xref: /titanic_41/usr/src/uts/sun4v/io/vdc.c (revision d4660949aa62dd6a963f4913b7120b383cf473c4)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 /*
30  * LDoms virtual disk client (vdc) device driver
31  *
32  * This driver runs on a guest logical domain and communicates with the virtual
33  * disk server (vds) driver running on the service domain which is exporting
34  * virtualized "disks" to the guest logical domain.
35  *
36  * The driver can be divided into four sections:
37  *
38  * 1) generic device driver housekeeping
39  *	_init, _fini, attach, detach, ops structures, etc.
40  *
41  * 2) communication channel setup
42  *	Setup the communications link over the LDC channel that vdc uses to
43  *	talk to the vDisk server. Initialise the descriptor ring which
44  *	allows the LDC clients to transfer data via memory mappings.
45  *
46  * 3) Support exported to upper layers (filesystems, etc)
47  *	The upper layers call into vdc via strategy(9E) and DKIO(7I)
48  *	ioctl calls. vdc will copy the data to be written to the descriptor
49  *	ring or maps the buffer to store the data read by the vDisk
50  *	server into the descriptor ring. It then sends a message to the
51  *	vDisk server requesting it to complete the operation.
52  *
53  * 4) Handling responses from vDisk server.
54  *	The vDisk server will ACK some or all of the messages vdc sends to it
55  *	(this is configured during the handshake). Upon receipt of an ACK
56  *	vdc will check the descriptor ring and signal to the upper layer
57  *	code waiting on the IO.
58  */
59 
60 #include <sys/atomic.h>
61 #include <sys/conf.h>
62 #include <sys/disp.h>
63 #include <sys/ddi.h>
64 #include <sys/dkio.h>
65 #include <sys/efi_partition.h>
66 #include <sys/fcntl.h>
67 #include <sys/file.h>
68 #include <sys/kstat.h>
69 #include <sys/mach_descrip.h>
70 #include <sys/modctl.h>
71 #include <sys/mdeg.h>
72 #include <sys/note.h>
73 #include <sys/open.h>
74 #include <sys/sdt.h>
75 #include <sys/stat.h>
76 #include <sys/sunddi.h>
77 #include <sys/types.h>
78 #include <sys/promif.h>
79 #include <sys/var.h>
80 #include <sys/vtoc.h>
81 #include <sys/archsystm.h>
82 #include <sys/sysmacros.h>
83 
84 #include <sys/cdio.h>
85 #include <sys/dktp/fdisk.h>
86 #include <sys/dktp/dadkio.h>
87 #include <sys/mhd.h>
88 #include <sys/scsi/generic/sense.h>
89 #include <sys/scsi/impl/uscsi.h>
90 #include <sys/scsi/impl/services.h>
91 #include <sys/scsi/targets/sddef.h>
92 
93 #include <sys/ldoms.h>
94 #include <sys/ldc.h>
95 #include <sys/vio_common.h>
96 #include <sys/vio_mailbox.h>
97 #include <sys/vio_util.h>
98 #include <sys/vdsk_common.h>
99 #include <sys/vdsk_mailbox.h>
100 #include <sys/vdc.h>
101 
102 /*
103  * function prototypes
104  */
105 
106 /* standard driver functions */
107 static int	vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred);
108 static int	vdc_close(dev_t dev, int flag, int otyp, cred_t *cred);
109 static int	vdc_strategy(struct buf *buf);
110 static int	vdc_print(dev_t dev, char *str);
111 static int	vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk);
112 static int	vdc_read(dev_t dev, struct uio *uio, cred_t *cred);
113 static int	vdc_write(dev_t dev, struct uio *uio, cred_t *cred);
114 static int	vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode,
115 			cred_t *credp, int *rvalp);
116 static int	vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred);
117 static int	vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred);
118 
119 static int	vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,
120 			void *arg, void **resultp);
121 static int	vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd);
122 static int	vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd);
123 
124 /* setup */
125 static void	vdc_min(struct buf *bufp);
126 static int	vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen);
127 static int	vdc_do_ldc_init(vdc_t *vdc, vdc_server_t *srvr);
128 static int	vdc_start_ldc_connection(vdc_t *vdc);
129 static int	vdc_create_device_nodes(vdc_t *vdc);
130 static int	vdc_create_device_nodes_efi(vdc_t *vdc);
131 static int	vdc_create_device_nodes_vtoc(vdc_t *vdc);
132 static int	vdc_create_device_nodes_props(vdc_t *vdc);
133 static void	vdc_create_io_kstats(vdc_t *vdc);
134 static void	vdc_create_err_kstats(vdc_t *vdc);
135 static void	vdc_set_err_kstats(vdc_t *vdc);
136 static int	vdc_get_md_node(dev_info_t *dip, md_t **mdpp,
137 		    mde_cookie_t *vd_nodep);
138 static int	vdc_init_ports(vdc_t *vdc, md_t *mdp, mde_cookie_t vd_nodep);
139 static void	vdc_fini_ports(vdc_t *vdc);
140 static void	vdc_switch_server(vdc_t *vdcp);
141 static int	vdc_do_ldc_up(vdc_t *vdc);
142 static void	vdc_terminate_ldc(vdc_t *vdc, vdc_server_t *srvr);
143 static int	vdc_init_descriptor_ring(vdc_t *vdc);
144 static void	vdc_destroy_descriptor_ring(vdc_t *vdc);
145 static int	vdc_setup_devid(vdc_t *vdc);
146 static void	vdc_store_label_efi(vdc_t *, efi_gpt_t *, efi_gpe_t *);
147 static void	vdc_store_label_vtoc(vdc_t *, struct dk_geom *, struct vtoc *);
148 static void	vdc_store_label_unk(vdc_t *vdc);
149 static boolean_t vdc_is_opened(vdc_t *vdc);
150 
151 /* handshake with vds */
152 static int		vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver);
153 static int		vdc_ver_negotiation(vdc_t *vdcp);
154 static int		vdc_init_attr_negotiation(vdc_t *vdc);
155 static int		vdc_attr_negotiation(vdc_t *vdcp);
156 static int		vdc_init_dring_negotiate(vdc_t *vdc);
157 static int		vdc_dring_negotiation(vdc_t *vdcp);
158 static int		vdc_send_rdx(vdc_t *vdcp);
159 static int		vdc_rdx_exchange(vdc_t *vdcp);
160 static boolean_t	vdc_is_supported_version(vio_ver_msg_t *ver_msg);
161 
162 /* processing incoming messages from vDisk server */
163 static void	vdc_process_msg_thread(vdc_t *vdc);
164 static int	vdc_recv(vdc_t *vdc, vio_msg_t *msgp, size_t *nbytesp);
165 
166 static uint_t	vdc_handle_cb(uint64_t event, caddr_t arg);
167 static int	vdc_process_data_msg(vdc_t *vdc, vio_msg_t *msg);
168 static int	vdc_handle_ver_msg(vdc_t *vdc, vio_ver_msg_t *ver_msg);
169 static int	vdc_handle_attr_msg(vdc_t *vdc, vd_attr_msg_t *attr_msg);
170 static int	vdc_handle_dring_reg_msg(vdc_t *vdc, vio_dring_reg_msg_t *msg);
171 static int 	vdc_send_request(vdc_t *vdcp, int operation,
172 		    caddr_t addr, size_t nbytes, int slice, diskaddr_t offset,
173 		    int cb_type, void *cb_arg, vio_desc_direction_t dir);
174 static int	vdc_map_to_shared_dring(vdc_t *vdcp, int idx);
175 static int 	vdc_populate_descriptor(vdc_t *vdcp, int operation,
176 		    caddr_t addr, size_t nbytes, int slice, diskaddr_t offset,
177 		    int cb_type, void *cb_arg, vio_desc_direction_t dir);
178 static int 	vdc_do_sync_op(vdc_t *vdcp, int operation, caddr_t addr,
179 		    size_t nbytes, int slice, diskaddr_t offset, int cb_type,
180 		    void *cb_arg, vio_desc_direction_t dir, boolean_t);
181 
182 static int	vdc_wait_for_response(vdc_t *vdcp, vio_msg_t *msgp);
183 static int	vdc_drain_response(vdc_t *vdcp);
184 static int	vdc_depopulate_descriptor(vdc_t *vdc, uint_t idx);
185 static int	vdc_populate_mem_hdl(vdc_t *vdcp, vdc_local_desc_t *ldep);
186 static int	vdc_verify_seq_num(vdc_t *vdc, vio_dring_msg_t *dring_msg);
187 
188 /* dkio */
189 static int	vd_process_ioctl(dev_t dev, int cmd, caddr_t arg, int mode,
190 		    int *rvalp);
191 static int	vd_process_efi_ioctl(void *vdisk, int cmd, uintptr_t arg);
192 static void	vdc_create_fake_geometry(vdc_t *vdc);
193 static int	vdc_validate_geometry(vdc_t *vdc);
194 static void	vdc_validate(vdc_t *vdc);
195 static void	vdc_validate_task(void *arg);
196 static int	vdc_null_copy_func(vdc_t *vdc, void *from, void *to,
197 		    int mode, int dir);
198 static int	vdc_get_wce_convert(vdc_t *vdc, void *from, void *to,
199 		    int mode, int dir);
200 static int	vdc_set_wce_convert(vdc_t *vdc, void *from, void *to,
201 		    int mode, int dir);
202 static int	vdc_get_vtoc_convert(vdc_t *vdc, void *from, void *to,
203 		    int mode, int dir);
204 static int	vdc_set_vtoc_convert(vdc_t *vdc, void *from, void *to,
205 		    int mode, int dir);
206 static int	vdc_get_geom_convert(vdc_t *vdc, void *from, void *to,
207 		    int mode, int dir);
208 static int	vdc_set_geom_convert(vdc_t *vdc, void *from, void *to,
209 		    int mode, int dir);
210 static int	vdc_get_efi_convert(vdc_t *vdc, void *from, void *to,
211 		    int mode, int dir);
212 static int	vdc_set_efi_convert(vdc_t *vdc, void *from, void *to,
213 		    int mode, int dir);
214 
215 static void 	vdc_ownership_update(vdc_t *vdc, int ownership_flags);
216 static int	vdc_access_set(vdc_t *vdc, uint64_t flags, int mode);
217 static vdc_io_t	*vdc_failfast_io_queue(vdc_t *vdc, struct buf *buf);
218 static int	vdc_failfast_check_resv(vdc_t *vdc);
219 
220 /*
221  * Module variables
222  */
223 
224 /*
225  * Tunable variables to control how long vdc waits before timing out on
226  * various operations
227  */
228 static int	vdc_hshake_retries = 3;
229 
230 static int	vdc_timeout = 0; /* units: seconds */
231 static int 	vdc_ldcup_timeout = 1; /* units: seconds */
232 
233 static uint64_t vdc_hz_min_ldc_delay;
234 static uint64_t vdc_min_timeout_ldc = 1 * MILLISEC;
235 static uint64_t vdc_hz_max_ldc_delay;
236 static uint64_t vdc_max_timeout_ldc = 100 * MILLISEC;
237 
238 static uint64_t vdc_ldc_read_init_delay = 1 * MILLISEC;
239 static uint64_t vdc_ldc_read_max_delay = 100 * MILLISEC;
240 
241 /* values for dumping - need to run in a tighter loop */
242 static uint64_t	vdc_usec_timeout_dump = 100 * MILLISEC;	/* 0.1s units: ns */
243 static int	vdc_dump_retries = 100;
244 
245 static uint16_t	vdc_scsi_timeout = 60;	/* 60s units: seconds  */
246 
247 static uint64_t vdc_ownership_delay = 6 * MICROSEC; /* 6s units: usec */
248 
249 /* Count of the number of vdc instances attached */
250 static volatile uint32_t	vdc_instance_count = 0;
251 
252 /* Tunable to log all SCSI errors */
253 static boolean_t vdc_scsi_log_error = B_FALSE;
254 
255 /* Soft state pointer */
256 static void	*vdc_state;
257 
258 /*
259  * Controlling the verbosity of the error/debug messages
260  *
261  * vdc_msglevel - controls level of messages
262  * vdc_matchinst - 64-bit variable where each bit corresponds
263  *                 to the vdc instance the vdc_msglevel applies.
264  */
265 int		vdc_msglevel = 0x0;
266 uint64_t	vdc_matchinst = 0ull;
267 
268 /*
269  * Supported vDisk protocol version pairs.
270  *
271  * The first array entry is the latest and preferred version.
272  */
273 static const vio_ver_t	vdc_version[] = {{1, 1}};
274 
275 static struct cb_ops vdc_cb_ops = {
276 	vdc_open,	/* cb_open */
277 	vdc_close,	/* cb_close */
278 	vdc_strategy,	/* cb_strategy */
279 	vdc_print,	/* cb_print */
280 	vdc_dump,	/* cb_dump */
281 	vdc_read,	/* cb_read */
282 	vdc_write,	/* cb_write */
283 	vdc_ioctl,	/* cb_ioctl */
284 	nodev,		/* cb_devmap */
285 	nodev,		/* cb_mmap */
286 	nodev,		/* cb_segmap */
287 	nochpoll,	/* cb_chpoll */
288 	ddi_prop_op,	/* cb_prop_op */
289 	NULL,		/* cb_str */
290 	D_MP | D_64BIT,	/* cb_flag */
291 	CB_REV,		/* cb_rev */
292 	vdc_aread,	/* cb_aread */
293 	vdc_awrite	/* cb_awrite */
294 };
295 
296 static struct dev_ops vdc_ops = {
297 	DEVO_REV,	/* devo_rev */
298 	0,		/* devo_refcnt */
299 	vdc_getinfo,	/* devo_getinfo */
300 	nulldev,	/* devo_identify */
301 	nulldev,	/* devo_probe */
302 	vdc_attach,	/* devo_attach */
303 	vdc_detach,	/* devo_detach */
304 	nodev,		/* devo_reset */
305 	&vdc_cb_ops,	/* devo_cb_ops */
306 	NULL,		/* devo_bus_ops */
307 	nulldev		/* devo_power */
308 };
309 
310 static struct modldrv modldrv = {
311 	&mod_driverops,
312 	"virtual disk client",
313 	&vdc_ops,
314 };
315 
316 static struct modlinkage modlinkage = {
317 	MODREV_1,
318 	&modldrv,
319 	NULL
320 };
321 
322 /* -------------------------------------------------------------------------- */
323 
324 /*
325  * Device Driver housekeeping and setup
326  */
327 
328 int
329 _init(void)
330 {
331 	int	status;
332 
333 	if ((status = ddi_soft_state_init(&vdc_state, sizeof (vdc_t), 1)) != 0)
334 		return (status);
335 	if ((status = mod_install(&modlinkage)) != 0)
336 		ddi_soft_state_fini(&vdc_state);
337 	return (status);
338 }
339 
340 int
341 _info(struct modinfo *modinfop)
342 {
343 	return (mod_info(&modlinkage, modinfop));
344 }
345 
346 int
347 _fini(void)
348 {
349 	int	status;
350 
351 	if ((status = mod_remove(&modlinkage)) != 0)
352 		return (status);
353 	ddi_soft_state_fini(&vdc_state);
354 	return (0);
355 }
356 
357 static int
358 vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,  void *arg, void **resultp)
359 {
360 	_NOTE(ARGUNUSED(dip))
361 
362 	int	instance = VDCUNIT((dev_t)arg);
363 	vdc_t	*vdc = NULL;
364 
365 	switch (cmd) {
366 	case DDI_INFO_DEVT2DEVINFO:
367 		if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
368 			*resultp = NULL;
369 			return (DDI_FAILURE);
370 		}
371 		*resultp = vdc->dip;
372 		return (DDI_SUCCESS);
373 	case DDI_INFO_DEVT2INSTANCE:
374 		*resultp = (void *)(uintptr_t)instance;
375 		return (DDI_SUCCESS);
376 	default:
377 		*resultp = NULL;
378 		return (DDI_FAILURE);
379 	}
380 }
381 
382 static int
383 vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
384 {
385 	kt_did_t failfast_tid, ownership_tid;
386 	int	instance;
387 	int	rv;
388 	vdc_t	*vdc = NULL;
389 
390 	switch (cmd) {
391 	case DDI_DETACH:
392 		/* the real work happens below */
393 		break;
394 	case DDI_SUSPEND:
395 		/* nothing to do for this non-device */
396 		return (DDI_SUCCESS);
397 	default:
398 		return (DDI_FAILURE);
399 	}
400 
401 	ASSERT(cmd == DDI_DETACH);
402 	instance = ddi_get_instance(dip);
403 	DMSGX(1, "[%d] Entered\n", instance);
404 
405 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
406 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
407 		return (DDI_FAILURE);
408 	}
409 
410 	/*
411 	 * This function is called when vdc is detached or if it has failed to
412 	 * attach. In that case, the attach may have fail before the vdisk type
413 	 * has been set so we can't call vdc_is_opened(). However as the attach
414 	 * has failed, we know that the vdisk is not opened and we can safely
415 	 * detach.
416 	 */
417 	if (vdc->vdisk_type != VD_DISK_TYPE_UNK && vdc_is_opened(vdc)) {
418 		DMSG(vdc, 0, "[%d] Cannot detach: device is open", instance);
419 		return (DDI_FAILURE);
420 	}
421 
422 	if (vdc->dkio_flush_pending) {
423 		DMSG(vdc, 0,
424 		    "[%d] Cannot detach: %d outstanding DKIO flushes\n",
425 		    instance, vdc->dkio_flush_pending);
426 		return (DDI_FAILURE);
427 	}
428 
429 	if (vdc->validate_pending) {
430 		DMSG(vdc, 0,
431 		    "[%d] Cannot detach: %d outstanding validate request\n",
432 		    instance, vdc->validate_pending);
433 		return (DDI_FAILURE);
434 	}
435 
436 	DMSG(vdc, 0, "[%d] proceeding...\n", instance);
437 
438 	/* If we took ownership, release ownership */
439 	mutex_enter(&vdc->ownership_lock);
440 	if (vdc->ownership & VDC_OWNERSHIP_GRANTED) {
441 		rv = vdc_access_set(vdc, VD_ACCESS_SET_CLEAR, FKIOCTL);
442 		if (rv == 0) {
443 			vdc_ownership_update(vdc, VDC_OWNERSHIP_NONE);
444 		}
445 	}
446 	mutex_exit(&vdc->ownership_lock);
447 
448 	/* mark instance as detaching */
449 	vdc->lifecycle	= VDC_LC_DETACHING;
450 
451 	/*
452 	 * try and disable callbacks to prevent another handshake
453 	 */
454 	if (vdc->curr_server != NULL) {
455 		rv = ldc_set_cb_mode(vdc->curr_server->ldc_handle,
456 		    LDC_CB_DISABLE);
457 		DMSG(vdc, 0, "callback disabled (rv=%d)\n", rv);
458 	}
459 
460 	if (vdc->initialized & VDC_THREAD) {
461 		mutex_enter(&vdc->read_lock);
462 		if ((vdc->read_state == VDC_READ_WAITING) ||
463 		    (vdc->read_state == VDC_READ_RESET)) {
464 			vdc->read_state = VDC_READ_RESET;
465 			cv_signal(&vdc->read_cv);
466 		}
467 
468 		mutex_exit(&vdc->read_lock);
469 
470 		/* wake up any thread waiting for connection to come online */
471 		mutex_enter(&vdc->lock);
472 		if (vdc->state == VDC_STATE_INIT_WAITING) {
473 			DMSG(vdc, 0,
474 			    "[%d] write reset - move to resetting state...\n",
475 			    instance);
476 			vdc->state = VDC_STATE_RESETTING;
477 			cv_signal(&vdc->initwait_cv);
478 		}
479 		mutex_exit(&vdc->lock);
480 
481 		/* now wait until state transitions to VDC_STATE_DETACH */
482 		thread_join(vdc->msg_proc_thr->t_did);
483 		ASSERT(vdc->state == VDC_STATE_DETACH);
484 		DMSG(vdc, 0, "[%d] Reset thread exit and join ..\n",
485 		    vdc->instance);
486 	}
487 
488 	mutex_enter(&vdc->lock);
489 
490 	if (vdc->initialized & VDC_DRING)
491 		vdc_destroy_descriptor_ring(vdc);
492 
493 	vdc_fini_ports(vdc);
494 
495 	if (vdc->failfast_thread) {
496 		failfast_tid = vdc->failfast_thread->t_did;
497 		vdc->failfast_interval = 0;
498 		cv_signal(&vdc->failfast_cv);
499 	} else {
500 		failfast_tid = 0;
501 	}
502 
503 	if (vdc->ownership & VDC_OWNERSHIP_WANTED) {
504 		ownership_tid = vdc->ownership_thread->t_did;
505 		vdc->ownership = VDC_OWNERSHIP_NONE;
506 		cv_signal(&vdc->ownership_cv);
507 	} else {
508 		ownership_tid = 0;
509 	}
510 
511 	mutex_exit(&vdc->lock);
512 
513 	if (failfast_tid != 0)
514 		thread_join(failfast_tid);
515 
516 	if (ownership_tid != 0)
517 		thread_join(ownership_tid);
518 
519 	if (vdc->initialized & VDC_MINOR) {
520 		ddi_prop_remove_all(dip);
521 		ddi_remove_minor_node(dip, NULL);
522 	}
523 
524 	if (vdc->io_stats) {
525 		kstat_delete(vdc->io_stats);
526 		vdc->io_stats = NULL;
527 	}
528 
529 	if (vdc->err_stats) {
530 		kstat_delete(vdc->err_stats);
531 		vdc->err_stats = NULL;
532 	}
533 
534 	if (vdc->initialized & VDC_LOCKS) {
535 		mutex_destroy(&vdc->lock);
536 		mutex_destroy(&vdc->read_lock);
537 		mutex_destroy(&vdc->ownership_lock);
538 		cv_destroy(&vdc->initwait_cv);
539 		cv_destroy(&vdc->dring_free_cv);
540 		cv_destroy(&vdc->membind_cv);
541 		cv_destroy(&vdc->sync_pending_cv);
542 		cv_destroy(&vdc->sync_blocked_cv);
543 		cv_destroy(&vdc->read_cv);
544 		cv_destroy(&vdc->running_cv);
545 		cv_destroy(&vdc->ownership_cv);
546 		cv_destroy(&vdc->failfast_cv);
547 		cv_destroy(&vdc->failfast_io_cv);
548 	}
549 
550 	if (vdc->minfo)
551 		kmem_free(vdc->minfo, sizeof (struct dk_minfo));
552 
553 	if (vdc->cinfo)
554 		kmem_free(vdc->cinfo, sizeof (struct dk_cinfo));
555 
556 	if (vdc->vtoc)
557 		kmem_free(vdc->vtoc, sizeof (struct vtoc));
558 
559 	if (vdc->geom)
560 		kmem_free(vdc->geom, sizeof (struct dk_geom));
561 
562 	if (vdc->devid) {
563 		ddi_devid_unregister(dip);
564 		ddi_devid_free(vdc->devid);
565 	}
566 
567 	if (vdc->initialized & VDC_SOFT_STATE)
568 		ddi_soft_state_free(vdc_state, instance);
569 
570 	DMSG(vdc, 0, "[%d] End %p\n", instance, (void *)vdc);
571 
572 	return (DDI_SUCCESS);
573 }
574 
575 
576 static int
577 vdc_do_attach(dev_info_t *dip)
578 {
579 	int		instance;
580 	vdc_t		*vdc = NULL;
581 	int		status;
582 	md_t		*mdp;
583 	mde_cookie_t	vd_node;
584 
585 	ASSERT(dip != NULL);
586 
587 	instance = ddi_get_instance(dip);
588 	if (ddi_soft_state_zalloc(vdc_state, instance) != DDI_SUCCESS) {
589 		cmn_err(CE_NOTE, "[%d] Couldn't alloc state structure",
590 		    instance);
591 		return (DDI_FAILURE);
592 	}
593 
594 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
595 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
596 		return (DDI_FAILURE);
597 	}
598 
599 	/*
600 	 * We assign the value to initialized in this case to zero out the
601 	 * variable and then set bits in it to indicate what has been done
602 	 */
603 	vdc->initialized = VDC_SOFT_STATE;
604 
605 	vdc_hz_min_ldc_delay = drv_usectohz(vdc_min_timeout_ldc);
606 	vdc_hz_max_ldc_delay = drv_usectohz(vdc_max_timeout_ldc);
607 
608 	vdc->dip	= dip;
609 	vdc->instance	= instance;
610 	vdc->vdisk_type	= VD_DISK_TYPE_UNK;
611 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
612 	vdc->state	= VDC_STATE_INIT;
613 	vdc->lifecycle	= VDC_LC_ATTACHING;
614 	vdc->session_id = 0;
615 	vdc->block_size = DEV_BSIZE;
616 	vdc->max_xfer_sz = maxphys / DEV_BSIZE;
617 
618 	/*
619 	 * We assume, for now, that the vDisk server will export 'read'
620 	 * operations to us at a minimum (this is needed because of checks
621 	 * in vdc for supported operations early in the handshake process).
622 	 * The vDisk server will return ENOTSUP if this is not the case.
623 	 * The value will be overwritten during the attribute exchange with
624 	 * the bitmask of operations exported by server.
625 	 */
626 	vdc->operations = VD_OP_MASK_READ;
627 
628 	vdc->vtoc = NULL;
629 	vdc->geom = NULL;
630 	vdc->cinfo = NULL;
631 	vdc->minfo = NULL;
632 
633 	mutex_init(&vdc->lock, NULL, MUTEX_DRIVER, NULL);
634 	cv_init(&vdc->initwait_cv, NULL, CV_DRIVER, NULL);
635 	cv_init(&vdc->dring_free_cv, NULL, CV_DRIVER, NULL);
636 	cv_init(&vdc->membind_cv, NULL, CV_DRIVER, NULL);
637 	cv_init(&vdc->running_cv, NULL, CV_DRIVER, NULL);
638 
639 	vdc->threads_pending = 0;
640 	vdc->sync_op_pending = B_FALSE;
641 	vdc->sync_op_blocked = B_FALSE;
642 	cv_init(&vdc->sync_pending_cv, NULL, CV_DRIVER, NULL);
643 	cv_init(&vdc->sync_blocked_cv, NULL, CV_DRIVER, NULL);
644 
645 	mutex_init(&vdc->ownership_lock, NULL, MUTEX_DRIVER, NULL);
646 	cv_init(&vdc->ownership_cv, NULL, CV_DRIVER, NULL);
647 	cv_init(&vdc->failfast_cv, NULL, CV_DRIVER, NULL);
648 	cv_init(&vdc->failfast_io_cv, NULL, CV_DRIVER, NULL);
649 
650 	/* init blocking msg read functionality */
651 	mutex_init(&vdc->read_lock, NULL, MUTEX_DRIVER, NULL);
652 	cv_init(&vdc->read_cv, NULL, CV_DRIVER, NULL);
653 	vdc->read_state = VDC_READ_IDLE;
654 
655 	vdc->initialized |= VDC_LOCKS;
656 
657 	/* get device and port MD node for this disk instance */
658 	if (vdc_get_md_node(dip, &mdp, &vd_node) != 0) {
659 		cmn_err(CE_NOTE, "[%d] Could not get machine description node",
660 		    instance);
661 		return (DDI_FAILURE);
662 	}
663 
664 	if (vdc_init_ports(vdc, mdp, vd_node) != 0) {
665 		cmn_err(CE_NOTE, "[%d] Error initialising ports", instance);
666 		return (DDI_FAILURE);
667 	}
668 
669 	(void) md_fini_handle(mdp);
670 
671 	/* initialize the thread responsible for managing state with server */
672 	vdc->msg_proc_thr = thread_create(NULL, 0, vdc_process_msg_thread,
673 	    vdc, 0, &p0, TS_RUN, minclsyspri);
674 	if (vdc->msg_proc_thr == NULL) {
675 		cmn_err(CE_NOTE, "[%d] Failed to create msg processing thread",
676 		    instance);
677 		return (DDI_FAILURE);
678 	}
679 
680 	vdc->initialized |= VDC_THREAD;
681 
682 	/* Create the kstats for saving the I/O statistics used by iostat(1M) */
683 	vdc_create_io_kstats(vdc);
684 	vdc_create_err_kstats(vdc);
685 
686 	atomic_inc_32(&vdc_instance_count);
687 
688 	/*
689 	 * Check the disk label. This will send requests and do the handshake.
