit in 'initialized'
2593 	 * at the start.
2594 	 */
2595 	vdc->initialized |= VDC_DRING_ENTRY;
2596 	for (i = 0; i < vdc->dring_len; i++) {
2597 		dep = VDC_GET_DRING_ENTRY_PTR(vdc, i);
2598 		dep->hdr.dstate = VIO_DESC_FREE;
2599 
2600 		status = ldc_mem_alloc_handle(vdc->ldc_handle,
2601 		    &vdc->local_dring[i].desc_mhdl);
2602 		if (status != 0) {
2603 			DMSG(vdc, 0, "![%d] Failed to alloc mem handle for"
2604 			    " descriptor %d", vdc->instance, i);
2605 			return (status);
2606 		}
2607 		vdc->local_dring[i].is_free = B_TRUE;
2608 		vdc->local_dring[i].dep = dep;
2609 	}
2610 
2611 	/* Initialize the starting index */
2612 	vdc->dring_curr_idx = 0;
2613 
2614 	return (status);
2615 }
2616 
2617 /*
2618  * Function:
2619  *	vdc_destroy_descriptor_ring()
2620  *
2621  * Description:
2622  *
2623  * Arguments:
2624  *	vdc	- soft state pointer for this instance of the device driver.
2625  *
2626  * Return Code:
2627  *	None
2628  */
2629 static void
2630 vdc_destroy_descriptor_ring(vdc_t *vdc)
2631 {
2632 	vdc_local_desc_t	*ldep = NULL;	/* Local Dring Entry Pointer */
2633 	ldc_mem_handle_t	mhdl = NULL;
2634 	ldc_mem_info_t		minfo;
2635 	int			status = -1;
2636 	int			i;	/* loop */
2637 
2638 	ASSERT(vdc != NULL);
2639 	ASSERT(mutex_owned(&vdc->lock));
2640 
2641 	DMSG(vdc, 0, "[%d] Entered\n", vdc->instance);
2642 
2643 	if (vdc->initialized & VDC_DRING_ENTRY) {
2644 		DMSG(vdc, 0,
2645 		    "[%d] Removing Local DRing entries\n", vdc->instance);
2646 		for (i = 0; i < vdc->dring_len; i++) {
2647 			ldep = &vdc->local_dring[i];
2648 			mhdl = ldep->desc_mhdl;
2649 
2650 			if (mhdl == NULL)
2651 				continue;
2652 
2653 			if ((status = ldc_mem_info(mhdl, &minfo)) != 0) {
2654 				DMSG(vdc, 0,
2655 				    "ldc_mem_info returned an error: %d\n",
2656 				    status);
2657 
2658 				/*
2659 				 * This must mean that the mem handle
2660 				 * is not valid. Clear it out so that
2661 				 * no one tries to use it.
2662 				 */
2663 				ldep->desc_mhdl = NULL;
2664 				continue;
2665 			}
2666 
2667 			if (minfo.status == LDC_BOUND) {
2668 				(void) ldc_mem_unbind_handle(mhdl);
2669 			}
2670 
2671 			(void) ldc_mem_free_handle(mhdl);
2672 
2673 			ldep->desc_mhdl = NULL;
2674 		}
2675 		vdc->initialized &= ~VDC_DRING_ENTRY;
2676 	}
2677 
2678 	if (vdc->initialized & VDC_DRING_LOCAL) {
2679 		DMSG(vdc, 0, "[%d] Freeing Local DRing\n", vdc->instance);
2680 		kmem_free(vdc->local_dring,
2681 		    vdc->dring_len * sizeof (vdc_local_desc_t));
2682 		vdc->initialized &= ~VDC_DRING_LOCAL;
2683 	}
2684 
2685 	if (vdc->initialized & VDC_DRING_BOUND) {
2686 		DMSG(vdc, 0, "[%d] Unbinding DRing\n", vdc->instance);
2687 		status = ldc_mem_dring_unbind(vdc->ldc_dring_hdl);
2688 		if (status == 0) {
2689 			vdc->initialized &= ~VDC_DRING_BOUND;
2690 		} else {
2691 			DMSG(vdc, 0, "[%d] Error %d unbinding DRing %lx",
2692 			    vdc->instance, status, vdc->ldc_dring_hdl);
2693 		}
2694 		kmem_free(vdc->dring_cookie, sizeof (ldc_mem_cookie_t));
2695 	}
2696 
2697 	if (vdc->initialized & VDC_DRING_INIT) {
2698 		DMSG(vdc, 0, "[%d] Destroying DRing\n", vdc->instance);
2699 		status = ldc_mem_dring_destroy(vdc->ldc_dring_hdl);
2700 		if (status == 0) {
2701 			vdc->ldc_dring_hdl = NULL;
2702 			bzero(&vdc->dring_mem_info, sizeof (ldc_mem_info_t));
2703 			vdc->initialized &= ~VDC_DRING_INIT;
2704 		} else {
2705 			DMSG(vdc, 0, "[%d] Error %d destroying DRing (%lx)",
2706 			    vdc->instance, status, vdc->ldc_dring_hdl);
2707 		}
2708 	}
2709 }
2710 
2711 /*
2712  * Function:
2713  *	vdc_map_to_shared_ring()
2714  *
2715  * Description:
2716  *	Copy contents of the local descriptor to the shared
2717  *	memory descriptor.
2718  *
2719  * Arguments:
2720  *	vdcp	- soft state pointer for this instance of the device driver.
2721  *	idx	- descriptor ring index
2722  *
2723  * Return Code:
2724  *	None
2725  */
2726 static int
2727 vdc_map_to_shared_dring(vdc_t *vdcp, int idx)
2728 {
2729 	vdc_local_desc_t	*ldep;
2730 	vd_dring_entry_t	*dep;
2731 	int			rv;
2732 
2733 	ldep = &(vdcp->local_dring[idx]);
2734 
2735 	/* for now leave in the old pop_mem_hdl stuff */
2736 	if (ldep->nbytes > 0) {
2737 		rv = vdc_populate_mem_hdl(vdcp, ldep);
2738 		if (rv) {
2739 			DMSG(vdcp, 0, "[%d] Cannot populate mem handle\n",
2740 			    vdcp->instance);
2741 			return (rv);
2742 		}
2743 	}
2744 
2745 	/*
2746 	 * fill in the data details into the DRing
2747 	 */
2748 	dep = ldep->dep;
2749 	ASSERT(dep != NULL);
2750 
2751 	dep->payload.req_id = VDC_GET_NEXT_REQ_ID(vdcp);
2752 	dep->payload.operation = ldep->operation;
2753 	dep->payload.addr = ldep->offset;
2754 	dep->payload.nbytes = ldep->nbytes;
2755 	dep->payload.status = (uint32_t)-1;	/* vds will set valid value */
2756 	dep->payload.slice = ldep->slice;
2757 	dep->hdr.dstate = VIO_DESC_READY;
2758 	dep->hdr.ack = 1;		/* request an ACK for every message */
2759 
2760 	return (0);
2761 }
2762 
2763 /*
2764  * Function:
2765  *	vdc_send_request
2766  *
2767  * Description:
2768  *	This routine writes the data to be transmitted to vds into the
2769  *	descriptor, notifies vds that the ring has been updated and
2770  *	then waits for the request to be processed.
2771  *
2772  * Arguments:
2773  *	vdcp	  - the soft state pointer
2774  *	operation - operation we want vds to perform (VD_OP_XXX)
2775  *	addr	  - address of data buf to be read/written.
2776  *	nbytes	  - number of bytes to read/write
2777  *	slice	  - the disk slice this request is for
2778  *	offset	  - relative disk offset
2779  *	cb_type   - type of call - STRATEGY or SYNC
2780  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
2781  *			. mode for ioctl(9e)
2782  *			. LP64 diskaddr_t (block I/O)
2783  *	dir	  - direction of operation (READ/WRITE/BOTH)
2784  *
2785  * Return Codes:
2786  *	0
2787  *	ENXIO
2788  */
2789 static int
2790 vdc_send_request(vdc_t *vdcp, int operation, caddr_t addr,
2791     size_t nbytes, int slice, diskaddr_t offset, int cb_type,
2792     void *cb_arg, vio_desc_direction_t dir)
2793 {
2794 	int	rv = 0;
2795 
2796 	ASSERT(vdcp != NULL);
2797 	ASSERT(slice == VD_SLICE_NONE || slice < V_NUMPAR);
2798 
2799 	mutex_enter(&vdcp->lock);
2800 
2801 	/*
2802 	 * If this is a block read/write operation we update the I/O statistics
2803 	 * to indicate that the request is being put on the waitq to be
2804 	 * serviced.
2805 	 *
2806 	 * We do it here (a common routine for both synchronous and strategy
2807 	 * calls) for performance reasons - we are already holding vdc->lock
2808 	 * so there is no extra locking overhead. We would have to explicitly
2809 	 * grab the 'lock' mutex to update the stats if we were to do this
2810 	 * higher up the stack in vdc_strategy() et. al.
2811 	 */
2812 	if ((operation == VD_OP_BREAD) || (operation == VD_OP_BWRITE)) {
2813 		DTRACE_IO1(start, buf_t *, cb_arg);
2814 		VD_KSTAT_WAITQ_ENTER(vdcp->io_stats);
2815 	}
2816 
2817 	do {
2818 		while (vdcp->state != VDC_STATE_RUNNING) {
2819 
2820 			/* return error if detaching */
2821 			if (vdcp->state == VDC_STATE_DETACH) {
2822 				rv = ENXIO;
2823 				goto done;
2824 			}
2825 
2826 			/* fail request if connection timeout is reached */
2827 			if (vdcp->ctimeout_reached) {
2828 				rv = EIO;
2829 				goto done;
2830 			}
2831 
2832 			/*
2833 			 * If we are panicking and the disk is not ready then
2834 			 * we can't send any request because we can't complete
2835 			 * the handshake now.
2836 			 */
2837 			if (ddi_in_panic()) {
2838 				rv = EIO;
2839 				goto done;
2840 			}
2841 
2842 			cv_wait(&vdcp->running_cv, &vdcp->lock);
2843 		}
2844 
2845 	} while (vdc_populate_descriptor(vdcp, operation, addr,
2846 	    nbytes, slice, offset, cb_type, cb_arg, dir));
2847 
2848 done:
2849 	/*
2850 	 * If this is a block read/write we update the I/O statistics kstat
2851 	 * to indicate that this request has been placed on the queue for
2852 	 * processing (i.e sent to the vDisk server) - iostat(1M) will
2853 	 * report the time waiting for the vDisk server under the %b column
2854 	 * In the case of an error we simply take it off the wait queue.
2855 	 */
2856 	if ((operation == VD_OP_BREAD) || (operation == VD_OP_BWRITE)) {
2857 		if (rv == 0) {
2858 			VD_KSTAT_WAITQ_TO_RUNQ(vdcp->io_stats);
2859 			DTRACE_PROBE1(send, buf_t *, cb_arg);
2860 		} else {
2861 			VD_UPDATE_ERR_STATS(vdcp, vd_transerrs);
2862 			VD_KSTAT_WAITQ_EXIT(vdcp->io_stats);
2863 			DTRACE_IO1(done, buf_t *, cb_arg);
2864 		}
2865 	}
2866 
2867 	mutex_exit(&vdcp->lock);
2868 
2869 	return (rv);
2870 }
2871 
2872 
2873 /*
2874  * Function:
2875  *	vdc_populate_descriptor
2876  *
2877  * Description:
2878  *	This routine writes the data to be transmitted to vds into the
2879  *	descriptor, notifies vds that the ring has been updated and
2880  *	then waits for the request to be processed.
2881  *
2882  * Arguments:
2883  *	vdcp	  - the soft state pointer
2884  *	operation - operation we want vds to perform (VD_OP_XXX)
2885  *	addr	  - address of data buf to be read/written.
2886  *	nbytes	  - number of bytes to read/write
2887  *	slice	  - the disk slice this request is for
2888  *	offset	  - relative disk offset
2889  *	cb_type   - type of call - STRATEGY or SYNC
2890  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
2891  *			. mode for ioctl(9e)
2892  *			. LP64 diskaddr_t (block I/O)
2893  *	dir	  - direction of operation (READ/WRITE/BOTH)
2894  *
2895  * Return Codes:
2896  *	0
2897  *	EAGAIN
2898  *	ECONNRESET
2899  *	ENXIO
2900  */
2901 static int
2902 vdc_populate_descriptor(vdc_t *vdcp, int operation, caddr_t addr,
2903     size_t nbytes, int slice, diskaddr_t offset, int cb_type,
2904     void *cb_arg, vio_desc_direction_t dir)
2905 {
2906 	vdc_local_desc_t	*local_dep = NULL; /* Local Dring Pointer */
2907 	int			idx;		/* Index of DRing entry used */
2908 	int			next_idx;
2909 	vio_dring_msg_t		dmsg;
2910 	size_t			msglen;
2911 	int			rv;
2912 
2913 	ASSERT(MUTEX_HELD(&vdcp->lock));
2914 	vdcp->threads_pending++;
2915 loop:
2916 	DMSG(vdcp, 2, ": dring_curr_idx = %d\n", vdcp->dring_curr_idx);
2917 
2918 	/* Get next available D-Ring entry */
2919 	idx = vdcp->dring_curr_idx;
2920 	local_dep = &(vdcp->local_dring[idx]);
2921 
2922 	if (!local_dep->is_free) {
2923 		DMSG(vdcp, 2, "[%d]: dring full - waiting for space\n",
2924 		    vdcp->instance);
2925 		cv_wait(&vdcp->dring_free_cv, &vdcp->lock);
2926 		if (vdcp->state == VDC_STATE_RUNNING ||
2927 		    vdcp->state == VDC_STATE_HANDLE_PENDING) {
2928 			goto loop;
2929 		}
2930 		vdcp->threads_pending--;
2931 		return (ECONNRESET);
2932 	}
2933 
2934 	next_idx = idx + 1;
2935 	if (next_idx >= vdcp->dring_len)
2936 		next_idx = 0;
2937 	vdcp->dring_curr_idx = next_idx;
2938 
2939 	ASSERT(local_dep->is_free);
2940 
2941 	local_dep->operation = operation;
2942 	local_dep->addr = addr;
2943 	local_dep->nbytes = nbytes;
2944 	local_dep->slice = slice;
2945 	local_dep->offset = offset;
2946 	local_dep->cb_type = cb_type;
2947 	local_dep->cb_arg = cb_arg;
2948 	local_dep->dir = dir;
2949 
2950 	local_dep->is_free = B_FALSE;
2951 
2952 	rv = vdc_map_to_shared_dring(vdcp, idx);
2953 	if (rv) {
2954 		DMSG(vdcp, 0, "[%d]: cannot bind memory - waiting ..\n",
2955 		    vdcp->instance);
2956 		/* free the descriptor */
2957 		local_dep->is_free = B_TRUE;
2958 		vdcp->dring_curr_idx = idx;
2959 		cv_wait(&vdcp->membind_cv, &vdcp->lock);
2960 		if (vdcp->state == VDC_STATE_RUNNING ||
2961 		    vdcp->state == VDC_STATE_HANDLE_PENDING) {
2962 			goto loop;
2963 		}
2964 		vdcp->threads_pending--;
2965 		return (ECONNRESET);
2966 	}
2967 
2968 	/*
2969 	 * Send a msg with the DRing details to vds
2970 	 */
2971 	VIO_INIT_DRING_DATA_TAG(dmsg);
2972 	VDC_INIT_DRING_DATA_MSG_IDS(dmsg, vdcp);
2973 	dmsg.dring_ident = vdcp->dring_ident;
2974 	dmsg.start_idx = idx;
2975 	dmsg.end_idx = idx;
2976 	vdcp->seq_num++;
2977 
2978 	DTRACE_PROBE2(populate, int, vdcp->instance,
2979 	    vdc_local_desc_t *, local_dep);
2980 	DMSG(vdcp, 2, "ident=0x%lx, st=%u, end=%u, seq=%ld\n",
2981 	    vdcp->dring_ident, dmsg.start_idx, dmsg.end_idx, dmsg.seq_num);
2982 
2983 	/*
2984 	 * note we're still holding the lock here to
2985 	 * make sure the message goes out in order !!!...
2986 	 */
2987 	msglen = sizeof (dmsg);
2988 	rv = vdc_send(vdcp, (caddr_t)&dmsg, &msglen);
2989 	switch (rv) {
2990 	case ECONNRESET:
2991 		/*
2992 		 * vdc_send initiates the reset on failure.
2993 		 * Since the transaction has already been put
2994 		 * on the local dring, it will automatically get
2995 		 * retried when the channel is reset. Given that,
2996 		 * it is ok to just return success even though the
2997 		 * send failed.
2998 		 */
2999 		rv = 0;
3000 		break;
3001 
3002 	case 0: /* EOK */
3003 		DMSG(vdcp, 1, "sent via LDC: rv=%d\n", rv);
3004 		break;
3005 
3006 	default:
3007 		goto cleanup_and_exit;
3008 	}
3009 
3010 	vdcp->threads_pending--;
3011 	return (rv);
3012 
3013 cleanup_and_exit:
3014 	DMSG(vdcp, 0, "unexpected error, rv=%d\n", rv);
3015 	return (ENXIO);
3016 }
3017 
3018 /*
3019  * Function:
3020  *	vdc_do_sync_op
3021  *
3022  * Description:
3023  * 	Wrapper around vdc_populate_descriptor that blocks until the
3024  * 	response to the message is available.
3025  *
3026  * Arguments:
3027  *	vdcp	  - the soft state pointer
3028  *	operation - operation we want vds to perform (VD_OP_XXX)
3029  *	addr	  - address of data buf to be read/written.
3030  *	nbytes	  - number of bytes to read/write
3031  *	slice	  - the disk slice this request is for
3032  *	offset	  - relative disk offset
3033  *	cb_type   - type of call - STRATEGY or SYNC
3034  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
3035  *			. mode for ioctl(9e)
3036  *			. LP64 diskaddr_t (block I/O)
3037  *	dir	  - direction of operation (READ/WRITE/BOTH)
3038  *	rconflict - check for reservation conflict in case of failure
3039  *
3040  * rconflict should be set to B_TRUE by most callers. Callers invoking the
3041  * VD_OP_SCSICMD operation can set rconflict to B_FALSE if they check the
3042  * result of a successful operation with vd_scsi_status().
3043  *
3044  * Return Codes:
3045  *	0
3046  *	EAGAIN
3047  *	EFAULT
3048  *	ENXIO
3049  *	EIO
3050  */
3051 static int
3052 vdc_do_sync_op(vdc_t *vdcp, int operation, caddr_t addr, size_t nbytes,
3053     int slice, diskaddr_t offset, int cb_type, void *cb_arg,
3054     vio_desc_direction_t dir, boolean_t rconflict)
3055 {
3056 	int status;
3057 	vdc_io_t *vio;
3058 	boolean_t check_resv_conflict = B_FALSE;
3059 
3060 	ASSERT(cb_type == CB_SYNC);
3061 
3062 	/*
3063 	 * Grab the lock, if blocked wait until the server
3064 	 * response causes us to wake up again.
3065 	 */
3066 	mutex_enter(&vdcp->lock);
3067 	vdcp->sync_op_cnt++;
3068 	while (vdcp->sync_op_blocked && vdcp->state != VDC_STATE_DETACH)
3069 		cv_wait(&vdcp->sync_blocked_cv, &vdcp->lock);
3070 
3071 	if (vdcp->state == VDC_STATE_DETACH) {
3072 		cv_broadcast(&vdcp->sync_blocked_cv);
3073 		vdcp->sync_op_cnt--;
3074 		mutex_exit(&vdcp->lock);
3075 		return (ENXIO);
3076 	}
3077 
3078 	/* now block anyone other thread entering after us */
3079 	vdcp->sync_op_blocked = B_TRUE;
3080 	vdcp->sync_op_pending = B_TRUE;
3081 	mutex_exit(&vdcp->lock);
3082 
3083 	status = vdc_send_request(vdcp, operation, addr,
3084 	    nbytes, slice, offset, cb_type, cb_arg, dir);
3085 
3086 	mutex_enter(&vdcp->lock);
3087 
3088 	if (status != 0) {
3089 		vdcp->sync_op_pending = B_FALSE;
3090 	} else {
3091 		/*
3092 		 * block until our transaction completes.
3093 		 * Also anyone else waiting also gets to go next.
3094 		 */
3095 		while (vdcp->sync_op_pending && vdcp->state != VDC_STATE_DETACH)
3096 			cv_wait(&vdcp->sync_pending_cv, &vdcp->lock);
3097 
3098 		DMSG(vdcp, 2, ": operation returned %d\n",
3099 		    vdcp->sync_op_status);
3100 		if (vdcp->state == VDC_STATE_DETACH) {
3101 			vdcp->sync_op_pending = B_FALSE;
3102 			status = ENXIO;
3103 		} else {
3104 			status = vdcp->sync_op_status;
3105 			if (status != 0 && vdcp->failfast_interval != 0) {
3106 				/*
3107 				 * Operation has failed and failfast is enabled.
3108 				 * We need to check if the failure is due to a
3109 				 * reservation conflict if this was requested.
3110 				 */
3111 				check_resv_conflict = rconflict;
3112 			}
3113 
3114 		}
3115 	}
3116 
3117 	vdcp->sync_op_status = 0;
3118 	vdcp->sync_op_blocked = B_FALSE;
3119 	vdcp->sync_op_cnt--;
3120 
3121 	/* signal the next waiting thread */
3122 	cv_signal(&vdcp->sync_blocked_cv);
3123 
3124 	/*
3125 	 * We have to check for reservation conflict after unblocking sync
3126 	 * operations because some sync operations will be used to do this
3127 	 * check.
3128 	 */
3129 	if (check_resv_conflict) {
3130 		vio = vdc_failfast_io_queue(vdcp, NULL);
3131 		while (vio->vio_qtime != 0)
3132 			cv_wait(&vdcp->failfast_io_cv, &vdcp->lock);
3133 		kmem_free(vio, sizeof (vdc_io_t));
3134 	}
3135 
3136 	mutex_exit(&vdcp->lock);
3137 
3138 	return (status);
3139 }
3140 
3141 
3142 /*
3143  * Function:
3144  *	vdc_drain_response()
3145  *
3146  * Description:
3147  * 	When a guest is panicking, the completion of requests needs to be
3148  * 	handled differently because interrupts are disabled and vdc
3149  * 	will not get messages. We have to poll for the messages instead.
3150  *
3151  *	Note: since we don't have a buf_t available we cannot implement
3152  *	the io:::done DTrace probe in this specific case.
3153  *
3154  * Arguments:
3155  *	vdc	- soft state pointer for this instance of the device driver.
3156  *
3157  * Return Code:
3158  *	0	- Success
3159  */
3160 static int
3161 vdc_drain_response(vdc_t *vdc)
3162 {
3163 	int 			rv, idx, retries;
3164 	size_t			msglen;
3165 	vdc_local_desc_t 	*ldep = NULL;	/* Local Dring Entry Pointer */
3166 	vio_dring_msg_t		dmsg;
3167 
3168 	mutex_enter(&vdc->lock);
3169 
3170 	retries = 0;
3171 	for (;;) {
3172 		msglen = sizeof (dmsg);
3173 		rv = ldc_read(vdc->ldc_handle, (caddr_t)&dmsg, &msglen);
3174 		if (rv) {
3175 			rv = EINVAL;
3176 			break;
3177 		}
3178 
3179 		/*
3180 		 * if there are no packets wait and check again
3181 		 */
3182 		if ((rv == 0) && (msglen == 0)) {
3183 			if (retries++ > vdc_dump_retries) {
3184 				rv = EAGAIN;
3185 				break;
3186 			}
3187 
3188 			drv_usecwait(vdc_usec_timeout_dump);
3189 			continue;
3190 		}
3191 
3192 		/*
3193 		 * Ignore all messages that are not ACKs/NACKs to
3194 		 * DRing requests.
3195 		 */
3196 		if ((dmsg.tag.vio_msgtype != VIO_TYPE_DATA) ||
3197 		    (dmsg.tag.vio_subtype_env != VIO_DRING_DATA)) {
3198 			DMSG(vdc, 0, "discard pkt: type=%d sub=%d env=%d\n",
3199 			    dmsg.tag.vio_msgtype,
3200 			    dmsg.tag.vio_subtype,
3201 			    dmsg.tag.vio_subtype_env);
3202 			continue;
3203 		}
3204 
3205 		/*
3206 		 * set the appropriate return value for the current request.
