xref: /titanic_41/usr/src/uts/sun4v/io/vdc.c (revision 7f667e74610492ddbce8ce60f52ece95d2401949)
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 2009 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /*
28  * LDoms virtual disk client (vdc) device driver
29  *
30  * This driver runs on a guest logical domain and communicates with the virtual
31  * disk server (vds) driver running on the service domain which is exporting
32  * virtualized "disks" to the guest logical domain.
33  *
34  * The driver can be divided into four sections:
35  *
36  * 1) generic device driver housekeeping
37  *	_init, _fini, attach, detach, ops structures, etc.
38  *
39  * 2) communication channel setup
40  *	Setup the communications link over the LDC channel that vdc uses to
41  *	talk to the vDisk server. Initialise the descriptor ring which
42  *	allows the LDC clients to transfer data via memory mappings.
43  *
44  * 3) Support exported to upper layers (filesystems, etc)
45  *	The upper layers call into vdc via strategy(9E) and DKIO(7I)
46  *	ioctl calls. vdc will copy the data to be written to the descriptor
47  *	ring or maps the buffer to store the data read by the vDisk
48  *	server into the descriptor ring. It then sends a message to the
49  *	vDisk server requesting it to complete the operation.
50  *
51  * 4) Handling responses from vDisk server.
52  *	The vDisk server will ACK some or all of the messages vdc sends to it
53  *	(this is configured during the handshake). Upon receipt of an ACK
54  *	vdc will check the descriptor ring and signal to the upper layer
55  *	code waiting on the IO.
56  */
57 
58 #include <sys/atomic.h>
59 #include <sys/conf.h>
60 #include <sys/disp.h>
61 #include <sys/ddi.h>
62 #include <sys/dkio.h>
63 #include <sys/efi_partition.h>
64 #include <sys/fcntl.h>
65 #include <sys/file.h>
66 #include <sys/kstat.h>
67 #include <sys/mach_descrip.h>
68 #include <sys/modctl.h>
69 #include <sys/mdeg.h>
70 #include <sys/note.h>
71 #include <sys/open.h>
72 #include <sys/sdt.h>
73 #include <sys/stat.h>
74 #include <sys/sunddi.h>
75 #include <sys/types.h>
76 #include <sys/promif.h>
77 #include <sys/var.h>
78 #include <sys/vtoc.h>
79 #include <sys/archsystm.h>
80 #include <sys/sysmacros.h>
81 
82 #include <sys/cdio.h>
83 #include <sys/dktp/fdisk.h>
84 #include <sys/dktp/dadkio.h>
85 #include <sys/mhd.h>
86 #include <sys/scsi/generic/sense.h>
87 #include <sys/scsi/impl/uscsi.h>
88 #include <sys/scsi/impl/services.h>
89 #include <sys/scsi/targets/sddef.h>
90 
91 #include <sys/ldoms.h>
92 #include <sys/ldc.h>
93 #include <sys/vio_common.h>
94 #include <sys/vio_mailbox.h>
95 #include <sys/vio_util.h>
96 #include <sys/vdsk_common.h>
97 #include <sys/vdsk_mailbox.h>
98 #include <sys/vdc.h>
99 
100 #define	VD_OLDVTOC_LIMIT	0x7fffffff
101 
102 /*
103  * function prototypes
104  */
105 
106 /* standard driver functions */
107 static int	vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred);
108 static int	vdc_close(dev_t dev, int flag, int otyp, cred_t *cred);
109 static int	vdc_strategy(struct buf *buf);
110 static int	vdc_print(dev_t dev, char *str);
111 static int	vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk);
112 static int	vdc_read(dev_t dev, struct uio *uio, cred_t *cred);
113 static int	vdc_write(dev_t dev, struct uio *uio, cred_t *cred);
114 static int	vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode,
115 			cred_t *credp, int *rvalp);
116 static int	vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred);
117 static int	vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred);
118 
119 static int	vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,
120 			void *arg, void **resultp);
121 static int	vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd);
122 static int	vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd);
123 static int	vdc_prop_op(dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op,
124 		    int mod_flags, char *name, caddr_t valuep, int *lengthp);
125 
126 /* setup */
127 static void	vdc_min(struct buf *bufp);
128 static int	vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen);
129 static int	vdc_do_ldc_init(vdc_t *vdc, vdc_server_t *srvr);
130 static int	vdc_start_ldc_connection(vdc_t *vdc);
131 static int	vdc_create_device_nodes(vdc_t *vdc);
132 static int	vdc_create_device_nodes_efi(vdc_t *vdc);
133 static int	vdc_create_device_nodes_vtoc(vdc_t *vdc);
134 static void	vdc_create_io_kstats(vdc_t *vdc);
135 static void	vdc_create_err_kstats(vdc_t *vdc);
136 static void	vdc_set_err_kstats(vdc_t *vdc);
137 static int	vdc_get_md_node(dev_info_t *dip, md_t **mdpp,
138 		    mde_cookie_t *vd_nodep);
139 static int	vdc_init_ports(vdc_t *vdc, md_t *mdp, mde_cookie_t vd_nodep);
140 static void	vdc_fini_ports(vdc_t *vdc);
141 static void	vdc_switch_server(vdc_t *vdcp);
142 static int	vdc_do_ldc_up(vdc_t *vdc);
143 static void	vdc_terminate_ldc(vdc_t *vdc, vdc_server_t *srvr);
144 static int	vdc_init_descriptor_ring(vdc_t *vdc);
145 static void	vdc_destroy_descriptor_ring(vdc_t *vdc);
146 static int	vdc_setup_devid(vdc_t *vdc);
147 static void	vdc_store_label_efi(vdc_t *, efi_gpt_t *, efi_gpe_t *);
148 static void	vdc_store_label_vtoc(vdc_t *, struct dk_geom *,
149 		    struct extvtoc *);
150 static void	vdc_store_label_unk(vdc_t *vdc);
151 static boolean_t vdc_is_opened(vdc_t *vdc);
152 static void	vdc_update_size(vdc_t *vdc, size_t, size_t, size_t);
153 
154 /* handshake with vds */
155 static int		vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver);
156 static int		vdc_ver_negotiation(vdc_t *vdcp);
157 static int		vdc_init_attr_negotiation(vdc_t *vdc);
158 static int		vdc_attr_negotiation(vdc_t *vdcp);
159 static int		vdc_init_dring_negotiate(vdc_t *vdc);
160 static int		vdc_dring_negotiation(vdc_t *vdcp);
161 static int		vdc_send_rdx(vdc_t *vdcp);
162 static int		vdc_rdx_exchange(vdc_t *vdcp);
163 static boolean_t	vdc_is_supported_version(vio_ver_msg_t *ver_msg);
164 
165 /* processing incoming messages from vDisk server */
166 static void	vdc_process_msg_thread(vdc_t *vdc);
167 static int	vdc_recv(vdc_t *vdc, vio_msg_t *msgp, size_t *nbytesp);
168 
169 static uint_t	vdc_handle_cb(uint64_t event, caddr_t arg);
170 static int	vdc_process_data_msg(vdc_t *vdc, vio_msg_t *msg);
171 static int	vdc_handle_ver_msg(vdc_t *vdc, vio_ver_msg_t *ver_msg);
172 static int	vdc_handle_attr_msg(vdc_t *vdc, vd_attr_msg_t *attr_msg);
173 static int	vdc_handle_dring_reg_msg(vdc_t *vdc, vio_dring_reg_msg_t *msg);
174 static int 	vdc_send_request(vdc_t *vdcp, int operation,
175 		    caddr_t addr, size_t nbytes, int slice, diskaddr_t offset,
176 		    int cb_type, void *cb_arg, vio_desc_direction_t dir);
177 static int	vdc_map_to_shared_dring(vdc_t *vdcp, int idx);
178 static int 	vdc_populate_descriptor(vdc_t *vdcp, int operation,
179 		    caddr_t addr, size_t nbytes, int slice, diskaddr_t offset,
180 		    int cb_type, void *cb_arg, vio_desc_direction_t dir);
181 static int 	vdc_do_sync_op(vdc_t *vdcp, int operation, caddr_t addr,
182 		    size_t nbytes, int slice, diskaddr_t offset, int cb_type,
183 		    void *cb_arg, vio_desc_direction_t dir, boolean_t);
184 
185 static int	vdc_wait_for_response(vdc_t *vdcp, vio_msg_t *msgp);
186 static int	vdc_drain_response(vdc_t *vdcp, vio_cb_type_t cb_type,
187 		    struct buf *buf);
188 static int	vdc_depopulate_descriptor(vdc_t *vdc, uint_t idx);
189 static int	vdc_populate_mem_hdl(vdc_t *vdcp, vdc_local_desc_t *ldep);
190 static int	vdc_verify_seq_num(vdc_t *vdc, vio_dring_msg_t *dring_msg);
191 
192 /* dkio */
193 static int	vd_process_ioctl(dev_t dev, int cmd, caddr_t arg, int mode,
194 		    int *rvalp);
195 static int	vd_process_efi_ioctl(void *vdisk, int cmd, uintptr_t arg);
196 static void	vdc_create_fake_geometry(vdc_t *vdc);
197 static int	vdc_validate_geometry(vdc_t *vdc);
198 static void	vdc_validate(vdc_t *vdc);
199 static void	vdc_validate_task(void *arg);
200 static int	vdc_null_copy_func(vdc_t *vdc, void *from, void *to,
201 		    int mode, int dir);
202 static int	vdc_get_wce_convert(vdc_t *vdc, void *from, void *to,
203 		    int mode, int dir);
204 static int	vdc_set_wce_convert(vdc_t *vdc, void *from, void *to,
205 		    int mode, int dir);
206 static int	vdc_get_vtoc_convert(vdc_t *vdc, void *from, void *to,
207 		    int mode, int dir);
208 static int	vdc_set_vtoc_convert(vdc_t *vdc, void *from, void *to,
209 		    int mode, int dir);
210 static int	vdc_get_extvtoc_convert(vdc_t *vdc, void *from, void *to,
211 		    int mode, int dir);
212 static int	vdc_set_extvtoc_convert(vdc_t *vdc, void *from, void *to,
213 		    int mode, int dir);
214 static int	vdc_get_geom_convert(vdc_t *vdc, void *from, void *to,
215 		    int mode, int dir);
216 static int	vdc_set_geom_convert(vdc_t *vdc, void *from, void *to,
217 		    int mode, int dir);
218 static int	vdc_get_efi_convert(vdc_t *vdc, void *from, void *to,
219 		    int mode, int dir);
220 static int	vdc_set_efi_convert(vdc_t *vdc, void *from, void *to,
221 		    int mode, int dir);
222 
223 static void 	vdc_ownership_update(vdc_t *vdc, int ownership_flags);
224 static int	vdc_access_set(vdc_t *vdc, uint64_t flags, int mode);
225 static vdc_io_t	*vdc_failfast_io_queue(vdc_t *vdc, struct buf *buf);
226 static int	vdc_failfast_check_resv(vdc_t *vdc);
227 
228 /*
229  * Module variables
230  */
231 
232 /*
233  * Tunable variables to control how long vdc waits before timing out on
234  * various operations
235  */
236 static int	vdc_hshake_retries = 3;
237 
238 static int	vdc_timeout = 0; /* units: seconds */
239 static int 	vdc_ldcup_timeout = 1; /* units: seconds */
240 
241 static uint64_t vdc_hz_min_ldc_delay;
242 static uint64_t vdc_min_timeout_ldc = 1 * MILLISEC;
243 static uint64_t vdc_hz_max_ldc_delay;
244 static uint64_t vdc_max_timeout_ldc = 100 * MILLISEC;
245 
246 static uint64_t vdc_ldc_read_init_delay = 1 * MILLISEC;
247 static uint64_t vdc_ldc_read_max_delay = 100 * MILLISEC;
248 
249 /* values for dumping - need to run in a tighter loop */
250 static uint64_t	vdc_usec_timeout_dump = 100 * MILLISEC;	/* 0.1s units: ns */
251 static int	vdc_dump_retries = 100;
252 
253 static uint16_t	vdc_scsi_timeout = 60;	/* 60s units: seconds  */
254 
255 static uint64_t vdc_ownership_delay = 6 * MICROSEC; /* 6s units: usec */
256 
257 /* Count of the number of vdc instances attached */
258 static volatile uint32_t	vdc_instance_count = 0;
259 
260 /* Tunable to log all SCSI errors */
261 static boolean_t vdc_scsi_log_error = B_FALSE;
262 
263 /* Soft state pointer */
264 static void	*vdc_state;
265 
266 /*
267  * Controlling the verbosity of the error/debug messages
268  *
269  * vdc_msglevel - controls level of messages
270  * vdc_matchinst - 64-bit variable where each bit corresponds
271  *                 to the vdc instance the vdc_msglevel applies.
272  */
273 int		vdc_msglevel = 0x0;
274 uint64_t	vdc_matchinst = 0ull;
275 
276 /*
277  * Supported vDisk protocol version pairs.
278  *
279  * The first array entry is the latest and preferred version.
280  */
281 static const vio_ver_t	vdc_version[] = {{1, 1}};
282 
283 static struct cb_ops vdc_cb_ops = {
284 	vdc_open,	/* cb_open */
285 	vdc_close,	/* cb_close */
286 	vdc_strategy,	/* cb_strategy */
287 	vdc_print,	/* cb_print */
288 	vdc_dump,	/* cb_dump */
289 	vdc_read,	/* cb_read */
290 	vdc_write,	/* cb_write */
291 	vdc_ioctl,	/* cb_ioctl */
292 	nodev,		/* cb_devmap */
293 	nodev,		/* cb_mmap */
294 	nodev,		/* cb_segmap */
295 	nochpoll,	/* cb_chpoll */
296 	vdc_prop_op,	/* cb_prop_op */
297 	NULL,		/* cb_str */
298 	D_MP | D_64BIT,	/* cb_flag */
299 	CB_REV,		/* cb_rev */
300 	vdc_aread,	/* cb_aread */
301 	vdc_awrite	/* cb_awrite */
302 };
303 
304 static struct dev_ops vdc_ops = {
305 	DEVO_REV,	/* devo_rev */
306 	0,		/* devo_refcnt */
307 	vdc_getinfo,	/* devo_getinfo */
308 	nulldev,	/* devo_identify */
309 	nulldev,	/* devo_probe */
310 	vdc_attach,	/* devo_attach */
311 	vdc_detach,	/* devo_detach */
312 	nodev,		/* devo_reset */
313 	&vdc_cb_ops,	/* devo_cb_ops */
314 	NULL,		/* devo_bus_ops */
315 	nulldev,	/* devo_power */
316 	ddi_quiesce_not_needed,	/* devo_quiesce */
317 };
318 
319 static struct modldrv modldrv = {
320 	&mod_driverops,
321 	"virtual disk client",
322 	&vdc_ops,
323 };
324 
325 static struct modlinkage modlinkage = {
326 	MODREV_1,
327 	&modldrv,
328 	NULL
329 };
330 
331 /* -------------------------------------------------------------------------- */
332 
333 /*
334  * Device Driver housekeeping and setup
335  */
336 
337 int
338 _init(void)
339 {
340 	int	status;
341 
342 	if ((status = ddi_soft_state_init(&vdc_state, sizeof (vdc_t), 1)) != 0)
343 		return (status);
344 	if ((status = mod_install(&modlinkage)) != 0)
345 		ddi_soft_state_fini(&vdc_state);
346 	return (status);
347 }
348 
349 int
350 _info(struct modinfo *modinfop)
351 {
352 	return (mod_info(&modlinkage, modinfop));
353 }
354 
355 int
356 _fini(void)
357 {
358 	int	status;
359 
360 	if ((status = mod_remove(&modlinkage)) != 0)
361 		return (status);
362 	ddi_soft_state_fini(&vdc_state);
363 	return (0);
364 }
365 
366 static int
367 vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,  void *arg, void **resultp)
368 {
369 	_NOTE(ARGUNUSED(dip))
370 
371 	int	instance = VDCUNIT((dev_t)arg);
372 	vdc_t	*vdc = NULL;
373 
374 	switch (cmd) {
375 	case DDI_INFO_DEVT2DEVINFO:
376 		if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
377 			*resultp = NULL;
378 			return (DDI_FAILURE);
379 		}
380 		*resultp = vdc->dip;
381 		return (DDI_SUCCESS);
382 	case DDI_INFO_DEVT2INSTANCE:
383 		*resultp = (void *)(uintptr_t)instance;
384 		return (DDI_SUCCESS);
385 	default:
386 		*resultp = NULL;
387 		return (DDI_FAILURE);
388 	}
389 }
390 
391 static int
392 vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
393 {
394 	kt_did_t failfast_tid, ownership_tid;
395 	int	instance;
396 	int	rv;
397 	vdc_server_t *srvr;
398 	vdc_t	*vdc = NULL;
399 
400 	switch (cmd) {
401 	case DDI_DETACH:
402 		/* the real work happens below */
403 		break;
404 	case DDI_SUSPEND:
405 		/* nothing to do for this non-device */
406 		return (DDI_SUCCESS);
407 	default:
408 		return (DDI_FAILURE);
409 	}
410 
411 	ASSERT(cmd == DDI_DETACH);
412 	instance = ddi_get_instance(dip);
413 	DMSGX(1, "[%d] Entered\n", instance);
414 
415 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
416 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
417 		return (DDI_FAILURE);
418 	}
419 
420 	/*
421 	 * This function is called when vdc is detached or if it has failed to
422 	 * attach. In that case, the attach may have fail before the vdisk type
423 	 * has been set so we can't call vdc_is_opened(). However as the attach
424 	 * has failed, we know that the vdisk is not opened and we can safely
425 	 * detach.
426 	 */
427 	if (vdc->vdisk_type != VD_DISK_TYPE_UNK && vdc_is_opened(vdc)) {
428 		DMSG(vdc, 0, "[%d] Cannot detach: device is open", instance);
429 		return (DDI_FAILURE);
430 	}
431 
432 	if (vdc->dkio_flush_pending) {
433 		DMSG(vdc, 0,
434 		    "[%d] Cannot detach: %d outstanding DKIO flushes\n",
435 		    instance, vdc->dkio_flush_pending);
436 		return (DDI_FAILURE);
437 	}
438 
439 	if (vdc->validate_pending) {
440 		DMSG(vdc, 0,
441 		    "[%d] Cannot detach: %d outstanding validate request\n",
442 		    instance, vdc->validate_pending);
443 		return (DDI_FAILURE);
444 	}
445 
446 	DMSG(vdc, 0, "[%d] proceeding...\n", instance);
447 
448 	/* If we took ownership, release ownership */
449 	mutex_enter(&vdc->ownership_lock);
450 	if (vdc->ownership & VDC_OWNERSHIP_GRANTED) {
451 		rv = vdc_access_set(vdc, VD_ACCESS_SET_CLEAR, FKIOCTL);
452 		if (rv == 0) {
453 			vdc_ownership_update(vdc, VDC_OWNERSHIP_NONE);
454 		}
455 	}
456 	mutex_exit(&vdc->ownership_lock);
457 
458 	/* mark instance as detaching */
459 	vdc->lifecycle	= VDC_LC_DETACHING;
460 
461 	/*
462 	 * Try and disable callbacks to prevent another handshake. We have to
463 	 * disable callbacks for all servers.
464 	 */
465 	for (srvr = vdc->server_list; srvr != NULL; srvr = srvr->next) {
466 		rv = ldc_set_cb_mode(srvr->ldc_handle, LDC_CB_DISABLE);
467 		DMSG(vdc, 0, "callback disabled (ldc=%lu, rv=%d)\n",
468 		    srvr->ldc_id, rv);
469 	}
470 
471 	if (vdc->initialized & VDC_THREAD) {
472 		mutex_enter(&vdc->read_lock);
473 		if ((vdc->read_state == VDC_READ_WAITING) ||
474 		    (vdc->read_state == VDC_READ_RESET)) {
475 			vdc->read_state = VDC_READ_RESET;
476 			cv_signal(&vdc->read_cv);
477 		}
478 
479 		mutex_exit(&vdc->read_lock);
480 
481 		/* wake up any thread waiting for connection to come online */
482 		mutex_enter(&vdc->lock);
483 		if (vdc->state == VDC_STATE_INIT_WAITING) {
484 			DMSG(vdc, 0,
485 			    "[%d] write reset - move to resetting state...\n",
486 			    instance);
487 			vdc->state = VDC_STATE_RESETTING;
488 			cv_signal(&vdc->initwait_cv);
489 		}
490 		mutex_exit(&vdc->lock);
491 
492 		/* now wait until state transitions to VDC_STATE_DETACH */
493 		thread_join(vdc->msg_proc_thr->t_did);
494 		ASSERT(vdc->state == VDC_STATE_DETACH);
495 		DMSG(vdc, 0, "[%d] Reset thread exit and join ..\n",
496 		    vdc->instance);
497 	}
498 
499 	mutex_enter(&vdc->lock);
500 
501 	if (vdc->initialized & VDC_DRING)
502 		vdc_destroy_descriptor_ring(vdc);
503 
504 	vdc_fini_ports(vdc);
505 
506 	if (vdc->failfast_thread) {
507 		failfast_tid = vdc->failfast_thread->t_did;
508 		vdc->failfast_interval = 0;
509 		cv_signal(&vdc->failfast_cv);
510 	} else {
511 		failfast_tid = 0;
512 	}
513 
514 	if (vdc->ownership & VDC_OWNERSHIP_WANTED) {
515 		ownership_tid = vdc->ownership_thread->t_did;
516 		vdc->ownership = VDC_OWNERSHIP_NONE;
517 		cv_signal(&vdc->ownership_cv);
518 	} else {
519 		ownership_tid = 0;
520 	}
521 
522 	mutex_exit(&vdc->lock);
523 
524 	if (failfast_tid != 0)
525 		thread_join(failfast_tid);
526 
527 	if (ownership_tid != 0)
528 		thread_join(ownership_tid);
529 
530 	if (vdc->initialized & VDC_MINOR)
531 		ddi_remove_minor_node(dip, NULL);
532 
533 	if (vdc->io_stats) {
534 		kstat_delete(vdc->io_stats);
535 		vdc->io_stats = NULL;
536 	}
537 
538 	if (vdc->err_stats) {
539 		kstat_delete(vdc->err_stats);
540 		vdc->err_stats = NULL;
541 	}
542 
543 	if (vdc->initialized & VDC_LOCKS) {
544 		mutex_destroy(&vdc->lock);
545 		mutex_destroy(&vdc->read_lock);
546 		mutex_destroy(&vdc->ownership_lock);
547 		cv_destroy(&vdc->initwait_cv);
548 		cv_destroy(&vdc->dring_free_cv);
549 		cv_destroy(&vdc->membind_cv);
550 		cv_destroy(&vdc->sync_pending_cv);
551 		cv_destroy(&vdc->sync_blocked_cv);
552 		cv_destroy(&vdc->read_cv);
553 		cv_destroy(&vdc->running_cv);
554 		cv_destroy(&vdc->ownership_cv);
555 		cv_destroy(&vdc->failfast_cv);
556 		cv_destroy(&vdc->failfast_io_cv);
557 	}
558 
559 	if (vdc->minfo)
560 		kmem_free(vdc->minfo, sizeof (struct dk_minfo));
561 
562 	if (vdc->cinfo)
563 		kmem_free(vdc->cinfo, sizeof (struct dk_cinfo));
564 
565 	if (vdc->vtoc)
566 		kmem_free(vdc->vtoc, sizeof (struct extvtoc));
567 
568 	if (vdc->geom)
569 		kmem_free(vdc->geom, sizeof (struct dk_geom));
570 
571 	if (vdc->devid) {
572 		ddi_devid_unregister(dip);
573 		ddi_devid_free(vdc->devid);
574 	}
575 
576 	if (vdc->initialized & VDC_SOFT_STATE)
577 		ddi_soft_state_free(vdc_state, instance);
578 
579 	DMSG(vdc, 0, "[%d] End %p\n", instance, (void *)vdc);
580 
581 	return (DDI_SUCCESS);
582 }
583 
584 
585 static int
586 vdc_do_attach(dev_info_t *dip)
587 {
588 	int		instance;
589 	vdc_t		*vdc = NULL;
590 	int		status;
591 	md_t		*mdp;
592 	mde_cookie_t	vd_node;
593 
594 	ASSERT(dip != NULL);
595 
596 	instance = ddi_get_instance(dip);
597 	if (ddi_soft_state_zalloc(vdc_state, instance) != DDI_SUCCESS) {
598 		cmn_err(CE_NOTE, "[%d] Couldn't alloc state structure",
599 		    instance);
600 		return (DDI_FAILURE);
601 	}
602 
603 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
604 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
605 		return (DDI_FAILURE);
606 	}
607 
608 	/*
609 	 * We assign the value to initialized in this case to zero out the
610 	 * variable and then set bits in it to indicate what has been done
611 	 */
612 	vdc->initialized = VDC_SOFT_STATE;
613 
614 	vdc_hz_min_ldc_delay = drv_usectohz(vdc_min_timeout_ldc);
615 	vdc_hz_max_ldc_delay = drv_usectohz(vdc_max_timeout_ldc);
616 
617 	vdc->dip	= dip;
618 	vdc->instance	= instance;
619 	vdc->vdisk_type	= VD_DISK_TYPE_UNK;
620 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
621 	vdc->state	= VDC_STATE_INIT;
622 	vdc->lifecycle	= VDC_LC_ATTACHING;
623 	vdc->session_id = 0;
624 	vdc->block_size = DEV_BSIZE;
625 	vdc->max_xfer_sz = maxphys / DEV_BSIZE;
626 
627 	/*
628 	 * We assume, for now, that the vDisk server will export 'read'
629 	 * operations to us at a minimum (this is needed because of checks
630 	 * in vdc for supported operations early in the handshake process).
631 	 * The vDisk server will return ENOTSUP if this is not the case.
632 	 * The value will be overwritten during the attribute exchange with
633 	 * the bitmask of operations exported by server.
634 	 */
635 	vdc->operations = VD_OP_MASK_READ;
636 
637 	vdc->vtoc = NULL;
638 	vdc->geom = NULL;
639 	vdc->cinfo = NULL;
640 	vdc->minfo = NULL;
641 
642 	mutex_init(&vdc->lock, NULL, MUTEX_DRIVER, NULL);
643 	cv_init(&vdc->initwait_cv, NULL, CV_DRIVER, NULL);
644 	cv_init(&vdc->dring_free_cv, NULL, CV_DRIVER, NULL);
645 	cv_init(&vdc->membind_cv, NULL, CV_DRIVER, NULL);
646 	cv_init(&vdc->running_cv, NULL, CV_DRIVER, NULL);
647 
648 	vdc->threads_pending = 0;
649 	vdc->sync_op_pending = B_FALSE;
650 	vdc->sync_op_blocked = B_FALSE;
651 	cv_init(&vdc->sync_pending_cv, NULL, CV_DRIVER, NULL);
652 	cv_init(&vdc->sync_blocked_cv, NULL, CV_DRIVER, NULL);
653 
654 	mutex_init(&vdc->ownership_lock, NULL, MUTEX_DRIVER, NULL);
655 	cv_init(&vdc->ownership_cv, NULL, CV_DRIVER, NULL);
656 	cv_init(&vdc->failfast_cv, NULL, CV_DRIVER, NULL);
657 	cv_init(&vdc->failfast_io_cv, NULL, CV_DRIVER, NULL);
658 
659 	/* init blocking msg read functionality */
660 	mutex_init(&vdc->read_lock, NULL, MUTEX_DRIVER, NULL);
661 	cv_init(&vdc->read_cv, NULL, CV_DRIVER, NULL);
662 	vdc->read_state = VDC_READ_IDLE;
663 
664 	vdc->initialized |= VDC_LOCKS;
665 
666 	/* get device and port MD node for this disk instance */
667 	if (vdc_get_md_node(dip, &mdp, &vd_node) != 0) {
668 		cmn_err(CE_NOTE, "[%d] Could not get machine description node",
669 		    instance);
670 		return (DDI_FAILURE);
671 	}
672 
673 	if (vdc_init_ports(vdc, mdp, vd_node) != 0) {
674 		cmn_err(CE_NOTE, "[%d] Error initialising ports", instance);
675 		return (DDI_FAILURE);
676 	}
677 
678 	(void) md_fini_handle(mdp);
679 
680 	/* Create the kstats for saving the I/O statistics used by iostat(1M) */
681 	vdc_create_io_kstats(vdc);
682 	vdc_create_err_kstats(vdc);
683 
684 	/* Initialize remaining structures before starting the msg thread */
685 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
686 	vdc->vtoc = kmem_zalloc(sizeof (struct extvtoc), KM_SLEEP);
687 	vdc->geom = kmem_zalloc(sizeof (struct dk_geom), KM_SLEEP);
688 	vdc->minfo = kmem_zalloc(sizeof (struct dk_minfo), KM_SLEEP);
689 
690 	/* initialize the thread responsible for managing state with server */
691 	vdc->msg_proc_thr = thread_create(NULL, 0, vdc_process_msg_thread,
692 	    vdc, 0, &p0, TS_RUN, minclsyspri);
693 	if (vdc->msg_proc_thr == NULL) {
694 		cmn_err(CE_NOTE, "[%d] Failed to create msg processing thread",
695 		    instance);
696 		return (DDI_FAILURE);
697 	}
698 
699 	vdc->initialized |= VDC_THREAD;
700 
701 	atomic_inc_32(&vdc_instance_count);
702 
703 	/*
704 	 * Check the disk label. This will send requests and do the handshake.
705 	 * We don't really care about the disk label now. What we really need is
706 	 * the handshake do be done so that we know the type of the disk (slice
707 	 * or full disk) and the appropriate device nodes can be created.
708 	 */
709 
710 	mutex_enter(&vdc->lock);
711 	(void) vdc_validate_geometry(vdc);
712 	mutex_exit(&vdc->lock);
713 
714 	/*
715 	 * Now that we have the device info we can create the device nodes
716 	 */
717 	status = vdc_create_device_nodes(vdc);
718 	if (status) {
719 		DMSG(vdc, 0, "[%d] Failed to create device nodes",
720 		    instance);
721 		goto return_status;
722 	}
723 
724 	/*
725 	 * Setup devid
726 	 */
727 	if (vdc_setup_devid(vdc)) {
728 		DMSG(vdc, 0, "[%d] No device id available\n", instance);
729 	}
730 
731 	/*
732 	 * Fill in the fields of the error statistics kstat that were not
733 	 * available when creating the kstat
734 	 */
735 	vdc_set_err_kstats(vdc);
736 
737 	ddi_report_dev(dip);
738 	vdc->lifecycle	= VDC_LC_ONLINE;
739 	DMSG(vdc, 0, "[%d] Attach tasks successful\n", instance);
740 
741 return_status:
742 	DMSG(vdc, 0, "[%d] Attach completed\n", instance);
743 	return (status);
744 }
745 
746 static int
747 vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
748 {
749 	int	status;
750 
751 	switch (cmd) {
752 	case DDI_ATTACH:
753 		if ((status = vdc_do_attach(dip)) != 0)
754 			(void) vdc_detach(dip, DDI_DETACH);
755 		return (status);
756 	case DDI_RESUME:
757 		/* nothing to do for this non-device */
758 		return (DDI_SUCCESS);
759 	default:
760 		return (DDI_FAILURE);
761 	}
762 }
763 
764 static int
765 vdc_do_ldc_init(vdc_t *vdc, vdc_server_t *srvr)
766 {
767 	int			status = 0;
768 	ldc_status_t		ldc_state;
769 	ldc_attr_t		ldc_attr;
770 
771 	ASSERT(vdc != NULL);
772 	ASSERT(srvr != NULL);
773 
774 	ldc_attr.devclass = LDC_DEV_BLK;
775 	ldc_attr.instance = vdc->instance;
776 	ldc_attr.mode = LDC_MODE_UNRELIABLE;	/* unreliable transport */
777 	ldc_attr.mtu = VD_LDC_MTU;
778 
779 	if ((srvr->state & VDC_LDC_INIT) == 0) {
780 		status = ldc_init(srvr->ldc_id, &ldc_attr,
781 		    &srvr->ldc_handle);
782 		if (status != 0) {
783 			DMSG(vdc, 0, "[%d] ldc_init(chan %ld) returned %d",
784 			    vdc->instance, srvr->ldc_id, status);
785 			return (status);
786 		}
787 		srvr->state |= VDC_LDC_INIT;
788 	}
789 	status = ldc_status(srvr->ldc_handle, &ldc_state);
790 	if (status != 0) {
791 		DMSG(vdc, 0, "[%d] Cannot discover LDC status [err=%d]",
792 		    vdc->instance, status);
793 		goto init_exit;
794 	}
795 	srvr->ldc_state = ldc_state;
796 
797 	if ((srvr->state & VDC_LDC_CB) == 0) {
798 		status = ldc_reg_callback(srvr->ldc_handle, vdc_handle_cb,
799 		    (caddr_t)srvr);
800 		if (status != 0) {
801 			DMSG(vdc, 0, "[%d] LDC callback reg. failed (%d)",
802 			    vdc->instance, status);
803 			goto init_exit;
804 		}
805 		srvr->state |= VDC_LDC_CB;
806 	}
807 
808 	/*
809 	 * At this stage we have initialised LDC, we will now try and open
810 	 * the connection.
811 	 */
812 	if (srvr->ldc_state == LDC_INIT) {
813 		status = ldc_open(srvr->ldc_handle);
814 		if (status != 0) {
815 			DMSG(vdc, 0, "[%d] ldc_open(chan %ld) returned %d",
816 			    vdc->instance, srvr->ldc_id, status);
817 			goto init_exit;
818 		}
819 		srvr->state |= VDC_LDC_OPEN;
820 	}
821 
822 init_exit:
823 	if (status) {
824 		vdc_terminate_ldc(vdc, srvr);
825 	}
826 
827 	return (status);
828 }
829 
830 static int
831 vdc_start_ldc_connection(vdc_t *vdc)
832 {
833 	int		status = 0;
834 
835 	ASSERT(vdc != NULL);
836 
837 	ASSERT(MUTEX_HELD(&vdc->lock));
838 
839 	status = vdc_do_ldc_up(vdc);
840 
841 	DMSG(vdc, 0, "[%d] Finished bringing up LDC\n", vdc->instance);
842 
843 	return (status);
844 }
845 
846 static int
847 vdc_stop_ldc_connection(vdc_t *vdcp)
848 {
849 	int	status;
850 
851 	ASSERT(vdcp != NULL);
852 
853 	ASSERT(MUTEX_HELD(&vdcp->lock));
854 
855 	DMSG(vdcp, 0, ": Resetting connection to vDisk server : state %d\n",
856 	    vdcp->state);
857 
858 	status = ldc_down(vdcp->curr_server->ldc_handle);
859 	DMSG(vdcp, 0, "ldc_down() = %d\n", status);
860 
861 	vdcp->initialized &= ~VDC_HANDSHAKE;
862 	DMSG(vdcp, 0, "initialized=%x\n", vdcp->initialized);
863 
864 	return (status);
865 }
866 
867 static void
868 vdc_create_io_kstats(vdc_t *vdc)
869 {
870 	if (vdc->io_stats != NULL) {
871 		DMSG(vdc, 0, "[%d] I/O kstat already exists\n", vdc->instance);
872 		return;
873 	}
874 
875 	vdc->io_stats = kstat_create(VDC_DRIVER_NAME, vdc->instance, NULL,
876 	    "disk", KSTAT_TYPE_IO, 1, KSTAT_FLAG_PERSISTENT);
877 	if (vdc->io_stats != NULL) {
878 		vdc->io_stats->ks_lock = &vdc->lock;
879 		kstat_install(vdc->io_stats);
880 	} else {
881 		cmn_err(CE_NOTE, "[%d] Failed to create kstat: I/O statistics"
882 		    " will not be gathered", vdc->instance);
883 	}
884 }
885 
886 static void
887 vdc_create_err_kstats(vdc_t *vdc)
888 {
889 	vd_err_stats_t	*stp;
890 	char	kstatmodule_err[KSTAT_STRLEN];
891 	char	kstatname[KSTAT_STRLEN];
892 	int	ndata = (sizeof (vd_err_stats_t) / sizeof (kstat_named_t));
893 	int	instance = vdc->instance;
894 
895 	if (vdc->err_stats != NULL) {
896 		DMSG(vdc, 0, "[%d] ERR kstat already exists\n", vdc->instance);
897 		return;
898 	}
899 
900 	(void) snprintf(kstatmodule_err, sizeof (kstatmodule_err),
901 	    "%serr", VDC_DRIVER_NAME);
902 	(void) snprintf(kstatname, sizeof (kstatname),
903 	    "%s%d,err", VDC_DRIVER_NAME, instance);
904 
905 	vdc->err_stats = kstat_create(kstatmodule_err, instance, kstatname,
906 	    "device_error", KSTAT_TYPE_NAMED, ndata, KSTAT_FLAG_PERSISTENT);
907 
908 	if (vdc->err_stats == NULL) {
909 		cmn_err(CE_NOTE, "[%d] Failed to create kstat: Error statistics"
910 		    " will not be gathered", instance);
911 		return;
912 	}
913 
914 	stp = (vd_err_stats_t *)vdc->err_stats->ks_data;
915 	kstat_named_init(&stp->vd_softerrs,	"Soft Errors",
916 	    KSTAT_DATA_UINT32);
917 	kstat_named_init(&stp->vd_transerrs,	"Transport Errors",
918 	    KSTAT_DATA_UINT32);
919 	kstat_named_init(&stp->vd_protoerrs,	"Protocol Errors",
920 	    KSTAT_DATA_UINT32);
921 	kstat_named_init(&stp->vd_vid,		"Vendor",
922 	    KSTAT_DATA_CHAR);
923 	kstat_named_init(&stp->vd_pid,		"Product",
924 	    KSTAT_DATA_CHAR);
925 	kstat_named_init(&stp->vd_capacity,	"Size",
926 	    KSTAT_DATA_ULONGLONG);
927 
928 	vdc->err_stats->ks_update  = nulldev;
929 
930 	kstat_install(vdc->err_stats);
931 }
932 
933 static void
934 vdc_set_err_kstats(vdc_t *vdc)
935 {
936 	vd_err_stats_t  *stp;
937 
938 	if (vdc->err_stats == NULL)
939 		return;
940 
941 	mutex_enter(&vdc->lock);
942 
943 	stp = (vd_err_stats_t *)vdc->err_stats->ks_data;
944 	ASSERT(stp != NULL);
945 
946 	stp->vd_capacity.value.ui64 = vdc->vdisk_size * vdc->block_size;
947 	(void) strcpy(stp->vd_vid.value.c, "SUN");
948 	(void) strcpy(stp->vd_pid.value.c, "VDSK");
949 
950 	mutex_exit(&vdc->lock);
951 }
952 
953 static int
954 vdc_create_device_nodes_efi(vdc_t *vdc)
955 {
956 	ddi_remove_minor_node(vdc->dip, "h");
957 	ddi_remove_minor_node(vdc->dip, "h,raw");
958 
959 	if (ddi_create_minor_node(vdc->dip, "wd", S_IFBLK,
960 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
961 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
962 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'wd'",
963 		    vdc->instance);
964 		return (EIO);
965 	}
966 
967 	/* if any device node is created we set this flag */
968 	vdc->initialized |= VDC_MINOR;
969 
970 	if (ddi_create_minor_node(vdc->dip, "wd,raw", S_IFCHR,
971 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
972 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
973 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'wd,raw'",
974 		    vdc->instance);
975 		return (EIO);
976 	}
977 
978 	return (0);
979 }
980 
981 static int
982 vdc_create_device_nodes_vtoc(vdc_t *vdc)
983 {
984 	ddi_remove_minor_node(vdc->dip, "wd");
985 	ddi_remove_minor_node(vdc->dip, "wd,raw");
986 
987 	if (ddi_create_minor_node(vdc->dip, "h", S_IFBLK,
988 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
989 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
990 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'h'",
991 		    vdc->instance);
992 		return (EIO);
993 	}
994 
995 	/* if any device node is created we set this flag */
996 	vdc->initialized |= VDC_MINOR;
997 
998 	if (ddi_create_minor_node(vdc->dip, "h,raw", S_IFCHR,
999 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
1000 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
1001 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'h,raw'",
1002 		    vdc->instance);
1003 		return (EIO);
1004 	}
1005 
1006 	return (0);
1007 }
1008 
1009 /*
1010  * Function:
1011  *	vdc_create_device_nodes
1012  *
1013  * Description:
1014  *	This function creates the block and character device nodes under
1015  *	/devices. It is called as part of the attach(9E) of the instance
1016  *	during the handshake with vds after vds has sent the attributes
1017  *	to vdc.