690 	 * We don't really care about the disk label now. What we really need is
691 	 * the handshake do be done so that we know the type of the disk (slice
692 	 * or full disk) and the appropriate device nodes can be created.
693 	 */
694 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
695 	vdc->vtoc = kmem_zalloc(sizeof (struct vtoc), KM_SLEEP);
696 	vdc->geom = kmem_zalloc(sizeof (struct dk_geom), KM_SLEEP);
697 	vdc->minfo = kmem_zalloc(sizeof (struct dk_minfo), KM_SLEEP);
698 
699 	mutex_enter(&vdc->lock);
700 	(void) vdc_validate_geometry(vdc);
701 	mutex_exit(&vdc->lock);
702 
703 	/*
704 	 * Now that we have the device info we can create the
705 	 * device nodes and properties
706 	 */
707 	status = vdc_create_device_nodes(vdc);
708 	if (status) {
709 		DMSG(vdc, 0, "[%d] Failed to create device nodes",
710 		    instance);
711 		goto return_status;
712 	}
713 	status = vdc_create_device_nodes_props(vdc);
714 	if (status) {
715 		DMSG(vdc, 0, "[%d] Failed to create device nodes"
716 		    " properties (%d)", instance, status);
717 		goto return_status;
718 	}
719 
720 	/*
721 	 * Setup devid
722 	 */
723 	if (vdc_setup_devid(vdc)) {
724 		DMSG(vdc, 0, "[%d] No device id available\n", instance);
725 	}
726 
727 	/*
728 	 * Fill in the fields of the error statistics kstat that were not
729 	 * available when creating the kstat
730 	 */
731 	vdc_set_err_kstats(vdc);
732 
733 	ddi_report_dev(dip);
734 	vdc->lifecycle	= VDC_LC_ONLINE;
735 	DMSG(vdc, 0, "[%d] Attach tasks successful\n", instance);
736 
737 return_status:
738 	DMSG(vdc, 0, "[%d] Attach completed\n", instance);
739 	return (status);
740 }
741 
742 static int
743 vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
744 {
745 	int	status;
746 
747 	switch (cmd) {
748 	case DDI_ATTACH:
749 		if ((status = vdc_do_attach(dip)) != 0)
750 			(void) vdc_detach(dip, DDI_DETACH);
751 		return (status);
752 	case DDI_RESUME:
753 		/* nothing to do for this non-device */
754 		return (DDI_SUCCESS);
755 	default:
756 		return (DDI_FAILURE);
757 	}
758 }
759 
760 static int
761 vdc_do_ldc_init(vdc_t *vdc, vdc_server_t *srvr)
762 {
763 	int			status = 0;
764 	ldc_status_t		ldc_state;
765 	ldc_attr_t		ldc_attr;
766 
767 	ASSERT(vdc != NULL);
768 	ASSERT(srvr != NULL);
769 
770 	ldc_attr.devclass = LDC_DEV_BLK;
771 	ldc_attr.instance = vdc->instance;
772 	ldc_attr.mode = LDC_MODE_UNRELIABLE;	/* unreliable transport */
773 	ldc_attr.mtu = VD_LDC_MTU;
774 
775 	if ((srvr->state & VDC_LDC_INIT) == 0) {
776 		status = ldc_init(srvr->ldc_id, &ldc_attr,
777 		    &srvr->ldc_handle);
778 		if (status != 0) {
779 			DMSG(vdc, 0, "[%d] ldc_init(chan %ld) returned %d",
780 			    vdc->instance, srvr->ldc_id, status);
781 			return (status);
782 		}
783 		srvr->state |= VDC_LDC_INIT;
784 	}
785 	status = ldc_status(srvr->ldc_handle, &ldc_state);
786 	if (status != 0) {
787 		DMSG(vdc, 0, "[%d] Cannot discover LDC status [err=%d]",
788 		    vdc->instance, status);
789 		goto init_exit;
790 	}
791 	srvr->ldc_state = ldc_state;
792 
793 	if ((srvr->state & VDC_LDC_CB) == 0) {
794 		status = ldc_reg_callback(srvr->ldc_handle, vdc_handle_cb,
795 		    (caddr_t)srvr);
796 		if (status != 0) {
797 			DMSG(vdc, 0, "[%d] LDC callback reg. failed (%d)",
798 			    vdc->instance, status);
799 			goto init_exit;
800 		}
801 		srvr->state |= VDC_LDC_CB;
802 	}
803 
804 	/*
805 	 * At this stage we have initialised LDC, we will now try and open
806 	 * the connection.
807 	 */
808 	if (srvr->ldc_state == LDC_INIT) {
809 		status = ldc_open(srvr->ldc_handle);
810 		if (status != 0) {
811 			DMSG(vdc, 0, "[%d] ldc_open(chan %ld) returned %d",
812 			    vdc->instance, srvr->ldc_id, status);
813 			goto init_exit;
814 		}
815 		srvr->state |= VDC_LDC_OPEN;
816 	}
817 
818 init_exit:
819 	if (status) {
820 		vdc_terminate_ldc(vdc, srvr);
821 	}
822 
823 	return (status);
824 }
825 
826 static int
827 vdc_start_ldc_connection(vdc_t *vdc)
828 {
829 	int		status = 0;
830 
831 	ASSERT(vdc != NULL);
832 
833 	ASSERT(MUTEX_HELD(&vdc->lock));
834 
835 	status = vdc_do_ldc_up(vdc);
836 
837 	DMSG(vdc, 0, "[%d] Finished bringing up LDC\n", vdc->instance);
838 
839 	return (status);
840 }
841 
842 static int
843 vdc_stop_ldc_connection(vdc_t *vdcp)
844 {
845 	int	status;
846 
847 	ASSERT(vdcp != NULL);
848 
849 	ASSERT(MUTEX_HELD(&vdcp->lock));
850 
851 	DMSG(vdcp, 0, ": Resetting connection to vDisk server : state %d\n",
852 	    vdcp->state);
853 
854 	status = ldc_down(vdcp->curr_server->ldc_handle);
855 	DMSG(vdcp, 0, "ldc_down() = %d\n", status);
856 
857 	vdcp->initialized &= ~VDC_HANDSHAKE;
858 	DMSG(vdcp, 0, "initialized=%x\n", vdcp->initialized);
859 
860 	return (status);
861 }
862 
863 static void
864 vdc_create_io_kstats(vdc_t *vdc)
865 {
866 	if (vdc->io_stats != NULL) {
867 		DMSG(vdc, 0, "[%d] I/O kstat already exists\n", vdc->instance);
868 		return;
869 	}
870 
871 	vdc->io_stats = kstat_create(VDC_DRIVER_NAME, vdc->instance, NULL,
872 	    "disk", KSTAT_TYPE_IO, 1, KSTAT_FLAG_PERSISTENT);
873 	if (vdc->io_stats != NULL) {
874 		vdc->io_stats->ks_lock = &vdc->lock;
875 		kstat_install(vdc->io_stats);
876 	} else {
877 		cmn_err(CE_NOTE, "[%d] Failed to create kstat: I/O statistics"
878 		    " will not be gathered", vdc->instance);
879 	}
880 }
881 
882 static void
883 vdc_create_err_kstats(vdc_t *vdc)
884 {
885 	vd_err_stats_t	*stp;
886 	char	kstatmodule_err[KSTAT_STRLEN];
887 	char	kstatname[KSTAT_STRLEN];
888 	int	ndata = (sizeof (vd_err_stats_t) / sizeof (kstat_named_t));
889 	int	instance = vdc->instance;
890 
891 	if (vdc->err_stats != NULL) {
892 		DMSG(vdc, 0, "[%d] ERR kstat already exists\n", vdc->instance);
893 		return;
894 	}
895 
896 	(void) snprintf(kstatmodule_err, sizeof (kstatmodule_err),
897 	    "%serr", VDC_DRIVER_NAME);
898 	(void) snprintf(kstatname, sizeof (kstatname),
899 	    "%s%d,err", VDC_DRIVER_NAME, instance);
900 
901 	vdc->err_stats = kstat_create(kstatmodule_err, instance, kstatname,
902 	    "device_error", KSTAT_TYPE_NAMED, ndata, KSTAT_FLAG_PERSISTENT);
903 
904 	if (vdc->err_stats == NULL) {
905 		cmn_err(CE_NOTE, "[%d] Failed to create kstat: Error statistics"
906 		    " will not be gathered", instance);
907 		return;
908 	}
909 
910 	stp = (vd_err_stats_t *)vdc->err_stats->ks_data;
911 	kstat_named_init(&stp->vd_softerrs,	"Soft Errors",
912 	    KSTAT_DATA_UINT32);
913 	kstat_named_init(&stp->vd_transerrs,	"Transport Errors",
914 	    KSTAT_DATA_UINT32);
915 	kstat_named_init(&stp->vd_protoerrs,	"Protocol Errors",
916 	    KSTAT_DATA_UINT32);
917 	kstat_named_init(&stp->vd_vid,		"Vendor",
918 	    KSTAT_DATA_CHAR);
919 	kstat_named_init(&stp->vd_pid,		"Product",
920 	    KSTAT_DATA_CHAR);
921 	kstat_named_init(&stp->vd_capacity,	"Size",
922 	    KSTAT_DATA_ULONGLONG);
923 
924 	vdc->err_stats->ks_update  = nulldev;
925 
926 	kstat_install(vdc->err_stats);
927 }
928 
929 static void
930 vdc_set_err_kstats(vdc_t *vdc)
931 {
932 	vd_err_stats_t  *stp;
933 
934 	if (vdc->err_stats == NULL)
935 		return;
936 
937 	mutex_enter(&vdc->lock);
938 
939 	stp = (vd_err_stats_t *)vdc->err_stats->ks_data;
940 	ASSERT(stp != NULL);
941 
942 	stp->vd_capacity.value.ui64 = vdc->vdisk_size * vdc->block_size;
943 	(void) strcpy(stp->vd_vid.value.c, "SUN");
944 	(void) strcpy(stp->vd_pid.value.c, "VDSK");
945 
946 	mutex_exit(&vdc->lock);
947 }
948 
949 static int
950 vdc_create_device_nodes_efi(vdc_t *vdc)
951 {
952 	ddi_remove_minor_node(vdc->dip, "h");
953 	ddi_remove_minor_node(vdc->dip, "h,raw");
954 
955 	if (ddi_create_minor_node(vdc->dip, "wd", S_IFBLK,
956 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
957 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
958 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'wd'",
959 		    vdc->instance);
960 		return (EIO);
961 	}
962 
963 	/* if any device node is created we set this flag */
964 	vdc->initialized |= VDC_MINOR;
965 
966 	if (ddi_create_minor_node(vdc->dip, "wd,raw", S_IFCHR,
967 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
968 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
969 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'wd,raw'",
970 		    vdc->instance);
971 		return (EIO);
972 	}
973 
974 	return (0);
975 }
976 
977 static int
978 vdc_create_device_nodes_vtoc(vdc_t *vdc)
979 {
980 	ddi_remove_minor_node(vdc->dip, "wd");
981 	ddi_remove_minor_node(vdc->dip, "wd,raw");
982 
983 	if (ddi_create_minor_node(vdc->dip, "h", S_IFBLK,
984 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
985 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
986 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'h'",
987 		    vdc->instance);
988 		return (EIO);
989 	}
990 
991 	/* if any device node is created we set this flag */
992 	vdc->initialized |= VDC_MINOR;
993 
994 	if (ddi_create_minor_node(vdc->dip, "h,raw", S_IFCHR,
995 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
996 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
997 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'h,raw'",
998 		    vdc->instance);
999 		return (EIO);
1000 	}
1001 
1002 	return (0);
1003 }
1004 
1005 /*
1006  * Function:
1007  *	vdc_create_device_nodes
1008  *
1009  * Description:
1010  *	This function creates the block and character device nodes under
1011  *	/devices along with the node properties. It is called as part of
1012  *	the attach(9E) of the instance during the handshake with vds after
1013  *	vds has sent the attributes to vdc.
1014  *
1015  *	If the device is of type VD_DISK_TYPE_SLICE then the minor node
1016  *	of 2 is used in keeping with the Solaris convention that slice 2
1017  *	refers to a whole disk. Slices start at 'a'
1018  *
1019  * Parameters:
1020  *	vdc 		- soft state pointer
1021  *
1022  * Return Values
1023  *	0		- Success
1024  *	EIO		- Failed to create node
1025  *	EINVAL		- Unknown type of disk exported
1026  */
1027 static int
1028 vdc_create_device_nodes(vdc_t *vdc)
1029 {
1030 	char		name[sizeof ("s,raw")];
1031 	dev_info_t	*dip = NULL;
1032 	int		instance, status;
1033 	int		num_slices = 1;
1034 	int		i;
1035 
1036 	ASSERT(vdc != NULL);
1037 
1038 	instance = vdc->instance;
1039 	dip = vdc->dip;
1040 
1041 	switch (vdc->vdisk_type) {
1042 	case VD_DISK_TYPE_DISK:
1043 		num_slices = V_NUMPAR;
1044 		break;
1045 	case VD_DISK_TYPE_SLICE:
1046 		num_slices = 1;
1047 		break;
1048 	case VD_DISK_TYPE_UNK:
1049 	default:
1050 		return (EINVAL);
1051 	}
1052 
1053 	/*
1054 	 * Minor nodes are different for EFI disks: EFI disks do not have
1055 	 * a minor node 'g' for the minor number corresponding to slice
1056 	 * VD_EFI_WD_SLICE (slice 7) instead they have a minor node 'wd'
1057 	 * representing the whole disk.
1058 	 */
1059 	for (i = 0; i < num_slices; i++) {
1060 
1061 		if (i == VD_EFI_WD_SLICE) {
1062 			if (vdc->vdisk_label == VD_DISK_LABEL_EFI)
1063 				status = vdc_create_device_nodes_efi(vdc);
1064 			else
1065 				status = vdc_create_device_nodes_vtoc(vdc);
1066 			if (status != 0)
1067 				return (status);
1068 			continue;
1069 		}
1070 
1071 		(void) snprintf(name, sizeof (name), "%c", 'a' + i);
1072 		if (ddi_create_minor_node(dip, name, S_IFBLK,
1073 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
1074 			cmn_err(CE_NOTE, "[%d] Couldn't add block node '%s'",
1075 			    instance, name);
1076 			return (EIO);
1077 		}
1078 
1079 		/* if any device node is created we set this flag */
1080 		vdc->initialized |= VDC_MINOR;
1081 
1082 		(void) snprintf(name, sizeof (name), "%c%s", 'a' + i, ",raw");
1083 
1084 		if (ddi_create_minor_node(dip, name, S_IFCHR,
1085 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
1086 			cmn_err(CE_NOTE, "[%d] Couldn't add raw node '%s'",
1087 			    instance, name);
1088 			return (EIO);
1089 		}
1090 	}
1091 
1092 	return (0);
1093 }
1094 
1095 /*
1096  * Function:
1097  *	vdc_create_device_nodes_props
1098  *
1099  * Description:
1100  *	This function creates the block and character device nodes under
1101  *	/devices along with the node properties. It is called as part of
1102  *	the attach(9E) of the instance during the handshake with vds after
1103  *	vds has sent the attributes to vdc.
1104  *
1105  * Parameters:
1106  *	vdc 		- soft state pointer
1107  *
1108  * Return Values
1109  *	0		- Success
1110  *	EIO		- Failed to create device node property
1111  *	EINVAL		- Unknown type of disk exported
1112  */
1113 static int
1114 vdc_create_device_nodes_props(vdc_t *vdc)
1115 {
1116 	dev_info_t	*dip = NULL;
1117 	int		instance;
1118 	int		num_slices = 1;
1119 	int64_t		size = 0;
1120 	dev_t		dev;
1121 	int		rv;
1122 	int		i;
1123 
1124 	ASSERT(vdc != NULL);
1125 
1126 	instance = vdc->instance;
1127 	dip = vdc->dip;
1128 
1129 	switch (vdc->vdisk_type) {
1130 	case VD_DISK_TYPE_DISK:
1131 		num_slices = V_NUMPAR;
1132 		break;
1133 	case VD_DISK_TYPE_SLICE:
1134 		num_slices = 1;
1135 		break;
1136 	case VD_DISK_TYPE_UNK:
1137 	default:
1138 		return (EINVAL);
1139 	}
1140 
1141 	if (vdc->vdisk_label == VD_DISK_LABEL_UNK) {
1142 		/* remove all properties */
1143 		for (i = 0; i < num_slices; i++) {
1144 			dev = makedevice(ddi_driver_major(dip),
1145 			    VD_MAKE_DEV(instance, i));
1146 			(void) ddi_prop_remove(dev, dip, VDC_SIZE_PROP_NAME);
1147 			(void) ddi_prop_remove(dev, dip, VDC_NBLOCKS_PROP_NAME);
1148 		}
1149 		return (0);
1150 	}
1151 
1152 	for (i = 0; i < num_slices; i++) {
1153 		dev = makedevice(ddi_driver_major(dip),
1154 		    VD_MAKE_DEV(instance, i));
1155 
1156 		size = vdc->slice[i].nblocks * vdc->block_size;
1157 		DMSG(vdc, 0, "[%d] sz %ld (%ld Mb)  p_size %lx\n",
1158 		    instance, size, size / (1024 * 1024),
1159 		    vdc->slice[i].nblocks);
1160 
1161 		rv = ddi_prop_update_int64(dev, dip, VDC_SIZE_PROP_NAME, size);
1162 		if (rv != DDI_PROP_SUCCESS) {
1163 			cmn_err(CE_NOTE, "[%d] Couldn't add '%s' prop of [%ld]",
1164 			    instance, VDC_SIZE_PROP_NAME, size);
1165 			return (EIO);
1166 		}
1167 
1168 		rv = ddi_prop_update_int64(dev, dip, VDC_NBLOCKS_PROP_NAME,
1169 		    lbtodb(size));
1170 		if (rv != DDI_PROP_SUCCESS) {
1171 			cmn_err(CE_NOTE, "[%d] Couldn't add '%s' prop [%llu]",
1172 			    instance, VDC_NBLOCKS_PROP_NAME, lbtodb(size));
1173 			return (EIO);
1174 		}
1175 	}
1176 
1177 	return (0);
1178 }
1179 
1180 /*
1181  * Function:
1182  *	vdc_is_opened
1183  *
1184  * Description:
1185  *	This function checks if any slice of a given virtual disk is
1186  *	currently opened.
1187  *
1188  * Parameters:
1189  *	vdc 		- soft state pointer
1190  *
1191  * Return Values
1192  *	B_TRUE		- at least one slice is opened.
1193  *	B_FALSE		- no slice is opened.
1194  */
1195 static boolean_t
1196 vdc_is_opened(vdc_t *vdc)
1197 {
1198 	int i, nslices;
1199 
1200 	switch (vdc->vdisk_type) {
1201 	case VD_DISK_TYPE_DISK:
1202 		nslices = V_NUMPAR;
1203 		break;
1204 	case VD_DISK_TYPE_SLICE:
1205 		nslices = 1;
1206 		break;
1207 	case VD_DISK_TYPE_UNK:
1208 	default:
1209 		ASSERT(0);
1210 	}
1211 
1212 	/* check if there's any layered open */
1213 	for (i = 0; i < nslices; i++) {
1214 		if (vdc->open_lyr[i] > 0)
1215 			return (B_TRUE);
1216 	}
1217 
1218 	/* check if there is any other kind of open */
1219 	for (i = 0; i < OTYPCNT; i++) {
1220 		if (vdc->open[i] != 0)
1221 			return (B_TRUE);
1222 	}
1223 
1224 	return (B_FALSE);
1225 }
1226 
1227 static int
1228 vdc_mark_opened(vdc_t *vdc, int slice, int flag, int otyp)
1229 {
1230 	uint8_t slicemask;
1231 	int i;
1232 
1233 	ASSERT(otyp < OTYPCNT);
1234 	ASSERT(slice < V_NUMPAR);
1235 	ASSERT(MUTEX_HELD(&vdc->lock));
1236 
1237 	slicemask = 1 << slice;
1238 
1239 	/* check if slice is already exclusively opened */
1240 	if (vdc->open_excl & slicemask)
1241 		return (EBUSY);
1242 
1243 	/* if open exclusive, check if slice is already opened */
1244 	if (flag & FEXCL) {
1245 		if (vdc->open_lyr[slice] > 0)
1246 			return (EBUSY);
1247 		for (i = 0; i < OTYPCNT; i++) {
1248 			if (vdc->open[i] & slicemask)
1249 				return (EBUSY);
1250 		}
1251 		vdc->open_excl |= slicemask;
1252 	}
1253 
1254 	/* mark slice as opened */
1255 	if (otyp == OTYP_LYR) {
1256 		vdc->open_lyr[slice]++;
1257 	} else {
1258 		vdc->open[otyp] |= slicemask;
1259 	}
1260 
1261 	return (0);
1262 }
1263 
1264 static void
1265 vdc_mark_closed(vdc_t *vdc, int slice, int flag, int otyp)
1266 {
1267 	uint8_t slicemask;
1268 
1269 	ASSERT(otyp < OTYPCNT);
1270 	ASSERT(slice < V_NUMPAR);
1271 	ASSERT(MUTEX_HELD(&vdc->lock));
1272 
1273 	slicemask = 1 << slice;
1274 
1275 	if (otyp == OTYP_LYR) {
1276 		ASSERT(vdc->open_lyr[slice] > 0);
1277 		vdc->open_lyr[slice]--;
1278 	} else {
1279 		vdc->open[otyp] &= ~slicemask;
1280 	}
1281 
1282 	if (flag & FEXCL)
1283 		vdc->open_excl &= ~slicemask;
1284 }
1285 
1286 static int
1287 vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred)
1288 {
1289 	_NOTE(ARGUNUSED(cred))
1290 
1291 	int	instance, nodelay;
1292 	int	slice, status = 0;
1293 	vdc_t	*vdc;
1294 
1295 	ASSERT(dev != NULL);
1296 	instance = VDCUNIT(*dev);
1297 
1298 	if (otyp >= OTYPCNT)
1299 		return (EINVAL);
1300 
1301 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1302 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1303 		return (ENXIO);
1304 	}
1305 
1306 	DMSG(vdc, 0, "minor = %d flag = %x, otyp = %x\n",
1307 	    getminor(*dev), flag, otyp);
1308 
1309 	slice = VDCPART(*dev);
1310 
1311 	nodelay = flag & (FNDELAY | FNONBLOCK);
1312 
1313 	if ((flag & FWRITE) && (!nodelay) &&
1314 	    !(VD_OP_SUPPORTED(vdc->operations, VD_OP_BWRITE))) {
1315 		return (EROFS);
1316 	}
1317 
1318 	mutex_enter(&vdc->lock);
1319 
1320 	status = vdc_mark_opened(vdc, slice, flag, otyp);
1321 
1322 	if (status != 0) {
1323 		mutex_exit(&vdc->lock);
1324 		return (status);
1325 	}
1326 
1327 	if (nodelay) {
1328 
1329 		/* don't resubmit a validate request if there's already one */
1330 		if (vdc->validate_pending > 0) {
1331 			mutex_exit(&vdc->lock);
1332 			return (0);
1333 		}
1334 
1335 		/* call vdc_validate() asynchronously to avoid blocking */
1336 		if (taskq_dispatch(system_taskq, vdc_validate_task,
1337 		    (void *)vdc, TQ_NOSLEEP) == NULL) {
1338 			vdc_mark_closed(vdc, slice, flag, otyp);
1339 			mutex_exit(&vdc->lock);
1340 			return (ENXIO);
1341 		}
1342 
1343 		vdc->validate_pending++;
1344 		mutex_exit(&vdc->lock);
1345 		return (0);
1346 	}
1347 
1348 	mutex_exit(&vdc->lock);
1349 
1350 	vdc_validate(vdc);
1351 
1352 	mutex_enter(&vdc->lock);
1353 
1354 	if (vdc->vdisk_label == VD_DISK_LABEL_UNK ||
1355 	    vdc->slice[slice].nblocks == 0) {
1356 		vdc_mark_closed(vdc, slice, flag, otyp);
1357 		status = EIO;
1358 	}
1359 
1360 	mutex_exit(&vdc->lock);
1361 
1362 	return (status);
1363 }
1364 
1365 static int
1366 vdc_close(dev_t dev, int flag, int otyp, cred_t *cred)
1367 {
1368 	_NOTE(ARGUNUSED(cred))
1369 
1370 	int	instance;
1371 	int	slice;
1372 	int	rv, rval;
1373 	vdc_t	*vdc;
1374 
1375 	instance = VDCUNIT(dev);
1376 
1377 	if (otyp >= OTYPCNT)
1378 		return (EINVAL);
1379 
1380 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1381 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1382 		return (ENXIO);
1383 	}
1384 
1385 	DMSG(vdc, 0, "[%d] flag = %x, otyp = %x\n", instance, flag, otyp);
1386 
1387 	slice = VDCPART(dev);
1388 
1389 	/*
1390 	 * Attempt to flush the W$ on a close operation. If this is
1391 	 * not a supported IOCTL command or the backing device is read-only
1392 	 * do not fail the close operation.