3207 		 */
3208 		switch (dmsg.tag.vio_subtype) {
3209 		case VIO_SUBTYPE_ACK:
3210 			rv = 0;
3211 			break;
3212 		case VIO_SUBTYPE_NACK:
3213 			rv = EAGAIN;
3214 			break;
3215 		default:
3216 			continue;
3217 		}
3218 
3219 		idx = dmsg.start_idx;
3220 		if (idx >= vdc->dring_len) {
3221 			DMSG(vdc, 0, "[%d] Bogus ack data : start %d\n",
3222 			    vdc->instance, idx);
3223 			continue;
3224 		}
3225 		ldep = &vdc->local_dring[idx];
3226 		if (ldep->dep->hdr.dstate != VIO_DESC_DONE) {
3227 			DMSG(vdc, 0, "[%d] Entry @ %d - state !DONE %d\n",
3228 			    vdc->instance, idx, ldep->dep->hdr.dstate);
3229 			continue;
3230 		}
3231 
3232 		DMSG(vdc, 1, "[%d] Depopulating idx=%d state=%d\n",
3233 		    vdc->instance, idx, ldep->dep->hdr.dstate);
3234 
3235 		rv = vdc_depopulate_descriptor(vdc, idx);
3236 		if (rv) {
3237 			DMSG(vdc, 0,
3238 			    "[%d] Entry @ %d - depopulate failed ..\n",
3239 			    vdc->instance, idx);
3240 		}
3241 
3242 		/* if this is the last descriptor - break out of loop */
3243 		if ((idx + 1) % vdc->dring_len == vdc->dring_curr_idx)
3244 			break;
3245 	}
3246 
3247 	mutex_exit(&vdc->lock);
3248 	DMSG(vdc, 0, "End idx=%d\n", idx);
3249 
3250 	return (rv);
3251 }
3252 
3253 
3254 /*
3255  * Function:
3256  *	vdc_depopulate_descriptor()
3257  *
3258  * Description:
3259  *
3260  * Arguments:
3261  *	vdc	- soft state pointer for this instance of the device driver.
3262  *	idx	- Index of the Descriptor Ring entry being modified
3263  *
3264  * Return Code:
3265  *	0	- Success
3266  */
3267 static int
3268 vdc_depopulate_descriptor(vdc_t *vdc, uint_t idx)
3269 {
3270 	vd_dring_entry_t *dep = NULL;		/* Dring Entry Pointer */
3271 	vdc_local_desc_t *ldep = NULL;		/* Local Dring Entry Pointer */
3272 	int		status = ENXIO;
3273 	int		rv = 0;
3274 
3275 	ASSERT(vdc != NULL);
3276 	ASSERT(idx < vdc->dring_len);
3277 	ldep = &vdc->local_dring[idx];
3278 	ASSERT(ldep != NULL);
3279 	ASSERT(MUTEX_HELD(&vdc->lock));
3280 
3281 	DTRACE_PROBE2(depopulate, int, vdc->instance, vdc_local_desc_t *, ldep);
3282 	DMSG(vdc, 2, ": idx = %d\n", idx);
3283 
3284 	dep = ldep->dep;
3285 	ASSERT(dep != NULL);
3286 	ASSERT((dep->hdr.dstate == VIO_DESC_DONE) ||
3287 	    (dep->payload.status == ECANCELED));
3288 
3289 	VDC_MARK_DRING_ENTRY_FREE(vdc, idx);
3290 
3291 	ldep->is_free = B_TRUE;
3292 	status = dep->payload.status;
3293 	DMSG(vdc, 2, ": is_free = %d : status = %d\n", ldep->is_free, status);
3294 
3295 	/*
3296 	 * If no buffers were used to transfer information to the server when
3297 	 * populating the descriptor then no memory handles need to be unbound
3298 	 * and we can return now.
3299 	 */
3300 	if (ldep->nbytes == 0) {
3301 		cv_signal(&vdc->dring_free_cv);
3302 		return (status);
3303 	}
3304 
3305 	/*
3306 	 * If the upper layer passed in a misaligned address we copied the
3307 	 * data into an aligned buffer before sending it to LDC - we now
3308 	 * copy it back to the original buffer.
3309 	 */
3310 	if (ldep->align_addr) {
3311 		ASSERT(ldep->addr != NULL);
3312 
3313 		if (dep->payload.nbytes > 0)
3314 			bcopy(ldep->align_addr, ldep->addr,
3315 			    dep->payload.nbytes);
3316 		kmem_free(ldep->align_addr,
3317 		    sizeof (caddr_t) * P2ROUNDUP(ldep->nbytes, 8));
3318 		ldep->align_addr = NULL;
3319 	}
3320 
3321 	rv = ldc_mem_unbind_handle(ldep->desc_mhdl);
3322 	if (rv != 0) {
3323 		DMSG(vdc, 0, "?[%d] unbind mhdl 0x%lx @ idx %d failed (%d)",
3324 		    vdc->instance, ldep->desc_mhdl, idx, rv);
3325 		/*
3326 		 * The error returned by the vDisk server is more informative
3327 		 * and thus has a higher priority but if it isn't set we ensure
3328 		 * that this function returns an error.
3329 		 */
3330 		if (status == 0)
3331 			status = EINVAL;
3332 	}
3333 
3334 	cv_signal(&vdc->membind_cv);
3335 	cv_signal(&vdc->dring_free_cv);
3336 
3337 	return (status);
3338 }
3339 
3340 /*
3341  * Function:
3342  *	vdc_populate_mem_hdl()
3343  *
3344  * Description:
3345  *
3346  * Arguments:
3347  *	vdc	- soft state pointer for this instance of the device driver.
3348  *	idx	- Index of the Descriptor Ring entry being modified
3349  *	addr	- virtual address being mapped in
3350  *	nybtes	- number of bytes in 'addr'
3351  *	operation - the vDisk operation being performed (VD_OP_xxx)
3352  *
3353  * Return Code:
3354  *	0	- Success
3355  */
3356 static int
3357 vdc_populate_mem_hdl(vdc_t *vdcp, vdc_local_desc_t *ldep)
3358 {
3359 	vd_dring_entry_t	*dep = NULL;
3360 	ldc_mem_handle_t	mhdl;
3361 	caddr_t			vaddr;
3362 	size_t			nbytes;
3363 	uint8_t			perm = LDC_MEM_RW;
3364 	uint8_t			maptype;
3365 	int			rv = 0;
3366 	int			i;
3367 
3368 	ASSERT(vdcp != NULL);
3369 
3370 	dep = ldep->dep;
3371 	mhdl = ldep->desc_mhdl;
3372 
3373 	switch (ldep->dir) {
3374 	case VIO_read_dir:
3375 		perm = LDC_MEM_W;
3376 		break;
3377 
3378 	case VIO_write_dir:
3379 		perm = LDC_MEM_R;
3380 		break;
3381 
3382 	case VIO_both_dir:
3383 		perm = LDC_MEM_RW;
3384 		break;
3385 
3386 	default:
3387 		ASSERT(0);	/* catch bad programming in vdc */
3388 	}
3389 
3390 	/*
3391 	 * LDC expects any addresses passed in to be 8-byte aligned. We need
3392 	 * to copy the contents of any misaligned buffers to a newly allocated
3393 	 * buffer and bind it instead (and copy the the contents back to the
3394 	 * original buffer passed in when depopulating the descriptor)
3395 	 */
3396 	vaddr = ldep->addr;
3397 	nbytes = ldep->nbytes;
3398 	if (((uint64_t)vaddr & 0x7) != 0) {
3399 		ASSERT(ldep->align_addr == NULL);
3400 		ldep->align_addr =
3401 		    kmem_alloc(sizeof (caddr_t) *
3402 		    P2ROUNDUP(nbytes, 8), KM_SLEEP);
3403 		DMSG(vdcp, 0, "[%d] Misaligned address %p reallocating "
3404 		    "(buf=%p nb=%ld op=%d)\n",
3405 		    vdcp->instance, (void *)vaddr, (void *)ldep->align_addr,
3406 		    nbytes, ldep->operation);
3407 		if (perm != LDC_MEM_W)
3408 			bcopy(vaddr, ldep->align_addr, nbytes);
3409 		vaddr = ldep->align_addr;
3410 	}
3411 
3412 	maptype = LDC_IO_MAP|LDC_SHADOW_MAP|LDC_DIRECT_MAP;
3413 	rv = ldc_mem_bind_handle(mhdl, vaddr, P2ROUNDUP(nbytes, 8),
3414 	    maptype, perm, &dep->payload.cookie[0], &dep->payload.ncookies);
3415 	DMSG(vdcp, 2, "[%d] bound mem handle; ncookies=%d\n",
3416 	    vdcp->instance, dep->payload.ncookies);
3417 	if (rv != 0) {
3418 		DMSG(vdcp, 0, "[%d] Failed to bind LDC memory handle "
3419 		    "(mhdl=%p, buf=%p, err=%d)\n",
3420 		    vdcp->instance, (void *)mhdl, (void *)vaddr, rv);
3421 		if (ldep->align_addr) {
3422 			kmem_free(ldep->align_addr,
3423 			    sizeof (caddr_t) * P2ROUNDUP(nbytes, 8));
3424 			ldep->align_addr = NULL;
3425 		}
3426 		return (EAGAIN);
3427 	}
3428 
3429 	/*
3430 	 * Get the other cookies (if any).
3431 	 */
3432 	for (i = 1; i < dep->payload.ncookies; i++) {
3433 		rv = ldc_mem_nextcookie(mhdl, &dep->payload.cookie[i]);
3434 		if (rv != 0) {
3435 			(void) ldc_mem_unbind_handle(mhdl);
3436 			DMSG(vdcp, 0, "?[%d] Failed to get next cookie "
3437 			    "(mhdl=%lx cnum=%d), err=%d",
3438 			    vdcp->instance, mhdl, i, rv);
3439 			if (ldep->align_addr) {
3440 				kmem_free(ldep->align_addr,
3441 				    sizeof (caddr_t) * ldep->nbytes);
3442 				ldep->align_addr = NULL;
3443 			}
3444 			return (EAGAIN);
3445 		}
3446 	}
3447 
3448 	return (rv);
3449 }
3450 
3451 /*
3452  * Interrupt handlers for messages from LDC
3453  */
3454 
3455 /*
3456  * Function:
3457  *	vdc_handle_cb()
3458  *
3459  * Description:
3460  *
3461  * Arguments:
3462  *	event	- Type of event (LDC_EVT_xxx) that triggered the callback
3463  *	arg	- soft state pointer for this instance of the device driver.
3464  *
3465  * Return Code:
3466  *	0	- Success
3467  */
3468 static uint_t
3469 vdc_handle_cb(uint64_t event, caddr_t arg)
3470 {
3471 	ldc_status_t	ldc_state;
3472 	int		rv = 0;
3473 
3474 	vdc_t	*vdc = (vdc_t *)(void *)arg;
3475 
3476 	ASSERT(vdc != NULL);
3477 
3478 	DMSG(vdc, 1, "evt=%lx seqID=%ld\n", event, vdc->seq_num);
3479 
3480 	/*
3481 	 * Depending on the type of event that triggered this callback,
3482 	 * we modify the handshake state or read the data.
3483 	 *
3484 	 * NOTE: not done as a switch() as event could be triggered by
3485 	 * a state change and a read request. Also the ordering	of the
3486 	 * check for the event types is deliberate.
3487 	 */
3488 	if (event & LDC_EVT_UP) {
3489 		DMSG(vdc, 0, "[%d] Received LDC_EVT_UP\n", vdc->instance);
3490 
3491 		mutex_enter(&vdc->lock);
3492 
3493 		/* get LDC state */
3494 		rv = ldc_status(vdc->ldc_handle, &ldc_state);
3495 		if (rv != 0) {
3496 			DMSG(vdc, 0, "[%d] Couldn't get LDC status %d",
3497 			    vdc->instance, rv);
3498 			return (LDC_SUCCESS);
3499 		}
3500 		if (vdc->ldc_state != LDC_UP && ldc_state == LDC_UP) {
3501 			/*
3502 			 * Reset the transaction sequence numbers when
3503 			 * LDC comes up. We then kick off the handshake
3504 			 * negotiation with the vDisk server.
3505 			 */
3506 			vdc->seq_num = 1;
3507 			vdc->seq_num_reply = 0;
3508 			vdc->ldc_state = ldc_state;
3509 			cv_signal(&vdc->initwait_cv);
3510 		}
3511 
3512 		mutex_exit(&vdc->lock);
3513 	}
3514 
3515 	if (event & LDC_EVT_READ) {
3516 		DMSG(vdc, 1, "[%d] Received LDC_EVT_READ\n", vdc->instance);
3517 		mutex_enter(&vdc->read_lock);
3518 		cv_signal(&vdc->read_cv);
3519 		vdc->read_state = VDC_READ_PENDING;
3520 		mutex_exit(&vdc->read_lock);
3521 
3522 		/* that's all we have to do - no need to handle DOWN/RESET */
3523 		return (LDC_SUCCESS);
3524 	}
3525 
3526 	if (event & (LDC_EVT_RESET|LDC_EVT_DOWN)) {
3527 
3528 		DMSG(vdc, 0, "[%d] Received LDC RESET event\n", vdc->instance);
3529 
3530 		mutex_enter(&vdc->lock);
3531 		/*
3532 		 * Need to wake up any readers so they will
3533 		 * detect that a reset has occurred.
3534 		 */
3535 		mutex_enter(&vdc->read_lock);
3536 		if ((vdc->read_state == VDC_READ_WAITING) ||
3537 		    (vdc->read_state == VDC_READ_RESET))
3538 			cv_signal(&vdc->read_cv);
3539 		vdc->read_state = VDC_READ_RESET;
3540 		mutex_exit(&vdc->read_lock);
3541 
3542 		/* wake up any threads waiting for connection to come up */
3543 		if (vdc->state == VDC_STATE_INIT_WAITING) {
3544 			vdc->state = VDC_STATE_RESETTING;
3545 			cv_signal(&vdc->initwait_cv);
3546 		}
3547 
3548 		mutex_exit(&vdc->lock);
3549 	}
3550 
3551 	if (event & ~(LDC_EVT_UP | LDC_EVT_RESET | LDC_EVT_DOWN | LDC_EVT_READ))
3552 		DMSG(vdc, 0, "![%d] Unexpected LDC event (%lx) received",
3553 		    vdc->instance, event);
3554 
3555 	return (LDC_SUCCESS);
3556 }
3557 
3558 /*
3559  * Function:
3560  *	vdc_wait_for_response()
3561  *
3562  * Description:
3563  *	Block waiting for a response from the server. If there is
3564  *	no data the thread block on the read_cv that is signalled
3565  *	by the callback when an EVT_READ occurs.
3566  *
3567  * Arguments:
3568  *	vdcp	- soft state pointer for this instance of the device driver.
3569  *
3570  * Return Code:
3571  *	0	- Success
3572  */
3573 static int
3574 vdc_wait_for_response(vdc_t *vdcp, vio_msg_t *msgp)
3575 {
3576 	size_t		nbytes = sizeof (*msgp);
3577 	int		status;
3578 
3579 	ASSERT(vdcp != NULL);
3580 
3581 	DMSG(vdcp, 1, "[%d] Entered\n", vdcp->instance);
3582 
3583 	status = vdc_recv(vdcp, msgp, &nbytes);
3584 	DMSG(vdcp, 3, "vdc_read() done.. status=0x%x size=0x%x\n",
3585 	    status, (int)nbytes);
3586 	if (status) {
3587 		DMSG(vdcp, 0, "?[%d] Error %d reading LDC msg\n",
3588 		    vdcp->instance, status);
3589 		return (status);
3590 	}
3591 
3592 	if (nbytes < sizeof (vio_msg_tag_t)) {
3593 		DMSG(vdcp, 0, "?[%d] Expect %lu bytes; recv'd %lu\n",
3594 		    vdcp->instance, sizeof (vio_msg_tag_t), nbytes);
3595 		return (ENOMSG);
3596 	}
3597 
3598 	DMSG(vdcp, 2, "[%d] (%x/%x/%x)\n", vdcp->instance,
3599 	    msgp->tag.vio_msgtype,
3600 	    msgp->tag.vio_subtype,
3601 	    msgp->tag.vio_subtype_env);
3602 
3603 	/*
3604 	 * Verify the Session ID of the message
3605 	 *
3606 	 * Every message after the Version has been negotiated should
3607 	 * have the correct session ID set.
3608 	 */
3609 	if ((msgp->tag.vio_sid != vdcp->session_id) &&
3610 	    (msgp->tag.vio_subtype_env != VIO_VER_INFO)) {
3611 		DMSG(vdcp, 0, "[%d] Invalid SID: received 0x%x, "
3612 		    "expected 0x%lx [seq num %lx @ %d]",
3613 		    vdcp->instance, msgp->tag.vio_sid,
3614 		    vdcp->session_id,
3615 		    ((vio_dring_msg_t *)msgp)->seq_num,
3616 		    ((vio_dring_msg_t *)msgp)->start_idx);
3617 		return (ENOMSG);
3618 	}
3619 	return (0);
3620 }
3621 
3622 
3623 /*
3624  * Function:
3625  *	vdc_resubmit_backup_dring()
3626  *
3627  * Description:
3628  *	Resubmit each descriptor in the backed up dring to
3629  * 	vDisk server. The Dring was backed up during connection
3630  *	reset.
3631  *
3632  * Arguments:
3633  *	vdcp	- soft state pointer for this instance of the device driver.
3634  *
3635  * Return Code:
3636  *	0	- Success
3637  */
3638 static int
3639 vdc_resubmit_backup_dring(vdc_t *vdcp)
3640 {
3641 	int		count;
3642 	int		b_idx;
3643 	int		rv;
3644 	int		dring_size;
3645 	int		status;
3646 	vio_msg_t	vio_msg;
3647 	vdc_local_desc_t	*curr_ldep;
3648 
3649 	ASSERT(MUTEX_NOT_HELD(&vdcp->lock));
3650 	ASSERT(vdcp->state == VDC_STATE_HANDLE_PENDING);
3651 
3652 	if (vdcp->local_dring_backup == NULL) {
3653 		/* the pending requests have already been processed */
3654 		return (0);
3655 	}
3656 
3657 	DMSG(vdcp, 1, "restoring pending dring entries (len=%d, tail=%d)\n",
3658 	    vdcp->local_dring_backup_len, vdcp->local_dring_backup_tail);
3659 
3660 	/*
3661 	 * Walk the backup copy of the local descriptor ring and
3662 	 * resubmit all the outstanding transactions.
3663 	 */
3664 	b_idx = vdcp->local_dring_backup_tail;
3665 	for (count = 0; count < vdcp->local_dring_backup_len; count++) {
3666 
3667 		curr_ldep = &(vdcp->local_dring_backup[b_idx]);
3668 
3669 		/* only resubmit outstanding transactions */
3670 		if (!curr_ldep->is_free) {
3671 
3672 			DMSG(vdcp, 1, "resubmitting entry idx=%x\n", b_idx);
3673 			mutex_enter(&vdcp->lock);
3674 			rv = vdc_populate_descriptor(vdcp, curr_ldep->operation,
3675 			    curr_ldep->addr, curr_ldep->nbytes,
3676 			    curr_ldep->slice, curr_ldep->offset,
3677 			    curr_ldep->cb_type, curr_ldep->cb_arg,
3678 			    curr_ldep->dir);
3679 			mutex_exit(&vdcp->lock);
3680 			if (rv) {
3681 				DMSG(vdcp, 1, "[%d] cannot resubmit entry %d\n",
3682 				    vdcp->instance, b_idx);
3683 				return (rv);
3684 			}
3685 
3686 			/* Wait for the response message. */
3687 			DMSG(vdcp, 1, "waiting for response to idx=%x\n",
3688 			    b_idx);
3689 			status = vdc_wait_for_response(vdcp, &vio_msg);
3690 			if (status) {
3691 				DMSG(vdcp, 1, "[%d] wait_for_response "
3692 				    "returned err=%d\n", vdcp->instance,
3693 				    status);
3694 				return (status);
3695 			}
3696 
3697 			DMSG(vdcp, 1, "processing msg for idx=%x\n", b_idx);
3698 			status = vdc_process_data_msg(vdcp, &vio_msg);
3699 			if (status) {
3700 				DMSG(vdcp, 1, "[%d] process_data_msg "
3701 				    "returned err=%d\n", vdcp->instance,
3702 				    status);
3703 				return (status);
3704 			}
3705 		}
3706 
3707 		/* get the next element to submit */
3708 		if (++b_idx >= vdcp->local_dring_backup_len)
3709 			b_idx = 0;
3710 	}
3711 
3712 	/* all done - now clear up pending dring copy */
3713 	dring_size = vdcp->local_dring_backup_len *
3714 	    sizeof (vdcp->local_dring_backup[0]);
3715 
3716 	(void) kmem_free(vdcp->local_dring_backup, dring_size);
3717 
3718 	vdcp->local_dring_backup = NULL;
3719 
3720 	return (0);
3721 }
3722 
3723 /*
3724  * Function:
3725  *	vdc_cancel_backup_dring
3726  *
3727  * Description:
3728  *	Cancel each descriptor in the backed up dring to vDisk server.
3729  *	The Dring was backed up during connection reset.
3730  *
3731  * Arguments:
3732  *	vdcp	- soft state pointer for this instance of the device driver.
3733  *
3734  * Return Code:
3735  *	None
3736  */
3737 void
3738 vdc_cancel_backup_ring(vdc_t *vdcp)
3739 {
3740 	vdc_local_desc_t *ldep;
3741 	struct buf 	*bufp;
3742 	int		count;
3743 	int		b_idx;
3744 	int		dring_size;
3745 
3746 	ASSERT(MUTEX_HELD(&vdcp->lock));
3747 	ASSERT(vdcp->state == VDC_STATE_INIT ||
3748 	    vdcp->state == VDC_STATE_INIT_WAITING ||
3749 	    vdcp->state == VDC_STATE_NEGOTIATE ||
3750 	    vdcp->state == VDC_STATE_RESETTING);
3751 
3752 	if (vdcp->local_dring_backup == NULL) {
3753 		/* the pending requests have already been processed */
3754 		return;
3755 	}
3756 
3757 	DMSG(vdcp, 1, "cancelling pending dring entries (len=%d, tail=%d)\n",
3758 	    vdcp->local_dring_backup_len, vdcp->local_dring_backup_tail);
3759 
3760 	/*
3761 	 * Walk the backup copy of the local descriptor ring and
3762 	 * cancel all the outstanding transactions.
3763 	 */
3764 	b_idx = vdcp->local_dring_backup_tail;
3765 	for (count = 0; count < vdcp->local_dring_backup_len; count++) {
3766 
3767 		ldep = &(vdcp->local_dring_backup[b_idx]);
3768 
3769 		/* only cancel outstanding transactions */
3770 		if (!ldep->is_free) {
3771 
3772 			DMSG(vdcp, 1, "cancelling entry idx=%x\n", b_idx);
3773 
3774 			/*
3775 			 * All requests have already been cleared from the
3776 			 * local descriptor ring and the LDC channel has been
3777 			 * reset so we will never get any reply for these
3778 			 * requests. Now we just have to notify threads waiting
3779 			 * for replies that the request has failed.
3780 			 */
3781 			switch (ldep->cb_type) {
3782 			case CB_SYNC:
3783 				ASSERT(vdcp->sync_op_pending);
3784 				vdcp->sync_op_status = EIO;
3785 				vdcp->sync_op_pending = B_FALSE;
3786 				cv_signal(&vdcp->sync_pending_cv);
3787 				break;
3788 
3789 			case CB_STRATEGY:
3790 				bufp = ldep->cb_arg;
3791 				ASSERT(bufp != NULL);
3792 				bufp->b_resid = bufp->b_bcount;
3793 				VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
3794 				VD_KSTAT_RUNQ_EXIT(vdcp->io_stats);
3795 				DTRACE_IO1(done, buf_t *, bufp);
3796 				bioerror(bufp, EIO);
3797 				biodone(bufp);
3798 				break;
3799 
3800 			default:
3801 				ASSERT(0);
3802 			}
3803 
3804 		}
3805 
3806 		/* get the next element to cancel */
3807 		if (++b_idx >= vdcp->local_dring_backup_len)
3808 			b_idx = 0;
3809 	}
3810 
3811 	/* all done - now clear up pending dring copy */
3812 	dring_size = vdcp->local_dring_backup_len *
3813 	    sizeof (vdcp->local_dring_backup[0]);
3814 
3815 	(void) kmem_free(vdcp->local_dring_backup, dring_size);
3816 
3817 	vdcp->local_dring_backup = NULL;
3818 
3819 	DTRACE_PROBE2(processed, int, count, vdc_t *, vdcp);
3820 }
3821 
3822 /*
3823  * Function:
3824  *	vdc_connection_timeout
3825  *
3826  * Description:
3827  *	This function is invoked if the timeout set to establish the connection
3828  *	with vds expires. This will happen if we spend too much time in the
3829  *	VDC_STATE_INIT_WAITING or VDC_STATE_NEGOTIATE states. Then we will
3830  *	cancel any pending request and mark them as failed.
3831  *
3832  *	If the timeout does not expire, it will be cancelled when we reach the
3833  *	VDC_STATE_HANDLE_PENDING or VDC_STATE_RESETTING state. This function can
3834  *	be invoked while we are in the VDC_STATE_HANDLE_PENDING or
3835  *	VDC_STATE_RESETTING state in which case we do nothing because the
3836  *	timeout is being cancelled.
3837  *
3838  * Arguments:
3839  *	arg	- argument of the timeout function actually a soft state
3840  *		  pointer for the instance of the device driver.
3841  *
3842  * Return Code:
3843  *	None
3844  */
3845 void
3846 vdc_connection_timeout(void *arg)
3847 {
3848 	vdc_t 		*vdcp = (vdc_t *)arg;
3849 
3850 	mutex_enter(&vdcp->lock);
3851 
3852 	if (vdcp->state == VDC_STATE_HANDLE_PENDING ||
3853 	    vdcp->state == VDC_STATE_DETACH) {
3854 		/*
3855 		 * The connection has just been re-established or
3856 		 * we are detaching.