1018  *
1019  *	If the device is of type VD_DISK_TYPE_SLICE then the minor node
1020  *	of 2 is used in keeping with the Solaris convention that slice 2
1021  *	refers to a whole disk. Slices start at 'a'
1022  *
1023  * Parameters:
1024  *	vdc 		- soft state pointer
1025  *
1026  * Return Values
1027  *	0		- Success
1028  *	EIO		- Failed to create node
1029  *	EINVAL		- Unknown type of disk exported
1030  */
1031 static int
1032 vdc_create_device_nodes(vdc_t *vdc)
1033 {
1034 	char		name[sizeof ("s,raw")];
1035 	dev_info_t	*dip = NULL;
1036 	int		instance, status;
1037 	int		num_slices = 1;
1038 	int		i;
1039 
1040 	ASSERT(vdc != NULL);
1041 
1042 	instance = vdc->instance;
1043 	dip = vdc->dip;
1044 
1045 	switch (vdc->vdisk_type) {
1046 	case VD_DISK_TYPE_DISK:
1047 		num_slices = V_NUMPAR;
1048 		break;
1049 	case VD_DISK_TYPE_SLICE:
1050 		num_slices = 1;
1051 		break;
1052 	case VD_DISK_TYPE_UNK:
1053 	default:
1054 		return (EINVAL);
1055 	}
1056 
1057 	/*
1058 	 * Minor nodes are different for EFI disks: EFI disks do not have
1059 	 * a minor node 'g' for the minor number corresponding to slice
1060 	 * VD_EFI_WD_SLICE (slice 7) instead they have a minor node 'wd'
1061 	 * representing the whole disk.
1062 	 */
1063 	for (i = 0; i < num_slices; i++) {
1064 
1065 		if (i == VD_EFI_WD_SLICE) {
1066 			if (vdc->vdisk_label == VD_DISK_LABEL_EFI)
1067 				status = vdc_create_device_nodes_efi(vdc);
1068 			else
1069 				status = vdc_create_device_nodes_vtoc(vdc);
1070 			if (status != 0)
1071 				return (status);
1072 			continue;
1073 		}
1074 
1075 		(void) snprintf(name, sizeof (name), "%c", 'a' + i);
1076 		if (ddi_create_minor_node(dip, name, S_IFBLK,
1077 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
1078 			cmn_err(CE_NOTE, "[%d] Couldn't add block node '%s'",
1079 			    instance, name);
1080 			return (EIO);
1081 		}
1082 
1083 		/* if any device node is created we set this flag */
1084 		vdc->initialized |= VDC_MINOR;
1085 
1086 		(void) snprintf(name, sizeof (name), "%c%s", 'a' + i, ",raw");
1087 
1088 		if (ddi_create_minor_node(dip, name, S_IFCHR,
1089 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
1090 			cmn_err(CE_NOTE, "[%d] Couldn't add raw node '%s'",
1091 			    instance, name);
1092 			return (EIO);
1093 		}
1094 	}
1095 
1096 	return (0);
1097 }
1098 
1099 /*
1100  * Driver prop_op(9e) entry point function. Return the number of blocks for
1101  * the partition in question or forward the request to the property facilities.
1102  */
1103 static int
1104 vdc_prop_op(dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op, int mod_flags,
1105     char *name, caddr_t valuep, int *lengthp)
1106 {
1107 	int instance = ddi_get_instance(dip);
1108 	vdc_t *vdc;
1109 	uint64_t nblocks;
1110 	uint_t blksize;
1111 
1112 	vdc = ddi_get_soft_state(vdc_state, instance);
1113 
1114 	if (dev == DDI_DEV_T_ANY || vdc == NULL) {
1115 		return (ddi_prop_op(dev, dip, prop_op, mod_flags,
1116 		    name, valuep, lengthp));
1117 	}
1118 
1119 	mutex_enter(&vdc->lock);
1120 	(void) vdc_validate_geometry(vdc);
1121 	if (vdc->vdisk_label == VD_DISK_LABEL_UNK) {
1122 		mutex_exit(&vdc->lock);
1123 		return (ddi_prop_op(dev, dip, prop_op, mod_flags,
1124 		    name, valuep, lengthp));
1125 	}
1126 	nblocks = vdc->slice[VDCPART(dev)].nblocks;
1127 	blksize = vdc->block_size;
1128 	mutex_exit(&vdc->lock);
1129 
1130 	return (ddi_prop_op_nblocks_blksize(dev, dip, prop_op, mod_flags,
1131 	    name, valuep, lengthp, nblocks, blksize));
1132 }
1133 
1134 /*
1135  * Function:
1136  *	vdc_is_opened
1137  *
1138  * Description:
1139  *	This function checks if any slice of a given virtual disk is
1140  *	currently opened.
1141  *
1142  * Parameters:
1143  *	vdc 		- soft state pointer
1144  *
1145  * Return Values
1146  *	B_TRUE		- at least one slice is opened.
1147  *	B_FALSE		- no slice is opened.
1148  */
1149 static boolean_t
1150 vdc_is_opened(vdc_t *vdc)
1151 {
1152 	int i, nslices;
1153 
1154 	switch (vdc->vdisk_type) {
1155 	case VD_DISK_TYPE_DISK:
1156 		nslices = V_NUMPAR;
1157 		break;
1158 	case VD_DISK_TYPE_SLICE:
1159 		nslices = 1;
1160 		break;
1161 	case VD_DISK_TYPE_UNK:
1162 	default:
1163 		ASSERT(0);
1164 	}
1165 
1166 	/* check if there's any layered open */
1167 	for (i = 0; i < nslices; i++) {
1168 		if (vdc->open_lyr[i] > 0)
1169 			return (B_TRUE);
1170 	}
1171 
1172 	/* check if there is any other kind of open */
1173 	for (i = 0; i < OTYPCNT; i++) {
1174 		if (vdc->open[i] != 0)
1175 			return (B_TRUE);
1176 	}
1177 
1178 	return (B_FALSE);
1179 }
1180 
1181 static int
1182 vdc_mark_opened(vdc_t *vdc, int slice, int flag, int otyp)
1183 {
1184 	uint8_t slicemask;
1185 	int i;
1186 
1187 	ASSERT(otyp < OTYPCNT);
1188 	ASSERT(slice < V_NUMPAR);
1189 	ASSERT(MUTEX_HELD(&vdc->lock));
1190 
1191 	slicemask = 1 << slice;
1192 
1193 	/* check if slice is already exclusively opened */
1194 	if (vdc->open_excl & slicemask)
1195 		return (EBUSY);
1196 
1197 	/* if open exclusive, check if slice is already opened */
1198 	if (flag & FEXCL) {
1199 		if (vdc->open_lyr[slice] > 0)
1200 			return (EBUSY);
1201 		for (i = 0; i < OTYPCNT; i++) {
1202 			if (vdc->open[i] & slicemask)
1203 				return (EBUSY);
1204 		}
1205 		vdc->open_excl |= slicemask;
1206 	}
1207 
1208 	/* mark slice as opened */
1209 	if (otyp == OTYP_LYR) {
1210 		vdc->open_lyr[slice]++;
1211 	} else {
1212 		vdc->open[otyp] |= slicemask;
1213 	}
1214 
1215 	return (0);
1216 }
1217 
1218 static void
1219 vdc_mark_closed(vdc_t *vdc, int slice, int flag, int otyp)
1220 {
1221 	uint8_t slicemask;
1222 
1223 	ASSERT(otyp < OTYPCNT);
1224 	ASSERT(slice < V_NUMPAR);
1225 	ASSERT(MUTEX_HELD(&vdc->lock));
1226 
1227 	slicemask = 1 << slice;
1228 
1229 	if (otyp == OTYP_LYR) {
1230 		ASSERT(vdc->open_lyr[slice] > 0);
1231 		vdc->open_lyr[slice]--;
1232 	} else {
1233 		vdc->open[otyp] &= ~slicemask;
1234 	}
1235 
1236 	if (flag & FEXCL)
1237 		vdc->open_excl &= ~slicemask;
1238 }
1239 
1240 static int
1241 vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred)
1242 {
1243 	_NOTE(ARGUNUSED(cred))
1244 
1245 	int	instance, nodelay;
1246 	int	slice, status = 0;
1247 	vdc_t	*vdc;
1248 
1249 	ASSERT(dev != NULL);
1250 	instance = VDCUNIT(*dev);
1251 
1252 	if (otyp >= OTYPCNT)
1253 		return (EINVAL);
1254 
1255 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1256 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1257 		return (ENXIO);
1258 	}
1259 
1260 	DMSG(vdc, 0, "minor = %d flag = %x, otyp = %x\n",
1261 	    getminor(*dev), flag, otyp);
1262 
1263 	slice = VDCPART(*dev);
1264 
1265 	nodelay = flag & (FNDELAY | FNONBLOCK);
1266 
1267 	if ((flag & FWRITE) && (!nodelay) &&
1268 	    !(VD_OP_SUPPORTED(vdc->operations, VD_OP_BWRITE))) {
1269 		return (EROFS);
1270 	}
1271 
1272 	mutex_enter(&vdc->lock);
1273 
1274 	status = vdc_mark_opened(vdc, slice, flag, otyp);
1275 
1276 	if (status != 0) {
1277 		mutex_exit(&vdc->lock);
1278 		return (status);
1279 	}
1280 
1281 	if (nodelay) {
1282 
1283 		/* don't resubmit a validate request if there's already one */
1284 		if (vdc->validate_pending > 0) {
1285 			mutex_exit(&vdc->lock);
1286 			return (0);
1287 		}
1288 
1289 		/* call vdc_validate() asynchronously to avoid blocking */
1290 		if (taskq_dispatch(system_taskq, vdc_validate_task,
1291 		    (void *)vdc, TQ_NOSLEEP) == NULL) {
1292 			vdc_mark_closed(vdc, slice, flag, otyp);
1293 			mutex_exit(&vdc->lock);
1294 			return (ENXIO);
1295 		}
1296 
1297 		vdc->validate_pending++;
1298 		mutex_exit(&vdc->lock);
1299 		return (0);
1300 	}
1301 
1302 	mutex_exit(&vdc->lock);
1303 
1304 	vdc_validate(vdc);
1305 
1306 	mutex_enter(&vdc->lock);
1307 
1308 	if (vdc->vdisk_label == VD_DISK_LABEL_UNK ||
1309 	    vdc->slice[slice].nblocks == 0) {
1310 		vdc_mark_closed(vdc, slice, flag, otyp);
1311 		status = EIO;
1312 	}
1313 
1314 	mutex_exit(&vdc->lock);
1315 
1316 	return (status);
1317 }
1318 
1319 static int
1320 vdc_close(dev_t dev, int flag, int otyp, cred_t *cred)
1321 {
1322 	_NOTE(ARGUNUSED(cred))
1323 
1324 	int	instance;
1325 	int	slice;
1326 	int	rv, rval;
1327 	vdc_t	*vdc;
1328 
1329 	instance = VDCUNIT(dev);
1330 
1331 	if (otyp >= OTYPCNT)
1332 		return (EINVAL);
1333 
1334 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1335 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1336 		return (ENXIO);
1337 	}
1338 
1339 	DMSG(vdc, 0, "[%d] flag = %x, otyp = %x\n", instance, flag, otyp);
1340 
1341 	slice = VDCPART(dev);
1342 
1343 	/*
1344 	 * Attempt to flush the W$ on a close operation. If this is
1345 	 * not a supported IOCTL command or the backing device is read-only
1346 	 * do not fail the close operation.
1347 	 */
1348 	rv = vd_process_ioctl(dev, DKIOCFLUSHWRITECACHE, NULL, FKIOCTL, &rval);
1349 
1350 	if (rv != 0 && rv != ENOTSUP && rv != ENOTTY && rv != EROFS) {
1351 		DMSG(vdc, 0, "[%d] flush failed with error %d on close\n",
1352 		    instance, rv);
1353 		return (EIO);
1354 	}
1355 
1356 	mutex_enter(&vdc->lock);
1357 	vdc_mark_closed(vdc, slice, flag, otyp);
1358 	mutex_exit(&vdc->lock);
1359 
1360 	return (0);
1361 }
1362 
1363 static int
1364 vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
1365 {
1366 	_NOTE(ARGUNUSED(credp))
1367 
1368 	return (vd_process_ioctl(dev, cmd, (caddr_t)arg, mode, rvalp));
1369 }
1370 
1371 static int
1372 vdc_print(dev_t dev, char *str)
1373 {
1374 	cmn_err(CE_NOTE, "vdc%d:  %s", VDCUNIT(dev), str);
1375 	return (0);
1376 }
1377 
1378 static int
1379 vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk)
1380 {
1381 	int	rv;
1382 	size_t	nbytes = nblk * DEV_BSIZE;
1383 	int	instance = VDCUNIT(dev);
1384 	vdc_t	*vdc = NULL;
1385 
1386 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1387 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1388 		return (ENXIO);
1389 	}
1390 
1391 	DMSG(vdc, 2, "[%d] dump %ld bytes at block 0x%lx : addr=0x%p\n",
1392 	    instance, nbytes, blkno, (void *)addr);
1393 	rv = vdc_send_request(vdc, VD_OP_BWRITE, addr, nbytes,
1394 	    VDCPART(dev), blkno, CB_STRATEGY, 0, VIO_write_dir);
1395 	if (rv) {
1396 		DMSG(vdc, 0, "Failed to do a disk dump (err=%d)\n", rv);
1397 		return (rv);
1398 	}
1399 
1400 	if (ddi_in_panic())
1401 		(void) vdc_drain_response(vdc, CB_STRATEGY, NULL);
1402 
1403 	DMSG(vdc, 0, "[%d] End\n", instance);
1404 
1405 	return (0);
1406 }
1407 
1408 /* -------------------------------------------------------------------------- */
1409 
1410 /*
1411  * Disk access routines
1412  *
1413  */
1414 
1415 /*
1416  * vdc_strategy()
1417  *
1418  * Return Value:
1419  *	0:	As per strategy(9E), the strategy() function must return 0
1420  *		[ bioerror(9f) sets b_flags to the proper error code ]
1421  */
1422 static int
1423 vdc_strategy(struct buf *buf)
1424 {
1425 	int	rv = -1;
1426 	vdc_t	*vdc = NULL;
1427 	int	instance = VDCUNIT(buf->b_edev);
1428 	int	op = (buf->b_flags & B_READ) ? VD_OP_BREAD : VD_OP_BWRITE;
1429 	int	slice;
1430 
1431 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1432 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1433 		bioerror(buf, ENXIO);
1434 		biodone(buf);
1435 		return (0);
1436 	}
1437 
1438 	DMSG(vdc, 2, "[%d] %s %ld bytes at block %llx : b_addr=0x%p\n",
1439 	    instance, (buf->b_flags & B_READ) ? "Read" : "Write",
1440 	    buf->b_bcount, buf->b_lblkno, (void *)buf->b_un.b_addr);
1441 
1442 	bp_mapin(buf);
1443 
1444 	if ((long)buf->b_private == VD_SLICE_NONE) {
1445 		/* I/O using an absolute disk offset */
1446 		slice = VD_SLICE_NONE;
1447 	} else {
1448 		slice = VDCPART(buf->b_edev);
1449 	}
1450 
1451 	rv = vdc_send_request(vdc, op, (caddr_t)buf->b_un.b_addr,
1452 	    buf->b_bcount, slice, buf->b_lblkno,
1453 	    CB_STRATEGY, buf, (op == VD_OP_BREAD) ? VIO_read_dir :
1454 	    VIO_write_dir);
1455 
1456 	/*
1457 	 * If the request was successfully sent, the strategy call returns and
1458 	 * the ACK handler calls the bioxxx functions when the vDisk server is
1459 	 * done otherwise we handle the error here.
1460 	 */
1461 	if (rv) {
1462 		DMSG(vdc, 0, "Failed to read/write (err=%d)\n", rv);
1463 		bioerror(buf, rv);
1464 		biodone(buf);
1465 	} else if (ddi_in_panic()) {
1466 		rv = vdc_drain_response(vdc, CB_STRATEGY, buf);
1467 		if (rv != 0) {
1468 			bioerror(buf, EIO);
1469 			biodone(buf);
1470 		}
1471 	}
1472 
1473 	return (0);
1474 }
1475 
1476 /*
1477  * Function:
1478  *	vdc_min
1479  *
1480  * Description:
1481  *	Routine to limit the size of a data transfer. Used in
1482  *	conjunction with physio(9F).
1483  *
1484  * Arguments:
1485  *	bp - pointer to the indicated buf(9S) struct.
1486  *
1487  */
1488 static void
1489 vdc_min(struct buf *bufp)
1490 {
1491 	vdc_t	*vdc = NULL;
1492 	int	instance = VDCUNIT(bufp->b_edev);
1493 
1494 	vdc = ddi_get_soft_state(vdc_state, instance);
1495 	VERIFY(vdc != NULL);
1496 
1497 	if (bufp->b_bcount > (vdc->max_xfer_sz * vdc->block_size)) {
1498 		bufp->b_bcount = vdc->max_xfer_sz * vdc->block_size;
1499 	}
1500 }
1501 
1502 static int
1503 vdc_read(dev_t dev, struct uio *uio, cred_t *cred)
1504 {
1505 	_NOTE(ARGUNUSED(cred))
1506 
1507 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1508 	return (physio(vdc_strategy, NULL, dev, B_READ, vdc_min, uio));
1509 }
1510 
1511 static int
1512 vdc_write(dev_t dev, struct uio *uio, cred_t *cred)
1513 {
1514 	_NOTE(ARGUNUSED(cred))
1515 
1516 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1517 	return (physio(vdc_strategy, NULL, dev, B_WRITE, vdc_min, uio));
1518 }
1519 
1520 static int
1521 vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred)
1522 {
1523 	_NOTE(ARGUNUSED(cred))
1524 
1525 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1526 	return (aphysio(vdc_strategy, anocancel, dev, B_READ, vdc_min, aio));
1527 }
1528 
1529 static int
1530 vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred)
1531 {
1532 	_NOTE(ARGUNUSED(cred))
1533 
1534 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1535 	return (aphysio(vdc_strategy, anocancel, dev, B_WRITE, vdc_min, aio));
1536 }
1537 
1538 
1539 /* -------------------------------------------------------------------------- */
1540 
1541 /*
1542  * Handshake support
1543  */
1544 
1545 
1546 /*
1547  * Function:
1548  *	vdc_init_ver_negotiation()
1549  *
1550  * Description:
1551  *
1552  * Arguments:
1553  *	vdc	- soft state pointer for this instance of the device driver.
1554  *
1555  * Return Code:
1556  *	0	- Success
1557  */
1558 static int
1559 vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver)
1560 {
1561 	vio_ver_msg_t	pkt;
1562 	size_t		msglen = sizeof (pkt);
1563 	int		status = -1;
1564 
1565 	ASSERT(vdc != NULL);
1566 	ASSERT(mutex_owned(&vdc->lock));
1567 
1568 	DMSG(vdc, 0, "[%d] Entered.\n", vdc->instance);
1569 
1570 	/*
1571 	 * set the Session ID to a unique value
1572 	 * (the lower 32 bits of the clock tick)
1573 	 */
1574 	vdc->session_id = ((uint32_t)gettick() & 0xffffffff);
1575 	DMSG(vdc, 0, "[%d] Set SID to 0x%lx\n", vdc->instance, vdc->session_id);
1576 
1577 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1578 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1579 	pkt.tag.vio_subtype_env = VIO_VER_INFO;
1580 	pkt.tag.vio_sid = vdc->session_id;
1581 	pkt.dev_class = VDEV_DISK;
1582 	pkt.ver_major = ver.major;
1583 	pkt.ver_minor = ver.minor;
1584 
1585 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1586 	DMSG(vdc, 0, "[%d] Ver info sent (status = %d)\n",
1587 	    vdc->instance, status);
1588 	if ((status != 0) || (msglen != sizeof (vio_ver_msg_t))) {
1589 		DMSG(vdc, 0, "[%d] Failed to send Ver negotiation info: "
1590 		    "id(%lx) rv(%d) size(%ld)", vdc->instance,
1591 		    vdc->curr_server->ldc_handle, status, msglen);
1592 		if (msglen != sizeof (vio_ver_msg_t))
1593 			status = ENOMSG;
1594 	}
1595 
1596 	return (status);
1597 }
1598 
1599 /*
1600  * Function:
1601  *	vdc_ver_negotiation()
1602  *
1603  * Description:
1604  *
1605  * Arguments:
1606  *	vdcp	- soft state pointer for this instance of the device driver.
1607  *
1608  * Return Code:
1609  *	0	- Success
1610  */
1611 static int
1612 vdc_ver_negotiation(vdc_t *vdcp)
1613 {
1614 	vio_msg_t vio_msg;
1615 	int status;
1616 
1617 	if (status = vdc_init_ver_negotiation(vdcp, vdc_version[0]))
1618 		return (status);
1619 
1620 	/* release lock and wait for response */
1621 	mutex_exit(&vdcp->lock);
1622 	status = vdc_wait_for_response(vdcp, &vio_msg);
1623 	mutex_enter(&vdcp->lock);
1624 	if (status) {
1625 		DMSG(vdcp, 0,
1626 		    "[%d] Failed waiting for Ver negotiation response, rv(%d)",
1627 		    vdcp->instance, status);
1628 		return (status);
1629 	}
1630 
1631 	/* check type and sub_type ... */
1632 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1633 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1634 		DMSG(vdcp, 0, "[%d] Invalid ver negotiation response\n",
1635 		    vdcp->instance);
1636 		return (EPROTO);
1637 	}
1638 
1639 	return (vdc_handle_ver_msg(vdcp, (vio_ver_msg_t *)&vio_msg));
1640 }
1641 
1642 /*
1643  * Function:
1644  *	vdc_init_attr_negotiation()
1645  *
1646  * Description:
1647  *
1648  * Arguments:
1649  *	vdc	- soft state pointer for this instance of the device driver.
1650  *
1651  * Return Code:
1652  *	0	- Success
1653  */
1654 static int
1655 vdc_init_attr_negotiation(vdc_t *vdc)
1656 {
1657 	vd_attr_msg_t	pkt;
1658 	size_t		msglen = sizeof (pkt);
1659 	int		status;
1660 
1661 	ASSERT(vdc != NULL);
1662 	ASSERT(mutex_owned(&vdc->lock));
1663 
1664 	DMSG(vdc, 0, "[%d] entered\n", vdc->instance);
1665 
1666 	/* fill in tag */
1667 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1668 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1669 	pkt.tag.vio_subtype_env = VIO_ATTR_INFO;
1670 	pkt.tag.vio_sid = vdc->session_id;
1671 	/* fill in payload */
1672 	pkt.max_xfer_sz = vdc->max_xfer_sz;
1673 	pkt.vdisk_block_size = vdc->block_size;
1674 	pkt.xfer_mode = VIO_DRING_MODE_V1_0;
1675 	pkt.operations = 0;	/* server will set bits of valid operations */
1676 	pkt.vdisk_type = 0;	/* server will set to valid device type */
1677 	pkt.vdisk_media = 0;	/* server will set to valid media type */
1678 	pkt.vdisk_size = 0;	/* server will set to valid size */
1679 
1680 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1681 	DMSG(vdc, 0, "Attr info sent (status = %d)\n", status);
1682 
1683 	if ((status != 0) || (msglen != sizeof (vd_attr_msg_t))) {
1684 		DMSG(vdc, 0, "[%d] Failed to send Attr negotiation info: "
1685 		    "id(%lx) rv(%d) size(%ld)", vdc->instance,
1686 		    vdc->curr_server->ldc_handle, status, msglen);
1687 		if (msglen != sizeof (vd_attr_msg_t))
1688 			status = ENOMSG;
1689 	}
1690 
1691 	return (status);
1692 }
1693 
1694 /*
1695  * Function:
1696  *	vdc_attr_negotiation()
1697  *
1698  * Description:
1699  *
1700  * Arguments:
1701  *	vdc	- soft state pointer for this instance of the device driver.
1702  *
1703  * Return Code:
1704  *	0	- Success
1705  */
1706 static int
1707 vdc_attr_negotiation(vdc_t *vdcp)
1708 {
1709 	int status;
1710 	vio_msg_t vio_msg;
1711 
1712 	if (status = vdc_init_attr_negotiation(vdcp))
1713 		return (status);
1714 
1715 	/* release lock and wait for response */
1716 	mutex_exit(&vdcp->lock);
1717 	status = vdc_wait_for_response(vdcp, &vio_msg);
1718 	mutex_enter(&vdcp->lock);
1719 	if (status) {
1720 		DMSG(vdcp, 0,
1721 		    "[%d] Failed waiting for Attr negotiation response, rv(%d)",
1722 		    vdcp->instance, status);
1723 		return (status);
1724 	}
1725 
1726 	/* check type and sub_type ... */
1727 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1728 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1729 		DMSG(vdcp, 0, "[%d] Invalid attr negotiation response\n",
1730 		    vdcp->instance);
1731 		return (EPROTO);
1732 	}
1733 
1734 	return (vdc_handle_attr_msg(vdcp, (vd_attr_msg_t *)&vio_msg));
1735 }
1736 
1737 
1738 /*
1739  * Function:
1740  *	vdc_init_dring_negotiate()
1741  *
1742  * Description:
1743  *
1744  * Arguments:
1745  *	vdc	- soft state pointer for this instance of the device driver.
1746  *
1747  * Return Code:
1748  *	0	- Success
1749  */
1750 static int
1751 vdc_init_dring_negotiate(vdc_t *vdc)
1752 {
1753 	vio_dring_reg_msg_t	pkt;
1754 	size_t			msglen = sizeof (pkt);
1755 	int			status = -1;
1756 	int			retry;
1757 	int			nretries = 10;
1758 
1759 	ASSERT(vdc != NULL);
1760 	ASSERT(mutex_owned(&vdc->lock));
1761 
1762 	for (retry = 0; retry < nretries; retry++) {
1763 		status = vdc_init_descriptor_ring(vdc);
1764 		if (status != EAGAIN)
1765 			break;
1766 		drv_usecwait(vdc_min_timeout_ldc);
1767 	}
1768 
1769 	if (status != 0) {
1770 		DMSG(vdc, 0, "[%d] Failed to init DRing (status = %d)\n",
1771 		    vdc->instance, status);
1772 		return (status);
1773 	}
1774 
1775 	DMSG(vdc, 0, "[%d] Init of descriptor ring completed (status = %d)\n",
1776 	    vdc->instance, status);
1777 
1778 	/* fill in tag */
1779 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1780 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1781 	pkt.tag.vio_subtype_env = VIO_DRING_REG;
1782 	pkt.tag.vio_sid = vdc->session_id;
1783 	/* fill in payload */
1784 	pkt.dring_ident = 0;
1785 	pkt.num_descriptors = vdc->dring_len;
1786 	pkt.descriptor_size = vdc->dring_entry_size;
1787 	pkt.options = (VIO_TX_DRING | VIO_RX_DRING);
1788 	pkt.ncookies = vdc->dring_cookie_count;
1789 	pkt.cookie[0] = vdc->dring_cookie[0];	/* for now just one cookie */
1790 
1791 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1792 	if (status != 0) {
1793 		DMSG(vdc, 0, "[%d] Failed to register DRing (err = %d)",
1794 		    vdc->instance, status);
1795 	}
1796 
1797 	return (status);
1798 }
1799 
1800 
1801 /*
1802  * Function:
1803  *	vdc_dring_negotiation()
1804  *
1805  * Description:
1806  *
1807  * Arguments:
1808  *	vdc	- soft state pointer for this instance of the device driver.
1809  *
1810  * Return Code:
1811  *	0	- Success
1812  */
1813 static int
1814 vdc_dring_negotiation(vdc_t *vdcp)
1815 {
1816 	int status;
1817 	vio_msg_t vio_msg;
1818 
1819 	if (status = vdc_init_dring_negotiate(vdcp))
1820 		return (status);
1821 
1822 	/* release lock and wait for response */
1823 	mutex_exit(&vdcp->lock);
1824 	status = vdc_wait_for_response(vdcp, &vio_msg);
1825 	mutex_enter(&vdcp->lock);
1826 	if (status) {
1827 		DMSG(vdcp, 0,
1828 		    "[%d] Failed waiting for Dring negotiation response,"
1829 		    " rv(%d)", vdcp->instance, status);
1830 		return (status);
1831 	}
1832 
1833 	/* check type and sub_type ... */
1834 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1835 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1836 		DMSG(vdcp, 0, "[%d] Invalid Dring negotiation response\n",
1837 		    vdcp->instance);
1838 		return (EPROTO);
1839 	}
1840 
1841 	return (vdc_handle_dring_reg_msg(vdcp,
1842 	    (vio_dring_reg_msg_t *)&vio_msg));
1843 }
1844 
1845 
1846 /*
1847  * Function:
1848  *	vdc_send_rdx()
1849  *
1850  * Description:
1851  *
1852  * Arguments:
1853  *	vdc	- soft state pointer for this instance of the device driver.
1854  *
1855  * Return Code:
1856  *	0	- Success
1857  */
1858 static int
1859 vdc_send_rdx(vdc_t *vdcp)
1860 {
1861 	vio_msg_t	msg;
1862 	size_t		msglen = sizeof (vio_msg_t);
1863 	int		status;
1864 
1865 	/*
1866 	 * Send an RDX message to vds to indicate we are ready
1867 	 * to send data
1868 	 */
1869 	msg.tag.vio_msgtype = VIO_TYPE_CTRL;
1870 	msg.tag.vio_subtype = VIO_SUBTYPE_INFO;
1871 	msg.tag.vio_subtype_env = VIO_RDX;
1872 	msg.tag.vio_sid = vdcp->session_id;
1873 	status = vdc_send(vdcp, (caddr_t)&msg, &msglen);
1874 	if (status != 0) {
1875 		DMSG(vdcp, 0, "[%d] Failed to send RDX message (%d)",
1876 		    vdcp->instance, status);
1877 	}
1878 
1879 	return (status);
1880 }
1881 
1882 /*
1883  * Function:
1884  *	vdc_handle_rdx()
1885  *
1886  * Description:
1887  *
1888  * Arguments:
1889  *	vdc	- soft state pointer for this instance of the device driver.
1890  *	msgp	- received msg
1891  *
1892  * Return Code:
1893  *	0	- Success
1894  */
1895 static int
1896 vdc_handle_rdx(vdc_t *vdcp, vio_rdx_msg_t *msgp)
1897 {
1898 	_NOTE(ARGUNUSED(vdcp))
1899 	_NOTE(ARGUNUSED(msgp))
1900 
1901 	ASSERT(msgp->tag.vio_msgtype == VIO_TYPE_CTRL);
1902 	ASSERT(msgp->tag.vio_subtype == VIO_SUBTYPE_ACK);
1903 	ASSERT(msgp->tag.vio_subtype_env == VIO_RDX);
1904 
1905 	DMSG(vdcp, 1, "[%d] Got an RDX msg", vdcp->instance);
1906 
1907 	return (0);
1908 }
1909 
1910 /*
1911  * Function:
1912  *	vdc_rdx_exchange()
1913  *
1914  * Description:
1915  *
1916  * Arguments:
1917  *	vdc	- soft state pointer for this instance of the device driver.
1918  *
1919  * Return Code:
1920  *	0	- Success
1921  */
1922 static int
1923 vdc_rdx_exchange(vdc_t *vdcp)
1924 {
1925 	int status;
1926 	vio_msg_t vio_msg;
1927 
1928 	if (status = vdc_send_rdx(vdcp))
1929 		return (status);
1930 
1931 	/* release lock and wait for response */
1932 	mutex_exit(&vdcp->lock);
1933 	status = vdc_wait_for_response(vdcp, &vio_msg);
1934 	mutex_enter(&vdcp->lock);
1935 	if (status) {
1936 		DMSG(vdcp, 0, "[%d] Failed waiting for RDX response, rv(%d)",
1937 		    vdcp->instance, status);
1938 		return (status);
1939 	}
1940 
1941 	/* check type and sub_type ... */
1942 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1943 	    vio_msg.tag.vio_subtype != VIO_SUBTYPE_ACK) {
1944 		DMSG(vdcp, 0, "[%d] Invalid RDX response\n", vdcp->instance);
1945 		return (EPROTO);
1946 	}
1947 
1948 	return (vdc_handle_rdx(vdcp, (vio_rdx_msg_t *)&vio_msg));
1949 }
1950 
1951 
1952 /* -------------------------------------------------------------------------- */
1953 
1954 /*
1955  * LDC helper routines
1956  */
1957 
1958 static int
1959 vdc_recv(vdc_t *vdc, vio_msg_t *msgp, size_t *nbytesp)
1960 {
1961 	int		status;
1962 	uint64_t	delay_time;
1963 	size_t		len;
1964 
1965 	/*
1966 	 * Until we get a blocking ldc read we have to retry until the entire
1967 	 * LDC message has arrived before ldc_read() will return that message.
1968 	 * If ldc_read() succeed but returns a zero length message then that
1969 	 * means that the LDC queue is empty and we have to wait for a
1970 	 * notification from the LDC callback which will set the read_state to
1971 	 * VDC_READ_PENDING. Note we also bail out if the channel is reset or
1972 	 * goes away.
1973 	 */
1974 	delay_time = vdc_ldc_read_init_delay;
1975 
1976 	for (;;) {
1977 
1978 		len = *nbytesp;
1979 		/*
1980 		 * vdc->curr_server is protected by vdc->lock but to avoid
1981 		 * contentions we don't take the lock here. We can do this
1982 		 * safely because vdc_recv() is only called from thread
1983 		 * process_msg_thread() which is also the only thread that
1984 		 * can change vdc->curr_server.
1985 		 */
1986 		status = ldc_read(vdc->curr_server->ldc_handle,
1987 		    (caddr_t)msgp, &len);
1988 
1989 		if (status == EAGAIN) {
1990 			delay_time *= 2;
1991 			if (delay_time >= vdc_ldc_read_max_delay)
1992 				delay_time = vdc_ldc_read_max_delay;
1993 			delay(delay_time);
1994 			continue;
1995 		}
1996 
1997 		if (status != 0) {
1998 			DMSG(vdc, 0, "ldc_read returned %d\n", status);
1999 			break;
2000 		}
2001 
2002 		if (len != 0) {
2003 			*nbytesp = len;
2004 			break;
2005 		}
2006 
2007 		mutex_enter(&vdc->read_lock);
2008 
2009 		while (vdc->read_state != VDC_READ_PENDING) {
2010 
2011 			/* detect if the connection has been reset */
2012 			if (vdc->read_state == VDC_READ_RESET) {
2013 				mutex_exit(&vdc->read_lock);
2014 				return (ECONNRESET);
2015 			}
2016 
2017 			vdc->read_state = VDC_READ_WAITING;
2018 			cv_wait(&vdc->read_cv, &vdc->read_lock);
2019 		}
2020 
2021 		vdc->read_state = VDC_READ_IDLE;
2022 		mutex_exit(&vdc->read_lock);
2023 
2024 		delay_time = vdc_ldc_read_init_delay;
2025 	}
2026 
2027 	return (status);
2028 }
2029 
2030 
2031 
2032 #ifdef DEBUG
2033 void
2034 vdc_decode_tag(vdc_t *vdcp, vio_msg_t *msg)
2035 {
2036 	char *ms, *ss, *ses;
2037 	switch (msg->tag.vio_msgtype) {
2038 #define	Q(_s)	case _s : ms = #_s; break;
2039 	Q(VIO_TYPE_CTRL)
2040 	Q(VIO_TYPE_DATA)
2041 	Q(VIO_TYPE_ERR)
2042 #undef Q
2043 	default: ms = "unknown"; break;
2044 	}
2045 
2046 	switch (msg->tag.vio_subtype) {
2047 #define	Q(_s)	case _s : ss = #_s; break;
2048 	Q(VIO_SUBTYPE_INFO)
2049 	Q(VIO_SUBTYPE_ACK)
2050 	Q(VIO_SUBTYPE_NACK)
2051 #undef Q
2052 	default: ss = "unknown"; break;
2053 	}
2054 
2055 	switch (msg->tag.vio_subtype_env) {
2056 #define	Q(_s)	case _s : ses = #_s; break;
2057 	Q(VIO_VER_INFO)
2058 	Q(VIO_ATTR_INFO)
2059 	Q(VIO_DRING_REG)
2060 	Q(VIO_DRING_UNREG)
2061 	Q(VIO_RDX)
2062 	Q(VIO_PKT_DATA)
2063 	Q(VIO_DESC_DATA)
2064 	Q(VIO_DRING_DATA)
2065 #undef Q
2066 	default: ses = "unknown"; break;
2067 	}
2068 
2069 	DMSG(vdcp, 3, "(%x/%x/%x) message : (%s/%s/%s)\n",
2070 	    msg->tag.vio_msgtype, msg->tag.vio_subtype,
2071 	    msg->tag.vio_subtype_env, ms, ss, ses);
2072 }
2073 #endif
2074 
2075 /*
2076  * Function:
2077  *	vdc_send()
2078  *
2079  * Description:
2080  *	The function encapsulates the call to write a message using LDC.
2081  *	If LDC indicates that the call failed due to the queue being full,
2082  *	we retry the ldc_write(), otherwise we return the error returned by LDC.
2083  *
2084  * Arguments:
2085  *	ldc_handle	- LDC handle for the channel this instance of vdc uses
2086  *	pkt		- address of LDC message to be sent
2087  *	msglen		- the size of the message being sent. When the function
2088  *			  returns, this contains the number of bytes written.
2089  *
2090  * Return Code:
2091  *	0		- Success.
2092  *	EINVAL		- pkt or msglen were NULL
2093  *	ECONNRESET	- The connection was not up.
2094  *	EWOULDBLOCK	- LDC queue is full
2095  *	xxx		- other error codes returned by ldc_write
2096  */
2097 static int
2098 vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen)
2099 {
2100 	size_t	size = 0;
2101 	int	status = 0;
2102 	clock_t delay_ticks;
2103 
2104 	ASSERT(vdc != NULL);
2105 	ASSERT(mutex_owned(&vdc->lock));
2106 	ASSERT(msglen != NULL);
2107 	ASSERT(*msglen != 0);
2108 
2109 #ifdef DEBUG
2110 	vdc_decode_tag(vdc, (vio_msg_t *)(uintptr_t)pkt);
2111 #endif
2112 	/*
2113 	 * Wait indefinitely to send if channel
2114 	 * is busy, but bail out if we succeed or
2115 	 * if the channel closes or is reset.