1393 	 */
1394 	rv = vd_process_ioctl(dev, DKIOCFLUSHWRITECACHE, NULL, FKIOCTL, &rval);
1395 
1396 	if (rv != 0 && rv != ENOTSUP && rv != ENOTTY && rv != EROFS) {
1397 		DMSG(vdc, 0, "[%d] flush failed with error %d on close\n",
1398 		    instance, rv);
1399 		return (EIO);
1400 	}
1401 
1402 	mutex_enter(&vdc->lock);
1403 	vdc_mark_closed(vdc, slice, flag, otyp);
1404 	mutex_exit(&vdc->lock);
1405 
1406 	return (0);
1407 }
1408 
1409 static int
1410 vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
1411 {
1412 	_NOTE(ARGUNUSED(credp))
1413 
1414 	return (vd_process_ioctl(dev, cmd, (caddr_t)arg, mode, rvalp));
1415 }
1416 
1417 static int
1418 vdc_print(dev_t dev, char *str)
1419 {
1420 	cmn_err(CE_NOTE, "vdc%d:  %s", VDCUNIT(dev), str);
1421 	return (0);
1422 }
1423 
1424 static int
1425 vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk)
1426 {
1427 	int	rv;
1428 	size_t	nbytes = nblk * DEV_BSIZE;
1429 	int	instance = VDCUNIT(dev);
1430 	vdc_t	*vdc = NULL;
1431 
1432 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1433 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1434 		return (ENXIO);
1435 	}
1436 
1437 	DMSG(vdc, 2, "[%d] dump %ld bytes at block 0x%lx : addr=0x%p\n",
1438 	    instance, nbytes, blkno, (void *)addr);
1439 	rv = vdc_send_request(vdc, VD_OP_BWRITE, addr, nbytes,
1440 	    VDCPART(dev), blkno, CB_STRATEGY, 0, VIO_write_dir);
1441 	if (rv) {
1442 		DMSG(vdc, 0, "Failed to do a disk dump (err=%d)\n", rv);
1443 		return (rv);
1444 	}
1445 
1446 	if (ddi_in_panic())
1447 		(void) vdc_drain_response(vdc);
1448 
1449 	DMSG(vdc, 0, "[%d] End\n", instance);
1450 
1451 	return (0);
1452 }
1453 
1454 /* -------------------------------------------------------------------------- */
1455 
1456 /*
1457  * Disk access routines
1458  *
1459  */
1460 
1461 /*
1462  * vdc_strategy()
1463  *
1464  * Return Value:
1465  *	0:	As per strategy(9E), the strategy() function must return 0
1466  *		[ bioerror(9f) sets b_flags to the proper error code ]
1467  */
1468 static int
1469 vdc_strategy(struct buf *buf)
1470 {
1471 	int	rv = -1;
1472 	vdc_t	*vdc = NULL;
1473 	int	instance = VDCUNIT(buf->b_edev);
1474 	int	op = (buf->b_flags & B_READ) ? VD_OP_BREAD : VD_OP_BWRITE;
1475 	int	slice;
1476 
1477 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1478 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1479 		bioerror(buf, ENXIO);
1480 		biodone(buf);
1481 		return (0);
1482 	}
1483 
1484 	DMSG(vdc, 2, "[%d] %s %ld bytes at block %llx : b_addr=0x%p\n",
1485 	    instance, (buf->b_flags & B_READ) ? "Read" : "Write",
1486 	    buf->b_bcount, buf->b_lblkno, (void *)buf->b_un.b_addr);
1487 
1488 	bp_mapin(buf);
1489 
1490 	if ((long)buf->b_private == VD_SLICE_NONE) {
1491 		/* I/O using an absolute disk offset */
1492 		slice = VD_SLICE_NONE;
1493 	} else {
1494 		slice = VDCPART(buf->b_edev);
1495 	}
1496 
1497 	rv = vdc_send_request(vdc, op, (caddr_t)buf->b_un.b_addr,
1498 	    buf->b_bcount, slice, buf->b_lblkno,
1499 	    CB_STRATEGY, buf, (op == VD_OP_BREAD) ? VIO_read_dir :
1500 	    VIO_write_dir);
1501 
1502 	/*
1503 	 * If the request was successfully sent, the strategy call returns and
1504 	 * the ACK handler calls the bioxxx functions when the vDisk server is
1505 	 * done otherwise we handle the error here.
1506 	 */
1507 	if (rv) {
1508 		DMSG(vdc, 0, "Failed to read/write (err=%d)\n", rv);
1509 		bioerror(buf, rv);
1510 		biodone(buf);
1511 	}
1512 
1513 	return (0);
1514 }
1515 
1516 /*
1517  * Function:
1518  *	vdc_min
1519  *
1520  * Description:
1521  *	Routine to limit the size of a data transfer. Used in
1522  *	conjunction with physio(9F).
1523  *
1524  * Arguments:
1525  *	bp - pointer to the indicated buf(9S) struct.
1526  *
1527  */
1528 static void
1529 vdc_min(struct buf *bufp)
1530 {
1531 	vdc_t	*vdc = NULL;
1532 	int	instance = VDCUNIT(bufp->b_edev);
1533 
1534 	vdc = ddi_get_soft_state(vdc_state, instance);
1535 	VERIFY(vdc != NULL);
1536 
1537 	if (bufp->b_bcount > (vdc->max_xfer_sz * vdc->block_size)) {
1538 		bufp->b_bcount = vdc->max_xfer_sz * vdc->block_size;
1539 	}
1540 }
1541 
1542 static int
1543 vdc_read(dev_t dev, struct uio *uio, cred_t *cred)
1544 {
1545 	_NOTE(ARGUNUSED(cred))
1546 
1547 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1548 	return (physio(vdc_strategy, NULL, dev, B_READ, vdc_min, uio));
1549 }
1550 
1551 static int
1552 vdc_write(dev_t dev, struct uio *uio, cred_t *cred)
1553 {
1554 	_NOTE(ARGUNUSED(cred))
1555 
1556 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1557 	return (physio(vdc_strategy, NULL, dev, B_WRITE, vdc_min, uio));
1558 }
1559 
1560 static int
1561 vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred)
1562 {
1563 	_NOTE(ARGUNUSED(cred))
1564 
1565 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1566 	return (aphysio(vdc_strategy, anocancel, dev, B_READ, vdc_min, aio));
1567 }
1568 
1569 static int
1570 vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred)
1571 {
1572 	_NOTE(ARGUNUSED(cred))
1573 
1574 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1575 	return (aphysio(vdc_strategy, anocancel, dev, B_WRITE, vdc_min, aio));
1576 }
1577 
1578 
1579 /* -------------------------------------------------------------------------- */
1580 
1581 /*
1582  * Handshake support
1583  */
1584 
1585 
1586 /*
1587  * Function:
1588  *	vdc_init_ver_negotiation()
1589  *
1590  * Description:
1591  *
1592  * Arguments:
1593  *	vdc	- soft state pointer for this instance of the device driver.
1594  *
1595  * Return Code:
1596  *	0	- Success
1597  */
1598 static int
1599 vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver)
1600 {
1601 	vio_ver_msg_t	pkt;
1602 	size_t		msglen = sizeof (pkt);
1603 	int		status = -1;
1604 
1605 	ASSERT(vdc != NULL);
1606 	ASSERT(mutex_owned(&vdc->lock));
1607 
1608 	DMSG(vdc, 0, "[%d] Entered.\n", vdc->instance);
1609 
1610 	/*
1611 	 * set the Session ID to a unique value
1612 	 * (the lower 32 bits of the clock tick)
1613 	 */
1614 	vdc->session_id = ((uint32_t)gettick() & 0xffffffff);
1615 	DMSG(vdc, 0, "[%d] Set SID to 0x%lx\n", vdc->instance, vdc->session_id);
1616 
1617 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1618 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1619 	pkt.tag.vio_subtype_env = VIO_VER_INFO;
1620 	pkt.tag.vio_sid = vdc->session_id;
1621 	pkt.dev_class = VDEV_DISK;
1622 	pkt.ver_major = ver.major;
1623 	pkt.ver_minor = ver.minor;
1624 
1625 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1626 	DMSG(vdc, 0, "[%d] Ver info sent (status = %d)\n",
1627 	    vdc->instance, status);
1628 	if ((status != 0) || (msglen != sizeof (vio_ver_msg_t))) {
1629 		DMSG(vdc, 0, "[%d] Failed to send Ver negotiation info: "
1630 		    "id(%lx) rv(%d) size(%ld)", vdc->instance,
1631 		    vdc->curr_server->ldc_handle, status, msglen);
1632 		if (msglen != sizeof (vio_ver_msg_t))
1633 			status = ENOMSG;
1634 	}
1635 
1636 	return (status);
1637 }
1638 
1639 /*
1640  * Function:
1641  *	vdc_ver_negotiation()
1642  *
1643  * Description:
1644  *
1645  * Arguments:
1646  *	vdcp	- soft state pointer for this instance of the device driver.
1647  *
1648  * Return Code:
1649  *	0	- Success
1650  */
1651 static int
1652 vdc_ver_negotiation(vdc_t *vdcp)
1653 {
1654 	vio_msg_t vio_msg;
1655 	int status;
1656 
1657 	if (status = vdc_init_ver_negotiation(vdcp, vdc_version[0]))
1658 		return (status);
1659 
1660 	/* release lock and wait for response */
1661 	mutex_exit(&vdcp->lock);
1662 	status = vdc_wait_for_response(vdcp, &vio_msg);
1663 	mutex_enter(&vdcp->lock);
1664 	if (status) {
1665 		DMSG(vdcp, 0,
1666 		    "[%d] Failed waiting for Ver negotiation response, rv(%d)",
1667 		    vdcp->instance, status);
1668 		return (status);
1669 	}
1670 
1671 	/* check type and sub_type ... */
1672 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1673 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1674 		DMSG(vdcp, 0, "[%d] Invalid ver negotiation response\n",
1675 		    vdcp->instance);
1676 		return (EPROTO);
1677 	}
1678 
1679 	return (vdc_handle_ver_msg(vdcp, (vio_ver_msg_t *)&vio_msg));
1680 }
1681 
1682 /*
1683  * Function:
1684  *	vdc_init_attr_negotiation()
1685  *
1686  * Description:
1687  *
1688  * Arguments:
1689  *	vdc	- soft state pointer for this instance of the device driver.
1690  *
1691  * Return Code:
1692  *	0	- Success
1693  */
1694 static int
1695 vdc_init_attr_negotiation(vdc_t *vdc)
1696 {
1697 	vd_attr_msg_t	pkt;
1698 	size_t		msglen = sizeof (pkt);
1699 	int		status;
1700 
1701 	ASSERT(vdc != NULL);
1702 	ASSERT(mutex_owned(&vdc->lock));
1703 
1704 	DMSG(vdc, 0, "[%d] entered\n", vdc->instance);
1705 
1706 	/* fill in tag */
1707 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1708 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1709 	pkt.tag.vio_subtype_env = VIO_ATTR_INFO;
1710 	pkt.tag.vio_sid = vdc->session_id;
1711 	/* fill in payload */
1712 	pkt.max_xfer_sz = vdc->max_xfer_sz;
1713 	pkt.vdisk_block_size = vdc->block_size;
1714 	pkt.xfer_mode = VIO_DRING_MODE_V1_0;
1715 	pkt.operations = 0;	/* server will set bits of valid operations */
1716 	pkt.vdisk_type = 0;	/* server will set to valid device type */
1717 	pkt.vdisk_media = 0;	/* server will set to valid media type */
1718 	pkt.vdisk_size = 0;	/* server will set to valid size */
1719 
1720 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1721 	DMSG(vdc, 0, "Attr info sent (status = %d)\n", status);
1722 
1723 	if ((status != 0) || (msglen != sizeof (vio_ver_msg_t))) {
1724 		DMSG(vdc, 0, "[%d] Failed to send Attr negotiation info: "
1725 		    "id(%lx) rv(%d) size(%ld)", vdc->instance,
1726 		    vdc->curr_server->ldc_handle, status, msglen);
1727 		if (msglen != sizeof (vio_ver_msg_t))
1728 			status = ENOMSG;
1729 	}
1730 
1731 	return (status);
1732 }
1733 
1734 /*
1735  * Function:
1736  *	vdc_attr_negotiation()
1737  *
1738  * Description:
1739  *
1740  * Arguments:
1741  *	vdc	- soft state pointer for this instance of the device driver.
1742  *
1743  * Return Code:
1744  *	0	- Success
1745  */
1746 static int
1747 vdc_attr_negotiation(vdc_t *vdcp)
1748 {
1749 	int status;
1750 	vio_msg_t vio_msg;
1751 
1752 	if (status = vdc_init_attr_negotiation(vdcp))
1753 		return (status);
1754 
1755 	/* release lock and wait for response */
1756 	mutex_exit(&vdcp->lock);
1757 	status = vdc_wait_for_response(vdcp, &vio_msg);
1758 	mutex_enter(&vdcp->lock);
1759 	if (status) {
1760 		DMSG(vdcp, 0,
1761 		    "[%d] Failed waiting for Attr negotiation response, rv(%d)",
1762 		    vdcp->instance, status);
1763 		return (status);
1764 	}
1765 
1766 	/* check type and sub_type ... */
1767 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1768 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1769 		DMSG(vdcp, 0, "[%d] Invalid attr negotiation response\n",
1770 		    vdcp->instance);
1771 		return (EPROTO);
1772 	}
1773 
1774 	return (vdc_handle_attr_msg(vdcp, (vd_attr_msg_t *)&vio_msg));
1775 }
1776 
1777 
1778 /*
1779  * Function:
1780  *	vdc_init_dring_negotiate()
1781  *
1782  * Description:
1783  *
1784  * Arguments:
1785  *	vdc	- soft state pointer for this instance of the device driver.
1786  *
1787  * Return Code:
1788  *	0	- Success
1789  */
1790 static int
1791 vdc_init_dring_negotiate(vdc_t *vdc)
1792 {
1793 	vio_dring_reg_msg_t	pkt;
1794 	size_t			msglen = sizeof (pkt);
1795 	int			status = -1;
1796 	int			retry;
1797 	int			nretries = 10;
1798 
1799 	ASSERT(vdc != NULL);
1800 	ASSERT(mutex_owned(&vdc->lock));
1801 
1802 	for (retry = 0; retry < nretries; retry++) {
1803 		status = vdc_init_descriptor_ring(vdc);
1804 		if (status != EAGAIN)
1805 			break;
1806 		drv_usecwait(vdc_min_timeout_ldc);
1807 	}
1808 
1809 	if (status != 0) {
1810 		DMSG(vdc, 0, "[%d] Failed to init DRing (status = %d)\n",
1811 		    vdc->instance, status);
1812 		return (status);
1813 	}
1814 
1815 	DMSG(vdc, 0, "[%d] Init of descriptor ring completed (status = %d)\n",
1816 	    vdc->instance, status);
1817 
1818 	/* fill in tag */
1819 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1820 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1821 	pkt.tag.vio_subtype_env = VIO_DRING_REG;
1822 	pkt.tag.vio_sid = vdc->session_id;
1823 	/* fill in payload */
1824 	pkt.dring_ident = 0;
1825 	pkt.num_descriptors = vdc->dring_len;
1826 	pkt.descriptor_size = vdc->dring_entry_size;
1827 	pkt.options = (VIO_TX_DRING | VIO_RX_DRING);
1828 	pkt.ncookies = vdc->dring_cookie_count;
1829 	pkt.cookie[0] = vdc->dring_cookie[0];	/* for now just one cookie */
1830 
1831 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1832 	if (status != 0) {
1833 		DMSG(vdc, 0, "[%d] Failed to register DRing (err = %d)",
1834 		    vdc->instance, status);
1835 	}
1836 
1837 	return (status);
1838 }
1839 
1840 
1841 /*
1842  * Function:
1843  *	vdc_dring_negotiation()
1844  *
1845  * Description:
1846  *
1847  * Arguments:
1848  *	vdc	- soft state pointer for this instance of the device driver.
1849  *
1850  * Return Code:
1851  *	0	- Success
1852  */
1853 static int
1854 vdc_dring_negotiation(vdc_t *vdcp)
1855 {
1856 	int status;
1857 	vio_msg_t vio_msg;
1858 
1859 	if (status = vdc_init_dring_negotiate(vdcp))
1860 		return (status);
1861 
1862 	/* release lock and wait for response */
1863 	mutex_exit(&vdcp->lock);
1864 	status = vdc_wait_for_response(vdcp, &vio_msg);
1865 	mutex_enter(&vdcp->lock);
1866 	if (status) {
1867 		DMSG(vdcp, 0,
1868 		    "[%d] Failed waiting for Dring negotiation response,"
1869 		    " rv(%d)", vdcp->instance, status);
1870 		return (status);
1871 	}
1872 
1873 	/* check type and sub_type ... */
1874 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1875 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1876 		DMSG(vdcp, 0, "[%d] Invalid Dring negotiation response\n",
1877 		    vdcp->instance);
1878 		return (EPROTO);
1879 	}
1880 
1881 	return (vdc_handle_dring_reg_msg(vdcp,
1882 	    (vio_dring_reg_msg_t *)&vio_msg));
1883 }
1884 
1885 
1886 /*
1887  * Function:
1888  *	vdc_send_rdx()
1889  *
1890  * Description:
1891  *
1892  * Arguments:
1893  *	vdc	- soft state pointer for this instance of the device driver.
1894  *
1895  * Return Code:
1896  *	0	- Success
1897  */
1898 static int
1899 vdc_send_rdx(vdc_t *vdcp)
1900 {
1901 	vio_msg_t	msg;
1902 	size_t		msglen = sizeof (vio_msg_t);
1903 	int		status;
1904 
1905 	/*
1906 	 * Send an RDX message to vds to indicate we are ready
1907 	 * to send data
1908 	 */
1909 	msg.tag.vio_msgtype = VIO_TYPE_CTRL;
1910 	msg.tag.vio_subtype = VIO_SUBTYPE_INFO;
1911 	msg.tag.vio_subtype_env = VIO_RDX;
1912 	msg.tag.vio_sid = vdcp->session_id;
1913 	status = vdc_send(vdcp, (caddr_t)&msg, &msglen);
1914 	if (status != 0) {
1915 		DMSG(vdcp, 0, "[%d] Failed to send RDX message (%d)",
1916 		    vdcp->instance, status);
1917 	}
1918 
1919 	return (status);
1920 }
1921 
1922 /*
1923  * Function:
1924  *	vdc_handle_rdx()
1925  *
1926  * Description:
1927  *
1928  * Arguments:
1929  *	vdc	- soft state pointer for this instance of the device driver.
1930  *	msgp	- received msg
1931  *
1932  * Return Code:
1933  *	0	- Success
1934  */
1935 static int
1936 vdc_handle_rdx(vdc_t *vdcp, vio_rdx_msg_t *msgp)
1937 {
1938 	_NOTE(ARGUNUSED(vdcp))
1939 	_NOTE(ARGUNUSED(msgp))
1940 
1941 	ASSERT(msgp->tag.vio_msgtype == VIO_TYPE_CTRL);
1942 	ASSERT(msgp->tag.vio_subtype == VIO_SUBTYPE_ACK);
1943 	ASSERT(msgp->tag.vio_subtype_env == VIO_RDX);
1944 
1945 	DMSG(vdcp, 1, "[%d] Got an RDX msg", vdcp->instance);
1946 
1947 	return (0);
1948 }
1949 
1950 /*
1951  * Function:
1952  *	vdc_rdx_exchange()
1953  *
1954  * Description:
1955  *
1956  * Arguments:
1957  *	vdc	- soft state pointer for this instance of the device driver.
1958  *
1959  * Return Code:
1960  *	0	- Success
1961  */
1962 static int
1963 vdc_rdx_exchange(vdc_t *vdcp)
1964 {
1965 	int status;
1966 	vio_msg_t vio_msg;
1967 
1968 	if (status = vdc_send_rdx(vdcp))
1969 		return (status);
1970 
1971 	/* release lock and wait for response */
1972 	mutex_exit(&vdcp->lock);
1973 	status = vdc_wait_for_response(vdcp, &vio_msg);
1974 	mutex_enter(&vdcp->lock);
1975 	if (status) {
1976 		DMSG(vdcp, 0, "[%d] Failed waiting for RDX response, rv(%d)",
1977 		    vdcp->instance, status);
1978 		return (status);
1979 	}
1980 
1981 	/* check type and sub_type ... */
1982 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1983 	    vio_msg.tag.vio_subtype != VIO_SUBTYPE_ACK) {
1984 		DMSG(vdcp, 0, "[%d] Invalid RDX response\n", vdcp->instance);
1985 		return (EPROTO);
1986 	}
1987 
1988 	return (vdc_handle_rdx(vdcp, (vio_rdx_msg_t *)&vio_msg));
1989 }
1990 
1991 
1992 /* -------------------------------------------------------------------------- */
1993 
1994 /*
1995  * LDC helper routines
1996  */
1997 
1998 static int
1999 vdc_recv(vdc_t *vdc, vio_msg_t *msgp, size_t *nbytesp)
2000 {
2001 	int		status;
2002 	boolean_t	q_has_pkts = B_FALSE;
2003 	uint64_t	delay_time;
2004 	size_t		len;
2005 
2006 	mutex_enter(&vdc->read_lock);
2007 
2008 	if (vdc->read_state == VDC_READ_IDLE)
2009 		vdc->read_state = VDC_READ_WAITING;
2010 
2011 	while (vdc->read_state != VDC_READ_PENDING) {
2012 
2013 		/* detect if the connection has been reset */
2014 		if (vdc->read_state == VDC_READ_RESET) {
2015 			status = ECONNRESET;
2016 			goto done;
2017 		}
2018 
2019 		cv_wait(&vdc->read_cv, &vdc->read_lock);
2020 	}
2021 
2022 	/*
2023 	 * Until we get a blocking ldc read we have to retry
2024 	 * until the entire LDC message has arrived before
2025 	 * ldc_read() will succeed. Note we also bail out if
2026 	 * the channel is reset or goes away.
2027 	 */
2028 	delay_time = vdc_ldc_read_init_delay;
2029 loop:
2030 	len = *nbytesp;
2031 	status = ldc_read(vdc->curr_server->ldc_handle, (caddr_t)msgp, &len);
2032 	switch (status) {
2033 	case EAGAIN:
2034 		delay_time *= 2;
2035 		if (delay_time >= vdc_ldc_read_max_delay)
2036 			delay_time = vdc_ldc_read_max_delay;
2037 		delay(delay_time);
2038 		goto loop;
2039 
2040 	case 0:
2041 		if (len == 0) {
2042 			DMSG(vdc, 1, "[%d] ldc_read returned 0 bytes with "
2043 			    "no error!\n", vdc->instance);
2044 			goto loop;
2045 		}
2046 
2047 		*nbytesp = len;
2048 
2049 		/*
2050 		 * If there are pending messages, leave the
2051 		 * read state as pending. Otherwise, set the state
2052 		 * back to idle.
2053 		 */
2054 		status = ldc_chkq(vdc->curr_server->ldc_handle, &q_has_pkts);
2055 		if (status == 0 && !q_has_pkts)
2056 			vdc->read_state = VDC_READ_IDLE;
2057 
2058 		break;
2059 	default:
2060 		DMSG(vdc, 0, "ldc_read returned %d\n", status);
2061 		break;
2062 	}
2063 
2064 done:
2065 	mutex_exit(&vdc->read_lock);
2066 
2067 	return (status);
2068 }
2069 
2070 
2071 
2072 #ifdef DEBUG
2073 void
2074 vdc_decode_tag(vdc_t *vdcp, vio_msg_t *msg)
2075 {
2076 	char *ms, *ss, *ses;
2077 	switch (msg->tag.vio_msgtype) {
2078 #define	Q(_s)	case _s : ms = #_s; break;
2079 	Q(VIO_TYPE_CTRL)
2080 	Q(VIO_TYPE_DATA)
2081 	Q(VIO_TYPE_ERR)
2082 #undef Q
2083 	default: ms = "unknown"; break;
2084 	}
2085 
2086 	switch (msg->tag.vio_subtype) {
2087 #define	Q(_s)	case _s : ss = #_s; break;
2088 	Q(VIO_SUBTYPE_INFO)
2089 	Q(VIO_SUBTYPE_ACK)
2090 	Q(VIO_SUBTYPE_NACK)
2091 #undef Q
2092 	default: ss = "unknown"; break;
2093 	}
2094 
2095 	switch (msg->tag.vio_subtype_env) {
2096 #define	Q(_s)	case _s : ses = #_s; break;
2097 	Q(VIO_VER_INFO)
2098 	Q(VIO_ATTR_INFO)
2099 	Q(VIO_DRING_REG)
2100 	Q(VIO_DRING_UNREG)
2101 	Q(VIO_RDX)
2102 	Q(VIO_PKT_DATA)
2103 	Q(VIO_DESC_DATA)
2104 	Q(VIO_DRING_DATA)
2105 #undef Q
2106 	default: ses = "unknown"; break;
2107 	}
2108 
2109 	DMSG(vdcp, 3, "(%x/%x/%x) message : (%s/%s/%s)\n",
2110 	    msg->tag.vio_msgtype, msg->tag.vio_subtype,
2111 	    msg->tag.vio_subtype_env, ms, ss, ses);
2112 }
2113 #endif
2114 
2115 /*
2116  * Function:
2117  *	vdc_send()
2118  *
2119  * Description:
2120  *	The function encapsulates the call to write a message using LDC.
2121  *	If LDC indicates that the call failed due to the queue being full,
2122  *	we retry the ldc_write(), otherwise we return the error returned by LDC.
2123  *
2124  * Arguments:
2125  *	ldc_handle	- LDC handle for the channel this instance of vdc uses
2126  *	pkt		- address of LDC message to be sent
2127  *	msglen		- the size of the message being sent. When the function
2128  *			  returns, this contains the number of bytes written.
2129  *
2130  * Return Code:
2131  *	0		- Success.
2132  *	EINVAL		- pkt or msglen were NULL
2133  *	ECONNRESET	- The connection was not up.
2134  *	EWOULDBLOCK	- LDC queue is full
2135  *	xxx		- other error codes returned by ldc_write
2136  */
2137 static int
2138 vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen)
2139 {
2140 	size_t	size = 0;
2141 	int	status = 0;
2142 	clock_t delay_ticks;
2143 
2144 	ASSERT(vdc != NULL);
2145 	ASSERT(mutex_owned(&vdc->lock));
2146 	ASSERT(msglen != NULL);
2147 	ASSERT(*msglen != 0);
2148 
2149 #ifdef DEBUG
2150 	vdc_decode_tag(vdc, (vio_msg_t *)(uintptr_t)pkt);
2151 #endif
2152 	/*
2153 	 * Wait indefinitely to send if channel
2154 	 * is busy, but bail out if we succeed or
2155 	 * if the channel closes or is reset.
2156 	 */
2157 	delay_ticks = vdc_hz_min_ldc_delay;
2158 	do {
2159 		size = *msglen;
2160 		status = ldc_write(vdc->curr_server->ldc_handle, pkt, &size);
2161 		if (status == EWOULDBLOCK) {
2162 			delay(delay_ticks);
2163 			/* geometric backoff */
2164 			delay_ticks *= 2;
2165 			if (delay_ticks > vdc_hz_max_ldc_delay)
2166 				delay_ticks = vdc_hz_max_ldc_delay;
2167 		}
2168 	} while (status == EWOULDBLOCK);
2169 
2170 	/* if LDC had serious issues --- reset vdc state */
2171 	if (status == EIO || status == ECONNRESET) {
2172 		/* LDC had serious issues --- reset vdc state */
2173 		mutex_enter(&vdc->read_lock);
2174 		if ((vdc->read_state == VDC_READ_WAITING) ||
2175 		    (vdc->read_state == VDC_READ_RESET))
2176 			cv_signal(&vdc->read_cv);
2177 		vdc->read_state = VDC_READ_RESET;
2178 		mutex_exit(&vdc->read_lock);
2179 
2180 		/* wake up any waiters in the reset thread */
2181 		if (vdc->state == VDC_STATE_INIT_WAITING) {
2182 			DMSG(vdc, 0, "[%d] write reset - "
2183 			    "vdc is resetting ..\n", vdc->instance);
2184 			vdc->state = VDC_STATE_RESETTING;
2185 			cv_signal(&vdc->initwait_cv);
2186 		}
2187 
2188 		return (ECONNRESET);
2189 	}
2190 
2191 	/* return the last size written */
2192 	*msglen = size;
2193 
2194 	return (status);
2195 }
2196 
2197 /*
2198  * Function:
2199  *	vdc_get_md_node
2200  *
2201  * Description:
2202  *	Get the MD, the device node for the given disk instance. The
2203  *	caller is responsible for cleaning up the reference to the
2204  *	returned MD (mdpp) by calling md_fini_handle().
2205  *
2206  * Arguments:
2207  *	dip	- dev info pointer for this instance of the device driver.
2208  *	mdpp	- the returned MD.
2209  *	vd_nodep - the returned device node.
2210  *
2211  * Return Code:
2212  *	0	- Success.
2213  *	ENOENT	- Expected node or property did not exist.