3857 		 */
3858 		vdcp->ctimeout_reached = B_FALSE;
3859 		mutex_exit(&vdcp->lock);
3860 		return;
3861 	}
3862 
3863 	vdcp->ctimeout_reached = B_TRUE;
3864 
3865 	/* notify requests waiting for sending */
3866 	cv_broadcast(&vdcp->running_cv);
3867 
3868 	/* cancel requests waiting for a result */
3869 	vdc_cancel_backup_ring(vdcp);
3870 
3871 	mutex_exit(&vdcp->lock);
3872 
3873 	cmn_err(CE_NOTE, "[%d] connection to service domain timeout",
3874 	    vdcp->instance);
3875 }
3876 
3877 /*
3878  * Function:
3879  *	vdc_backup_local_dring()
3880  *
3881  * Description:
3882  *	Backup the current dring in the event of a reset. The Dring
3883  *	transactions will be resubmitted to the server when the
3884  *	connection is restored.
3885  *
3886  * Arguments:
3887  *	vdcp	- soft state pointer for this instance of the device driver.
3888  *
3889  * Return Code:
3890  *	NONE
3891  */
3892 static void
3893 vdc_backup_local_dring(vdc_t *vdcp)
3894 {
3895 	int dring_size;
3896 
3897 	ASSERT(MUTEX_HELD(&vdcp->lock));
3898 	ASSERT(vdcp->state == VDC_STATE_RESETTING);
3899 
3900 	/*
3901 	 * If the backup dring is stil around, it means
3902 	 * that the last restore did not complete. However,
3903 	 * since we never got back into the running state,
3904 	 * the backup copy we have is still valid.
3905 	 */
3906 	if (vdcp->local_dring_backup != NULL) {
3907 		DMSG(vdcp, 1, "reusing local descriptor ring backup "
3908 		    "(len=%d, tail=%d)\n", vdcp->local_dring_backup_len,
3909 		    vdcp->local_dring_backup_tail);
3910 		return;
3911 	}
3912 
3913 	/*
3914 	 * The backup dring can be NULL and the local dring may not be
3915 	 * initialized. This can happen if we had a reset while establishing
3916 	 * a new connection but after the connection has timed out. In that
3917 	 * case the backup dring is NULL because the requests have been
3918 	 * cancelled and the request occured before the local dring is
3919 	 * initialized.
3920 	 */
3921 	if (!(vdcp->initialized & VDC_DRING_LOCAL))
3922 		return;
3923 
3924 	DMSG(vdcp, 1, "backing up the local descriptor ring (len=%d, "
3925 	    "tail=%d)\n", vdcp->dring_len, vdcp->dring_curr_idx);
3926 
3927 	dring_size = vdcp->dring_len * sizeof (vdcp->local_dring[0]);
3928 
3929 	vdcp->local_dring_backup = kmem_alloc(dring_size, KM_SLEEP);
3930 	bcopy(vdcp->local_dring, vdcp->local_dring_backup, dring_size);
3931 
3932 	vdcp->local_dring_backup_tail = vdcp->dring_curr_idx;
3933 	vdcp->local_dring_backup_len = vdcp->dring_len;
3934 }
3935 
3936 /* -------------------------------------------------------------------------- */
3937 
3938 /*
3939  * The following functions process the incoming messages from vds
3940  */
3941 
3942 /*
3943  * Function:
3944  *      vdc_process_msg_thread()
3945  *
3946  * Description:
3947  *
3948  *	Main VDC message processing thread. Each vDisk instance
3949  * 	consists of a copy of this thread. This thread triggers
3950  * 	all the handshakes and data exchange with the server. It
3951  * 	also handles all channel resets
3952  *
3953  * Arguments:
3954  *      vdc     - soft state pointer for this instance of the device driver.
3955  *
3956  * Return Code:
3957  *      None
3958  */
3959 static void
3960 vdc_process_msg_thread(vdc_t *vdcp)
3961 {
3962 	int	status;
3963 	int	ctimeout;
3964 	timeout_id_t tmid = 0;
3965 
3966 	mutex_enter(&vdcp->lock);
3967 
3968 	for (;;) {
3969 
3970 #define	Q(_s)	(vdcp->state == _s) ? #_s :
3971 		DMSG(vdcp, 3, "state = %d (%s)\n", vdcp->state,
3972 		    Q(VDC_STATE_INIT)
3973 		    Q(VDC_STATE_INIT_WAITING)
3974 		    Q(VDC_STATE_NEGOTIATE)
3975 		    Q(VDC_STATE_HANDLE_PENDING)
3976 		    Q(VDC_STATE_RUNNING)
3977 		    Q(VDC_STATE_RESETTING)
3978 		    Q(VDC_STATE_DETACH)
3979 		    "UNKNOWN");
3980 
3981 		switch (vdcp->state) {
3982 		case VDC_STATE_INIT:
3983 
3984 			/*
3985 			 * If requested, start a timeout to check if the
3986 			 * connection with vds is established in the
3987 			 * specified delay. If the timeout expires, we
3988 			 * will cancel any pending request.
3989 			 *
3990 			 * If some reset have occurred while establishing
3991 			 * the connection, we already have a timeout armed
3992 			 * and in that case we don't need to arm a new one.
3993 			 */
3994 			ctimeout = (vdc_timeout != 0)?
3995 			    vdc_timeout : vdcp->ctimeout;
3996 
3997 			if (ctimeout != 0 && tmid == 0) {
3998 				tmid = timeout(vdc_connection_timeout, vdcp,
3999 				    ctimeout * drv_usectohz(1000000));
4000 			}
4001 
4002 			/* Check if have re-initializing repeatedly */
4003 			if (vdcp->hshake_cnt++ > vdc_hshake_retries &&
4004 			    vdcp->lifecycle != VDC_LC_ONLINE) {
4005 				cmn_err(CE_NOTE, "[%d] disk access failed.\n",
4006 				    vdcp->instance);
4007 				vdcp->state = VDC_STATE_DETACH;
4008 				break;
4009 			}
4010 
4011 			/* Bring up connection with vds via LDC */
4012 			status = vdc_start_ldc_connection(vdcp);
4013 			if (status == EINVAL) {
4014 				DMSG(vdcp, 0, "[%d] Could not start LDC",
4015 				    vdcp->instance);
4016 				vdcp->state = VDC_STATE_DETACH;
4017 			} else {
4018 				vdcp->state = VDC_STATE_INIT_WAITING;
4019 			}
4020 			break;
4021 
4022 		case VDC_STATE_INIT_WAITING:
4023 
4024 			/*
4025 			 * Let the callback event move us on
4026 			 * when channel is open to server
4027 			 */
4028 			while (vdcp->ldc_state != LDC_UP) {
4029 				cv_wait(&vdcp->initwait_cv, &vdcp->lock);
4030 				if (vdcp->state != VDC_STATE_INIT_WAITING) {
4031 					DMSG(vdcp, 0,
4032 				"state moved to %d out from under us...\n",
4033 					    vdcp->state);
4034 
4035 					break;
4036 				}
4037 			}
4038 			if (vdcp->state == VDC_STATE_INIT_WAITING &&
4039 			    vdcp->ldc_state == LDC_UP) {
4040 				vdcp->state = VDC_STATE_NEGOTIATE;
4041 			}
4042 			break;
4043 
4044 		case VDC_STATE_NEGOTIATE:
4045 			switch (status = vdc_ver_negotiation(vdcp)) {
4046 			case 0:
4047 				break;
4048 			default:
4049 				DMSG(vdcp, 0, "ver negotiate failed (%d)..\n",
4050 				    status);
4051 				goto reset;
4052 			}
4053 
4054 			switch (status = vdc_attr_negotiation(vdcp)) {
4055 			case 0:
4056 				break;
4057 			default:
4058 				DMSG(vdcp, 0, "attr negotiate failed (%d)..\n",
4059 				    status);
4060 				goto reset;
4061 			}
4062 
4063 			switch (status = vdc_dring_negotiation(vdcp)) {
4064 			case 0:
4065 				break;
4066 			default:
4067 				DMSG(vdcp, 0, "dring negotiate failed (%d)..\n",
4068 				    status);
4069 				goto reset;
4070 			}
4071 
4072 			switch (status = vdc_rdx_exchange(vdcp)) {
4073 			case 0:
4074 				vdcp->state = VDC_STATE_HANDLE_PENDING;
4075 				goto done;
4076 			default:
4077 				DMSG(vdcp, 0, "RDX xchg failed ..(%d)\n",
4078 				    status);
4079 				goto reset;
4080 			}
4081 reset:
4082 			DMSG(vdcp, 0, "negotiation failed: resetting (%d)\n",
4083 			    status);
4084 			vdcp->state = VDC_STATE_RESETTING;
4085 			vdcp->self_reset = B_TRUE;
4086 done:
4087 			DMSG(vdcp, 0, "negotiation complete (state=0x%x)...\n",
4088 			    vdcp->state);
4089 			break;
4090 
4091 		case VDC_STATE_HANDLE_PENDING:
4092 
4093 			if (vdcp->ctimeout_reached) {
4094 				/*
4095 				 * The connection timeout had been reached so
4096 				 * pending requests have been cancelled. Now
4097 				 * that the connection is back we can reset
4098 				 * the timeout.
4099 				 */
4100 				ASSERT(vdcp->local_dring_backup == NULL);
4101 				ASSERT(tmid != 0);
4102 				tmid = 0;
4103 				vdcp->ctimeout_reached = B_FALSE;
4104 				vdcp->state = VDC_STATE_RUNNING;
4105 				DMSG(vdcp, 0, "[%d] connection to service "
4106 				    "domain is up", vdcp->instance);
4107 				break;
4108 			}
4109 
4110 			mutex_exit(&vdcp->lock);
4111 			if (tmid != 0) {
4112 				(void) untimeout(tmid);
4113 				tmid = 0;
4114 			}
4115 			status = vdc_resubmit_backup_dring(vdcp);
4116 			mutex_enter(&vdcp->lock);
4117 
4118 			if (status)
4119 				vdcp->state = VDC_STATE_RESETTING;
4120 			else
4121 				vdcp->state = VDC_STATE_RUNNING;
4122 
4123 			break;
4124 
4125 		/* enter running state */
4126 		case VDC_STATE_RUNNING:
4127 			/*
4128 			 * Signal anyone waiting for the connection
4129 			 * to come on line.
4130 			 */
4131 			vdcp->hshake_cnt = 0;
4132 			cv_broadcast(&vdcp->running_cv);
4133 
4134 			/* failfast has to been checked after reset */
4135 			cv_signal(&vdcp->failfast_cv);
4136 
4137 			/* ownership is lost during reset */
4138 			if (vdcp->ownership & VDC_OWNERSHIP_WANTED)
4139 				vdcp->ownership |= VDC_OWNERSHIP_RESET;
4140 			cv_signal(&vdcp->ownership_cv);
4141 
4142 			mutex_exit(&vdcp->lock);
4143 
4144 			for (;;) {
4145 				vio_msg_t msg;
4146 				status = vdc_wait_for_response(vdcp, &msg);
4147 				if (status) break;
4148 
4149 				DMSG(vdcp, 1, "[%d] new pkt(s) available\n",
4150 				    vdcp->instance);
4151 				status = vdc_process_data_msg(vdcp, &msg);
4152 				if (status) {
4153 					DMSG(vdcp, 1, "[%d] process_data_msg "
4154 					    "returned err=%d\n", vdcp->instance,
4155 					    status);
4156 					break;
4157 				}
4158 
4159 			}
4160 
4161 			mutex_enter(&vdcp->lock);
4162 
4163 			vdcp->state = VDC_STATE_RESETTING;
4164 			vdcp->self_reset = B_TRUE;
4165 			break;
4166 
4167 		case VDC_STATE_RESETTING:
4168 			/*
4169 			 * When we reach this state, we either come from the
4170 			 * VDC_STATE_RUNNING state and we can have pending
4171 			 * request but no timeout is armed; or we come from
4172 			 * the VDC_STATE_INIT_WAITING, VDC_NEGOTIATE or
4173 			 * VDC_HANDLE_PENDING state and there is no pending
4174 			 * request or pending requests have already been copied
4175 			 * into the backup dring. So we can safely keep the
4176 			 * connection timeout armed while we are in this state.
4177 			 */
4178 
4179 			DMSG(vdcp, 0, "Initiating channel reset "
4180 			    "(pending = %d)\n", (int)vdcp->threads_pending);
4181 
4182 			if (vdcp->self_reset) {
4183 				DMSG(vdcp, 0,
4184 				    "[%d] calling stop_ldc_connection.\n",
4185 				    vdcp->instance);
4186 				status = vdc_stop_ldc_connection(vdcp);
4187 				vdcp->self_reset = B_FALSE;
4188 			}
4189 
4190 			/*
4191 			 * Wait for all threads currently waiting
4192 			 * for a free dring entry to use.
4193 			 */
4194 			while (vdcp->threads_pending) {
4195 				cv_broadcast(&vdcp->membind_cv);
4196 				cv_broadcast(&vdcp->dring_free_cv);
4197 				mutex_exit(&vdcp->lock);
4198 				/* give the waiters enough time to wake up */
4199 				delay(vdc_hz_min_ldc_delay);
4200 				mutex_enter(&vdcp->lock);
4201 			}
4202 
4203 			ASSERT(vdcp->threads_pending == 0);
4204 
4205 			/* Sanity check that no thread is receiving */
4206 			ASSERT(vdcp->read_state != VDC_READ_WAITING);
4207 
4208 			vdcp->read_state = VDC_READ_IDLE;
4209 
4210 			vdc_backup_local_dring(vdcp);
4211 
4212 			/* cleanup the old d-ring */
4213 			vdc_destroy_descriptor_ring(vdcp);
4214 
4215 			/* go and start again */
4216 			vdcp->state = VDC_STATE_INIT;
4217 
4218 			break;
4219 
4220 		case VDC_STATE_DETACH:
4221 			DMSG(vdcp, 0, "[%d] Reset thread exit cleanup ..\n",
4222 			    vdcp->instance);
4223 
4224 			/* cancel any pending timeout */
4225 			mutex_exit(&vdcp->lock);
4226 			if (tmid != 0) {
4227 				(void) untimeout(tmid);
4228 				tmid = 0;
4229 			}
4230 			mutex_enter(&vdcp->lock);
4231 
4232 			/*
4233 			 * Signal anyone waiting for connection
4234 			 * to come online
4235 			 */
4236 			cv_broadcast(&vdcp->running_cv);
4237 
4238 			while (vdcp->sync_op_pending) {
4239 				cv_signal(&vdcp->sync_pending_cv);
4240 				cv_signal(&vdcp->sync_blocked_cv);
4241 				mutex_exit(&vdcp->lock);
4242 				/* give the waiters enough time to wake up */
4243 				delay(vdc_hz_min_ldc_delay);
4244 				mutex_enter(&vdcp->lock);
4245 			}
4246 
4247 			mutex_exit(&vdcp->lock);
4248 
4249 			DMSG(vdcp, 0, "[%d] Msg processing thread exiting ..\n",
4250 			    vdcp->instance);
4251 			thread_exit();
4252 			break;
4253 		}
4254 	}
4255 }
4256 
4257 
4258 /*
4259  * Function:
4260  *	vdc_process_data_msg()
4261  *
4262  * Description:
4263  *	This function is called by the message processing thread each time
4264  *	a message with a msgtype of VIO_TYPE_DATA is received. It will either
4265  *	be an ACK or NACK from vds[1] which vdc handles as follows.
4266  *		ACK	- wake up the waiting thread
4267  *		NACK	- resend any messages necessary
4268  *
4269  *	[1] Although the message format allows it, vds should not send a
4270  *	    VIO_SUBTYPE_INFO message to vdc asking it to read data; if for
4271  *	    some bizarre reason it does, vdc will reset the connection.
4272  *
4273  * Arguments:
4274  *	vdc	- soft state pointer for this instance of the device driver.
4275  *	msg	- the LDC message sent by vds
4276  *
4277  * Return Code:
4278  *	0	- Success.
4279  *	> 0	- error value returned by LDC
4280  */
4281 static int
4282 vdc_process_data_msg(vdc_t *vdcp, vio_msg_t *msg)
4283 {
4284 	int			status = 0;
4285 	vio_dring_msg_t		*dring_msg;
4286 	vdc_local_desc_t	*ldep = NULL;
4287 	int			start, end;
4288 	int			idx;
4289 
4290 	dring_msg = (vio_dring_msg_t *)msg;
4291 
4292 	ASSERT(msg->tag.vio_msgtype == VIO_TYPE_DATA);
4293 	ASSERT(vdcp != NULL);
4294 
4295 	mutex_enter(&vdcp->lock);
4296 
4297 	/*
4298 	 * Check to see if the message has bogus data
4299 	 */
4300 	idx = start = dring_msg->start_idx;
4301 	end = dring_msg->end_idx;
4302 	if ((start >= vdcp->dring_len) ||
4303 	    (end >= vdcp->dring_len) || (end < -1)) {
4304 		DMSG(vdcp, 0, "[%d] Bogus ACK data : start %d, end %d\n",
4305 		    vdcp->instance, start, end);
4306 		VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4307 		mutex_exit(&vdcp->lock);
4308 		return (EINVAL);
4309 	}
4310 
4311 	/*
4312 	 * Verify that the sequence number is what vdc expects.
4313 	 */
4314 	switch (vdc_verify_seq_num(vdcp, dring_msg)) {
4315 	case VDC_SEQ_NUM_TODO:
4316 		break;	/* keep processing this message */
4317 	case VDC_SEQ_NUM_SKIP:
4318 		mutex_exit(&vdcp->lock);
4319 		return (0);
4320 	case VDC_SEQ_NUM_INVALID:
4321 		DMSG(vdcp, 0, "[%d] invalid seqno\n", vdcp->instance);
4322 		VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4323 		mutex_exit(&vdcp->lock);
4324 		return (ENXIO);
4325 	}
4326 
4327 	if (msg->tag.vio_subtype == VIO_SUBTYPE_NACK) {
4328 		DMSG(vdcp, 0, "[%d] DATA NACK\n", vdcp->instance);
4329 		VDC_DUMP_DRING_MSG(dring_msg);
4330 		VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4331 		mutex_exit(&vdcp->lock);
4332 		return (EIO);
4333 
4334 	} else if (msg->tag.vio_subtype == VIO_SUBTYPE_INFO) {
4335 		VD_UPDATE_ERR_STATS(vdcp, vd_protoerrs);
4336 		mutex_exit(&vdcp->lock);
4337 		return (EPROTO);
4338 	}
4339 
4340 	DMSG(vdcp, 1, ": start %d end %d\n", start, end);
4341 	ASSERT(start == end);
4342 
4343 	ldep = &vdcp->local_dring[idx];
4344 
4345 	DMSG(vdcp, 1, ": state 0x%x - cb_type 0x%x\n",
4346 	    ldep->dep->hdr.dstate, ldep->cb_type);
4347 
4348 	if (ldep->dep->hdr.dstate == VIO_DESC_DONE) {
4349 		struct buf *bufp;
4350 
4351 		switch (ldep->cb_type) {
4352 		case CB_SYNC:
4353 			ASSERT(vdcp->sync_op_pending);
4354 
4355 			status = vdc_depopulate_descriptor(vdcp, idx);
4356 			vdcp->sync_op_status = status;
4357 			vdcp->sync_op_pending = B_FALSE;
4358 			cv_signal(&vdcp->sync_pending_cv);
4359 			break;
4360 
4361 		case CB_STRATEGY:
4362 			bufp = ldep->cb_arg;
4363 			ASSERT(bufp != NULL);
4364 			bufp->b_resid =
4365 			    bufp->b_bcount - ldep->dep->payload.nbytes;
4366 			status = ldep->dep->payload.status; /* Future:ntoh */
4367 			if (status != 0) {
4368 				DMSG(vdcp, 1, "strategy status=%d\n", status);
4369 				VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4370 				bioerror(bufp, status);
4371 			}
4372 
4373 			(void) vdc_depopulate_descriptor(vdcp, idx);
4374 
4375 			DMSG(vdcp, 1,
4376 			    "strategy complete req=%ld bytes resp=%ld bytes\n",
4377 			    bufp->b_bcount, ldep->dep->payload.nbytes);
4378 
4379 			if (status != 0 && vdcp->failfast_interval != 0) {
4380 				/*
4381 				 * The I/O has failed and failfast is enabled.
4382 				 * We need the failfast thread to check if the
4383 				 * failure is due to a reservation conflict.
4384 				 */
4385 				(void) vdc_failfast_io_queue(vdcp, bufp);
4386 			} else {
4387 				if (status == 0) {
4388 					int op = (bufp->b_flags & B_READ) ?
4389 					    VD_OP_BREAD : VD_OP_BWRITE;
4390 					VD_UPDATE_IO_STATS(vdcp, op,
4391 					    ldep->dep->payload.nbytes);
4392 				}
4393 				VD_KSTAT_RUNQ_EXIT(vdcp->io_stats);
4394 				DTRACE_IO1(done, buf_t *, bufp);
4395 				biodone(bufp);
4396 			}
4397 			break;
4398 
4399 		default:
4400 			ASSERT(0);
4401 		}
4402 	}
4403 
4404 	/* let the arrival signal propogate */
4405 	mutex_exit(&vdcp->lock);
4406 
4407 	/* probe gives the count of how many entries were processed */
4408 	DTRACE_PROBE2(processed, int, 1, vdc_t *, vdcp);
4409 
4410 	return (0);
4411 }
4412 
4413 
4414 /*
4415  * Function:
4416  *	vdc_handle_ver_msg()
4417  *
4418  * Description:
4419  *
4420  * Arguments:
4421  *	vdc	- soft state pointer for this instance of the device driver.
4422  *	ver_msg	- LDC message sent by vDisk server
4423  *
4424  * Return Code:
4425  *	0	- Success
4426  */
4427 static int
4428 vdc_handle_ver_msg(vdc_t *vdc, vio_ver_msg_t *ver_msg)
4429 {
4430 	int status = 0;
4431 
4432 	ASSERT(vdc != NULL);
4433 	ASSERT(mutex_owned(&vdc->lock));
4434 
4435 	if (ver_msg->tag.vio_subtype_env != VIO_VER_INFO) {
4436 		return (EPROTO);
4437 	}
4438 
4439 	if (ver_msg->dev_class != VDEV_DISK_SERVER) {
4440 		return (EINVAL);
4441 	}
4442 
4443 	switch (ver_msg->tag.vio_subtype) {
4444 	case VIO_SUBTYPE_ACK:
4445 		/*
4446 		 * We check to see if the version returned is indeed supported
4447 		 * (The server may have also adjusted the minor number downwards
4448 		 * and if so 'ver_msg' will contain the actual version agreed)
4449 		 */
4450 		if (vdc_is_supported_version(ver_msg)) {
4451 			vdc->ver.major = ver_msg->ver_major;
4452 			vdc->ver.minor = ver_msg->ver_minor;
4453 			ASSERT(vdc->ver.major > 0);
4454 		} else {
4455 			status = EPROTO;
4456 		}
4457 		break;
4458 
4459 	case VIO_SUBTYPE_NACK:
4460 		/*
4461 		 * call vdc_is_supported_version() which will return the next
4462 		 * supported version (if any) in 'ver_msg'
4463 		 */
4464 		(void) vdc_is_supported_version(ver_msg);
4465 		if (ver_msg->ver_major > 0) {
4466 			size_t len = sizeof (*ver_msg);
4467 
4468 			ASSERT(vdc->ver.major > 0);
4469 
4470 			/* reset the necessary fields and resend */
4471 			ver_msg->tag.vio_subtype = VIO_SUBTYPE_INFO;
4472 			ver_msg->dev_class = VDEV_DISK;
4473 
4474 			status = vdc_send(vdc, (caddr_t)ver_msg, &len);
4475 			DMSG(vdc, 0, "[%d] Resend VER info (LDC status = %d)\n",
4476 			    vdc->instance, status);
4477 			if (len != sizeof (*ver_msg))
4478 				status = EBADMSG;
4479 		} else {
4480 			DMSG(vdc, 0, "[%d] No common version with vDisk server",
4481 			    vdc->instance);
4482 			status = ENOTSUP;
4483 		}
4484 
4485 		break;
4486 	case VIO_SUBTYPE_INFO:
4487 		/*
4488 		 * Handle the case where vds starts handshake
4489 		 * (for now only vdc is the instigator)
4490 		 */
4491 		status = ENOTSUP;
4492 		break;
4493 
4494 	default:
4495 		status = EINVAL;
4496 		break;
4497 	}
4498 
4499 	return (status);
4500 }
4501 
4502 /*
4503  * Function:
4504  *	vdc_handle_attr_msg()
4505  *
4506  * Description:
4507  *
4508  * Arguments:
4509  *	vdc	- soft state pointer for this instance of the device driver.