2116 	 */
2117 	delay_ticks = vdc_hz_min_ldc_delay;
2118 	do {
2119 		size = *msglen;
2120 		status = ldc_write(vdc->curr_server->ldc_handle, pkt, &size);
2121 		if (status == EWOULDBLOCK) {
2122 			delay(delay_ticks);
2123 			/* geometric backoff */
2124 			delay_ticks *= 2;
2125 			if (delay_ticks > vdc_hz_max_ldc_delay)
2126 				delay_ticks = vdc_hz_max_ldc_delay;
2127 		}
2128 	} while (status == EWOULDBLOCK);
2129 
2130 	/* if LDC had serious issues --- reset vdc state */
2131 	if (status == EIO || status == ECONNRESET) {
2132 		/* LDC had serious issues --- reset vdc state */
2133 		mutex_enter(&vdc->read_lock);
2134 		if ((vdc->read_state == VDC_READ_WAITING) ||
2135 		    (vdc->read_state == VDC_READ_RESET))
2136 			cv_signal(&vdc->read_cv);
2137 		vdc->read_state = VDC_READ_RESET;
2138 		mutex_exit(&vdc->read_lock);
2139 
2140 		/* wake up any waiters in the reset thread */
2141 		if (vdc->state == VDC_STATE_INIT_WAITING) {
2142 			DMSG(vdc, 0, "[%d] write reset - "
2143 			    "vdc is resetting ..\n", vdc->instance);
2144 			vdc->state = VDC_STATE_RESETTING;
2145 			cv_signal(&vdc->initwait_cv);
2146 		}
2147 
2148 		return (ECONNRESET);
2149 	}
2150 
2151 	/* return the last size written */
2152 	*msglen = size;
2153 
2154 	return (status);
2155 }
2156 
2157 /*
2158  * Function:
2159  *	vdc_get_md_node
2160  *
2161  * Description:
2162  *	Get the MD, the device node for the given disk instance. The
2163  *	caller is responsible for cleaning up the reference to the
2164  *	returned MD (mdpp) by calling md_fini_handle().
2165  *
2166  * Arguments:
2167  *	dip	- dev info pointer for this instance of the device driver.
2168  *	mdpp	- the returned MD.
2169  *	vd_nodep - the returned device node.
2170  *
2171  * Return Code:
2172  *	0	- Success.
2173  *	ENOENT	- Expected node or property did not exist.
2174  *	ENXIO	- Unexpected error communicating with MD framework
2175  */
2176 static int
2177 vdc_get_md_node(dev_info_t *dip, md_t **mdpp, mde_cookie_t *vd_nodep)
2178 {
2179 	int		status = ENOENT;
2180 	char		*node_name = NULL;
2181 	md_t		*mdp = NULL;
2182 	int		num_nodes;
2183 	int		num_vdevs;
2184 	mde_cookie_t	rootnode;
2185 	mde_cookie_t	*listp = NULL;
2186 	boolean_t	found_inst = B_FALSE;
2187 	int		listsz;
2188 	int		idx;
2189 	uint64_t	md_inst;
2190 	int		obp_inst;
2191 	int		instance = ddi_get_instance(dip);
2192 
2193 	/*
2194 	 * Get the OBP instance number for comparison with the MD instance
2195 	 *
2196 	 * The "cfg-handle" property of a vdc node in an MD contains the MD's
2197 	 * notion of "instance", or unique identifier, for that node; OBP
2198 	 * stores the value of the "cfg-handle" MD property as the value of
2199 	 * the "reg" property on the node in the device tree it builds from
2200 	 * the MD and passes to Solaris.  Thus, we look up the devinfo node's
2201 	 * "reg" property value to uniquely identify this device instance.
2202 	 * If the "reg" property cannot be found, the device tree state is
2203 	 * presumably so broken that there is no point in continuing.
2204 	 */
2205 	if (!ddi_prop_exists(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS, OBP_REG)) {
2206 		cmn_err(CE_WARN, "'%s' property does not exist", OBP_REG);
2207 		return (ENOENT);
2208 	}
2209 	obp_inst = ddi_prop_get_int(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
2210 	    OBP_REG, -1);
2211 	DMSGX(1, "[%d] OBP inst=%d\n", instance, obp_inst);
2212 
2213 	/*
2214 	 * We now walk the MD nodes to find the node for this vdisk.
2215 	 */
2216 	if ((mdp = md_get_handle()) == NULL) {
2217 		cmn_err(CE_WARN, "unable to init machine description");
2218 		return (ENXIO);
2219 	}
2220 
2221 	num_nodes = md_node_count(mdp);
2222 	ASSERT(num_nodes > 0);
2223 
2224 	listsz = num_nodes * sizeof (mde_cookie_t);
2225 
2226 	/* allocate memory for nodes */
2227 	listp = kmem_zalloc(listsz, KM_SLEEP);
2228 
2229 	rootnode = md_root_node(mdp);
2230 	ASSERT(rootnode != MDE_INVAL_ELEM_COOKIE);
2231 
2232 	/*
2233 	 * Search for all the virtual devices, we will then check to see which
2234 	 * ones are disk nodes.
2235 	 */
2236 	num_vdevs = md_scan_dag(mdp, rootnode,
2237 	    md_find_name(mdp, VDC_MD_VDEV_NAME),
2238 	    md_find_name(mdp, "fwd"), listp);
2239 
2240 	if (num_vdevs <= 0) {
2241 		cmn_err(CE_NOTE, "No '%s' node found", VDC_MD_VDEV_NAME);
2242 		status = ENOENT;
2243 		goto done;
2244 	}
2245 
2246 	DMSGX(1, "[%d] num_vdevs=%d\n", instance, num_vdevs);
2247 	for (idx = 0; idx < num_vdevs; idx++) {
2248 		status = md_get_prop_str(mdp, listp[idx], "name", &node_name);
2249 		if ((status != 0) || (node_name == NULL)) {
2250 			cmn_err(CE_NOTE, "Unable to get name of node type '%s'"
2251 			    ": err %d", VDC_MD_VDEV_NAME, status);
2252 			continue;
2253 		}
2254 
2255 		DMSGX(1, "[%d] Found node '%s'\n", instance, node_name);
2256 		if (strcmp(VDC_MD_DISK_NAME, node_name) == 0) {
2257 			status = md_get_prop_val(mdp, listp[idx],
2258 			    VDC_MD_CFG_HDL, &md_inst);
2259 			DMSGX(1, "[%d] vdc inst in MD=%lx\n",
2260 			    instance, md_inst);
2261 			if ((status == 0) && (md_inst == obp_inst)) {
2262 				found_inst = B_TRUE;
2263 				break;
2264 			}
2265 		}
2266 	}
2267 
2268 	if (!found_inst) {
2269 		DMSGX(0, "Unable to find correct '%s' node", VDC_MD_DISK_NAME);
2270 		status = ENOENT;
2271 		goto done;
2272 	}
2273 	DMSGX(0, "[%d] MD inst=%lx\n", instance, md_inst);
2274 
2275 	*vd_nodep = listp[idx];
2276 	*mdpp = mdp;
2277 done:
2278 	kmem_free(listp, listsz);
2279 	return (status);
2280 }
2281 
2282 /*
2283  * Function:
2284  *	vdc_init_ports
2285  *
2286  * Description:
2287  *	Initialize all the ports for this vdisk instance.
2288  *
2289  * Arguments:
2290  *	vdc	- soft state pointer for this instance of the device driver.
2291  *	mdp	- md pointer
2292  *	vd_nodep - device md node.
2293  *
2294  * Return Code:
2295  *	0	- Success.
2296  *	ENOENT	- Expected node or property did not exist.
2297  */
2298 static int
2299 vdc_init_ports(vdc_t *vdc, md_t *mdp, mde_cookie_t vd_nodep)
2300 {
2301 	int		status = 0;
2302 	int		idx;
2303 	int		num_nodes;
2304 	int		num_vports;
2305 	int		num_chans;
2306 	int		listsz;
2307 	mde_cookie_t	vd_port;
2308 	mde_cookie_t	*chanp = NULL;
2309 	mde_cookie_t	*portp = NULL;
2310 	vdc_server_t	*srvr;
2311 	vdc_server_t	*prev_srvr = NULL;
2312 
2313 	/*
2314 	 * We now walk the MD nodes to find the port nodes for this vdisk.
2315 	 */
2316 	num_nodes = md_node_count(mdp);
2317 	ASSERT(num_nodes > 0);
2318 
2319 	listsz = num_nodes * sizeof (mde_cookie_t);
2320 
2321 	/* allocate memory for nodes */
2322 	portp = kmem_zalloc(listsz, KM_SLEEP);
2323 	chanp = kmem_zalloc(listsz, KM_SLEEP);
2324 
2325 	num_vports = md_scan_dag(mdp, vd_nodep,
2326 	    md_find_name(mdp, VDC_MD_PORT_NAME),
2327 	    md_find_name(mdp, "fwd"), portp);
2328 	if (num_vports == 0) {
2329 		DMSGX(0, "Found no '%s' node for '%s' port\n",
2330 		    VDC_MD_PORT_NAME, VDC_MD_VDEV_NAME);
2331 		status = ENOENT;
2332 		goto done;
2333 	}
2334 
2335 	DMSGX(1, "Found %d '%s' node(s) for '%s' port\n",
2336 	    num_vports, VDC_MD_PORT_NAME, VDC_MD_VDEV_NAME);
2337 
2338 	vdc->num_servers = 0;
2339 	for (idx = 0; idx < num_vports; idx++) {
2340 
2341 		/* initialize this port */
2342 		vd_port = portp[idx];
2343 		srvr = kmem_zalloc(sizeof (vdc_server_t), KM_SLEEP);
2344 		srvr->vdcp = vdc;
2345 
2346 		/* get port id */
2347 		if (md_get_prop_val(mdp, vd_port, VDC_MD_ID, &srvr->id) != 0) {
2348 			cmn_err(CE_NOTE, "vDisk port '%s' property not found",
2349 			    VDC_MD_ID);
2350 			kmem_free(srvr, sizeof (vdc_server_t));
2351 			continue;
2352 		}
2353 
2354 		/* set the connection timeout */
2355 		if (md_get_prop_val(mdp, vd_port, VDC_MD_TIMEOUT,
2356 		    &srvr->ctimeout) != 0) {
2357 			srvr->ctimeout = 0;
2358 		}
2359 
2360 		/* get the ldc id */
2361 		num_chans = md_scan_dag(mdp, vd_port,
2362 		    md_find_name(mdp, VDC_MD_CHAN_NAME),
2363 		    md_find_name(mdp, "fwd"), chanp);
2364 
2365 		/* expecting at least one channel */
2366 		if (num_chans <= 0) {
2367 			cmn_err(CE_NOTE, "No '%s' node for '%s' port",
2368 			    VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME);
2369 			kmem_free(srvr, sizeof (vdc_server_t));
2370 			continue;
2371 		} else if (num_chans != 1) {
2372 			DMSGX(0, "Expected 1 '%s' node for '%s' port, "
2373 			    "found %d\n", VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME,
2374 			    num_chans);
2375 		}
2376 
2377 		/*
2378 		 * We use the first channel found (index 0), irrespective of how
2379 		 * many are there in total.
2380 		 */
2381 		if (md_get_prop_val(mdp, chanp[0], VDC_MD_ID,
2382 		    &srvr->ldc_id) != 0) {
2383 			cmn_err(CE_NOTE, "Channel '%s' property not found",
2384 			    VDC_MD_ID);
2385 			kmem_free(srvr, sizeof (vdc_server_t));
2386 			continue;
2387 		}
2388 
2389 		/*
2390 		 * now initialise LDC channel which will be used to
2391 		 * communicate with this server
2392 		 */
2393 		if (vdc_do_ldc_init(vdc, srvr) != 0) {
2394 			kmem_free(srvr, sizeof (vdc_server_t));
2395 			continue;
2396 		}
2397 
2398 		/* add server to list */
2399 		if (prev_srvr)
2400 			prev_srvr->next = srvr;
2401 		else
2402 			vdc->server_list = srvr;
2403 
2404 		prev_srvr = srvr;
2405 
2406 		/* inc numbers of servers */
2407 		vdc->num_servers++;
2408 	}
2409 
2410 	/*
2411 	 * Adjust the max number of handshake retries to match
2412 	 * the number of vdisk servers.
2413 	 */
2414 	if (vdc_hshake_retries < vdc->num_servers)
2415 		vdc_hshake_retries = vdc->num_servers;
2416 
2417 	/* pick first server as current server */
2418 	if (vdc->server_list != NULL) {
2419 		vdc->curr_server = vdc->server_list;
2420 		status = 0;
2421 	} else {
2422 		status = ENOENT;
2423 	}
2424 
2425 done:
2426 	kmem_free(chanp, listsz);
2427 	kmem_free(portp, listsz);
2428 	return (status);
2429 }
2430 
2431 
2432 /*
2433  * Function:
2434  *	vdc_do_ldc_up
2435  *
2436  * Description:
2437  *	Bring the channel for the current server up.
2438  *
2439  * Arguments:
2440  *	vdc	- soft state pointer for this instance of the device driver.
2441  *
2442  * Return Code:
2443  *	0		- Success.
2444  *	EINVAL		- Driver is detaching / LDC error
2445  *	ECONNREFUSED	- Other end is not listening
2446  */
2447 static int
2448 vdc_do_ldc_up(vdc_t *vdc)
2449 {
2450 	int		status;
2451 	ldc_status_t	ldc_state;
2452 
2453 	ASSERT(MUTEX_HELD(&vdc->lock));
2454 
2455 	DMSG(vdc, 0, "[%d] Bringing up channel %lx\n",
2456 	    vdc->instance, vdc->curr_server->ldc_id);
2457 
2458 	if (vdc->lifecycle == VDC_LC_DETACHING)
2459 		return (EINVAL);
2460 
2461 	if ((status = ldc_up(vdc->curr_server->ldc_handle)) != 0) {
2462 		switch (status) {
2463 		case ECONNREFUSED:	/* listener not ready at other end */
2464 			DMSG(vdc, 0, "[%d] ldc_up(%lx,...) return %d\n",
2465 			    vdc->instance, vdc->curr_server->ldc_id, status);
2466 			status = 0;
2467 			break;
2468 		default:
2469 			DMSG(vdc, 0, "[%d] Failed to bring up LDC: "
2470 			    "channel=%ld, err=%d", vdc->instance,
2471 			    vdc->curr_server->ldc_id, status);
2472 			break;
2473 		}
2474 	}
2475 
2476 	if (ldc_status(vdc->curr_server->ldc_handle, &ldc_state) == 0) {
2477 		vdc->curr_server->ldc_state = ldc_state;
2478 		if (ldc_state == LDC_UP) {
2479 			DMSG(vdc, 0, "[%d] LDC channel already up\n",
2480 			    vdc->instance);
2481 			vdc->seq_num = 1;
2482 			vdc->seq_num_reply = 0;
2483 		}
2484 	}
2485 
2486 	return (status);
2487 }
2488 
2489 /*
2490  * Function:
2491  *	vdc_terminate_ldc()
2492  *
2493  * Description:
2494  *
2495  * Arguments:
2496  *	vdc	- soft state pointer for this instance of the device driver.
2497  *	srvr	- vdc per-server info structure
2498  *
2499  * Return Code:
2500  *	None
2501  */
2502 static void
2503 vdc_terminate_ldc(vdc_t *vdc, vdc_server_t *srvr)
2504 {
2505 	int	instance = ddi_get_instance(vdc->dip);
2506 
2507 	if (srvr->state & VDC_LDC_OPEN) {
2508 		DMSG(vdc, 0, "[%d] ldc_close()\n", instance);
2509 		(void) ldc_close(srvr->ldc_handle);
2510 	}
2511 	if (srvr->state & VDC_LDC_CB) {
2512 		DMSG(vdc, 0, "[%d] ldc_unreg_callback()\n", instance);
2513 		(void) ldc_unreg_callback(srvr->ldc_handle);
2514 	}
2515 	if (srvr->state & VDC_LDC_INIT) {
2516 		DMSG(vdc, 0, "[%d] ldc_fini()\n", instance);
2517 		(void) ldc_fini(srvr->ldc_handle);
2518 		srvr->ldc_handle = NULL;
2519 	}
2520 
2521 	srvr->state &= ~(VDC_LDC_INIT | VDC_LDC_CB | VDC_LDC_OPEN);
2522 }
2523 
2524 /*
2525  * Function:
2526  *	vdc_fini_ports()
2527  *
2528  * Description:
2529  *	Finalize all ports by closing the channel associated with each
2530  *	port and also freeing the server structure.
2531  *
2532  * Arguments:
2533  *	vdc	- soft state pointer for this instance of the device driver.
2534  *
2535  * Return Code:
2536  *	None
2537  */
2538 static void
2539 vdc_fini_ports(vdc_t *vdc)
2540 {
2541 	int		instance = ddi_get_instance(vdc->dip);
2542 	vdc_server_t	*srvr, *prev_srvr;
2543 
2544 	ASSERT(vdc != NULL);
2545 	ASSERT(mutex_owned(&vdc->lock));
2546 
2547 	DMSG(vdc, 0, "[%d] initialized=%x\n", instance, vdc->initialized);
2548 
2549 	srvr = vdc->server_list;
2550 
2551 	while (srvr) {
2552 
2553 		vdc_terminate_ldc(vdc, srvr);
2554 
2555 		/* next server */
2556 		prev_srvr = srvr;
2557 		srvr = srvr->next;
2558 
2559 		/* free server */
2560 		kmem_free(prev_srvr, sizeof (vdc_server_t));
2561 	}
2562 
2563 	vdc->server_list = NULL;
2564 }
2565 
2566 /* -------------------------------------------------------------------------- */
2567 
2568 /*
2569  * Descriptor Ring helper routines
2570  */
2571 
2572 /*
2573  * Function:
2574  *	vdc_init_descriptor_ring()
2575  *
2576  * Description:
2577  *
2578  * Arguments:
2579  *	vdc	- soft state pointer for this instance of the device driver.
2580  *
2581  * Return Code:
2582  *	0	- Success
2583  */
2584 static int
2585 vdc_init_descriptor_ring(vdc_t *vdc)
2586 {
2587 	vd_dring_entry_t	*dep = NULL;	/* DRing Entry pointer */
2588 	int	status = 0;
2589 	int	i;
2590 
2591 	DMSG(vdc, 0, "[%d] initialized=%x\n", vdc->instance, vdc->initialized);
2592 
2593 	ASSERT(vdc != NULL);
2594 	ASSERT(mutex_owned(&vdc->lock));
2595 
2596 	/* ensure we have enough room to store max sized block */
2597 	ASSERT(maxphys <= VD_MAX_BLOCK_SIZE);
2598 
2599 	if ((vdc->initialized & VDC_DRING_INIT) == 0) {
2600 		DMSG(vdc, 0, "[%d] ldc_mem_dring_create\n", vdc->instance);
2601 		/*
2602 		 * Calculate the maximum block size we can transmit using one
2603 		 * Descriptor Ring entry from the attributes returned by the
2604 		 * vDisk server. This is subject to a minimum of 'maxphys'
2605 		 * as we do not have the capability to split requests over
2606 		 * multiple DRing entries.
2607 		 */
2608 		if ((vdc->max_xfer_sz * vdc->block_size) < maxphys) {
2609 			DMSG(vdc, 0, "[%d] using minimum DRing size\n",
2610 			    vdc->instance);
2611 			vdc->dring_max_cookies = maxphys / PAGESIZE;
2612 		} else {
2613 			vdc->dring_max_cookies =
2614 			    (vdc->max_xfer_sz * vdc->block_size) / PAGESIZE;
2615 		}
2616 		vdc->dring_entry_size = (sizeof (vd_dring_entry_t) +
2617 		    (sizeof (ldc_mem_cookie_t) *
2618 		    (vdc->dring_max_cookies - 1)));
2619 		vdc->dring_len = VD_DRING_LEN;
2620 
2621 		status = ldc_mem_dring_create(vdc->dring_len,
2622 		    vdc->dring_entry_size, &vdc->dring_hdl);
2623 		if ((vdc->dring_hdl == NULL) || (status != 0)) {
2624 			DMSG(vdc, 0, "[%d] Descriptor ring creation failed",
2625 			    vdc->instance);
2626 			return (status);
2627 		}
2628 		vdc->initialized |= VDC_DRING_INIT;
2629 	}
2630 
2631 	if ((vdc->initialized & VDC_DRING_BOUND) == 0) {
2632 		DMSG(vdc, 0, "[%d] ldc_mem_dring_bind\n", vdc->instance);
2633 		vdc->dring_cookie =
2634 		    kmem_zalloc(sizeof (ldc_mem_cookie_t), KM_SLEEP);
2635 
2636 		status = ldc_mem_dring_bind(vdc->curr_server->ldc_handle,
2637 		    vdc->dring_hdl,
2638 		    LDC_SHADOW_MAP|LDC_DIRECT_MAP, LDC_MEM_RW,
2639 		    &vdc->dring_cookie[0],
2640 		    &vdc->dring_cookie_count);
2641 		if (status != 0) {
2642 			DMSG(vdc, 0, "[%d] Failed to bind descriptor ring "
2643 			    "(%lx) to channel (%lx) status=%d\n",
2644 			    vdc->instance, vdc->dring_hdl,
2645 			    vdc->curr_server->ldc_handle, status);
2646 			return (status);
2647 		}
2648 		ASSERT(vdc->dring_cookie_count == 1);
2649 		vdc->initialized |= VDC_DRING_BOUND;
2650 	}
2651 
2652 	status = ldc_mem_dring_info(vdc->dring_hdl, &vdc->dring_mem_info);
2653 	if (status != 0) {
2654 		DMSG(vdc, 0,
2655 		    "[%d] Failed to get info for descriptor ring (%lx)\n",
2656 		    vdc->instance, vdc->dring_hdl);
2657 		return (status);
2658 	}
2659 
2660 	if ((vdc->initialized & VDC_DRING_LOCAL) == 0) {
2661 		DMSG(vdc, 0, "[%d] local dring\n", vdc->instance);
2662 
2663 		/* Allocate the local copy of this dring */
2664 		vdc->local_dring =
2665 		    kmem_zalloc(vdc->dring_len * sizeof (vdc_local_desc_t),
2666 		    KM_SLEEP);
2667 		vdc->initialized |= VDC_DRING_LOCAL;
2668 	}
2669 
2670 	/*
2671 	 * Mark all DRing entries as free and initialize the private
2672 	 * descriptor's memory handles. If any entry is initialized,
2673 	 * we need to free it later so we set the bit in 'initialized'
2674 	 * at the start.
2675 	 */
2676 	vdc->initialized |= VDC_DRING_ENTRY;
2677 	for (i = 0; i < vdc->dring_len; i++) {
2678 		dep = VDC_GET_DRING_ENTRY_PTR(vdc, i);
2679 		dep->hdr.dstate = VIO_DESC_FREE;
2680 
2681 		status = ldc_mem_alloc_handle(vdc->curr_server->ldc_handle,
2682 		    &vdc->local_dring[i].desc_mhdl);
2683 		if (status != 0) {
2684 			DMSG(vdc, 0, "![%d] Failed to alloc mem handle for"
2685 			    " descriptor %d", vdc->instance, i);
2686 			return (status);
2687 		}
2688 		vdc->local_dring[i].is_free = B_TRUE;
2689 		vdc->local_dring[i].dep = dep;
2690 	}
2691 
2692 	/* Initialize the starting index */
2693 	vdc->dring_curr_idx = 0;
2694 
2695 	return (status);
2696 }
2697 
2698 /*
2699  * Function:
2700  *	vdc_destroy_descriptor_ring()
2701  *
2702  * Description:
2703  *
2704  * Arguments:
2705  *	vdc	- soft state pointer for this instance of the device driver.
2706  *
2707  * Return Code:
2708  *	None
2709  */
2710 static void
2711 vdc_destroy_descriptor_ring(vdc_t *vdc)
2712 {
2713 	vdc_local_desc_t	*ldep = NULL;	/* Local Dring Entry Pointer */
2714 	ldc_mem_handle_t	mhdl = NULL;
2715 	ldc_mem_info_t		minfo;
2716 	int			status = -1;
2717 	int			i;	/* loop */
2718 
2719 	ASSERT(vdc != NULL);
2720 	ASSERT(mutex_owned(&vdc->lock));
2721 
2722 	DMSG(vdc, 0, "[%d] Entered\n", vdc->instance);
2723 
2724 	if (vdc->initialized & VDC_DRING_ENTRY) {
2725 		DMSG(vdc, 0,
2726 		    "[%d] Removing Local DRing entries\n", vdc->instance);
2727 		for (i = 0; i < vdc->dring_len; i++) {
2728 			ldep = &vdc->local_dring[i];
2729 			mhdl = ldep->desc_mhdl;
2730 
2731 			if (mhdl == NULL)
2732 				continue;
2733 
2734 			if ((status = ldc_mem_info(mhdl, &minfo)) != 0) {
2735 				DMSG(vdc, 0,
2736 				    "ldc_mem_info returned an error: %d\n",
2737 				    status);
2738 
2739 				/*
2740 				 * This must mean that the mem handle
2741 				 * is not valid. Clear it out so that
2742 				 * no one tries to use it.
2743 				 */
2744 				ldep->desc_mhdl = NULL;
2745 				continue;
2746 			}
2747 
2748 			if (minfo.status == LDC_BOUND) {
2749 				(void) ldc_mem_unbind_handle(mhdl);
2750 			}
2751 
2752 			(void) ldc_mem_free_handle(mhdl);
2753 
2754 			ldep->desc_mhdl = NULL;
2755 		}
2756 		vdc->initialized &= ~VDC_DRING_ENTRY;
2757 	}
2758 
2759 	if (vdc->initialized & VDC_DRING_LOCAL) {
2760 		DMSG(vdc, 0, "[%d] Freeing Local DRing\n", vdc->instance);
2761 		kmem_free(vdc->local_dring,
2762 		    vdc->dring_len * sizeof (vdc_local_desc_t));
2763 		vdc->initialized &= ~VDC_DRING_LOCAL;
2764 	}
2765 
2766 	if (vdc->initialized & VDC_DRING_BOUND) {
2767 		DMSG(vdc, 0, "[%d] Unbinding DRing\n", vdc->instance);
2768 		status = ldc_mem_dring_unbind(vdc->dring_hdl);
2769 		if (status == 0) {
2770 			vdc->initialized &= ~VDC_DRING_BOUND;
2771 		} else {
2772 			DMSG(vdc, 0, "[%d] Error %d unbinding DRing %lx",
2773 			    vdc->instance, status, vdc->dring_hdl);
2774 		}
2775 		kmem_free(vdc->dring_cookie, sizeof (ldc_mem_cookie_t));
2776 	}
2777 
2778 	if (vdc->initialized & VDC_DRING_INIT) {
2779 		DMSG(vdc, 0, "[%d] Destroying DRing\n", vdc->instance);
2780 		status = ldc_mem_dring_destroy(vdc->dring_hdl);
2781 		if (status == 0) {
2782 			vdc->dring_hdl = NULL;
2783 			bzero(&vdc->dring_mem_info, sizeof (ldc_mem_info_t));
2784 			vdc->initialized &= ~VDC_DRING_INIT;
2785 		} else {
2786 			DMSG(vdc, 0, "[%d] Error %d destroying DRing (%lx)",
2787 			    vdc->instance, status, vdc->dring_hdl);
2788 		}
2789 	}
2790 }
2791 
2792 /*
2793  * Function:
2794  *	vdc_map_to_shared_dring()
2795  *
2796  * Description:
2797  *	Copy contents of the local descriptor to the shared
2798  *	memory descriptor.
2799  *
2800  * Arguments:
2801  *	vdcp	- soft state pointer for this instance of the device driver.
2802  *	idx	- descriptor ring index
2803  *
2804  * Return Code:
2805  *	None
2806  */
2807 static int
2808 vdc_map_to_shared_dring(vdc_t *vdcp, int idx)
2809 {
2810 	vdc_local_desc_t	*ldep;
2811 	vd_dring_entry_t	*dep;
2812 	int			rv;
2813 
2814 	ldep = &(vdcp->local_dring[idx]);
2815 
2816 	/* for now leave in the old pop_mem_hdl stuff */
2817 	if (ldep->nbytes > 0) {
2818 		rv = vdc_populate_mem_hdl(vdcp, ldep);
2819 		if (rv) {
2820 			DMSG(vdcp, 0, "[%d] Cannot populate mem handle\n",
2821 			    vdcp->instance);
2822 			return (rv);
2823 		}
2824 	}
2825 
2826 	/*
2827 	 * fill in the data details into the DRing
2828 	 */
2829 	dep = ldep->dep;
2830 	ASSERT(dep != NULL);
2831 
2832 	dep->payload.req_id = VDC_GET_NEXT_REQ_ID(vdcp);
2833 	dep->payload.operation = ldep->operation;
2834 	dep->payload.addr = ldep->offset;
2835 	dep->payload.nbytes = ldep->nbytes;
2836 	dep->payload.status = (uint32_t)-1;	/* vds will set valid value */
2837 	dep->payload.slice = ldep->slice;
2838 	dep->hdr.dstate = VIO_DESC_READY;
2839 	dep->hdr.ack = 1;		/* request an ACK for every message */
2840 
2841 	return (0);
2842 }
2843 
2844 /*
2845  * Function:
2846  *	vdc_send_request
2847  *
2848  * Description:
2849  *	This routine writes the data to be transmitted to vds into the
2850  *	descriptor, notifies vds that the ring has been updated and
2851  *	then waits for the request to be processed.
2852  *
2853  * Arguments:
2854  *	vdcp	  - the soft state pointer
2855  *	operation - operation we want vds to perform (VD_OP_XXX)
2856  *	addr	  - address of data buf to be read/written.
2857  *	nbytes	  - number of bytes to read/write
2858  *	slice	  - the disk slice this request is for
2859  *	offset	  - relative disk offset
2860  *	cb_type   - type of call - STRATEGY or SYNC
2861  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
2862  *			. mode for ioctl(9e)
2863  *			. LP64 diskaddr_t (block I/O)
2864  *	dir	  - direction of operation (READ/WRITE/BOTH)
2865  *
2866  * Return Codes:
2867  *	0
2868  *	ENXIO
2869  */
2870 static int
2871 vdc_send_request(vdc_t *vdcp, int operation, caddr_t addr,
2872     size_t nbytes, int slice, diskaddr_t offset, int cb_type,
2873     void *cb_arg, vio_desc_direction_t dir)
2874 {
2875 	int	rv = 0;
2876 
2877 	ASSERT(vdcp != NULL);
2878 	ASSERT(slice == VD_SLICE_NONE || slice < V_NUMPAR);
2879 
2880 	mutex_enter(&vdcp->lock);
2881 
2882 	/*
2883 	 * If this is a block read/write operation we update the I/O statistics
2884 	 * to indicate that the request is being put on the waitq to be
2885 	 * serviced.
2886 	 *
2887 	 * We do it here (a common routine for both synchronous and strategy
2888 	 * calls) for performance reasons - we are already holding vdc->lock
2889 	 * so there is no extra locking overhead. We would have to explicitly
2890 	 * grab the 'lock' mutex to update the stats if we were to do this
2891 	 * higher up the stack in vdc_strategy() et. al.
2892 	 */
2893 	if ((operation == VD_OP_BREAD) || (operation == VD_OP_BWRITE)) {
2894 		DTRACE_IO1(start, buf_t *, cb_arg);
2895 		VD_KSTAT_WAITQ_ENTER(vdcp);
2896 	}
2897 
2898 	do {
2899 		while (vdcp->state != VDC_STATE_RUNNING) {
2900 
2901 			/* return error if detaching */
2902 			if (vdcp->state == VDC_STATE_DETACH) {
2903 				rv = ENXIO;
2904 				goto done;
2905 			}
2906 
2907 			/* fail request if connection timeout is reached */
2908 			if (vdcp->ctimeout_reached) {
2909 				rv = EIO;
2910 				goto done;
2911 			}
2912 
2913 			/*
2914 			 * If we are panicking and the disk is not ready then
2915 			 * we can't send any request because we can't complete
2916 			 * the handshake now.
2917 			 */
2918 			if (ddi_in_panic()) {
2919 				rv = EIO;
2920 				goto done;
2921 			}
2922 
2923 			cv_wait(&vdcp->running_cv, &vdcp->lock);
2924 		}
2925 
2926 	} while (vdc_populate_descriptor(vdcp, operation, addr,
2927 	    nbytes, slice, offset, cb_type, cb_arg, dir));
2928 
2929 done:
2930 	/*
2931 	 * If this is a block read/write we update the I/O statistics kstat
2932 	 * to indicate that this request has been placed on the queue for
2933 	 * processing (i.e sent to the vDisk server) - iostat(1M) will
2934 	 * report the time waiting for the vDisk server under the %b column
2935 	 * In the case of an error we simply take it off the wait queue.
2936 	 */
2937 	if ((operation == VD_OP_BREAD) || (operation == VD_OP_BWRITE)) {
2938 		if (rv == 0) {
2939 			VD_KSTAT_WAITQ_TO_RUNQ(vdcp);
2940 			DTRACE_PROBE1(send, buf_t *, cb_arg);
2941 		} else {
2942 			VD_UPDATE_ERR_STATS(vdcp, vd_transerrs);
2943 			VD_KSTAT_WAITQ_EXIT(vdcp);
2944 			DTRACE_IO1(done, buf_t *, cb_arg);
2945 		}
2946 	}
2947 
2948 	mutex_exit(&vdcp->lock);
2949 
2950 	return (rv);
2951 }
2952 
2953 
2954 /*
2955  * Function:
2956  *	vdc_populate_descriptor
2957  *
2958  * Description:
2959  *	This routine writes the data to be transmitted to vds into the
2960  *	descriptor, notifies vds that the ring has been updated and
2961  *	then waits for the request to be processed.
2962  *
2963  * Arguments:
2964  *	vdcp	  - the soft state pointer
2965  *	operation - operation we want vds to perform (VD_OP_XXX)
2966  *	addr	  - address of data buf to be read/written.
2967  *	nbytes	  - number of bytes to read/write
2968  *	slice	  - the disk slice this request is for
2969  *	offset	  - relative disk offset
2970  *	cb_type   - type of call - STRATEGY or SYNC
2971  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
2972  *			. mode for ioctl(9e)
2973  *			. LP64 diskaddr_t (block I/O)
2974  *	dir	  - direction of operation (READ/WRITE/BOTH)
2975  *
2976  * Return Codes:
2977  *	0
2978  *	EAGAIN
2979  *	ECONNRESET
2980  *	ENXIO
2981  */
2982 static int
2983 vdc_populate_descriptor(vdc_t *vdcp, int operation, caddr_t addr,
2984     size_t nbytes, int slice, diskaddr_t offset, int cb_type,
2985     void *cb_arg, vio_desc_direction_t dir)
2986 {
2987 	vdc_local_desc_t	*local_dep = NULL; /* Local Dring Pointer */
2988 	int			idx;		/* Index of DRing entry used */
2989 	int			next_idx;
2990 	vio_dring_msg_t		dmsg;
2991 	size_t			msglen;
2992 	int			rv;
2993 
2994 	ASSERT(MUTEX_HELD(&vdcp->lock));
2995 	vdcp->threads_pending++;
2996 loop:
2997 	DMSG(vdcp, 2, ": dring_curr_idx = %d\n", vdcp->dring_curr_idx);
2998 
2999 	/* Get next available D-Ring entry */
3000 	idx = vdcp->dring_curr_idx;
3001 	local_dep = &(vdcp->local_dring[idx]);
3002 
3003 	if (!local_dep->is_free) {
3004 		DMSG(vdcp, 2, "[%d]: dring full - waiting for space\n",
3005 		    vdcp->instance);
3006 		cv_wait(&vdcp->dring_free_cv, &vdcp->lock);
3007 		if (vdcp->state == VDC_STATE_RUNNING ||
3008 		    vdcp->state == VDC_STATE_HANDLE_PENDING) {
3009 			goto loop;
3010 		}
3011 		vdcp->threads_pending--;
3012 		return (ECONNRESET);
3013 	}
3014 
3015 	next_idx = idx + 1;
3016 	if (next_idx >= vdcp->dring_len)
3017 		next_idx = 0;
3018 	vdcp->dring_curr_idx = next_idx;
3019 
3020 	ASSERT(local_dep->is_free);
3021 
3022 	local_dep->operation = operation;
3023 	local_dep->addr = addr;
3024 	local_dep->nbytes = nbytes;
3025 	local_dep->slice = slice;
3026 	local_dep->offset = offset;
3027 	local_dep->cb_type = cb_type;
3028 	local_dep->cb_arg = cb_arg;
3029 	local_dep->dir = dir;
3030 
3031 	local_dep->is_free = B_FALSE;
3032 
3033 	rv = vdc_map_to_shared_dring(vdcp, idx);
3034 	if (rv) {
3035 		DMSG(vdcp, 0, "[%d]: cannot bind memory - waiting ..\n",
3036 		    vdcp->instance);
3037 		/* free the descriptor */
3038 		local_dep->is_free = B_TRUE;
3039 		vdcp->dring_curr_idx = idx;
3040 		cv_wait(&vdcp->membind_cv, &vdcp->lock);
3041 		if (vdcp->state == VDC_STATE_RUNNING ||
3042 		    vdcp->state == VDC_STATE_HANDLE_PENDING) {
3043 			goto loop;
3044 		}
3045 		vdcp->threads_pending--;
3046 		return (ECONNRESET);
3047 	}
3048 
3049 	/*
3050 	 * Send a msg with the DRing details to vds
3051 	 */
3052 	VIO_INIT_DRING_DATA_TAG(dmsg);
3053 	VDC_INIT_DRING_DATA_MSG_IDS(dmsg, vdcp);
3054 	dmsg.dring_ident = vdcp->dring_ident;
3055 	dmsg.start_idx = idx;
3056 	dmsg.end_idx = idx;
3057 	vdcp->seq_num++;
3058 
3059 	DTRACE_PROBE2(populate, int, vdcp->instance,
3060 	    vdc_local_desc_t *, local_dep);
3061 	DMSG(vdcp, 2, "ident=0x%lx, st=%u, end=%u, seq=%ld\n",
3062 	    vdcp->dring_ident, dmsg.start_idx, dmsg.end_idx, dmsg.seq_num);
3063 
3064 	/*
3065 	 * note we're still holding the lock here to
3066 	 * make sure the message goes out in order !!!...
3067 	 */
3068 	msglen = sizeof (dmsg);
3069 	rv = vdc_send(vdcp, (caddr_t)&dmsg, &msglen);
3070 	switch (rv) {
3071 	case ECONNRESET:
3072 		/*
3073 		 * vdc_send initiates the reset on failure.
3074 		 * Since the transaction has already been put
3075 		 * on the local dring, it will automatically get
3076 		 * retried when the channel is reset. Given that,
3077 		 * it is ok to just return success even though the
3078 		 * send failed.
3079 		 */
3080 		rv = 0;
3081 		break;
3082 
3083 	case 0: /* EOK */
3084 		DMSG(vdcp, 1, "sent via LDC: rv=%d\n", rv);
3085 		break;
3086 
3087 	default:
3088 		goto cleanup_and_exit;
3089 	}
3090 
3091 	vdcp->threads_pending--;
3092 	return (rv);
3093 
3094 cleanup_and_exit:
3095 	DMSG(vdcp, 0, "unexpected error, rv=%d\n", rv);
3096 	return (ENXIO);
3097 }
3098 
3099 /*
3100  * Function:
3101  *	vdc_do_sync_op
3102  *
3103  * Description:
3104  * 	Wrapper around vdc_populate_descriptor that blocks until the
3105  * 	response to the message is available.
3106  *
3107  * Arguments:
3108  *	vdcp	  - the soft state pointer
3109  *	operation - operation we want vds to perform (VD_OP_XXX)
3110  *	addr	  - address of data buf to be read/written.
3111  *	nbytes	  - number of bytes to read/write
3112  *	slice	  - the disk slice this request is for
3113  *	offset	  - relative disk offset
3114  *	cb_type   - type of call - STRATEGY or SYNC
3115  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
3116  *			. mode for ioctl(9e)
3117  *			. LP64 diskaddr_t (block I/O)
3118  *	dir	  - direction of operation (READ/WRITE/BOTH)
3119  *	rconflict - check for reservation conflict in case of failure
3120  *
3121  * rconflict should be set to B_TRUE by most callers. Callers invoking the
3122  * VD_OP_SCSICMD operation can set rconflict to B_FALSE if they check the
3123  * result of a successful operation with vd_scsi_status().