2214  *	ENXIO	- Unexpected error communicating with MD framework
2215  */
2216 static int
2217 vdc_get_md_node(dev_info_t *dip, md_t **mdpp, mde_cookie_t *vd_nodep)
2218 {
2219 	int		status = ENOENT;
2220 	char		*node_name = NULL;
2221 	md_t		*mdp = NULL;
2222 	int		num_nodes;
2223 	int		num_vdevs;
2224 	mde_cookie_t	rootnode;
2225 	mde_cookie_t	*listp = NULL;
2226 	boolean_t	found_inst = B_FALSE;
2227 	int		listsz;
2228 	int		idx;
2229 	uint64_t	md_inst;
2230 	int		obp_inst;
2231 	int		instance = ddi_get_instance(dip);
2232 
2233 	/*
2234 	 * Get the OBP instance number for comparison with the MD instance
2235 	 *
2236 	 * The "cfg-handle" property of a vdc node in an MD contains the MD's
2237 	 * notion of "instance", or unique identifier, for that node; OBP
2238 	 * stores the value of the "cfg-handle" MD property as the value of
2239 	 * the "reg" property on the node in the device tree it builds from
2240 	 * the MD and passes to Solaris.  Thus, we look up the devinfo node's
2241 	 * "reg" property value to uniquely identify this device instance.
2242 	 * If the "reg" property cannot be found, the device tree state is
2243 	 * presumably so broken that there is no point in continuing.
2244 	 */
2245 	if (!ddi_prop_exists(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS, OBP_REG)) {
2246 		cmn_err(CE_WARN, "'%s' property does not exist", OBP_REG);
2247 		return (ENOENT);
2248 	}
2249 	obp_inst = ddi_prop_get_int(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
2250 	    OBP_REG, -1);
2251 	DMSGX(1, "[%d] OBP inst=%d\n", instance, obp_inst);
2252 
2253 	/*
2254 	 * We now walk the MD nodes to find the node for this vdisk.
2255 	 */
2256 	if ((mdp = md_get_handle()) == NULL) {
2257 		cmn_err(CE_WARN, "unable to init machine description");
2258 		return (ENXIO);
2259 	}
2260 
2261 	num_nodes = md_node_count(mdp);
2262 	ASSERT(num_nodes > 0);
2263 
2264 	listsz = num_nodes * sizeof (mde_cookie_t);
2265 
2266 	/* allocate memory for nodes */
2267 	listp = kmem_zalloc(listsz, KM_SLEEP);
2268 
2269 	rootnode = md_root_node(mdp);
2270 	ASSERT(rootnode != MDE_INVAL_ELEM_COOKIE);
2271 
2272 	/*
2273 	 * Search for all the virtual devices, we will then check to see which
2274 	 * ones are disk nodes.
2275 	 */
2276 	num_vdevs = md_scan_dag(mdp, rootnode,
2277 	    md_find_name(mdp, VDC_MD_VDEV_NAME),
2278 	    md_find_name(mdp, "fwd"), listp);
2279 
2280 	if (num_vdevs <= 0) {
2281 		cmn_err(CE_NOTE, "No '%s' node found", VDC_MD_VDEV_NAME);
2282 		status = ENOENT;
2283 		goto done;
2284 	}
2285 
2286 	DMSGX(1, "[%d] num_vdevs=%d\n", instance, num_vdevs);
2287 	for (idx = 0; idx < num_vdevs; idx++) {
2288 		status = md_get_prop_str(mdp, listp[idx], "name", &node_name);
2289 		if ((status != 0) || (node_name == NULL)) {
2290 			cmn_err(CE_NOTE, "Unable to get name of node type '%s'"
2291 			    ": err %d", VDC_MD_VDEV_NAME, status);
2292 			continue;
2293 		}
2294 
2295 		DMSGX(1, "[%d] Found node '%s'\n", instance, node_name);
2296 		if (strcmp(VDC_MD_DISK_NAME, node_name) == 0) {
2297 			status = md_get_prop_val(mdp, listp[idx],
2298 			    VDC_MD_CFG_HDL, &md_inst);
2299 			DMSGX(1, "[%d] vdc inst in MD=%lx\n",
2300 			    instance, md_inst);
2301 			if ((status == 0) && (md_inst == obp_inst)) {
2302 				found_inst = B_TRUE;
2303 				break;
2304 			}
2305 		}
2306 	}
2307 
2308 	if (!found_inst) {
2309 		DMSGX(0, "Unable to find correct '%s' node", VDC_MD_DISK_NAME);
2310 		status = ENOENT;
2311 		goto done;
2312 	}
2313 	DMSGX(0, "[%d] MD inst=%lx\n", instance, md_inst);
2314 
2315 	*vd_nodep = listp[idx];
2316 	*mdpp = mdp;
2317 done:
2318 	kmem_free(listp, listsz);
2319 	return (status);
2320 }
2321 
2322 /*
2323  * Function:
2324  *	vdc_init_ports
2325  *
2326  * Description:
2327  *	Initialize all the ports for this vdisk instance.
2328  *
2329  * Arguments:
2330  *	vdc	- soft state pointer for this instance of the device driver.
2331  *	mdp	- md pointer
2332  *	vd_nodep - device md node.
2333  *
2334  * Return Code:
2335  *	0	- Success.
2336  *	ENOENT	- Expected node or property did not exist.
2337  */
2338 static int
2339 vdc_init_ports(vdc_t *vdc, md_t *mdp, mde_cookie_t vd_nodep)
2340 {
2341 	int		status = 0;
2342 	int		idx;
2343 	int		num_nodes;
2344 	int		num_vports;
2345 	int		num_chans;
2346 	int		listsz;
2347 	mde_cookie_t	vd_port;
2348 	mde_cookie_t	*chanp = NULL;
2349 	mde_cookie_t	*portp = NULL;
2350 	vdc_server_t	*srvr;
2351 	vdc_server_t	*prev_srvr = NULL;
2352 
2353 	/*
2354 	 * We now walk the MD nodes to find the port nodes for this vdisk.
2355 	 */
2356 	num_nodes = md_node_count(mdp);
2357 	ASSERT(num_nodes > 0);
2358 
2359 	listsz = num_nodes * sizeof (mde_cookie_t);
2360 
2361 	/* allocate memory for nodes */
2362 	portp = kmem_zalloc(listsz, KM_SLEEP);
2363 	chanp = kmem_zalloc(listsz, KM_SLEEP);
2364 
2365 	num_vports = md_scan_dag(mdp, vd_nodep,
2366 	    md_find_name(mdp, VDC_MD_PORT_NAME),
2367 	    md_find_name(mdp, "fwd"), portp);
2368 	if (num_vports == 0) {
2369 		DMSGX(0, "Found no '%s' node for '%s' port\n",
2370 		    VDC_MD_PORT_NAME, VDC_MD_VDEV_NAME);
2371 		status = ENOENT;
2372 		goto done;
2373 	}
2374 
2375 	DMSGX(1, "Found %d '%s' node(s) for '%s' port\n",
2376 	    num_vports, VDC_MD_PORT_NAME, VDC_MD_VDEV_NAME);
2377 
2378 	vdc->num_servers = 0;
2379 	for (idx = 0; idx < num_vports; idx++) {
2380 
2381 		/* initialize this port */
2382 		vd_port = portp[idx];
2383 		srvr = kmem_zalloc(sizeof (vdc_server_t), KM_SLEEP);
2384 		srvr->vdcp = vdc;
2385 
2386 		/* get port id */
2387 		if (md_get_prop_val(mdp, vd_port, VDC_MD_ID, &srvr->id) != 0) {
2388 			cmn_err(CE_NOTE, "vDisk port '%s' property not found",
2389 			    VDC_MD_ID);
2390 			kmem_free(srvr, sizeof (vdc_server_t));
2391 			continue;
2392 		}
2393 
2394 		/* set the connection timeout */
2395 		if (md_get_prop_val(mdp, vd_port, VDC_MD_TIMEOUT,
2396 		    &srvr->ctimeout) != 0) {
2397 			srvr->ctimeout = 0;
2398 		}
2399 
2400 		/* get the ldc id */
2401 		num_chans = md_scan_dag(mdp, vd_port,
2402 		    md_find_name(mdp, VDC_MD_CHAN_NAME),
2403 		    md_find_name(mdp, "fwd"), chanp);
2404 
2405 		/* expecting at least one channel */
2406 		if (num_chans <= 0) {
2407 			cmn_err(CE_NOTE, "No '%s' node for '%s' port",
2408 			    VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME);
2409 			kmem_free(srvr, sizeof (vdc_server_t));
2410 			continue;
2411 		} else if (num_chans != 1) {
2412 			DMSGX(0, "Expected 1 '%s' node for '%s' port, "
2413 			    "found %d\n", VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME,
2414 			    num_chans);
2415 		}
2416 
2417 		/*
2418 		 * We use the first channel found (index 0), irrespective of how
2419 		 * many are there in total.
2420 		 */
2421 		if (md_get_prop_val(mdp, chanp[0], VDC_MD_ID,
2422 		    &srvr->ldc_id) != 0) {
2423 			cmn_err(CE_NOTE, "Channel '%s' property not found",
2424 			    VDC_MD_ID);
2425 			kmem_free(srvr, sizeof (vdc_server_t));
2426 			continue;
2427 		}
2428 
2429 		/*
2430 		 * now initialise LDC channel which will be used to
2431 		 * communicate with this server
2432 		 */
2433 		if (vdc_do_ldc_init(vdc, srvr) != 0) {
2434 			kmem_free(srvr, sizeof (vdc_server_t));
2435 			continue;
2436 		}
2437 
2438 		/* add server to list */
2439 		if (prev_srvr) {
2440 			prev_srvr->next = srvr;
2441 		} else {
2442 			vdc->server_list = srvr;
2443 			prev_srvr = srvr;
2444 		}
2445 
2446 		/* inc numbers of servers */
2447 		vdc->num_servers++;
2448 	}
2449 
2450 	/*
2451 	 * Adjust the max number of handshake retries to match
2452 	 * the number of vdisk servers.
2453 	 */
2454 	if (vdc_hshake_retries < vdc->num_servers)
2455 		vdc_hshake_retries = vdc->num_servers;
2456 
2457 	/* pick first server as current server */
2458 	if (vdc->server_list != NULL) {
2459 		vdc->curr_server = vdc->server_list;
2460 		status = 0;
2461 	} else {
2462 		status = ENOENT;
2463 	}
2464 
2465 done:
2466 	kmem_free(chanp, listsz);
2467 	kmem_free(portp, listsz);
2468 	return (status);
2469 }
2470 
2471 
2472 /*
2473  * Function:
2474  *	vdc_do_ldc_up
2475  *
2476  * Description:
2477  *	Bring the channel for the current server up.
2478  *
2479  * Arguments:
2480  *	vdc	- soft state pointer for this instance of the device driver.
2481  *
2482  * Return Code:
2483  *	0		- Success.
2484  *	EINVAL		- Driver is detaching / LDC error
2485  *	ECONNREFUSED	- Other end is not listening
2486  */
2487 static int
2488 vdc_do_ldc_up(vdc_t *vdc)
2489 {
2490 	int		status;
2491 	ldc_status_t	ldc_state;
2492 
2493 	ASSERT(MUTEX_HELD(&vdc->lock));
2494 
2495 	DMSG(vdc, 0, "[%d] Bringing up channel %lx\n",
2496 	    vdc->instance, vdc->curr_server->ldc_id);
2497 
2498 	if (vdc->lifecycle == VDC_LC_DETACHING)
2499 		return (EINVAL);
2500 
2501 	if ((status = ldc_up(vdc->curr_server->ldc_handle)) != 0) {
2502 		switch (status) {
2503 		case ECONNREFUSED:	/* listener not ready at other end */
2504 			DMSG(vdc, 0, "[%d] ldc_up(%lx,...) return %d\n",
2505 			    vdc->instance, vdc->curr_server->ldc_id, status);
2506 			status = 0;
2507 			break;
2508 		default:
2509 			DMSG(vdc, 0, "[%d] Failed to bring up LDC: "
2510 			    "channel=%ld, err=%d", vdc->instance,
2511 			    vdc->curr_server->ldc_id, status);
2512 			break;
2513 		}
2514 	}
2515 
2516 	if (ldc_status(vdc->curr_server->ldc_handle, &ldc_state) == 0) {
2517 		vdc->curr_server->ldc_state = ldc_state;
2518 		if (ldc_state == LDC_UP) {
2519 			DMSG(vdc, 0, "[%d] LDC channel already up\n",
2520 			    vdc->instance);
2521 			vdc->seq_num = 1;
2522 			vdc->seq_num_reply = 0;
2523 		}
2524 	}
2525 
2526 	return (status);
2527 }
2528 
2529 /*
2530  * Function:
2531  *	vdc_terminate_ldc()
2532  *
2533  * Description:
2534  *
2535  * Arguments:
2536  *	vdc	- soft state pointer for this instance of the device driver.
2537  *	srvr	- vdc per-server info structure
2538  *
2539  * Return Code:
2540  *	None
2541  */
2542 static void
2543 vdc_terminate_ldc(vdc_t *vdc, vdc_server_t *srvr)
2544 {
2545 	int	instance = ddi_get_instance(vdc->dip);
2546 
2547 	if (srvr->state & VDC_LDC_OPEN) {
2548 		DMSG(vdc, 0, "[%d] ldc_close()\n", instance);
2549 		(void) ldc_close(srvr->ldc_handle);
2550 	}
2551 	if (srvr->state & VDC_LDC_CB) {
2552 		DMSG(vdc, 0, "[%d] ldc_unreg_callback()\n", instance);
2553 		(void) ldc_unreg_callback(srvr->ldc_handle);
2554 	}
2555 	if (srvr->state & VDC_LDC_INIT) {
2556 		DMSG(vdc, 0, "[%d] ldc_fini()\n", instance);
2557 		(void) ldc_fini(srvr->ldc_handle);
2558 		srvr->ldc_handle = NULL;
2559 	}
2560 
2561 	srvr->state &= ~(VDC_LDC_INIT | VDC_LDC_CB | VDC_LDC_OPEN);
2562 }
2563 
2564 /*
2565  * Function:
2566  *	vdc_fini_ports()
2567  *
2568  * Description:
2569  *	Finalize all ports by closing the channel associated with each
2570  *	port and also freeing the server structure.
2571  *
2572  * Arguments:
2573  *	vdc	- soft state pointer for this instance of the device driver.
2574  *
2575  * Return Code:
2576  *	None
2577  */
2578 static void
2579 vdc_fini_ports(vdc_t *vdc)
2580 {
2581 	int		instance = ddi_get_instance(vdc->dip);
2582 	vdc_server_t	*srvr, *prev_srvr;
2583 
2584 	ASSERT(vdc != NULL);
2585 	ASSERT(mutex_owned(&vdc->lock));
2586 
2587 	DMSG(vdc, 0, "[%d] initialized=%x\n", instance, vdc->initialized);
2588 
2589 	srvr = vdc->server_list;
2590 
2591 	while (srvr) {
2592 
2593 		vdc_terminate_ldc(vdc, srvr);
2594 
2595 		/* next server */
2596 		prev_srvr = srvr;
2597 		srvr = srvr->next;
2598 
2599 		/* free server */
2600 		kmem_free(prev_srvr, sizeof (vdc_server_t));
2601 	}
2602 
2603 	vdc->server_list = NULL;
2604 }
2605 
2606 /* -------------------------------------------------------------------------- */
2607 
2608 /*
2609  * Descriptor Ring helper routines
2610  */
2611 
2612 /*
2613  * Function:
2614  *	vdc_init_descriptor_ring()
2615  *
2616  * Description:
2617  *
2618  * Arguments:
2619  *	vdc	- soft state pointer for this instance of the device driver.
2620  *
2621  * Return Code:
2622  *	0	- Success
2623  */
2624 static int
2625 vdc_init_descriptor_ring(vdc_t *vdc)
2626 {
2627 	vd_dring_entry_t	*dep = NULL;	/* DRing Entry pointer */
2628 	int	status = 0;
2629 	int	i;
2630 
2631 	DMSG(vdc, 0, "[%d] initialized=%x\n", vdc->instance, vdc->initialized);
2632 
2633 	ASSERT(vdc != NULL);
2634 	ASSERT(mutex_owned(&vdc->lock));
2635 
2636 	/* ensure we have enough room to store max sized block */
2637 	ASSERT(maxphys <= VD_MAX_BLOCK_SIZE);
2638 
2639 	if ((vdc->initialized & VDC_DRING_INIT) == 0) {
2640 		DMSG(vdc, 0, "[%d] ldc_mem_dring_create\n", vdc->instance);
2641 		/*
2642 		 * Calculate the maximum block size we can transmit using one
2643 		 * Descriptor Ring entry from the attributes returned by the
2644 		 * vDisk server. This is subject to a minimum of 'maxphys'
2645 		 * as we do not have the capability to split requests over
2646 		 * multiple DRing entries.
2647 		 */
2648 		if ((vdc->max_xfer_sz * vdc->block_size) < maxphys) {
2649 			DMSG(vdc, 0, "[%d] using minimum DRing size\n",
2650 			    vdc->instance);
2651 			vdc->dring_max_cookies = maxphys / PAGESIZE;
2652 		} else {
2653 			vdc->dring_max_cookies =
2654 			    (vdc->max_xfer_sz * vdc->block_size) / PAGESIZE;
2655 		}
2656 		vdc->dring_entry_size = (sizeof (vd_dring_entry_t) +
2657 		    (sizeof (ldc_mem_cookie_t) *
2658 		    (vdc->dring_max_cookies - 1)));
2659 		vdc->dring_len = VD_DRING_LEN;
2660 
2661 		status = ldc_mem_dring_create(vdc->dring_len,
2662 		    vdc->dring_entry_size, &vdc->dring_hdl);
2663 		if ((vdc->dring_hdl == NULL) || (status != 0)) {
2664 			DMSG(vdc, 0, "[%d] Descriptor ring creation failed",
2665 			    vdc->instance);
2666 			return (status);
2667 		}
2668 		vdc->initialized |= VDC_DRING_INIT;
2669 	}
2670 
2671 	if ((vdc->initialized & VDC_DRING_BOUND) == 0) {
2672 		DMSG(vdc, 0, "[%d] ldc_mem_dring_bind\n", vdc->instance);
2673 		vdc->dring_cookie =
2674 		    kmem_zalloc(sizeof (ldc_mem_cookie_t), KM_SLEEP);
2675 
2676 		status = ldc_mem_dring_bind(vdc->curr_server->ldc_handle,
2677 		    vdc->dring_hdl,
2678 		    LDC_SHADOW_MAP|LDC_DIRECT_MAP, LDC_MEM_RW,
2679 		    &vdc->dring_cookie[0],
2680 		    &vdc->dring_cookie_count);
2681 		if (status != 0) {
2682 			DMSG(vdc, 0, "[%d] Failed to bind descriptor ring "
2683 			    "(%lx) to channel (%lx) status=%d\n",
2684 			    vdc->instance, vdc->dring_hdl,
2685 			    vdc->curr_server->ldc_handle, status);
2686 			return (status);
2687 		}
2688 		ASSERT(vdc->dring_cookie_count == 1);
2689 		vdc->initialized |= VDC_DRING_BOUND;
2690 	}
2691 
2692 	status = ldc_mem_dring_info(vdc->dring_hdl, &vdc->dring_mem_info);
2693 	if (status != 0) {
2694 		DMSG(vdc, 0,
2695 		    "[%d] Failed to get info for descriptor ring (%lx)\n",
2696 		    vdc->instance, vdc->dring_hdl);
2697 		return (status);
2698 	}
2699 
2700 	if ((vdc->initialized & VDC_DRING_LOCAL) == 0) {
2701 		DMSG(vdc, 0, "[%d] local dring\n", vdc->instance);
2702 
2703 		/* Allocate the local copy of this dring */
2704 		vdc->local_dring =
2705 		    kmem_zalloc(vdc->dring_len * sizeof (vdc_local_desc_t),
2706 		    KM_SLEEP);
2707 		vdc->initialized |= VDC_DRING_LOCAL;
2708 	}
2709 
2710 	/*
2711 	 * Mark all DRing entries as free and initialize the private
2712 	 * descriptor's memory handles. If any entry is initialized,
2713 	 * we need to free it later so we set the bit in 'initialized'
2714 	 * at the start.
2715 	 */
2716 	vdc->initialized |= VDC_DRING_ENTRY;
2717 	for (i = 0; i < vdc->dring_len; i++) {
2718 		dep = VDC_GET_DRING_ENTRY_PTR(vdc, i);
2719 		dep->hdr.dstate = VIO_DESC_FREE;
2720 
2721 		status = ldc_mem_alloc_handle(vdc->curr_server->ldc_handle,
2722 		    &vdc->local_dring[i].desc_mhdl);
2723 		if (status != 0) {
2724 			DMSG(vdc, 0, "![%d] Failed to alloc mem handle for"
2725 			    " descriptor %d", vdc->instance, i);
2726 			return (status);
2727 		}
2728 		vdc->local_dring[i].is_free = B_TRUE;
2729 		vdc->local_dring[i].dep = dep;
2730 	}
2731 
2732 	/* Initialize the starting index */
2733 	vdc->dring_curr_idx = 0;
2734 
2735 	return (status);
2736 }
2737 
2738 /*
2739  * Function:
2740  *	vdc_destroy_descriptor_ring()
2741  *
2742  * Description:
2743  *
2744  * Arguments:
2745  *	vdc	- soft state pointer for this instance of the device driver.
2746  *
2747  * Return Code:
2748  *	None
2749  */
2750 static void
2751 vdc_destroy_descriptor_ring(vdc_t *vdc)
2752 {
2753 	vdc_local_desc_t	*ldep = NULL;	/* Local Dring Entry Pointer */
2754 	ldc_mem_handle_t	mhdl = NULL;
2755 	ldc_mem_info_t		minfo;
2756 	int			status = -1;
2757 	int			i;	/* loop */
2758 
2759 	ASSERT(vdc != NULL);
2760 	ASSERT(mutex_owned(&vdc->lock));
2761 
2762 	DMSG(vdc, 0, "[%d] Entered\n", vdc->instance);
2763 
2764 	if (vdc->initialized & VDC_DRING_ENTRY) {
2765 		DMSG(vdc, 0,
2766 		    "[%d] Removing Local DRing entries\n", vdc->instance);
2767 		for (i = 0; i < vdc->dring_len; i++) {
2768 			ldep = &vdc->local_dring[i];
2769 			mhdl = ldep->desc_mhdl;
2770 
2771 			if (mhdl == NULL)
2772 				continue;
2773 
2774 			if ((status = ldc_mem_info(mhdl, &minfo)) != 0) {
2775 				DMSG(vdc, 0,
2776 				    "ldc_mem_info returned an error: %d\n",
2777 				    status);
2778 
2779 				/*
2780 				 * This must mean that the mem handle
2781 				 * is not valid. Clear it out so that
2782 				 * no one tries to use it.
2783 				 */
2784 				ldep->desc_mhdl = NULL;
2785 				continue;
2786 			}
2787 
2788 			if (minfo.status == LDC_BOUND) {
2789 				(void) ldc_mem_unbind_handle(mhdl);
2790 			}
2791 
2792 			(void) ldc_mem_free_handle(mhdl);
2793 
2794 			ldep->desc_mhdl = NULL;
2795 		}
2796 		vdc->initialized &= ~VDC_DRING_ENTRY;
2797 	}
2798 
2799 	if (vdc->initialized & VDC_DRING_LOCAL) {
2800 		DMSG(vdc, 0, "[%d] Freeing Local DRing\n", vdc->instance);
2801 		kmem_free(vdc->local_dring,
2802 		    vdc->dring_len * sizeof (vdc_local_desc_t));
2803 		vdc->initialized &= ~VDC_DRING_LOCAL;
2804 	}
2805 
2806 	if (vdc->initialized & VDC_DRING_BOUND) {
2807 		DMSG(vdc, 0, "[%d] Unbinding DRing\n", vdc->instance);
2808 		status = ldc_mem_dring_unbind(vdc->dring_hdl);
2809 		if (status == 0) {
2810 			vdc->initialized &= ~VDC_DRING_BOUND;
2811 		} else {
2812 			DMSG(vdc, 0, "[%d] Error %d unbinding DRing %lx",
2813 			    vdc->instance, status, vdc->dring_hdl);
2814 		}
2815 		kmem_free(vdc->dring_cookie, sizeof (ldc_mem_cookie_t));
2816 	}
2817 
2818 	if (vdc->initialized & VDC_DRING_INIT) {
2819 		DMSG(vdc, 0, "[%d] Destroying DRing\n", vdc->instance);
2820 		status = ldc_mem_dring_destroy(vdc->dring_hdl);
2821 		if (status == 0) {
2822 			vdc->dring_hdl = NULL;
2823 			bzero(&vdc->dring_mem_info, sizeof (ldc_mem_info_t));
2824 			vdc->initialized &= ~VDC_DRING_INIT;
2825 		} else {
2826 			DMSG(vdc, 0, "[%d] Error %d destroying DRing (%lx)",
2827 			    vdc->instance, status, vdc->dring_hdl);
2828 		}
2829 	}
2830 }
2831 
2832 /*
2833  * Function:
2834  *	vdc_map_to_shared_dring()
2835  *
2836  * Description:
2837  *	Copy contents of the local descriptor to the shared
2838  *	memory descriptor.
2839  *
2840  * Arguments:
2841  *	vdcp	- soft state pointer for this instance of the device driver.
2842  *	idx	- descriptor ring index
2843  *
2844  * Return Code:
2845  *	None
2846  */
2847 static int
2848 vdc_map_to_shared_dring(vdc_t *vdcp, int idx)
2849 {
2850 	vdc_local_desc_t	*ldep;
2851 	vd_dring_entry_t	*dep;
2852 	int			rv;
2853 
2854 	ldep = &(vdcp->local_dring[idx]);
2855 
2856 	/* for now leave in the old pop_mem_hdl stuff */
2857 	if (ldep->nbytes > 0) {
2858 		rv = vdc_populate_mem_hdl(vdcp, ldep);
2859 		if (rv) {
2860 			DMSG(vdcp, 0, "[%d] Cannot populate mem handle\n",
2861 			    vdcp->instance);
2862 			return (rv);
2863 		}
2864 	}
2865 
2866 	/*
2867 	 * fill in the data details into the DRing
2868 	 */
2869 	dep = ldep->dep;
2870 	ASSERT(dep != NULL);
2871 
2872 	dep->payload.req_id = VDC_GET_NEXT_REQ_ID(vdcp);
2873 	dep->payload.operation = ldep->operation;
2874 	dep->payload.addr = ldep->offset;
2875 	dep->payload.nbytes = ldep->nbytes;
2876 	dep->payload.status = (uint32_t)-1;	/* vds will set valid value */
2877 	dep->payload.slice = ldep->slice;
2878 	dep->hdr.dstate = VIO_DESC_READY;
2879 	dep->hdr.ack = 1;		/* request an ACK for every message */
2880 
2881 	return (0);
2882 }
2883 
2884 /*
2885  * Function:
2886  *	vdc_send_request
2887  *
2888  * Description:
2889  *	This routine writes the data to be transmitted to vds into the
2890  *	descriptor, notifies vds that the ring has been updated and
2891  *	then waits for the request to be processed.
2892  *
2893  * Arguments:
2894  *	vdcp	  - the soft state pointer
2895  *	operation - operation we want vds to perform (VD_OP_XXX)
2896  *	addr	  - address of data buf to be read/written.