4510  *	attr_msg	- LDC message sent by vDisk server
4511  *
4512  * Return Code:
4513  *	0	- Success
4514  */
4515 static int
4516 vdc_handle_attr_msg(vdc_t *vdc, vd_attr_msg_t *attr_msg)
4517 {
4518 	int status = 0;
4519 
4520 	ASSERT(vdc != NULL);
4521 	ASSERT(mutex_owned(&vdc->lock));
4522 
4523 	if (attr_msg->tag.vio_subtype_env != VIO_ATTR_INFO) {
4524 		return (EPROTO);
4525 	}
4526 
4527 	switch (attr_msg->tag.vio_subtype) {
4528 	case VIO_SUBTYPE_ACK:
4529 		/*
4530 		 * We now verify the attributes sent by vds.
4531 		 */
4532 		if (attr_msg->vdisk_size == 0) {
4533 			DMSG(vdc, 0, "[%d] Invalid disk size from vds",
4534 			    vdc->instance);
4535 			status = EINVAL;
4536 			break;
4537 		}
4538 
4539 		if (attr_msg->max_xfer_sz == 0) {
4540 			DMSG(vdc, 0, "[%d] Invalid transfer size from vds",
4541 			    vdc->instance);
4542 			status = EINVAL;
4543 			break;
4544 		}
4545 
4546 		if (attr_msg->vdisk_size == VD_SIZE_UNKNOWN) {
4547 			DMSG(vdc, 0, "[%d] Unknown disk size from vds",
4548 			    vdc->instance);
4549 			attr_msg->vdisk_size = 0;
4550 		}
4551 
4552 		/*
4553 		 * If the disk size is already set check that it hasn't changed.
4554 		 */
4555 		if ((vdc->vdisk_size != 0) && (attr_msg->vdisk_size != 0) &&
4556 		    (vdc->vdisk_size != attr_msg->vdisk_size)) {
4557 			DMSG(vdc, 0, "[%d] Different disk size from vds "
4558 			    "(old=0x%lx - new=0x%lx", vdc->instance,
4559 			    vdc->vdisk_size, attr_msg->vdisk_size)
4560 			status = EINVAL;
4561 			break;
4562 		}
4563 
4564 		vdc->vdisk_size = attr_msg->vdisk_size;
4565 		vdc->vdisk_type = attr_msg->vdisk_type;
4566 		vdc->operations = attr_msg->operations;
4567 		if (vio_ver_is_supported(vdc->ver, 1, 1))
4568 			vdc->vdisk_media = attr_msg->vdisk_media;
4569 		else
4570 			vdc->vdisk_media = 0;
4571 
4572 		DMSG(vdc, 0, "[%d] max_xfer_sz: sent %lx acked %lx\n",
4573 		    vdc->instance, vdc->max_xfer_sz, attr_msg->max_xfer_sz);
4574 		DMSG(vdc, 0, "[%d] vdisk_block_size: sent %lx acked %x\n",
4575 		    vdc->instance, vdc->block_size,
4576 		    attr_msg->vdisk_block_size);
4577 
4578 		/*
4579 		 * We don't know at compile time what the vDisk server will
4580 		 * think are good values but we apply a large (arbitrary)
4581 		 * upper bound to prevent memory exhaustion in vdc if it was
4582 		 * allocating a DRing based of huge values sent by the server.
4583 		 * We probably will never exceed this except if the message
4584 		 * was garbage.
4585 		 */
4586 		if ((attr_msg->max_xfer_sz * attr_msg->vdisk_block_size) <=
4587 		    (PAGESIZE * DEV_BSIZE)) {
4588 			vdc->max_xfer_sz = attr_msg->max_xfer_sz;
4589 			vdc->block_size = attr_msg->vdisk_block_size;
4590 		} else {
4591 			DMSG(vdc, 0, "[%d] vds block transfer size too big;"
4592 			    " using max supported by vdc", vdc->instance);
4593 		}
4594 
4595 		if ((attr_msg->xfer_mode != VIO_DRING_MODE_V1_0) ||
4596 		    (attr_msg->vdisk_size > INT64_MAX) ||
4597 		    (attr_msg->operations == 0) ||
4598 		    (attr_msg->vdisk_type > VD_DISK_TYPE_DISK)) {
4599 			DMSG(vdc, 0, "[%d] Invalid attributes from vds",
4600 			    vdc->instance);
4601 			status = EINVAL;
4602 			break;
4603 		}
4604 
4605 		/*
4606 		 * Now that we have received all attributes we can create a
4607 		 * fake geometry for the disk.
4608 		 */
4609 		vdc_create_fake_geometry(vdc);
4610 		break;
4611 
4612 	case VIO_SUBTYPE_NACK:
4613 		/*
4614 		 * vds could not handle the attributes we sent so we
4615 		 * stop negotiating.
4616 		 */
4617 		status = EPROTO;
4618 		break;
4619 
4620 	case VIO_SUBTYPE_INFO:
4621 		/*
4622 		 * Handle the case where vds starts the handshake
4623 		 * (for now; vdc is the only supported instigatior)
4624 		 */
4625 		status = ENOTSUP;
4626 		break;
4627 
4628 	default:
4629 		status = ENOTSUP;
4630 		break;
4631 	}
4632 
4633 	return (status);
4634 }
4635 
4636 /*
4637  * Function:
4638  *	vdc_handle_dring_reg_msg()
4639  *
4640  * Description:
4641  *
4642  * Arguments:
4643  *	vdc		- soft state pointer for this instance of the driver.
4644  *	dring_msg	- LDC message sent by vDisk server
4645  *
4646  * Return Code:
4647  *	0	- Success
4648  */
4649 static int
4650 vdc_handle_dring_reg_msg(vdc_t *vdc, vio_dring_reg_msg_t *dring_msg)
4651 {
4652 	int		status = 0;
4653 
4654 	ASSERT(vdc != NULL);
4655 	ASSERT(mutex_owned(&vdc->lock));
4656 
4657 	if (dring_msg->tag.vio_subtype_env != VIO_DRING_REG) {
4658 		return (EPROTO);
4659 	}
4660 
4661 	switch (dring_msg->tag.vio_subtype) {
4662 	case VIO_SUBTYPE_ACK:
4663 		/* save the received dring_ident */
4664 		vdc->dring_ident = dring_msg->dring_ident;
4665 		DMSG(vdc, 0, "[%d] Received dring ident=0x%lx\n",
4666 		    vdc->instance, vdc->dring_ident);
4667 		break;
4668 
4669 	case VIO_SUBTYPE_NACK:
4670 		/*
4671 		 * vds could not handle the DRing info we sent so we
4672 		 * stop negotiating.
4673 		 */
4674 		DMSG(vdc, 0, "[%d] server could not register DRing\n",
4675 		    vdc->instance);
4676 		status = EPROTO;
4677 		break;
4678 
4679 	case VIO_SUBTYPE_INFO:
4680 		/*
4681 		 * Handle the case where vds starts handshake
4682 		 * (for now only vdc is the instigatior)
4683 		 */
4684 		status = ENOTSUP;
4685 		break;
4686 	default:
4687 		status = ENOTSUP;
4688 	}
4689 
4690 	return (status);
4691 }
4692 
4693 /*
4694  * Function:
4695  *	vdc_verify_seq_num()
4696  *
4697  * Description:
4698  *	This functions verifies that the sequence number sent back by the vDisk
4699  *	server with the latest message is what is expected (i.e. it is greater
4700  *	than the last seq num sent by the vDisk server and less than or equal
4701  *	to the last seq num generated by vdc).
4702  *
4703  *	It then checks the request ID to see if any requests need processing
4704  *	in the DRing.
4705  *
4706  * Arguments:
4707  *	vdc		- soft state pointer for this instance of the driver.
4708  *	dring_msg	- pointer to the LDC message sent by vds
4709  *
4710  * Return Code:
4711  *	VDC_SEQ_NUM_TODO	- Message needs to be processed
4712  *	VDC_SEQ_NUM_SKIP	- Message has already been processed
4713  *	VDC_SEQ_NUM_INVALID	- The seq numbers are so out of sync,
4714  *				  vdc cannot deal with them
4715  */
4716 static int
4717 vdc_verify_seq_num(vdc_t *vdc, vio_dring_msg_t *dring_msg)
4718 {
4719 	ASSERT(vdc != NULL);
4720 	ASSERT(dring_msg != NULL);
4721 	ASSERT(mutex_owned(&vdc->lock));
4722 
4723 	/*
4724 	 * Check to see if the messages were responded to in the correct
4725 	 * order by vds.
4726 	 */
4727 	if ((dring_msg->seq_num <= vdc->seq_num_reply) ||
4728 	    (dring_msg->seq_num > vdc->seq_num)) {
4729 		DMSG(vdc, 0, "?[%d] Bogus sequence_number %lu: "
4730 		    "%lu > expected <= %lu (last proc req %lu sent %lu)\n",
4731 		    vdc->instance, dring_msg->seq_num,
4732 		    vdc->seq_num_reply, vdc->seq_num,
4733 		    vdc->req_id_proc, vdc->req_id);
4734 		return (VDC_SEQ_NUM_INVALID);
4735 	}
4736 	vdc->seq_num_reply = dring_msg->seq_num;
4737 
4738 	if (vdc->req_id_proc < vdc->req_id)
4739 		return (VDC_SEQ_NUM_TODO);
4740 	else
4741 		return (VDC_SEQ_NUM_SKIP);
4742 }
4743 
4744 
4745 /*
4746  * Function:
4747  *	vdc_is_supported_version()
4748  *
4749  * Description:
4750  *	This routine checks if the major/minor version numbers specified in
4751  *	'ver_msg' are supported. If not it finds the next version that is
4752  *	in the supported version list 'vdc_version[]' and sets the fields in
4753  *	'ver_msg' to those values
4754  *
4755  * Arguments:
4756  *	ver_msg	- LDC message sent by vDisk server
4757  *
4758  * Return Code:
4759  *	B_TRUE	- Success
4760  *	B_FALSE	- Version not supported
4761  */
4762 static boolean_t
4763 vdc_is_supported_version(vio_ver_msg_t *ver_msg)
4764 {
4765 	int vdc_num_versions = sizeof (vdc_version) / sizeof (vdc_version[0]);
4766 
4767 	for (int i = 0; i < vdc_num_versions; i++) {
4768 		ASSERT(vdc_version[i].major > 0);
4769 		ASSERT((i == 0) ||
4770 		    (vdc_version[i].major < vdc_version[i-1].major));
4771 
4772 		/*
4773 		 * If the major versions match, adjust the minor version, if
4774 		 * necessary, down to the highest value supported by this
4775 		 * client. The server should support all minor versions lower
4776 		 * than the value it sent
4777 		 */
4778 		if (ver_msg->ver_major == vdc_version[i].major) {
4779 			if (ver_msg->ver_minor > vdc_version[i].minor) {
4780 				DMSGX(0,
4781 				    "Adjusting minor version from %u to %u",
4782 				    ver_msg->ver_minor, vdc_version[i].minor);
4783 				ver_msg->ver_minor = vdc_version[i].minor;
4784 			}
4785 			return (B_TRUE);
4786 		}
4787 
4788 		/*
4789 		 * If the message contains a higher major version number, set
4790 		 * the message's major/minor versions to the current values
4791 		 * and return false, so this message will get resent with
4792 		 * these values, and the server will potentially try again
4793 		 * with the same or a lower version
4794 		 */
4795 		if (ver_msg->ver_major > vdc_version[i].major) {
4796 			ver_msg->ver_major = vdc_version[i].major;
4797 			ver_msg->ver_minor = vdc_version[i].minor;
4798 			DMSGX(0, "Suggesting major/minor (0x%x/0x%x)\n",
4799 			    ver_msg->ver_major, ver_msg->ver_minor);
4800 
4801 			return (B_FALSE);
4802 		}
4803 
4804 		/*
4805 		 * Otherwise, the message's major version is less than the
4806 		 * current major version, so continue the loop to the next
4807 		 * (lower) supported version
4808 		 */
4809 	}
4810 
4811 	/*
4812 	 * No common version was found; "ground" the version pair in the
4813 	 * message to terminate negotiation
4814 	 */
4815 	ver_msg->ver_major = 0;
4816 	ver_msg->ver_minor = 0;
4817 
4818 	return (B_FALSE);
4819 }
4820 /* -------------------------------------------------------------------------- */
4821 
4822 /*
4823  * DKIO(7) support
4824  */
4825 
4826 typedef struct vdc_dk_arg {
4827 	struct dk_callback	dkc;
4828 	int			mode;
4829 	dev_t			dev;
4830 	vdc_t			*vdc;
4831 } vdc_dk_arg_t;
4832 
4833 /*
4834  * Function:
4835  * 	vdc_dkio_flush_cb()
4836  *
4837  * Description:
4838  *	This routine is a callback for DKIOCFLUSHWRITECACHE which can be called
4839  *	by kernel code.
4840  *
4841  * Arguments:
4842  *	arg	- a pointer to a vdc_dk_arg_t structure.
4843  */
4844 void
4845 vdc_dkio_flush_cb(void *arg)
4846 {
4847 	struct vdc_dk_arg	*dk_arg = (struct vdc_dk_arg *)arg;
4848 	struct dk_callback	*dkc = NULL;
4849 	vdc_t			*vdc = NULL;
4850 	int			rv;
4851 
4852 	if (dk_arg == NULL) {
4853 		cmn_err(CE_NOTE, "?[Unk] DKIOCFLUSHWRITECACHE arg is NULL\n");
4854 		return;
4855 	}
4856 	dkc = &dk_arg->dkc;
4857 	vdc = dk_arg->vdc;
4858 	ASSERT(vdc != NULL);
4859 
4860 	rv = vdc_do_sync_op(vdc, VD_OP_FLUSH, NULL, 0,
4861 	    VDCPART(dk_arg->dev), 0, CB_SYNC, 0, VIO_both_dir, B_TRUE);
4862 	if (rv != 0) {
4863 		DMSG(vdc, 0, "[%d] DKIOCFLUSHWRITECACHE failed %d : model %x\n",
4864 		    vdc->instance, rv,
4865 		    ddi_model_convert_from(dk_arg->mode & FMODELS));
4866 	}
4867 
4868 	/*
4869 	 * Trigger the call back to notify the caller the the ioctl call has
4870 	 * been completed.
4871 	 */
4872 	if ((dk_arg->mode & FKIOCTL) &&
4873 	    (dkc != NULL) &&
4874 	    (dkc->dkc_callback != NULL)) {
4875 		ASSERT(dkc->dkc_cookie != NULL);
4876 		(*dkc->dkc_callback)(dkc->dkc_cookie, rv);
4877 	}
4878 
4879 	/* Indicate that one less DKIO write flush is outstanding */
4880 	mutex_enter(&vdc->lock);
4881 	vdc->dkio_flush_pending--;
4882 	ASSERT(vdc->dkio_flush_pending >= 0);
4883 	mutex_exit(&vdc->lock);
4884 
4885 	/* free the mem that was allocated when the callback was dispatched */
4886 	kmem_free(arg, sizeof (vdc_dk_arg_t));
4887 }
4888 
4889 /*
4890  * Function:
4891  * 	vdc_dkio_get_partition()
4892  *
4893  * Description:
4894  *	This function implements the DKIOCGAPART ioctl.
4895  *
4896  * Arguments:
4897  *	vdc	- soft state pointer
4898  *	arg	- a pointer to a dk_map[NDKMAP] or dk_map32[NDKMAP] structure
4899  *	flag	- ioctl flags
4900  */
4901 static int
4902 vdc_dkio_get_partition(vdc_t *vdc, caddr_t arg, int flag)
4903 {
4904 	struct dk_geom *geom;
4905 	struct vtoc *vtoc;
4906 	union {
4907 		struct dk_map map[NDKMAP];
4908 		struct dk_map32 map32[NDKMAP];
4909 	} data;
4910 	int i, rv, size;
4911 
4912 	mutex_enter(&vdc->lock);
4913 
4914 	if ((rv = vdc_validate_geometry(vdc)) != 0) {
4915 		mutex_exit(&vdc->lock);
4916 		return (rv);
4917 	}
4918 
4919 	vtoc = vdc->vtoc;
4920 	geom = vdc->geom;
4921 
4922 	if (ddi_model_convert_from(flag & FMODELS) == DDI_MODEL_ILP32) {
4923 
4924 		for (i = 0; i < vtoc->v_nparts; i++) {
4925 			data.map32[i].dkl_cylno = vtoc->v_part[i].p_start /
4926 			    (geom->dkg_nhead * geom->dkg_nsect);
4927 			data.map32[i].dkl_nblk = vtoc->v_part[i].p_size;
4928 		}
4929 		size = NDKMAP * sizeof (struct dk_map32);
4930 
4931 	} else {
4932 
4933 		for (i = 0; i < vtoc->v_nparts; i++) {
4934 			data.map[i].dkl_cylno = vtoc->v_part[i].p_start /
4935 			    (geom->dkg_nhead * geom->dkg_nsect);
4936 			data.map[i].dkl_nblk = vtoc->v_part[i].p_size;
4937 		}
4938 		size = NDKMAP * sizeof (struct dk_map);
4939 
4940 	}
4941 
4942 	mutex_exit(&vdc->lock);
4943 
4944 	if (ddi_copyout(&data, arg, size, flag) != 0)
4945 		return (EFAULT);
4946 
4947 	return (0);
4948 }
4949 
4950 /*
4951  * Function:
4952  * 	vdc_dioctl_rwcmd()
4953  *
4954  * Description:
4955  *	This function implements the DIOCTL_RWCMD ioctl. This ioctl is used
4956  *	for DKC_DIRECT disks to read or write at an absolute disk offset.
4957  *
4958  * Arguments:
4959  *	dev	- device
4960  *	arg	- a pointer to a dadkio_rwcmd or dadkio_rwcmd32 structure
4961  *	flag	- ioctl flags
4962  */
4963 static int
4964 vdc_dioctl_rwcmd(dev_t dev, caddr_t arg, int flag)
4965 {
4966 	struct dadkio_rwcmd32 rwcmd32;
4967 	struct dadkio_rwcmd rwcmd;
4968 	struct iovec aiov;
4969 	struct uio auio;
4970 	int rw, status;
4971 	struct buf *buf;
4972 
4973 	if (ddi_model_convert_from(flag & FMODELS) == DDI_MODEL_ILP32) {
4974 		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd32,
4975 		    sizeof (struct dadkio_rwcmd32), flag)) {
4976 			return (EFAULT);
4977 		}
4978 		rwcmd.cmd = rwcmd32.cmd;
4979 		rwcmd.flags = rwcmd32.flags;
4980 		rwcmd.blkaddr = (daddr_t)rwcmd32.blkaddr;
4981 		rwcmd.buflen = rwcmd32.buflen;
4982 		rwcmd.bufaddr = (caddr_t)(uintptr_t)rwcmd32.bufaddr;
4983 	} else {
4984 		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd,
4985 		    sizeof (struct dadkio_rwcmd), flag)) {
4986 			return (EFAULT);
4987 		}
4988 	}
4989 
4990 	switch (rwcmd.cmd) {
4991 	case DADKIO_RWCMD_READ:
4992 		rw = B_READ;
4993 		break;
4994 	case DADKIO_RWCMD_WRITE:
4995 		rw = B_WRITE;
4996 		break;
4997 	default:
4998 		return (EINVAL);
4999 	}
5000 
5001 	bzero((caddr_t)&aiov, sizeof (struct iovec));
5002 	aiov.iov_base   = rwcmd.bufaddr;
5003 	aiov.iov_len    = rwcmd.buflen;
5004 
5005 	bzero((caddr_t)&auio, sizeof (struct uio));
5006 	auio.uio_iov    = &aiov;
5007 	auio.uio_iovcnt = 1;
5008 	auio.uio_loffset = rwcmd.blkaddr * DEV_BSIZE;
5009 	auio.uio_resid  = rwcmd.buflen;
5010 	auio.uio_segflg = flag & FKIOCTL ? UIO_SYSSPACE : UIO_USERSPACE;
5011 
5012 	buf = kmem_alloc(sizeof (buf_t), KM_SLEEP);
5013 	bioinit(buf);
5014 	/*
5015 	 * We use the private field of buf to specify that this is an
5016 	 * I/O using an absolute offset.
5017 	 */
5018 	buf->b_private = (void *)VD_SLICE_NONE;
5019 
5020 	status = physio(vdc_strategy, buf, dev, rw, vdc_min, &auio);
5021 
5022 	biofini(buf);
5023 	kmem_free(buf, sizeof (buf_t));
5024 
5025 	return (status);
5026 }
5027 
5028 /*
5029  * Allocate a buffer for a VD_OP_SCSICMD operation. The size of the allocated
5030  * buffer is returned in alloc_len.
5031  */
5032 static vd_scsi_t *
5033 vdc_scsi_alloc(int cdb_len, int sense_len, int datain_len, int dataout_len,
5034     int *alloc_len)
5035 {
5036 	vd_scsi_t *vd_scsi;
5037 	int vd_scsi_len = VD_SCSI_SIZE;
5038 
5039 	vd_scsi_len += P2ROUNDUP(cdb_len, sizeof (uint64_t));
5040 	vd_scsi_len += P2ROUNDUP(sense_len, sizeof (uint64_t));
5041 	vd_scsi_len += P2ROUNDUP(datain_len, sizeof (uint64_t));
5042 	vd_scsi_len += P2ROUNDUP(dataout_len, sizeof (uint64_t));
5043 
5044 	ASSERT(vd_scsi_len % sizeof (uint64_t) == 0);
5045 
5046 	vd_scsi = kmem_zalloc(vd_scsi_len, KM_SLEEP);
5047 
5048 	vd_scsi->cdb_len = cdb_len;
5049 	vd_scsi->sense_len = sense_len;
5050 	vd_scsi->datain_len = datain_len;
5051 	vd_scsi->dataout_len = dataout_len;
5052 
5053 	*alloc_len = vd_scsi_len;
5054 
5055 	return (vd_scsi);
5056 }
5057 
5058 /*
5059  * Convert the status of a SCSI command to a Solaris return code.
5060  *
5061  * Arguments:
5062  *	vd_scsi		- The SCSI operation buffer.
5063  *	log_error	- indicate if an error message should be logged.
5064  *
5065  * Note that our SCSI error messages are rather primitive for the moment
5066  * and could be improved by decoding some data like the SCSI command and
5067  * the sense key.
5068  *
5069  * Return value:
5070  *	0		- Status is good.
5071  *	EACCES		- Status reports a reservation conflict.
5072  *	ENOTSUP		- Status reports a check condition and sense key
5073  *			  reports an illegal request.
5074  *	EIO		- Any other status.
5075  */
5076 static int
5077 vdc_scsi_status(vdc_t *vdc, vd_scsi_t *vd_scsi, boolean_t log_error)
5078 {
5079 	int rv;
5080 	char path_str[MAXPATHLEN];
5081 	char panic_str[VDC_RESV_CONFLICT_FMT_LEN + MAXPATHLEN];
5082 	union scsi_cdb *cdb;
5083 	struct scsi_extended_sense *sense;
5084 
5085 	if (vd_scsi->cmd_status == STATUS_GOOD)
5086 		/* no error */
5087 		return (0);
5088 
5089 	/* when the tunable vdc_scsi_log_error is true we log all errors */
5090 	if (vdc_scsi_log_error)
5091 		log_error = B_TRUE;
5092 
5093 	if (log_error) {
5094 		cmn_err(CE_WARN, "%s (vdc%d):\tError for Command: 0x%x)\n",
5095 		    ddi_pathname(vdc->dip, path_str), vdc->instance,
5096 		    GETCMD(VD_SCSI_DATA_CDB(vd_scsi)));
5097 	}
5098 
5099 	/* default returned value */
5100 	rv = EIO;
5101 
5102 	switch (vd_scsi->cmd_status) {
5103 
5104 	case STATUS_CHECK:
5105 	case STATUS_TERMINATED:
5106 		if (log_error)
5107 			cmn_err(CE_CONT, "\tCheck Condition Error\n");
5108 
5109 		/* check sense buffer */
5110 		if (vd_scsi->sense_len == 0 ||
5111 		    vd_scsi->sense_status != STATUS_GOOD) {
5112 			if (log_error)
5113 				cmn_err(CE_CONT, "\tNo Sense Data Available\n");
5114 			break;
5115 		}
5116 
5117 		sense = VD_SCSI_DATA_SENSE(vd_scsi);
5118 
5119 		if (log_error) {
5120 			cmn_err(CE_CONT, "\tSense Key:  0x%x\n"
5121 			    "\tASC: 0x%x, ASCQ: 0x%x\n",
5122 			    scsi_sense_key((uint8_t *)sense),
5123 			    scsi_sense_asc((uint8_t *)sense),
5124 			    scsi_sense_ascq((uint8_t *)sense));
5125 		}
5126 
5127 		if (scsi_sense_key((uint8_t *)sense) == KEY_ILLEGAL_REQUEST)
5128 			rv = ENOTSUP;
5129 		break;
5130 
5131 	case STATUS_BUSY:
5132 		if (log_error)
5133 			cmn_err(CE_NOTE, "\tDevice Busy\n");
5134 		break;
5135 
5136 	case STATUS_RESERVATION_CONFLICT:
5137 		/*
5138 		 * If the command was PERSISTENT_RESERVATION_[IN|OUT] then
5139 		 * reservation conflict could be due to various reasons like
5140 		 * incorrect keys, not registered or not reserved etc. So,
5141 		 * we should not panic in that case.