3124  *
3125  * Return Codes:
3126  *	0
3127  *	EAGAIN
3128  *	EFAULT
3129  *	ENXIO
3130  *	EIO
3131  */
3132 static int
3133 vdc_do_sync_op(vdc_t *vdcp, int operation, caddr_t addr, size_t nbytes,
3134     int slice, diskaddr_t offset, int cb_type, void *cb_arg,
3135     vio_desc_direction_t dir, boolean_t rconflict)
3136 {
3137 	int status;
3138 	vdc_io_t *vio;
3139 	boolean_t check_resv_conflict = B_FALSE;
3140 
3141 	ASSERT(cb_type == CB_SYNC);
3142 
3143 	/*
3144 	 * Grab the lock, if blocked wait until the server
3145 	 * response causes us to wake up again.
3146 	 */
3147 	mutex_enter(&vdcp->lock);
3148 	vdcp->sync_op_cnt++;
3149 	while (vdcp->sync_op_blocked && vdcp->state != VDC_STATE_DETACH) {
3150 		if (ddi_in_panic()) {
3151 			/* don't block if we are panicking */
3152 			vdcp->sync_op_cnt--;
3153 			mutex_exit(&vdcp->lock);
3154 			return (EIO);
3155 		} else {
3156 			cv_wait(&vdcp->sync_blocked_cv, &vdcp->lock);
3157 		}
3158 	}
3159 
3160 	if (vdcp->state == VDC_STATE_DETACH) {
3161 		cv_broadcast(&vdcp->sync_blocked_cv);
3162 		vdcp->sync_op_cnt--;
3163 		mutex_exit(&vdcp->lock);
3164 		return (ENXIO);
3165 	}
3166 
3167 	/* now block anyone other thread entering after us */
3168 	vdcp->sync_op_blocked = B_TRUE;
3169 	vdcp->sync_op_pending = B_TRUE;
3170 	mutex_exit(&vdcp->lock);
3171 
3172 	status = vdc_send_request(vdcp, operation, addr,
3173 	    nbytes, slice, offset, cb_type, cb_arg, dir);
3174 
3175 	mutex_enter(&vdcp->lock);
3176 
3177 	if (status != 0) {
3178 		vdcp->sync_op_pending = B_FALSE;
3179 	} else if (ddi_in_panic()) {
3180 		if (vdc_drain_response(vdcp, CB_SYNC, NULL) == 0) {
3181 			status = vdcp->sync_op_status;
3182 		} else {
3183 			vdcp->sync_op_pending = B_FALSE;
3184 			status = EIO;
3185 		}
3186 	} else {
3187 		/*
3188 		 * block until our transaction completes.
3189 		 * Also anyone else waiting also gets to go next.
3190 		 */
3191 		while (vdcp->sync_op_pending && vdcp->state != VDC_STATE_DETACH)
3192 			cv_wait(&vdcp->sync_pending_cv, &vdcp->lock);
3193 
3194 		DMSG(vdcp, 2, ": operation returned %d\n",
3195 		    vdcp->sync_op_status);
3196 		if (vdcp->state == VDC_STATE_DETACH) {
3197 			vdcp->sync_op_pending = B_FALSE;
3198 			status = ENXIO;
3199 		} else {
3200 			status = vdcp->sync_op_status;
3201 			if (status != 0 && vdcp->failfast_interval != 0) {
3202 				/*
3203 				 * Operation has failed and failfast is enabled.
3204 				 * We need to check if the failure is due to a
3205 				 * reservation conflict if this was requested.
3206 				 */
3207 				check_resv_conflict = rconflict;
3208 			}
3209 
3210 		}
3211 	}
3212 
3213 	vdcp->sync_op_status = 0;
3214 	vdcp->sync_op_blocked = B_FALSE;
3215 	vdcp->sync_op_cnt--;
3216 
3217 	/* signal the next waiting thread */
3218 	cv_signal(&vdcp->sync_blocked_cv);
3219 
3220 	/*
3221 	 * We have to check for reservation conflict after unblocking sync
3222 	 * operations because some sync operations will be used to do this
3223 	 * check.
3224 	 */
3225 	if (check_resv_conflict) {
3226 		vio = vdc_failfast_io_queue(vdcp, NULL);
3227 		while (vio->vio_qtime != 0)
3228 			cv_wait(&vdcp->failfast_io_cv, &vdcp->lock);
3229 		kmem_free(vio, sizeof (vdc_io_t));
3230 	}
3231 
3232 	mutex_exit(&vdcp->lock);
3233 
3234 	return (status);
3235 }
3236 
3237 
3238 /*
3239  * Function:
3240  *	vdc_drain_response()
3241  *
3242  * Description:
3243  * 	When a guest is panicking, the completion of requests needs to be
3244  * 	handled differently because interrupts are disabled and vdc
3245  * 	will not get messages. We have to poll for the messages instead.
3246  *
3247  *	Note: since we are panicking we don't implement	the io:::done
3248  *	DTrace probe or update the I/O statistics kstats.
3249  *
3250  * Arguments:
3251  *	vdc	- soft state pointer for this instance of the device driver.
3252  *	cb_type	- the type of request we want to drain. If type is CB_SYNC
3253  *		  then we drain all responses until we find a CB_SYNC request.
3254  *		  If the type is CB_STRATEGY then the behavior depends on the
3255  *		  value of the buf argument.
3256  *	buf	- if the cb_type argument is CB_SYNC then the buf argument
3257  *		  must be NULL. If the cb_type argument is CB_STRATEGY and
3258  *		  if buf is NULL then we drain all responses, otherwise we
3259  *		  poll until we receive a ACK/NACK for the specific I/O
3260  *		  described by buf.
3261  *
3262  * Return Code:
3263  *	0	- Success. If we were expecting a response to a particular
3264  *		  CB_SYNC or CB_STRATEGY request then this means that a
3265  *		  response has been received.
3266  */
3267 static int
3268 vdc_drain_response(vdc_t *vdc, vio_cb_type_t cb_type, struct buf *buf)
3269 {
3270 	int 			rv, idx, retries;
3271 	size_t			msglen;
3272 	vdc_local_desc_t 	*ldep = NULL;	/* Local Dring Entry Pointer */
3273 	vio_dring_msg_t		dmsg;
3274 	struct buf		*mbuf;
3275 	boolean_t		ack;
3276 
3277 	ASSERT(cb_type == CB_STRATEGY || cb_type == CB_SYNC);
3278 
3279 	mutex_enter(&vdc->lock);
3280 
3281 	retries = 0;
3282 	for (;;) {
3283 		msglen = sizeof (dmsg);
3284 		rv = ldc_read(vdc->curr_server->ldc_handle, (caddr_t)&dmsg,
3285 		    &msglen);
3286 		if (rv) {
3287 			rv = EINVAL;
3288 			break;
3289 		}
3290 
3291 		/*
3292 		 * if there are no packets wait and check again
3293 		 */
3294 		if ((rv == 0) && (msglen == 0)) {
3295 			if (retries++ > vdc_dump_retries) {
3296 				rv = EAGAIN;
3297 				break;
3298 			}
3299 
3300 			drv_usecwait(vdc_usec_timeout_dump);
3301 			continue;
3302 		}
3303 
3304 		/*
3305 		 * Ignore all messages that are not ACKs/NACKs to
3306 		 * DRing requests.
3307 		 */
3308 		if ((dmsg.tag.vio_msgtype != VIO_TYPE_DATA) ||
3309 		    (dmsg.tag.vio_subtype_env != VIO_DRING_DATA)) {
3310 			DMSG(vdc, 0, "discard pkt: type=%d sub=%d env=%d\n",
3311 			    dmsg.tag.vio_msgtype,
3312 			    dmsg.tag.vio_subtype,
3313 			    dmsg.tag.vio_subtype_env);
3314 			continue;
3315 		}
3316 
3317 		/*
3318 		 * Record if the packet was ACK'ed or not. If the packet was not
3319 		 * ACK'ed then we will just mark the request as failed; we don't
3320 		 * want to reset the connection at this point.
3321 		 */
3322 		switch (dmsg.tag.vio_subtype) {
3323 		case VIO_SUBTYPE_ACK:
3324 			ack = B_TRUE;
3325 			break;
3326 		case VIO_SUBTYPE_NACK:
3327 			ack = B_FALSE;
3328 			break;
3329 		default:
3330 			continue;
3331 		}
3332 
3333 		idx = dmsg.start_idx;
3334 		if (idx >= vdc->dring_len) {
3335 			DMSG(vdc, 0, "[%d] Bogus ack data : start %d\n",
3336 			    vdc->instance, idx);
3337 			continue;
3338 		}
3339 		ldep = &vdc->local_dring[idx];
3340 		if (ldep->dep->hdr.dstate != VIO_DESC_DONE) {
3341 			DMSG(vdc, 0, "[%d] Entry @ %d - state !DONE %d\n",
3342 			    vdc->instance, idx, ldep->dep->hdr.dstate);
3343 			continue;
3344 		}
3345 
3346 		switch (ldep->cb_type) {
3347 
3348 		case CB_STRATEGY:
3349 			mbuf = ldep->cb_arg;
3350 			if (mbuf != NULL) {
3351 				mbuf->b_resid = mbuf->b_bcount -
3352 				    ldep->dep->payload.nbytes;
3353 				bioerror(mbuf,
3354 				    ack ? ldep->dep->payload.status : EIO);
3355 				biodone(mbuf);
3356 			}
3357 			rv = vdc_depopulate_descriptor(vdc, idx);
3358 			if (buf != NULL && buf == mbuf) {
3359 				rv = 0;
3360 				goto done;
3361 			}
3362 			break;
3363 
3364 		case CB_SYNC:
3365 			rv = vdc_depopulate_descriptor(vdc, idx);
3366 			vdc->sync_op_status = ack ? rv : EIO;
3367 			vdc->sync_op_pending = B_FALSE;
3368 			cv_signal(&vdc->sync_pending_cv);
3369 			if (cb_type == CB_SYNC) {
3370 				rv = 0;
3371 				goto done;
3372 			}
3373 			break;
3374 		}
3375 
3376 		/* if this is the last descriptor - break out of loop */
3377 		if ((idx + 1) % vdc->dring_len == vdc->dring_curr_idx) {
3378 			/*
3379 			 * If we were expecting a response for a particular
3380 			 * request then we return with an error otherwise we
3381 			 * have successfully completed the drain.
3382 			 */
3383 			rv = (buf != NULL || cb_type == CB_SYNC)? ESRCH: 0;
3384 			break;
3385 		}
3386 	}
3387 
3388 done:
3389 	mutex_exit(&vdc->lock);
3390 	DMSG(vdc, 0, "End idx=%d\n", idx);
3391 
3392 	return (rv);
3393 }
3394 
3395 
3396 /*
3397  * Function:
3398  *	vdc_depopulate_descriptor()
3399  *
3400  * Description:
3401  *
3402  * Arguments:
3403  *	vdc	- soft state pointer for this instance of the device driver.
3404  *	idx	- Index of the Descriptor Ring entry being modified
3405  *
3406  * Return Code:
3407  *	0	- Success
3408  */
3409 static int
3410 vdc_depopulate_descriptor(vdc_t *vdc, uint_t idx)
3411 {
3412 	vd_dring_entry_t *dep = NULL;		/* Dring Entry Pointer */
3413 	vdc_local_desc_t *ldep = NULL;		/* Local Dring Entry Pointer */
3414 	int		status = ENXIO;
3415 	int		rv = 0;
3416 
3417 	ASSERT(vdc != NULL);
3418 	ASSERT(idx < vdc->dring_len);
3419 	ldep = &vdc->local_dring[idx];
3420 	ASSERT(ldep != NULL);
3421 	ASSERT(MUTEX_HELD(&vdc->lock));
3422 
3423 	DTRACE_PROBE2(depopulate, int, vdc->instance, vdc_local_desc_t *, ldep);
3424 	DMSG(vdc, 2, ": idx = %d\n", idx);
3425 
3426 	dep = ldep->dep;
3427 	ASSERT(dep != NULL);
3428 	ASSERT((dep->hdr.dstate == VIO_DESC_DONE) ||
3429 	    (dep->payload.status == ECANCELED));
3430 
3431 	VDC_MARK_DRING_ENTRY_FREE(vdc, idx);
3432 
3433 	ldep->is_free = B_TRUE;
3434 	status = dep->payload.status;
3435 	DMSG(vdc, 2, ": is_free = %d : status = %d\n", ldep->is_free, status);
3436 
3437 	/*
3438 	 * If no buffers were used to transfer information to the server when
3439 	 * populating the descriptor then no memory handles need to be unbound
3440 	 * and we can return now.
3441 	 */
3442 	if (ldep->nbytes == 0) {
3443 		cv_signal(&vdc->dring_free_cv);
3444 		return (status);
3445 	}
3446 
3447 	/*
3448 	 * If the upper layer passed in a misaligned address we copied the
3449 	 * data into an aligned buffer before sending it to LDC - we now
3450 	 * copy it back to the original buffer.
3451 	 */
3452 	if (ldep->align_addr) {
3453 		ASSERT(ldep->addr != NULL);
3454 
3455 		if (dep->payload.nbytes > 0)
3456 			bcopy(ldep->align_addr, ldep->addr,
3457 			    dep->payload.nbytes);
3458 		kmem_free(ldep->align_addr,
3459 		    sizeof (caddr_t) * P2ROUNDUP(ldep->nbytes, 8));
3460 		ldep->align_addr = NULL;
3461 	}
3462 
3463 	rv = ldc_mem_unbind_handle(ldep->desc_mhdl);
3464 	if (rv != 0) {
3465 		DMSG(vdc, 0, "?[%d] unbind mhdl 0x%lx @ idx %d failed (%d)",
3466 		    vdc->instance, ldep->desc_mhdl, idx, rv);
3467 		/*
3468 		 * The error returned by the vDisk server is more informative
3469 		 * and thus has a higher priority but if it isn't set we ensure
3470 		 * that this function returns an error.
3471 		 */
3472 		if (status == 0)
3473 			status = EINVAL;
3474 	}
3475 
3476 	cv_signal(&vdc->membind_cv);
3477 	cv_signal(&vdc->dring_free_cv);
3478 
3479 	return (status);
3480 }
3481 
3482 /*
3483  * Function:
3484  *	vdc_populate_mem_hdl()
3485  *
3486  * Description:
3487  *
3488  * Arguments:
3489  *	vdc	- soft state pointer for this instance of the device driver.
3490  *	idx	- Index of the Descriptor Ring entry being modified
3491  *	addr	- virtual address being mapped in
3492  *	nybtes	- number of bytes in 'addr'
3493  *	operation - the vDisk operation being performed (VD_OP_xxx)
3494  *
3495  * Return Code:
3496  *	0	- Success
3497  */
3498 static int
3499 vdc_populate_mem_hdl(vdc_t *vdcp, vdc_local_desc_t *ldep)
3500 {
3501 	vd_dring_entry_t	*dep = NULL;
3502 	ldc_mem_handle_t	mhdl;
3503 	caddr_t			vaddr;
3504 	size_t			nbytes;
3505 	uint8_t			perm = LDC_MEM_RW;
3506 	uint8_t			maptype;
3507 	int			rv = 0;
3508 	int			i;
3509 
3510 	ASSERT(vdcp != NULL);
3511 
3512 	dep = ldep->dep;
3513 	mhdl = ldep->desc_mhdl;
3514 
3515 	switch (ldep->dir) {
3516 	case VIO_read_dir:
3517 		perm = LDC_MEM_W;
3518 		break;
3519 
3520 	case VIO_write_dir:
3521 		perm = LDC_MEM_R;
3522 		break;
3523 
3524 	case VIO_both_dir:
3525 		perm = LDC_MEM_RW;
3526 		break;
3527 
3528 	default:
3529 		ASSERT(0);	/* catch bad programming in vdc */
3530 	}
3531 
3532 	/*
3533 	 * LDC expects any addresses passed in to be 8-byte aligned. We need
3534 	 * to copy the contents of any misaligned buffers to a newly allocated
3535 	 * buffer and bind it instead (and copy the the contents back to the
3536 	 * original buffer passed in when depopulating the descriptor)
3537 	 */
3538 	vaddr = ldep->addr;
3539 	nbytes = ldep->nbytes;
3540 	if (((uint64_t)vaddr & 0x7) != 0) {
3541 		ASSERT(ldep->align_addr == NULL);
3542 		ldep->align_addr =
3543 		    kmem_alloc(sizeof (caddr_t) *
3544 		    P2ROUNDUP(nbytes, 8), KM_SLEEP);
3545 		DMSG(vdcp, 0, "[%d] Misaligned address %p reallocating "
3546 		    "(buf=%p nb=%ld op=%d)\n",
3547 		    vdcp->instance, (void *)vaddr, (void *)ldep->align_addr,
3548 		    nbytes, ldep->operation);
3549 		if (perm != LDC_MEM_W)
3550 			bcopy(vaddr, ldep->align_addr, nbytes);
3551 		vaddr = ldep->align_addr;
3552 	}
3553 
3554 	maptype = LDC_IO_MAP|LDC_SHADOW_MAP|LDC_DIRECT_MAP;
3555 	rv = ldc_mem_bind_handle(mhdl, vaddr, P2ROUNDUP(nbytes, 8),
3556 	    maptype, perm, &dep->payload.cookie[0], &dep->payload.ncookies);
3557 	DMSG(vdcp, 2, "[%d] bound mem handle; ncookies=%d\n",
3558 	    vdcp->instance, dep->payload.ncookies);
3559 	if (rv != 0) {
3560 		DMSG(vdcp, 0, "[%d] Failed to bind LDC memory handle "
3561 		    "(mhdl=%p, buf=%p, err=%d)\n",
3562 		    vdcp->instance, (void *)mhdl, (void *)vaddr, rv);
3563 		if (ldep->align_addr) {
3564 			kmem_free(ldep->align_addr,
3565 			    sizeof (caddr_t) * P2ROUNDUP(nbytes, 8));
3566 			ldep->align_addr = NULL;
3567 		}
3568 		return (EAGAIN);
3569 	}
3570 
3571 	/*
3572 	 * Get the other cookies (if any).
3573 	 */
3574 	for (i = 1; i < dep->payload.ncookies; i++) {
3575 		rv = ldc_mem_nextcookie(mhdl, &dep->payload.cookie[i]);
3576 		if (rv != 0) {
3577 			(void) ldc_mem_unbind_handle(mhdl);
3578 			DMSG(vdcp, 0, "?[%d] Failed to get next cookie "
3579 			    "(mhdl=%lx cnum=%d), err=%d",
3580 			    vdcp->instance, mhdl, i, rv);
3581 			if (ldep->align_addr) {
3582 				kmem_free(ldep->align_addr,
3583 				    sizeof (caddr_t) * ldep->nbytes);
3584 				ldep->align_addr = NULL;
3585 			}
3586 			return (EAGAIN);
3587 		}
3588 	}
3589 
3590 	return (rv);
3591 }
3592 
3593 /*
3594  * Interrupt handlers for messages from LDC
3595  */
3596 
3597 /*
3598  * Function:
3599  *	vdc_handle_cb()
3600  *
3601  * Description:
3602  *
3603  * Arguments:
3604  *	event	- Type of event (LDC_EVT_xxx) that triggered the callback
3605  *	arg	- soft state pointer for this instance of the device driver.
3606  *
3607  * Return Code:
3608  *	0	- Success
3609  */
3610 static uint_t
3611 vdc_handle_cb(uint64_t event, caddr_t arg)
3612 {
3613 	ldc_status_t	ldc_state;
3614 	int		rv = 0;
3615 	vdc_server_t	*srvr = (vdc_server_t *)(void *)arg;
3616 	vdc_t		*vdc = srvr->vdcp;
3617 
3618 	ASSERT(vdc != NULL);
3619 
3620 	DMSG(vdc, 1, "evt=%lx seqID=%ld\n", event, vdc->seq_num);
3621 
3622 	/* If callback is not for the current server, ignore it */
3623 	mutex_enter(&vdc->lock);
3624 
3625 	if (vdc->curr_server != srvr) {
3626 		DMSG(vdc, 0, "[%d] Ignoring event 0x%lx for port@%ld\n",
3627 		    vdc->instance, event, srvr->id);
3628 		mutex_exit(&vdc->lock);
3629 		return (LDC_SUCCESS);
3630 	}
3631 
3632 	/*
3633 	 * Depending on the type of event that triggered this callback,
3634 	 * we modify the handshake state or read the data.
3635 	 *
3636 	 * NOTE: not done as a switch() as event could be triggered by
3637 	 * a state change and a read request. Also the ordering	of the
3638 	 * check for the event types is deliberate.
3639 	 */
3640 	if (event & LDC_EVT_UP) {
3641 		DMSG(vdc, 0, "[%d] Received LDC_EVT_UP\n", vdc->instance);
3642 
3643 		/* get LDC state */
3644 		rv = ldc_status(srvr->ldc_handle, &ldc_state);
3645 		if (rv != 0) {
3646 			DMSG(vdc, 0, "[%d] Couldn't get LDC status %d",
3647 			    vdc->instance, rv);
3648 			mutex_exit(&vdc->lock);
3649 			return (LDC_SUCCESS);
3650 		}
3651 		if (srvr->ldc_state != LDC_UP &&
3652 		    ldc_state == LDC_UP) {
3653 			/*
3654 			 * Reset the transaction sequence numbers when
3655 			 * LDC comes up. We then kick off the handshake
3656 			 * negotiation with the vDisk server.
3657 			 */
3658 			vdc->seq_num = 1;
3659 			vdc->seq_num_reply = 0;
3660 			srvr->ldc_state = ldc_state;
3661 			cv_signal(&vdc->initwait_cv);
3662 		}
3663 	}
3664 
3665 	if (event & LDC_EVT_READ) {
3666 		DMSG(vdc, 1, "[%d] Received LDC_EVT_READ\n", vdc->instance);
3667 		mutex_enter(&vdc->read_lock);
3668 		cv_signal(&vdc->read_cv);
3669 		vdc->read_state = VDC_READ_PENDING;
3670 		mutex_exit(&vdc->read_lock);
3671 		mutex_exit(&vdc->lock);
3672 
3673 		/* that's all we have to do - no need to handle DOWN/RESET */
3674 		return (LDC_SUCCESS);
3675 	}
3676 
3677 	if (event & (LDC_EVT_RESET|LDC_EVT_DOWN)) {
3678 
3679 		DMSG(vdc, 0, "[%d] Received LDC RESET event\n", vdc->instance);
3680 
3681 		/*
3682 		 * Need to wake up any readers so they will
3683 		 * detect that a reset has occurred.
3684 		 */
3685 		mutex_enter(&vdc->read_lock);
3686 		if ((vdc->read_state == VDC_READ_WAITING) ||
3687 		    (vdc->read_state == VDC_READ_RESET))
3688 			cv_signal(&vdc->read_cv);
3689 		vdc->read_state = VDC_READ_RESET;
3690 		mutex_exit(&vdc->read_lock);
3691 
3692 		/* wake up any threads waiting for connection to come up */
3693 		if (vdc->state == VDC_STATE_INIT_WAITING) {
3694 			vdc->state = VDC_STATE_RESETTING;
3695 			cv_signal(&vdc->initwait_cv);
3696 		}
3697 
3698 	}
3699 
3700 	mutex_exit(&vdc->lock);
3701 
3702 	if (event & ~(LDC_EVT_UP | LDC_EVT_RESET | LDC_EVT_DOWN | LDC_EVT_READ))
3703 		DMSG(vdc, 0, "![%d] Unexpected LDC event (%lx) received",
3704 		    vdc->instance, event);
3705 
3706 	return (LDC_SUCCESS);
3707 }
3708 
3709 /*
3710  * Function:
3711  *	vdc_wait_for_response()
3712  *
3713  * Description:
3714  *	Block waiting for a response from the server. If there is
3715  *	no data the thread block on the read_cv that is signalled
3716  *	by the callback when an EVT_READ occurs.
3717  *
3718  * Arguments:
3719  *	vdcp	- soft state pointer for this instance of the device driver.
3720  *
3721  * Return Code:
3722  *	0	- Success
3723  */
3724 static int
3725 vdc_wait_for_response(vdc_t *vdcp, vio_msg_t *msgp)
3726 {
3727 	size_t		nbytes = sizeof (*msgp);
3728 	int		status;
3729 
3730 	ASSERT(vdcp != NULL);
3731 
3732 	DMSG(vdcp, 1, "[%d] Entered\n", vdcp->instance);
3733 
3734 	status = vdc_recv(vdcp, msgp, &nbytes);
3735 	DMSG(vdcp, 3, "vdc_read() done.. status=0x%x size=0x%x\n",
3736 	    status, (int)nbytes);
3737 	if (status) {
3738 		DMSG(vdcp, 0, "?[%d] Error %d reading LDC msg\n",
3739 		    vdcp->instance, status);
3740 		return (status);
3741 	}
3742 
3743 	if (nbytes < sizeof (vio_msg_tag_t)) {
3744 		DMSG(vdcp, 0, "?[%d] Expect %lu bytes; recv'd %lu\n",
3745 		    vdcp->instance, sizeof (vio_msg_tag_t), nbytes);
3746 		return (ENOMSG);
3747 	}
3748 
3749 	DMSG(vdcp, 2, "[%d] (%x/%x/%x)\n", vdcp->instance,
3750 	    msgp->tag.vio_msgtype,
3751 	    msgp->tag.vio_subtype,
3752 	    msgp->tag.vio_subtype_env);
3753 
3754 	/*
3755 	 * Verify the Session ID of the message
3756 	 *
3757 	 * Every message after the Version has been negotiated should
3758 	 * have the correct session ID set.
3759 	 */
3760 	if ((msgp->tag.vio_sid != vdcp->session_id) &&
3761 	    (msgp->tag.vio_subtype_env != VIO_VER_INFO)) {
3762 		DMSG(vdcp, 0, "[%d] Invalid SID: received 0x%x, "
3763 		    "expected 0x%lx [seq num %lx @ %d]",
3764 		    vdcp->instance, msgp->tag.vio_sid,
3765 		    vdcp->session_id,
3766 		    ((vio_dring_msg_t *)msgp)->seq_num,
3767 		    ((vio_dring_msg_t *)msgp)->start_idx);
3768 		return (ENOMSG);
3769 	}
3770 	return (0);
3771 }
3772 
3773 
3774 /*
3775  * Function:
3776  *	vdc_resubmit_backup_dring()
3777  *
3778  * Description:
3779  *	Resubmit each descriptor in the backed up dring to
3780  * 	vDisk server. The Dring was backed up during connection
3781  *	reset.
3782  *
3783  * Arguments:
3784  *	vdcp	- soft state pointer for this instance of the device driver.
3785  *
3786  * Return Code:
3787  *	0	- Success
3788  */
3789 static int
3790 vdc_resubmit_backup_dring(vdc_t *vdcp)
3791 {
3792 	int		processed = 0;
3793 	int		count;
3794 	int		b_idx;
3795 	int		rv = 0;
3796 	int		dring_size;
3797 	int		op;
3798 	vio_msg_t	vio_msg;
3799 	vdc_local_desc_t	*curr_ldep;
3800 
3801 	ASSERT(MUTEX_NOT_HELD(&vdcp->lock));
3802 	ASSERT(vdcp->state == VDC_STATE_HANDLE_PENDING);
3803 
3804 	if (vdcp->local_dring_backup == NULL) {
3805 		/* the pending requests have already been processed */
3806 		return (0);
3807 	}
3808 
3809 	DMSG(vdcp, 1, "restoring pending dring entries (len=%d, tail=%d)\n",
3810 	    vdcp->local_dring_backup_len, vdcp->local_dring_backup_tail);
3811 
3812 	/*
3813 	 * Walk the backup copy of the local descriptor ring and
3814 	 * resubmit all the outstanding transactions.
3815 	 */
3816 	b_idx = vdcp->local_dring_backup_tail;
3817 	for (count = 0; count < vdcp->local_dring_backup_len; count++) {
3818 
3819 		curr_ldep = &(vdcp->local_dring_backup[b_idx]);
3820 
3821 		/* only resubmit outstanding transactions */
3822 		if (!curr_ldep->is_free) {
3823 			/*
3824 			 * If we are retrying a block read/write operation we
3825 			 * need to update the I/O statistics to indicate that
3826 			 * the request is being put back on the waitq to be
3827 			 * serviced (it will have been taken off after the
3828 			 * error was reported).
3829 			 */
3830 			mutex_enter(&vdcp->lock);
3831 			op = curr_ldep->operation;
3832 			if ((op == VD_OP_BREAD) || (op == VD_OP_BWRITE)) {
3833 				DTRACE_IO1(start, buf_t *, curr_ldep->cb_arg);
3834 				VD_KSTAT_WAITQ_ENTER(vdcp);
3835 			}
3836 
3837 			DMSG(vdcp, 1, "resubmitting entry idx=%x\n", b_idx);
3838 			rv = vdc_populate_descriptor(vdcp, op,
3839 			    curr_ldep->addr, curr_ldep->nbytes,
3840 			    curr_ldep->slice, curr_ldep->offset,
3841 			    curr_ldep->cb_type, curr_ldep->cb_arg,
3842 			    curr_ldep->dir);
3843 
3844 			if (rv) {
3845 				if (op == VD_OP_BREAD || op == VD_OP_BWRITE) {
3846 					VD_UPDATE_ERR_STATS(vdcp, vd_transerrs);
3847 					VD_KSTAT_WAITQ_EXIT(vdcp);
3848 					DTRACE_IO1(done, buf_t *,
3849 					    curr_ldep->cb_arg);
3850 				}
3851 				DMSG(vdcp, 1, "[%d] cannot resubmit entry %d\n",
3852 				    vdcp->instance, b_idx);
3853 				mutex_exit(&vdcp->lock);
3854 				goto done;
3855 			}
3856 
3857 			/*
3858 			 * If this is a block read/write we update the I/O
3859 			 * statistics kstat to indicate that the request
3860 			 * has been sent back to the vDisk server and should
3861 			 * now be put on the run queue.
3862 			 */
3863 			if ((op == VD_OP_BREAD) || (op == VD_OP_BWRITE)) {
3864 				DTRACE_PROBE1(send, buf_t *, curr_ldep->cb_arg);
3865 				VD_KSTAT_WAITQ_TO_RUNQ(vdcp);
3866 			}
3867 			mutex_exit(&vdcp->lock);
3868 
3869 			/* Wait for the response message. */
3870 			DMSG(vdcp, 1, "waiting for response to idx=%x\n",
3871 			    b_idx);
3872 			rv = vdc_wait_for_response(vdcp, &vio_msg);
3873 			if (rv) {
3874 				/*
3875 				 * If this is a block read/write we update
3876 				 * the I/O statistics kstat to take it
3877 				 * off the run queue.
3878 				 */
3879 				mutex_enter(&vdcp->lock);
3880 				if (op == VD_OP_BREAD || op == VD_OP_BWRITE) {
3881 					VD_UPDATE_ERR_STATS(vdcp, vd_transerrs);
3882 					VD_KSTAT_RUNQ_EXIT(vdcp);
3883 					DTRACE_IO1(done, buf_t *,
3884 					    curr_ldep->cb_arg);
3885 				}
3886 				DMSG(vdcp, 1, "[%d] wait_for_response "
3887 				    "returned err=%d\n", vdcp->instance,
3888 				    rv);
3889 				mutex_exit(&vdcp->lock);
3890 				goto done;
3891 			}
3892 
3893 			DMSG(vdcp, 1, "processing msg for idx=%x\n", b_idx);
3894 			rv = vdc_process_data_msg(vdcp, &vio_msg);
3895 			if (rv) {
3896 				DMSG(vdcp, 1, "[%d] process_data_msg "
3897 				    "returned err=%d\n", vdcp->instance,
3898 				    rv);
3899 				goto done;
3900 			}
3901 			/*
3902 			 * Mark this entry as free so that we will not resubmit
3903 			 * this "done" request again, if we were to use the same
3904 			 * backup_dring again in future. This could happen when
3905 			 * a reset happens while processing the backup_dring.
3906 			 */
3907 			curr_ldep->is_free = B_TRUE;
3908 			processed++;
3909 		}
3910 
3911 		/* get the next element to submit */
3912 		if (++b_idx >= vdcp->local_dring_backup_len)
3913 			b_idx = 0;
3914 	}
3915 
3916 	/* all done - now clear up pending dring copy */
3917 	dring_size = vdcp->local_dring_backup_len *
3918 	    sizeof (vdcp->local_dring_backup[0]);
3919 
3920 	(void) kmem_free(vdcp->local_dring_backup, dring_size);
3921 
3922 	vdcp->local_dring_backup = NULL;
3923 
3924 done:
3925 	DTRACE_PROBE2(processed, int, processed, vdc_t *, vdcp);
3926 
3927 	return (rv);
3928 }
3929 
3930 /*
3931  * Function:
3932  *	vdc_cancel_backup_dring
3933  *
3934  * Description:
3935  *	Cancel each descriptor in the backed up dring to vDisk server.
3936  *	The Dring was backed up during connection reset.
3937  *
3938  * Arguments:
3939  *	vdcp	- soft state pointer for this instance of the device driver.
3940  *
3941  * Return Code:
3942  *	None
3943  */
3944 void
3945 vdc_cancel_backup_dring(vdc_t *vdcp)
3946 {
3947 	vdc_local_desc_t *ldep;
3948 	struct buf 	*bufp;
3949 	int		count;
3950 	int		b_idx;
3951 	int		dring_size;
3952 	int		cancelled = 0;
3953 
3954 	ASSERT(MUTEX_HELD(&vdcp->lock));
3955 	ASSERT(vdcp->state == VDC_STATE_INIT ||
3956 	    vdcp->state == VDC_STATE_INIT_WAITING ||
3957 	    vdcp->state == VDC_STATE_NEGOTIATE ||
3958 	    vdcp->state == VDC_STATE_RESETTING);
3959 
3960 	if (vdcp->local_dring_backup == NULL) {
3961 		/* the pending requests have already been processed */
3962 		return;
3963 	}
3964 
3965 	DMSG(vdcp, 1, "cancelling pending dring entries (len=%d, tail=%d)\n",
3966 	    vdcp->local_dring_backup_len, vdcp->local_dring_backup_tail);
3967 
3968 	/*
3969 	 * Walk the backup copy of the local descriptor ring and
3970 	 * cancel all the outstanding transactions.
3971 	 */
3972 	b_idx = vdcp->local_dring_backup_tail;
3973 	for (count = 0; count < vdcp->local_dring_backup_len; count++) {
3974 
3975 		ldep = &(vdcp->local_dring_backup[b_idx]);
3976 
3977 		/* only cancel outstanding transactions */
3978 		if (!ldep->is_free) {
3979 
3980 			DMSG(vdcp, 1, "cancelling entry idx=%x\n", b_idx);
3981 			cancelled++;
3982 
3983 			/*
3984 			 * All requests have already been cleared from the
3985 			 * local descriptor ring and the LDC channel has been
3986 			 * reset so we will never get any reply for these
3987 			 * requests. Now we just have to notify threads waiting
3988 			 * for replies that the request has failed.
3989 			 */
3990 			switch (ldep->cb_type) {
3991 			case CB_SYNC:
3992 				ASSERT(vdcp->sync_op_pending);
3993 				vdcp->sync_op_status = EIO;
3994 				vdcp->sync_op_pending = B_FALSE;
3995 				cv_signal(&vdcp->sync_pending_cv);
3996 				break;
3997 
3998 			case CB_STRATEGY:
3999 				bufp = ldep->cb_arg;
4000 				ASSERT(bufp != NULL);
4001 				bufp->b_resid = bufp->b_bcount;
4002 				VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4003 				VD_KSTAT_RUNQ_EXIT(vdcp);
4004 				DTRACE_IO1(done, buf_t *, bufp);
4005 				bioerror(bufp, EIO);
4006 				biodone(bufp);
4007 				break;
4008 
4009 			default:
4010 				ASSERT(0);
4011 			}
4012 
4013 		}
4014 
4015 		/* get the next element to cancel */
4016 		if (++b_idx >= vdcp->local_dring_backup_len)
4017 			b_idx = 0;
4018 	}
4019 
4020 	/* all done - now clear up pending dring copy */
4021 	dring_size = vdcp->local_dring_backup_len *
4022 	    sizeof (vdcp->local_dring_backup[0]);
4023 
4024 	(void) kmem_free(vdcp->local_dring_backup, dring_size);
4025 
4026 	vdcp->local_dring_backup = NULL;
4027 
4028 	DTRACE_PROBE2(cancelled, int, cancelled, vdc_t *, vdcp);
4029 }
4030 
4031 /*
4032  * Function:
4033  *	vdc_connection_timeout
4034  *
4035  * Description:
4036  *	This function is invoked if the timeout set to establish the connection
4037  *	with vds expires. This will happen if we spend too much time in the
4038  *	VDC_STATE_INIT_WAITING or VDC_STATE_NEGOTIATE states. Then we will
4039  *	cancel any pending request and mark them as failed.
4040  *
4041  *	If the timeout does not expire, it will be cancelled when we reach the
4042  *	VDC_STATE_HANDLE_PENDING or VDC_STATE_RESETTING state. This function can
4043  *	be invoked while we are in the VDC_STATE_HANDLE_PENDING or
4044  *	VDC_STATE_RESETTING state in which case we do nothing because the
4045  *	timeout is being cancelled.
4046  *
4047  * Arguments:
4048  *	arg	- argument of the timeout function actually a soft state
4049  *		  pointer for the instance of the device driver.
4050  *
4051  * Return Code:
4052  *	None
4053  */
4054 void
4055 vdc_connection_timeout(void *arg)
4056 {
4057 	vdc_t 		*vdcp = (vdc_t *)arg;
4058 
4059 	mutex_enter(&vdcp->lock);
4060 
4061 	if (vdcp->state == VDC_STATE_HANDLE_PENDING ||
4062 	    vdcp->state == VDC_STATE_DETACH) {
4063 		/*
4064 		 * The connection has just been re-established or
4065 		 * we are detaching.
4066 		 */
4067 		vdcp->ctimeout_reached = B_FALSE;
4068 		mutex_exit(&vdcp->lock);
4069 		return;
4070 	}
4071 
4072 	vdcp->ctimeout_reached = B_TRUE;
4073 
4074 	/* notify requests waiting for sending */
4075 	cv_broadcast(&vdcp->running_cv);
4076 
4077 	/* cancel requests waiting for a result */
4078 	vdc_cancel_backup_dring(vdcp);
4079 
4080 	mutex_exit(&vdcp->lock);
4081 
4082 	cmn_err(CE_NOTE, "[%d] connection to service domain timeout",
4083 	    vdcp->instance);
4084 }
4085 
4086 /*
4087  * Function:
4088  *	vdc_backup_local_dring()
4089  *
4090  * Description:
4091  *	Backup the current dring in the event of a reset. The Dring
4092  *	transactions will be resubmitted to the server when the
4093  *	connection is restored.
4094  *
4095  * Arguments:
4096  *	vdcp	- soft state pointer for this instance of the device driver.
4097  *
4098  * Return Code:
4099  *	NONE
4100  */
4101 static void
4102 vdc_backup_local_dring(vdc_t *vdcp)
4103 {
4104 	int dring_size;
4105 
4106 	ASSERT(MUTEX_HELD(&vdcp->lock));
4107 	ASSERT(vdcp->state == VDC_STATE_RESETTING);
4108 
4109 	/*
4110 	 * If the backup dring is stil around, it means
4111 	 * that the last restore did not complete. However,
4112 	 * since we never got back into the running state,
4113 	 * the backup copy we have is still valid.
4114 	 */
4115 	if (vdcp->local_dring_backup != NULL) {
4116 		DMSG(vdcp, 1, "reusing local descriptor ring backup "
4117 		    "(len=%d, tail=%d)\n", vdcp->local_dring_backup_len,
4118 		    vdcp->local_dring_backup_tail);
4119 		return;
4120 	}
4121 
4122 	/*
4123 	 * The backup dring can be NULL and the local dring may not be
4124 	 * initialized. This can happen if we had a reset while establishing
4125 	 * a new connection but after the connection has timed out. In that
4126 	 * case the backup dring is NULL because the requests have been
4127 	 * cancelled and the request occured before the local dring is
4128 	 * initialized.