2897  *	nbytes	  - number of bytes to read/write
2898  *	slice	  - the disk slice this request is for
2899  *	offset	  - relative disk offset
2900  *	cb_type   - type of call - STRATEGY or SYNC
2901  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
2902  *			. mode for ioctl(9e)
2903  *			. LP64 diskaddr_t (block I/O)
2904  *	dir	  - direction of operation (READ/WRITE/BOTH)
2905  *
2906  * Return Codes:
2907  *	0
2908  *	ENXIO
2909  */
2910 static int
2911 vdc_send_request(vdc_t *vdcp, int operation, caddr_t addr,
2912     size_t nbytes, int slice, diskaddr_t offset, int cb_type,
2913     void *cb_arg, vio_desc_direction_t dir)
2914 {
2915 	int	rv = 0;
2916 
2917 	ASSERT(vdcp != NULL);
2918 	ASSERT(slice == VD_SLICE_NONE || slice < V_NUMPAR);
2919 
2920 	mutex_enter(&vdcp->lock);
2921 
2922 	/*
2923 	 * If this is a block read/write operation we update the I/O statistics
2924 	 * to indicate that the request is being put on the waitq to be
2925 	 * serviced.
2926 	 *
2927 	 * We do it here (a common routine for both synchronous and strategy
2928 	 * calls) for performance reasons - we are already holding vdc->lock
2929 	 * so there is no extra locking overhead. We would have to explicitly
2930 	 * grab the 'lock' mutex to update the stats if we were to do this
2931 	 * higher up the stack in vdc_strategy() et. al.
2932 	 */
2933 	if ((operation == VD_OP_BREAD) || (operation == VD_OP_BWRITE)) {
2934 		DTRACE_IO1(start, buf_t *, cb_arg);
2935 		VD_KSTAT_WAITQ_ENTER(vdcp);
2936 	}
2937 
2938 	do {
2939 		while (vdcp->state != VDC_STATE_RUNNING) {
2940 
2941 			/* return error if detaching */
2942 			if (vdcp->state == VDC_STATE_DETACH) {
2943 				rv = ENXIO;
2944 				goto done;
2945 			}
2946 
2947 			/* fail request if connection timeout is reached */
2948 			if (vdcp->ctimeout_reached) {
2949 				rv = EIO;
2950 				goto done;
2951 			}
2952 
2953 			/*
2954 			 * If we are panicking and the disk is not ready then
2955 			 * we can't send any request because we can't complete
2956 			 * the handshake now.
2957 			 */
2958 			if (ddi_in_panic()) {
2959 				rv = EIO;
2960 				goto done;
2961 			}
2962 
2963 			cv_wait(&vdcp->running_cv, &vdcp->lock);
2964 		}
2965 
2966 	} while (vdc_populate_descriptor(vdcp, operation, addr,
2967 	    nbytes, slice, offset, cb_type, cb_arg, dir));
2968 
2969 done:
2970 	/*
2971 	 * If this is a block read/write we update the I/O statistics kstat
2972 	 * to indicate that this request has been placed on the queue for
2973 	 * processing (i.e sent to the vDisk server) - iostat(1M) will
2974 	 * report the time waiting for the vDisk server under the %b column
2975 	 * In the case of an error we simply take it off the wait queue.
2976 	 */
2977 	if ((operation == VD_OP_BREAD) || (operation == VD_OP_BWRITE)) {
2978 		if (rv == 0) {
2979 			VD_KSTAT_WAITQ_TO_RUNQ(vdcp);
2980 			DTRACE_PROBE1(send, buf_t *, cb_arg);
2981 		} else {
2982 			VD_UPDATE_ERR_STATS(vdcp, vd_transerrs);
2983 			VD_KSTAT_WAITQ_EXIT(vdcp);
2984 			DTRACE_IO1(done, buf_t *, cb_arg);
2985 		}
2986 	}
2987 
2988 	mutex_exit(&vdcp->lock);
2989 
2990 	return (rv);
2991 }
2992 
2993 
2994 /*
2995  * Function:
2996  *	vdc_populate_descriptor
2997  *
2998  * Description:
2999  *	This routine writes the data to be transmitted to vds into the
3000  *	descriptor, notifies vds that the ring has been updated and
3001  *	then waits for the request to be processed.
3002  *
3003  * Arguments:
3004  *	vdcp	  - the soft state pointer
3005  *	operation - operation we want vds to perform (VD_OP_XXX)
3006  *	addr	  - address of data buf to be read/written.
3007  *	nbytes	  - number of bytes to read/write
3008  *	slice	  - the disk slice this request is for
3009  *	offset	  - relative disk offset
3010  *	cb_type   - type of call - STRATEGY or SYNC
3011  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
3012  *			. mode for ioctl(9e)
3013  *			. LP64 diskaddr_t (block I/O)
3014  *	dir	  - direction of operation (READ/WRITE/BOTH)
3015  *
3016  * Return Codes:
3017  *	0
3018  *	EAGAIN
3019  *	ECONNRESET
3020  *	ENXIO
3021  */
3022 static int
3023 vdc_populate_descriptor(vdc_t *vdcp, int operation, caddr_t addr,
3024     size_t nbytes, int slice, diskaddr_t offset, int cb_type,
3025     void *cb_arg, vio_desc_direction_t dir)
3026 {
3027 	vdc_local_desc_t	*local_dep = NULL; /* Local Dring Pointer */
3028 	int			idx;		/* Index of DRing entry used */
3029 	int			next_idx;
3030 	vio_dring_msg_t		dmsg;
3031 	size_t			msglen;
3032 	int			rv;
3033 
3034 	ASSERT(MUTEX_HELD(&vdcp->lock));
3035 	vdcp->threads_pending++;
3036 loop:
3037 	DMSG(vdcp, 2, ": dring_curr_idx = %d\n", vdcp->dring_curr_idx);
3038 
3039 	/* Get next available D-Ring entry */
3040 	idx = vdcp->dring_curr_idx;
3041 	local_dep = &(vdcp->local_dring[idx]);
3042 
3043 	if (!local_dep->is_free) {
3044 		DMSG(vdcp, 2, "[%d]: dring full - waiting for space\n",
3045 		    vdcp->instance);
3046 		cv_wait(&vdcp->dring_free_cv, &vdcp->lock);
3047 		if (vdcp->state == VDC_STATE_RUNNING ||
3048 		    vdcp->state == VDC_STATE_HANDLE_PENDING) {
3049 			goto loop;
3050 		}
3051 		vdcp->threads_pending--;
3052 		return (ECONNRESET);
3053 	}
3054 
3055 	next_idx = idx + 1;
3056 	if (next_idx >= vdcp->dring_len)
3057 		next_idx = 0;
3058 	vdcp->dring_curr_idx = next_idx;
3059 
3060 	ASSERT(local_dep->is_free);
3061 
3062 	local_dep->operation = operation;
3063 	local_dep->addr = addr;
3064 	local_dep->nbytes = nbytes;
3065 	local_dep->slice = slice;
3066 	local_dep->offset = offset;
3067 	local_dep->cb_type = cb_type;
3068 	local_dep->cb_arg = cb_arg;
3069 	local_dep->dir = dir;
3070 
3071 	local_dep->is_free = B_FALSE;
3072 
3073 	rv = vdc_map_to_shared_dring(vdcp, idx);
3074 	if (rv) {
3075 		DMSG(vdcp, 0, "[%d]: cannot bind memory - waiting ..\n",
3076 		    vdcp->instance);
3077 		/* free the descriptor */
3078 		local_dep->is_free = B_TRUE;
3079 		vdcp->dring_curr_idx = idx;
3080 		cv_wait(&vdcp->membind_cv, &vdcp->lock);
3081 		if (vdcp->state == VDC_STATE_RUNNING ||
3082 		    vdcp->state == VDC_STATE_HANDLE_PENDING) {
3083 			goto loop;
3084 		}
3085 		vdcp->threads_pending--;
3086 		return (ECONNRESET);
3087 	}
3088 
3089 	/*
3090 	 * Send a msg with the DRing details to vds
3091 	 */
3092 	VIO_INIT_DRING_DATA_TAG(dmsg);
3093 	VDC_INIT_DRING_DATA_MSG_IDS(dmsg, vdcp);
3094 	dmsg.dring_ident = vdcp->dring_ident;
3095 	dmsg.start_idx = idx;
3096 	dmsg.end_idx = idx;
3097 	vdcp->seq_num++;
3098 
3099 	DTRACE_PROBE2(populate, int, vdcp->instance,
3100 	    vdc_local_desc_t *, local_dep);
3101 	DMSG(vdcp, 2, "ident=0x%lx, st=%u, end=%u, seq=%ld\n",
3102 	    vdcp->dring_ident, dmsg.start_idx, dmsg.end_idx, dmsg.seq_num);
3103 
3104 	/*
3105 	 * note we're still holding the lock here to
3106 	 * make sure the message goes out in order !!!...
3107 	 */
3108 	msglen = sizeof (dmsg);
3109 	rv = vdc_send(vdcp, (caddr_t)&dmsg, &msglen);
3110 	switch (rv) {
3111 	case ECONNRESET:
3112 		/*
3113 		 * vdc_send initiates the reset on failure.
3114 		 * Since the transaction has already been put
3115 		 * on the local dring, it will automatically get
3116 		 * retried when the channel is reset. Given that,
3117 		 * it is ok to just return success even though the
3118 		 * send failed.
3119 		 */
3120 		rv = 0;
3121 		break;
3122 
3123 	case 0: /* EOK */
3124 		DMSG(vdcp, 1, "sent via LDC: rv=%d\n", rv);
3125 		break;
3126 
3127 	default:
3128 		goto cleanup_and_exit;
3129 	}
3130 
3131 	vdcp->threads_pending--;
3132 	return (rv);
3133 
3134 cleanup_and_exit:
3135 	DMSG(vdcp, 0, "unexpected error, rv=%d\n", rv);
3136 	return (ENXIO);
3137 }
3138 
3139 /*
3140  * Function:
3141  *	vdc_do_sync_op
3142  *
3143  * Description:
3144  * 	Wrapper around vdc_populate_descriptor that blocks until the
3145  * 	response to the message is available.
3146  *
3147  * Arguments:
3148  *	vdcp	  - the soft state pointer
3149  *	operation - operation we want vds to perform (VD_OP_XXX)
3150  *	addr	  - address of data buf to be read/written.
3151  *	nbytes	  - number of bytes to read/write
3152  *	slice	  - the disk slice this request is for
3153  *	offset	  - relative disk offset
3154  *	cb_type   - type of call - STRATEGY or SYNC
3155  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
3156  *			. mode for ioctl(9e)
3157  *			. LP64 diskaddr_t (block I/O)
3158  *	dir	  - direction of operation (READ/WRITE/BOTH)
3159  *	rconflict - check for reservation conflict in case of failure
3160  *
3161  * rconflict should be set to B_TRUE by most callers. Callers invoking the
3162  * VD_OP_SCSICMD operation can set rconflict to B_FALSE if they check the
3163  * result of a successful operation with vd_scsi_status().
3164  *
3165  * Return Codes:
3166  *	0
3167  *	EAGAIN
3168  *	EFAULT
3169  *	ENXIO
3170  *	EIO
3171  */
3172 static int
3173 vdc_do_sync_op(vdc_t *vdcp, int operation, caddr_t addr, size_t nbytes,
3174     int slice, diskaddr_t offset, int cb_type, void *cb_arg,
3175     vio_desc_direction_t dir, boolean_t rconflict)
3176 {
3177 	int status;
3178 	vdc_io_t *vio;
3179 	boolean_t check_resv_conflict = B_FALSE;
3180 
3181 	ASSERT(cb_type == CB_SYNC);
3182 
3183 	/*
3184 	 * Grab the lock, if blocked wait until the server
3185 	 * response causes us to wake up again.
3186 	 */
3187 	mutex_enter(&vdcp->lock);
3188 	vdcp->sync_op_cnt++;
3189 	while (vdcp->sync_op_blocked && vdcp->state != VDC_STATE_DETACH)
3190 		cv_wait(&vdcp->sync_blocked_cv, &vdcp->lock);
3191 
3192 	if (vdcp->state == VDC_STATE_DETACH) {
3193 		cv_broadcast(&vdcp->sync_blocked_cv);
3194 		vdcp->sync_op_cnt--;
3195 		mutex_exit(&vdcp->lock);
3196 		return (ENXIO);
3197 	}
3198 
3199 	/* now block anyone other thread entering after us */
3200 	vdcp->sync_op_blocked = B_TRUE;
3201 	vdcp->sync_op_pending = B_TRUE;
3202 	mutex_exit(&vdcp->lock);
3203 
3204 	status = vdc_send_request(vdcp, operation, addr,
3205 	    nbytes, slice, offset, cb_type, cb_arg, dir);
3206 
3207 	mutex_enter(&vdcp->lock);
3208 
3209 	if (status != 0) {
3210 		vdcp->sync_op_pending = B_FALSE;
3211 	} else {
3212 		/*
3213 		 * block until our transaction completes.
3214 		 * Also anyone else waiting also gets to go next.
3215 		 */
3216 		while (vdcp->sync_op_pending && vdcp->state != VDC_STATE_DETACH)
3217 			cv_wait(&vdcp->sync_pending_cv, &vdcp->lock);
3218 
3219 		DMSG(vdcp, 2, ": operation returned %d\n",
3220 		    vdcp->sync_op_status);
3221 		if (vdcp->state == VDC_STATE_DETACH) {
3222 			vdcp->sync_op_pending = B_FALSE;
3223 			status = ENXIO;
3224 		} else {
3225 			status = vdcp->sync_op_status;
3226 			if (status != 0 && vdcp->failfast_interval != 0) {
3227 				/*
3228 				 * Operation has failed and failfast is enabled.
3229 				 * We need to check if the failure is due to a
3230 				 * reservation conflict if this was requested.
3231 				 */
3232 				check_resv_conflict = rconflict;
3233 			}
3234 
3235 		}
3236 	}
3237 
3238 	vdcp->sync_op_status = 0;
3239 	vdcp->sync_op_blocked = B_FALSE;
3240 	vdcp->sync_op_cnt--;
3241 
3242 	/* signal the next waiting thread */
3243 	cv_signal(&vdcp->sync_blocked_cv);
3244 
3245 	/*
3246 	 * We have to check for reservation conflict after unblocking sync
3247 	 * operations because some sync operations will be used to do this
3248 	 * check.
3249 	 */
3250 	if (check_resv_conflict) {
3251 		vio = vdc_failfast_io_queue(vdcp, NULL);
3252 		while (vio->vio_qtime != 0)
3253 			cv_wait(&vdcp->failfast_io_cv, &vdcp->lock);
3254 		kmem_free(vio, sizeof (vdc_io_t));
3255 	}
3256 
3257 	mutex_exit(&vdcp->lock);
3258 
3259 	return (status);
3260 }
3261 
3262 
3263 /*
3264  * Function:
3265  *	vdc_drain_response()
3266  *
3267  * Description:
3268  * 	When a guest is panicking, the completion of requests needs to be
3269  * 	handled differently because interrupts are disabled and vdc
3270  * 	will not get messages. We have to poll for the messages instead.
3271  *
3272  *	Note: since we don't have a buf_t available we cannot implement
3273  *	the io:::done DTrace probe in this specific case.
3274  *
3275  * Arguments:
3276  *	vdc	- soft state pointer for this instance of the device driver.
3277  *
3278  * Return Code:
3279  *	0	- Success
3280  */
3281 static int
3282 vdc_drain_response(vdc_t *vdc)
3283 {
3284 	int 			rv, idx, retries;
3285 	size_t			msglen;
3286 	vdc_local_desc_t 	*ldep = NULL;	/* Local Dring Entry Pointer */
3287 	vio_dring_msg_t		dmsg;
3288 
3289 	mutex_enter(&vdc->lock);
3290 
3291 	retries = 0;
3292 	for (;;) {
3293 		msglen = sizeof (dmsg);
3294 		rv = ldc_read(vdc->curr_server->ldc_handle, (caddr_t)&dmsg,
3295 		    &msglen);
3296 		if (rv) {
3297 			rv = EINVAL;
3298 			break;
3299 		}
3300 
3301 		/*
3302 		 * if there are no packets wait and check again
3303 		 */
3304 		if ((rv == 0) && (msglen == 0)) {
3305 			if (retries++ > vdc_dump_retries) {
3306 				rv = EAGAIN;
3307 				break;
3308 			}
3309 
3310 			drv_usecwait(vdc_usec_timeout_dump);
3311 			continue;
3312 		}
3313 
3314 		/*
3315 		 * Ignore all messages that are not ACKs/NACKs to
3316 		 * DRing requests.
3317 		 */
3318 		if ((dmsg.tag.vio_msgtype != VIO_TYPE_DATA) ||
3319 		    (dmsg.tag.vio_subtype_env != VIO_DRING_DATA)) {
3320 			DMSG(vdc, 0, "discard pkt: type=%d sub=%d env=%d\n",
3321 			    dmsg.tag.vio_msgtype,
3322 			    dmsg.tag.vio_subtype,
3323 			    dmsg.tag.vio_subtype_env);
3324 			continue;
3325 		}
3326 
3327 		/*
3328 		 * set the appropriate return value for the current request.
3329 		 */
3330 		switch (dmsg.tag.vio_subtype) {
3331 		case VIO_SUBTYPE_ACK:
3332 			rv = 0;
3333 			break;
3334 		case VIO_SUBTYPE_NACK:
3335 			rv = EAGAIN;
3336 			break;
3337 		default:
3338 			continue;
3339 		}
3340 
3341 		idx = dmsg.start_idx;
3342 		if (idx >= vdc->dring_len) {
3343 			DMSG(vdc, 0, "[%d] Bogus ack data : start %d\n",
3344 			    vdc->instance, idx);
3345 			continue;
3346 		}
3347 		ldep = &vdc->local_dring[idx];
3348 		if (ldep->dep->hdr.dstate != VIO_DESC_DONE) {
3349 			DMSG(vdc, 0, "[%d] Entry @ %d - state !DONE %d\n",
3350 			    vdc->instance, idx, ldep->dep->hdr.dstate);
3351 			continue;
3352 		}
3353 
3354 		DMSG(vdc, 1, "[%d] Depopulating idx=%d state=%d\n",
3355 		    vdc->instance, idx, ldep->dep->hdr.dstate);
3356 
3357 		rv = vdc_depopulate_descriptor(vdc, idx);
3358 		if (rv) {
3359 			DMSG(vdc, 0,
3360 			    "[%d] Entry @ %d - depopulate failed ..\n",
3361 			    vdc->instance, idx);
3362 		}
3363 
3364 		/* if this is the last descriptor - break out of loop */
3365 		if ((idx + 1) % vdc->dring_len == vdc->dring_curr_idx)
3366 			break;
3367 	}
3368 
3369 	mutex_exit(&vdc->lock);
3370 	DMSG(vdc, 0, "End idx=%d\n", idx);
3371 
3372 	return (rv);
3373 }
3374 
3375 
3376 /*
3377  * Function:
3378  *	vdc_depopulate_descriptor()
3379  *
3380  * Description:
3381  *
3382  * Arguments:
3383  *	vdc	- soft state pointer for this instance of the device driver.
3384  *	idx	- Index of the Descriptor Ring entry being modified
3385  *
3386  * Return Code:
3387  *	0	- Success
3388  */
3389 static int
3390 vdc_depopulate_descriptor(vdc_t *vdc, uint_t idx)
3391 {
3392 	vd_dring_entry_t *dep = NULL;		/* Dring Entry Pointer */
3393 	vdc_local_desc_t *ldep = NULL;		/* Local Dring Entry Pointer */
3394 	int		status = ENXIO;
3395 	int		rv = 0;
3396 
3397 	ASSERT(vdc != NULL);
3398 	ASSERT(idx < vdc->dring_len);
3399 	ldep = &vdc->local_dring[idx];
3400 	ASSERT(ldep != NULL);
3401 	ASSERT(MUTEX_HELD(&vdc->lock));
3402 
3403 	DTRACE_PROBE2(depopulate, int, vdc->instance, vdc_local_desc_t *, ldep);
3404 	DMSG(vdc, 2, ": idx = %d\n", idx);
3405 
3406 	dep = ldep->dep;
3407 	ASSERT(dep != NULL);
3408 	ASSERT((dep->hdr.dstate == VIO_DESC_DONE) ||
3409 	    (dep->payload.status == ECANCELED));
3410 
3411 	VDC_MARK_DRING_ENTRY_FREE(vdc, idx);
3412 
3413 	ldep->is_free = B_TRUE;
3414 	status = dep->payload.status;
3415 	DMSG(vdc, 2, ": is_free = %d : status = %d\n", ldep->is_free, status);
3416 
3417 	/*
3418 	 * If no buffers were used to transfer information to the server when
3419 	 * populating the descriptor then no memory handles need to be unbound
3420 	 * and we can return now.
3421 	 */
3422 	if (ldep->nbytes == 0) {
3423 		cv_signal(&vdc->dring_free_cv);
3424 		return (status);
3425 	}
3426 
3427 	/*
3428 	 * If the upper layer passed in a misaligned address we copied the
3429 	 * data into an aligned buffer before sending it to LDC - we now
3430 	 * copy it back to the original buffer.
3431 	 */
3432 	if (ldep->align_addr) {
3433 		ASSERT(ldep->addr != NULL);
3434 
3435 		if (dep->payload.nbytes > 0)
3436 			bcopy(ldep->align_addr, ldep->addr,
3437 			    dep->payload.nbytes);
3438 		kmem_free(ldep->align_addr,
3439 		    sizeof (caddr_t) * P2ROUNDUP(ldep->nbytes, 8));
3440 		ldep->align_addr = NULL;
3441 	}
3442 
3443 	rv = ldc_mem_unbind_handle(ldep->desc_mhdl);
3444 	if (rv != 0) {
3445 		DMSG(vdc, 0, "?[%d] unbind mhdl 0x%lx @ idx %d failed (%d)",
3446 		    vdc->instance, ldep->desc_mhdl, idx, rv);
3447 		/*
3448 		 * The error returned by the vDisk server is more informative
3449 		 * and thus has a higher priority but if it isn't set we ensure
3450 		 * that this function returns an error.
3451 		 */
3452 		if (status == 0)
3453 			status = EINVAL;
3454 	}
3455 
3456 	cv_signal(&vdc->membind_cv);
3457 	cv_signal(&vdc->dring_free_cv);
3458 
3459 	return (status);
3460 }
3461 
3462 /*
3463  * Function:
3464  *	vdc_populate_mem_hdl()
3465  *
3466  * Description:
3467  *
3468  * Arguments:
3469  *	vdc	- soft state pointer for this instance of the device driver.
3470  *	idx	- Index of the Descriptor Ring entry being modified
3471  *	addr	- virtual address being mapped in
3472  *	nybtes	- number of bytes in 'addr'
3473  *	operation - the vDisk operation being performed (VD_OP_xxx)
3474  *
3475  * Return Code:
3476  *	0	- Success
3477  */
3478 static int
3479 vdc_populate_mem_hdl(vdc_t *vdcp, vdc_local_desc_t *ldep)
3480 {
3481 	vd_dring_entry_t	*dep = NULL;
3482 	ldc_mem_handle_t	mhdl;
3483 	caddr_t			vaddr;
3484 	size_t			nbytes;
3485 	uint8_t			perm = LDC_MEM_RW;
3486 	uint8_t			maptype;
3487 	int			rv = 0;
3488 	int			i;
3489 
3490 	ASSERT(vdcp != NULL);
3491 
3492 	dep = ldep->dep;
3493 	mhdl = ldep->desc_mhdl;
3494 
3495 	switch (ldep->dir) {
3496 	case VIO_read_dir:
3497 		perm = LDC_MEM_W;
3498 		break;
3499 
3500 	case VIO_write_dir:
3501 		perm = LDC_MEM_R;
3502 		break;
3503 
3504 	case VIO_both_dir:
3505 		perm = LDC_MEM_RW;
3506 		break;
3507 
3508 	default:
3509 		ASSERT(0);	/* catch bad programming in vdc */
3510 	}
3511 
3512 	/*
3513 	 * LDC expects any addresses passed in to be 8-byte aligned. We need
3514 	 * to copy the contents of any misaligned buffers to a newly allocated
3515 	 * buffer and bind it instead (and copy the the contents back to the
3516 	 * original buffer passed in when depopulating the descriptor)
3517 	 */
3518 	vaddr = ldep->addr;
3519 	nbytes = ldep->nbytes;
3520 	if (((uint64_t)vaddr & 0x7) != 0) {
3521 		ASSERT(ldep->align_addr == NULL);
3522 		ldep->align_addr =
3523 		    kmem_alloc(sizeof (caddr_t) *
3524 		    P2ROUNDUP(nbytes, 8), KM_SLEEP);
3525 		DMSG(vdcp, 0, "[%d] Misaligned address %p reallocating "
3526 		    "(buf=%p nb=%ld op=%d)\n",
3527 		    vdcp->instance, (void *)vaddr, (void *)ldep->align_addr,
3528 		    nbytes, ldep->operation);
3529 		if (perm != LDC_MEM_W)
3530 			bcopy(vaddr, ldep->align_addr, nbytes);
3531 		vaddr = ldep->align_addr;
3532 	}
3533 
3534 	maptype = LDC_IO_MAP|LDC_SHADOW_MAP|LDC_DIRECT_MAP;
3535 	rv = ldc_mem_bind_handle(mhdl, vaddr, P2ROUNDUP(nbytes, 8),
3536 	    maptype, perm, &dep->payload.cookie[0], &dep->payload.ncookies);
3537 	DMSG(vdcp, 2, "[%d] bound mem handle; ncookies=%d\n",
3538 	    vdcp->instance, dep->payload.ncookies);
3539 	if (rv != 0) {
3540 		DMSG(vdcp, 0, "[%d] Failed to bind LDC memory handle "
3541 		    "(mhdl=%p, buf=%p, err=%d)\n",
3542 		    vdcp->instance, (void *)mhdl, (void *)vaddr, rv);
3543 		if (ldep->align_addr) {
3544 			kmem_free(ldep->align_addr,
3545 			    sizeof (caddr_t) * P2ROUNDUP(nbytes, 8));
3546 			ldep->align_addr = NULL;
3547 		}
3548 		return (EAGAIN);
3549 	}
3550 
3551 	/*
3552 	 * Get the other cookies (if any).
3553 	 */
3554 	for (i = 1; i < dep->payload.ncookies; i++) {
3555 		rv = ldc_mem_nextcookie(mhdl, &dep->payload.cookie[i]);
3556 		if (rv != 0) {
3557 			(void) ldc_mem_unbind_handle(mhdl);
3558 			DMSG(vdcp, 0, "?[%d] Failed to get next cookie "
3559 			    "(mhdl=%lx cnum=%d), err=%d",
3560 			    vdcp->instance, mhdl, i, rv);
3561 			if (ldep->align_addr) {
3562 				kmem_free(ldep->align_addr,
3563 				    sizeof (caddr_t) * ldep->nbytes);
3564 				ldep->align_addr = NULL;
3565 			}
3566 			return (EAGAIN);
3567 		}
3568 	}
3569 
3570 	return (rv);
3571 }
3572 
3573 /*
3574  * Interrupt handlers for messages from LDC
3575  */
3576 
3577 /*
3578  * Function:
3579  *	vdc_handle_cb()
3580  *
3581  * Description:
3582  *
3583  * Arguments:
3584  *	event	- Type of event (LDC_EVT_xxx) that triggered the callback
3585  *	arg	- soft state pointer for this instance of the device driver.