5142 		 */
5143 		cdb = VD_SCSI_DATA_CDB(vd_scsi);
5144 		if (vdc->failfast_interval != 0 &&
5145 		    cdb->scc_cmd != SCMD_PERSISTENT_RESERVE_IN &&
5146 		    cdb->scc_cmd != SCMD_PERSISTENT_RESERVE_OUT) {
5147 			/* failfast is enabled so we have to panic */
5148 			(void) snprintf(panic_str, sizeof (panic_str),
5149 			    VDC_RESV_CONFLICT_FMT_STR "%s",
5150 			    ddi_pathname(vdc->dip, path_str));
5151 			panic(panic_str);
5152 		}
5153 		if (log_error)
5154 			cmn_err(CE_NOTE, "\tReservation Conflict\n");
5155 		rv = EACCES;
5156 		break;
5157 
5158 	case STATUS_QFULL:
5159 		if (log_error)
5160 			cmn_err(CE_NOTE, "\tQueue Full\n");
5161 		break;
5162 
5163 	case STATUS_MET:
5164 	case STATUS_INTERMEDIATE:
5165 	case STATUS_SCSI2:
5166 	case STATUS_INTERMEDIATE_MET:
5167 	case STATUS_ACA_ACTIVE:
5168 		if (log_error)
5169 			cmn_err(CE_CONT,
5170 			    "\tUnexpected SCSI status received: 0x%x\n",
5171 			    vd_scsi->cmd_status);
5172 		break;
5173 
5174 	default:
5175 		if (log_error)
5176 			cmn_err(CE_CONT,
5177 			    "\tInvalid SCSI status received: 0x%x\n",
5178 			    vd_scsi->cmd_status);
5179 		break;
5180 	}
5181 
5182 	return (rv);
5183 }
5184 
5185 /*
5186  * Implemented the USCSICMD uscsi(7I) ioctl. This ioctl is converted to
5187  * a VD_OP_SCSICMD operation which is sent to the vdisk server. If a SCSI
5188  * reset is requested (i.e. a flag USCSI_RESET* is set) then the ioctl is
5189  * converted to a VD_OP_RESET operation.
5190  */
5191 static int
5192 vdc_uscsi_cmd(vdc_t *vdc, caddr_t arg, int mode)
5193 {
5194 	struct uscsi_cmd 	uscsi;
5195 	struct uscsi_cmd32	uscsi32;
5196 	vd_scsi_t 		*vd_scsi;
5197 	int 			vd_scsi_len;
5198 	union scsi_cdb		*cdb;
5199 	struct scsi_extended_sense *sense;
5200 	char 			*datain, *dataout;
5201 	size_t			cdb_len, datain_len, dataout_len, sense_len;
5202 	int 			rv;
5203 
5204 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5205 		if (ddi_copyin(arg, &uscsi32, sizeof (struct uscsi_cmd32),
5206 		    mode) != 0)
5207 			return (EFAULT);
5208 		uscsi_cmd32touscsi_cmd((&uscsi32), (&uscsi));
5209 	} else {
5210 		if (ddi_copyin(arg, &uscsi, sizeof (struct uscsi_cmd),
5211 		    mode) != 0)
5212 			return (EFAULT);
5213 	}
5214 
5215 	/* a uscsi reset is converted to a VD_OP_RESET operation */
5216 	if (uscsi.uscsi_flags & (USCSI_RESET | USCSI_RESET_LUN |
5217 	    USCSI_RESET_ALL)) {
5218 		rv = vdc_do_sync_op(vdc, VD_OP_RESET, NULL, 0, 0, 0, CB_SYNC,
5219 		    (void *)(uint64_t)mode, VIO_both_dir, B_TRUE);
5220 		return (rv);
5221 	}
5222 
5223 	/* cdb buffer length */
5224 	cdb_len = uscsi.uscsi_cdblen;
5225 
5226 	/* data in and out buffers length */
5227 	if (uscsi.uscsi_flags & USCSI_READ) {
5228 		datain_len = uscsi.uscsi_buflen;
5229 		dataout_len = 0;
5230 	} else {
5231 		datain_len = 0;
5232 		dataout_len = uscsi.uscsi_buflen;
5233 	}
5234 
5235 	/* sense buffer length */
5236 	if (uscsi.uscsi_flags & USCSI_RQENABLE)
5237 		sense_len = uscsi.uscsi_rqlen;
5238 	else
5239 		sense_len = 0;
5240 
5241 	/* allocate buffer for the VD_SCSICMD_OP operation */
5242 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, datain_len, dataout_len,
5243 	    &vd_scsi_len);
5244 
5245 	/*
5246 	 * The documentation of USCSI_ISOLATE and USCSI_DIAGNOSE is very vague,
5247 	 * but basically they prevent a SCSI command from being retried in case
5248 	 * of an error.
5249 	 */
5250 	if ((uscsi.uscsi_flags & USCSI_ISOLATE) ||
5251 	    (uscsi.uscsi_flags & USCSI_DIAGNOSE))
5252 		vd_scsi->options |= VD_SCSI_OPT_NORETRY;
5253 
5254 	/* set task attribute */
5255 	if (uscsi.uscsi_flags & USCSI_NOTAG) {
5256 		vd_scsi->task_attribute = 0;
5257 	} else {
5258 		if (uscsi.uscsi_flags & USCSI_HEAD)
5259 			vd_scsi->task_attribute = VD_SCSI_TASK_ACA;
5260 		else if (uscsi.uscsi_flags & USCSI_HTAG)
5261 			vd_scsi->task_attribute = VD_SCSI_TASK_HQUEUE;
5262 		else if (uscsi.uscsi_flags & USCSI_OTAG)
5263 			vd_scsi->task_attribute = VD_SCSI_TASK_ORDERED;
5264 		else
5265 			vd_scsi->task_attribute = 0;
5266 	}
5267 
5268 	/* set timeout */
5269 	vd_scsi->timeout = uscsi.uscsi_timeout;
5270 
5271 	/* copy-in cdb data */
5272 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5273 	if (ddi_copyin(uscsi.uscsi_cdb, cdb, cdb_len, mode) != 0) {
5274 		rv = EFAULT;
5275 		goto done;
5276 	}
5277 
5278 	/* keep a pointer to the sense buffer */
5279 	sense = VD_SCSI_DATA_SENSE(vd_scsi);
5280 
5281 	/* keep a pointer to the data-in buffer */
5282 	datain = (char *)VD_SCSI_DATA_IN(vd_scsi);
5283 
5284 	/* copy-in request data to the data-out buffer */
5285 	dataout = (char *)VD_SCSI_DATA_OUT(vd_scsi);
5286 	if (!(uscsi.uscsi_flags & USCSI_READ)) {
5287 		if (ddi_copyin(uscsi.uscsi_bufaddr, dataout, dataout_len,
5288 		    mode)) {
5289 			rv = EFAULT;
5290 			goto done;
5291 		}
5292 	}
5293 
5294 	/* submit the request */
5295 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5296 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5297 
5298 	if (rv != 0)
5299 		goto done;
5300 
5301 	/* update scsi status */
5302 	uscsi.uscsi_status = vd_scsi->cmd_status;
5303 
5304 	/* update sense data */
5305 	if ((uscsi.uscsi_flags & USCSI_RQENABLE) &&
5306 	    (uscsi.uscsi_status == STATUS_CHECK ||
5307 	    uscsi.uscsi_status == STATUS_TERMINATED)) {
5308 
5309 		uscsi.uscsi_rqstatus = vd_scsi->sense_status;
5310 
5311 		if (uscsi.uscsi_rqstatus == STATUS_GOOD) {
5312 			uscsi.uscsi_rqresid = uscsi.uscsi_rqlen -
5313 			    vd_scsi->sense_len;
5314 			if (ddi_copyout(sense, uscsi.uscsi_rqbuf,
5315 			    vd_scsi->sense_len, mode) != 0) {
5316 				rv = EFAULT;
5317 				goto done;
5318 			}
5319 		}
5320 	}
5321 
5322 	/* update request data */
5323 	if (uscsi.uscsi_status == STATUS_GOOD) {
5324 		if (uscsi.uscsi_flags & USCSI_READ) {
5325 			uscsi.uscsi_resid = uscsi.uscsi_buflen -
5326 			    vd_scsi->datain_len;
5327 			if (ddi_copyout(datain, uscsi.uscsi_bufaddr,
5328 			    vd_scsi->datain_len, mode) != 0) {
5329 				rv = EFAULT;
5330 				goto done;
5331 			}
5332 		} else {
5333 			uscsi.uscsi_resid = uscsi.uscsi_buflen -
5334 			    vd_scsi->dataout_len;
5335 		}
5336 	}
5337 
5338 	/* copy-out result */
5339 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5340 		uscsi_cmdtouscsi_cmd32((&uscsi), (&uscsi32));
5341 		if (ddi_copyout(&uscsi32, arg, sizeof (struct uscsi_cmd32),
5342 		    mode) != 0) {
5343 			rv = EFAULT;
5344 			goto done;
5345 		}
5346 	} else {
5347 		if (ddi_copyout(&uscsi, arg, sizeof (struct uscsi_cmd),
5348 		    mode) != 0) {
5349 			rv = EFAULT;
5350 			goto done;
5351 		}
5352 	}
5353 
5354 	/* get the return code from the SCSI command status */
5355 	rv = vdc_scsi_status(vdc, vd_scsi,
5356 	    !(uscsi.uscsi_flags & USCSI_SILENT));
5357 
5358 done:
5359 	kmem_free(vd_scsi, vd_scsi_len);
5360 	return (rv);
5361 }
5362 
5363 /*
5364  * Create a VD_OP_SCSICMD buffer for a SCSI PERSISTENT IN command.
5365  *
5366  * Arguments:
5367  *	cmd		- SCSI PERSISTENT IN command
5368  *	len		- length of the SCSI input buffer
5369  *	vd_scsi_len	- return the length of the allocated buffer
5370  *
5371  * Returned Value:
5372  *	a pointer to the allocated VD_OP_SCSICMD buffer.
5373  */
5374 static vd_scsi_t *
5375 vdc_scsi_alloc_persistent_in(uchar_t cmd, int len, int *vd_scsi_len)
5376 {
5377 	int cdb_len, sense_len, datain_len, dataout_len;
5378 	vd_scsi_t *vd_scsi;
5379 	union scsi_cdb *cdb;
5380 
5381 	cdb_len = CDB_GROUP1;
5382 	sense_len = sizeof (struct scsi_extended_sense);
5383 	datain_len = len;
5384 	dataout_len = 0;
5385 
5386 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, datain_len, dataout_len,
5387 	    vd_scsi_len);
5388 
5389 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5390 
5391 	/* set cdb */
5392 	cdb->scc_cmd = SCMD_PERSISTENT_RESERVE_IN;
5393 	cdb->cdb_opaque[1] = cmd;
5394 	FORMG1COUNT(cdb, datain_len);
5395 
5396 	vd_scsi->timeout = vdc_scsi_timeout;
5397 
5398 	return (vd_scsi);
5399 }
5400 
5401 /*
5402  * Create a VD_OP_SCSICMD buffer for a SCSI PERSISTENT OUT command.
5403  *
5404  * Arguments:
5405  *	cmd		- SCSI PERSISTENT OUT command
5406  *	len		- length of the SCSI output buffer
5407  *	vd_scsi_len	- return the length of the allocated buffer
5408  *
5409  * Returned Code:
5410  *	a pointer to the allocated VD_OP_SCSICMD buffer.
5411  */
5412 static vd_scsi_t *
5413 vdc_scsi_alloc_persistent_out(uchar_t cmd, int len, int *vd_scsi_len)
5414 {
5415 	int cdb_len, sense_len, datain_len, dataout_len;
5416 	vd_scsi_t *vd_scsi;
5417 	union scsi_cdb *cdb;
5418 
5419 	cdb_len = CDB_GROUP1;
5420 	sense_len = sizeof (struct scsi_extended_sense);
5421 	datain_len = 0;
5422 	dataout_len = len;
5423 
5424 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, datain_len, dataout_len,
5425 	    vd_scsi_len);
5426 
5427 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5428 
5429 	/* set cdb */
5430 	cdb->scc_cmd = SCMD_PERSISTENT_RESERVE_OUT;
5431 	cdb->cdb_opaque[1] = cmd;
5432 	FORMG1COUNT(cdb, dataout_len);
5433 
5434 	vd_scsi->timeout = vdc_scsi_timeout;
5435 
5436 	return (vd_scsi);
5437 }
5438 
5439 /*
5440  * Implement the MHIOCGRP_INKEYS mhd(7i) ioctl. The ioctl is converted
5441  * to a SCSI PERSISTENT IN READ KEYS command which is sent to the vdisk
5442  * server with a VD_OP_SCSICMD operation.
5443  */
5444 static int
5445 vdc_mhd_inkeys(vdc_t *vdc, caddr_t arg, int mode)
5446 {
5447 	vd_scsi_t *vd_scsi;
5448 	mhioc_inkeys_t inkeys;
5449 	mhioc_key_list_t klist;
5450 	struct mhioc_inkeys32 inkeys32;
5451 	struct mhioc_key_list32 klist32;
5452 	sd_prin_readkeys_t *scsi_keys;
5453 	void *user_keys;
5454 	int vd_scsi_len;
5455 	int listsize, listlen, rv;
5456 
5457 	/* copyin arguments */
5458 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5459 		rv = ddi_copyin(arg, &inkeys32, sizeof (inkeys32), mode);
5460 		if (rv != 0)
5461 			return (EFAULT);
5462 
5463 		rv = ddi_copyin((caddr_t)(uintptr_t)inkeys32.li, &klist32,
5464 		    sizeof (klist32), mode);
5465 		if (rv != 0)
5466 			return (EFAULT);
5467 
5468 		listsize = klist32.listsize;
5469 	} else {
5470 		rv = ddi_copyin(arg, &inkeys, sizeof (inkeys), mode);
5471 		if (rv != 0)
5472 			return (EFAULT);
5473 
5474 		rv = ddi_copyin(inkeys.li, &klist, sizeof (klist), mode);
5475 		if (rv != 0)
5476 			return (EFAULT);
5477 
5478 		listsize = klist.listsize;
5479 	}
5480 
5481 	/* build SCSI VD_OP request */
5482 	vd_scsi = vdc_scsi_alloc_persistent_in(SD_READ_KEYS,
5483 	    sizeof (sd_prin_readkeys_t) - sizeof (caddr_t) +
5484 	    (sizeof (mhioc_resv_key_t) * listsize), &vd_scsi_len);
5485 
5486 	scsi_keys = (sd_prin_readkeys_t *)VD_SCSI_DATA_IN(vd_scsi);
5487 
5488 	/* submit the request */
5489 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5490 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5491 
5492 	if (rv != 0)
5493 		goto done;
5494 
5495 	listlen = scsi_keys->len / MHIOC_RESV_KEY_SIZE;
5496 
5497 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5498 		inkeys32.generation = scsi_keys->generation;
5499 		rv = ddi_copyout(&inkeys32, arg, sizeof (inkeys32), mode);
5500 		if (rv != 0) {
5501 			rv = EFAULT;
5502 			goto done;
5503 		}
5504 
5505 		klist32.listlen = listlen;
5506 		rv = ddi_copyout(&klist32, (caddr_t)(uintptr_t)inkeys32.li,
5507 		    sizeof (klist32), mode);
5508 		if (rv != 0) {
5509 			rv = EFAULT;
5510 			goto done;
5511 		}
5512 
5513 		user_keys = (caddr_t)(uintptr_t)klist32.list;
5514 	} else {
5515 		inkeys.generation = scsi_keys->generation;
5516 		rv = ddi_copyout(&inkeys, arg, sizeof (inkeys), mode);
5517 		if (rv != 0) {
5518 			rv = EFAULT;
5519 			goto done;
5520 		}
5521 
5522 		klist.listlen = listlen;
5523 		rv = ddi_copyout(&klist, inkeys.li, sizeof (klist), mode);
5524 		if (rv != 0) {
5525 			rv = EFAULT;
5526 			goto done;
5527 		}
5528 
5529 		user_keys = klist.list;
5530 	}
5531 
5532 	/* copy out keys */
5533 	if (listlen > 0 && listsize > 0) {
5534 		if (listsize < listlen)
5535 			listlen = listsize;
5536 		rv = ddi_copyout(&scsi_keys->keylist, user_keys,
5537 		    listlen * MHIOC_RESV_KEY_SIZE, mode);
5538 		if (rv != 0)
5539 			rv = EFAULT;
5540 	}
5541 
5542 	if (rv == 0)
5543 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
5544 
5545 done:
5546 	kmem_free(vd_scsi, vd_scsi_len);
5547 
5548 	return (rv);
5549 }
5550 
5551 /*
5552  * Implement the MHIOCGRP_INRESV mhd(7i) ioctl. The ioctl is converted
5553  * to a SCSI PERSISTENT IN READ RESERVATION command which is sent to
5554  * the vdisk server with a VD_OP_SCSICMD operation.
5555  */
5556 static int
5557 vdc_mhd_inresv(vdc_t *vdc, caddr_t arg, int mode)
5558 {
5559 	vd_scsi_t *vd_scsi;
5560 	mhioc_inresvs_t inresv;
5561 	mhioc_resv_desc_list_t rlist;
5562 	struct mhioc_inresvs32 inresv32;
5563 	struct mhioc_resv_desc_list32 rlist32;
5564 	mhioc_resv_desc_t mhd_resv;
5565 	sd_prin_readresv_t *scsi_resv;
5566 	sd_readresv_desc_t *resv;
5567 	mhioc_resv_desc_t *user_resv;
5568 	int vd_scsi_len;
5569 	int listsize, listlen, i, rv;
5570 
5571 	/* copyin arguments */
5572 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5573 		rv = ddi_copyin(arg, &inresv32, sizeof (inresv32), mode);
5574 		if (rv != 0)
5575 			return (EFAULT);
5576 
5577 		rv = ddi_copyin((caddr_t)(uintptr_t)inresv32.li, &rlist32,
5578 		    sizeof (rlist32), mode);
5579 		if (rv != 0)
5580 			return (EFAULT);
5581 
5582 		listsize = rlist32.listsize;
5583 	} else {
5584 		rv = ddi_copyin(arg, &inresv, sizeof (inresv), mode);
5585 		if (rv != 0)
5586 			return (EFAULT);
5587 
5588 		rv = ddi_copyin(inresv.li, &rlist, sizeof (rlist), mode);
5589 		if (rv != 0)
5590 			return (EFAULT);
5591 
5592 		listsize = rlist.listsize;
5593 	}
5594 
5595 	/* build SCSI VD_OP request */
5596 	vd_scsi = vdc_scsi_alloc_persistent_in(SD_READ_RESV,
5597 	    sizeof (sd_prin_readresv_t) - sizeof (caddr_t) +
5598 	    (SCSI3_RESV_DESC_LEN * listsize), &vd_scsi_len);
5599 
5600 	scsi_resv = (sd_prin_readresv_t *)VD_SCSI_DATA_IN(vd_scsi);
5601 
5602 	/* submit the request */
5603 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5604 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5605 
5606 	if (rv != 0)
5607 		goto done;
5608 
5609 	listlen = scsi_resv->len / SCSI3_RESV_DESC_LEN;
5610 
5611 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5612 		inresv32.generation = scsi_resv->generation;
5613 		rv = ddi_copyout(&inresv32, arg, sizeof (inresv32), mode);
5614 		if (rv != 0) {
5615 			rv = EFAULT;
5616 			goto done;
5617 		}
5618 
5619 		rlist32.listlen = listlen;
5620 		rv = ddi_copyout(&rlist32, (caddr_t)(uintptr_t)inresv32.li,
5621 		    sizeof (rlist32), mode);
5622 		if (rv != 0) {
5623 			rv = EFAULT;
5624 			goto done;
5625 		}
5626 
5627 		user_resv = (mhioc_resv_desc_t *)(uintptr_t)rlist32.list;
5628 	} else {
5629 		inresv.generation = scsi_resv->generation;
5630 		rv = ddi_copyout(&inresv, arg, sizeof (inresv), mode);
5631 		if (rv != 0) {
5632 			rv = EFAULT;
5633 			goto done;
5634 		}
5635 
5636 		rlist.listlen = listlen;
5637 		rv = ddi_copyout(&rlist, inresv.li, sizeof (rlist), mode);
5638 		if (rv != 0) {
5639 			rv = EFAULT;
5640 			goto done;
5641 		}
5642 
5643 		user_resv = rlist.list;
5644 	}
5645 
5646 	/* copy out reservations */
5647 	if (listsize > 0 && listlen > 0) {
5648 		if (listsize < listlen)
5649 			listlen = listsize;
5650 		resv = (sd_readresv_desc_t *)&scsi_resv->readresv_desc;
5651 
5652 		for (i = 0; i < listlen; i++) {
5653 			mhd_resv.type = resv->type;
5654 			mhd_resv.scope = resv->scope;
5655 			mhd_resv.scope_specific_addr =
5656 			    BE_32(resv->scope_specific_addr);
5657 			bcopy(&resv->resvkey, &mhd_resv.key,
5658 			    MHIOC_RESV_KEY_SIZE);
5659 
5660 			rv = ddi_copyout(&mhd_resv, user_resv,
5661 			    sizeof (mhd_resv), mode);
5662 			if (rv != 0) {
5663 				rv = EFAULT;
5664 				goto done;
5665 			}
5666 			resv++;
5667 			user_resv++;
5668 		}
5669 	}
5670 
5671 	if (rv == 0)
5672 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
5673 
5674 done:
5675 	kmem_free(vd_scsi, vd_scsi_len);
5676 	return (rv);
5677 }
5678 
5679 /*
5680  * Implement the MHIOCGRP_REGISTER mhd(7i) ioctl. The ioctl is converted
5681  * to a SCSI PERSISTENT OUT REGISTER command which is sent to the vdisk
5682  * server with a VD_OP_SCSICMD operation.
5683  */
5684 static int
5685 vdc_mhd_register(vdc_t *vdc, caddr_t arg, int mode)
5686 {
5687 	vd_scsi_t *vd_scsi;
5688 	sd_prout_t *scsi_prout;
5689 	mhioc_register_t mhd_reg;
5690 	int vd_scsi_len, rv;
5691 
5692 	/* copyin arguments */
5693 	rv = ddi_copyin(arg, &mhd_reg, sizeof (mhd_reg), mode);
5694 	if (rv != 0)
5695 		return (EFAULT);
5696 
5697 	/* build SCSI VD_OP request */
5698 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_REGISTER,
5699 	    sizeof (sd_prout_t), &vd_scsi_len);
5700 
5701 	/* set parameters */
5702 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
5703 	bcopy(mhd_reg.oldkey.key, scsi_prout->res_key, MHIOC_RESV_KEY_SIZE);
5704 	bcopy(mhd_reg.newkey.key, scsi_prout->service_key, MHIOC_RESV_KEY_SIZE);
5705 	scsi_prout->aptpl = (uchar_t)mhd_reg.aptpl;
5706 
5707 	/* submit the request */
5708 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5709 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5710 
5711 	if (rv == 0)
5712 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
5713 
5714 	kmem_free(vd_scsi, vd_scsi_len);
5715 	return (rv);
5716 }
5717 
5718 /*
5719  * Implement the MHIOCGRP_RESERVE mhd(7i) ioctl. The ioctl is converted
5720  * to a SCSI PERSISTENT OUT RESERVE command which is sent to the vdisk
5721  * server with a VD_OP_SCSICMD operation.
5722  */
5723 static int
5724 vdc_mhd_reserve(vdc_t *vdc, caddr_t arg, int mode)
5725 {
5726 	union scsi_cdb *cdb;
5727 	vd_scsi_t *vd_scsi;
5728 	sd_prout_t *scsi_prout;
5729 	mhioc_resv_desc_t mhd_resv;
5730 	int vd_scsi_len, rv;
5731 
5732 	/* copyin arguments */
5733 	rv = ddi_copyin(arg, &mhd_resv, sizeof (mhd_resv), mode);
5734 	if (rv != 0)
5735 		return (EFAULT);
5736 
5737 	/* build SCSI VD_OP request */
5738 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_RESERVE,
5739 	    sizeof (sd_prout_t), &vd_scsi_len);
5740 
5741 	/* set parameters */
5742 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5743 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
5744 	bcopy(mhd_resv.key.key, scsi_prout->res_key, MHIOC_RESV_KEY_SIZE);
5745 	scsi_prout->scope_address = mhd_resv.scope_specific_addr;
5746 	cdb->cdb_opaque[2] = mhd_resv.type;
5747 
5748 	/* submit the request */
5749 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5750 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5751 
5752 	if (rv == 0)
5753 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
5754 
5755 	kmem_free(vd_scsi, vd_scsi_len);
5756 	return (rv);
5757 }
5758 
5759 /*
5760  * Implement the MHIOCGRP_PREEMPTANDABORT mhd(7i) ioctl. The ioctl is
5761  * converted to a SCSI PERSISTENT OUT PREEMPT AND ABORT command which
5762  * is sent to the vdisk server with a VD_OP_SCSICMD operation.