4129 	 */
4130 	if (!(vdcp->initialized & VDC_DRING_LOCAL))
4131 		return;
4132 
4133 	DMSG(vdcp, 1, "backing up the local descriptor ring (len=%d, "
4134 	    "tail=%d)\n", vdcp->dring_len, vdcp->dring_curr_idx);
4135 
4136 	dring_size = vdcp->dring_len * sizeof (vdcp->local_dring[0]);
4137 
4138 	vdcp->local_dring_backup = kmem_alloc(dring_size, KM_SLEEP);
4139 	bcopy(vdcp->local_dring, vdcp->local_dring_backup, dring_size);
4140 
4141 	vdcp->local_dring_backup_tail = vdcp->dring_curr_idx;
4142 	vdcp->local_dring_backup_len = vdcp->dring_len;
4143 }
4144 
4145 static void
4146 vdc_switch_server(vdc_t *vdcp)
4147 {
4148 	int		rv;
4149 	vdc_server_t 	*curr_server, *new_server;
4150 
4151 	ASSERT(MUTEX_HELD(&vdcp->lock));
4152 
4153 	/* if there is only one server return back */
4154 	if (vdcp->num_servers == 1) {
4155 		return;
4156 	}
4157 
4158 	/* Get current and next server */
4159 	curr_server = vdcp->curr_server;
4160 	new_server =
4161 	    (curr_server->next) ? curr_server->next : vdcp->server_list;
4162 	ASSERT(curr_server != new_server);
4163 
4164 	/* bring current server's channel down */
4165 	rv = ldc_down(curr_server->ldc_handle);
4166 	if (rv) {
4167 		DMSG(vdcp, 0, "[%d] Cannot bring channel down, port %ld\n",
4168 		    vdcp->instance, curr_server->id);
4169 		return;
4170 	}
4171 
4172 	/* switch the server */
4173 	vdcp->curr_server = new_server;
4174 
4175 	DMSG(vdcp, 0, "[%d] Switched to next vdisk server, port@%ld, ldc@%ld\n",
4176 	    vdcp->instance, vdcp->curr_server->id, vdcp->curr_server->ldc_id);
4177 }
4178 
4179 /* -------------------------------------------------------------------------- */
4180 
4181 /*
4182  * The following functions process the incoming messages from vds
4183  */
4184 
4185 /*
4186  * Function:
4187  *      vdc_process_msg_thread()
4188  *
4189  * Description:
4190  *
4191  *	Main VDC message processing thread. Each vDisk instance
4192  * 	consists of a copy of this thread. This thread triggers
4193  * 	all the handshakes and data exchange with the server. It
4194  * 	also handles all channel resets
4195  *
4196  * Arguments:
4197  *      vdc     - soft state pointer for this instance of the device driver.
4198  *
4199  * Return Code:
4200  *      None
4201  */
4202 static void
4203 vdc_process_msg_thread(vdc_t *vdcp)
4204 {
4205 	int		status;
4206 	int		ctimeout;
4207 	timeout_id_t	tmid = 0;
4208 	clock_t		ldcup_timeout = 0;
4209 
4210 	mutex_enter(&vdcp->lock);
4211 
4212 	for (;;) {
4213 
4214 #define	Q(_s)	(vdcp->state == _s) ? #_s :
4215 		DMSG(vdcp, 3, "state = %d (%s)\n", vdcp->state,
4216 		    Q(VDC_STATE_INIT)
4217 		    Q(VDC_STATE_INIT_WAITING)
4218 		    Q(VDC_STATE_NEGOTIATE)
4219 		    Q(VDC_STATE_HANDLE_PENDING)
4220 		    Q(VDC_STATE_RUNNING)
4221 		    Q(VDC_STATE_RESETTING)
4222 		    Q(VDC_STATE_DETACH)
4223 		    "UNKNOWN");
4224 
4225 		switch (vdcp->state) {
4226 		case VDC_STATE_INIT:
4227 
4228 			/*
4229 			 * If requested, start a timeout to check if the
4230 			 * connection with vds is established in the
4231 			 * specified delay. If the timeout expires, we
4232 			 * will cancel any pending request.
4233 			 *
4234 			 * If some reset have occurred while establishing
4235 			 * the connection, we already have a timeout armed
4236 			 * and in that case we don't need to arm a new one.
4237 			 *
4238 			 * The same rule applies when there are multiple vds'.
4239 			 * If either a connection cannot be established or
4240 			 * the handshake times out, the connection thread will
4241 			 * try another server. The 'ctimeout' will report
4242 			 * back an error after it expires irrespective of
4243 			 * whether the vdisk is trying to connect to just
4244 			 * one or multiple servers.
4245 			 */
4246 			ctimeout = (vdc_timeout != 0)?
4247 			    vdc_timeout : vdcp->curr_server->ctimeout;
4248 
4249 			if (ctimeout != 0 && tmid == 0) {
4250 				tmid = timeout(vdc_connection_timeout, vdcp,
4251 				    ctimeout * drv_usectohz(MICROSEC));
4252 			}
4253 
4254 			/* Check if we are re-initializing repeatedly */
4255 			if (vdcp->hshake_cnt > vdc_hshake_retries &&
4256 			    vdcp->lifecycle != VDC_LC_ONLINE) {
4257 
4258 				DMSG(vdcp, 0, "[%d] too many handshakes,cnt=%d",
4259 				    vdcp->instance, vdcp->hshake_cnt);
4260 				cmn_err(CE_NOTE, "[%d] disk access failed.\n",
4261 				    vdcp->instance);
4262 				vdcp->state = VDC_STATE_DETACH;
4263 				break;
4264 			}
4265 
4266 			/* Switch to STATE_DETACH if drv is detaching */
4267 			if (vdcp->lifecycle == VDC_LC_DETACHING) {
4268 				vdcp->state = VDC_STATE_DETACH;
4269 				break;
4270 			}
4271 
4272 			/* Switch server */
4273 			if (vdcp->hshake_cnt > 0)
4274 				vdc_switch_server(vdcp);
4275 			vdcp->hshake_cnt++;
4276 
4277 			/* Bring up connection with vds via LDC */
4278 			status = vdc_start_ldc_connection(vdcp);
4279 			if (status != EINVAL) {
4280 				vdcp->state = VDC_STATE_INIT_WAITING;
4281 			}
4282 			break;
4283 
4284 		case VDC_STATE_INIT_WAITING:
4285 
4286 			/* if channel is UP, start negotiation */
4287 			if (vdcp->curr_server->ldc_state == LDC_UP) {
4288 				vdcp->state = VDC_STATE_NEGOTIATE;
4289 				break;
4290 			}
4291 
4292 			/* check if only one server exists */
4293 			if (vdcp->num_servers == 1) {
4294 				cv_wait(&vdcp->initwait_cv, &vdcp->lock);
4295 			} else {
4296 				/*
4297 				 * wait for LDC_UP, if it times out, switch
4298 				 * to another server.
4299 				 */
4300 				ldcup_timeout = ddi_get_lbolt() +
4301 				    (vdc_ldcup_timeout *
4302 				    drv_usectohz(MICROSEC));
4303 				status = cv_timedwait(&vdcp->initwait_cv,
4304 				    &vdcp->lock, ldcup_timeout);
4305 				if (status == -1 &&
4306 				    vdcp->state == VDC_STATE_INIT_WAITING &&
4307 				    vdcp->curr_server->ldc_state != LDC_UP) {
4308 					/* timed out & still waiting */
4309 					vdcp->state = VDC_STATE_INIT;
4310 					break;
4311 				}
4312 			}
4313 
4314 			if (vdcp->state != VDC_STATE_INIT_WAITING) {
4315 				DMSG(vdcp, 0,
4316 				    "state moved to %d out from under us...\n",
4317 				    vdcp->state);
4318 			}
4319 			break;
4320 
4321 		case VDC_STATE_NEGOTIATE:
4322 			switch (status = vdc_ver_negotiation(vdcp)) {
4323 			case 0:
4324 				break;
4325 			default:
4326 				DMSG(vdcp, 0, "ver negotiate failed (%d)..\n",
4327 				    status);
4328 				goto reset;
4329 			}
4330 
4331 			switch (status = vdc_attr_negotiation(vdcp)) {
4332 			case 0:
4333 				break;
4334 			default:
4335 				DMSG(vdcp, 0, "attr negotiate failed (%d)..\n",
4336 				    status);
4337 				goto reset;
4338 			}
4339 
4340 			switch (status = vdc_dring_negotiation(vdcp)) {
4341 			case 0:
4342 				break;
4343 			default:
4344 				DMSG(vdcp, 0, "dring negotiate failed (%d)..\n",
4345 				    status);
4346 				goto reset;
4347 			}
4348 
4349 			switch (status = vdc_rdx_exchange(vdcp)) {
4350 			case 0:
4351 				vdcp->state = VDC_STATE_HANDLE_PENDING;
4352 				goto done;
4353 			default:
4354 				DMSG(vdcp, 0, "RDX xchg failed ..(%d)\n",
4355 				    status);
4356 				goto reset;
4357 			}
4358 reset:
4359 			DMSG(vdcp, 0, "negotiation failed: resetting (%d)\n",
4360 			    status);
4361 			vdcp->state = VDC_STATE_RESETTING;
4362 			vdcp->self_reset = B_TRUE;
4363 done:
4364 			DMSG(vdcp, 0, "negotiation complete (state=0x%x)...\n",
4365 			    vdcp->state);
4366 			break;
4367 
4368 		case VDC_STATE_HANDLE_PENDING:
4369 
4370 			if (vdcp->ctimeout_reached) {
4371 				/*
4372 				 * The connection timeout had been reached so
4373 				 * pending requests have been cancelled. Now
4374 				 * that the connection is back we can reset
4375 				 * the timeout.
4376 				 */
4377 				ASSERT(vdcp->local_dring_backup == NULL);
4378 				ASSERT(tmid != 0);
4379 				tmid = 0;
4380 				vdcp->ctimeout_reached = B_FALSE;
4381 				vdcp->state = VDC_STATE_RUNNING;
4382 				DMSG(vdcp, 0, "[%d] connection to service "
4383 				    "domain is up", vdcp->instance);
4384 				break;
4385 			}
4386 
4387 			mutex_exit(&vdcp->lock);
4388 			if (tmid != 0) {
4389 				(void) untimeout(tmid);
4390 				tmid = 0;
4391 			}
4392 			status = vdc_resubmit_backup_dring(vdcp);
4393 			mutex_enter(&vdcp->lock);
4394 
4395 			if (status)
4396 				vdcp->state = VDC_STATE_RESETTING;
4397 			else
4398 				vdcp->state = VDC_STATE_RUNNING;
4399 
4400 			break;
4401 
4402 		/* enter running state */
4403 		case VDC_STATE_RUNNING:
4404 			/*
4405 			 * Signal anyone waiting for the connection
4406 			 * to come on line.
4407 			 */
4408 			vdcp->hshake_cnt = 0;
4409 			cv_broadcast(&vdcp->running_cv);
4410 
4411 			/* failfast has to been checked after reset */
4412 			cv_signal(&vdcp->failfast_cv);
4413 
4414 			/* ownership is lost during reset */
4415 			if (vdcp->ownership & VDC_OWNERSHIP_WANTED)
4416 				vdcp->ownership |= VDC_OWNERSHIP_RESET;
4417 			cv_signal(&vdcp->ownership_cv);
4418 
4419 			cmn_err(CE_CONT, "?vdisk@%d is online using "
4420 			    "ldc@%ld,%ld\n", vdcp->instance,
4421 			    vdcp->curr_server->ldc_id, vdcp->curr_server->id);
4422 
4423 			mutex_exit(&vdcp->lock);
4424 
4425 			for (;;) {
4426 				vio_msg_t msg;
4427 				status = vdc_wait_for_response(vdcp, &msg);
4428 				if (status) break;
4429 
4430 				DMSG(vdcp, 1, "[%d] new pkt(s) available\n",
4431 				    vdcp->instance);
4432 				status = vdc_process_data_msg(vdcp, &msg);
4433 				if (status) {
4434 					DMSG(vdcp, 1, "[%d] process_data_msg "
4435 					    "returned err=%d\n", vdcp->instance,
4436 					    status);
4437 					break;
4438 				}
4439 
4440 			}
4441 
4442 			mutex_enter(&vdcp->lock);
4443 
4444 			cmn_err(CE_CONT, "?vdisk@%d is offline\n",
4445 			    vdcp->instance);
4446 
4447 			vdcp->state = VDC_STATE_RESETTING;
4448 			vdcp->self_reset = B_TRUE;
4449 			break;
4450 
4451 		case VDC_STATE_RESETTING:
4452 			/*
4453 			 * When we reach this state, we either come from the
4454 			 * VDC_STATE_RUNNING state and we can have pending
4455 			 * request but no timeout is armed; or we come from
4456 			 * the VDC_STATE_INIT_WAITING, VDC_NEGOTIATE or
4457 			 * VDC_HANDLE_PENDING state and there is no pending
4458 			 * request or pending requests have already been copied
4459 			 * into the backup dring. So we can safely keep the
4460 			 * connection timeout armed while we are in this state.
4461 			 */
4462 
4463 			DMSG(vdcp, 0, "Initiating channel reset "
4464 			    "(pending = %d)\n", (int)vdcp->threads_pending);
4465 
4466 			if (vdcp->self_reset) {
4467 				DMSG(vdcp, 0,
4468 				    "[%d] calling stop_ldc_connection.\n",
4469 				    vdcp->instance);
4470 				status = vdc_stop_ldc_connection(vdcp);
4471 				vdcp->self_reset = B_FALSE;
4472 			}
4473 
4474 			/*
4475 			 * Wait for all threads currently waiting
4476 			 * for a free dring entry to use.
4477 			 */
4478 			while (vdcp->threads_pending) {
4479 				cv_broadcast(&vdcp->membind_cv);
4480 				cv_broadcast(&vdcp->dring_free_cv);
4481 				mutex_exit(&vdcp->lock);
4482 				/* give the waiters enough time to wake up */
4483 				delay(vdc_hz_min_ldc_delay);
4484 				mutex_enter(&vdcp->lock);
4485 			}
4486 
4487 			ASSERT(vdcp->threads_pending == 0);
4488 
4489 			/* Sanity check that no thread is receiving */
4490 			ASSERT(vdcp->read_state != VDC_READ_WAITING);
4491 
4492 			vdcp->read_state = VDC_READ_IDLE;
4493 
4494 			vdc_backup_local_dring(vdcp);
4495 
4496 			/* cleanup the old d-ring */
4497 			vdc_destroy_descriptor_ring(vdcp);
4498 
4499 			/* go and start again */
4500 			vdcp->state = VDC_STATE_INIT;
4501 
4502 			break;
4503 
4504 		case VDC_STATE_DETACH:
4505 			DMSG(vdcp, 0, "[%d] Reset thread exit cleanup ..\n",
4506 			    vdcp->instance);
4507 
4508 			/* cancel any pending timeout */
4509 			mutex_exit(&vdcp->lock);
4510 			if (tmid != 0) {
4511 				(void) untimeout(tmid);
4512 				tmid = 0;
4513 			}
4514 			mutex_enter(&vdcp->lock);
4515 
4516 			/*
4517 			 * Signal anyone waiting for connection
4518 			 * to come online
4519 			 */
4520 			cv_broadcast(&vdcp->running_cv);
4521 
4522 			while (vdcp->sync_op_pending) {
4523 				cv_signal(&vdcp->sync_pending_cv);
4524 				cv_signal(&vdcp->sync_blocked_cv);
4525 				mutex_exit(&vdcp->lock);
4526 				/* give the waiters enough time to wake up */
4527 				delay(vdc_hz_min_ldc_delay);
4528 				mutex_enter(&vdcp->lock);
4529 			}
4530 
4531 			mutex_exit(&vdcp->lock);
4532 
4533 			DMSG(vdcp, 0, "[%d] Msg processing thread exiting ..\n",
4534 			    vdcp->instance);
4535 			thread_exit();
4536 			break;
4537 		}
4538 	}
4539 }
4540 
4541 
4542 /*
4543  * Function:
4544  *	vdc_process_data_msg()
4545  *
4546  * Description:
4547  *	This function is called by the message processing thread each time
4548  *	a message with a msgtype of VIO_TYPE_DATA is received. It will either
4549  *	be an ACK or NACK from vds[1] which vdc handles as follows.
4550  *		ACK	- wake up the waiting thread
4551  *		NACK	- resend any messages necessary
4552  *
4553  *	[1] Although the message format allows it, vds should not send a
4554  *	    VIO_SUBTYPE_INFO message to vdc asking it to read data; if for
4555  *	    some bizarre reason it does, vdc will reset the connection.
4556  *
4557  * Arguments:
4558  *	vdc	- soft state pointer for this instance of the device driver.
4559  *	msg	- the LDC message sent by vds
4560  *
4561  * Return Code:
4562  *	0	- Success.
4563  *	> 0	- error value returned by LDC
4564  */
4565 static int
4566 vdc_process_data_msg(vdc_t *vdcp, vio_msg_t *msg)
4567 {
4568 	int			status = 0;
4569 	vio_dring_msg_t		*dring_msg;
4570 	vdc_local_desc_t	*ldep = NULL;
4571 	int			start, end;
4572 	int			idx;
4573 	int			op;
4574 
4575 	dring_msg = (vio_dring_msg_t *)msg;
4576 
4577 	ASSERT(msg->tag.vio_msgtype == VIO_TYPE_DATA);
4578 	ASSERT(vdcp != NULL);
4579 
4580 	mutex_enter(&vdcp->lock);
4581 
4582 	/*
4583 	 * Check to see if the message has bogus data
4584 	 */
4585 	idx = start = dring_msg->start_idx;
4586 	end = dring_msg->end_idx;
4587 	if ((start >= vdcp->dring_len) ||
4588 	    (end >= vdcp->dring_len) || (end < -1)) {
4589 		/*
4590 		 * Update the I/O statistics to indicate that an error ocurred.
4591 		 * No need to update the wait/run queues as no specific read or
4592 		 * write request is being completed in response to this 'msg'.
4593 		 */
4594 		VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4595 		DMSG(vdcp, 0, "[%d] Bogus ACK data : start %d, end %d\n",
4596 		    vdcp->instance, start, end);
4597 		mutex_exit(&vdcp->lock);
4598 		return (EINVAL);
4599 	}
4600 
4601 	/*
4602 	 * Verify that the sequence number is what vdc expects.
4603 	 */
4604 	switch (vdc_verify_seq_num(vdcp, dring_msg)) {
4605 	case VDC_SEQ_NUM_TODO:
4606 		break;	/* keep processing this message */
4607 	case VDC_SEQ_NUM_SKIP:
4608 		mutex_exit(&vdcp->lock);
4609 		return (0);
4610 	case VDC_SEQ_NUM_INVALID:
4611 		/*
4612 		 * Update the I/O statistics to indicate that an error ocurred.
4613 		 * No need to update the wait/run queues as no specific read or
4614 		 * write request is being completed in response to this 'msg'.
4615 		 */
4616 		VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4617 		DMSG(vdcp, 0, "[%d] invalid seqno\n", vdcp->instance);
4618 		mutex_exit(&vdcp->lock);
4619 		return (ENXIO);
4620 	}
4621 
4622 	if (msg->tag.vio_subtype == VIO_SUBTYPE_NACK) {
4623 		/*
4624 		 * Update the I/O statistics to indicate that an error ocurred.
4625 		 *
4626 		 * We need to update the run queue if a read or write request
4627 		 * is being NACKed - otherwise there will appear to be an
4628 		 * indefinite outstanding request and statistics reported by
4629 		 * iostat(1M) will be incorrect. The transaction will be
4630 		 * resubmitted from the backup DRing following the reset
4631 		 * and the wait/run queues will be entered again.
4632 		 */
4633 		ldep = &vdcp->local_dring[idx];
4634 		op = ldep->operation;
4635 		if ((op == VD_OP_BREAD) || (op == VD_OP_BWRITE)) {
4636 			DTRACE_IO1(done, buf_t *, ldep->cb_arg);
4637 			VD_KSTAT_RUNQ_EXIT(vdcp);
4638 		}
4639 		VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4640 		VDC_DUMP_DRING_MSG(dring_msg);
4641 		DMSG(vdcp, 0, "[%d] DATA NACK\n", vdcp->instance);
4642 		mutex_exit(&vdcp->lock);
4643 		return (EIO);
4644 
4645 	} else if (msg->tag.vio_subtype == VIO_SUBTYPE_INFO) {
4646 		/*
4647 		 * Update the I/O statistics to indicate that an error occurred.
4648 		 * No need to update the wait/run queues as no specific read or
4649 		 * write request is being completed in response to this 'msg'.
4650 		 */
4651 		VD_UPDATE_ERR_STATS(vdcp, vd_protoerrs);
4652 		mutex_exit(&vdcp->lock);
4653 		return (EPROTO);
4654 	}
4655 
4656 	DMSG(vdcp, 1, ": start %d end %d\n", start, end);
4657 	ASSERT(start == end);
4658 
4659 	ldep = &vdcp->local_dring[idx];
4660 
4661 	DMSG(vdcp, 1, ": state 0x%x - cb_type 0x%x\n",
4662 	    ldep->dep->hdr.dstate, ldep->cb_type);
4663 
4664 	if (ldep->dep->hdr.dstate == VIO_DESC_DONE) {
4665 		struct buf *bufp;
4666 
4667 		switch (ldep->cb_type) {
4668 		case CB_SYNC:
4669 			ASSERT(vdcp->sync_op_pending);
4670 
4671 			status = vdc_depopulate_descriptor(vdcp, idx);
4672 			vdcp->sync_op_status = status;
4673 			vdcp->sync_op_pending = B_FALSE;
4674 			cv_signal(&vdcp->sync_pending_cv);
4675 			break;
4676 
4677 		case CB_STRATEGY:
4678 			bufp = ldep->cb_arg;
4679 			ASSERT(bufp != NULL);
4680 			bufp->b_resid =
4681 			    bufp->b_bcount - ldep->dep->payload.nbytes;
4682 			status = ldep->dep->payload.status; /* Future:ntoh */
4683 			if (status != 0) {
4684 				DMSG(vdcp, 1, "strategy status=%d\n", status);
4685 				VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4686 				bioerror(bufp, status);
4687 			}
4688 
4689 			(void) vdc_depopulate_descriptor(vdcp, idx);
4690 
4691 			DMSG(vdcp, 1,
4692 			    "strategy complete req=%ld bytes resp=%ld bytes\n",
4693 			    bufp->b_bcount, ldep->dep->payload.nbytes);
4694 
4695 			if (status != 0 && vdcp->failfast_interval != 0) {
4696 				/*
4697 				 * The I/O has failed and failfast is enabled.
4698 				 * We need the failfast thread to check if the
4699 				 * failure is due to a reservation conflict.
4700 				 */
4701 				(void) vdc_failfast_io_queue(vdcp, bufp);
4702 			} else {
4703 				if (status == 0) {
4704 					op = (bufp->b_flags & B_READ) ?
4705 					    VD_OP_BREAD : VD_OP_BWRITE;
4706 					VD_UPDATE_IO_STATS(vdcp, op,
4707 					    ldep->dep->payload.nbytes);
4708 				}
4709 				VD_KSTAT_RUNQ_EXIT(vdcp);
4710 				DTRACE_IO1(done, buf_t *, bufp);
4711 				biodone(bufp);
4712 			}
4713 			break;
4714 
4715 		default:
4716 			ASSERT(0);
4717 		}
4718 	}
4719 
4720 	/* let the arrival signal propogate */
4721 	mutex_exit(&vdcp->lock);
4722 
4723 	/* probe gives the count of how many entries were processed */
4724 	DTRACE_PROBE2(processed, int, 1, vdc_t *, vdcp);
4725 
4726 	return (0);
4727 }
4728 
4729 
4730 /*
4731  * Function:
4732  *	vdc_handle_ver_msg()
4733  *
4734  * Description:
4735  *
4736  * Arguments:
4737  *	vdc	- soft state pointer for this instance of the device driver.
4738  *	ver_msg	- LDC message sent by vDisk server
4739  *
4740  * Return Code:
4741  *	0	- Success
4742  */
4743 static int
4744 vdc_handle_ver_msg(vdc_t *vdc, vio_ver_msg_t *ver_msg)
4745 {
4746 	int status = 0;
4747 
4748 	ASSERT(vdc != NULL);
4749 	ASSERT(mutex_owned(&vdc->lock));
4750 
4751 	if (ver_msg->tag.vio_subtype_env != VIO_VER_INFO) {
4752 		return (EPROTO);
4753 	}
4754 
4755 	if (ver_msg->dev_class != VDEV_DISK_SERVER) {
4756 		return (EINVAL);
4757 	}
4758 
4759 	switch (ver_msg->tag.vio_subtype) {
4760 	case VIO_SUBTYPE_ACK:
4761 		/*
4762 		 * We check to see if the version returned is indeed supported
4763 		 * (The server may have also adjusted the minor number downwards
4764 		 * and if so 'ver_msg' will contain the actual version agreed)
4765 		 */
4766 		if (vdc_is_supported_version(ver_msg)) {
4767 			vdc->ver.major = ver_msg->ver_major;
4768 			vdc->ver.minor = ver_msg->ver_minor;
4769 			ASSERT(vdc->ver.major > 0);
4770 		} else {
4771 			status = EPROTO;
4772 		}
4773 		break;
4774 
4775 	case VIO_SUBTYPE_NACK:
4776 		/*
4777 		 * call vdc_is_supported_version() which will return the next
4778 		 * supported version (if any) in 'ver_msg'
4779 		 */
4780 		(void) vdc_is_supported_version(ver_msg);
4781 		if (ver_msg->ver_major > 0) {
4782 			size_t len = sizeof (*ver_msg);
4783 
4784 			ASSERT(vdc->ver.major > 0);
4785 
4786 			/* reset the necessary fields and resend */
4787 			ver_msg->tag.vio_subtype = VIO_SUBTYPE_INFO;
4788 			ver_msg->dev_class = VDEV_DISK;
4789 
4790 			status = vdc_send(vdc, (caddr_t)ver_msg, &len);
4791 			DMSG(vdc, 0, "[%d] Resend VER info (LDC status = %d)\n",
4792 			    vdc->instance, status);
4793 			if (len != sizeof (*ver_msg))
4794 				status = EBADMSG;
4795 		} else {
4796 			DMSG(vdc, 0, "[%d] No common version with vDisk server",
4797 			    vdc->instance);
4798 			status = ENOTSUP;
4799 		}
4800 
4801 		break;
4802 	case VIO_SUBTYPE_INFO:
4803 		/*
4804 		 * Handle the case where vds starts handshake
4805 		 * (for now only vdc is the instigator)
4806 		 */
4807 		status = ENOTSUP;
4808 		break;
4809 
4810 	default:
4811 		status = EINVAL;
4812 		break;
4813 	}
4814 
4815 	return (status);
4816 }
4817 
4818 /*
4819  * Function:
4820  *	vdc_handle_attr_msg()
4821  *
4822  * Description:
4823  *
4824  * Arguments:
4825  *	vdc	- soft state pointer for this instance of the device driver.
4826  *	attr_msg	- LDC message sent by vDisk server
4827  *
4828  * Return Code:
4829  *	0	- Success
4830  */
4831 static int
4832 vdc_handle_attr_msg(vdc_t *vdc, vd_attr_msg_t *attr_msg)
4833 {
4834 	int status = 0;
4835 
4836 	ASSERT(vdc != NULL);
4837 	ASSERT(mutex_owned(&vdc->lock));
4838 
4839 	if (attr_msg->tag.vio_subtype_env != VIO_ATTR_INFO) {
4840 		return (EPROTO);
4841 	}
4842 
4843 	switch (attr_msg->tag.vio_subtype) {
4844 	case VIO_SUBTYPE_ACK:
4845 		/*
4846 		 * We now verify the attributes sent by vds.
4847 		 */
4848 		if (attr_msg->vdisk_size == 0) {
4849 			DMSG(vdc, 0, "[%d] Invalid disk size from vds",
4850 			    vdc->instance);
4851 			status = EINVAL;
4852 			break;
4853 		}
4854 
4855 		if (attr_msg->max_xfer_sz == 0) {
4856 			DMSG(vdc, 0, "[%d] Invalid transfer size from vds",
4857 			    vdc->instance);
4858 			status = EINVAL;
4859 			break;
4860 		}
4861 
4862 		if (attr_msg->vdisk_size == VD_SIZE_UNKNOWN) {
4863 			DMSG(vdc, 0, "[%d] Unknown disk size from vds",
4864 			    vdc->instance);
4865 			attr_msg->vdisk_size = 0;
4866 		}
4867 		/* update disk, block and transfer sizes */
4868 		vdc_update_size(vdc, attr_msg->vdisk_size,
4869 		    attr_msg->vdisk_block_size, attr_msg->max_xfer_sz);
4870 		vdc->vdisk_type = attr_msg->vdisk_type;
4871 		vdc->operations = attr_msg->operations;
4872 		if (vio_ver_is_supported(vdc->ver, 1, 1))
4873 			vdc->vdisk_media = attr_msg->vdisk_media;
4874 		else
4875 			vdc->vdisk_media = 0;
4876 
4877 		DMSG(vdc, 0, "[%d] max_xfer_sz: sent %lx acked %lx\n",
4878 		    vdc->instance, vdc->max_xfer_sz, attr_msg->max_xfer_sz);
4879 		DMSG(vdc, 0, "[%d] vdisk_block_size: sent %lx acked %x\n",
4880 		    vdc->instance, vdc->block_size,
4881 		    attr_msg->vdisk_block_size);
4882 
4883 		if ((attr_msg->xfer_mode != VIO_DRING_MODE_V1_0) ||
4884 		    (attr_msg->vdisk_size > INT64_MAX) ||
4885 		    (attr_msg->operations == 0) ||
4886 		    (attr_msg->vdisk_type > VD_DISK_TYPE_DISK)) {
4887 			DMSG(vdc, 0, "[%d] Invalid attributes from vds",
4888 			    vdc->instance);
4889 			status = EINVAL;
4890 			break;
4891 		}
4892 
4893 		/*
4894 		 * Now that we have received all attributes we can create a
4895 		 * fake geometry for the disk.
4896 		 */
4897 		vdc_create_fake_geometry(vdc);
4898 		break;
4899 
4900 	case VIO_SUBTYPE_NACK:
4901 		/*
4902 		 * vds could not handle the attributes we sent so we
4903 		 * stop negotiating.
4904 		 */
4905 		status = EPROTO;
4906 		break;
4907 
4908 	case VIO_SUBTYPE_INFO:
4909 		/*
4910 		 * Handle the case where vds starts the handshake
4911 		 * (for now; vdc is the only supported instigatior)
4912 		 */
4913 		status = ENOTSUP;
4914 		break;
4915 
4916 	default:
4917 		status = ENOTSUP;
4918 		break;
4919 	}
4920 
4921 	return (status);
4922 }
4923 
4924 /*
4925  * Function:
4926  *	vdc_handle_dring_reg_msg()
4927  *
4928  * Description:
4929  *
4930  * Arguments:
4931  *	vdc		- soft state pointer for this instance of the driver.
4932  *	dring_msg	- LDC message sent by vDisk server
4933  *
4934  * Return Code:
4935  *	0	- Success
4936  */
4937 static int
4938 vdc_handle_dring_reg_msg(vdc_t *vdc, vio_dring_reg_msg_t *dring_msg)
4939 {
4940 	int		status = 0;
4941 
4942 	ASSERT(vdc != NULL);
4943 	ASSERT(mutex_owned(&vdc->lock));
4944 
4945 	if (dring_msg->tag.vio_subtype_env != VIO_DRING_REG) {
4946 		return (EPROTO);
4947 	}
4948 
4949 	switch (dring_msg->tag.vio_subtype) {
4950 	case VIO_SUBTYPE_ACK:
4951 		/* save the received dring_ident */
4952 		vdc->dring_ident = dring_msg->dring_ident;
4953 		DMSG(vdc, 0, "[%d] Received dring ident=0x%lx\n",
4954 		    vdc->instance, vdc->dring_ident);
4955 		break;
4956 
4957 	case VIO_SUBTYPE_NACK:
4958 		/*
4959 		 * vds could not handle the DRing info we sent so we
4960 		 * stop negotiating.
4961 		 */
4962 		DMSG(vdc, 0, "[%d] server could not register DRing\n",
4963 		    vdc->instance);
4964 		status = EPROTO;
4965 		break;
4966 
4967 	case VIO_SUBTYPE_INFO:
4968 		/*
4969 		 * Handle the case where vds starts handshake
4970 		 * (for now only vdc is the instigatior)
4971 		 */
4972 		status = ENOTSUP;
4973 		break;
4974 	default:
4975 		status = ENOTSUP;
4976 	}
4977 
4978 	return (status);
4979 }
4980 
4981 /*
4982  * Function:
4983  *	vdc_verify_seq_num()
4984  *
4985  * Description:
4986  *	This functions verifies that the sequence number sent back by the vDisk
4987  *	server with the latest message is what is expected (i.e. it is greater
4988  *	than the last seq num sent by the vDisk server and less than or equal
4989  *	to the last seq num generated by vdc).
4990  *
4991  *	It then checks the request ID to see if any requests need processing
4992  *	in the DRing.
4993  *
4994  * Arguments:
4995  *	vdc		- soft state pointer for this instance of the driver.
4996  *	dring_msg	- pointer to the LDC message sent by vds
4997  *
4998  * Return Code:
4999  *	VDC_SEQ_NUM_TODO	- Message needs to be processed
5000  *	VDC_SEQ_NUM_SKIP	- Message has already been processed
5001  *	VDC_SEQ_NUM_INVALID	- The seq numbers are so out of sync,
5002  *				  vdc cannot deal with them
5003  */
5004 static int
5005 vdc_verify_seq_num(vdc_t *vdc, vio_dring_msg_t *dring_msg)
5006 {
5007 	ASSERT(vdc != NULL);
5008 	ASSERT(dring_msg != NULL);
5009 	ASSERT(mutex_owned(&vdc->lock));
5010 
5011 	/*
5012 	 * Check to see if the messages were responded to in the correct
5013 	 * order by vds.
5014 	 */
5015 	if ((dring_msg->seq_num <= vdc->seq_num_reply) ||
5016 	    (dring_msg->seq_num > vdc->seq_num)) {
5017 		DMSG(vdc, 0, "?[%d] Bogus sequence_number %lu: "
5018 		    "%lu > expected <= %lu (last proc req %lu sent %lu)\n",
5019 		    vdc->instance, dring_msg->seq_num,
5020 		    vdc->seq_num_reply, vdc->seq_num,
5021 		    vdc->req_id_proc, vdc->req_id);
5022 		return (VDC_SEQ_NUM_INVALID);
5023 	}
5024 	vdc->seq_num_reply = dring_msg->seq_num;
5025 
5026 	if (vdc->req_id_proc < vdc->req_id)
5027 		return (VDC_SEQ_NUM_TODO);
5028 	else
5029 		return (VDC_SEQ_NUM_SKIP);
5030 }
5031 
5032 
5033 /*
5034  * Function:
5035  *	vdc_is_supported_version()
5036  *
5037  * Description:
5038  *	This routine checks if the major/minor version numbers specified in
5039  *	'ver_msg' are supported. If not it finds the next version that is
5040  *	in the supported version list 'vdc_version[]' and sets the fields in
5041  *	'ver_msg' to those values
5042  *
5043  * Arguments:
5044  *	ver_msg	- LDC message sent by vDisk server
5045  *
5046  * Return Code:
5047  *	B_TRUE	- Success
5048  *	B_FALSE	- Version not supported
5049  */
5050 static boolean_t
5051 vdc_is_supported_version(vio_ver_msg_t *ver_msg)
5052 {
5053 	int vdc_num_versions = sizeof (vdc_version) / sizeof (vdc_version[0]);
5054 
5055 	for (int i = 0; i < vdc_num_versions; i++) {
5056 		ASSERT(vdc_version[i].major > 0);
5057 		ASSERT((i == 0) ||
5058 		    (vdc_version[i].major < vdc_version[i-1].major));
5059 
5060 		/*
5061 		 * If the major versions match, adjust the minor version, if
5062 		 * necessary, down to the highest value supported by this
5063 		 * client. The server should support all minor versions lower
5064 		 * than the value it sent
5065 		 */
5066 		if (ver_msg->ver_major == vdc_version[i].major) {
5067 			if (ver_msg->ver_minor > vdc_version[i].minor) {
5068 				DMSGX(0,
5069 				    "Adjusting minor version from %u to %u",
5070 				    ver_msg->ver_minor, vdc_version[i].minor);
5071 				ver_msg->ver_minor = vdc_version[i].minor;
5072 			}
5073 			return (B_TRUE);
5074 		}
5075 
5076 		/*
5077 		 * If the message contains a higher major version number, set
5078 		 * the message's major/minor versions to the current values
5079 		 * and return false, so this message will get resent with
5080 		 * these values, and the server will potentially try again
5081 		 * with the same or a lower version
5082 		 */
5083 		if (ver_msg->ver_major > vdc_version[i].major) {
5084 			ver_msg->ver_major = vdc_version[i].major;
5085 			ver_msg->ver_minor = vdc_version[i].minor;
5086 			DMSGX(0, "Suggesting major/minor (0x%x/0x%x)\n",
5087 			    ver_msg->ver_major, ver_msg->ver_minor);
5088 
5089 			return (B_FALSE);
5090 		}
5091 
5092 		/*
5093 		 * Otherwise, the message's major version is less than the
5094 		 * current major version, so continue the loop to the next
5095 		 * (lower) supported version
5096 		 */
5097 	}
5098 
5099 	/*
5100 	 * No common version was found; "ground" the version pair in the
5101 	 * message to terminate negotiation
5102 	 */
5103 	ver_msg->ver_major = 0;
5104 	ver_msg->ver_minor = 0;
5105 
5106 	return (B_FALSE);
5107 }
5108 /* -------------------------------------------------------------------------- */
5109 
5110 /*
5111  * DKIO(7) support
5112  */
5113 
5114 typedef struct vdc_dk_arg {
5115 	struct dk_callback	dkc;
5116 	int			mode;
5117 	dev_t			dev;
5118 	vdc_t			*vdc;
5119 } vdc_dk_arg_t;
5120 
5121 /*
5122  * Function:
5123  * 	vdc_dkio_flush_cb()
5124  *
5125  * Description:
5126  *	This routine is a callback for DKIOCFLUSHWRITECACHE which can be called
5127  *	by kernel code.
5128  *
5129  * Arguments:
5130  *	arg	- a pointer to a vdc_dk_arg_t structure.
5131  */
5132 void
5133 vdc_dkio_flush_cb(void *arg)
5134 {
5135 	struct vdc_dk_arg	*dk_arg = (struct vdc_dk_arg *)arg;
5136 	struct dk_callback	*dkc = NULL;
5137 	vdc_t			*vdc = NULL;
5138 	int			rv;
5139 
5140 	if (dk_arg == NULL) {
5141 		cmn_err(CE_NOTE, "?[Unk] DKIOCFLUSHWRITECACHE arg is NULL\n");
5142 		return;
5143 	}
5144 	dkc = &dk_arg->dkc;
5145 	vdc = dk_arg->vdc;
5146 	ASSERT(vdc != NULL);
5147 
5148 	rv = vdc_do_sync_op(vdc, VD_OP_FLUSH, NULL, 0,
5149 	    VDCPART(dk_arg->dev), 0, CB_SYNC, 0, VIO_both_dir, B_TRUE);
5150 	if (rv != 0) {
5151 		DMSG(vdc, 0, "[%d] DKIOCFLUSHWRITECACHE failed %d : model %x\n",
5152 		    vdc->instance, rv,
5153 		    ddi_model_convert_from(dk_arg->mode & FMODELS));
5154 	}
5155 
5156 	/*
5157 	 * Trigger the call back to notify the caller the the ioctl call has
5158 	 * been completed.