3586  *
3587  * Return Code:
3588  *	0	- Success
3589  */
3590 static uint_t
3591 vdc_handle_cb(uint64_t event, caddr_t arg)
3592 {
3593 	ldc_status_t	ldc_state;
3594 	int		rv = 0;
3595 	vdc_server_t	*srvr = (vdc_server_t *)(void *)arg;
3596 	vdc_t		*vdc = srvr->vdcp;
3597 
3598 	ASSERT(vdc != NULL);
3599 
3600 	DMSG(vdc, 1, "evt=%lx seqID=%ld\n", event, vdc->seq_num);
3601 
3602 	/* If callback is not for the current server, ignore it */
3603 	mutex_enter(&vdc->lock);
3604 
3605 	if (vdc->curr_server != srvr) {
3606 		DMSG(vdc, 0, "[%d] Ignoring event 0x%lx for port@%ld\n",
3607 		    vdc->instance, event, srvr->id);
3608 		mutex_exit(&vdc->lock);
3609 		return (LDC_SUCCESS);
3610 	}
3611 
3612 	/*
3613 	 * Depending on the type of event that triggered this callback,
3614 	 * we modify the handshake state or read the data.
3615 	 *
3616 	 * NOTE: not done as a switch() as event could be triggered by
3617 	 * a state change and a read request. Also the ordering	of the
3618 	 * check for the event types is deliberate.
3619 	 */
3620 	if (event & LDC_EVT_UP) {
3621 		DMSG(vdc, 0, "[%d] Received LDC_EVT_UP\n", vdc->instance);
3622 
3623 		/* get LDC state */
3624 		rv = ldc_status(srvr->ldc_handle, &ldc_state);
3625 		if (rv != 0) {
3626 			DMSG(vdc, 0, "[%d] Couldn't get LDC status %d",
3627 			    vdc->instance, rv);
3628 			mutex_exit(&vdc->lock);
3629 			return (LDC_SUCCESS);
3630 		}
3631 		if (srvr->ldc_state != LDC_UP &&
3632 		    ldc_state == LDC_UP) {
3633 			/*
3634 			 * Reset the transaction sequence numbers when
3635 			 * LDC comes up. We then kick off the handshake
3636 			 * negotiation with the vDisk server.
3637 			 */
3638 			vdc->seq_num = 1;
3639 			vdc->seq_num_reply = 0;
3640 			srvr->ldc_state = ldc_state;
3641 			cv_signal(&vdc->initwait_cv);
3642 		}
3643 	}
3644 
3645 	if (event & LDC_EVT_READ) {
3646 		DMSG(vdc, 1, "[%d] Received LDC_EVT_READ\n", vdc->instance);
3647 		mutex_enter(&vdc->read_lock);
3648 		cv_signal(&vdc->read_cv);
3649 		vdc->read_state = VDC_READ_PENDING;
3650 		mutex_exit(&vdc->read_lock);
3651 		mutex_exit(&vdc->lock);
3652 
3653 		/* that's all we have to do - no need to handle DOWN/RESET */
3654 		return (LDC_SUCCESS);
3655 	}
3656 
3657 	if (event & (LDC_EVT_RESET|LDC_EVT_DOWN)) {
3658 
3659 		DMSG(vdc, 0, "[%d] Received LDC RESET event\n", vdc->instance);
3660 
3661 		/*
3662 		 * Need to wake up any readers so they will
3663 		 * detect that a reset has occurred.
3664 		 */
3665 		mutex_enter(&vdc->read_lock);
3666 		if ((vdc->read_state == VDC_READ_WAITING) ||
3667 		    (vdc->read_state == VDC_READ_RESET))
3668 			cv_signal(&vdc->read_cv);
3669 		vdc->read_state = VDC_READ_RESET;
3670 		mutex_exit(&vdc->read_lock);
3671 
3672 		/* wake up any threads waiting for connection to come up */
3673 		if (vdc->state == VDC_STATE_INIT_WAITING) {
3674 			vdc->state = VDC_STATE_RESETTING;
3675 			cv_signal(&vdc->initwait_cv);
3676 		}
3677 
3678 	}
3679 
3680 	mutex_exit(&vdc->lock);
3681 
3682 	if (event & ~(LDC_EVT_UP | LDC_EVT_RESET | LDC_EVT_DOWN | LDC_EVT_READ))
3683 		DMSG(vdc, 0, "![%d] Unexpected LDC event (%lx) received",
3684 		    vdc->instance, event);
3685 
3686 	return (LDC_SUCCESS);
3687 }
3688 
3689 /*
3690  * Function:
3691  *	vdc_wait_for_response()
3692  *
3693  * Description:
3694  *	Block waiting for a response from the server. If there is
3695  *	no data the thread block on the read_cv that is signalled
3696  *	by the callback when an EVT_READ occurs.
3697  *
3698  * Arguments:
3699  *	vdcp	- soft state pointer for this instance of the device driver.
3700  *
3701  * Return Code:
3702  *	0	- Success
3703  */
3704 static int
3705 vdc_wait_for_response(vdc_t *vdcp, vio_msg_t *msgp)
3706 {
3707 	size_t		nbytes = sizeof (*msgp);
3708 	int		status;
3709 
3710 	ASSERT(vdcp != NULL);
3711 
3712 	DMSG(vdcp, 1, "[%d] Entered\n", vdcp->instance);
3713 
3714 	status = vdc_recv(vdcp, msgp, &nbytes);
3715 	DMSG(vdcp, 3, "vdc_read() done.. status=0x%x size=0x%x\n",
3716 	    status, (int)nbytes);
3717 	if (status) {
3718 		DMSG(vdcp, 0, "?[%d] Error %d reading LDC msg\n",
3719 		    vdcp->instance, status);
3720 		return (status);
3721 	}
3722 
3723 	if (nbytes < sizeof (vio_msg_tag_t)) {
3724 		DMSG(vdcp, 0, "?[%d] Expect %lu bytes; recv'd %lu\n",
3725 		    vdcp->instance, sizeof (vio_msg_tag_t), nbytes);
3726 		return (ENOMSG);
3727 	}
3728 
3729 	DMSG(vdcp, 2, "[%d] (%x/%x/%x)\n", vdcp->instance,
3730 	    msgp->tag.vio_msgtype,
3731 	    msgp->tag.vio_subtype,
3732 	    msgp->tag.vio_subtype_env);
3733 
3734 	/*
3735 	 * Verify the Session ID of the message
3736 	 *
3737 	 * Every message after the Version has been negotiated should
3738 	 * have the correct session ID set.
3739 	 */
3740 	if ((msgp->tag.vio_sid != vdcp->session_id) &&
3741 	    (msgp->tag.vio_subtype_env != VIO_VER_INFO)) {
3742 		DMSG(vdcp, 0, "[%d] Invalid SID: received 0x%x, "
3743 		    "expected 0x%lx [seq num %lx @ %d]",
3744 		    vdcp->instance, msgp->tag.vio_sid,
3745 		    vdcp->session_id,
3746 		    ((vio_dring_msg_t *)msgp)->seq_num,
3747 		    ((vio_dring_msg_t *)msgp)->start_idx);
3748 		return (ENOMSG);
3749 	}
3750 	return (0);
3751 }
3752 
3753 
3754 /*
3755  * Function:
3756  *	vdc_resubmit_backup_dring()
3757  *
3758  * Description:
3759  *	Resubmit each descriptor in the backed up dring to
3760  * 	vDisk server. The Dring was backed up during connection
3761  *	reset.
3762  *
3763  * Arguments:
3764  *	vdcp	- soft state pointer for this instance of the device driver.
3765  *
3766  * Return Code:
3767  *	0	- Success
3768  */
3769 static int
3770 vdc_resubmit_backup_dring(vdc_t *vdcp)
3771 {
3772 	int		processed = 0;
3773 	int		count;
3774 	int		b_idx;
3775 	int		rv = 0;
3776 	int		dring_size;
3777 	int		op;
3778 	vio_msg_t	vio_msg;
3779 	vdc_local_desc_t	*curr_ldep;
3780 
3781 	ASSERT(MUTEX_NOT_HELD(&vdcp->lock));
3782 	ASSERT(vdcp->state == VDC_STATE_HANDLE_PENDING);
3783 
3784 	if (vdcp->local_dring_backup == NULL) {
3785 		/* the pending requests have already been processed */
3786 		return (0);
3787 	}
3788 
3789 	DMSG(vdcp, 1, "restoring pending dring entries (len=%d, tail=%d)\n",
3790 	    vdcp->local_dring_backup_len, vdcp->local_dring_backup_tail);
3791 
3792 	/*
3793 	 * Walk the backup copy of the local descriptor ring and
3794 	 * resubmit all the outstanding transactions.
3795 	 */
3796 	b_idx = vdcp->local_dring_backup_tail;
3797 	for (count = 0; count < vdcp->local_dring_backup_len; count++) {
3798 
3799 		curr_ldep = &(vdcp->local_dring_backup[b_idx]);
3800 
3801 		/* only resubmit outstanding transactions */
3802 		if (!curr_ldep->is_free) {
3803 			/*
3804 			 * If we are retrying a block read/write operation we
3805 			 * need to update the I/O statistics to indicate that
3806 			 * the request is being put back on the waitq to be
3807 			 * serviced (it will have been taken off after the
3808 			 * error was reported).
3809 			 */
3810 			mutex_enter(&vdcp->lock);
3811 			op = curr_ldep->operation;
3812 			if ((op == VD_OP_BREAD) || (op == VD_OP_BWRITE)) {
3813 				DTRACE_IO1(start, buf_t *, curr_ldep->cb_arg);
3814 				VD_KSTAT_WAITQ_ENTER(vdcp);
3815 			}
3816 
3817 			DMSG(vdcp, 1, "resubmitting entry idx=%x\n", b_idx);
3818 			rv = vdc_populate_descriptor(vdcp, op,
3819 			    curr_ldep->addr, curr_ldep->nbytes,
3820 			    curr_ldep->slice, curr_ldep->offset,
3821 			    curr_ldep->cb_type, curr_ldep->cb_arg,
3822 			    curr_ldep->dir);
3823 
3824 			if (rv) {
3825 				if (op == VD_OP_BREAD || op == VD_OP_BWRITE) {
3826 					VD_UPDATE_ERR_STATS(vdcp, vd_transerrs);
3827 					VD_KSTAT_WAITQ_EXIT(vdcp);
3828 					DTRACE_IO1(done, buf_t *,
3829 					    curr_ldep->cb_arg);
3830 				}
3831 				DMSG(vdcp, 1, "[%d] cannot resubmit entry %d\n",
3832 				    vdcp->instance, b_idx);
3833 				mutex_exit(&vdcp->lock);
3834 				goto done;
3835 			}
3836 
3837 			/*
3838 			 * If this is a block read/write we update the I/O
3839 			 * statistics kstat to indicate that the request
3840 			 * has been sent back to the vDisk server and should
3841 			 * now be put on the run queue.
3842 			 */
3843 			if ((op == VD_OP_BREAD) || (op == VD_OP_BWRITE)) {
3844 				DTRACE_PROBE1(send, buf_t *, curr_ldep->cb_arg);
3845 				VD_KSTAT_WAITQ_TO_RUNQ(vdcp);
3846 			}
3847 			mutex_exit(&vdcp->lock);
3848 
3849 			/* Wait for the response message. */
3850 			DMSG(vdcp, 1, "waiting for response to idx=%x\n",
3851 			    b_idx);
3852 			rv = vdc_wait_for_response(vdcp, &vio_msg);
3853 			if (rv) {
3854 				/*
3855 				 * If this is a block read/write we update
3856 				 * the I/O statistics kstat to take it
3857 				 * off the run queue.
3858 				 */
3859 				mutex_enter(&vdcp->lock);
3860 				if (op == VD_OP_BREAD || op == VD_OP_BWRITE) {
3861 					VD_UPDATE_ERR_STATS(vdcp, vd_transerrs);
3862 					VD_KSTAT_RUNQ_EXIT(vdcp);
3863 					DTRACE_IO1(done, buf_t *,
3864 					    curr_ldep->cb_arg);
3865 				}
3866 				DMSG(vdcp, 1, "[%d] wait_for_response "
3867 				    "returned err=%d\n", vdcp->instance,
3868 				    rv);
3869 				mutex_exit(&vdcp->lock);
3870 				goto done;
3871 			}
3872 
3873 			DMSG(vdcp, 1, "processing msg for idx=%x\n", b_idx);
3874 			rv = vdc_process_data_msg(vdcp, &vio_msg);
3875 			if (rv) {
3876 				DMSG(vdcp, 1, "[%d] process_data_msg "
3877 				    "returned err=%d\n", vdcp->instance,
3878 				    rv);
3879 				goto done;
3880 			}
3881 			processed++;
3882 		}
3883 
3884 		/* get the next element to submit */
3885 		if (++b_idx >= vdcp->local_dring_backup_len)
3886 			b_idx = 0;
3887 	}
3888 
3889 	/* all done - now clear up pending dring copy */
3890 	dring_size = vdcp->local_dring_backup_len *
3891 	    sizeof (vdcp->local_dring_backup[0]);
3892 
3893 	(void) kmem_free(vdcp->local_dring_backup, dring_size);
3894 
3895 	vdcp->local_dring_backup = NULL;
3896 
3897 done:
3898 	DTRACE_PROBE2(processed, int, processed, vdc_t *, vdcp);
3899 
3900 	return (rv);
3901 }
3902 
3903 /*
3904  * Function:
3905  *	vdc_cancel_backup_dring
3906  *
3907  * Description:
3908  *	Cancel each descriptor in the backed up dring to vDisk server.
3909  *	The Dring was backed up during connection reset.
3910  *
3911  * Arguments:
3912  *	vdcp	- soft state pointer for this instance of the device driver.
3913  *
3914  * Return Code:
3915  *	None
3916  */
3917 void
3918 vdc_cancel_backup_dring(vdc_t *vdcp)
3919 {
3920 	vdc_local_desc_t *ldep;
3921 	struct buf 	*bufp;
3922 	int		count;
3923 	int		b_idx;
3924 	int		dring_size;
3925 	int		cancelled = 0;
3926 
3927 	ASSERT(MUTEX_HELD(&vdcp->lock));
3928 	ASSERT(vdcp->state == VDC_STATE_INIT ||
3929 	    vdcp->state == VDC_STATE_INIT_WAITING ||
3930 	    vdcp->state == VDC_STATE_NEGOTIATE ||
3931 	    vdcp->state == VDC_STATE_RESETTING);
3932 
3933 	if (vdcp->local_dring_backup == NULL) {
3934 		/* the pending requests have already been processed */
3935 		return;
3936 	}
3937 
3938 	DMSG(vdcp, 1, "cancelling pending dring entries (len=%d, tail=%d)\n",
3939 	    vdcp->local_dring_backup_len, vdcp->local_dring_backup_tail);
3940 
3941 	/*
3942 	 * Walk the backup copy of the local descriptor ring and
3943 	 * cancel all the outstanding transactions.
3944 	 */
3945 	b_idx = vdcp->local_dring_backup_tail;
3946 	for (count = 0; count < vdcp->local_dring_backup_len; count++) {
3947 
3948 		ldep = &(vdcp->local_dring_backup[b_idx]);
3949 
3950 		/* only cancel outstanding transactions */
3951 		if (!ldep->is_free) {
3952 
3953 			DMSG(vdcp, 1, "cancelling entry idx=%x\n", b_idx);
3954 			cancelled++;
3955 
3956 			/*
3957 			 * All requests have already been cleared from the
3958 			 * local descriptor ring and the LDC channel has been
3959 			 * reset so we will never get any reply for these
3960 			 * requests. Now we just have to notify threads waiting
3961 			 * for replies that the request has failed.
3962 			 */
3963 			switch (ldep->cb_type) {
3964 			case CB_SYNC:
3965 				ASSERT(vdcp->sync_op_pending);
3966 				vdcp->sync_op_status = EIO;
3967 				vdcp->sync_op_pending = B_FALSE;
3968 				cv_signal(&vdcp->sync_pending_cv);
3969 				break;
3970 
3971 			case CB_STRATEGY:
3972 				bufp = ldep->cb_arg;
3973 				ASSERT(bufp != NULL);
3974 				bufp->b_resid = bufp->b_bcount;
3975 				VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
3976 				VD_KSTAT_RUNQ_EXIT(vdcp);
3977 				DTRACE_IO1(done, buf_t *, bufp);
3978 				bioerror(bufp, EIO);
3979 				biodone(bufp);
3980 				break;
3981 
3982 			default:
3983 				ASSERT(0);
3984 			}
3985 
3986 		}
3987 
3988 		/* get the next element to cancel */
3989 		if (++b_idx >= vdcp->local_dring_backup_len)
3990 			b_idx = 0;
3991 	}
3992 
3993 	/* all done - now clear up pending dring copy */
3994 	dring_size = vdcp->local_dring_backup_len *
3995 	    sizeof (vdcp->local_dring_backup[0]);
3996 
3997 	(void) kmem_free(vdcp->local_dring_backup, dring_size);
3998 
3999 	vdcp->local_dring_backup = NULL;
4000 
4001 	DTRACE_PROBE2(cancelled, int, cancelled, vdc_t *, vdcp);
4002 }
4003 
4004 /*
4005  * Function:
4006  *	vdc_connection_timeout
4007  *
4008  * Description:
4009  *	This function is invoked if the timeout set to establish the connection
4010  *	with vds expires. This will happen if we spend too much time in the
4011  *	VDC_STATE_INIT_WAITING or VDC_STATE_NEGOTIATE states. Then we will
4012  *	cancel any pending request and mark them as failed.
4013  *
4014  *	If the timeout does not expire, it will be cancelled when we reach the
4015  *	VDC_STATE_HANDLE_PENDING or VDC_STATE_RESETTING state. This function can
4016  *	be invoked while we are in the VDC_STATE_HANDLE_PENDING or
4017  *	VDC_STATE_RESETTING state in which case we do nothing because the
4018  *	timeout is being cancelled.
4019  *
4020  * Arguments:
4021  *	arg	- argument of the timeout function actually a soft state
4022  *		  pointer for the instance of the device driver.
4023  *
4024  * Return Code:
4025  *	None
4026  */
4027 void
4028 vdc_connection_timeout(void *arg)
4029 {
4030 	vdc_t 		*vdcp = (vdc_t *)arg;
4031 
4032 	mutex_enter(&vdcp->lock);
4033 
4034 	if (vdcp->state == VDC_STATE_HANDLE_PENDING ||
4035 	    vdcp->state == VDC_STATE_DETACH) {
4036 		/*
4037 		 * The connection has just been re-established or
4038 		 * we are detaching.
4039 		 */
4040 		vdcp->ctimeout_reached = B_FALSE;
4041 		mutex_exit(&vdcp->lock);
4042 		return;
4043 	}
4044 
4045 	vdcp->ctimeout_reached = B_TRUE;
4046 
4047 	/* notify requests waiting for sending */
4048 	cv_broadcast(&vdcp->running_cv);
4049 
4050 	/* cancel requests waiting for a result */
4051 	vdc_cancel_backup_dring(vdcp);
4052 
4053 	mutex_exit(&vdcp->lock);
4054 
4055 	cmn_err(CE_NOTE, "[%d] connection to service domain timeout",
4056 	    vdcp->instance);
4057 }
4058 
4059 /*
4060  * Function:
4061  *	vdc_backup_local_dring()
4062  *
4063  * Description:
4064  *	Backup the current dring in the event of a reset. The Dring
4065  *	transactions will be resubmitted to the server when the
4066  *	connection is restored.
4067  *
4068  * Arguments:
4069  *	vdcp	- soft state pointer for this instance of the device driver.
4070  *
4071  * Return Code:
4072  *	NONE
4073  */
4074 static void
4075 vdc_backup_local_dring(vdc_t *vdcp)
4076 {
4077 	int dring_size;
4078 
4079 	ASSERT(MUTEX_HELD(&vdcp->lock));
4080 	ASSERT(vdcp->state == VDC_STATE_RESETTING);
4081 
4082 	/*
4083 	 * If the backup dring is stil around, it means
4084 	 * that the last restore did not complete. However,
4085 	 * since we never got back into the running state,
4086 	 * the backup copy we have is still valid.
4087 	 */
4088 	if (vdcp->local_dring_backup != NULL) {
4089 		DMSG(vdcp, 1, "reusing local descriptor ring backup "
4090 		    "(len=%d, tail=%d)\n", vdcp->local_dring_backup_len,
4091 		    vdcp->local_dring_backup_tail);
4092 		return;
4093 	}
4094 
4095 	/*
4096 	 * The backup dring can be NULL and the local dring may not be
4097 	 * initialized. This can happen if we had a reset while establishing
4098 	 * a new connection but after the connection has timed out. In that
4099 	 * case the backup dring is NULL because the requests have been
4100 	 * cancelled and the request occured before the local dring is
4101 	 * initialized.
4102 	 */
4103 	if (!(vdcp->initialized & VDC_DRING_LOCAL))
4104 		return;
4105 
4106 	DMSG(vdcp, 1, "backing up the local descriptor ring (len=%d, "
4107 	    "tail=%d)\n", vdcp->dring_len, vdcp->dring_curr_idx);
4108 
4109 	dring_size = vdcp->dring_len * sizeof (vdcp->local_dring[0]);
4110 
4111 	vdcp->local_dring_backup = kmem_alloc(dring_size, KM_SLEEP);
4112 	bcopy(vdcp->local_dring, vdcp->local_dring_backup, dring_size);
4113 
4114 	vdcp->local_dring_backup_tail = vdcp->dring_curr_idx;
4115 	vdcp->local_dring_backup_len = vdcp->dring_len;
4116 }
4117 
4118 static void
4119 vdc_switch_server(vdc_t *vdcp)
4120 {
4121 	int		rv;
4122 	vdc_server_t 	*curr_server, *new_server;
4123 
4124 	ASSERT(MUTEX_HELD(&vdcp->lock));
4125 
4126 	/* if there is only one server return back */
4127 	if (vdcp->num_servers == 1) {
4128 		return;
4129 	}
4130 
4131 	/* Get current and next server */
4132 	curr_server = vdcp->curr_server;
4133 	new_server =
4134 	    (curr_server->next) ? curr_server->next : vdcp->server_list;
4135 	ASSERT(curr_server != new_server);
4136 
4137 	/* bring current server's channel down */
4138 	rv = ldc_down(curr_server->ldc_handle);
4139 	if (rv) {
4140 		DMSG(vdcp, 0, "[%d] Cannot bring channel down, port %ld\n",
4141 		    vdcp->instance, curr_server->id);
4142 		return;
4143 	}
4144 
4145 	/* switch the server */
4146 	vdcp->curr_server = new_server;
4147 
4148 	cmn_err(CE_NOTE, "Successfully failed over from VDS on port@%ld to "
4149 	    "VDS on port@%ld.\n", curr_server->id, new_server->id);
4150 	DMSG(vdcp, 0, "[%d] Switched to next vdisk server, port@%ld, ldc@%ld\n",
4151 	    vdcp->instance, vdcp->curr_server->id, vdcp->curr_server->ldc_id);
4152 }
4153 
4154 /* -------------------------------------------------------------------------- */
4155 
4156 /*
4157  * The following functions process the incoming messages from vds
4158  */
4159 
4160 /*
4161  * Function:
4162  *      vdc_process_msg_thread()
4163  *
4164  * Description:
4165  *
4166  *	Main VDC message processing thread. Each vDisk instance
4167  * 	consists of a copy of this thread. This thread triggers
4168  * 	all the handshakes and data exchange with the server. It
4169  * 	also handles all channel resets
4170  *
4171  * Arguments:
4172  *      vdc     - soft state pointer for this instance of the device driver.
4173  *
4174  * Return Code:
4175  *      None
4176  */
4177 static void
4178 vdc_process_msg_thread(vdc_t *vdcp)
4179 {
4180 	int		status;
4181 	int		ctimeout;
4182 	timeout_id_t	tmid = 0;
4183 	clock_t		ldcup_timeout = 0;
4184 
4185 	mutex_enter(&vdcp->lock);
4186 
4187 	for (;;) {
4188 
4189 #define	Q(_s)	(vdcp->state == _s) ? #_s :
4190 		DMSG(vdcp, 3, "state = %d (%s)\n", vdcp->state,
4191 		    Q(VDC_STATE_INIT)
4192 		    Q(VDC_STATE_INIT_WAITING)
4193 		    Q(VDC_STATE_NEGOTIATE)
4194 		    Q(VDC_STATE_HANDLE_PENDING)
4195 		    Q(VDC_STATE_RUNNING)
4196 		    Q(VDC_STATE_RESETTING)
4197 		    Q(VDC_STATE_DETACH)
4198 		    "UNKNOWN");
4199 
4200 		switch (vdcp->state) {
4201 		case VDC_STATE_INIT:
4202 
4203 			/*
4204 			 * If requested, start a timeout to check if the
4205 			 * connection with vds is established in the
4206 			 * specified delay. If the timeout expires, we
4207 			 * will cancel any pending request.
4208 			 *
4209 			 * If some reset have occurred while establishing
4210 			 * the connection, we already have a timeout armed
4211 			 * and in that case we don't need to arm a new one.
4212 			 *
4213 			 * The same rule applies when there are multiple vds'.
4214 			 * If either a connection cannot be established or
4215 			 * the handshake times out, the connection thread will
4216 			 * try another server. The 'ctimeout' will report
4217 			 * back an error after it expires irrespective of
4218 			 * whether the vdisk is trying to connect to just
4219 			 * one or multiple servers.
4220 			 */
4221 			ctimeout = (vdc_timeout != 0)?
4222 			    vdc_timeout : vdcp->curr_server->ctimeout;
4223 
4224 			if (ctimeout != 0 && tmid == 0) {
4225 				tmid = timeout(vdc_connection_timeout, vdcp,
4226 				    ctimeout * drv_usectohz(MICROSEC));
4227 			}
4228 
4229 			/* Check if we are re-initializing repeatedly */
4230 			if (vdcp->hshake_cnt > vdc_hshake_retries &&
4231 			    vdcp->lifecycle != VDC_LC_ONLINE) {
4232 
4233 				DMSG(vdcp, 0, "[%d] too many handshakes,cnt=%d",
4234 				    vdcp->instance, vdcp->hshake_cnt);
4235 				cmn_err(CE_NOTE, "[%d] disk access failed.\n",
4236 				    vdcp->instance);
4237 				vdcp->state = VDC_STATE_DETACH;
4238 				break;
4239 			}
4240 
4241 			/* Switch to STATE_DETACH if drv is detaching */
4242 			if (vdcp->lifecycle == VDC_LC_DETACHING) {
4243 				vdcp->state = VDC_STATE_DETACH;
4244 				break;
4245 			}
4246 
4247 			/* Switch server */
4248 			if (vdcp->hshake_cnt > 0)
4249 				vdc_switch_server(vdcp);
4250 			vdcp->hshake_cnt++;
4251 
4252 			/* Bring up connection with vds via LDC */
4253 			status = vdc_start_ldc_connection(vdcp);
4254 			if (status != EINVAL) {
4255 				vdcp->state = VDC_STATE_INIT_WAITING;
4256 			}
4257 			break;
4258 
4259 		case VDC_STATE_INIT_WAITING:
4260 
4261 			/* if channel is UP, start negotiation */
4262 			if (vdcp->curr_server->ldc_state == LDC_UP) {
4263 				vdcp->state = VDC_STATE_NEGOTIATE;
4264 				break;
4265 			}
4266 
4267 			/* check if only one server exists */
4268 			if (vdcp->num_servers == 1) {
4269 				cv_wait(&vdcp->initwait_cv, &vdcp->lock);
4270 			} else {
4271 				/*
4272 				 * wait for LDC_UP, if it times out, switch
4273 				 * to another server.