5763  */
5764 static int
5765 vdc_mhd_preemptabort(vdc_t *vdc, caddr_t arg, int mode)
5766 {
5767 	union scsi_cdb *cdb;
5768 	vd_scsi_t *vd_scsi;
5769 	sd_prout_t *scsi_prout;
5770 	mhioc_preemptandabort_t mhd_preempt;
5771 	int vd_scsi_len, rv;
5772 
5773 	/* copyin arguments */
5774 	rv = ddi_copyin(arg, &mhd_preempt, sizeof (mhd_preempt), mode);
5775 	if (rv != 0)
5776 		return (EFAULT);
5777 
5778 	/* build SCSI VD_OP request */
5779 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_PREEMPTANDABORT,
5780 	    sizeof (sd_prout_t), &vd_scsi_len);
5781 
5782 	/* set parameters */
5783 	vd_scsi->task_attribute = VD_SCSI_TASK_ACA;
5784 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5785 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
5786 	bcopy(mhd_preempt.resvdesc.key.key, scsi_prout->res_key,
5787 	    MHIOC_RESV_KEY_SIZE);
5788 	bcopy(mhd_preempt.victim_key.key, scsi_prout->service_key,
5789 	    MHIOC_RESV_KEY_SIZE);
5790 	scsi_prout->scope_address = mhd_preempt.resvdesc.scope_specific_addr;
5791 	cdb->cdb_opaque[2] = mhd_preempt.resvdesc.type;
5792 
5793 	/* submit the request */
5794 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5795 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5796 
5797 	if (rv == 0)
5798 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
5799 
5800 	kmem_free(vd_scsi, vd_scsi_len);
5801 	return (rv);
5802 }
5803 
5804 /*
5805  * Implement the MHIOCGRP_REGISTERANDIGNOREKEY mhd(7i) ioctl. The ioctl
5806  * is converted to a SCSI PERSISTENT OUT REGISTER AND IGNORE EXISTING KEY
5807  * command which is sent to the vdisk server with a VD_OP_SCSICMD operation.
5808  */
5809 static int
5810 vdc_mhd_registerignore(vdc_t *vdc, caddr_t arg, int mode)
5811 {
5812 	vd_scsi_t *vd_scsi;
5813 	sd_prout_t *scsi_prout;
5814 	mhioc_registerandignorekey_t mhd_regi;
5815 	int vd_scsi_len, rv;
5816 
5817 	/* copyin arguments */
5818 	rv = ddi_copyin(arg, &mhd_regi, sizeof (mhd_regi), mode);
5819 	if (rv != 0)
5820 		return (EFAULT);
5821 
5822 	/* build SCSI VD_OP request */
5823 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_REGISTERANDIGNOREKEY,
5824 	    sizeof (sd_prout_t), &vd_scsi_len);
5825 
5826 	/* set parameters */
5827 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
5828 	bcopy(mhd_regi.newkey.key, scsi_prout->service_key,
5829 	    MHIOC_RESV_KEY_SIZE);
5830 	scsi_prout->aptpl = (uchar_t)mhd_regi.aptpl;
5831 
5832 	/* submit the request */
5833 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5834 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5835 
5836 	if (rv == 0)
5837 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
5838 
5839 	kmem_free(vd_scsi, vd_scsi_len);
5840 	return (rv);
5841 }
5842 
5843 /*
5844  * This function is used by the failfast mechanism to send a SCSI command
5845  * to check for reservation conflict.
5846  */
5847 static int
5848 vdc_failfast_scsi_cmd(vdc_t *vdc, uchar_t scmd)
5849 {
5850 	int cdb_len, sense_len, vd_scsi_len;
5851 	vd_scsi_t *vd_scsi;
5852 	union scsi_cdb *cdb;
5853 	int rv;
5854 
5855 	ASSERT(scmd == SCMD_TEST_UNIT_READY || scmd == SCMD_WRITE_G1);
5856 
5857 	if (scmd == SCMD_WRITE_G1)
5858 		cdb_len = CDB_GROUP1;
5859 	else
5860 		cdb_len = CDB_GROUP0;
5861 
5862 	sense_len = sizeof (struct scsi_extended_sense);
5863 
5864 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, 0, 0, &vd_scsi_len);
5865 
5866 	/* set cdb */
5867 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5868 	cdb->scc_cmd = scmd;
5869 
5870 	vd_scsi->timeout = vdc_scsi_timeout;
5871 
5872 	/*
5873 	 * Submit the request. The last argument has to be B_FALSE so that
5874 	 * vdc_do_sync_op does not loop checking for reservation conflict if
5875 	 * the operation returns an error.
5876 	 */
5877 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5878 	    0, 0, CB_SYNC, (void *)(uint64_t)FKIOCTL, VIO_both_dir, B_FALSE);
5879 
5880 	if (rv == 0)
5881 		(void) vdc_scsi_status(vdc, vd_scsi, B_FALSE);
5882 
5883 	kmem_free(vd_scsi, vd_scsi_len);
5884 	return (rv);
5885 }
5886 
5887 /*
5888  * This function is used by the failfast mechanism to check for reservation
5889  * conflict. It sends some SCSI commands which will fail with a reservation
5890  * conflict error if the system does not have access to the disk and this
5891  * will panic the system.
5892  *
5893  * Returned Code:
5894  *	0	- disk is accessible without reservation conflict error
5895  *	!= 0	- unable to check if disk is accessible
5896  */
5897 int
5898 vdc_failfast_check_resv(vdc_t *vdc)
5899 {
5900 	int failure = 0;
5901 
5902 	/*
5903 	 * Send a TEST UNIT READY command. The command will panic
5904 	 * the system if it fails with a reservation conflict.
5905 	 */
5906 	if (vdc_failfast_scsi_cmd(vdc, SCMD_TEST_UNIT_READY) != 0)
5907 		failure++;
5908 
5909 	/*
5910 	 * With SPC-3 compliant devices TEST UNIT READY will succeed on
5911 	 * a reserved device, so we also do a WRITE(10) of zero byte in
5912 	 * order to provoke a Reservation Conflict status on those newer
5913 	 * devices.
5914 	 */
5915 	if (vdc_failfast_scsi_cmd(vdc, SCMD_WRITE_G1) != 0)
5916 		failure++;
5917 
5918 	return (failure);
5919 }
5920 
5921 /*
5922  * Add a pending I/O to the failfast I/O queue. An I/O is added to this
5923  * queue when it has failed and failfast is enabled. Then we have to check
5924  * if it has failed because of a reservation conflict in which case we have
5925  * to panic the system.
5926  *
5927  * Async I/O should be queued with their block I/O data transfer structure
5928  * (buf). Sync I/O should be queued with buf = NULL.
5929  */
5930 static vdc_io_t *
5931 vdc_failfast_io_queue(vdc_t *vdc, struct buf *buf)
5932 {
5933 	vdc_io_t *vio;
5934 
5935 	ASSERT(MUTEX_HELD(&vdc->lock));
5936 
5937 	vio = kmem_alloc(sizeof (vdc_io_t), KM_SLEEP);
5938 	vio->vio_next = vdc->failfast_io_queue;
5939 	vio->vio_buf = buf;
5940 	vio->vio_qtime = ddi_get_lbolt();
5941 
5942 	vdc->failfast_io_queue = vio;
5943 
5944 	/* notify the failfast thread that a new I/O is queued */
5945 	cv_signal(&vdc->failfast_cv);
5946 
5947 	return (vio);
5948 }
5949 
5950 /*
5951  * Remove and complete I/O in the failfast I/O queue which have been
5952  * added after the indicated deadline. A deadline of 0 means that all
5953  * I/O have to be unqueued and marked as completed.
5954  */
5955 static void
5956 vdc_failfast_io_unqueue(vdc_t *vdc, clock_t deadline)
5957 {
5958 	vdc_io_t *vio, *vio_tmp;
5959 
5960 	ASSERT(MUTEX_HELD(&vdc->lock));
5961 
5962 	vio_tmp = NULL;
5963 	vio = vdc->failfast_io_queue;
5964 
5965 	if (deadline != 0) {
5966 		/*
5967 		 * Skip any io queued after the deadline. The failfast
5968 		 * I/O queue is ordered starting with the last I/O added
5969 		 * to the queue.
5970 		 */
5971 		while (vio != NULL && vio->vio_qtime > deadline) {
5972 			vio_tmp = vio;
5973 			vio = vio->vio_next;
5974 		}
5975 	}
5976 
5977 	if (vio == NULL)
5978 		/* nothing to unqueue */
5979 		return;
5980 
5981 	/* update the queue */
5982 	if (vio_tmp == NULL)
5983 		vdc->failfast_io_queue = NULL;
5984 	else
5985 		vio_tmp->vio_next = NULL;
5986 
5987 	/*
5988 	 * Complete unqueued I/O. Async I/O have a block I/O data transfer
5989 	 * structure (buf) and they are completed by calling biodone(). Sync
5990 	 * I/O do not have a buf and they are completed by setting the
5991 	 * vio_qtime to zero and signaling failfast_io_cv. In that case, the
5992 	 * thread waiting for the I/O to complete is responsible for freeing
5993 	 * the vio structure.
5994 	 */
5995 	while (vio != NULL) {
5996 		vio_tmp = vio->vio_next;
5997 		if (vio->vio_buf != NULL) {
5998 			VD_KSTAT_RUNQ_EXIT(vdc->io_stats);
5999 			DTRACE_IO1(done, buf_t *, vio->vio_buf);
6000 			biodone(vio->vio_buf);
6001 			kmem_free(vio, sizeof (vdc_io_t));
6002 		} else {
6003 			vio->vio_qtime = 0;
6004 		}
6005 		vio = vio_tmp;
6006 	}
6007 
6008 	cv_broadcast(&vdc->failfast_io_cv);
6009 }
6010 
6011 /*
6012  * Failfast Thread.
6013  *
6014  * While failfast is enabled, the failfast thread sends a TEST UNIT READY
6015  * and a zero size WRITE(10) SCSI commands on a regular basis to check that
6016  * we still have access to the disk. If a command fails with a RESERVATION
6017  * CONFLICT error then the system will immediatly panic.
6018  *
6019  * The failfast thread is also woken up when an I/O has failed. It then check
6020  * the access to the disk to ensure that the I/O failure was not due to a
6021  * reservation conflict.
6022  *
6023  * There is one failfast thread for each virtual disk for which failfast is
6024  * enabled. We could have only one thread sending requests for all disks but
6025  * this would need vdc to send asynchronous requests and to have callbacks to
6026  * process replies.
6027  */
6028 static void
6029 vdc_failfast_thread(void *arg)
6030 {
6031 	int status;
6032 	vdc_t *vdc = (vdc_t *)arg;
6033 	clock_t timeout, starttime;
6034 
6035 	mutex_enter(&vdc->lock);
6036 
6037 	while (vdc->failfast_interval != 0) {
6038 
6039 		starttime = ddi_get_lbolt();
6040 
6041 		mutex_exit(&vdc->lock);
6042 
6043 		/* check for reservation conflict */
6044 		status = vdc_failfast_check_resv(vdc);
6045 
6046 		mutex_enter(&vdc->lock);
6047 		/*
6048 		 * We have dropped the lock to send the SCSI command so we have
6049 		 * to check that failfast is still enabled.
6050 		 */
6051 		if (vdc->failfast_interval == 0)
6052 			break;
6053 
6054 		/*
6055 		 * If we have successfully check the disk access and there was
6056 		 * no reservation conflict then we can complete any I/O queued
6057 		 * before the last check.
6058 		 */
6059 		if (status == 0)
6060 			vdc_failfast_io_unqueue(vdc, starttime);
6061 
6062 		/* proceed again if some I/O are still in the queue */
6063 		if (vdc->failfast_io_queue != NULL)
6064 			continue;
6065 
6066 		timeout = ddi_get_lbolt() +
6067 		    drv_usectohz(vdc->failfast_interval);
6068 		(void) cv_timedwait(&vdc->failfast_cv, &vdc->lock, timeout);
6069 	}
6070 
6071 	/*
6072 	 * Failfast is being stop so we can complete any queued I/O.
6073 	 */
6074 	vdc_failfast_io_unqueue(vdc, 0);
6075 	vdc->failfast_thread = NULL;
6076 	mutex_exit(&vdc->lock);
6077 	thread_exit();
6078 }
6079 
6080 /*
6081  * Implement the MHIOCENFAILFAST mhd(7i) ioctl.
6082  */
6083 static int
6084 vdc_failfast(vdc_t *vdc, caddr_t arg, int mode)
6085 {
6086 	unsigned int mh_time;
6087 
6088 	if (ddi_copyin((void *)arg, &mh_time, sizeof (int), mode))
6089 		return (EFAULT);
6090 
6091 	mutex_enter(&vdc->lock);
6092 	if (mh_time != 0 && vdc->failfast_thread == NULL) {
6093 		vdc->failfast_thread = thread_create(NULL, 0,
6094 		    vdc_failfast_thread, vdc, 0, &p0, TS_RUN,
6095 		    v.v_maxsyspri - 2);
6096 	}
6097 
6098 	vdc->failfast_interval = mh_time * 1000;
6099 	cv_signal(&vdc->failfast_cv);
6100 	mutex_exit(&vdc->lock);
6101 
6102 	return (0);
6103 }
6104 
6105 /*
6106  * Implement the MHIOCTKOWN and MHIOCRELEASE mhd(7i) ioctls. These ioctls are
6107  * converted to VD_OP_SET_ACCESS operations.
6108  */
6109 static int
6110 vdc_access_set(vdc_t *vdc, uint64_t flags, int mode)
6111 {
6112 	int rv;
6113 
6114 	/* submit owership command request */
6115 	rv = vdc_do_sync_op(vdc, VD_OP_SET_ACCESS, (caddr_t)&flags,
6116 	    sizeof (uint64_t), 0, 0, CB_SYNC, (void *)(uint64_t)mode,
6117 	    VIO_both_dir, B_TRUE);
6118 
6119 	return (rv);
6120 }
6121 
6122 /*
6123  * Implement the MHIOCSTATUS mhd(7i) ioctl. This ioctl is converted to a
6124  * VD_OP_GET_ACCESS operation.
6125  */
6126 static int
6127 vdc_access_get(vdc_t *vdc, uint64_t *status, int mode)
6128 {
6129 	int rv;
6130 
6131 	/* submit owership command request */
6132 	rv = vdc_do_sync_op(vdc, VD_OP_GET_ACCESS, (caddr_t)status,
6133 	    sizeof (uint64_t), 0, 0, CB_SYNC, (void *)(uint64_t)mode,
6134 	    VIO_both_dir, B_TRUE);
6135 
6136 	return (rv);
6137 }
6138 
6139 /*
6140  * Disk Ownership Thread.
6141  *
6142  * When we have taken the ownership of a disk, this thread waits to be
6143  * notified when the LDC channel is reset so that it can recover the
6144  * ownership.
6145  *
6146  * Note that the thread handling the LDC reset (vdc_process_msg_thread())
6147  * can not be used to do the ownership recovery because it has to be
6148  * running to handle the reply message to the ownership operation.
6149  */
6150 static void
6151 vdc_ownership_thread(void *arg)
6152 {
6153 	vdc_t *vdc = (vdc_t *)arg;
6154 	clock_t timeout;
6155 	uint64_t status;
6156 
6157 	mutex_enter(&vdc->ownership_lock);
6158 	mutex_enter(&vdc->lock);
6159 
6160 	while (vdc->ownership & VDC_OWNERSHIP_WANTED) {
6161 
6162 		if ((vdc->ownership & VDC_OWNERSHIP_RESET) ||
6163 		    !(vdc->ownership & VDC_OWNERSHIP_GRANTED)) {
6164 			/*
6165 			 * There was a reset so the ownership has been lost,
6166 			 * try to recover. We do this without using the preempt
6167 			 * option so that we don't steal the ownership from
6168 			 * someone who has preempted us.
6169 			 */
6170 			DMSG(vdc, 0, "[%d] Ownership lost, recovering",
6171 			    vdc->instance);
6172 
6173 			vdc->ownership &= ~(VDC_OWNERSHIP_RESET |
6174 			    VDC_OWNERSHIP_GRANTED);
6175 
6176 			mutex_exit(&vdc->lock);
6177 
6178 			status = vdc_access_set(vdc, VD_ACCESS_SET_EXCLUSIVE |
6179 			    VD_ACCESS_SET_PRESERVE, FKIOCTL);
6180 
6181 			mutex_enter(&vdc->lock);
6182 
6183 			if (status == 0) {
6184 				DMSG(vdc, 0, "[%d] Ownership recovered",
6185 				    vdc->instance);
6186 				vdc->ownership |= VDC_OWNERSHIP_GRANTED;
6187 			} else {
6188 				DMSG(vdc, 0, "[%d] Fail to recover ownership",
6189 				    vdc->instance);
6190 			}
6191 
6192 		}
6193 
6194 		/*
6195 		 * If we have the ownership then we just wait for an event
6196 		 * to happen (LDC reset), otherwise we will retry to recover
6197 		 * after a delay.
6198 		 */
6199 		if (vdc->ownership & VDC_OWNERSHIP_GRANTED)
6200 			timeout = 0;
6201 		else
6202 			timeout = ddi_get_lbolt() +
6203 			    drv_usectohz(vdc_ownership_delay);
6204 
6205 		/* Release the ownership_lock and wait on the vdc lock */
6206 		mutex_exit(&vdc->ownership_lock);
6207 
6208 		if (timeout == 0)
6209 			(void) cv_wait(&vdc->ownership_cv, &vdc->lock);
6210 		else
6211 			(void) cv_timedwait(&vdc->ownership_cv,
6212 			    &vdc->lock, timeout);
6213 
6214 		mutex_exit(&vdc->lock);
6215 
6216 		mutex_enter(&vdc->ownership_lock);
6217 		mutex_enter(&vdc->lock);
6218 	}
6219 
6220 	vdc->ownership_thread = NULL;
6221 	mutex_exit(&vdc->lock);
6222 	mutex_exit(&vdc->ownership_lock);
6223 
6224 	thread_exit();
6225 }
6226 
6227 static void
6228 vdc_ownership_update(vdc_t *vdc, int ownership_flags)
6229 {
6230 	ASSERT(MUTEX_HELD(&vdc->ownership_lock));
6231 
6232 	mutex_enter(&vdc->lock);
6233 	vdc->ownership = ownership_flags;
6234 	if ((vdc->ownership & VDC_OWNERSHIP_WANTED) &&
6235 	    vdc->ownership_thread == NULL) {
6236 		/* start ownership thread */
6237 		vdc->ownership_thread = thread_create(NULL, 0,
6238 		    vdc_ownership_thread, vdc, 0, &p0, TS_RUN,
6239 		    v.v_maxsyspri - 2);
6240 	} else {
6241 		/* notify the ownership thread */
6242 		cv_signal(&vdc->ownership_cv);
6243 	}
6244 	mutex_exit(&vdc->lock);
6245 }
6246 
6247 /*
6248  * Get the size and the block size of a virtual disk from the vdisk server.
6249  * We need to use this operation when the vdisk_size attribute was not
6250  * available during the handshake with the vdisk server.
6251  */
6252 static int
6253 vdc_check_capacity(vdc_t *vdc)
6254 {
6255 	int rv = 0;
6256 	size_t alloc_len;
6257 	vd_capacity_t *vd_cap;
6258 
6259 	if (vdc->vdisk_size != 0)
6260 		return (0);
6261 
6262 	alloc_len = P2ROUNDUP(sizeof (vd_capacity_t), sizeof (uint64_t));
6263 
6264 	vd_cap = kmem_zalloc(alloc_len, KM_SLEEP);
6265 
6266 	rv = vdc_do_sync_op(vdc, VD_OP_GET_CAPACITY, (caddr_t)vd_cap, alloc_len,
6267 	    0, 0, CB_SYNC, (void *)(uint64_t)FKIOCTL, VIO_both_dir, B_TRUE);
6268 
6269 	if (rv == 0) {
6270 		if (vd_cap->vdisk_block_size != vdc->block_size ||
6271 		    vd_cap->vdisk_size == VD_SIZE_UNKNOWN ||
6272 		    vd_cap->vdisk_size == 0)
6273 			rv = EINVAL;
6274 		else
6275 			vdc->vdisk_size = vd_cap->vdisk_size;
6276 	}
6277 
6278 	kmem_free(vd_cap, alloc_len);
6279 	return (rv);
6280 }
6281 
6282 /*
6283  * This structure is used in the DKIO(7I) array below.
6284  */
6285 typedef struct vdc_dk_ioctl {
6286 	uint8_t		op;		/* VD_OP_XXX value */
6287 	int		cmd;		/* Solaris ioctl operation number */
6288 	size_t		nbytes;		/* size of structure to be copied */
6289 
6290 	/* function to convert between vDisk and Solaris structure formats */
6291 	int	(*convert)(vdc_t *vdc, void *vd_buf, void *ioctl_arg,
6292 	    int mode, int dir);
6293 } vdc_dk_ioctl_t;
6294 
6295 /*
6296  * Subset of DKIO(7I) operations currently supported
6297  */
6298 static vdc_dk_ioctl_t	dk_ioctl[] = {
6299 	{VD_OP_FLUSH,		DKIOCFLUSHWRITECACHE,	0,
6300 		vdc_null_copy_func},
6301 	{VD_OP_GET_WCE,		DKIOCGETWCE,		sizeof (int),
6302 		vdc_get_wce_convert},
6303 	{VD_OP_SET_WCE,		DKIOCSETWCE,		sizeof (int),
6304 		vdc_set_wce_convert},
6305 	{VD_OP_GET_VTOC,	DKIOCGVTOC,		sizeof (vd_vtoc_t),
6306 		vdc_get_vtoc_convert},
6307 	{VD_OP_SET_VTOC,	DKIOCSVTOC,		sizeof (vd_vtoc_t),
6308 		vdc_set_vtoc_convert},
6309 	{VD_OP_GET_DISKGEOM,	DKIOCGGEOM,		sizeof (vd_geom_t),
6310 		vdc_get_geom_convert},
6311 	{VD_OP_GET_DISKGEOM,	DKIOCG_PHYGEOM,		sizeof (vd_geom_t),
6312 		vdc_get_geom_convert},
6313 	{VD_OP_GET_DISKGEOM, 	DKIOCG_VIRTGEOM,	sizeof (vd_geom_t),
6314 		vdc_get_geom_convert},
6315 	{VD_OP_SET_DISKGEOM,	DKIOCSGEOM,		sizeof (vd_geom_t),
6316 		vdc_set_geom_convert},
6317 	{VD_OP_GET_EFI,		DKIOCGETEFI,		0,
6318 		vdc_get_efi_convert},
6319 	{VD_OP_SET_EFI,		DKIOCSETEFI,		0,
6320 		vdc_set_efi_convert},
6321 
6322 	/* DIOCTL_RWCMD is converted to a read or a write */
6323 	{0, DIOCTL_RWCMD,  sizeof (struct dadkio_rwcmd), NULL},
6324 
6325 	/* mhd(7I) non-shared multihost disks ioctls */
6326 	{0, MHIOCTKOWN,				0, vdc_null_copy_func},
6327 	{0, MHIOCRELEASE,			0, vdc_null_copy_func},
6328 	{0, MHIOCSTATUS,			0, vdc_null_copy_func},
6329 	{0, MHIOCQRESERVE,			0, vdc_null_copy_func},
6330 
6331 	/* mhd(7I) shared multihost disks ioctls */
6332 	{0, MHIOCGRP_INKEYS,			0, vdc_null_copy_func},
6333 	{0, MHIOCGRP_INRESV,			0, vdc_null_copy_func},
6334 	{0, MHIOCGRP_REGISTER,			0, vdc_null_copy_func},
6335 	{0, MHIOCGRP_RESERVE, 			0, vdc_null_copy_func},
6336 	{0, MHIOCGRP_PREEMPTANDABORT,		0, vdc_null_copy_func},
6337 	{0, MHIOCGRP_REGISTERANDIGNOREKEY,	0, vdc_null_copy_func},
6338 
6339 	/* mhd(7I) failfast ioctl */
6340 	{0, MHIOCENFAILFAST,			0, vdc_null_copy_func},
6341 
6342 	/*
6343 	 * These particular ioctls are not sent to the server - vdc fakes up
6344 	 * the necessary info.
6345 	 */
6346 	{0, DKIOCINFO, sizeof (struct dk_cinfo), vdc_null_copy_func},
6347 	{0, DKIOCGMEDIAINFO, sizeof (struct dk_minfo), vdc_null_copy_func},
6348 	{0, USCSICMD,	sizeof (struct uscsi_cmd), vdc_null_copy_func},
6349 	{0, DKIOCGAPART, 0, vdc_null_copy_func },
6350 	{0, DKIOCREMOVABLE, 0, vdc_null_copy_func},
6351 	{0, CDROMREADOFFSET, 0, vdc_null_copy_func}
6352 };
6353 
6354 /*
6355  * This function handles ioctl requests from the vd_efi_alloc_and_read()
6356  * function and forward them to the vdisk.
6357  */
6358 static int
6359 vd_process_efi_ioctl(void *vdisk, int cmd, uintptr_t arg)
6360 {
6361 	vdc_t *vdc = (vdc_t *)vdisk;
6362 	dev_t dev;
6363 	int rval;
6364 
6365 	dev = makedevice(ddi_driver_major(vdc->dip),
6366 	    VD_MAKE_DEV(vdc->instance, 0));
6367 
6368 	return (vd_process_ioctl(dev, cmd, (caddr_t)arg, FKIOCTL, &rval));
6369 }
6370 
6371 /*
6372  * Function:
6373  *	vd_process_ioctl()
6374  *
6375  * Description:
6376  *	This routine processes disk specific ioctl calls
6377  *
6378  * Arguments:
6379  *	dev	- the device number
6380  *	cmd	- the operation [dkio(7I)] to be processed
6381  *	arg	- pointer to user provided structure
6382  *		  (contains data to be set or reference parameter for get)
6383  *	mode	- bit flag, indicating open settings, 32/64 bit type, etc
6384  *	rvalp	- pointer to return value for calling process.