5159 	 */
5160 	if ((dk_arg->mode & FKIOCTL) &&
5161 	    (dkc != NULL) &&
5162 	    (dkc->dkc_callback != NULL)) {
5163 		ASSERT(dkc->dkc_cookie != NULL);
5164 		(*dkc->dkc_callback)(dkc->dkc_cookie, rv);
5165 	}
5166 
5167 	/* Indicate that one less DKIO write flush is outstanding */
5168 	mutex_enter(&vdc->lock);
5169 	vdc->dkio_flush_pending--;
5170 	ASSERT(vdc->dkio_flush_pending >= 0);
5171 	mutex_exit(&vdc->lock);
5172 
5173 	/* free the mem that was allocated when the callback was dispatched */
5174 	kmem_free(arg, sizeof (vdc_dk_arg_t));
5175 }
5176 
5177 /*
5178  * Function:
5179  * 	vdc_dkio_gapart()
5180  *
5181  * Description:
5182  *	This function implements the DKIOCGAPART ioctl.
5183  *
5184  * Arguments:
5185  *	vdc	- soft state pointer
5186  *	arg	- a pointer to a dk_map[NDKMAP] or dk_map32[NDKMAP] structure
5187  *	flag	- ioctl flags
5188  */
5189 static int
5190 vdc_dkio_gapart(vdc_t *vdc, caddr_t arg, int flag)
5191 {
5192 	struct dk_geom *geom;
5193 	struct extvtoc *vtoc;
5194 	union {
5195 		struct dk_map map[NDKMAP];
5196 		struct dk_map32 map32[NDKMAP];
5197 	} data;
5198 	int i, rv, size;
5199 
5200 	mutex_enter(&vdc->lock);
5201 
5202 	if ((rv = vdc_validate_geometry(vdc)) != 0) {
5203 		mutex_exit(&vdc->lock);
5204 		return (rv);
5205 	}
5206 
5207 	if (vdc->vdisk_size > VD_OLDVTOC_LIMIT) {
5208 		mutex_exit(&vdc->lock);
5209 		return (EOVERFLOW);
5210 	}
5211 
5212 	vtoc = vdc->vtoc;
5213 	geom = vdc->geom;
5214 
5215 	if (ddi_model_convert_from(flag & FMODELS) == DDI_MODEL_ILP32) {
5216 
5217 		for (i = 0; i < vtoc->v_nparts; i++) {
5218 			data.map32[i].dkl_cylno = vtoc->v_part[i].p_start /
5219 			    (geom->dkg_nhead * geom->dkg_nsect);
5220 			data.map32[i].dkl_nblk = vtoc->v_part[i].p_size;
5221 		}
5222 		size = NDKMAP * sizeof (struct dk_map32);
5223 
5224 	} else {
5225 
5226 		for (i = 0; i < vtoc->v_nparts; i++) {
5227 			data.map[i].dkl_cylno = vtoc->v_part[i].p_start /
5228 			    (geom->dkg_nhead * geom->dkg_nsect);
5229 			data.map[i].dkl_nblk = vtoc->v_part[i].p_size;
5230 		}
5231 		size = NDKMAP * sizeof (struct dk_map);
5232 
5233 	}
5234 
5235 	mutex_exit(&vdc->lock);
5236 
5237 	if (ddi_copyout(&data, arg, size, flag) != 0)
5238 		return (EFAULT);
5239 
5240 	return (0);
5241 }
5242 
5243 /*
5244  * Function:
5245  * 	vdc_dkio_partition()
5246  *
5247  * Description:
5248  *	This function implements the DKIOCPARTITION ioctl.
5249  *
5250  * Arguments:
5251  *	vdc	- soft state pointer
5252  *	arg	- a pointer to a struct partition64 structure
5253  *	flag	- ioctl flags
5254  */
5255 static int
5256 vdc_dkio_partition(vdc_t *vdc, caddr_t arg, int flag)
5257 {
5258 	struct partition64 p64;
5259 	efi_gpt_t *gpt;
5260 	efi_gpe_t *gpe;
5261 	vd_efi_dev_t edev;
5262 	uint_t partno;
5263 	int rv;
5264 
5265 	if (ddi_copyin(arg, &p64, sizeof (struct partition64), flag)) {
5266 		return (EFAULT);
5267 	}
5268 
5269 	VD_EFI_DEV_SET(edev, vdc, vd_process_efi_ioctl);
5270 
5271 	if ((rv = vd_efi_alloc_and_read(&edev, &gpt, &gpe)) != 0) {
5272 		return (rv);
5273 	}
5274 
5275 	partno = p64.p_partno;
5276 
5277 	if (partno >= gpt->efi_gpt_NumberOfPartitionEntries) {
5278 		vd_efi_free(&edev, gpt, gpe);
5279 		return (ESRCH);
5280 	}
5281 
5282 	bcopy(&gpe[partno].efi_gpe_PartitionTypeGUID, &p64.p_type,
5283 	    sizeof (struct uuid));
5284 	p64.p_start = gpe[partno].efi_gpe_StartingLBA;
5285 	p64.p_size = gpe[partno].efi_gpe_EndingLBA - p64.p_start + 1;
5286 
5287 	if (ddi_copyout(&p64, arg, sizeof (struct partition64), flag)) {
5288 		vd_efi_free(&edev, gpt, gpe);
5289 		return (EFAULT);
5290 	}
5291 
5292 	vd_efi_free(&edev, gpt, gpe);
5293 	return (0);
5294 }
5295 
5296 /*
5297  * Function:
5298  * 	vdc_dioctl_rwcmd()
5299  *
5300  * Description:
5301  *	This function implements the DIOCTL_RWCMD ioctl. This ioctl is used
5302  *	for DKC_DIRECT disks to read or write at an absolute disk offset.
5303  *
5304  * Arguments:
5305  *	dev	- device
5306  *	arg	- a pointer to a dadkio_rwcmd or dadkio_rwcmd32 structure
5307  *	flag	- ioctl flags
5308  */
5309 static int
5310 vdc_dioctl_rwcmd(dev_t dev, caddr_t arg, int flag)
5311 {
5312 	struct dadkio_rwcmd32 rwcmd32;
5313 	struct dadkio_rwcmd rwcmd;
5314 	struct iovec aiov;
5315 	struct uio auio;
5316 	int rw, status;
5317 	struct buf *buf;
5318 
5319 	if (ddi_model_convert_from(flag & FMODELS) == DDI_MODEL_ILP32) {
5320 		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd32,
5321 		    sizeof (struct dadkio_rwcmd32), flag)) {
5322 			return (EFAULT);
5323 		}
5324 		rwcmd.cmd = rwcmd32.cmd;
5325 		rwcmd.flags = rwcmd32.flags;
5326 		rwcmd.blkaddr = (daddr_t)rwcmd32.blkaddr;
5327 		rwcmd.buflen = rwcmd32.buflen;
5328 		rwcmd.bufaddr = (caddr_t)(uintptr_t)rwcmd32.bufaddr;
5329 	} else {
5330 		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd,
5331 		    sizeof (struct dadkio_rwcmd), flag)) {
5332 			return (EFAULT);
5333 		}
5334 	}
5335 
5336 	switch (rwcmd.cmd) {
5337 	case DADKIO_RWCMD_READ:
5338 		rw = B_READ;
5339 		break;
5340 	case DADKIO_RWCMD_WRITE:
5341 		rw = B_WRITE;
5342 		break;
5343 	default:
5344 		return (EINVAL);
5345 	}
5346 
5347 	bzero((caddr_t)&aiov, sizeof (struct iovec));
5348 	aiov.iov_base   = rwcmd.bufaddr;
5349 	aiov.iov_len    = rwcmd.buflen;
5350 
5351 	bzero((caddr_t)&auio, sizeof (struct uio));
5352 	auio.uio_iov    = &aiov;
5353 	auio.uio_iovcnt = 1;
5354 	auio.uio_loffset = rwcmd.blkaddr * DEV_BSIZE;
5355 	auio.uio_resid  = rwcmd.buflen;
5356 	auio.uio_segflg = flag & FKIOCTL ? UIO_SYSSPACE : UIO_USERSPACE;
5357 
5358 	buf = kmem_alloc(sizeof (buf_t), KM_SLEEP);
5359 	bioinit(buf);
5360 	/*
5361 	 * We use the private field of buf to specify that this is an
5362 	 * I/O using an absolute offset.
5363 	 */
5364 	buf->b_private = (void *)VD_SLICE_NONE;
5365 
5366 	status = physio(vdc_strategy, buf, dev, rw, vdc_min, &auio);
5367 
5368 	biofini(buf);
5369 	kmem_free(buf, sizeof (buf_t));
5370 
5371 	return (status);
5372 }
5373 
5374 /*
5375  * Allocate a buffer for a VD_OP_SCSICMD operation. The size of the allocated
5376  * buffer is returned in alloc_len.
5377  */
5378 static vd_scsi_t *
5379 vdc_scsi_alloc(int cdb_len, int sense_len, int datain_len, int dataout_len,
5380     int *alloc_len)
5381 {
5382 	vd_scsi_t *vd_scsi;
5383 	int vd_scsi_len = VD_SCSI_SIZE;
5384 
5385 	vd_scsi_len += P2ROUNDUP(cdb_len, sizeof (uint64_t));
5386 	vd_scsi_len += P2ROUNDUP(sense_len, sizeof (uint64_t));
5387 	vd_scsi_len += P2ROUNDUP(datain_len, sizeof (uint64_t));
5388 	vd_scsi_len += P2ROUNDUP(dataout_len, sizeof (uint64_t));
5389 
5390 	ASSERT(vd_scsi_len % sizeof (uint64_t) == 0);
5391 
5392 	vd_scsi = kmem_zalloc(vd_scsi_len, KM_SLEEP);
5393 
5394 	vd_scsi->cdb_len = cdb_len;
5395 	vd_scsi->sense_len = sense_len;
5396 	vd_scsi->datain_len = datain_len;
5397 	vd_scsi->dataout_len = dataout_len;
5398 
5399 	*alloc_len = vd_scsi_len;
5400 
5401 	return (vd_scsi);
5402 }
5403 
5404 /*
5405  * Convert the status of a SCSI command to a Solaris return code.
5406  *
5407  * Arguments:
5408  *	vd_scsi		- The SCSI operation buffer.
5409  *	log_error	- indicate if an error message should be logged.
5410  *
5411  * Note that our SCSI error messages are rather primitive for the moment
5412  * and could be improved by decoding some data like the SCSI command and
5413  * the sense key.
5414  *
5415  * Return value:
5416  *	0		- Status is good.
5417  *	EACCES		- Status reports a reservation conflict.
5418  *	ENOTSUP		- Status reports a check condition and sense key
5419  *			  reports an illegal request.
5420  *	EIO		- Any other status.
5421  */
5422 static int
5423 vdc_scsi_status(vdc_t *vdc, vd_scsi_t *vd_scsi, boolean_t log_error)
5424 {
5425 	int rv;
5426 	char path_str[MAXPATHLEN];
5427 	char panic_str[VDC_RESV_CONFLICT_FMT_LEN + MAXPATHLEN];
5428 	union scsi_cdb *cdb;
5429 	struct scsi_extended_sense *sense;
5430 
5431 	if (vd_scsi->cmd_status == STATUS_GOOD)
5432 		/* no error */
5433 		return (0);
5434 
5435 	/* when the tunable vdc_scsi_log_error is true we log all errors */
5436 	if (vdc_scsi_log_error)
5437 		log_error = B_TRUE;
5438 
5439 	if (log_error) {
5440 		cmn_err(CE_WARN, "%s (vdc%d):\tError for Command: 0x%x)\n",
5441 		    ddi_pathname(vdc->dip, path_str), vdc->instance,
5442 		    GETCMD(VD_SCSI_DATA_CDB(vd_scsi)));
5443 	}
5444 
5445 	/* default returned value */
5446 	rv = EIO;
5447 
5448 	switch (vd_scsi->cmd_status) {
5449 
5450 	case STATUS_CHECK:
5451 	case STATUS_TERMINATED:
5452 		if (log_error)
5453 			cmn_err(CE_CONT, "\tCheck Condition Error\n");
5454 
5455 		/* check sense buffer */
5456 		if (vd_scsi->sense_len == 0 ||
5457 		    vd_scsi->sense_status != STATUS_GOOD) {
5458 			if (log_error)
5459 				cmn_err(CE_CONT, "\tNo Sense Data Available\n");
5460 			break;
5461 		}
5462 
5463 		sense = VD_SCSI_DATA_SENSE(vd_scsi);
5464 
5465 		if (log_error) {
5466 			cmn_err(CE_CONT, "\tSense Key:  0x%x\n"
5467 			    "\tASC: 0x%x, ASCQ: 0x%x\n",
5468 			    scsi_sense_key((uint8_t *)sense),
5469 			    scsi_sense_asc((uint8_t *)sense),
5470 			    scsi_sense_ascq((uint8_t *)sense));
5471 		}
5472 
5473 		if (scsi_sense_key((uint8_t *)sense) == KEY_ILLEGAL_REQUEST)
5474 			rv = ENOTSUP;
5475 		break;
5476 
5477 	case STATUS_BUSY:
5478 		if (log_error)
5479 			cmn_err(CE_NOTE, "\tDevice Busy\n");
5480 		break;
5481 
5482 	case STATUS_RESERVATION_CONFLICT:
5483 		/*
5484 		 * If the command was PERSISTENT_RESERVATION_[IN|OUT] then
5485 		 * reservation conflict could be due to various reasons like
5486 		 * incorrect keys, not registered or not reserved etc. So,
5487 		 * we should not panic in that case.
5488 		 */
5489 		cdb = VD_SCSI_DATA_CDB(vd_scsi);
5490 		if (vdc->failfast_interval != 0 &&
5491 		    cdb->scc_cmd != SCMD_PERSISTENT_RESERVE_IN &&
5492 		    cdb->scc_cmd != SCMD_PERSISTENT_RESERVE_OUT) {
5493 			/* failfast is enabled so we have to panic */
5494 			(void) snprintf(panic_str, sizeof (panic_str),
5495 			    VDC_RESV_CONFLICT_FMT_STR "%s",
5496 			    ddi_pathname(vdc->dip, path_str));
5497 			panic(panic_str);
5498 		}
5499 		if (log_error)
5500 			cmn_err(CE_NOTE, "\tReservation Conflict\n");
5501 		rv = EACCES;
5502 		break;
5503 
5504 	case STATUS_QFULL:
5505 		if (log_error)
5506 			cmn_err(CE_NOTE, "\tQueue Full\n");
5507 		break;
5508 
5509 	case STATUS_MET:
5510 	case STATUS_INTERMEDIATE:
5511 	case STATUS_SCSI2:
5512 	case STATUS_INTERMEDIATE_MET:
5513 	case STATUS_ACA_ACTIVE:
5514 		if (log_error)
5515 			cmn_err(CE_CONT,
5516 			    "\tUnexpected SCSI status received: 0x%x\n",
5517 			    vd_scsi->cmd_status);
5518 		break;
5519 
5520 	default:
5521 		if (log_error)
5522 			cmn_err(CE_CONT,
5523 			    "\tInvalid SCSI status received: 0x%x\n",
5524 			    vd_scsi->cmd_status);
5525 		break;
5526 	}
5527 
5528 	return (rv);
5529 }
5530 
5531 /*
5532  * Implemented the USCSICMD uscsi(7I) ioctl. This ioctl is converted to
5533  * a VD_OP_SCSICMD operation which is sent to the vdisk server. If a SCSI
5534  * reset is requested (i.e. a flag USCSI_RESET* is set) then the ioctl is
5535  * converted to a VD_OP_RESET operation.
5536  */
5537 static int
5538 vdc_uscsi_cmd(vdc_t *vdc, caddr_t arg, int mode)
5539 {
5540 	struct uscsi_cmd 	uscsi;
5541 	struct uscsi_cmd32	uscsi32;
5542 	vd_scsi_t 		*vd_scsi;
5543 	int 			vd_scsi_len;
5544 	union scsi_cdb		*cdb;
5545 	struct scsi_extended_sense *sense;
5546 	char 			*datain, *dataout;
5547 	size_t			cdb_len, datain_len, dataout_len, sense_len;
5548 	int 			rv;
5549 
5550 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5551 		if (ddi_copyin(arg, &uscsi32, sizeof (struct uscsi_cmd32),
5552 		    mode) != 0)
5553 			return (EFAULT);
5554 		uscsi_cmd32touscsi_cmd((&uscsi32), (&uscsi));
5555 	} else {
5556 		if (ddi_copyin(arg, &uscsi, sizeof (struct uscsi_cmd),
5557 		    mode) != 0)
5558 			return (EFAULT);
5559 	}
5560 
5561 	/* a uscsi reset is converted to a VD_OP_RESET operation */
5562 	if (uscsi.uscsi_flags & (USCSI_RESET | USCSI_RESET_LUN |
5563 	    USCSI_RESET_ALL)) {
5564 		rv = vdc_do_sync_op(vdc, VD_OP_RESET, NULL, 0, 0, 0, CB_SYNC,
5565 		    (void *)(uint64_t)mode, VIO_both_dir, B_TRUE);
5566 		return (rv);
5567 	}
5568 
5569 	/* cdb buffer length */
5570 	cdb_len = uscsi.uscsi_cdblen;
5571 
5572 	/* data in and out buffers length */
5573 	if (uscsi.uscsi_flags & USCSI_READ) {
5574 		datain_len = uscsi.uscsi_buflen;
5575 		dataout_len = 0;
5576 	} else {
5577 		datain_len = 0;
5578 		dataout_len = uscsi.uscsi_buflen;
5579 	}
5580 
5581 	/* sense buffer length */
5582 	if (uscsi.uscsi_flags & USCSI_RQENABLE)
5583 		sense_len = uscsi.uscsi_rqlen;
5584 	else
5585 		sense_len = 0;
5586 
5587 	/* allocate buffer for the VD_SCSICMD_OP operation */
5588 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, datain_len, dataout_len,
5589 	    &vd_scsi_len);
5590 
5591 	/*
5592 	 * The documentation of USCSI_ISOLATE and USCSI_DIAGNOSE is very vague,
5593 	 * but basically they prevent a SCSI command from being retried in case
5594 	 * of an error.
5595 	 */
5596 	if ((uscsi.uscsi_flags & USCSI_ISOLATE) ||
5597 	    (uscsi.uscsi_flags & USCSI_DIAGNOSE))
5598 		vd_scsi->options |= VD_SCSI_OPT_NORETRY;
5599 
5600 	/* set task attribute */
5601 	if (uscsi.uscsi_flags & USCSI_NOTAG) {
5602 		vd_scsi->task_attribute = 0;
5603 	} else {
5604 		if (uscsi.uscsi_flags & USCSI_HEAD)
5605 			vd_scsi->task_attribute = VD_SCSI_TASK_ACA;
5606 		else if (uscsi.uscsi_flags & USCSI_HTAG)
5607 			vd_scsi->task_attribute = VD_SCSI_TASK_HQUEUE;
5608 		else if (uscsi.uscsi_flags & USCSI_OTAG)
5609 			vd_scsi->task_attribute = VD_SCSI_TASK_ORDERED;
5610 		else
5611 			vd_scsi->task_attribute = 0;
5612 	}
5613 
5614 	/* set timeout */
5615 	vd_scsi->timeout = uscsi.uscsi_timeout;
5616 
5617 	/* copy-in cdb data */
5618 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5619 	if (ddi_copyin(uscsi.uscsi_cdb, cdb, cdb_len, mode) != 0) {
5620 		rv = EFAULT;
5621 		goto done;
5622 	}
5623 
5624 	/* keep a pointer to the sense buffer */
5625 	sense = VD_SCSI_DATA_SENSE(vd_scsi);
5626 
5627 	/* keep a pointer to the data-in buffer */
5628 	datain = (char *)VD_SCSI_DATA_IN(vd_scsi);
5629 
5630 	/* copy-in request data to the data-out buffer */
5631 	dataout = (char *)VD_SCSI_DATA_OUT(vd_scsi);
5632 	if (!(uscsi.uscsi_flags & USCSI_READ)) {
5633 		if (ddi_copyin(uscsi.uscsi_bufaddr, dataout, dataout_len,
5634 		    mode)) {
5635 			rv = EFAULT;
5636 			goto done;
5637 		}
5638 	}
5639 
5640 	/* submit the request */
5641 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5642 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5643 
5644 	if (rv != 0)
5645 		goto done;
5646 
5647 	/* update scsi status */
5648 	uscsi.uscsi_status = vd_scsi->cmd_status;
5649 
5650 	/* update sense data */
5651 	if ((uscsi.uscsi_flags & USCSI_RQENABLE) &&
5652 	    (uscsi.uscsi_status == STATUS_CHECK ||
5653 	    uscsi.uscsi_status == STATUS_TERMINATED)) {
5654 
5655 		uscsi.uscsi_rqstatus = vd_scsi->sense_status;
5656 
5657 		if (uscsi.uscsi_rqstatus == STATUS_GOOD) {
5658 			uscsi.uscsi_rqresid = uscsi.uscsi_rqlen -
5659 			    vd_scsi->sense_len;
5660 			if (ddi_copyout(sense, uscsi.uscsi_rqbuf,
5661 			    vd_scsi->sense_len, mode) != 0) {
5662 				rv = EFAULT;
5663 				goto done;
5664 			}
5665 		}
5666 	}
5667 
5668 	/* update request data */
5669 	if (uscsi.uscsi_status == STATUS_GOOD) {
5670 		if (uscsi.uscsi_flags & USCSI_READ) {
5671 			uscsi.uscsi_resid = uscsi.uscsi_buflen -
5672 			    vd_scsi->datain_len;
5673 			if (ddi_copyout(datain, uscsi.uscsi_bufaddr,
5674 			    vd_scsi->datain_len, mode) != 0) {
5675 				rv = EFAULT;
5676 				goto done;
5677 			}
5678 		} else {
5679 			uscsi.uscsi_resid = uscsi.uscsi_buflen -
5680 			    vd_scsi->dataout_len;
5681 		}
5682 	}
5683 
5684 	/* copy-out result */
5685 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5686 		uscsi_cmdtouscsi_cmd32((&uscsi), (&uscsi32));
5687 		if (ddi_copyout(&uscsi32, arg, sizeof (struct uscsi_cmd32),
5688 		    mode) != 0) {
5689 			rv = EFAULT;
5690 			goto done;
5691 		}
5692 	} else {
5693 		if (ddi_copyout(&uscsi, arg, sizeof (struct uscsi_cmd),
5694 		    mode) != 0) {
5695 			rv = EFAULT;
5696 			goto done;
5697 		}
5698 	}
5699 
5700 	/* get the return code from the SCSI command status */
5701 	rv = vdc_scsi_status(vdc, vd_scsi,
5702 	    !(uscsi.uscsi_flags & USCSI_SILENT));
5703 
5704 done:
5705 	kmem_free(vd_scsi, vd_scsi_len);
5706 	return (rv);
5707 }
5708 
5709 /*
5710  * Create a VD_OP_SCSICMD buffer for a SCSI PERSISTENT IN command.
5711  *
5712  * Arguments:
5713  *	cmd		- SCSI PERSISTENT IN command
5714  *	len		- length of the SCSI input buffer
5715  *	vd_scsi_len	- return the length of the allocated buffer
5716  *
5717  * Returned Value:
5718  *	a pointer to the allocated VD_OP_SCSICMD buffer.
5719  */
5720 static vd_scsi_t *
5721 vdc_scsi_alloc_persistent_in(uchar_t cmd, int len, int *vd_scsi_len)
5722 {
5723 	int cdb_len, sense_len, datain_len, dataout_len;
5724 	vd_scsi_t *vd_scsi;
5725 	union scsi_cdb *cdb;
5726 
5727 	cdb_len = CDB_GROUP1;
5728 	sense_len = sizeof (struct scsi_extended_sense);
5729 	datain_len = len;
5730 	dataout_len = 0;
5731 
5732 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, datain_len, dataout_len,
5733 	    vd_scsi_len);
5734 
5735 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5736 
5737 	/* set cdb */
5738 	cdb->scc_cmd = SCMD_PERSISTENT_RESERVE_IN;
5739 	cdb->cdb_opaque[1] = cmd;
5740 	FORMG1COUNT(cdb, datain_len);
5741 
5742 	vd_scsi->timeout = vdc_scsi_timeout;
5743 
5744 	return (vd_scsi);
5745 }
5746 
5747 /*
5748  * Create a VD_OP_SCSICMD buffer for a SCSI PERSISTENT OUT command.
5749  *
5750  * Arguments:
5751  *	cmd		- SCSI PERSISTENT OUT command
5752  *	len		- length of the SCSI output buffer
5753  *	vd_scsi_len	- return the length of the allocated buffer
5754  *
5755  * Returned Code:
5756  *	a pointer to the allocated VD_OP_SCSICMD buffer.
5757  */
5758 static vd_scsi_t *
5759 vdc_scsi_alloc_persistent_out(uchar_t cmd, int len, int *vd_scsi_len)
5760 {
5761 	int cdb_len, sense_len, datain_len, dataout_len;
5762 	vd_scsi_t *vd_scsi;
5763 	union scsi_cdb *cdb;
5764 
5765 	cdb_len = CDB_GROUP1;
5766 	sense_len = sizeof (struct scsi_extended_sense);
5767 	datain_len = 0;
5768 	dataout_len = len;
5769 
5770 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, datain_len, dataout_len,
5771 	    vd_scsi_len);
5772 
5773 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5774 
5775 	/* set cdb */
5776 	cdb->scc_cmd = SCMD_PERSISTENT_RESERVE_OUT;
5777 	cdb->cdb_opaque[1] = cmd;
5778 	FORMG1COUNT(cdb, dataout_len);
5779 
5780 	vd_scsi->timeout = vdc_scsi_timeout;
5781 
5782 	return (vd_scsi);
5783 }
5784 
5785 /*
5786  * Implement the MHIOCGRP_INKEYS mhd(7i) ioctl. The ioctl is converted
5787  * to a SCSI PERSISTENT IN READ KEYS command which is sent to the vdisk
5788  * server with a VD_OP_SCSICMD operation.
5789  */
5790 static int
5791 vdc_mhd_inkeys(vdc_t *vdc, caddr_t arg, int mode)
5792 {
5793 	vd_scsi_t *vd_scsi;
5794 	mhioc_inkeys_t inkeys;
5795 	mhioc_key_list_t klist;
5796 	struct mhioc_inkeys32 inkeys32;
5797 	struct mhioc_key_list32 klist32;
5798 	sd_prin_readkeys_t *scsi_keys;
5799 	void *user_keys;
5800 	int vd_scsi_len;
5801 	int listsize, listlen, rv;
5802 
5803 	/* copyin arguments */
5804 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5805 		rv = ddi_copyin(arg, &inkeys32, sizeof (inkeys32), mode);
5806 		if (rv != 0)
5807 			return (EFAULT);
5808 
5809 		rv = ddi_copyin((caddr_t)(uintptr_t)inkeys32.li, &klist32,
5810 		    sizeof (klist32), mode);
5811 		if (rv != 0)
5812 			return (EFAULT);
5813 
5814 		listsize = klist32.listsize;
5815 	} else {
5816 		rv = ddi_copyin(arg, &inkeys, sizeof (inkeys), mode);
5817 		if (rv != 0)
5818 			return (EFAULT);
5819 
5820 		rv = ddi_copyin(inkeys.li, &klist, sizeof (klist), mode);
5821 		if (rv != 0)
5822 			return (EFAULT);
5823 
5824 		listsize = klist.listsize;
5825 	}
5826 
5827 	/* build SCSI VD_OP request */
5828 	vd_scsi = vdc_scsi_alloc_persistent_in(SD_READ_KEYS,
5829 	    sizeof (sd_prin_readkeys_t) - sizeof (caddr_t) +
5830 	    (sizeof (mhioc_resv_key_t) * listsize), &vd_scsi_len);
5831 
5832 	scsi_keys = (sd_prin_readkeys_t *)VD_SCSI_DATA_IN(vd_scsi);
5833 
5834 	/* submit the request */
5835 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5836 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5837 
5838 	if (rv != 0)
5839 		goto done;
5840 
5841 	listlen = scsi_keys->len / MHIOC_RESV_KEY_SIZE;
5842 
5843 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5844 		inkeys32.generation = scsi_keys->generation;
5845 		rv = ddi_copyout(&inkeys32, arg, sizeof (inkeys32), mode);
5846 		if (rv != 0) {
5847 			rv = EFAULT;
5848 			goto done;
5849 		}
5850 
5851 		klist32.listlen = listlen;
5852 		rv = ddi_copyout(&klist32, (caddr_t)(uintptr_t)inkeys32.li,
5853 		    sizeof (klist32), mode);
5854 		if (rv != 0) {
5855 			rv = EFAULT;
5856 			goto done;
5857 		}
5858 
5859 		user_keys = (caddr_t)(uintptr_t)klist32.list;
5860 	} else {
5861 		inkeys.generation = scsi_keys->generation;
5862 		rv = ddi_copyout(&inkeys, arg, sizeof (inkeys), mode);
5863 		if (rv != 0) {
5864 			rv = EFAULT;
5865 			goto done;
5866 		}
5867 
5868 		klist.listlen = listlen;
5869 		rv = ddi_copyout(&klist, inkeys.li, sizeof (klist), mode);
5870 		if (rv != 0) {
5871 			rv = EFAULT;
5872 			goto done;
5873 		}
5874 
5875 		user_keys = klist.list;
5876 	}
5877 
5878 	/* copy out keys */
5879 	if (listlen > 0 && listsize > 0) {
5880 		if (listsize < listlen)
5881 			listlen = listsize;
5882 		rv = ddi_copyout(&scsi_keys->keylist, user_keys,
5883 		    listlen * MHIOC_RESV_KEY_SIZE, mode);
5884 		if (rv != 0)
5885 			rv = EFAULT;
5886 	}
5887 
5888 	if (rv == 0)
5889 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
5890 
5891 done:
5892 	kmem_free(vd_scsi, vd_scsi_len);
5893 
5894 	return (rv);
5895 }
5896 
5897 /*
5898  * Implement the MHIOCGRP_INRESV mhd(7i) ioctl. The ioctl is converted
5899  * to a SCSI PERSISTENT IN READ RESERVATION command which is sent to
5900  * the vdisk server with a VD_OP_SCSICMD operation.
5901  */
5902 static int
5903 vdc_mhd_inresv(vdc_t *vdc, caddr_t arg, int mode)
5904 {
5905 	vd_scsi_t *vd_scsi;
5906 	mhioc_inresvs_t inresv;
5907 	mhioc_resv_desc_list_t rlist;
5908 	struct mhioc_inresvs32 inresv32;
5909 	struct mhioc_resv_desc_list32 rlist32;
5910 	mhioc_resv_desc_t mhd_resv;
5911 	sd_prin_readresv_t *scsi_resv;
5912 	sd_readresv_desc_t *resv;
5913 	mhioc_resv_desc_t *user_resv;
5914 	int vd_scsi_len;
5915 	int listsize, listlen, i, rv;
5916 
5917 	/* copyin arguments */
5918 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5919 		rv = ddi_copyin(arg, &inresv32, sizeof (inresv32), mode);
5920 		if (rv != 0)
5921 			return (EFAULT);
5922 
5923 		rv = ddi_copyin((caddr_t)(uintptr_t)inresv32.li, &rlist32,
5924 		    sizeof (rlist32), mode);
5925 		if (rv != 0)
5926 			return (EFAULT);
5927 
5928 		listsize = rlist32.listsize;
5929 	} else {
5930 		rv = ddi_copyin(arg, &inresv, sizeof (inresv), mode);
5931 		if (rv != 0)
5932 			return (EFAULT);
5933 
5934 		rv = ddi_copyin(inresv.li, &rlist, sizeof (rlist), mode);
5935 		if (rv != 0)
5936 			return (EFAULT);
5937 
5938 		listsize = rlist.listsize;
5939 	}
5940 
5941 	/* build SCSI VD_OP request */
5942 	vd_scsi = vdc_scsi_alloc_persistent_in(SD_READ_RESV,
5943 	    sizeof (sd_prin_readresv_t) - sizeof (caddr_t) +
5944 	    (SCSI3_RESV_DESC_LEN * listsize), &vd_scsi_len);
5945 
5946 	scsi_resv = (sd_prin_readresv_t *)VD_SCSI_DATA_IN(vd_scsi);
5947 
5948 	/* submit the request */
5949 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5950 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5951 
5952 	if (rv != 0)
5953 		goto done;
5954 
5955 	listlen = scsi_resv->len / SCSI3_RESV_DESC_LEN;
5956 
5957 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5958 		inresv32.generation = scsi_resv->generation;
5959 		rv = ddi_copyout(&inresv32, arg, sizeof (inresv32), mode);
5960 		if (rv != 0) {
5961 			rv = EFAULT;
5962 			goto done;
5963 		}
5964 
5965 		rlist32.listlen = listlen;
5966 		rv = ddi_copyout(&rlist32, (caddr_t)(uintptr_t)inresv32.li,
5967 		    sizeof (rlist32), mode);
5968 		if (rv != 0) {
5969 			rv = EFAULT;
5970 			goto done;
5971 		}
5972 
5973 		user_resv = (mhioc_resv_desc_t *)(uintptr_t)rlist32.list;
5974 	} else {
5975 		inresv.generation = scsi_resv->generation;
5976 		rv = ddi_copyout(&inresv, arg, sizeof (inresv), mode);
5977 		if (rv != 0) {
5978 			rv = EFAULT;
5979 			goto done;
5980 		}
5981 
5982 		rlist.listlen = listlen;
5983 		rv = ddi_copyout(&rlist, inresv.li, sizeof (rlist), mode);
5984 		if (rv != 0) {
5985 			rv = EFAULT;
5986 			goto done;
5987 		}
5988 
5989 		user_resv = rlist.list;
5990 	}
5991 
5992 	/* copy out reservations */
5993 	if (listsize > 0 && listlen > 0) {
5994 		if (listsize < listlen)
5995 			listlen = listsize;
5996 		resv = (sd_readresv_desc_t *)&scsi_resv->readresv_desc;
5997 
5998 		for (i = 0; i < listlen; i++) {
5999 			mhd_resv.type = resv->type;
6000 			mhd_resv.scope = resv->scope;
6001 			mhd_resv.scope_specific_addr =
6002 			    BE_32(resv->scope_specific_addr);
6003 			bcopy(&resv->resvkey, &mhd_resv.key,
6004 			    MHIOC_RESV_KEY_SIZE);
6005 
6006 			rv = ddi_copyout(&mhd_resv, user_resv,
6007 			    sizeof (mhd_resv), mode);
6008 			if (rv != 0) {
6009 				rv = EFAULT;
6010 				goto done;
6011 			}
6012 			resv++;
6013 			user_resv++;
6014 		}
6015 	}
6016 
6017 	if (rv == 0)
6018 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6019 
6020 done:
6021 	kmem_free(vd_scsi, vd_scsi_len);
6022 	return (rv);
6023 }
6024 
6025 /*
6026  * Implement the MHIOCGRP_REGISTER mhd(7i) ioctl. The ioctl is converted
6027  * to a SCSI PERSISTENT OUT REGISTER command which is sent to the vdisk
6028  * server with a VD_OP_SCSICMD operation.
6029  */
6030 static int
6031 vdc_mhd_register(vdc_t *vdc, caddr_t arg, int mode)
6032 {
6033 	vd_scsi_t *vd_scsi;
6034 	sd_prout_t *scsi_prout;
6035 	mhioc_register_t mhd_reg;
6036 	int vd_scsi_len, rv;
6037 
6038 	/* copyin arguments */
6039 	rv = ddi_copyin(arg, &mhd_reg, sizeof (mhd_reg), mode);
6040 	if (rv != 0)
6041 		return (EFAULT);
6042 
6043 	/* build SCSI VD_OP request */
6044 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_REGISTER,
6045 	    sizeof (sd_prout_t), &vd_scsi_len);
6046 
6047 	/* set parameters */
6048 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
6049 	bcopy(mhd_reg.oldkey.key, scsi_prout->res_key, MHIOC_RESV_KEY_SIZE);
6050 	bcopy(mhd_reg.newkey.key, scsi_prout->service_key, MHIOC_RESV_KEY_SIZE);
6051 	scsi_prout->aptpl = (uchar_t)mhd_reg.aptpl;
6052 
6053 	/* submit the request */
6054 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6055 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
6056 
6057 	if (rv == 0)
6058 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6059 
6060 	kmem_free(vd_scsi, vd_scsi_len);
6061 	return (rv);
6062 }
6063 
6064 /*
6065  * Implement the MHIOCGRP_RESERVE mhd(7i) ioctl. The ioctl is converted
6066  * to a SCSI PERSISTENT OUT RESERVE command which is sent to the vdisk
6067  * server with a VD_OP_SCSICMD operation.
6068  */
6069 static int
6070 vdc_mhd_reserve(vdc_t *vdc, caddr_t arg, int mode)
6071 {
6072 	union scsi_cdb *cdb;
6073 	vd_scsi_t *vd_scsi;
6074 	sd_prout_t *scsi_prout;
6075 	mhioc_resv_desc_t mhd_resv;
6076 	int vd_scsi_len, rv;
6077 
6078 	/* copyin arguments */
6079 	rv = ddi_copyin(arg, &mhd_resv, sizeof (mhd_resv), mode);
6080 	if (rv != 0)
6081 		return (EFAULT);
6082 
6083 	/* build SCSI VD_OP request */
6084 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_RESERVE,
6085 	    sizeof (sd_prout_t), &vd_scsi_len);
6086 
6087 	/* set parameters */
6088 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
6089 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
6090 	bcopy(mhd_resv.key.key, scsi_prout->res_key, MHIOC_RESV_KEY_SIZE);
6091 	scsi_prout->scope_address = mhd_resv.scope_specific_addr;
6092 	cdb->cdb_opaque[2] = mhd_resv.type;
6093 
6094 	/* submit the request */
6095 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6096 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
6097 
6098 	if (rv == 0)
6099 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6100 
6101 	kmem_free(vd_scsi, vd_scsi_len);
6102 	return (rv);
6103 }
6104 
6105 /*
6106  * Implement the MHIOCGRP_PREEMPTANDABORT mhd(7i) ioctl. The ioctl is
6107  * converted to a SCSI PERSISTENT OUT PREEMPT AND ABORT command which
6108  * is sent to the vdisk server with a VD_OP_SCSICMD operation.
6109  */
6110 static int
6111 vdc_mhd_preemptabort(vdc_t *vdc, caddr_t arg, int mode)
6112 {
6113 	union scsi_cdb *cdb;
6114 	vd_scsi_t *vd_scsi;
6115 	sd_prout_t *scsi_prout;
6116 	mhioc_preemptandabort_t mhd_preempt;
6117 	int vd_scsi_len, rv;
6118 
6119 	/* copyin arguments */
6120 	rv = ddi_copyin(arg, &mhd_preempt, sizeof (mhd_preempt), mode);
6121 	if (rv != 0)
6122 		return (EFAULT);
6123 
6124 	/* build SCSI VD_OP request */
6125 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_PREEMPTANDABORT,
6126 	    sizeof (sd_prout_t), &vd_scsi_len);
6127 
6128 	/* set parameters */
6129 	vd_scsi->task_attribute = VD_SCSI_TASK_ACA;
6130 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
6131 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
6132 	bcopy(mhd_preempt.resvdesc.key.key, scsi_prout->res_key,
6133 	    MHIOC_RESV_KEY_SIZE);
6134 	bcopy(mhd_preempt.victim_key.key, scsi_prout->service_key,
6135 	    MHIOC_RESV_KEY_SIZE);
6136 	scsi_prout->scope_address = mhd_preempt.resvdesc.scope_specific_addr;
6137 	cdb->cdb_opaque[2] = mhd_preempt.resvdesc.type;
6138 
6139 	/* submit the request */
6140 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6141 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
6142 
6143 	if (rv == 0)
6144 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6145 
6146 	kmem_free(vd_scsi, vd_scsi_len);
6147 	return (rv);
6148 }
6149 
6150 /*
6151  * Implement the MHIOCGRP_REGISTERANDIGNOREKEY mhd(7i) ioctl. The ioctl
6152  * is converted to a SCSI PERSISTENT OUT REGISTER AND IGNORE EXISTING KEY
6153  * command which is sent to the vdisk server with a VD_OP_SCSICMD operation.