4274 				 */
4275 				ldcup_timeout = ddi_get_lbolt() +
4276 				    (vdc_ldcup_timeout *
4277 				    drv_usectohz(MICROSEC));
4278 				status = cv_timedwait(&vdcp->initwait_cv,
4279 				    &vdcp->lock, ldcup_timeout);
4280 				if (status == -1 &&
4281 				    vdcp->state == VDC_STATE_INIT_WAITING &&
4282 				    vdcp->curr_server->ldc_state != LDC_UP) {
4283 					/* timed out & still waiting */
4284 					vdcp->state = VDC_STATE_INIT;
4285 					break;
4286 				}
4287 			}
4288 
4289 			if (vdcp->state != VDC_STATE_INIT_WAITING) {
4290 				DMSG(vdcp, 0,
4291 				    "state moved to %d out from under us...\n",
4292 				    vdcp->state);
4293 			}
4294 			break;
4295 
4296 		case VDC_STATE_NEGOTIATE:
4297 			switch (status = vdc_ver_negotiation(vdcp)) {
4298 			case 0:
4299 				break;
4300 			default:
4301 				DMSG(vdcp, 0, "ver negotiate failed (%d)..\n",
4302 				    status);
4303 				goto reset;
4304 			}
4305 
4306 			switch (status = vdc_attr_negotiation(vdcp)) {
4307 			case 0:
4308 				break;
4309 			default:
4310 				DMSG(vdcp, 0, "attr negotiate failed (%d)..\n",
4311 				    status);
4312 				goto reset;
4313 			}
4314 
4315 			switch (status = vdc_dring_negotiation(vdcp)) {
4316 			case 0:
4317 				break;
4318 			default:
4319 				DMSG(vdcp, 0, "dring negotiate failed (%d)..\n",
4320 				    status);
4321 				goto reset;
4322 			}
4323 
4324 			switch (status = vdc_rdx_exchange(vdcp)) {
4325 			case 0:
4326 				vdcp->state = VDC_STATE_HANDLE_PENDING;
4327 				goto done;
4328 			default:
4329 				DMSG(vdcp, 0, "RDX xchg failed ..(%d)\n",
4330 				    status);
4331 				goto reset;
4332 			}
4333 reset:
4334 			DMSG(vdcp, 0, "negotiation failed: resetting (%d)\n",
4335 			    status);
4336 			vdcp->state = VDC_STATE_RESETTING;
4337 			vdcp->self_reset = B_TRUE;
4338 done:
4339 			DMSG(vdcp, 0, "negotiation complete (state=0x%x)...\n",
4340 			    vdcp->state);
4341 			break;
4342 
4343 		case VDC_STATE_HANDLE_PENDING:
4344 
4345 			if (vdcp->ctimeout_reached) {
4346 				/*
4347 				 * The connection timeout had been reached so
4348 				 * pending requests have been cancelled. Now
4349 				 * that the connection is back we can reset
4350 				 * the timeout.
4351 				 */
4352 				ASSERT(vdcp->local_dring_backup == NULL);
4353 				ASSERT(tmid != 0);
4354 				tmid = 0;
4355 				vdcp->ctimeout_reached = B_FALSE;
4356 				vdcp->state = VDC_STATE_RUNNING;
4357 				DMSG(vdcp, 0, "[%d] connection to service "
4358 				    "domain is up", vdcp->instance);
4359 				break;
4360 			}
4361 
4362 			mutex_exit(&vdcp->lock);
4363 			if (tmid != 0) {
4364 				(void) untimeout(tmid);
4365 				tmid = 0;
4366 			}
4367 			status = vdc_resubmit_backup_dring(vdcp);
4368 			mutex_enter(&vdcp->lock);
4369 
4370 			if (status)
4371 				vdcp->state = VDC_STATE_RESETTING;
4372 			else
4373 				vdcp->state = VDC_STATE_RUNNING;
4374 
4375 			break;
4376 
4377 		/* enter running state */
4378 		case VDC_STATE_RUNNING:
4379 			/*
4380 			 * Signal anyone waiting for the connection
4381 			 * to come on line.
4382 			 */
4383 			vdcp->hshake_cnt = 0;
4384 			cv_broadcast(&vdcp->running_cv);
4385 
4386 			/* failfast has to been checked after reset */
4387 			cv_signal(&vdcp->failfast_cv);
4388 
4389 			/* ownership is lost during reset */
4390 			if (vdcp->ownership & VDC_OWNERSHIP_WANTED)
4391 				vdcp->ownership |= VDC_OWNERSHIP_RESET;
4392 			cv_signal(&vdcp->ownership_cv);
4393 
4394 			mutex_exit(&vdcp->lock);
4395 
4396 			for (;;) {
4397 				vio_msg_t msg;
4398 				status = vdc_wait_for_response(vdcp, &msg);
4399 				if (status) break;
4400 
4401 				DMSG(vdcp, 1, "[%d] new pkt(s) available\n",
4402 				    vdcp->instance);
4403 				status = vdc_process_data_msg(vdcp, &msg);
4404 				if (status) {
4405 					DMSG(vdcp, 1, "[%d] process_data_msg "
4406 					    "returned err=%d\n", vdcp->instance,
4407 					    status);
4408 					break;
4409 				}
4410 
4411 			}
4412 
4413 			mutex_enter(&vdcp->lock);
4414 
4415 			vdcp->state = VDC_STATE_RESETTING;
4416 			vdcp->self_reset = B_TRUE;
4417 			break;
4418 
4419 		case VDC_STATE_RESETTING:
4420 			/*
4421 			 * When we reach this state, we either come from the
4422 			 * VDC_STATE_RUNNING state and we can have pending
4423 			 * request but no timeout is armed; or we come from
4424 			 * the VDC_STATE_INIT_WAITING, VDC_NEGOTIATE or
4425 			 * VDC_HANDLE_PENDING state and there is no pending
4426 			 * request or pending requests have already been copied
4427 			 * into the backup dring. So we can safely keep the
4428 			 * connection timeout armed while we are in this state.
4429 			 */
4430 
4431 			DMSG(vdcp, 0, "Initiating channel reset "
4432 			    "(pending = %d)\n", (int)vdcp->threads_pending);
4433 
4434 			if (vdcp->self_reset) {
4435 				DMSG(vdcp, 0,
4436 				    "[%d] calling stop_ldc_connection.\n",
4437 				    vdcp->instance);
4438 				status = vdc_stop_ldc_connection(vdcp);
4439 				vdcp->self_reset = B_FALSE;
4440 			}
4441 
4442 			/*
4443 			 * Wait for all threads currently waiting
4444 			 * for a free dring entry to use.
4445 			 */
4446 			while (vdcp->threads_pending) {
4447 				cv_broadcast(&vdcp->membind_cv);
4448 				cv_broadcast(&vdcp->dring_free_cv);
4449 				mutex_exit(&vdcp->lock);
4450 				/* give the waiters enough time to wake up */
4451 				delay(vdc_hz_min_ldc_delay);
4452 				mutex_enter(&vdcp->lock);
4453 			}
4454 
4455 			ASSERT(vdcp->threads_pending == 0);
4456 
4457 			/* Sanity check that no thread is receiving */
4458 			ASSERT(vdcp->read_state != VDC_READ_WAITING);
4459 
4460 			vdcp->read_state = VDC_READ_IDLE;
4461 
4462 			vdc_backup_local_dring(vdcp);
4463 
4464 			/* cleanup the old d-ring */
4465 			vdc_destroy_descriptor_ring(vdcp);
4466 
4467 			/* go and start again */
4468 			vdcp->state = VDC_STATE_INIT;
4469 
4470 			break;
4471 
4472 		case VDC_STATE_DETACH:
4473 			DMSG(vdcp, 0, "[%d] Reset thread exit cleanup ..\n",
4474 			    vdcp->instance);
4475 
4476 			/* cancel any pending timeout */
4477 			mutex_exit(&vdcp->lock);
4478 			if (tmid != 0) {
4479 				(void) untimeout(tmid);
4480 				tmid = 0;
4481 			}
4482 			mutex_enter(&vdcp->lock);
4483 
4484 			/*
4485 			 * Signal anyone waiting for connection
4486 			 * to come online
4487 			 */
4488 			cv_broadcast(&vdcp->running_cv);
4489 
4490 			while (vdcp->sync_op_pending) {
4491 				cv_signal(&vdcp->sync_pending_cv);
4492 				cv_signal(&vdcp->sync_blocked_cv);
4493 				mutex_exit(&vdcp->lock);
4494 				/* give the waiters enough time to wake up */
4495 				delay(vdc_hz_min_ldc_delay);
4496 				mutex_enter(&vdcp->lock);
4497 			}
4498 
4499 			mutex_exit(&vdcp->lock);
4500 
4501 			DMSG(vdcp, 0, "[%d] Msg processing thread exiting ..\n",
4502 			    vdcp->instance);
4503 			thread_exit();
4504 			break;
4505 		}
4506 	}
4507 }
4508 
4509 
4510 /*
4511  * Function:
4512  *	vdc_process_data_msg()
4513  *
4514  * Description:
4515  *	This function is called by the message processing thread each time
4516  *	a message with a msgtype of VIO_TYPE_DATA is received. It will either
4517  *	be an ACK or NACK from vds[1] which vdc handles as follows.
4518  *		ACK	- wake up the waiting thread
4519  *		NACK	- resend any messages necessary
4520  *
4521  *	[1] Although the message format allows it, vds should not send a
4522  *	    VIO_SUBTYPE_INFO message to vdc asking it to read data; if for
4523  *	    some bizarre reason it does, vdc will reset the connection.
4524  *
4525  * Arguments:
4526  *	vdc	- soft state pointer for this instance of the device driver.
4527  *	msg	- the LDC message sent by vds
4528  *
4529  * Return Code:
4530  *	0	- Success.
4531  *	> 0	- error value returned by LDC
4532  */
4533 static int
4534 vdc_process_data_msg(vdc_t *vdcp, vio_msg_t *msg)
4535 {
4536 	int			status = 0;
4537 	vio_dring_msg_t		*dring_msg;
4538 	vdc_local_desc_t	*ldep = NULL;
4539 	int			start, end;
4540 	int			idx;
4541 	int			op;
4542 
4543 	dring_msg = (vio_dring_msg_t *)msg;
4544 
4545 	ASSERT(msg->tag.vio_msgtype == VIO_TYPE_DATA);
4546 	ASSERT(vdcp != NULL);
4547 
4548 	mutex_enter(&vdcp->lock);
4549 
4550 	/*
4551 	 * Check to see if the message has bogus data
4552 	 */
4553 	idx = start = dring_msg->start_idx;
4554 	end = dring_msg->end_idx;
4555 	if ((start >= vdcp->dring_len) ||
4556 	    (end >= vdcp->dring_len) || (end < -1)) {
4557 		/*
4558 		 * Update the I/O statistics to indicate that an error ocurred.
4559 		 * No need to update the wait/run queues as no specific read or
4560 		 * write request is being completed in response to this 'msg'.
4561 		 */
4562 		VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4563 		DMSG(vdcp, 0, "[%d] Bogus ACK data : start %d, end %d\n",
4564 		    vdcp->instance, start, end);
4565 		mutex_exit(&vdcp->lock);
4566 		return (EINVAL);
4567 	}
4568 
4569 	/*
4570 	 * Verify that the sequence number is what vdc expects.
4571 	 */
4572 	switch (vdc_verify_seq_num(vdcp, dring_msg)) {
4573 	case VDC_SEQ_NUM_TODO:
4574 		break;	/* keep processing this message */
4575 	case VDC_SEQ_NUM_SKIP:
4576 		mutex_exit(&vdcp->lock);
4577 		return (0);
4578 	case VDC_SEQ_NUM_INVALID:
4579 		/*
4580 		 * Update the I/O statistics to indicate that an error ocurred.
4581 		 * No need to update the wait/run queues as no specific read or
4582 		 * write request is being completed in response to this 'msg'.
4583 		 */
4584 		VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4585 		DMSG(vdcp, 0, "[%d] invalid seqno\n", vdcp->instance);
4586 		mutex_exit(&vdcp->lock);
4587 		return (ENXIO);
4588 	}
4589 
4590 	if (msg->tag.vio_subtype == VIO_SUBTYPE_NACK) {
4591 		/*
4592 		 * Update the I/O statistics to indicate that an error ocurred.
4593 		 *
4594 		 * We need to update the run queue if a read or write request
4595 		 * is being NACKed - otherwise there will appear to be an
4596 		 * indefinite outstanding request and statistics reported by
4597 		 * iostat(1M) will be incorrect. The transaction will be
4598 		 * resubmitted from the backup DRing following the reset
4599 		 * and the wait/run queues will be entered again.
4600 		 */
4601 		ldep = &vdcp->local_dring[idx];
4602 		op = ldep->operation;
4603 		if ((op == VD_OP_BREAD) || (op == VD_OP_BWRITE)) {
4604 			DTRACE_IO1(done, buf_t *, ldep->cb_arg);
4605 			VD_KSTAT_RUNQ_EXIT(vdcp);
4606 		}
4607 		VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4608 		VDC_DUMP_DRING_MSG(dring_msg);
4609 		DMSG(vdcp, 0, "[%d] DATA NACK\n", vdcp->instance);
4610 		mutex_exit(&vdcp->lock);
4611 		return (EIO);
4612 
4613 	} else if (msg->tag.vio_subtype == VIO_SUBTYPE_INFO) {
4614 		/*
4615 		 * Update the I/O statistics to indicate that an error occurred.
4616 		 * No need to update the wait/run queues as no specific read or
4617 		 * write request is being completed in response to this 'msg'.
4618 		 */
4619 		VD_UPDATE_ERR_STATS(vdcp, vd_protoerrs);
4620 		mutex_exit(&vdcp->lock);
4621 		return (EPROTO);
4622 	}
4623 
4624 	DMSG(vdcp, 1, ": start %d end %d\n", start, end);
4625 	ASSERT(start == end);
4626 
4627 	ldep = &vdcp->local_dring[idx];
4628 
4629 	DMSG(vdcp, 1, ": state 0x%x - cb_type 0x%x\n",
4630 	    ldep->dep->hdr.dstate, ldep->cb_type);
4631 
4632 	if (ldep->dep->hdr.dstate == VIO_DESC_DONE) {
4633 		struct buf *bufp;
4634 
4635 		switch (ldep->cb_type) {
4636 		case CB_SYNC:
4637 			ASSERT(vdcp->sync_op_pending);
4638 
4639 			status = vdc_depopulate_descriptor(vdcp, idx);
4640 			vdcp->sync_op_status = status;
4641 			vdcp->sync_op_pending = B_FALSE;
4642 			cv_signal(&vdcp->sync_pending_cv);
4643 			break;
4644 
4645 		case CB_STRATEGY:
4646 			bufp = ldep->cb_arg;
4647 			ASSERT(bufp != NULL);
4648 			bufp->b_resid =
4649 			    bufp->b_bcount - ldep->dep->payload.nbytes;
4650 			status = ldep->dep->payload.status; /* Future:ntoh */
4651 			if (status != 0) {
4652 				DMSG(vdcp, 1, "strategy status=%d\n", status);
4653 				VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4654 				bioerror(bufp, status);
4655 			}
4656 
4657 			(void) vdc_depopulate_descriptor(vdcp, idx);
4658 
4659 			DMSG(vdcp, 1,
4660 			    "strategy complete req=%ld bytes resp=%ld bytes\n",
4661 			    bufp->b_bcount, ldep->dep->payload.nbytes);
4662 
4663 			if (status != 0 && vdcp->failfast_interval != 0) {
4664 				/*
4665 				 * The I/O has failed and failfast is enabled.
4666 				 * We need the failfast thread to check if the
4667 				 * failure is due to a reservation conflict.
4668 				 */
4669 				(void) vdc_failfast_io_queue(vdcp, bufp);
4670 			} else {
4671 				if (status == 0) {
4672 					op = (bufp->b_flags & B_READ) ?
4673 					    VD_OP_BREAD : VD_OP_BWRITE;
4674 					VD_UPDATE_IO_STATS(vdcp, op,
4675 					    ldep->dep->payload.nbytes);
4676 				}
4677 				VD_KSTAT_RUNQ_EXIT(vdcp);
4678 				DTRACE_IO1(done, buf_t *, bufp);
4679 				biodone(bufp);
4680 			}
4681 			break;
4682 
4683 		default:
4684 			ASSERT(0);
4685 		}
4686 	}
4687 
4688 	/* let the arrival signal propogate */
4689 	mutex_exit(&vdcp->lock);
4690 
4691 	/* probe gives the count of how many entries were processed */
4692 	DTRACE_PROBE2(processed, int, 1, vdc_t *, vdcp);
4693 
4694 	return (0);
4695 }
4696 
4697 
4698 /*
4699  * Function:
4700  *	vdc_handle_ver_msg()
4701  *
4702  * Description:
4703  *
4704  * Arguments:
4705  *	vdc	- soft state pointer for this instance of the device driver.
4706  *	ver_msg	- LDC message sent by vDisk server
4707  *
4708  * Return Code:
4709  *	0	- Success
4710  */
4711 static int
4712 vdc_handle_ver_msg(vdc_t *vdc, vio_ver_msg_t *ver_msg)
4713 {
4714 	int status = 0;
4715 
4716 	ASSERT(vdc != NULL);
4717 	ASSERT(mutex_owned(&vdc->lock));
4718 
4719 	if (ver_msg->tag.vio_subtype_env != VIO_VER_INFO) {
4720 		return (EPROTO);
4721 	}
4722 
4723 	if (ver_msg->dev_class != VDEV_DISK_SERVER) {
4724 		return (EINVAL);
4725 	}
4726 
4727 	switch (ver_msg->tag.vio_subtype) {
4728 	case VIO_SUBTYPE_ACK:
4729 		/*
4730 		 * We check to see if the version returned is indeed supported
4731 		 * (The server may have also adjusted the minor number downwards
4732 		 * and if so 'ver_msg' will contain the actual version agreed)
4733 		 */
4734 		if (vdc_is_supported_version(ver_msg)) {
4735 			vdc->ver.major = ver_msg->ver_major;
4736 			vdc->ver.minor = ver_msg->ver_minor;
4737 			ASSERT(vdc->ver.major > 0);
4738 		} else {
4739 			status = EPROTO;
4740 		}
4741 		break;
4742 
4743 	case VIO_SUBTYPE_NACK:
4744 		/*
4745 		 * call vdc_is_supported_version() which will return the next
4746 		 * supported version (if any) in 'ver_msg'
4747 		 */
4748 		(void) vdc_is_supported_version(ver_msg);
4749 		if (ver_msg->ver_major > 0) {
4750 			size_t len = sizeof (*ver_msg);
4751 
4752 			ASSERT(vdc->ver.major > 0);
4753 
4754 			/* reset the necessary fields and resend */
4755 			ver_msg->tag.vio_subtype = VIO_SUBTYPE_INFO;
4756 			ver_msg->dev_class = VDEV_DISK;
4757 
4758 			status = vdc_send(vdc, (caddr_t)ver_msg, &len);
4759 			DMSG(vdc, 0, "[%d] Resend VER info (LDC status = %d)\n",
4760 			    vdc->instance, status);
4761 			if (len != sizeof (*ver_msg))
4762 				status = EBADMSG;
4763 		} else {
4764 			DMSG(vdc, 0, "[%d] No common version with vDisk server",
4765 			    vdc->instance);
4766 			status = ENOTSUP;
4767 		}
4768 
4769 		break;
4770 	case VIO_SUBTYPE_INFO:
4771 		/*
4772 		 * Handle the case where vds starts handshake
4773 		 * (for now only vdc is the instigator)
4774 		 */
4775 		status = ENOTSUP;
4776 		break;
4777 
4778 	default:
4779 		status = EINVAL;
4780 		break;
4781 	}
4782 
4783 	return (status);
4784 }
4785 
4786 /*
4787  * Function:
4788  *	vdc_handle_attr_msg()
4789  *
4790  * Description:
4791  *
4792  * Arguments:
4793  *	vdc	- soft state pointer for this instance of the device driver.
4794  *	attr_msg	- LDC message sent by vDisk server
4795  *
4796  * Return Code:
4797  *	0	- Success
4798  */
4799 static int
4800 vdc_handle_attr_msg(vdc_t *vdc, vd_attr_msg_t *attr_msg)
4801 {
4802 	int status = 0;
4803 
4804 	ASSERT(vdc != NULL);
4805 	ASSERT(mutex_owned(&vdc->lock));
4806 
4807 	if (attr_msg->tag.vio_subtype_env != VIO_ATTR_INFO) {
4808 		return (EPROTO);
4809 	}
4810 
4811 	switch (attr_msg->tag.vio_subtype) {
4812 	case VIO_SUBTYPE_ACK:
4813 		/*
4814 		 * We now verify the attributes sent by vds.
4815 		 */
4816 		if (attr_msg->vdisk_size == 0) {
4817 			DMSG(vdc, 0, "[%d] Invalid disk size from vds",
4818 			    vdc->instance);
4819 			status = EINVAL;
4820 			break;
4821 		}
4822 
4823 		if (attr_msg->max_xfer_sz == 0) {
4824 			DMSG(vdc, 0, "[%d] Invalid transfer size from vds",
4825 			    vdc->instance);
4826 			status = EINVAL;
4827 			break;
4828 		}
4829 
4830 		if (attr_msg->vdisk_size == VD_SIZE_UNKNOWN) {
4831 			DMSG(vdc, 0, "[%d] Unknown disk size from vds",
4832 			    vdc->instance);
4833 			attr_msg->vdisk_size = 0;
4834 		}
4835 
4836 		/*
4837 		 * If the disk size is already set check that it hasn't changed.
4838 		 */
4839 		if ((vdc->vdisk_size != 0) && (attr_msg->vdisk_size != 0) &&
4840 		    (vdc->vdisk_size != attr_msg->vdisk_size)) {
4841 			DMSG(vdc, 0, "[%d] Different disk size from vds "
4842 			    "(old=0x%lx - new=0x%lx", vdc->instance,
4843 			    vdc->vdisk_size, attr_msg->vdisk_size)
4844 			status = EINVAL;
4845 			break;
4846 		}
4847 
4848 		vdc->vdisk_size = attr_msg->vdisk_size;
4849 		vdc->vdisk_type = attr_msg->vdisk_type;
4850 		vdc->operations = attr_msg->operations;
4851 		if (vio_ver_is_supported(vdc->ver, 1, 1))
4852 			vdc->vdisk_media = attr_msg->vdisk_media;
4853 		else
4854 			vdc->vdisk_media = 0;
4855 
4856 		DMSG(vdc, 0, "[%d] max_xfer_sz: sent %lx acked %lx\n",
4857 		    vdc->instance, vdc->max_xfer_sz, attr_msg->max_xfer_sz);
4858 		DMSG(vdc, 0, "[%d] vdisk_block_size: sent %lx acked %x\n",
4859 		    vdc->instance, vdc->block_size,
4860 		    attr_msg->vdisk_block_size);
4861 
4862 		/*
4863 		 * We don't know at compile time what the vDisk server will
4864 		 * think are good values but we apply a large (arbitrary)
4865 		 * upper bound to prevent memory exhaustion in vdc if it was
4866 		 * allocating a DRing based of huge values sent by the server.
4867 		 * We probably will never exceed this except if the message
4868 		 * was garbage.
4869 		 */
4870 		if ((attr_msg->max_xfer_sz * attr_msg->vdisk_block_size) <=
4871 		    (PAGESIZE * DEV_BSIZE)) {
4872 			vdc->max_xfer_sz = attr_msg->max_xfer_sz;
4873 			vdc->block_size = attr_msg->vdisk_block_size;
4874 		} else {
4875 			DMSG(vdc, 0, "[%d] vds block transfer size too big;"
4876 			    " using max supported by vdc", vdc->instance);
4877 		}
4878 
4879 		if ((attr_msg->xfer_mode != VIO_DRING_MODE_V1_0) ||
4880 		    (attr_msg->vdisk_size > INT64_MAX) ||
4881 		    (attr_msg->operations == 0) ||
4882 		    (attr_msg->vdisk_type > VD_DISK_TYPE_DISK)) {
4883 			DMSG(vdc, 0, "[%d] Invalid attributes from vds",
4884 			    vdc->instance);
4885 			status = EINVAL;
4886 			break;
4887 		}
4888 
4889 		/*
4890 		 * Now that we have received all attributes we can create a
4891 		 * fake geometry for the disk.
4892 		 */
4893 		vdc_create_fake_geometry(vdc);
4894 		break;
4895 
4896 	case VIO_SUBTYPE_NACK:
4897 		/*
4898 		 * vds could not handle the attributes we sent so we
4899 		 * stop negotiating.
4900 		 */
4901 		status = EPROTO;
4902 		break;
4903 
4904 	case VIO_SUBTYPE_INFO:
4905 		/*
4906 		 * Handle the case where vds starts the handshake
4907 		 * (for now; vdc is the only supported instigatior)
4908 		 */
4909 		status = ENOTSUP;
4910 		break;
4911 
4912 	default:
4913 		status = ENOTSUP;
4914 		break;
4915 	}
4916 
4917 	return (status);
4918 }
4919 
4920 /*
4921  * Function:
4922  *	vdc_handle_dring_reg_msg()
4923  *
4924  * Description:
4925  *
4926  * Arguments:
4927  *	vdc		- soft state pointer for this instance of the driver.
4928  *	dring_msg	- LDC message sent by vDisk server
4929  *
4930  * Return Code:
4931  *	0	- Success
4932  */
4933 static int
4934 vdc_handle_dring_reg_msg(vdc_t *vdc, vio_dring_reg_msg_t *dring_msg)
4935 {
4936 	int		status = 0;
4937 
4938 	ASSERT(vdc != NULL);
4939 	ASSERT(mutex_owned(&vdc->lock));
4940 
4941 	if (dring_msg->tag.vio_subtype_env != VIO_DRING_REG) {
4942 		return (EPROTO);
4943 	}
4944 
4945 	switch (dring_msg->tag.vio_subtype) {
4946 	case VIO_SUBTYPE_ACK:
4947 		/* save the received dring_ident */
4948 		vdc->dring_ident = dring_msg->dring_ident;
4949 		DMSG(vdc, 0, "[%d] Received dring ident=0x%lx\n",
4950 		    vdc->instance, vdc->dring_ident);
4951 		break;
4952 
4953 	case VIO_SUBTYPE_NACK:
4954 		/*
4955 		 * vds could not handle the DRing info we sent so we
4956 		 * stop negotiating.
4957 		 */
4958 		DMSG(vdc, 0, "[%d] server could not register DRing\n",
4959 		    vdc->instance);
4960 		status = EPROTO;
4961 		break;
4962 
4963 	case VIO_SUBTYPE_INFO:
4964 		/*
4965 		 * Handle the case where vds starts handshake
4966 		 * (for now only vdc is the instigatior)
4967 		 */
4968 		status = ENOTSUP;
4969 		break;
4970 	default:
4971 		status = ENOTSUP;
4972 	}
4973 
4974 	return (status);
4975 }
4976 
4977 /*
4978  * Function:
4979  *	vdc_verify_seq_num()
4980  *
4981  * Description:
4982  *	This functions verifies that the sequence number sent back by the vDisk
4983  *	server with the latest message is what is expected (i.e. it is greater
4984  *	than the last seq num sent by the vDisk server and less than or equal
4985  *	to the last seq num generated by vdc).
4986  *
4987  *	It then checks the request ID to see if any requests need processing
4988  *	in the DRing.
4989  *
4990  * Arguments:
4991  *	vdc		- soft state pointer for this instance of the driver.
4992  *	dring_msg	- pointer to the LDC message sent by vds
4993  *
4994  * Return Code:
4995  *	VDC_SEQ_NUM_TODO	- Message needs to be processed
4996  *	VDC_SEQ_NUM_SKIP	- Message has already been processed
4997  *	VDC_SEQ_NUM_INVALID	- The seq numbers are so out of sync,
4998  *				  vdc cannot deal with them
4999  */
5000 static int
5001 vdc_verify_seq_num(vdc_t *vdc, vio_dring_msg_t *dring_msg)
5002 {
5003 	ASSERT(vdc != NULL);
5004 	ASSERT(dring_msg != NULL);
5005 	ASSERT(mutex_owned(&vdc->lock));
5006 
5007 	/*
5008 	 * Check to see if the messages were responded to in the correct
5009 	 * order by vds.