6385  *
6386  * Return Code:
6387  *	0
6388  *	EFAULT
6389  *	ENXIO
6390  *	EIO
6391  *	ENOTSUP
6392  */
6393 static int
6394 vd_process_ioctl(dev_t dev, int cmd, caddr_t arg, int mode, int *rvalp)
6395 {
6396 	int		instance = VDCUNIT(dev);
6397 	vdc_t		*vdc = NULL;
6398 	int		rv = -1;
6399 	int		idx = 0;		/* index into dk_ioctl[] */
6400 	size_t		len = 0;		/* #bytes to send to vds */
6401 	size_t		alloc_len = 0;		/* #bytes to allocate mem for */
6402 	caddr_t		mem_p = NULL;
6403 	size_t		nioctls = (sizeof (dk_ioctl)) / (sizeof (dk_ioctl[0]));
6404 	vdc_dk_ioctl_t	*iop;
6405 
6406 	vdc = ddi_get_soft_state(vdc_state, instance);
6407 	if (vdc == NULL) {
6408 		cmn_err(CE_NOTE, "![%d] Could not get soft state structure",
6409 		    instance);
6410 		return (ENXIO);
6411 	}
6412 
6413 	DMSG(vdc, 0, "[%d] Processing ioctl(%x) for dev %lx : model %x\n",
6414 	    instance, cmd, dev, ddi_model_convert_from(mode & FMODELS));
6415 
6416 	if (rvalp != NULL) {
6417 		/* the return value of the ioctl is 0 by default */
6418 		*rvalp = 0;
6419 	}
6420 
6421 	/*
6422 	 * Validate the ioctl operation to be performed.
6423 	 *
6424 	 * If we have looped through the array without finding a match then we
6425 	 * don't support this ioctl.
6426 	 */
6427 	for (idx = 0; idx < nioctls; idx++) {
6428 		if (cmd == dk_ioctl[idx].cmd)
6429 			break;
6430 	}
6431 
6432 	if (idx >= nioctls) {
6433 		DMSG(vdc, 0, "[%d] Unsupported ioctl (0x%x)\n",
6434 		    vdc->instance, cmd);
6435 		return (ENOTSUP);
6436 	}
6437 
6438 	iop = &(dk_ioctl[idx]);
6439 
6440 	if (cmd == DKIOCGETEFI || cmd == DKIOCSETEFI) {
6441 		/* size is not fixed for EFI ioctls, it depends on ioctl arg */
6442 		dk_efi_t	dk_efi;
6443 
6444 		rv = ddi_copyin(arg, &dk_efi, sizeof (dk_efi_t), mode);
6445 		if (rv != 0)
6446 			return (EFAULT);
6447 
6448 		len = sizeof (vd_efi_t) - 1 + dk_efi.dki_length;
6449 	} else {
6450 		len = iop->nbytes;
6451 	}
6452 
6453 	/* check if the ioctl is applicable */
6454 	switch (cmd) {
6455 	case CDROMREADOFFSET:
6456 	case DKIOCREMOVABLE:
6457 		return (ENOTTY);
6458 
6459 	case USCSICMD:
6460 	case MHIOCTKOWN:
6461 	case MHIOCSTATUS:
6462 	case MHIOCQRESERVE:
6463 	case MHIOCRELEASE:
6464 	case MHIOCGRP_INKEYS:
6465 	case MHIOCGRP_INRESV:
6466 	case MHIOCGRP_REGISTER:
6467 	case MHIOCGRP_RESERVE:
6468 	case MHIOCGRP_PREEMPTANDABORT:
6469 	case MHIOCGRP_REGISTERANDIGNOREKEY:
6470 	case MHIOCENFAILFAST:
6471 		if (vdc->cinfo == NULL)
6472 			return (ENXIO);
6473 		if (vdc->cinfo->dki_ctype != DKC_SCSI_CCS)
6474 			return (ENOTTY);
6475 		break;
6476 
6477 	case DIOCTL_RWCMD:
6478 		if (vdc->cinfo == NULL)
6479 			return (ENXIO);
6480 		if (vdc->cinfo->dki_ctype != DKC_DIRECT)
6481 			return (ENOTTY);
6482 		break;
6483 
6484 	case DKIOCINFO:
6485 		if (vdc->cinfo == NULL)
6486 			return (ENXIO);
6487 		break;
6488 
6489 	case DKIOCGMEDIAINFO:
6490 		if (vdc->minfo == NULL)
6491 			return (ENXIO);
6492 		if (vdc_check_capacity(vdc) != 0)
6493 			/* disk capacity is not available */
6494 			return (EIO);
6495 		break;
6496 	}
6497 
6498 	/*
6499 	 * Deal with ioctls which require a processing different than
6500 	 * converting ioctl arguments and sending a corresponding
6501 	 * VD operation.
6502 	 */
6503 	switch (cmd) {
6504 
6505 	case USCSICMD:
6506 	{
6507 		return (vdc_uscsi_cmd(vdc, arg, mode));
6508 	}
6509 
6510 	case MHIOCTKOWN:
6511 	{
6512 		mutex_enter(&vdc->ownership_lock);
6513 		/*
6514 		 * We have to set VDC_OWNERSHIP_WANTED now so that the ownership
6515 		 * can be flagged with VDC_OWNERSHIP_RESET if the LDC is reset
6516 		 * while we are processing the ioctl.
6517 		 */
6518 		vdc_ownership_update(vdc, VDC_OWNERSHIP_WANTED);
6519 
6520 		rv = vdc_access_set(vdc, VD_ACCESS_SET_EXCLUSIVE |
6521 		    VD_ACCESS_SET_PREEMPT | VD_ACCESS_SET_PRESERVE, mode);
6522 		if (rv == 0) {
6523 			vdc_ownership_update(vdc, VDC_OWNERSHIP_WANTED |
6524 			    VDC_OWNERSHIP_GRANTED);
6525 		} else {
6526 			vdc_ownership_update(vdc, VDC_OWNERSHIP_NONE);
6527 		}
6528 		mutex_exit(&vdc->ownership_lock);
6529 		return (rv);
6530 	}
6531 
6532 	case MHIOCRELEASE:
6533 	{
6534 		mutex_enter(&vdc->ownership_lock);
6535 		rv = vdc_access_set(vdc, VD_ACCESS_SET_CLEAR, mode);
6536 		if (rv == 0) {
6537 			vdc_ownership_update(vdc, VDC_OWNERSHIP_NONE);
6538 		}
6539 		mutex_exit(&vdc->ownership_lock);
6540 		return (rv);
6541 	}
6542 
6543 	case MHIOCSTATUS:
6544 	{
6545 		uint64_t status;
6546 
6547 		rv = vdc_access_get(vdc, &status, mode);
6548 		if (rv == 0 && rvalp != NULL)
6549 			*rvalp = (status & VD_ACCESS_ALLOWED)? 0 : 1;
6550 		return (rv);
6551 	}
6552 
6553 	case MHIOCQRESERVE:
6554 	{
6555 		rv = vdc_access_set(vdc, VD_ACCESS_SET_EXCLUSIVE, mode);
6556 		return (rv);
6557 	}
6558 
6559 	case MHIOCGRP_INKEYS:
6560 	{
6561 		return (vdc_mhd_inkeys(vdc, arg, mode));
6562 	}
6563 
6564 	case MHIOCGRP_INRESV:
6565 	{
6566 		return (vdc_mhd_inresv(vdc, arg, mode));
6567 	}
6568 
6569 	case MHIOCGRP_REGISTER:
6570 	{
6571 		return (vdc_mhd_register(vdc, arg, mode));
6572 	}
6573 
6574 	case MHIOCGRP_RESERVE:
6575 	{
6576 		return (vdc_mhd_reserve(vdc, arg, mode));
6577 	}
6578 
6579 	case MHIOCGRP_PREEMPTANDABORT:
6580 	{
6581 		return (vdc_mhd_preemptabort(vdc, arg, mode));
6582 	}
6583 
6584 	case MHIOCGRP_REGISTERANDIGNOREKEY:
6585 	{
6586 		return (vdc_mhd_registerignore(vdc, arg, mode));
6587 	}
6588 
6589 	case MHIOCENFAILFAST:
6590 	{
6591 		rv = vdc_failfast(vdc, arg, mode);
6592 		return (rv);
6593 	}
6594 
6595 	case DIOCTL_RWCMD:
6596 	{
6597 		return (vdc_dioctl_rwcmd(dev, arg, mode));
6598 	}
6599 
6600 	case DKIOCGAPART:
6601 	{
6602 		return (vdc_dkio_get_partition(vdc, arg, mode));
6603 	}
6604 
6605 	case DKIOCINFO:
6606 	{
6607 		struct dk_cinfo	cinfo;
6608 
6609 		bcopy(vdc->cinfo, &cinfo, sizeof (struct dk_cinfo));
6610 		cinfo.dki_partition = VDCPART(dev);
6611 
6612 		rv = ddi_copyout(&cinfo, (void *)arg,
6613 		    sizeof (struct dk_cinfo), mode);
6614 		if (rv != 0)
6615 			return (EFAULT);
6616 
6617 		return (0);
6618 	}
6619 
6620 	case DKIOCGMEDIAINFO:
6621 	{
6622 		ASSERT(vdc->vdisk_size != 0);
6623 		if (vdc->minfo->dki_capacity == 0)
6624 			vdc->minfo->dki_capacity = vdc->vdisk_size;
6625 		rv = ddi_copyout(vdc->minfo, (void *)arg,
6626 		    sizeof (struct dk_minfo), mode);
6627 		if (rv != 0)
6628 			return (EFAULT);
6629 
6630 		return (0);
6631 	}
6632 
6633 	case DKIOCFLUSHWRITECACHE:
6634 		{
6635 			struct dk_callback *dkc =
6636 			    (struct dk_callback *)(uintptr_t)arg;
6637 			vdc_dk_arg_t	*dkarg = NULL;
6638 
6639 			DMSG(vdc, 1, "[%d] Flush W$: mode %x\n",
6640 			    instance, mode);
6641 
6642 			/*
6643 			 * If arg is NULL, then there is no callback function
6644 			 * registered and the call operates synchronously; we
6645 			 * break and continue with the rest of the function and
6646 			 * wait for vds to return (i.e. after the request to
6647 			 * vds returns successfully, all writes completed prior
6648 			 * to the ioctl will have been flushed from the disk
6649 			 * write cache to persistent media.
6650 			 *
6651 			 * If a callback function is registered, we dispatch
6652 			 * the request on a task queue and return immediately.
6653 			 * The callback will deal with informing the calling
6654 			 * thread that the flush request is completed.
6655 			 */
6656 			if (dkc == NULL)
6657 				break;
6658 
6659 			/*
6660 			 * the asynchronous callback is only supported if
6661 			 * invoked from within the kernel
6662 			 */
6663 			if ((mode & FKIOCTL) == 0)
6664 				return (ENOTSUP);
6665 
6666 			dkarg = kmem_zalloc(sizeof (vdc_dk_arg_t), KM_SLEEP);
6667 
6668 			dkarg->mode = mode;
6669 			dkarg->dev = dev;
6670 			bcopy(dkc, &dkarg->dkc, sizeof (*dkc));
6671 
6672 			mutex_enter(&vdc->lock);
6673 			vdc->dkio_flush_pending++;
6674 			dkarg->vdc = vdc;
6675 			mutex_exit(&vdc->lock);
6676 
6677 			/* put the request on a task queue */
6678 			rv = taskq_dispatch(system_taskq, vdc_dkio_flush_cb,
6679 			    (void *)dkarg, DDI_SLEEP);
6680 			if (rv == NULL) {
6681 				/* clean up if dispatch fails */
6682 				mutex_enter(&vdc->lock);
6683 				vdc->dkio_flush_pending--;
6684 				mutex_exit(&vdc->lock);
6685 				kmem_free(dkarg, sizeof (vdc_dk_arg_t));
6686 			}
6687 
6688 			return (rv == NULL ? ENOMEM : 0);
6689 		}
6690 	}
6691 
6692 	/* catch programming error in vdc - should be a VD_OP_XXX ioctl */
6693 	ASSERT(iop->op != 0);
6694 
6695 	/* check if the vDisk server handles the operation for this vDisk */
6696 	if (VD_OP_SUPPORTED(vdc->operations, iop->op) == B_FALSE) {
6697 		DMSG(vdc, 0, "[%d] Unsupported VD_OP operation (0x%x)\n",
6698 		    vdc->instance, iop->op);
6699 		return (ENOTSUP);
6700 	}
6701 
6702 	/* LDC requires that the memory being mapped is 8-byte aligned */
6703 	alloc_len = P2ROUNDUP(len, sizeof (uint64_t));
6704 	DMSG(vdc, 1, "[%d] struct size %ld alloc %ld\n",
6705 	    instance, len, alloc_len);
6706 
6707 	if (alloc_len > 0)
6708 		mem_p = kmem_zalloc(alloc_len, KM_SLEEP);
6709 
6710 	/*
6711 	 * Call the conversion function for this ioctl which, if necessary,
6712 	 * converts from the Solaris format to the format ARC'ed
6713 	 * as part of the vDisk protocol (FWARC 2006/195)
6714 	 */
6715 	ASSERT(iop->convert != NULL);
6716 	rv = (iop->convert)(vdc, arg, mem_p, mode, VD_COPYIN);
6717 	if (rv != 0) {
6718 		DMSG(vdc, 0, "[%d] convert func returned %d for ioctl 0x%x\n",
6719 		    instance, rv, cmd);
6720 		if (mem_p != NULL)
6721 			kmem_free(mem_p, alloc_len);
6722 		return (rv);
6723 	}
6724 
6725 	/*
6726 	 * send request to vds to service the ioctl.
6727 	 */
6728 	rv = vdc_do_sync_op(vdc, iop->op, mem_p, alloc_len,
6729 	    VDCPART(dev), 0, CB_SYNC, (void *)(uint64_t)mode,
6730 	    VIO_both_dir, B_TRUE);
6731 
6732 	if (rv != 0) {
6733 		/*
6734 		 * This is not necessarily an error. The ioctl could
6735 		 * be returning a value such as ENOTTY to indicate
6736 		 * that the ioctl is not applicable.
6737 		 */
6738 		DMSG(vdc, 0, "[%d] vds returned %d for ioctl 0x%x\n",
6739 		    instance, rv, cmd);
6740 		if (mem_p != NULL)
6741 			kmem_free(mem_p, alloc_len);
6742 
6743 		return (rv);
6744 	}
6745 
6746 	/*
6747 	 * Call the conversion function (if it exists) for this ioctl
6748 	 * which converts from the format ARC'ed as part of the vDisk
6749 	 * protocol (FWARC 2006/195) back to a format understood by
6750 	 * the rest of Solaris.
6751 	 */
6752 	rv = (iop->convert)(vdc, mem_p, arg, mode, VD_COPYOUT);
6753 	if (rv != 0) {
6754 		DMSG(vdc, 0, "[%d] convert func returned %d for ioctl 0x%x\n",
6755 		    instance, rv, cmd);
6756 		if (mem_p != NULL)
6757 			kmem_free(mem_p, alloc_len);
6758 		return (rv);
6759 	}
6760 
6761 	if (mem_p != NULL)
6762 		kmem_free(mem_p, alloc_len);
6763 
6764 	return (rv);
6765 }
6766 
6767 /*
6768  * Function:
6769  *
6770  * Description:
6771  *	This is an empty conversion function used by ioctl calls which
6772  *	do not need to convert the data being passed in/out to userland
6773  */
6774 static int
6775 vdc_null_copy_func(vdc_t *vdc, void *from, void *to, int mode, int dir)
6776 {
6777 	_NOTE(ARGUNUSED(vdc))
6778 	_NOTE(ARGUNUSED(from))
6779 	_NOTE(ARGUNUSED(to))
6780 	_NOTE(ARGUNUSED(mode))
6781 	_NOTE(ARGUNUSED(dir))
6782 
6783 	return (0);
6784 }
6785 
6786 static int
6787 vdc_get_wce_convert(vdc_t *vdc, void *from, void *to,
6788     int mode, int dir)
6789 {
6790 	_NOTE(ARGUNUSED(vdc))
6791 
6792 	if (dir == VD_COPYIN)
6793 		return (0);		/* nothing to do */
6794 
6795 	if (ddi_copyout(from, to, sizeof (int), mode) != 0)
6796 		return (EFAULT);
6797 
6798 	return (0);
6799 }
6800 
6801 static int
6802 vdc_set_wce_convert(vdc_t *vdc, void *from, void *to,
6803     int mode, int dir)
6804 {
6805 	_NOTE(ARGUNUSED(vdc))
6806 
6807 	if (dir == VD_COPYOUT)
6808 		return (0);		/* nothing to do */
6809 
6810 	if (ddi_copyin(from, to, sizeof (int), mode) != 0)
6811 		return (EFAULT);
6812 
6813 	return (0);
6814 }
6815 
6816 /*
6817  * Function:
6818  *	vdc_get_vtoc_convert()
6819  *
6820  * Description:
6821  *	This routine performs the necessary convertions from the DKIOCGVTOC
6822  *	Solaris structure to the format defined in FWARC 2006/195.
6823  *
6824  *	In the struct vtoc definition, the timestamp field is marked as not
6825  *	supported so it is not part of vDisk protocol (FWARC 2006/195).
6826  *	However SVM uses that field to check it can write into the VTOC,
6827  *	so we fake up the info of that field.
6828  *
6829  * Arguments:
6830  *	vdc	- the vDisk client
6831  *	from	- the buffer containing the data to be copied from
6832  *	to	- the buffer to be copied to
6833  *	mode	- flags passed to ioctl() call
6834  *	dir	- the "direction" of the copy - VD_COPYIN or VD_COPYOUT
6835  *
6836  * Return Code:
6837  *	0	- Success
6838  *	ENXIO	- incorrect buffer passed in.
6839  *	EFAULT	- ddi_copyout routine encountered an error.
6840  */
6841 static int
6842 vdc_get_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
6843 {
6844 	int		i;
6845 	void		*tmp_mem = NULL;
6846 	void		*tmp_memp;
6847 	struct vtoc	vt;
6848 	struct vtoc32	vt32;
6849 	int		copy_len = 0;
6850 	int		rv = 0;
6851 
6852 	if (dir != VD_COPYOUT)
6853 		return (0);	/* nothing to do */
6854 
6855 	if ((from == NULL) || (to == NULL))
6856 		return (ENXIO);
6857 
6858 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32)
6859 		copy_len = sizeof (struct vtoc32);
6860 	else
6861 		copy_len = sizeof (struct vtoc);
6862 
6863 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
6864 
6865 	VD_VTOC2VTOC((vd_vtoc_t *)from, &vt);
6866 
6867 	/* fake the VTOC timestamp field */
6868 	for (i = 0; i < V_NUMPAR; i++) {
6869 		vt.timestamp[i] = vdc->vtoc->timestamp[i];
6870 	}
6871 
6872 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
6873 		/* LINTED E_ASSIGN_NARROW_CONV */
6874 		vtoctovtoc32(vt, vt32);
6875 		tmp_memp = &vt32;
6876 	} else {
6877 		tmp_memp = &vt;
6878 	}
6879 	rv = ddi_copyout(tmp_memp, to, copy_len, mode);
6880 	if (rv != 0)
6881 		rv = EFAULT;
6882 
6883 	kmem_free(tmp_mem, copy_len);
6884 	return (rv);
6885 }
6886 
6887 /*
6888  * Function:
6889  *	vdc_set_vtoc_convert()
6890  *
6891  * Description:
6892  *	This routine performs the necessary convertions from the DKIOCSVTOC
6893  *	Solaris structure to the format defined in FWARC 2006/195.
6894  *
6895  * Arguments:
6896  *	vdc	- the vDisk client
6897  *	from	- Buffer with data
6898  *	to	- Buffer where data is to be copied to
6899  *	mode	- flags passed to ioctl
6900  *	dir	- direction of copy (in or out)
6901  *
6902  * Return Code:
6903  *	0	- Success
6904  *	ENXIO	- Invalid buffer passed in
6905  *	EFAULT	- ddi_copyin of data failed
6906  */
6907 static int
6908 vdc_set_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
6909 {
6910 	_NOTE(ARGUNUSED(vdc))
6911 
6912 	void		*tmp_mem = NULL, *uvtoc;
6913 	struct vtoc	vt;
6914 	struct vtoc	*vtp = &vt;
6915 	vd_vtoc_t	vtvd;
6916 	int		copy_len = 0;
6917 	int		i, rv = 0;
6918 
6919 	if ((from == NULL) || (to == NULL))
6920 		return (ENXIO);
6921 
6922 	if (dir == VD_COPYIN)
6923 		uvtoc = from;
6924 	else
6925 		uvtoc = to;
6926 
6927 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32)
6928 		copy_len = sizeof (struct vtoc32);
6929 	else
6930 		copy_len = sizeof (struct vtoc);
6931 
6932 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
6933 
6934 	rv = ddi_copyin(uvtoc, tmp_mem, copy_len, mode);
6935 	if (rv != 0) {
6936 		kmem_free(tmp_mem, copy_len);
6937 		return (EFAULT);
6938 	}
6939 
6940 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
6941 		vtoc32tovtoc((*(struct vtoc32 *)tmp_mem), vt);
6942 	} else {
6943 		vtp = tmp_mem;
6944 	}
6945 
6946 	if (dir == VD_COPYOUT) {
6947 		/*
6948 		 * The disk label may have changed. Revalidate the disk
6949 		 * geometry. This will also update the device nodes and
6950 		 * properties.
6951 		 */
6952 		vdc_validate(vdc);
6953 
6954 		/*
6955 		 * We also need to keep track of the timestamp fields.
6956 		 */
6957 		for (i = 0; i < V_NUMPAR; i++) {
6958 			vdc->vtoc->timestamp[i] = vtp->timestamp[i];
6959 		}
6960 
6961 		return (0);
6962 	}
6963 
6964 	VTOC2VD_VTOC(vtp, &vtvd);
6965 	bcopy(&vtvd, to, sizeof (vd_vtoc_t));
6966 	kmem_free(tmp_mem, copy_len);
6967 
6968 	return (0);
6969 }
6970 
6971 /*
6972  * Function:
6973  *	vdc_get_geom_convert()
6974  *
6975  * Description:
6976  *	This routine performs the necessary convertions from the DKIOCGGEOM,
6977  *	DKIOCG_PHYSGEOM and DKIOG_VIRTGEOM Solaris structures to the format
6978  *	defined in FWARC 2006/195
6979  *
6980  * Arguments:
6981  *	vdc	- the vDisk client
6982  *	from	- Buffer with data
6983  *	to	- Buffer where data is to be copied to
6984  *	mode	- flags passed to ioctl
6985  *	dir	- direction of copy (in or out)
6986  *
6987  * Return Code:
6988  *	0	- Success
6989  *	ENXIO	- Invalid buffer passed in
6990  *	EFAULT	- ddi_copyout of data failed
6991  */
6992 static int
6993 vdc_get_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
6994 {
6995 	_NOTE(ARGUNUSED(vdc))
6996 
6997 	struct dk_geom	geom;
6998 	int	copy_len = sizeof (struct dk_geom);
6999 	int	rv = 0;
7000 
7001 	if (dir != VD_COPYOUT)
7002 		return (0);	/* nothing to do */
7003 
7004 	if ((from == NULL) || (to == NULL))
7005 		return (ENXIO);
7006 
7007 	VD_GEOM2DK_GEOM((vd_geom_t *)from, &geom);
7008 	rv = ddi_copyout(&geom, to, copy_len, mode);
7009 	if (rv != 0)
7010 		rv = EFAULT;
7011 
7012 	return (rv);
7013 }
7014 
7015 /*
7016  * Function:
7017  *	vdc_set_geom_convert()
7018  *
7019  * Description:
7020  *	This routine performs the necessary convertions from the DKIOCSGEOM
7021  *	Solaris structure to the format defined in FWARC 2006/195.