6154  */
6155 static int
6156 vdc_mhd_registerignore(vdc_t *vdc, caddr_t arg, int mode)
6157 {
6158 	vd_scsi_t *vd_scsi;
6159 	sd_prout_t *scsi_prout;
6160 	mhioc_registerandignorekey_t mhd_regi;
6161 	int vd_scsi_len, rv;
6162 
6163 	/* copyin arguments */
6164 	rv = ddi_copyin(arg, &mhd_regi, sizeof (mhd_regi), mode);
6165 	if (rv != 0)
6166 		return (EFAULT);
6167 
6168 	/* build SCSI VD_OP request */
6169 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_REGISTERANDIGNOREKEY,
6170 	    sizeof (sd_prout_t), &vd_scsi_len);
6171 
6172 	/* set parameters */
6173 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
6174 	bcopy(mhd_regi.newkey.key, scsi_prout->service_key,
6175 	    MHIOC_RESV_KEY_SIZE);
6176 	scsi_prout->aptpl = (uchar_t)mhd_regi.aptpl;
6177 
6178 	/* submit the request */
6179 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6180 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
6181 
6182 	if (rv == 0)
6183 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6184 
6185 	kmem_free(vd_scsi, vd_scsi_len);
6186 	return (rv);
6187 }
6188 
6189 /*
6190  * This function is used by the failfast mechanism to send a SCSI command
6191  * to check for reservation conflict.
6192  */
6193 static int
6194 vdc_failfast_scsi_cmd(vdc_t *vdc, uchar_t scmd)
6195 {
6196 	int cdb_len, sense_len, vd_scsi_len;
6197 	vd_scsi_t *vd_scsi;
6198 	union scsi_cdb *cdb;
6199 	int rv;
6200 
6201 	ASSERT(scmd == SCMD_TEST_UNIT_READY || scmd == SCMD_WRITE_G1);
6202 
6203 	if (scmd == SCMD_WRITE_G1)
6204 		cdb_len = CDB_GROUP1;
6205 	else
6206 		cdb_len = CDB_GROUP0;
6207 
6208 	sense_len = sizeof (struct scsi_extended_sense);
6209 
6210 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, 0, 0, &vd_scsi_len);
6211 
6212 	/* set cdb */
6213 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
6214 	cdb->scc_cmd = scmd;
6215 
6216 	vd_scsi->timeout = vdc_scsi_timeout;
6217 
6218 	/*
6219 	 * Submit the request. The last argument has to be B_FALSE so that
6220 	 * vdc_do_sync_op does not loop checking for reservation conflict if
6221 	 * the operation returns an error.
6222 	 */
6223 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6224 	    0, 0, CB_SYNC, (void *)(uint64_t)FKIOCTL, VIO_both_dir, B_FALSE);
6225 
6226 	if (rv == 0)
6227 		(void) vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6228 
6229 	kmem_free(vd_scsi, vd_scsi_len);
6230 	return (rv);
6231 }
6232 
6233 /*
6234  * This function is used by the failfast mechanism to check for reservation
6235  * conflict. It sends some SCSI commands which will fail with a reservation
6236  * conflict error if the system does not have access to the disk and this
6237  * will panic the system.
6238  *
6239  * Returned Code:
6240  *	0	- disk is accessible without reservation conflict error
6241  *	!= 0	- unable to check if disk is accessible
6242  */
6243 int
6244 vdc_failfast_check_resv(vdc_t *vdc)
6245 {
6246 	int failure = 0;
6247 
6248 	/*
6249 	 * Send a TEST UNIT READY command. The command will panic
6250 	 * the system if it fails with a reservation conflict.
6251 	 */
6252 	if (vdc_failfast_scsi_cmd(vdc, SCMD_TEST_UNIT_READY) != 0)
6253 		failure++;
6254 
6255 	/*
6256 	 * With SPC-3 compliant devices TEST UNIT READY will succeed on
6257 	 * a reserved device, so we also do a WRITE(10) of zero byte in
6258 	 * order to provoke a Reservation Conflict status on those newer
6259 	 * devices.
6260 	 */
6261 	if (vdc_failfast_scsi_cmd(vdc, SCMD_WRITE_G1) != 0)
6262 		failure++;
6263 
6264 	return (failure);
6265 }
6266 
6267 /*
6268  * Add a pending I/O to the failfast I/O queue. An I/O is added to this
6269  * queue when it has failed and failfast is enabled. Then we have to check
6270  * if it has failed because of a reservation conflict in which case we have
6271  * to panic the system.
6272  *
6273  * Async I/O should be queued with their block I/O data transfer structure
6274  * (buf). Sync I/O should be queued with buf = NULL.
6275  */
6276 static vdc_io_t *
6277 vdc_failfast_io_queue(vdc_t *vdc, struct buf *buf)
6278 {
6279 	vdc_io_t *vio;
6280 
6281 	ASSERT(MUTEX_HELD(&vdc->lock));
6282 
6283 	vio = kmem_alloc(sizeof (vdc_io_t), KM_SLEEP);
6284 	vio->vio_next = vdc->failfast_io_queue;
6285 	vio->vio_buf = buf;
6286 	vio->vio_qtime = ddi_get_lbolt();
6287 
6288 	vdc->failfast_io_queue = vio;
6289 
6290 	/* notify the failfast thread that a new I/O is queued */
6291 	cv_signal(&vdc->failfast_cv);
6292 
6293 	return (vio);
6294 }
6295 
6296 /*
6297  * Remove and complete I/O in the failfast I/O queue which have been
6298  * added after the indicated deadline. A deadline of 0 means that all
6299  * I/O have to be unqueued and marked as completed.
6300  */
6301 static void
6302 vdc_failfast_io_unqueue(vdc_t *vdc, clock_t deadline)
6303 {
6304 	vdc_io_t *vio, *vio_tmp;
6305 
6306 	ASSERT(MUTEX_HELD(&vdc->lock));
6307 
6308 	vio_tmp = NULL;
6309 	vio = vdc->failfast_io_queue;
6310 
6311 	if (deadline != 0) {
6312 		/*
6313 		 * Skip any io queued after the deadline. The failfast
6314 		 * I/O queue is ordered starting with the last I/O added
6315 		 * to the queue.
6316 		 */
6317 		while (vio != NULL && vio->vio_qtime > deadline) {
6318 			vio_tmp = vio;
6319 			vio = vio->vio_next;
6320 		}
6321 	}
6322 
6323 	if (vio == NULL)
6324 		/* nothing to unqueue */
6325 		return;
6326 
6327 	/* update the queue */
6328 	if (vio_tmp == NULL)
6329 		vdc->failfast_io_queue = NULL;
6330 	else
6331 		vio_tmp->vio_next = NULL;
6332 
6333 	/*
6334 	 * Complete unqueued I/O. Async I/O have a block I/O data transfer
6335 	 * structure (buf) and they are completed by calling biodone(). Sync
6336 	 * I/O do not have a buf and they are completed by setting the
6337 	 * vio_qtime to zero and signaling failfast_io_cv. In that case, the
6338 	 * thread waiting for the I/O to complete is responsible for freeing
6339 	 * the vio structure.
6340 	 */
6341 	while (vio != NULL) {
6342 		vio_tmp = vio->vio_next;
6343 		if (vio->vio_buf != NULL) {
6344 			VD_KSTAT_RUNQ_EXIT(vdc);
6345 			DTRACE_IO1(done, buf_t *, vio->vio_buf);
6346 			biodone(vio->vio_buf);
6347 			kmem_free(vio, sizeof (vdc_io_t));
6348 		} else {
6349 			vio->vio_qtime = 0;
6350 		}
6351 		vio = vio_tmp;
6352 	}
6353 
6354 	cv_broadcast(&vdc->failfast_io_cv);
6355 }
6356 
6357 /*
6358  * Failfast Thread.
6359  *
6360  * While failfast is enabled, the failfast thread sends a TEST UNIT READY
6361  * and a zero size WRITE(10) SCSI commands on a regular basis to check that
6362  * we still have access to the disk. If a command fails with a RESERVATION
6363  * CONFLICT error then the system will immediatly panic.
6364  *
6365  * The failfast thread is also woken up when an I/O has failed. It then check
6366  * the access to the disk to ensure that the I/O failure was not due to a
6367  * reservation conflict.
6368  *
6369  * There is one failfast thread for each virtual disk for which failfast is
6370  * enabled. We could have only one thread sending requests for all disks but
6371  * this would need vdc to send asynchronous requests and to have callbacks to
6372  * process replies.
6373  */
6374 static void
6375 vdc_failfast_thread(void *arg)
6376 {
6377 	int status;
6378 	vdc_t *vdc = (vdc_t *)arg;
6379 	clock_t timeout, starttime;
6380 
6381 	mutex_enter(&vdc->lock);
6382 
6383 	while (vdc->failfast_interval != 0) {
6384 
6385 		starttime = ddi_get_lbolt();
6386 
6387 		mutex_exit(&vdc->lock);
6388 
6389 		/* check for reservation conflict */
6390 		status = vdc_failfast_check_resv(vdc);
6391 
6392 		mutex_enter(&vdc->lock);
6393 		/*
6394 		 * We have dropped the lock to send the SCSI command so we have
6395 		 * to check that failfast is still enabled.
6396 		 */
6397 		if (vdc->failfast_interval == 0)
6398 			break;
6399 
6400 		/*
6401 		 * If we have successfully check the disk access and there was
6402 		 * no reservation conflict then we can complete any I/O queued
6403 		 * before the last check.
6404 		 */
6405 		if (status == 0)
6406 			vdc_failfast_io_unqueue(vdc, starttime);
6407 
6408 		/* proceed again if some I/O are still in the queue */
6409 		if (vdc->failfast_io_queue != NULL)
6410 			continue;
6411 
6412 		timeout = ddi_get_lbolt() +
6413 		    drv_usectohz(vdc->failfast_interval);
6414 		(void) cv_timedwait(&vdc->failfast_cv, &vdc->lock, timeout);
6415 	}
6416 
6417 	/*
6418 	 * Failfast is being stop so we can complete any queued I/O.
6419 	 */
6420 	vdc_failfast_io_unqueue(vdc, 0);
6421 	vdc->failfast_thread = NULL;
6422 	mutex_exit(&vdc->lock);
6423 	thread_exit();
6424 }
6425 
6426 /*
6427  * Implement the MHIOCENFAILFAST mhd(7i) ioctl.
6428  */
6429 static int
6430 vdc_failfast(vdc_t *vdc, caddr_t arg, int mode)
6431 {
6432 	unsigned int mh_time;
6433 
6434 	if (ddi_copyin((void *)arg, &mh_time, sizeof (int), mode))
6435 		return (EFAULT);
6436 
6437 	mutex_enter(&vdc->lock);
6438 	if (mh_time != 0 && vdc->failfast_thread == NULL) {
6439 		vdc->failfast_thread = thread_create(NULL, 0,
6440 		    vdc_failfast_thread, vdc, 0, &p0, TS_RUN,
6441 		    v.v_maxsyspri - 2);
6442 	}
6443 
6444 	vdc->failfast_interval = mh_time * 1000;
6445 	cv_signal(&vdc->failfast_cv);
6446 	mutex_exit(&vdc->lock);
6447 
6448 	return (0);
6449 }
6450 
6451 /*
6452  * Implement the MHIOCTKOWN and MHIOCRELEASE mhd(7i) ioctls. These ioctls are
6453  * converted to VD_OP_SET_ACCESS operations.
6454  */
6455 static int
6456 vdc_access_set(vdc_t *vdc, uint64_t flags, int mode)
6457 {
6458 	int rv;
6459 
6460 	/* submit owership command request */
6461 	rv = vdc_do_sync_op(vdc, VD_OP_SET_ACCESS, (caddr_t)&flags,
6462 	    sizeof (uint64_t), 0, 0, CB_SYNC, (void *)(uint64_t)mode,
6463 	    VIO_both_dir, B_TRUE);
6464 
6465 	return (rv);
6466 }
6467 
6468 /*
6469  * Implement the MHIOCSTATUS mhd(7i) ioctl. This ioctl is converted to a
6470  * VD_OP_GET_ACCESS operation.
6471  */
6472 static int
6473 vdc_access_get(vdc_t *vdc, uint64_t *status, int mode)
6474 {
6475 	int rv;
6476 
6477 	/* submit owership command request */
6478 	rv = vdc_do_sync_op(vdc, VD_OP_GET_ACCESS, (caddr_t)status,
6479 	    sizeof (uint64_t), 0, 0, CB_SYNC, (void *)(uint64_t)mode,
6480 	    VIO_both_dir, B_TRUE);
6481 
6482 	return (rv);
6483 }
6484 
6485 /*
6486  * Disk Ownership Thread.
6487  *
6488  * When we have taken the ownership of a disk, this thread waits to be
6489  * notified when the LDC channel is reset so that it can recover the
6490  * ownership.
6491  *
6492  * Note that the thread handling the LDC reset (vdc_process_msg_thread())
6493  * can not be used to do the ownership recovery because it has to be
6494  * running to handle the reply message to the ownership operation.
6495  */
6496 static void
6497 vdc_ownership_thread(void *arg)
6498 {
6499 	vdc_t *vdc = (vdc_t *)arg;
6500 	clock_t timeout;
6501 	uint64_t status;
6502 
6503 	mutex_enter(&vdc->ownership_lock);
6504 	mutex_enter(&vdc->lock);
6505 
6506 	while (vdc->ownership & VDC_OWNERSHIP_WANTED) {
6507 
6508 		if ((vdc->ownership & VDC_OWNERSHIP_RESET) ||
6509 		    !(vdc->ownership & VDC_OWNERSHIP_GRANTED)) {
6510 			/*
6511 			 * There was a reset so the ownership has been lost,
6512 			 * try to recover. We do this without using the preempt
6513 			 * option so that we don't steal the ownership from
6514 			 * someone who has preempted us.
6515 			 */
6516 			DMSG(vdc, 0, "[%d] Ownership lost, recovering",
6517 			    vdc->instance);
6518 
6519 			vdc->ownership &= ~(VDC_OWNERSHIP_RESET |
6520 			    VDC_OWNERSHIP_GRANTED);
6521 
6522 			mutex_exit(&vdc->lock);
6523 
6524 			status = vdc_access_set(vdc, VD_ACCESS_SET_EXCLUSIVE |
6525 			    VD_ACCESS_SET_PRESERVE, FKIOCTL);
6526 
6527 			mutex_enter(&vdc->lock);
6528 
6529 			if (status == 0) {
6530 				DMSG(vdc, 0, "[%d] Ownership recovered",
6531 				    vdc->instance);
6532 				vdc->ownership |= VDC_OWNERSHIP_GRANTED;
6533 			} else {
6534 				DMSG(vdc, 0, "[%d] Fail to recover ownership",
6535 				    vdc->instance);
6536 			}
6537 
6538 		}
6539 
6540 		/*
6541 		 * If we have the ownership then we just wait for an event
6542 		 * to happen (LDC reset), otherwise we will retry to recover
6543 		 * after a delay.
6544 		 */
6545 		if (vdc->ownership & VDC_OWNERSHIP_GRANTED)
6546 			timeout = 0;
6547 		else
6548 			timeout = ddi_get_lbolt() +
6549 			    drv_usectohz(vdc_ownership_delay);
6550 
6551 		/* Release the ownership_lock and wait on the vdc lock */
6552 		mutex_exit(&vdc->ownership_lock);
6553 
6554 		if (timeout == 0)
6555 			(void) cv_wait(&vdc->ownership_cv, &vdc->lock);
6556 		else
6557 			(void) cv_timedwait(&vdc->ownership_cv,
6558 			    &vdc->lock, timeout);
6559 
6560 		mutex_exit(&vdc->lock);
6561 
6562 		mutex_enter(&vdc->ownership_lock);
6563 		mutex_enter(&vdc->lock);
6564 	}
6565 
6566 	vdc->ownership_thread = NULL;
6567 	mutex_exit(&vdc->lock);
6568 	mutex_exit(&vdc->ownership_lock);
6569 
6570 	thread_exit();
6571 }
6572 
6573 static void
6574 vdc_ownership_update(vdc_t *vdc, int ownership_flags)
6575 {
6576 	ASSERT(MUTEX_HELD(&vdc->ownership_lock));
6577 
6578 	mutex_enter(&vdc->lock);
6579 	vdc->ownership = ownership_flags;
6580 	if ((vdc->ownership & VDC_OWNERSHIP_WANTED) &&
6581 	    vdc->ownership_thread == NULL) {
6582 		/* start ownership thread */
6583 		vdc->ownership_thread = thread_create(NULL, 0,
6584 		    vdc_ownership_thread, vdc, 0, &p0, TS_RUN,
6585 		    v.v_maxsyspri - 2);
6586 	} else {
6587 		/* notify the ownership thread */
6588 		cv_signal(&vdc->ownership_cv);
6589 	}
6590 	mutex_exit(&vdc->lock);
6591 }
6592 
6593 /*
6594  * Get the size and the block size of a virtual disk from the vdisk server.
6595  */
6596 static int
6597 vdc_get_capacity(vdc_t *vdc, size_t *dsk_size, size_t *blk_size)
6598 {
6599 	int rv = 0;
6600 	size_t alloc_len;
6601 	vd_capacity_t *vd_cap;
6602 
6603 	ASSERT(MUTEX_NOT_HELD(&vdc->lock));
6604 
6605 	alloc_len = P2ROUNDUP(sizeof (vd_capacity_t), sizeof (uint64_t));
6606 
6607 	vd_cap = kmem_zalloc(alloc_len, KM_SLEEP);
6608 
6609 	rv = vdc_do_sync_op(vdc, VD_OP_GET_CAPACITY, (caddr_t)vd_cap, alloc_len,
6610 	    0, 0, CB_SYNC, (void *)(uint64_t)FKIOCTL, VIO_both_dir, B_TRUE);
6611 
6612 	*dsk_size = vd_cap->vdisk_size;
6613 	*blk_size = vd_cap->vdisk_block_size;
6614 
6615 	kmem_free(vd_cap, alloc_len);
6616 	return (rv);
6617 }
6618 
6619 /*
6620  * Check the disk capacity. Disk size information is updated if size has
6621  * changed.
6622  *
6623  * Return 0 if the disk capacity is available, or non-zero if it is not.
6624  */
6625 static int
6626 vdc_check_capacity(vdc_t *vdc)
6627 {
6628 	size_t dsk_size, blk_size;
6629 	int rv;
6630 
6631 	/*
6632 	 * If the vdisk does not support the VD_OP_GET_CAPACITY operation
6633 	 * then the disk capacity has been retrieved during the handshake
6634 	 * and there's nothing more to do here.
6635 	 */
6636 	if (!VD_OP_SUPPORTED(vdc->operations, VD_OP_GET_CAPACITY))
6637 		return (0);
6638 
6639 	if ((rv = vdc_get_capacity(vdc, &dsk_size, &blk_size)) != 0)
6640 		return (rv);
6641 
6642 	if (dsk_size == VD_SIZE_UNKNOWN || dsk_size == 0)
6643 		return (EINVAL);
6644 
6645 	mutex_enter(&vdc->lock);
6646 	vdc_update_size(vdc, dsk_size, blk_size, vdc->max_xfer_sz);
6647 	mutex_exit(&vdc->lock);
6648 
6649 	return (0);
6650 }
6651 
6652 /*
6653  * This structure is used in the DKIO(7I) array below.
6654  */
6655 typedef struct vdc_dk_ioctl {
6656 	uint8_t		op;		/* VD_OP_XXX value */
6657 	int		cmd;		/* Solaris ioctl operation number */
6658 	size_t		nbytes;		/* size of structure to be copied */
6659 
6660 	/* function to convert between vDisk and Solaris structure formats */
6661 	int	(*convert)(vdc_t *vdc, void *vd_buf, void *ioctl_arg,
6662 	    int mode, int dir);
6663 } vdc_dk_ioctl_t;
6664 
6665 /*
6666  * Subset of DKIO(7I) operations currently supported
6667  */
6668 static vdc_dk_ioctl_t	dk_ioctl[] = {
6669 	{VD_OP_FLUSH,		DKIOCFLUSHWRITECACHE,	0,
6670 		vdc_null_copy_func},
6671 	{VD_OP_GET_WCE,		DKIOCGETWCE,		sizeof (int),
6672 		vdc_get_wce_convert},
6673 	{VD_OP_SET_WCE,		DKIOCSETWCE,		sizeof (int),
6674 		vdc_set_wce_convert},
6675 	{VD_OP_GET_VTOC,	DKIOCGVTOC,		sizeof (vd_vtoc_t),
6676 		vdc_get_vtoc_convert},
6677 	{VD_OP_SET_VTOC,	DKIOCSVTOC,		sizeof (vd_vtoc_t),
6678 		vdc_set_vtoc_convert},
6679 	{VD_OP_GET_VTOC,	DKIOCGEXTVTOC,		sizeof (vd_vtoc_t),
6680 		vdc_get_extvtoc_convert},
6681 	{VD_OP_SET_VTOC,	DKIOCSEXTVTOC,		sizeof (vd_vtoc_t),
6682 		vdc_set_extvtoc_convert},
6683 	{VD_OP_GET_DISKGEOM,	DKIOCGGEOM,		sizeof (vd_geom_t),
6684 		vdc_get_geom_convert},
6685 	{VD_OP_GET_DISKGEOM,	DKIOCG_PHYGEOM,		sizeof (vd_geom_t),
6686 		vdc_get_geom_convert},
6687 	{VD_OP_GET_DISKGEOM, 	DKIOCG_VIRTGEOM,	sizeof (vd_geom_t),
6688 		vdc_get_geom_convert},
6689 	{VD_OP_SET_DISKGEOM,	DKIOCSGEOM,		sizeof (vd_geom_t),
6690 		vdc_set_geom_convert},
6691 	{VD_OP_GET_EFI,		DKIOCGETEFI,		0,
6692 		vdc_get_efi_convert},
6693 	{VD_OP_SET_EFI,		DKIOCSETEFI,		0,
6694 		vdc_set_efi_convert},
6695 
6696 	/* DIOCTL_RWCMD is converted to a read or a write */
6697 	{0, DIOCTL_RWCMD,  sizeof (struct dadkio_rwcmd), NULL},
6698 
6699 	/* mhd(7I) non-shared multihost disks ioctls */
6700 	{0, MHIOCTKOWN,				0, vdc_null_copy_func},
6701 	{0, MHIOCRELEASE,			0, vdc_null_copy_func},
6702 	{0, MHIOCSTATUS,			0, vdc_null_copy_func},
6703 	{0, MHIOCQRESERVE,			0, vdc_null_copy_func},
6704 
6705 	/* mhd(7I) shared multihost disks ioctls */
6706 	{0, MHIOCGRP_INKEYS,			0, vdc_null_copy_func},
6707 	{0, MHIOCGRP_INRESV,			0, vdc_null_copy_func},
6708 	{0, MHIOCGRP_REGISTER,			0, vdc_null_copy_func},
6709 	{0, MHIOCGRP_RESERVE, 			0, vdc_null_copy_func},
6710 	{0, MHIOCGRP_PREEMPTANDABORT,		0, vdc_null_copy_func},
6711 	{0, MHIOCGRP_REGISTERANDIGNOREKEY,	0, vdc_null_copy_func},
6712 
6713 	/* mhd(7I) failfast ioctl */
6714 	{0, MHIOCENFAILFAST,			0, vdc_null_copy_func},
6715 
6716 	/*
6717 	 * These particular ioctls are not sent to the server - vdc fakes up
6718 	 * the necessary info.
6719 	 */
6720 	{0, DKIOCINFO, sizeof (struct dk_cinfo), vdc_null_copy_func},
6721 	{0, DKIOCGMEDIAINFO, sizeof (struct dk_minfo), vdc_null_copy_func},
6722 	{0, USCSICMD,	sizeof (struct uscsi_cmd), vdc_null_copy_func},
6723 	{0, DKIOCPARTITION, 0, vdc_null_copy_func },
6724 	{0, DKIOCGAPART, 0, vdc_null_copy_func },
6725 	{0, DKIOCREMOVABLE, 0, vdc_null_copy_func},
6726 	{0, CDROMREADOFFSET, 0, vdc_null_copy_func}
6727 };
6728 
6729 /*
6730  * This function handles ioctl requests from the vd_efi_alloc_and_read()
6731  * function and forward them to the vdisk.
6732  */
6733 static int
6734 vd_process_efi_ioctl(void *vdisk, int cmd, uintptr_t arg)
6735 {
6736 	vdc_t *vdc = (vdc_t *)vdisk;
6737 	dev_t dev;
6738 	int rval;
6739 
6740 	dev = makedevice(ddi_driver_major(vdc->dip),
6741 	    VD_MAKE_DEV(vdc->instance, 0));
6742 
6743 	return (vd_process_ioctl(dev, cmd, (caddr_t)arg, FKIOCTL, &rval));
6744 }
6745 
6746 /*
6747  * Function:
6748  *	vd_process_ioctl()
6749  *
6750  * Description:
6751  *	This routine processes disk specific ioctl calls
6752  *
6753  * Arguments:
6754  *	dev	- the device number
6755  *	cmd	- the operation [dkio(7I)] to be processed
6756  *	arg	- pointer to user provided structure
6757  *		  (contains data to be set or reference parameter for get)
6758  *	mode	- bit flag, indicating open settings, 32/64 bit type, etc
6759  *	rvalp	- pointer to return value for calling process.
6760  *
6761  * Return Code:
6762  *	0
6763  *	EFAULT
6764  *	ENXIO
6765  *	EIO
6766  *	ENOTSUP
6767  */
6768 static int
6769 vd_process_ioctl(dev_t dev, int cmd, caddr_t arg, int mode, int *rvalp)
6770 {
6771 	int		instance = VDCUNIT(dev);
6772 	vdc_t		*vdc = NULL;
6773 	int		rv = -1;
6774 	int		idx = 0;		/* index into dk_ioctl[] */
6775 	size_t		len = 0;		/* #bytes to send to vds */
6776 	size_t		alloc_len = 0;		/* #bytes to allocate mem for */
6777 	caddr_t		mem_p = NULL;
6778 	size_t		nioctls = (sizeof (dk_ioctl)) / (sizeof (dk_ioctl[0]));
6779 	vdc_dk_ioctl_t	*iop;
6780 
6781 	vdc = ddi_get_soft_state(vdc_state, instance);
6782 	if (vdc == NULL) {
6783 		cmn_err(CE_NOTE, "![%d] Could not get soft state structure",
6784 		    instance);
6785 		return (ENXIO);
6786 	}
6787 
6788 	DMSG(vdc, 0, "[%d] Processing ioctl(%x) for dev %lx : model %x\n",
6789 	    instance, cmd, dev, ddi_model_convert_from(mode & FMODELS));
6790 
6791 	if (rvalp != NULL) {
6792 		/* the return value of the ioctl is 0 by default */
6793 		*rvalp = 0;
6794 	}
6795 
6796 	/*
6797 	 * Validate the ioctl operation to be performed.
6798 	 *
6799 	 * If we have looped through the array without finding a match then we
6800 	 * don't support this ioctl.
6801 	 */
6802 	for (idx = 0; idx < nioctls; idx++) {
6803 		if (cmd == dk_ioctl[idx].cmd)
6804 			break;
6805 	}
6806 
6807 	if (idx >= nioctls) {
6808 		DMSG(vdc, 0, "[%d] Unsupported ioctl (0x%x)\n",
6809 		    vdc->instance, cmd);
6810 		return (ENOTSUP);
6811 	}
6812 
6813 	iop = &(dk_ioctl[idx]);
6814 
6815 	if (cmd == DKIOCGETEFI || cmd == DKIOCSETEFI) {
6816 		/* size is not fixed for EFI ioctls, it depends on ioctl arg */
6817 		dk_efi_t	dk_efi;
6818 
6819 		rv = ddi_copyin(arg, &dk_efi, sizeof (dk_efi_t), mode);
6820 		if (rv != 0)
6821 			return (EFAULT);
6822 
6823 		len = sizeof (vd_efi_t) - 1 + dk_efi.dki_length;
6824 	} else {
6825 		len = iop->nbytes;
6826 	}
6827 
6828 	/* check if the ioctl is applicable */
6829 	switch (cmd) {
6830 	case CDROMREADOFFSET:
6831 	case DKIOCREMOVABLE:
6832 		return (ENOTTY);
6833 
6834 	case USCSICMD:
6835 	case MHIOCTKOWN:
6836 	case MHIOCSTATUS:
6837 	case MHIOCQRESERVE:
6838 	case MHIOCRELEASE:
6839 	case MHIOCGRP_INKEYS:
6840 	case MHIOCGRP_INRESV:
6841 	case MHIOCGRP_REGISTER:
6842 	case MHIOCGRP_RESERVE:
6843 	case MHIOCGRP_PREEMPTANDABORT:
6844 	case MHIOCGRP_REGISTERANDIGNOREKEY:
6845 	case MHIOCENFAILFAST:
6846 		if (vdc->cinfo == NULL)
6847 			return (ENXIO);
6848 		if (vdc->cinfo->dki_ctype != DKC_SCSI_CCS)
6849 			return (ENOTTY);
6850 		break;
6851 
6852 	case DIOCTL_RWCMD:
6853 		if (vdc->cinfo == NULL)
6854 			return (ENXIO);
6855 		if (vdc->cinfo->dki_ctype != DKC_DIRECT)
6856 			return (ENOTTY);
6857 		break;
6858 
6859 	case DKIOCINFO:
6860 		if (vdc->cinfo == NULL)
6861 			return (ENXIO);
6862 		break;
6863 
6864 	case DKIOCGMEDIAINFO:
6865 		if (vdc->minfo == NULL)
6866 			return (ENXIO);
6867 		if (vdc_check_capacity(vdc) != 0)
6868 			/* disk capacity is not available */
6869 			return (EIO);
6870 		break;
6871 	}
6872 
6873 	/*
6874 	 * Deal with ioctls which require a processing different than
6875 	 * converting ioctl arguments and sending a corresponding
6876 	 * VD operation.
6877 	 */
6878 	switch (cmd) {
6879 
6880 	case USCSICMD:
6881 	{
6882 		return (vdc_uscsi_cmd(vdc, arg, mode));
6883 	}
6884 
6885 	case MHIOCTKOWN:
6886 	{
6887 		mutex_enter(&vdc->ownership_lock);
6888 		/*
6889 		 * We have to set VDC_OWNERSHIP_WANTED now so that the ownership
6890 		 * can be flagged with VDC_OWNERSHIP_RESET if the LDC is reset
6891 		 * while we are processing the ioctl.
6892 		 */
6893 		vdc_ownership_update(vdc, VDC_OWNERSHIP_WANTED);
6894 
6895 		rv = vdc_access_set(vdc, VD_ACCESS_SET_EXCLUSIVE |
6896 		    VD_ACCESS_SET_PREEMPT | VD_ACCESS_SET_PRESERVE, mode);
6897 		if (rv == 0) {
6898 			vdc_ownership_update(vdc, VDC_OWNERSHIP_WANTED |
6899 			    VDC_OWNERSHIP_GRANTED);
6900 		} else {
6901 			vdc_ownership_update(vdc, VDC_OWNERSHIP_NONE);
6902 		}
6903 		mutex_exit(&vdc->ownership_lock);
6904 		return (rv);
6905 	}
6906 
6907 	case MHIOCRELEASE:
6908 	{
6909 		mutex_enter(&vdc->ownership_lock);
6910 		rv = vdc_access_set(vdc, VD_ACCESS_SET_CLEAR, mode);
6911 		if (rv == 0) {
6912 			vdc_ownership_update(vdc, VDC_OWNERSHIP_NONE);
6913 		}
6914 		mutex_exit(&vdc->ownership_lock);
6915 		return (rv);
6916 	}
6917 
6918 	case MHIOCSTATUS:
6919 	{
6920 		uint64_t status;
6921 
6922 		rv = vdc_access_get(vdc, &status, mode);
6923 		if (rv == 0 && rvalp != NULL)
6924 			*rvalp = (status & VD_ACCESS_ALLOWED)? 0 : 1;
6925 		return (rv);
6926 	}
6927 
6928 	case MHIOCQRESERVE:
6929 	{
6930 		rv = vdc_access_set(vdc, VD_ACCESS_SET_EXCLUSIVE, mode);
6931 		return (rv);
6932 	}
6933 
6934 	case MHIOCGRP_INKEYS:
6935 	{
6936 		return (vdc_mhd_inkeys(vdc, arg, mode));
6937 	}
6938 
6939 	case MHIOCGRP_INRESV:
6940 	{
6941 		return (vdc_mhd_inresv(vdc, arg, mode));
6942 	}
6943 
6944 	case MHIOCGRP_REGISTER:
6945 	{
6946 		return (vdc_mhd_register(vdc, arg, mode));
6947 	}
6948 
6949 	case MHIOCGRP_RESERVE:
6950 	{
6951 		return (vdc_mhd_reserve(vdc, arg, mode));
6952 	}
6953 
6954 	case MHIOCGRP_PREEMPTANDABORT:
6955 	{
6956 		return (vdc_mhd_preemptabort(vdc, arg, mode));
6957 	}
6958 
6959 	case MHIOCGRP_REGISTERANDIGNOREKEY:
6960 	{
6961 		return (vdc_mhd_registerignore(vdc, arg, mode));
6962 	}
6963 
6964 	case MHIOCENFAILFAST:
6965 	{
6966 		rv = vdc_failfast(vdc, arg, mode);
6967 		return (rv);
6968 	}
6969 
6970 	case DIOCTL_RWCMD:
6971 	{
6972 		return (vdc_dioctl_rwcmd(dev, arg, mode));
6973 	}
6974 
6975 	case DKIOCGAPART:
6976 	{
6977 		return (vdc_dkio_gapart(vdc, arg, mode));
6978 	}
6979 
6980 	case DKIOCPARTITION:
6981 	{
6982 		return (vdc_dkio_partition(vdc, arg, mode));
6983 	}
6984 
6985 	case DKIOCINFO:
6986 	{
6987 		struct dk_cinfo	cinfo;
6988 
6989 		bcopy(vdc->cinfo, &cinfo, sizeof (struct dk_cinfo));
6990 		cinfo.dki_partition = VDCPART(dev);
6991 
6992 		rv = ddi_copyout(&cinfo, (void *)arg,
6993 		    sizeof (struct dk_cinfo), mode);
6994 		if (rv != 0)
6995 			return (EFAULT);
6996 
6997 		return (0);
6998 	}
6999 
7000 	case DKIOCGMEDIAINFO:
7001 	{
7002 		ASSERT(vdc->vdisk_size != 0);
7003 		ASSERT(vdc->minfo->dki_capacity != 0);
7004 		rv = ddi_copyout(vdc->minfo, (void *)arg,
7005 		    sizeof (struct dk_minfo), mode);
7006 		if (rv != 0)
7007 			return (EFAULT);
7008 
7009 		return (0);
7010 	}
7011 
7012 	case DKIOCFLUSHWRITECACHE:
7013 		{
7014 			struct dk_callback *dkc =
7015 			    (struct dk_callback *)(uintptr_t)arg;
7016 			vdc_dk_arg_t	*dkarg = NULL;
7017 
7018 			DMSG(vdc, 1, "[%d] Flush W$: mode %x\n",
7019 			    instance, mode);
7020 
7021 			/*
7022 			 * If arg is NULL, then there is no callback function
7023 			 * registered and the call operates synchronously; we
7024 			 * break and continue with the rest of the function and
7025 			 * wait for vds to return (i.e. after the request to
7026 			 * vds returns successfully, all writes completed prior
7027 			 * to the ioctl will have been flushed from the disk
7028 			 * write cache to persistent media.
7029 			 *
7030 			 * If a callback function is registered, we dispatch
7031 			 * the request on a task queue and return immediately.
7032 			 * The callback will deal with informing the calling
7033 			 * thread that the flush request is completed.
7034 			 */
7035 			if (dkc == NULL)
7036 				break;
7037 
7038 			/*
7039 			 * the asynchronous callback is only supported if
7040 			 * invoked from within the kernel
7041 			 */
7042 			if ((mode & FKIOCTL) == 0)
7043 				return (ENOTSUP);
7044 
7045 			dkarg = kmem_zalloc(sizeof (vdc_dk_arg_t), KM_SLEEP);
7046 
7047 			dkarg->mode = mode;
7048 			dkarg->dev = dev;
7049 			bcopy(dkc, &dkarg->dkc, sizeof (*dkc));
7050 
7051 			mutex_enter(&vdc->lock);
7052 			vdc->dkio_flush_pending++;
7053 			dkarg->vdc = vdc;
7054 			mutex_exit(&vdc->lock);
7055 
7056 			/* put the request on a task queue */
7057 			rv = taskq_dispatch(system_taskq, vdc_dkio_flush_cb,
7058 			    (void *)dkarg, DDI_SLEEP);
7059 			if (rv == NULL) {
7060 				/* clean up if dispatch fails */
7061 				mutex_enter(&vdc->lock);
7062 				vdc->dkio_flush_pending--;
7063 				mutex_exit(&vdc->lock);
7064 				kmem_free(dkarg, sizeof (vdc_dk_arg_t));
7065 			}
7066 
7067 			return (rv == NULL ? ENOMEM : 0);
7068 		}
7069 	}
7070 
7071 	/* catch programming error in vdc - should be a VD_OP_XXX ioctl */
7072 	ASSERT(iop->op != 0);
7073 
7074 	/* check if the vDisk server handles the operation for this vDisk */
7075 	if (VD_OP_SUPPORTED(vdc->operations, iop->op) == B_FALSE) {
7076 		DMSG(vdc, 0, "[%d] Unsupported VD_OP operation (0x%x)\n",
7077 		    vdc->instance, iop->op);
7078 		return (ENOTSUP);
7079 	}
7080 
7081 	/* LDC requires that the memory being mapped is 8-byte aligned */
7082 	alloc_len = P2ROUNDUP(len, sizeof (uint64_t));
7083 	DMSG(vdc, 1, "[%d] struct size %ld alloc %ld\n",
7084 	    instance, len, alloc_len);
7085 
7086 	if (alloc_len > 0)
7087 		mem_p = kmem_zalloc(alloc_len, KM_SLEEP);
7088 
7089 	/*
7090 	 * Call the conversion function for this ioctl which, if necessary,
7091 	 * converts from the Solaris format to the format ARC'ed
7092 	 * as part of the vDisk protocol (FWARC 2006/195)
7093 	 */
7094 	ASSERT(iop->convert != NULL);
7095 	rv = (iop->convert)(vdc, arg, mem_p, mode, VD_COPYIN);
7096 	if (rv != 0) {
7097 		DMSG(vdc, 0, "[%d] convert func returned %d for ioctl 0x%x\n",
7098 		    instance, rv, cmd);
7099 		if (mem_p != NULL)
7100 			kmem_free(mem_p, alloc_len);
7101 		return (rv);
7102 	}
7103 
7104 	/*
7105 	 * send request to vds to service the ioctl.
7106 	 */
7107 	rv = vdc_do_sync_op(vdc, iop->op, mem_p, alloc_len,
7108 	    VDCPART(dev), 0, CB_SYNC, (void *)(uint64_t)mode,
7109 	    VIO_both_dir, B_TRUE);
7110 
7111 	if (rv != 0) {
7112 		/*
7113 		 * This is not necessarily an error. The ioctl could
7114 		 * be returning a value such as ENOTTY to indicate
7115 		 * that the ioctl is not applicable.
7116 		 */
7117 		DMSG(vdc, 0, "[%d] vds returned %d for ioctl 0x%x\n",
7118 		    instance, rv, cmd);
7119 		if (mem_p != NULL)
7120 			kmem_free(mem_p, alloc_len);
7121 
7122 		return (rv);
7123 	}
7124 
7125 	/*
7126 	 * Call the conversion function (if it exists) for this ioctl
7127 	 * which converts from the format ARC'ed as part of the vDisk
7128 	 * protocol (FWARC 2006/195) back to a format understood by
7129 	 * the rest of Solaris.
7130 	 */
7131 	rv = (iop->convert)(vdc, mem_p, arg, mode, VD_COPYOUT);
7132 	if (rv != 0) {
7133 		DMSG(vdc, 0, "[%d] convert func returned %d for ioctl 0x%x\n",
7134 		    instance, rv, cmd);
7135 		if (mem_p != NULL)
7136 			kmem_free(mem_p, alloc_len);
7137 		return (rv);
7138 	}
7139 
7140 	if (mem_p != NULL)
7141 		kmem_free(mem_p, alloc_len);
7142 
7143 	return (rv);
7144 }
7145 
7146 /*
7147  * Function:
7148  *
7149  * Description:
7150  *	This is an empty conversion function used by ioctl calls which
7151  *	do not need to convert the data being passed in/out to userland
7152  */
7153 static int
7154 vdc_null_copy_func(vdc_t *vdc, void *from, void *to, int mode, int dir)
7155 {
7156 	_NOTE(ARGUNUSED(vdc))
7157 	_NOTE(ARGUNUSED(from))
7158 	_NOTE(ARGUNUSED(to))
7159 	_NOTE(ARGUNUSED(mode))
7160 	_NOTE(ARGUNUSED(dir))
7161 
7162 	return (0);
7163 }
7164 
7165 static int
7166 vdc_get_wce_convert(vdc_t *vdc, void *from, void *to,
7167     int mode, int dir)
7168 {
7169 	_NOTE(ARGUNUSED(vdc))
7170 
7171 	if (dir == VD_COPYIN)
7172 		return (0);		/* nothing to do */
7173 
7174 	if (ddi_copyout(from, to, sizeof (int), mode) != 0)
7175 		return (EFAULT);
7176 
7177 	return (0);
7178 }
7179 
7180 static int
7181 vdc_set_wce_convert(vdc_t *vdc, void *from, void *to,
7182     int mode, int dir)
7183 {
7184 	_NOTE(ARGUNUSED(vdc))
7185 
7186 	if (dir == VD_COPYOUT)
7187 		return (0);		/* nothing to do */
7188 
7189 	if (ddi_copyin(from, to, sizeof (int), mode) != 0)
7190 		return (EFAULT);
7191 
7192 	return (0);
7193 }
7194 
7195 /*
7196  * Function:
7197  *	vdc_get_vtoc_convert()
7198  *
7199  * Description:
7200  *	This routine performs the necessary convertions from the DKIOCGVTOC
7201  *	Solaris structure to the format defined in FWARC 2006/195.
7202  *
7203  *	In the struct vtoc definition, the timestamp field is marked as not
7204  *	supported so it is not part of vDisk protocol (FWARC 2006/195).
7205  *	However SVM uses that field to check it can write into the VTOC,
7206  *	so we fake up the info of that field.
7207  *
7208  * Arguments:
7209  *	vdc	- the vDisk client
7210  *	from	- the buffer containing the data to be copied from
7211  *	to	- the buffer to be copied to
7212  *	mode	- flags passed to ioctl() call
7213  *	dir	- the "direction" of the copy - VD_COPYIN or VD_COPYOUT
7214  *
7215  * Return Code:
7216  *	0	- Success
7217  *	ENXIO	- incorrect buffer passed in.
7218  *	EFAULT	- ddi_copyout routine encountered an error.
7219  */
7220 static int
7221 vdc_get_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7222 {
7223 	int		i;
7224 	struct vtoc	vtoc;
7225 	struct vtoc32	vtoc32;
7226 	struct extvtoc	evtoc;
7227 	int		rv;
7228 
7229 	if (dir != VD_COPYOUT)
7230 		return (0);	/* nothing to do */
7231 
7232 	if ((from == NULL) || (to == NULL))
7233 		return (ENXIO);
7234 
7235 	if (vdc->vdisk_size > VD_OLDVTOC_LIMIT)
7236 		return (EOVERFLOW);
7237 
7238 	VD_VTOC2VTOC((vd_vtoc_t *)from, &evtoc);
7239 
7240 	/* fake the VTOC timestamp field */
7241 	for (i = 0; i < V_NUMPAR; i++) {
7242 		evtoc.timestamp[i] = vdc->vtoc->timestamp[i];
7243 	}
7244 
7245 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
7246 		/* LINTED E_ASSIGN_NARROW_CONV */
7247 		extvtoctovtoc32(evtoc, vtoc32);
7248 		rv = ddi_copyout(&vtoc32, to, sizeof (vtoc32), mode);
7249 		if (rv != 0)
7250 			rv = EFAULT;
7251 	} else {
7252 		extvtoctovtoc(evtoc, vtoc);
7253 		rv = ddi_copyout(&vtoc, to, sizeof (vtoc), mode);
7254 		if (rv != 0)
7255 			rv = EFAULT;
7256 	}
7257 
7258 	return (rv);
7259 }
7260 
7261 /*
7262  * Function:
7263  *	vdc_set_vtoc_convert()
7264  *
7265  * Description:
7266  *	This routine performs the necessary convertions from the DKIOCSVTOC
7267  *	Solaris structure to the format defined in FWARC 2006/195.
7268  *
7269  * Arguments:
7270  *	vdc	- the vDisk client
7271  *	from	- Buffer with data
7272  *	to	- Buffer where data is to be copied to
7273  *	mode	- flags passed to ioctl
7274  *	dir	- direction of copy (in or out)
7275  *
7276  * Return Code:
7277  *	0	- Success
7278  *	ENXIO	- Invalid buffer passed in
7279  *	EFAULT	- ddi_copyin of data failed
7280  */
7281 static int
7282 vdc_set_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7283 {
7284 	void		*uvtoc;
7285 	struct vtoc	vtoc;
7286 	struct vtoc32	vtoc32;
7287 	struct extvtoc	evtoc;
7288 	int		i, rv;
7289 
7290 	if ((from == NULL) || (to == NULL))
7291 		return (ENXIO);
7292 
7293 	if (vdc->vdisk_size > VD_OLDVTOC_LIMIT)
7294 		return (EOVERFLOW);
7295 
7296 	uvtoc = (dir == VD_COPYIN)? from : to;
7297 
7298 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
7299 		rv = ddi_copyin(uvtoc, &vtoc32, sizeof (vtoc32), mode);
7300 		if (rv != 0)
7301 			return (EFAULT);
7302 		vtoc32toextvtoc(vtoc32, evtoc);
7303 	} else {
7304 		rv = ddi_copyin(uvtoc, &vtoc, sizeof (vtoc), mode);
7305 		if (rv != 0)
7306 			return (EFAULT);
7307 		vtoctoextvtoc(vtoc, evtoc);
7308 	}
7309 
7310 	if (dir == VD_COPYOUT) {
7311 		/*
7312 		 * The disk label may have changed. Revalidate the disk
7313 		 * geometry. This will also update the device nodes.
7314 		 */
7315 		vdc_validate(vdc);
7316 
7317 		/*
7318 		 * We also need to keep track of the timestamp fields.
7319 		 */
7320 		for (i = 0; i < V_NUMPAR; i++) {
7321 			vdc->vtoc->timestamp[i] = evtoc.timestamp[i];
7322 		}
7323 
7324 	} else {
7325 		VTOC2VD_VTOC(&evtoc, (vd_vtoc_t *)to);
7326 	}
7327 
7328 	return (0);
7329 }
7330 
7331 static int
7332 vdc_get_extvtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7333 {
7334 	int		i, rv;
7335 	struct extvtoc	evtoc;
7336 
7337 	if (dir != VD_COPYOUT)
7338 		return (0);	/* nothing to do */
7339 
7340 	if ((from == NULL) || (to == NULL))
7341 		return (ENXIO);
7342 
7343 	VD_VTOC2VTOC((vd_vtoc_t *)from, &evtoc);
7344 
7345 	/* fake the VTOC timestamp field */
7346 	for (i = 0; i < V_NUMPAR; i++) {
7347 		evtoc.timestamp[i] = vdc->vtoc->timestamp[i];
7348 	}
7349 
7350 	rv = ddi_copyout(&evtoc, to, sizeof (struct extvtoc), mode);
7351 	if (rv != 0)
7352 		rv = EFAULT;
7353 
7354 	return (rv);
7355 }
7356 
7357 static int
7358 vdc_set_extvtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7359 {
7360 	void		*uvtoc;
7361 	struct extvtoc	evtoc;
7362 	int		i, rv;
7363 
7364 	if ((from == NULL) || (to == NULL))
7365 		return (ENXIO);
7366 
7367 	uvtoc = (dir == VD_COPYIN)? from : to;
7368 
7369 	rv = ddi_copyin(uvtoc, &evtoc, sizeof (struct extvtoc), mode);
7370 	if (rv != 0)
7371 		return (EFAULT);
7372 
7373 	if (dir == VD_COPYOUT) {
7374 		/*
7375 		 * The disk label may have changed. Revalidate the disk
7376 		 * geometry. This will also update the device nodes.
7377 		 */
7378 		vdc_validate(vdc);
7379 
7380 		/*
7381 		 * We also need to keep track of the timestamp fields.
7382 		 */
7383 		for (i = 0; i < V_NUMPAR; i++) {
7384 			vdc->vtoc->timestamp[i] = evtoc.timestamp[i];
7385 		}
7386 
7387 	} else {
7388 		VTOC2VD_VTOC(&evtoc, (vd_vtoc_t *)to);
7389 	}
7390 
7391 	return (0);
7392 }
7393 
7394 /*
7395  * Function:
7396  *	vdc_get_geom_convert()
7397  *
7398  * Description:
7399  *	This routine performs the necessary convertions from the DKIOCGGEOM,
7400  *	DKIOCG_PHYSGEOM and DKIOG_VIRTGEOM Solaris structures to the format
7401  *	defined in FWARC 2006/195
7402  *
7403  * Arguments:
7404  *	vdc	- the vDisk client
7405  *	from	- Buffer with data
7406  *	to	- Buffer where data is to be copied to
7407  *	mode	- flags passed to ioctl
7408  *	dir	- direction of copy (in or out)
7409  *
7410  * Return Code:
7411  *	0	- Success
7412  *	ENXIO	- Invalid buffer passed in
7413  *	EFAULT	- ddi_copyout of data failed
7414  */
7415 static int
7416 vdc_get_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7417 {
7418 	_NOTE(ARGUNUSED(vdc))
7419 
7420 	struct dk_geom	geom;
7421 	int	copy_len = sizeof (struct dk_geom);
7422 	int	rv = 0;
7423 
7424 	if (dir != VD_COPYOUT)
7425 		return (0);	/* nothing to do */
7426 
7427 	if ((from == NULL) || (to == NULL))
7428 		return (ENXIO);
7429 
7430 	VD_GEOM2DK_GEOM((vd_geom_t *)from, &geom);
7431 	rv = ddi_copyout(&geom, to, copy_len, mode);
7432 	if (rv != 0)
7433 		rv = EFAULT;
7434 
7435 	return (rv);
7436 }
7437 
7438 /*
7439  * Function:
7440  *	vdc_set_geom_convert()
7441  *
7442  * Description:
7443  *	This routine performs the necessary convertions from the DKIOCSGEOM
7444  *	Solaris structure to the format defined in FWARC 2006/195.
7445  *
7446  * Arguments:
7447  *	vdc	- the vDisk client
7448  *	from	- Buffer with data
7449  *	to	- Buffer where data is to be copied to
7450  *	mode	- flags passed to ioctl
7451  *	dir	- direction of copy (in or out)
7452  *
7453  * Return Code:
7454  *	0	- Success
7455  *	ENXIO	- Invalid buffer passed in
7456  *	EFAULT	- ddi_copyin of data failed
7457  */
7458 static int
7459 vdc_set_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7460 {
7461 	_NOTE(ARGUNUSED(vdc))
7462 
7463 	vd_geom_t	vdgeom;
7464 	void		*tmp_mem = NULL;
7465 	int		copy_len = sizeof (struct dk_geom);
7466 	int		rv = 0;
7467 
7468 	if (dir != VD_COPYIN)
7469 		return (0);	/* nothing to do */
7470 
7471 	if ((from == NULL) || (to == NULL))
7472 		return (ENXIO);
7473 
7474 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
7475 
7476 	rv = ddi_copyin(from, tmp_mem, copy_len, mode);
7477 	if (rv != 0) {
7478 		kmem_free(tmp_mem, copy_len);
7479 		return (EFAULT);
7480 	}
7481 	DK_GEOM2VD_GEOM((struct dk_geom *)tmp_mem, &vdgeom);
7482 	bcopy(&vdgeom, to, sizeof (vdgeom));
7483 	kmem_free(tmp_mem, copy_len);
7484 
7485 	return (0);
7486 }
7487 
7488 static int
7489 vdc_get_efi_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7490 {
7491 	_NOTE(ARGUNUSED(vdc))
7492 
7493 	vd_efi_t	*vd_efi;
7494 	dk_efi_t	dk_efi;
7495 	int		rv = 0;
7496 	void		*uaddr;
7497 
7498 	if ((from == NULL) || (to == NULL))
7499 		return (ENXIO);
7500 
7501 	if (dir == VD_COPYIN) {
7502 
7503 		vd_efi = (vd_efi_t *)to;
7504 
7505 		rv = ddi_copyin(from, &dk_efi, sizeof (dk_efi_t), mode);
7506 		if (rv != 0)
7507 			return (EFAULT);
7508 
7509 		vd_efi->lba = dk_efi.dki_lba;
7510 		vd_efi->length = dk_efi.dki_length;
7511 		bzero(vd_efi->data, vd_efi->length);
7512 
7513 	} else {
7514 
7515 		rv = ddi_copyin(to, &dk_efi, sizeof (dk_efi_t), mode);
7516 		if (rv != 0)
7517 			return (EFAULT);
7518 
7519 		uaddr = dk_efi.dki_data;
7520 
7521 		dk_efi.dki_data = kmem_alloc(dk_efi.dki_length, KM_SLEEP);
7522 
7523 		VD_EFI2DK_EFI((vd_efi_t *)from, &dk_efi);
7524 
7525 		rv = ddi_copyout(dk_efi.dki_data, uaddr, dk_efi.dki_length,
7526 		    mode);
7527 		if (rv != 0)
7528 			return (EFAULT);
7529 
7530 		kmem_free(dk_efi.dki_data, dk_efi.dki_length);
7531 	}
7532 
7533 	return (0);
7534 }
7535 
7536 static int
7537 vdc_set_efi_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7538 {
7539 	_NOTE(ARGUNUSED(vdc))
7540 
7541 	dk_efi_t	dk_efi;
7542 	void		*uaddr;
7543 
7544 	if (dir == VD_COPYOUT) {
7545 		/*
7546 		 * The disk label may have changed. Revalidate the disk
7547 		 * geometry. This will also update the device nodes.
7548 		 */
7549 		vdc_validate(vdc);
7550 		return (0);
7551 	}
7552 
7553 	if ((from == NULL) || (to == NULL))
7554 		return (ENXIO);
7555 
7556 	if (ddi_copyin(from, &dk_efi, sizeof (dk_efi_t), mode) != 0)
7557 		return (EFAULT);
7558 
7559 	uaddr = dk_efi.dki_data;
7560 
7561 	dk_efi.dki_data = kmem_alloc(dk_efi.dki_length, KM_SLEEP);
7562 
7563 	if (ddi_copyin(uaddr, dk_efi.dki_data, dk_efi.dki_length, mode) != 0)
7564 		return (EFAULT);
7565 
7566 	DK_EFI2VD_EFI(&dk_efi, (vd_efi_t *)to);
7567 
7568 	kmem_free(dk_efi.dki_data, dk_efi.dki_length);
7569 
7570 	return (0);
7571 }
7572 
7573 
7574 /* -------------------------------------------------------------------------- */
7575 
7576 /*
7577  * Function:
7578  *	vdc_create_fake_geometry()
7579  *
7580  * Description:
7581  *	This routine fakes up the disk info needed for some DKIO ioctls such
7582  *	as DKIOCINFO and DKIOCGMEDIAINFO [just like lofi(7D) and ramdisk(7D) do]
7583  *
7584  *	Note: This function must not be called until the vDisk attributes have
7585  *	been exchanged as part of the handshake with the vDisk server.
7586  *
7587  * Arguments:
7588  *	vdc	- soft state pointer for this instance of the device driver.
7589  *
7590  * Return Code:
7591  *	none.
7592  */
7593 static void
7594 vdc_create_fake_geometry(vdc_t *vdc)
7595 {
7596 	ASSERT(vdc != NULL);
7597 	ASSERT(vdc->max_xfer_sz != 0);
7598 
7599 	/*
7600 	 * DKIOCINFO support
7601 	 */
7602 	if (vdc->cinfo == NULL)
7603 		vdc->cinfo = kmem_zalloc(sizeof (struct dk_cinfo), KM_SLEEP);
7604 
7605 	(void) strcpy(vdc->cinfo->dki_cname, VDC_DRIVER_NAME);
7606 	(void) strcpy(vdc->cinfo->dki_dname, VDC_DRIVER_NAME);
7607 	/* max_xfer_sz is #blocks so we don't need to divide by DEV_BSIZE */
7608 	vdc->cinfo->dki_maxtransfer = vdc->max_xfer_sz;
7609 
7610 	/*
7611 	 * We set the controller type to DKC_SCSI_CCS only if the VD_OP_SCSICMD
7612 	 * operation is supported, otherwise the controller type is DKC_DIRECT.
7613 	 * Version 1.0 does not support the VD_OP_SCSICMD operation, so the
7614 	 * controller type is always DKC_DIRECT in that case.
7615 	 *
7616 	 * If the virtual disk is backed by a physical CD/DVD device or
7617 	 * an ISO image, modify the controller type to indicate this
7618 	 */
7619 	switch (vdc->vdisk_media) {
7620 	case VD_MEDIA_CD:
7621 	case VD_MEDIA_DVD:
7622 		vdc->cinfo->dki_ctype = DKC_CDROM;
7623 		break;
7624 	case VD_MEDIA_FIXED:
7625 		if (VD_OP_SUPPORTED(vdc->operations, VD_OP_SCSICMD))
7626 			vdc->cinfo->dki_ctype = DKC_SCSI_CCS;
7627 		else
7628 			vdc->cinfo->dki_ctype = DKC_DIRECT;
7629 		break;
7630 	default:
7631 		/* in the case of v1.0 we default to a fixed disk */
7632 		vdc->cinfo->dki_ctype = DKC_DIRECT;
7633 		break;
7634 	}
7635 	vdc->cinfo->dki_flags = DKI_FMTVOL;
7636 	vdc->cinfo->dki_cnum = 0;
7637 	vdc->cinfo->dki_addr = 0;
7638 	vdc->cinfo->dki_space = 0;
7639 	vdc->cinfo->dki_prio = 0;
7640 	vdc->cinfo->dki_vec = 0;
7641 	vdc->cinfo->dki_unit = vdc->instance;
7642 	vdc->cinfo->dki_slave = 0;
7643 	/*
7644 	 * The partition number will be created on the fly depending on the
7645 	 * actual slice (i.e. minor node) that is used to request the data.
7646 	 */
7647 	vdc->cinfo->dki_partition = 0;
7648 
7649 	/*
7650 	 * DKIOCGMEDIAINFO support
7651 	 */
7652 	if (vdc->minfo == NULL)
7653 		vdc->minfo = kmem_zalloc(sizeof (struct dk_minfo), KM_SLEEP);
7654 
7655 	if (vio_ver_is_supported(vdc->ver, 1, 1)) {
7656 		vdc->minfo->dki_media_type =
7657 		    VD_MEDIATYPE2DK_MEDIATYPE(vdc->vdisk_media);
7658 	} else {
7659 		vdc->minfo->dki_media_type = DK_FIXED_DISK;
7660 	}
7661 
7662 	vdc->minfo->dki_capacity = vdc->vdisk_size;
7663 	vdc->minfo->dki_lbsize = vdc->block_size;
7664 }
7665 
7666 static ushort_t
7667 vdc_lbl2cksum(struct dk_label *label)
7668 {
7669 	int	count;
7670 	ushort_t sum, *sp;
7671 
7672 	count =	(sizeof (struct dk_label)) / (sizeof (short)) - 1;
7673 	sp = (ushort_t *)label;
7674 	sum = 0;
7675 	while (count--) {
7676 		sum ^= *sp++;
7677 	}
7678 
7679 	return (sum);
7680 }
7681 
7682 static void
7683 vdc_update_size(vdc_t *vdc, size_t dsk_size, size_t blk_size, size_t xfr_size)
7684 {
7685 	vd_err_stats_t  *stp;
7686 
7687 	ASSERT(MUTEX_HELD(&vdc->lock));
7688 	ASSERT(xfr_size != 0);
7689 
7690 	/*
7691 	 * If the disk size is unknown or sizes are unchanged then don't
7692 	 * update anything.
7693 	 */
7694 	if (dsk_size == VD_SIZE_UNKNOWN || dsk_size == 0 ||
7695 	    (blk_size == vdc->block_size && dsk_size == vdc->vdisk_size &&
7696 	    xfr_size == vdc->max_xfer_sz))
7697 		return;
7698 
7699 	/*
7700 	 * We don't know at compile time what the vDisk server will think
7701 	 * are good values but we apply a large (arbitrary) upper bound to
7702 	 * prevent memory exhaustion in vdc if it was allocating a DRing
7703 	 * based of huge values sent by the server. We probably will never
7704 	 * exceed this except if the message was garbage.
7705 	 */
7706 	if ((xfr_size * blk_size) > (PAGESIZE * DEV_BSIZE)) {
7707 		DMSG(vdc, 0, "[%d] vds block transfer size too big;"
7708 		    " using max supported by vdc", vdc->instance);
7709 		xfr_size = maxphys / DEV_BSIZE;
7710 		dsk_size = (dsk_size * blk_size) / DEV_BSIZE;
7711 		blk_size = DEV_BSIZE;
7712 	}
7713 
7714 	vdc->max_xfer_sz = xfr_size;
7715 	vdc->block_size = blk_size;
7716 	vdc->vdisk_size = dsk_size;
7717 
7718 	stp = (vd_err_stats_t *)vdc->err_stats->ks_data;
7719 	stp->vd_capacity.value.ui64 = dsk_size * blk_size;
7720 
7721 	vdc->minfo->dki_capacity = dsk_size;
7722 	vdc->minfo->dki_lbsize = (uint_t)blk_size;
7723 }
7724 
7725 /*
7726  * Function:
7727  *	vdc_validate_geometry
7728  *
7729  * Description:
7730  *	This routine discovers the label and geometry of the disk. It stores
7731  *	the disk label and related information in the vdc structure. If it
7732  *	fails to validate the geometry or to discover the disk label then
7733  *	the label is marked as unknown (VD_DISK_LABEL_UNK).
7734  *
7735  * Arguments:
7736  *	vdc	- soft state pointer for this instance of the device driver.
7737  *
7738  * Return Code:
7739  *	0	- success.
7740  *	EINVAL	- unknown disk label.
7741  *	ENOTSUP	- geometry not applicable (EFI label).
7742  *	EIO	- error accessing the disk.
7743  */
7744 static int
7745 vdc_validate_geometry(vdc_t *vdc)
7746 {
7747 	buf_t	*buf;	/* BREAD requests need to be in a buf_t structure */
7748 	dev_t	dev;
7749 	int	rv, rval;
7750 	struct dk_label label;
7751 	struct dk_geom geom;
7752 	struct extvtoc vtoc;
7753 	efi_gpt_t *gpt;
7754 	efi_gpe_t *gpe;
7755 	vd_efi_dev_t edev;
7756 
7757 	ASSERT(vdc != NULL);
7758 	ASSERT(vdc->vtoc != NULL && vdc->geom != NULL);
7759 	ASSERT(MUTEX_HELD(&vdc->lock));
7760 
7761 	mutex_exit(&vdc->lock);
7762 	/*
7763 	 * Check the disk capacity in case it has changed. If that fails then
7764 	 * we proceed and we will be using the disk size we currently have.
7765 	 */
7766 	(void) vdc_check_capacity(vdc);
7767 	dev = makedevice(ddi_driver_major(vdc->dip),
7768 	    VD_MAKE_DEV(vdc->instance, 0));
7769 
7770 	rv = vd_process_ioctl(dev, DKIOCGGEOM, (caddr_t)&geom, FKIOCTL, &rval);
7771 	if (rv == 0)
7772 		rv = vd_process_ioctl(dev, DKIOCGEXTVTOC, (caddr_t)&vtoc,
7773 		    FKIOCTL, &rval);
7774 
7775 	if (rv == ENOTSUP) {
7776 		/*
7777 		 * If the device does not support VTOC then we try
7778 		 * to read an EFI label.
7779 		 *
7780 		 * We need to know the block size and the disk size to
7781 		 * be able to read an EFI label.
7782 		 */
7783 		if (vdc->vdisk_size == 0) {
7784 			mutex_enter(&vdc->lock);
7785 			vdc_store_label_unk(vdc);
7786 			return (EIO);
7787 		}
7788 
7789 		VD_EFI_DEV_SET(edev, vdc, vd_process_efi_ioctl);
7790 
7791 		rv = vd_efi_alloc_and_read(&edev, &gpt, &gpe);
7792 
7793 		if (rv) {
7794 			DMSG(vdc, 0, "[%d] Failed to get EFI (err=%d)",
7795 			    vdc->instance, rv);
7796 			mutex_enter(&vdc->lock);
7797 			vdc_store_label_unk(vdc);
7798 			return (EIO);
7799 		}
7800 
7801 		mutex_enter(&vdc->lock);
7802 		vdc_store_label_efi(vdc, gpt, gpe);
7803 		vd_efi_free(&edev, gpt, gpe);
7804 		return (ENOTSUP);
7805 	}
7806 
7807 	if (rv != 0) {
7808 		DMSG(vdc, 0, "[%d] Failed to get VTOC (err=%d)",
7809 		    vdc->instance, rv);
7810 		mutex_enter(&vdc->lock);
7811 		vdc_store_label_unk(vdc);
7812 		if (rv != EINVAL)
7813 			rv = EIO;
7814 		return (rv);
7815 	}
7816 
7817 	/* check that geometry and vtoc are valid */
7818 	if (geom.dkg_nhead == 0 || geom.dkg_nsect == 0 ||
7819 	    vtoc.v_sanity != VTOC_SANE) {
7820 		mutex_enter(&vdc->lock);
7821 		vdc_store_label_unk(vdc);
7822 		return (EINVAL);
7823 	}
7824 
7825 	/*
7826 	 * We have a disk and a valid VTOC. However this does not mean
7827 	 * that the disk currently have a VTOC label. The returned VTOC may
7828 	 * be a default VTOC to be used for configuring the disk (this is
7829 	 * what is done for disk image). So we read the label from the
7830 	 * beginning of the disk to ensure we really have a VTOC label.
7831 	 *
7832 	 * FUTURE: This could be the default way for reading the VTOC
7833 	 * from the disk as opposed to sending the VD_OP_GET_VTOC
7834 	 * to the server. This will be the default if vdc is implemented
7835 	 * ontop of cmlb.
7836 	 */
7837 
7838 	/*
7839 	 * Single slice disk does not support read using an absolute disk
7840 	 * offset so we just rely on the DKIOCGVTOC ioctl in that case.
7841 	 */
7842 	if (vdc->vdisk_type == VD_DISK_TYPE_SLICE) {
7843 		mutex_enter(&vdc->lock);
7844 		if (vtoc.v_nparts != 1) {
7845 			vdc_store_label_unk(vdc);
7846 			return (EINVAL);
7847 		}
7848 		vdc_store_label_vtoc(vdc, &geom, &vtoc);
7849 		return (0);
7850 	}
7851 
7852 	if (vtoc.v_nparts != V_NUMPAR) {
7853 		mutex_enter(&vdc->lock);
7854 		vdc_store_label_unk(vdc);
7855 		return (EINVAL);
7856 	}
7857 
7858 	/*
7859 	 * Most CD/DVDs do not have a disk label and the label is
7860 	 * generated by the disk driver.  So the on-disk label check
7861 	 * below may fail and we return now to avoid this problem.
7862 	 */
7863 	if (vdc->vdisk_media == VD_MEDIA_CD ||
7864 	    vdc->vdisk_media == VD_MEDIA_DVD) {
7865 		mutex_enter(&vdc->lock);
7866 		vdc_store_label_vtoc(vdc, &geom, &vtoc);
7867 		return (0);
7868 	}
7869 
7870 	/*
7871 	 * Read disk label from start of disk
7872 	 */
7873 	buf = kmem_alloc(sizeof (buf_t), KM_SLEEP);
7874 	bioinit(buf);
7875 	buf->b_un.b_addr = (caddr_t)&label;
7876 	buf->b_bcount = DK_LABEL_SIZE;
7877 	buf->b_flags = B_BUSY | B_READ;
7878 	buf->b_dev = cmpdev(dev);
7879 	rv = vdc_send_request(vdc, VD_OP_BREAD, (caddr_t)&label,
7880 	    DK_LABEL_SIZE, VD_SLICE_NONE, 0, CB_STRATEGY, buf, VIO_read_dir);
7881 	if (rv) {
7882 		DMSG(vdc, 1, "[%d] Failed to read disk block 0\n",
7883 		    vdc->instance);
7884 	} else if (ddi_in_panic()) {
7885 		rv = vdc_drain_response(vdc, CB_STRATEGY, buf);
7886 		if (rv == 0) {
7887 			rv = geterror(buf);
7888 		}
7889 	} else {
7890 		rv = biowait(buf);
7891 	}
7892 	biofini(buf);
7893 	kmem_free(buf, sizeof (buf_t));
7894 
7895 	if (rv != 0 || label.dkl_magic != DKL_MAGIC ||
7896 	    label.dkl_cksum != vdc_lbl2cksum(&label)) {
7897 		DMSG(vdc, 1, "[%d] Got VTOC with invalid label\n",
7898 		    vdc->instance);
7899 		mutex_enter(&vdc->lock);
7900 		vdc_store_label_unk(vdc);
7901 		return (EINVAL);
7902 	}
7903 
7904 	mutex_enter(&vdc->lock);
7905 	vdc_store_label_vtoc(vdc, &geom, &vtoc);
7906 	return (0);
7907 }
7908 
7909 /*
7910  * Function:
7911  *	vdc_validate
7912  *
7913  * Description:
7914  *	This routine discovers the label of the disk and create the
7915  *	appropriate device nodes if the label has changed.
7916  *
7917  * Arguments:
7918  *	vdc	- soft state pointer for this instance of the device driver.
7919  *
7920  * Return Code:
7921  *	none.
7922  */
7923 static void
7924 vdc_validate(vdc_t *vdc)
7925 {
7926 	vd_disk_label_t old_label;
7927 	vd_slice_t old_slice[V_NUMPAR];
7928 	int rv;
7929 
7930 	ASSERT(!MUTEX_HELD(&vdc->lock));
7931 
7932 	mutex_enter(&vdc->lock);
7933 
7934 	/* save the current label and vtoc */
7935 	old_label = vdc->vdisk_label;
7936 	bcopy(vdc->slice, &old_slice, sizeof (vd_slice_t) * V_NUMPAR);
7937 
7938 	/* check the geometry */
7939 	(void) vdc_validate_geometry(vdc);
7940 
7941 	/* if the disk label has changed, update device nodes */
7942 	if (vdc->vdisk_label != old_label) {
7943 
7944 		if (vdc->vdisk_label == VD_DISK_LABEL_EFI)
7945 			rv = vdc_create_device_nodes_efi(vdc);
7946 		else
7947 			rv = vdc_create_device_nodes_vtoc(vdc);
7948 
7949 		if (rv != 0) {
7950 			DMSG(vdc, 0, "![%d] Failed to update device nodes",
7951 			    vdc->instance);
7952 		}
7953 	}
7954 
7955 	mutex_exit(&vdc->lock);
7956 }
7957 
7958 static void
7959 vdc_validate_task(void *arg)
7960 {
7961 	vdc_t *vdc = (vdc_t *)arg;
7962 
7963 	vdc_validate(vdc);
7964 
7965 	mutex_enter(&vdc->lock);
7966 	ASSERT(vdc->validate_pending > 0);
7967 	vdc->validate_pending--;
7968 	mutex_exit(&vdc->lock);
7969 }
7970 
7971 /*
7972  * Function:
7973  *	vdc_setup_devid()
7974  *
7975  * Description:
7976  *	This routine discovers the devid of a vDisk. It requests the devid of
7977  *	the underlying device from the vDisk server, builds an encapsulated
7978  *	devid based on the retrieved devid and registers that new devid to
7979  *	the vDisk.
7980  *
7981  * Arguments:
7982  *	vdc	- soft state pointer for this instance of the device driver.
7983  *
7984  * Return Code:
7985  *	0	- A devid was succesfully registered for the vDisk
7986  */
7987 static int
7988 vdc_setup_devid(vdc_t *vdc)
7989 {
7990 	int rv;
7991 	vd_devid_t *vd_devid;
7992 	size_t bufsize, bufid_len;
7993 
7994 	/*
7995 	 * At first sight, we don't know the size of the devid that the
7996 	 * server will return but this size will be encoded into the
7997 	 * reply. So we do a first request using a default size then we
7998 	 * check if this size was large enough. If not then we do a second
7999 	 * request with the correct size returned by the server. Note that
8000 	 * ldc requires size to be 8-byte aligned.
8001 	 */
8002 	bufsize = P2ROUNDUP(VD_DEVID_SIZE(VD_DEVID_DEFAULT_LEN),
8003 	    sizeof (uint64_t));
8004 	vd_devid = kmem_zalloc(bufsize, KM_SLEEP);
8005 	bufid_len = bufsize - sizeof (vd_efi_t) - 1;
8006 
8007 	rv = vdc_do_sync_op(vdc, VD_OP_GET_DEVID, (caddr_t)vd_devid,
8008 	    bufsize, 0, 0, CB_SYNC, 0, VIO_both_dir, B_TRUE);
8009 
8010 	DMSG(vdc, 2, "sync_op returned %d\n", rv);
8011 
8012 	if (rv) {
8013 		kmem_free(vd_devid, bufsize);
8014 		return (rv);
8015 	}
8016 
8017 	if (vd_devid->length > bufid_len) {
8018 		/*
8019 		 * The returned devid is larger than the buffer used. Try again
8020 		 * with a buffer with the right size.
8021 		 */
8022 		kmem_free(vd_devid, bufsize);
8023 		bufsize = P2ROUNDUP(VD_DEVID_SIZE(vd_devid->length),
8024 		    sizeof (uint64_t));
8025 		vd_devid = kmem_zalloc(bufsize, KM_SLEEP);
8026 		bufid_len = bufsize - sizeof (vd_efi_t) - 1;
8027 
8028 		rv = vdc_do_sync_op(vdc, VD_OP_GET_DEVID,
8029 		    (caddr_t)vd_devid, bufsize, 0, 0, CB_SYNC, 0,
8030 		    VIO_both_dir, B_TRUE);
8031 
8032 		if (rv) {
8033 			kmem_free(vd_devid, bufsize);
8034 			return (rv);
8035 		}
8036 	}
8037 
8038 	/*
8039 	 * The virtual disk should have the same device id as the one associated
8040 	 * with the physical disk it is mapped on, otherwise sharing a disk
8041 	 * between a LDom and a non-LDom may not work (for example for a shared
8042 	 * SVM disk set).
8043 	 *
8044 	 * The DDI framework does not allow creating a device id with any
8045 	 * type so we first create a device id of type DEVID_ENCAP and then
8046 	 * we restore the orignal type of the physical device.
8047 	 */
8048 
8049 	DMSG(vdc, 2, ": devid length = %d\n", vd_devid->length);
8050 
8051 	/* build an encapsulated devid based on the returned devid */
8052 	if (ddi_devid_init(vdc->dip, DEVID_ENCAP, vd_devid->length,
8053 	    vd_devid->id, &vdc->devid) != DDI_SUCCESS) {
8054 		DMSG(vdc, 1, "[%d] Fail to created devid\n", vdc->instance);
8055 		kmem_free(vd_devid, bufsize);
8056 		return (1);
8057 	}
8058 
8059 	DEVID_FORMTYPE((impl_devid_t *)vdc->devid, vd_devid->type);
8060 
8061 	ASSERT(ddi_devid_valid(vdc->devid) == DDI_SUCCESS);
8062 
8063 	kmem_free(vd_devid, bufsize);
8064 
8065 	if (ddi_devid_register(vdc->dip, vdc->devid) != DDI_SUCCESS) {
8066 		DMSG(vdc, 1, "[%d] Fail to register devid\n", vdc->instance);
8067 		return (1);
8068 	}
8069 
8070 	return (0);
8071 }
8072 
8073 static void
8074 vdc_store_label_efi(vdc_t *vdc, efi_gpt_t *gpt, efi_gpe_t *gpe)
8075 {
8076 	int i, nparts;
8077 
8078 	ASSERT(MUTEX_HELD(&vdc->lock));
8079 
8080 	vdc->vdisk_label = VD_DISK_LABEL_EFI;
8081 	bzero(vdc->vtoc, sizeof (struct extvtoc));
8082 	bzero(vdc->geom, sizeof (struct dk_geom));
8083 	bzero(vdc->slice, sizeof (vd_slice_t) * V_NUMPAR);
8084 
8085 	nparts = gpt->efi_gpt_NumberOfPartitionEntries;
8086 
8087 	for (i = 0; i < nparts && i < VD_EFI_WD_SLICE; i++) {
8088 
8089 		if (gpe[i].efi_gpe_StartingLBA == 0 &&
8090 		    gpe[i].efi_gpe_EndingLBA == 0) {
8091 			continue;
8092 		}
8093 
8094 		vdc->slice[i].start = gpe[i].efi_gpe_StartingLBA;
8095 		vdc->slice[i].nblocks = gpe[i].efi_gpe_EndingLBA -
8096 		    gpe[i].efi_gpe_StartingLBA + 1;
8097 	}
8098 
8099 	ASSERT(vdc->vdisk_size != 0);
8100 	vdc->slice[VD_EFI_WD_SLICE].start = 0;
8101 	vdc->slice[VD_EFI_WD_SLICE].nblocks = vdc->vdisk_size;
8102 
8103 }
8104 
8105 static void
8106 vdc_store_label_vtoc(vdc_t *vdc, struct dk_geom *geom, struct extvtoc *vtoc)
8107 {
8108 	int i;
8109 
8110 	ASSERT(MUTEX_HELD(&vdc->lock));
8111 	ASSERT(vdc->block_size == vtoc->v_sectorsz);
8112 
8113 	vdc->vdisk_label = VD_DISK_LABEL_VTOC;
8114 	bcopy(vtoc, vdc->vtoc, sizeof (struct extvtoc));
8115 	bcopy(geom, vdc->geom, sizeof (struct dk_geom));
8116 	bzero(vdc->slice, sizeof (vd_slice_t) * V_NUMPAR);
8117 
8118 	for (i = 0; i < vtoc->v_nparts; i++) {
8119 		vdc->slice[i].start = vtoc->v_part[i].p_start;
8120 		vdc->slice[i].nblocks = vtoc->v_part[i].p_size;
8121 	}
8122 }
8123 
8124 static void
8125 vdc_store_label_unk(vdc_t *vdc)
8126 {
8127 	ASSERT(MUTEX_HELD(&vdc->lock));
8128 
8129 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
8130 	bzero(vdc->vtoc, sizeof (struct extvtoc));
8131 	bzero(vdc->geom, sizeof (struct dk_geom));
8132 	bzero(vdc->slice, sizeof (vd_slice_t) * V_NUMPAR);
8133 }
8134