7022  *
7023  * Arguments:
7024  *	vdc	- the vDisk client
7025  *	from	- Buffer with data
7026  *	to	- Buffer where data is to be copied to
7027  *	mode	- flags passed to ioctl
7028  *	dir	- direction of copy (in or out)
7029  *
7030  * Return Code:
7031  *	0	- Success
7032  *	ENXIO	- Invalid buffer passed in
7033  *	EFAULT	- ddi_copyin of data failed
7034  */
7035 static int
7036 vdc_set_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7037 {
7038 	_NOTE(ARGUNUSED(vdc))
7039 
7040 	vd_geom_t	vdgeom;
7041 	void		*tmp_mem = NULL;
7042 	int		copy_len = sizeof (struct dk_geom);
7043 	int		rv = 0;
7044 
7045 	if (dir != VD_COPYIN)
7046 		return (0);	/* nothing to do */
7047 
7048 	if ((from == NULL) || (to == NULL))
7049 		return (ENXIO);
7050 
7051 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
7052 
7053 	rv = ddi_copyin(from, tmp_mem, copy_len, mode);
7054 	if (rv != 0) {
7055 		kmem_free(tmp_mem, copy_len);
7056 		return (EFAULT);
7057 	}
7058 	DK_GEOM2VD_GEOM((struct dk_geom *)tmp_mem, &vdgeom);
7059 	bcopy(&vdgeom, to, sizeof (vdgeom));
7060 	kmem_free(tmp_mem, copy_len);
7061 
7062 	return (0);
7063 }
7064 
7065 static int
7066 vdc_get_efi_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7067 {
7068 	_NOTE(ARGUNUSED(vdc))
7069 
7070 	vd_efi_t	*vd_efi;
7071 	dk_efi_t	dk_efi;
7072 	int		rv = 0;
7073 	void		*uaddr;
7074 
7075 	if ((from == NULL) || (to == NULL))
7076 		return (ENXIO);
7077 
7078 	if (dir == VD_COPYIN) {
7079 
7080 		vd_efi = (vd_efi_t *)to;
7081 
7082 		rv = ddi_copyin(from, &dk_efi, sizeof (dk_efi_t), mode);
7083 		if (rv != 0)
7084 			return (EFAULT);
7085 
7086 		vd_efi->lba = dk_efi.dki_lba;
7087 		vd_efi->length = dk_efi.dki_length;
7088 		bzero(vd_efi->data, vd_efi->length);
7089 
7090 	} else {
7091 
7092 		rv = ddi_copyin(to, &dk_efi, sizeof (dk_efi_t), mode);
7093 		if (rv != 0)
7094 			return (EFAULT);
7095 
7096 		uaddr = dk_efi.dki_data;
7097 
7098 		dk_efi.dki_data = kmem_alloc(dk_efi.dki_length, KM_SLEEP);
7099 
7100 		VD_EFI2DK_EFI((vd_efi_t *)from, &dk_efi);
7101 
7102 		rv = ddi_copyout(dk_efi.dki_data, uaddr, dk_efi.dki_length,
7103 		    mode);
7104 		if (rv != 0)
7105 			return (EFAULT);
7106 
7107 		kmem_free(dk_efi.dki_data, dk_efi.dki_length);
7108 	}
7109 
7110 	return (0);
7111 }
7112 
7113 static int
7114 vdc_set_efi_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7115 {
7116 	_NOTE(ARGUNUSED(vdc))
7117 
7118 	dk_efi_t	dk_efi;
7119 	void		*uaddr;
7120 
7121 	if (dir == VD_COPYOUT) {
7122 		/*
7123 		 * The disk label may have changed. Revalidate the disk
7124 		 * geometry. This will also update the device nodes and
7125 		 * properties.
7126 		 */
7127 		vdc_validate(vdc);
7128 		return (0);
7129 	}
7130 
7131 	if ((from == NULL) || (to == NULL))
7132 		return (ENXIO);
7133 
7134 	if (ddi_copyin(from, &dk_efi, sizeof (dk_efi_t), mode) != 0)
7135 		return (EFAULT);
7136 
7137 	uaddr = dk_efi.dki_data;
7138 
7139 	dk_efi.dki_data = kmem_alloc(dk_efi.dki_length, KM_SLEEP);
7140 
7141 	if (ddi_copyin(uaddr, dk_efi.dki_data, dk_efi.dki_length, mode) != 0)
7142 		return (EFAULT);
7143 
7144 	DK_EFI2VD_EFI(&dk_efi, (vd_efi_t *)to);
7145 
7146 	kmem_free(dk_efi.dki_data, dk_efi.dki_length);
7147 
7148 	return (0);
7149 }
7150 
7151 
7152 /* -------------------------------------------------------------------------- */
7153 
7154 /*
7155  * Function:
7156  *	vdc_create_fake_geometry()
7157  *
7158  * Description:
7159  *	This routine fakes up the disk info needed for some DKIO ioctls such
7160  *	as DKIOCINFO and DKIOCGMEDIAINFO [just like lofi(7D) and ramdisk(7D) do]
7161  *
7162  *	Note: This function must not be called until the vDisk attributes have
7163  *	been exchanged as part of the handshake with the vDisk server.
7164  *
7165  * Arguments:
7166  *	vdc	- soft state pointer for this instance of the device driver.
7167  *
7168  * Return Code:
7169  *	none.
7170  */
7171 static void
7172 vdc_create_fake_geometry(vdc_t *vdc)
7173 {
7174 	ASSERT(vdc != NULL);
7175 	ASSERT(vdc->max_xfer_sz != 0);
7176 
7177 	/*
7178 	 * DKIOCINFO support
7179 	 */
7180 	if (vdc->cinfo == NULL)
7181 		vdc->cinfo = kmem_zalloc(sizeof (struct dk_cinfo), KM_SLEEP);
7182 
7183 	(void) strcpy(vdc->cinfo->dki_cname, VDC_DRIVER_NAME);
7184 	(void) strcpy(vdc->cinfo->dki_dname, VDC_DRIVER_NAME);
7185 	/* max_xfer_sz is #blocks so we don't need to divide by DEV_BSIZE */
7186 	vdc->cinfo->dki_maxtransfer = vdc->max_xfer_sz;
7187 
7188 	/*
7189 	 * We set the controller type to DKC_SCSI_CCS only if the VD_OP_SCSICMD
7190 	 * operation is supported, otherwise the controller type is DKC_DIRECT.
7191 	 * Version 1.0 does not support the VD_OP_SCSICMD operation, so the
7192 	 * controller type is always DKC_DIRECT in that case.
7193 	 *
7194 	 * If the virtual disk is backed by a physical CD/DVD device or
7195 	 * an ISO image, modify the controller type to indicate this
7196 	 */
7197 	switch (vdc->vdisk_media) {
7198 	case VD_MEDIA_CD:
7199 	case VD_MEDIA_DVD:
7200 		vdc->cinfo->dki_ctype = DKC_CDROM;
7201 		break;
7202 	case VD_MEDIA_FIXED:
7203 		if (VD_OP_SUPPORTED(vdc->operations, VD_OP_SCSICMD))
7204 			vdc->cinfo->dki_ctype = DKC_SCSI_CCS;
7205 		else
7206 			vdc->cinfo->dki_ctype = DKC_DIRECT;
7207 		break;
7208 	default:
7209 		/* in the case of v1.0 we default to a fixed disk */
7210 		vdc->cinfo->dki_ctype = DKC_DIRECT;
7211 		break;
7212 	}
7213 	vdc->cinfo->dki_flags = DKI_FMTVOL;
7214 	vdc->cinfo->dki_cnum = 0;
7215 	vdc->cinfo->dki_addr = 0;
7216 	vdc->cinfo->dki_space = 0;
7217 	vdc->cinfo->dki_prio = 0;
7218 	vdc->cinfo->dki_vec = 0;
7219 	vdc->cinfo->dki_unit = vdc->instance;
7220 	vdc->cinfo->dki_slave = 0;
7221 	/*
7222 	 * The partition number will be created on the fly depending on the
7223 	 * actual slice (i.e. minor node) that is used to request the data.
7224 	 */
7225 	vdc->cinfo->dki_partition = 0;
7226 
7227 	/*
7228 	 * DKIOCGMEDIAINFO support
7229 	 */
7230 	if (vdc->minfo == NULL)
7231 		vdc->minfo = kmem_zalloc(sizeof (struct dk_minfo), KM_SLEEP);
7232 
7233 	if (vio_ver_is_supported(vdc->ver, 1, 1)) {
7234 		vdc->minfo->dki_media_type =
7235 		    VD_MEDIATYPE2DK_MEDIATYPE(vdc->vdisk_media);
7236 	} else {
7237 		vdc->minfo->dki_media_type = DK_FIXED_DISK;
7238 	}
7239 
7240 	vdc->minfo->dki_capacity = vdc->vdisk_size;
7241 	vdc->minfo->dki_lbsize = vdc->block_size;
7242 }
7243 
7244 static ushort_t
7245 vdc_lbl2cksum(struct dk_label *label)
7246 {
7247 	int	count;
7248 	ushort_t sum, *sp;
7249 
7250 	count =	(sizeof (struct dk_label)) / (sizeof (short)) - 1;
7251 	sp = (ushort_t *)label;
7252 	sum = 0;
7253 	while (count--) {
7254 		sum ^= *sp++;
7255 	}
7256 
7257 	return (sum);
7258 }
7259 
7260 /*
7261  * Function:
7262  *	vdc_validate_geometry
7263  *
7264  * Description:
7265  *	This routine discovers the label and geometry of the disk. It stores
7266  *	the disk label and related information in the vdc structure. If it
7267  *	fails to validate the geometry or to discover the disk label then
7268  *	the label is marked as unknown (VD_DISK_LABEL_UNK).
7269  *
7270  * Arguments:
7271  *	vdc	- soft state pointer for this instance of the device driver.
7272  *
7273  * Return Code:
7274  *	0	- success.
7275  *	EINVAL	- unknown disk label.
7276  *	ENOTSUP	- geometry not applicable (EFI label).
7277  *	EIO	- error accessing the disk.
7278  */
7279 static int
7280 vdc_validate_geometry(vdc_t *vdc)
7281 {
7282 	buf_t	*buf;	/* BREAD requests need to be in a buf_t structure */
7283 	dev_t	dev;
7284 	int	rv, rval;
7285 	struct dk_label label;
7286 	struct dk_geom geom;
7287 	struct vtoc vtoc;
7288 	efi_gpt_t *gpt;
7289 	efi_gpe_t *gpe;
7290 	vd_efi_dev_t edev;
7291 
7292 	ASSERT(vdc != NULL);
7293 	ASSERT(vdc->vtoc != NULL && vdc->geom != NULL);
7294 	ASSERT(MUTEX_HELD(&vdc->lock));
7295 
7296 	mutex_exit(&vdc->lock);
7297 
7298 	dev = makedevice(ddi_driver_major(vdc->dip),
7299 	    VD_MAKE_DEV(vdc->instance, 0));
7300 
7301 	rv = vd_process_ioctl(dev, DKIOCGGEOM, (caddr_t)&geom, FKIOCTL, &rval);
7302 	if (rv == 0)
7303 		rv = vd_process_ioctl(dev, DKIOCGVTOC, (caddr_t)&vtoc,
7304 		    FKIOCTL, &rval);
7305 
7306 	if (rv == ENOTSUP) {
7307 		/*
7308 		 * If the device does not support VTOC then we try
7309 		 * to read an EFI label.
7310 		 *
7311 		 * We need to know the block size and the disk size to
7312 		 * be able to read an EFI label.
7313 		 */
7314 		if (vdc->vdisk_size == 0) {
7315 			if ((rv = vdc_check_capacity(vdc)) != 0) {
7316 				mutex_enter(&vdc->lock);
7317 				vdc_store_label_unk(vdc);
7318 				return (rv);
7319 			}
7320 		}
7321 
7322 		VD_EFI_DEV_SET(edev, vdc, vd_process_efi_ioctl);
7323 
7324 		rv = vd_efi_alloc_and_read(&edev, &gpt, &gpe);
7325 
7326 		if (rv) {
7327 			DMSG(vdc, 0, "[%d] Failed to get EFI (err=%d)",
7328 			    vdc->instance, rv);
7329 			mutex_enter(&vdc->lock);
7330 			vdc_store_label_unk(vdc);
7331 			return (EIO);
7332 		}
7333 
7334 		mutex_enter(&vdc->lock);
7335 		vdc_store_label_efi(vdc, gpt, gpe);
7336 		vd_efi_free(&edev, gpt, gpe);
7337 		return (ENOTSUP);
7338 	}
7339 
7340 	if (rv != 0) {
7341 		DMSG(vdc, 0, "[%d] Failed to get VTOC (err=%d)",
7342 		    vdc->instance, rv);
7343 		mutex_enter(&vdc->lock);
7344 		vdc_store_label_unk(vdc);
7345 		if (rv != EINVAL)
7346 			rv = EIO;
7347 		return (rv);
7348 	}
7349 
7350 	/* check that geometry and vtoc are valid */
7351 	if (geom.dkg_nhead == 0 || geom.dkg_nsect == 0 ||
7352 	    vtoc.v_sanity != VTOC_SANE) {
7353 		mutex_enter(&vdc->lock);
7354 		vdc_store_label_unk(vdc);
7355 		return (EINVAL);
7356 	}
7357 
7358 	/*
7359 	 * We have a disk and a valid VTOC. However this does not mean
7360 	 * that the disk currently have a VTOC label. The returned VTOC may
7361 	 * be a default VTOC to be used for configuring the disk (this is
7362 	 * what is done for disk image). So we read the label from the
7363 	 * beginning of the disk to ensure we really have a VTOC label.
7364 	 *
7365 	 * FUTURE: This could be the default way for reading the VTOC
7366 	 * from the disk as opposed to sending the VD_OP_GET_VTOC
7367 	 * to the server. This will be the default if vdc is implemented
7368 	 * ontop of cmlb.
7369 	 */
7370 
7371 	/*
7372 	 * Single slice disk does not support read using an absolute disk
7373 	 * offset so we just rely on the DKIOCGVTOC ioctl in that case.
7374 	 */
7375 	if (vdc->vdisk_type == VD_DISK_TYPE_SLICE) {
7376 		mutex_enter(&vdc->lock);
7377 		if (vtoc.v_nparts != 1) {
7378 			vdc_store_label_unk(vdc);
7379 			return (EINVAL);
7380 		}
7381 		vdc_store_label_vtoc(vdc, &geom, &vtoc);
7382 		return (0);
7383 	}
7384 
7385 	if (vtoc.v_nparts != V_NUMPAR) {
7386 		mutex_enter(&vdc->lock);
7387 		vdc_store_label_unk(vdc);
7388 		return (EINVAL);
7389 	}
7390 
7391 	/*
7392 	 * Read disk label from start of disk
7393 	 */
7394 	buf = kmem_alloc(sizeof (buf_t), KM_SLEEP);
7395 	bioinit(buf);
7396 	buf->b_un.b_addr = (caddr_t)&label;
7397 	buf->b_bcount = DK_LABEL_SIZE;
7398 	buf->b_flags = B_BUSY | B_READ;
7399 	buf->b_dev = cmpdev(dev);
7400 	rv = vdc_send_request(vdc, VD_OP_BREAD, (caddr_t)&label,
7401 	    DK_LABEL_SIZE, VD_SLICE_NONE, 0, CB_STRATEGY, buf, VIO_read_dir);
7402 	if (rv) {
7403 		DMSG(vdc, 1, "[%d] Failed to read disk block 0\n",
7404 		    vdc->instance);
7405 	} else {
7406 		rv = biowait(buf);
7407 		biofini(buf);
7408 	}
7409 	kmem_free(buf, sizeof (buf_t));
7410 
7411 	if (rv != 0 || label.dkl_magic != DKL_MAGIC ||
7412 	    label.dkl_cksum != vdc_lbl2cksum(&label)) {
7413 		DMSG(vdc, 1, "[%d] Got VTOC with invalid label\n",
7414 		    vdc->instance);
7415 		mutex_enter(&vdc->lock);
7416 		vdc_store_label_unk(vdc);
7417 		return (EINVAL);
7418 	}
7419 
7420 	mutex_enter(&vdc->lock);
7421 	vdc_store_label_vtoc(vdc, &geom, &vtoc);
7422 	return (0);
7423 }
7424 
7425 /*
7426  * Function:
7427  *	vdc_validate
7428  *
7429  * Description:
7430  *	This routine discovers the label of the disk and create the
7431  *	appropriate device nodes if the label has changed.
7432  *
7433  * Arguments:
7434  *	vdc	- soft state pointer for this instance of the device driver.
7435  *
7436  * Return Code:
7437  *	none.
7438  */
7439 static void
7440 vdc_validate(vdc_t *vdc)
7441 {
7442 	vd_disk_label_t old_label;
7443 	vd_slice_t old_slice[V_NUMPAR];
7444 	int rv;
7445 
7446 	ASSERT(!MUTEX_HELD(&vdc->lock));
7447 
7448 	mutex_enter(&vdc->lock);
7449 
7450 	/* save the current label and vtoc */
7451 	old_label = vdc->vdisk_label;
7452 	bcopy(vdc->slice, &old_slice, sizeof (vd_slice_t) * V_NUMPAR);
7453 
7454 	/* check the geometry */
7455 	(void) vdc_validate_geometry(vdc);
7456 
7457 	/* if the disk label has changed, update device nodes */
7458 	if (vdc->vdisk_label != old_label) {
7459 
7460 		if (vdc->vdisk_label == VD_DISK_LABEL_EFI)
7461 			rv = vdc_create_device_nodes_efi(vdc);
7462 		else
7463 			rv = vdc_create_device_nodes_vtoc(vdc);
7464 
7465 		if (rv != 0) {
7466 			DMSG(vdc, 0, "![%d] Failed to update device nodes",
7467 			    vdc->instance);
7468 		}
7469 	}
7470 
7471 	/* if the vtoc has changed, update device nodes properties */
7472 	if (bcmp(vdc->slice, &old_slice, sizeof (vd_slice_t) * V_NUMPAR) != 0) {
7473 
7474 		if (vdc_create_device_nodes_props(vdc) != 0) {
7475 			DMSG(vdc, 0, "![%d] Failed to update device nodes"
7476 			    " properties", vdc->instance);
7477 		}
7478 	}
7479 
7480 	mutex_exit(&vdc->lock);
7481 }
7482 
7483 static void
7484 vdc_validate_task(void *arg)
7485 {
7486 	vdc_t *vdc = (vdc_t *)arg;
7487 
7488 	vdc_validate(vdc);
7489 
7490 	mutex_enter(&vdc->lock);
7491 	ASSERT(vdc->validate_pending > 0);
7492 	vdc->validate_pending--;
7493 	mutex_exit(&vdc->lock);
7494 }
7495 
7496 /*
7497  * Function:
7498  *	vdc_setup_devid()
7499  *
7500  * Description:
7501  *	This routine discovers the devid of a vDisk. It requests the devid of
7502  *	the underlying device from the vDisk server, builds an encapsulated
7503  *	devid based on the retrieved devid and registers that new devid to
7504  *	the vDisk.
7505  *
7506  * Arguments:
7507  *	vdc	- soft state pointer for this instance of the device driver.
7508  *
7509  * Return Code:
7510  *	0	- A devid was succesfully registered for the vDisk
7511  */
7512 static int
7513 vdc_setup_devid(vdc_t *vdc)
7514 {
7515 	int rv;
7516 	vd_devid_t *vd_devid;
7517 	size_t bufsize, bufid_len;
7518 
7519 	/*
7520 	 * At first sight, we don't know the size of the devid that the
7521 	 * server will return but this size will be encoded into the
7522 	 * reply. So we do a first request using a default size then we
7523 	 * check if this size was large enough. If not then we do a second
7524 	 * request with the correct size returned by the server. Note that
7525 	 * ldc requires size to be 8-byte aligned.
7526 	 */
7527 	bufsize = P2ROUNDUP(VD_DEVID_SIZE(VD_DEVID_DEFAULT_LEN),
7528 	    sizeof (uint64_t));
7529 	vd_devid = kmem_zalloc(bufsize, KM_SLEEP);
7530 	bufid_len = bufsize - sizeof (vd_efi_t) - 1;
7531 
7532 	rv = vdc_do_sync_op(vdc, VD_OP_GET_DEVID, (caddr_t)vd_devid,
7533 	    bufsize, 0, 0, CB_SYNC, 0, VIO_both_dir, B_TRUE);
7534 
7535 	DMSG(vdc, 2, "sync_op returned %d\n", rv);
7536 
7537 	if (rv) {
7538 		kmem_free(vd_devid, bufsize);
7539 		return (rv);
7540 	}
7541 
7542 	if (vd_devid->length > bufid_len) {
7543 		/*
7544 		 * The returned devid is larger than the buffer used. Try again
7545 		 * with a buffer with the right size.
7546 		 */
7547 		kmem_free(vd_devid, bufsize);
7548 		bufsize = P2ROUNDUP(VD_DEVID_SIZE(vd_devid->length),
7549 		    sizeof (uint64_t));
7550 		vd_devid = kmem_zalloc(bufsize, KM_SLEEP);
7551 		bufid_len = bufsize - sizeof (vd_efi_t) - 1;
7552 
7553 		rv = vdc_do_sync_op(vdc, VD_OP_GET_DEVID,
7554 		    (caddr_t)vd_devid, bufsize, 0, 0, CB_SYNC, 0,
7555 		    VIO_both_dir, B_TRUE);
7556 
7557 		if (rv) {
7558 			kmem_free(vd_devid, bufsize);
7559 			return (rv);
7560 		}
7561 	}
7562 
7563 	/*
7564 	 * The virtual disk should have the same device id as the one associated
7565 	 * with the physical disk it is mapped on, otherwise sharing a disk
7566 	 * between a LDom and a non-LDom may not work (for example for a shared
7567 	 * SVM disk set).
7568 	 *
7569 	 * The DDI framework does not allow creating a device id with any
7570 	 * type so we first create a device id of type DEVID_ENCAP and then
7571 	 * we restore the orignal type of the physical device.
7572 	 */
7573 
7574 	DMSG(vdc, 2, ": devid length = %d\n", vd_devid->length);
7575 
7576 	/* build an encapsulated devid based on the returned devid */
7577 	if (ddi_devid_init(vdc->dip, DEVID_ENCAP, vd_devid->length,
7578 	    vd_devid->id, &vdc->devid) != DDI_SUCCESS) {
7579 		DMSG(vdc, 1, "[%d] Fail to created devid\n", vdc->instance);
7580 		kmem_free(vd_devid, bufsize);
7581 		return (1);
7582 	}
7583 
7584 	DEVID_FORMTYPE((impl_devid_t *)vdc->devid, vd_devid->type);
7585 
7586 	ASSERT(ddi_devid_valid(vdc->devid) == DDI_SUCCESS);
7587 
7588 	kmem_free(vd_devid, bufsize);
7589 
7590 	if (ddi_devid_register(vdc->dip, vdc->devid) != DDI_SUCCESS) {
7591 		DMSG(vdc, 1, "[%d] Fail to register devid\n", vdc->instance);
7592 		return (1);
7593 	}
7594 
7595 	return (0);
7596 }
7597 
7598 static void
7599 vdc_store_label_efi(vdc_t *vdc, efi_gpt_t *gpt, efi_gpe_t *gpe)
7600 {
7601 	int i, nparts;
7602 
7603 	ASSERT(MUTEX_HELD(&vdc->lock));
7604 
7605 	vdc->vdisk_label = VD_DISK_LABEL_EFI;
7606 	bzero(vdc->vtoc, sizeof (struct vtoc));
7607 	bzero(vdc->geom, sizeof (struct dk_geom));
7608 	bzero(vdc->slice, sizeof (vd_slice_t) * V_NUMPAR);
7609 
7610 	nparts = gpt->efi_gpt_NumberOfPartitionEntries;
7611 
7612 	for (i = 0; i < nparts && i < VD_EFI_WD_SLICE; i++) {
7613 
7614 		if (gpe[i].efi_gpe_StartingLBA == 0 ||
7615 		    gpe[i].efi_gpe_EndingLBA == 0) {
7616 			continue;
7617 		}
7618 
7619 		vdc->slice[i].start = gpe[i].efi_gpe_StartingLBA;
7620 		vdc->slice[i].nblocks = gpe[i].efi_gpe_EndingLBA -
7621 		    gpe[i].efi_gpe_StartingLBA + 1;
7622 	}
7623 
7624 	ASSERT(vdc->vdisk_size != 0);
7625 	vdc->slice[VD_EFI_WD_SLICE].start = 0;
7626 	vdc->slice[VD_EFI_WD_SLICE].nblocks = vdc->vdisk_size;
7627 
7628 }
7629 
7630 static void
7631 vdc_store_label_vtoc(vdc_t *vdc, struct dk_geom *geom, struct vtoc *vtoc)
7632 {
7633 	int i;
7634 
7635 	ASSERT(MUTEX_HELD(&vdc->lock));
7636 	ASSERT(vdc->block_size == vtoc->v_sectorsz);
7637 
7638 	vdc->vdisk_label = VD_DISK_LABEL_VTOC;
7639 	bcopy(vtoc, vdc->vtoc, sizeof (struct vtoc));
7640 	bcopy(geom, vdc->geom, sizeof (struct dk_geom));
7641 	bzero(vdc->slice, sizeof (vd_slice_t) * V_NUMPAR);
7642 
7643 	for (i = 0; i < vtoc->v_nparts; i++) {
7644 		vdc->slice[i].start = vtoc->v_part[i].p_start;
7645 		vdc->slice[i].nblocks = vtoc->v_part[i].p_size;
7646 	}
7647 }
7648 
7649 static void
7650 vdc_store_label_unk(vdc_t *vdc)
7651 {
7652 	ASSERT(MUTEX_HELD(&vdc->lock));
7653 
7654 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
7655 	bzero(vdc->vtoc, sizeof (struct vtoc));
7656 	bzero(vdc->geom, sizeof (struct dk_geom));
7657 	bzero(vdc->slice, sizeof (vd_slice_t) * V_NUMPAR);
7658 }
7659