5010 	 */
5011 	if ((dring_msg->seq_num <= vdc->seq_num_reply) ||
5012 	    (dring_msg->seq_num > vdc->seq_num)) {
5013 		DMSG(vdc, 0, "?[%d] Bogus sequence_number %lu: "
5014 		    "%lu > expected <= %lu (last proc req %lu sent %lu)\n",
5015 		    vdc->instance, dring_msg->seq_num,
5016 		    vdc->seq_num_reply, vdc->seq_num,
5017 		    vdc->req_id_proc, vdc->req_id);
5018 		return (VDC_SEQ_NUM_INVALID);
5019 	}
5020 	vdc->seq_num_reply = dring_msg->seq_num;
5021 
5022 	if (vdc->req_id_proc < vdc->req_id)
5023 		return (VDC_SEQ_NUM_TODO);
5024 	else
5025 		return (VDC_SEQ_NUM_SKIP);
5026 }
5027 
5028 
5029 /*
5030  * Function:
5031  *	vdc_is_supported_version()
5032  *
5033  * Description:
5034  *	This routine checks if the major/minor version numbers specified in
5035  *	'ver_msg' are supported. If not it finds the next version that is
5036  *	in the supported version list 'vdc_version[]' and sets the fields in
5037  *	'ver_msg' to those values
5038  *
5039  * Arguments:
5040  *	ver_msg	- LDC message sent by vDisk server
5041  *
5042  * Return Code:
5043  *	B_TRUE	- Success
5044  *	B_FALSE	- Version not supported
5045  */
5046 static boolean_t
5047 vdc_is_supported_version(vio_ver_msg_t *ver_msg)
5048 {
5049 	int vdc_num_versions = sizeof (vdc_version) / sizeof (vdc_version[0]);
5050 
5051 	for (int i = 0; i < vdc_num_versions; i++) {
5052 		ASSERT(vdc_version[i].major > 0);
5053 		ASSERT((i == 0) ||
5054 		    (vdc_version[i].major < vdc_version[i-1].major));
5055 
5056 		/*
5057 		 * If the major versions match, adjust the minor version, if
5058 		 * necessary, down to the highest value supported by this
5059 		 * client. The server should support all minor versions lower
5060 		 * than the value it sent
5061 		 */
5062 		if (ver_msg->ver_major == vdc_version[i].major) {
5063 			if (ver_msg->ver_minor > vdc_version[i].minor) {
5064 				DMSGX(0,
5065 				    "Adjusting minor version from %u to %u",
5066 				    ver_msg->ver_minor, vdc_version[i].minor);
5067 				ver_msg->ver_minor = vdc_version[i].minor;
5068 			}
5069 			return (B_TRUE);
5070 		}
5071 
5072 		/*
5073 		 * If the message contains a higher major version number, set
5074 		 * the message's major/minor versions to the current values
5075 		 * and return false, so this message will get resent with
5076 		 * these values, and the server will potentially try again
5077 		 * with the same or a lower version
5078 		 */
5079 		if (ver_msg->ver_major > vdc_version[i].major) {
5080 			ver_msg->ver_major = vdc_version[i].major;
5081 			ver_msg->ver_minor = vdc_version[i].minor;
5082 			DMSGX(0, "Suggesting major/minor (0x%x/0x%x)\n",
5083 			    ver_msg->ver_major, ver_msg->ver_minor);
5084 
5085 			return (B_FALSE);
5086 		}
5087 
5088 		/*
5089 		 * Otherwise, the message's major version is less than the
5090 		 * current major version, so continue the loop to the next
5091 		 * (lower) supported version
5092 		 */
5093 	}
5094 
5095 	/*
5096 	 * No common version was found; "ground" the version pair in the
5097 	 * message to terminate negotiation
5098 	 */
5099 	ver_msg->ver_major = 0;
5100 	ver_msg->ver_minor = 0;
5101 
5102 	return (B_FALSE);
5103 }
5104 /* -------------------------------------------------------------------------- */
5105 
5106 /*
5107  * DKIO(7) support
5108  */
5109 
5110 typedef struct vdc_dk_arg {
5111 	struct dk_callback	dkc;
5112 	int			mode;
5113 	dev_t			dev;
5114 	vdc_t			*vdc;
5115 } vdc_dk_arg_t;
5116 
5117 /*
5118  * Function:
5119  * 	vdc_dkio_flush_cb()
5120  *
5121  * Description:
5122  *	This routine is a callback for DKIOCFLUSHWRITECACHE which can be called
5123  *	by kernel code.
5124  *
5125  * Arguments:
5126  *	arg	- a pointer to a vdc_dk_arg_t structure.
5127  */
5128 void
5129 vdc_dkio_flush_cb(void *arg)
5130 {
5131 	struct vdc_dk_arg	*dk_arg = (struct vdc_dk_arg *)arg;
5132 	struct dk_callback	*dkc = NULL;
5133 	vdc_t			*vdc = NULL;
5134 	int			rv;
5135 
5136 	if (dk_arg == NULL) {
5137 		cmn_err(CE_NOTE, "?[Unk] DKIOCFLUSHWRITECACHE arg is NULL\n");
5138 		return;
5139 	}
5140 	dkc = &dk_arg->dkc;
5141 	vdc = dk_arg->vdc;
5142 	ASSERT(vdc != NULL);
5143 
5144 	rv = vdc_do_sync_op(vdc, VD_OP_FLUSH, NULL, 0,
5145 	    VDCPART(dk_arg->dev), 0, CB_SYNC, 0, VIO_both_dir, B_TRUE);
5146 	if (rv != 0) {
5147 		DMSG(vdc, 0, "[%d] DKIOCFLUSHWRITECACHE failed %d : model %x\n",
5148 		    vdc->instance, rv,
5149 		    ddi_model_convert_from(dk_arg->mode & FMODELS));
5150 	}
5151 
5152 	/*
5153 	 * Trigger the call back to notify the caller the the ioctl call has
5154 	 * been completed.
5155 	 */
5156 	if ((dk_arg->mode & FKIOCTL) &&
5157 	    (dkc != NULL) &&
5158 	    (dkc->dkc_callback != NULL)) {
5159 		ASSERT(dkc->dkc_cookie != NULL);
5160 		(*dkc->dkc_callback)(dkc->dkc_cookie, rv);
5161 	}
5162 
5163 	/* Indicate that one less DKIO write flush is outstanding */
5164 	mutex_enter(&vdc->lock);
5165 	vdc->dkio_flush_pending--;
5166 	ASSERT(vdc->dkio_flush_pending >= 0);
5167 	mutex_exit(&vdc->lock);
5168 
5169 	/* free the mem that was allocated when the callback was dispatched */
5170 	kmem_free(arg, sizeof (vdc_dk_arg_t));
5171 }
5172 
5173 /*
5174  * Function:
5175  * 	vdc_dkio_gapart()
5176  *
5177  * Description:
5178  *	This function implements the DKIOCGAPART ioctl.
5179  *
5180  * Arguments:
5181  *	vdc	- soft state pointer
5182  *	arg	- a pointer to a dk_map[NDKMAP] or dk_map32[NDKMAP] structure
5183  *	flag	- ioctl flags
5184  */
5185 static int
5186 vdc_dkio_gapart(vdc_t *vdc, caddr_t arg, int flag)
5187 {
5188 	struct dk_geom *geom;
5189 	struct vtoc *vtoc;
5190 	union {
5191 		struct dk_map map[NDKMAP];
5192 		struct dk_map32 map32[NDKMAP];
5193 	} data;
5194 	int i, rv, size;
5195 
5196 	mutex_enter(&vdc->lock);
5197 
5198 	if ((rv = vdc_validate_geometry(vdc)) != 0) {
5199 		mutex_exit(&vdc->lock);
5200 		return (rv);
5201 	}
5202 
5203 	vtoc = vdc->vtoc;
5204 	geom = vdc->geom;
5205 
5206 	if (ddi_model_convert_from(flag & FMODELS) == DDI_MODEL_ILP32) {
5207 
5208 		for (i = 0; i < vtoc->v_nparts; i++) {
5209 			data.map32[i].dkl_cylno = vtoc->v_part[i].p_start /
5210 			    (geom->dkg_nhead * geom->dkg_nsect);
5211 			data.map32[i].dkl_nblk = vtoc->v_part[i].p_size;
5212 		}
5213 		size = NDKMAP * sizeof (struct dk_map32);
5214 
5215 	} else {
5216 
5217 		for (i = 0; i < vtoc->v_nparts; i++) {
5218 			data.map[i].dkl_cylno = vtoc->v_part[i].p_start /
5219 			    (geom->dkg_nhead * geom->dkg_nsect);
5220 			data.map[i].dkl_nblk = vtoc->v_part[i].p_size;
5221 		}
5222 		size = NDKMAP * sizeof (struct dk_map);
5223 
5224 	}
5225 
5226 	mutex_exit(&vdc->lock);
5227 
5228 	if (ddi_copyout(&data, arg, size, flag) != 0)
5229 		return (EFAULT);
5230 
5231 	return (0);
5232 }
5233 
5234 /*
5235  * Function:
5236  * 	vdc_dkio_partition()
5237  *
5238  * Description:
5239  *	This function implements the DKIOCPARTITION ioctl.
5240  *
5241  * Arguments:
5242  *	vdc	- soft state pointer
5243  *	arg	- a pointer to a struct partition64 structure
5244  *	flag	- ioctl flags
5245  */
5246 static int
5247 vdc_dkio_partition(vdc_t *vdc, caddr_t arg, int flag)
5248 {
5249 	struct partition64 p64;
5250 	efi_gpt_t *gpt;
5251 	efi_gpe_t *gpe;
5252 	vd_efi_dev_t edev;
5253 	uint_t partno;
5254 	int rv;
5255 
5256 	if (ddi_copyin(arg, &p64, sizeof (struct partition64), flag)) {
5257 		return (EFAULT);
5258 	}
5259 
5260 	VD_EFI_DEV_SET(edev, vdc, vd_process_efi_ioctl);
5261 
5262 	if ((rv = vd_efi_alloc_and_read(&edev, &gpt, &gpe)) != 0) {
5263 		return (rv);
5264 	}
5265 
5266 	partno = p64.p_partno;
5267 
5268 	if (partno >= gpt->efi_gpt_NumberOfPartitionEntries) {
5269 		vd_efi_free(&edev, gpt, gpe);
5270 		return (ESRCH);
5271 	}
5272 
5273 	bcopy(&gpe[partno].efi_gpe_PartitionTypeGUID, &p64.p_type,
5274 	    sizeof (struct uuid));
5275 	p64.p_start = gpe[partno].efi_gpe_StartingLBA;
5276 	p64.p_size = gpe[partno].efi_gpe_EndingLBA - p64.p_start + 1;
5277 
5278 	if (ddi_copyout(&p64, arg, sizeof (struct partition64), flag)) {
5279 		vd_efi_free(&edev, gpt, gpe);
5280 		return (EFAULT);
5281 	}
5282 
5283 	vd_efi_free(&edev, gpt, gpe);
5284 	return (0);
5285 }
5286 
5287 /*
5288  * Function:
5289  * 	vdc_dioctl_rwcmd()
5290  *
5291  * Description:
5292  *	This function implements the DIOCTL_RWCMD ioctl. This ioctl is used
5293  *	for DKC_DIRECT disks to read or write at an absolute disk offset.
5294  *
5295  * Arguments:
5296  *	dev	- device
5297  *	arg	- a pointer to a dadkio_rwcmd or dadkio_rwcmd32 structure
5298  *	flag	- ioctl flags
5299  */
5300 static int
5301 vdc_dioctl_rwcmd(dev_t dev, caddr_t arg, int flag)
5302 {
5303 	struct dadkio_rwcmd32 rwcmd32;
5304 	struct dadkio_rwcmd rwcmd;
5305 	struct iovec aiov;
5306 	struct uio auio;
5307 	int rw, status;
5308 	struct buf *buf;
5309 
5310 	if (ddi_model_convert_from(flag & FMODELS) == DDI_MODEL_ILP32) {
5311 		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd32,
5312 		    sizeof (struct dadkio_rwcmd32), flag)) {
5313 			return (EFAULT);
5314 		}
5315 		rwcmd.cmd = rwcmd32.cmd;
5316 		rwcmd.flags = rwcmd32.flags;
5317 		rwcmd.blkaddr = (daddr_t)rwcmd32.blkaddr;
5318 		rwcmd.buflen = rwcmd32.buflen;
5319 		rwcmd.bufaddr = (caddr_t)(uintptr_t)rwcmd32.bufaddr;
5320 	} else {
5321 		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd,
5322 		    sizeof (struct dadkio_rwcmd), flag)) {
5323 			return (EFAULT);
5324 		}
5325 	}
5326 
5327 	switch (rwcmd.cmd) {
5328 	case DADKIO_RWCMD_READ:
5329 		rw = B_READ;
5330 		break;
5331 	case DADKIO_RWCMD_WRITE:
5332 		rw = B_WRITE;
5333 		break;
5334 	default:
5335 		return (EINVAL);
5336 	}
5337 
5338 	bzero((caddr_t)&aiov, sizeof (struct iovec));
5339 	aiov.iov_base   = rwcmd.bufaddr;
5340 	aiov.iov_len    = rwcmd.buflen;
5341 
5342 	bzero((caddr_t)&auio, sizeof (struct uio));
5343 	auio.uio_iov    = &aiov;
5344 	auio.uio_iovcnt = 1;
5345 	auio.uio_loffset = rwcmd.blkaddr * DEV_BSIZE;
5346 	auio.uio_resid  = rwcmd.buflen;
5347 	auio.uio_segflg = flag & FKIOCTL ? UIO_SYSSPACE : UIO_USERSPACE;
5348 
5349 	buf = kmem_alloc(sizeof (buf_t), KM_SLEEP);
5350 	bioinit(buf);
5351 	/*
5352 	 * We use the private field of buf to specify that this is an
5353 	 * I/O using an absolute offset.
5354 	 */
5355 	buf->b_private = (void *)VD_SLICE_NONE;
5356 
5357 	status = physio(vdc_strategy, buf, dev, rw, vdc_min, &auio);
5358 
5359 	biofini(buf);
5360 	kmem_free(buf, sizeof (buf_t));
5361 
5362 	return (status);
5363 }
5364 
5365 /*
5366  * Allocate a buffer for a VD_OP_SCSICMD operation. The size of the allocated
5367  * buffer is returned in alloc_len.
5368  */
5369 static vd_scsi_t *
5370 vdc_scsi_alloc(int cdb_len, int sense_len, int datain_len, int dataout_len,
5371     int *alloc_len)
5372 {
5373 	vd_scsi_t *vd_scsi;
5374 	int vd_scsi_len = VD_SCSI_SIZE;
5375 
5376 	vd_scsi_len += P2ROUNDUP(cdb_len, sizeof (uint64_t));
5377 	vd_scsi_len += P2ROUNDUP(sense_len, sizeof (uint64_t));
5378 	vd_scsi_len += P2ROUNDUP(datain_len, sizeof (uint64_t));
5379 	vd_scsi_len += P2ROUNDUP(dataout_len, sizeof (uint64_t));
5380 
5381 	ASSERT(vd_scsi_len % sizeof (uint64_t) == 0);
5382 
5383 	vd_scsi = kmem_zalloc(vd_scsi_len, KM_SLEEP);
5384 
5385 	vd_scsi->cdb_len = cdb_len;
5386 	vd_scsi->sense_len = sense_len;
5387 	vd_scsi->datain_len = datain_len;
5388 	vd_scsi->dataout_len = dataout_len;
5389 
5390 	*alloc_len = vd_scsi_len;
5391 
5392 	return (vd_scsi);
5393 }
5394 
5395 /*
5396  * Convert the status of a SCSI command to a Solaris return code.
5397  *
5398  * Arguments:
5399  *	vd_scsi		- The SCSI operation buffer.
5400  *	log_error	- indicate if an error message should be logged.
5401  *
5402  * Note that our SCSI error messages are rather primitive for the moment
5403  * and could be improved by decoding some data like the SCSI command and
5404  * the sense key.
5405  *
5406  * Return value:
5407  *	0		- Status is good.
5408  *	EACCES		- Status reports a reservation conflict.
5409  *	ENOTSUP		- Status reports a check condition and sense key
5410  *			  reports an illegal request.
5411  *	EIO		- Any other status.
5412  */
5413 static int
5414 vdc_scsi_status(vdc_t *vdc, vd_scsi_t *vd_scsi, boolean_t log_error)
5415 {
5416 	int rv;
5417 	char path_str[MAXPATHLEN];
5418 	char panic_str[VDC_RESV_CONFLICT_FMT_LEN + MAXPATHLEN];
5419 	union scsi_cdb *cdb;
5420 	struct scsi_extended_sense *sense;
5421 
5422 	if (vd_scsi->cmd_status == STATUS_GOOD)
5423 		/* no error */
5424 		return (0);
5425 
5426 	/* when the tunable vdc_scsi_log_error is true we log all errors */
5427 	if (vdc_scsi_log_error)
5428 		log_error = B_TRUE;
5429 
5430 	if (log_error) {
5431 		cmn_err(CE_WARN, "%s (vdc%d):\tError for Command: 0x%x)\n",
5432 		    ddi_pathname(vdc->dip, path_str), vdc->instance,
5433 		    GETCMD(VD_SCSI_DATA_CDB(vd_scsi)));
5434 	}
5435 
5436 	/* default returned value */
5437 	rv = EIO;
5438 
5439 	switch (vd_scsi->cmd_status) {
5440 
5441 	case STATUS_CHECK:
5442 	case STATUS_TERMINATED:
5443 		if (log_error)
5444 			cmn_err(CE_CONT, "\tCheck Condition Error\n");
5445 
5446 		/* check sense buffer */
5447 		if (vd_scsi->sense_len == 0 ||
5448 		    vd_scsi->sense_status != STATUS_GOOD) {
5449 			if (log_error)
5450 				cmn_err(CE_CONT, "\tNo Sense Data Available\n");
5451 			break;
5452 		}
5453 
5454 		sense = VD_SCSI_DATA_SENSE(vd_scsi);
5455 
5456 		if (log_error) {
5457 			cmn_err(CE_CONT, "\tSense Key:  0x%x\n"
5458 			    "\tASC: 0x%x, ASCQ: 0x%x\n",
5459 			    scsi_sense_key((uint8_t *)sense),
5460 			    scsi_sense_asc((uint8_t *)sense),
5461 			    scsi_sense_ascq((uint8_t *)sense));
5462 		}
5463 
5464 		if (scsi_sense_key((uint8_t *)sense) == KEY_ILLEGAL_REQUEST)
5465 			rv = ENOTSUP;
5466 		break;
5467 
5468 	case STATUS_BUSY:
5469 		if (log_error)
5470 			cmn_err(CE_NOTE, "\tDevice Busy\n");
5471 		break;
5472 
5473 	case STATUS_RESERVATION_CONFLICT:
5474 		/*
5475 		 * If the command was PERSISTENT_RESERVATION_[IN|OUT] then
5476 		 * reservation conflict could be due to various reasons like
5477 		 * incorrect keys, not registered or not reserved etc. So,
5478 		 * we should not panic in that case.
5479 		 */
5480 		cdb = VD_SCSI_DATA_CDB(vd_scsi);
5481 		if (vdc->failfast_interval != 0 &&
5482 		    cdb->scc_cmd != SCMD_PERSISTENT_RESERVE_IN &&
5483 		    cdb->scc_cmd != SCMD_PERSISTENT_RESERVE_OUT) {
5484 			/* failfast is enabled so we have to panic */
5485 			(void) snprintf(panic_str, sizeof (panic_str),
5486 			    VDC_RESV_CONFLICT_FMT_STR "%s",
5487 			    ddi_pathname(vdc->dip, path_str));
5488 			panic(panic_str);
5489 		}
5490 		if (log_error)
5491 			cmn_err(CE_NOTE, "\tReservation Conflict\n");
5492 		rv = EACCES;
5493 		break;
5494 
5495 	case STATUS_QFULL:
5496 		if (log_error)
5497 			cmn_err(CE_NOTE, "\tQueue Full\n");
5498 		break;
5499 
5500 	case STATUS_MET:
5501 	case STATUS_INTERMEDIATE:
5502 	case STATUS_SCSI2:
5503 	case STATUS_INTERMEDIATE_MET:
5504 	case STATUS_ACA_ACTIVE:
5505 		if (log_error)
5506 			cmn_err(CE_CONT,
5507 			    "\tUnexpected SCSI status received: 0x%x\n",
5508 			    vd_scsi->cmd_status);
5509 		break;
5510 
5511 	default:
5512 		if (log_error)
5513 			cmn_err(CE_CONT,
5514 			    "\tInvalid SCSI status received: 0x%x\n",
5515 			    vd_scsi->cmd_status);
5516 		break;
5517 	}
5518 
5519 	return (rv);
5520 }
5521 
5522 /*
5523  * Implemented the USCSICMD uscsi(7I) ioctl. This ioctl is converted to
5524  * a VD_OP_SCSICMD operation which is sent to the vdisk server. If a SCSI
5525  * reset is requested (i.e. a flag USCSI_RESET* is set) then the ioctl is
5526  * converted to a VD_OP_RESET operation.
5527  */
5528 static int
5529 vdc_uscsi_cmd(vdc_t *vdc, caddr_t arg, int mode)
5530 {
5531 	struct uscsi_cmd 	uscsi;
5532 	struct uscsi_cmd32	uscsi32;
5533 	vd_scsi_t 		*vd_scsi;
5534 	int 			vd_scsi_len;
5535 	union scsi_cdb		*cdb;
5536 	struct scsi_extended_sense *sense;
5537 	char 			*datain, *dataout;
5538 	size_t			cdb_len, datain_len, dataout_len, sense_len;
5539 	int 			rv;
5540 
5541 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5542 		if (ddi_copyin(arg, &uscsi32, sizeof (struct uscsi_cmd32),
5543 		    mode) != 0)
5544 			return (EFAULT);
5545 		uscsi_cmd32touscsi_cmd((&uscsi32), (&uscsi));
5546 	} else {
5547 		if (ddi_copyin(arg, &uscsi, sizeof (struct uscsi_cmd),
5548 		    mode) != 0)
5549 			return (EFAULT);
5550 	}
5551 
5552 	/* a uscsi reset is converted to a VD_OP_RESET operation */
5553 	if (uscsi.uscsi_flags & (USCSI_RESET | USCSI_RESET_LUN |
5554 	    USCSI_RESET_ALL)) {
5555 		rv = vdc_do_sync_op(vdc, VD_OP_RESET, NULL, 0, 0, 0, CB_SYNC,
5556 		    (void *)(uint64_t)mode, VIO_both_dir, B_TRUE);
5557 		return (rv);
5558 	}
5559 
5560 	/* cdb buffer length */
5561 	cdb_len = uscsi.uscsi_cdblen;
5562 
5563 	/* data in and out buffers length */
5564 	if (uscsi.uscsi_flags & USCSI_READ) {
5565 		datain_len = uscsi.uscsi_buflen;
5566 		dataout_len = 0;
5567 	} else {
5568 		datain_len = 0;
5569 		dataout_len = uscsi.uscsi_buflen;
5570 	}
5571 
5572 	/* sense buffer length */
5573 	if (uscsi.uscsi_flags & USCSI_RQENABLE)
5574 		sense_len = uscsi.uscsi_rqlen;
5575 	else
5576 		sense_len = 0;
5577 
5578 	/* allocate buffer for the VD_SCSICMD_OP operation */
5579 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, datain_len, dataout_len,
5580 	    &vd_scsi_len);
5581 
5582 	/*
5583 	 * The documentation of USCSI_ISOLATE and USCSI_DIAGNOSE is very vague,
5584 	 * but basically they prevent a SCSI command from being retried in case
5585 	 * of an error.
5586 	 */
5587 	if ((uscsi.uscsi_flags & USCSI_ISOLATE) ||
5588 	    (uscsi.uscsi_flags & USCSI_DIAGNOSE))
5589 		vd_scsi->options |= VD_SCSI_OPT_NORETRY;
5590 
5591 	/* set task attribute */
5592 	if (uscsi.uscsi_flags & USCSI_NOTAG) {
5593 		vd_scsi->task_attribute = 0;
5594 	} else {
5595 		if (uscsi.uscsi_flags & USCSI_HEAD)
5596 			vd_scsi->task_attribute = VD_SCSI_TASK_ACA;
5597 		else if (uscsi.uscsi_flags & USCSI_HTAG)
5598 			vd_scsi->task_attribute = VD_SCSI_TASK_HQUEUE;
5599 		else if (uscsi.uscsi_flags & USCSI_OTAG)
5600 			vd_scsi->task_attribute = VD_SCSI_TASK_ORDERED;
5601 		else
5602 			vd_scsi->task_attribute = 0;
5603 	}
5604 
5605 	/* set timeout */
5606 	vd_scsi->timeout = uscsi.uscsi_timeout;
5607 
5608 	/* copy-in cdb data */
5609 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5610 	if (ddi_copyin(uscsi.uscsi_cdb, cdb, cdb_len, mode) != 0) {
5611 		rv = EFAULT;
5612 		goto done;
5613 	}
5614 
5615 	/* keep a pointer to the sense buffer */
5616 	sense = VD_SCSI_DATA_SENSE(vd_scsi);
5617 
5618 	/* keep a pointer to the data-in buffer */
5619 	datain = (char *)VD_SCSI_DATA_IN(vd_scsi);
5620 
5621 	/* copy-in request data to the data-out buffer */
5622 	dataout = (char *)VD_SCSI_DATA_OUT(vd_scsi);
5623 	if (!(uscsi.uscsi_flags & USCSI_READ)) {
5624 		if (ddi_copyin(uscsi.uscsi_bufaddr, dataout, dataout_len,
5625 		    mode)) {
5626 			rv = EFAULT;
5627 			goto done;
5628 		}
5629 	}
5630 
5631 	/* submit the request */
5632 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5633 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5634 
5635 	if (rv != 0)
5636 		goto done;
5637 
5638 	/* update scsi status */
5639 	uscsi.uscsi_status = vd_scsi->cmd_status;
5640 
5641 	/* update sense data */
5642 	if ((uscsi.uscsi_flags & USCSI_RQENABLE) &&
5643 	    (uscsi.uscsi_status == STATUS_CHECK ||
5644 	    uscsi.uscsi_status == STATUS_TERMINATED)) {
5645 
5646 		uscsi.uscsi_rqstatus = vd_scsi->sense_status;
5647 
5648 		if (uscsi.uscsi_rqstatus == STATUS_GOOD) {
5649 			uscsi.uscsi_rqresid = uscsi.uscsi_rqlen -
5650 			    vd_scsi->sense_len;
5651 			if (ddi_copyout(sense, uscsi.uscsi_rqbuf,
5652 			    vd_scsi->sense_len, mode) != 0) {
5653 				rv = EFAULT;
5654 				goto done;
5655 			}
5656 		}
5657 	}
5658 
5659 	/* update request data */
5660 	if (uscsi.uscsi_status == STATUS_GOOD) {
5661 		if (uscsi.uscsi_flags & USCSI_READ) {
5662 			uscsi.uscsi_resid = uscsi.uscsi_buflen -
5663 			    vd_scsi->datain_len;
5664 			if (ddi_copyout(datain, uscsi.uscsi_bufaddr,
5665 			    vd_scsi->datain_len, mode) != 0) {
5666 				rv = EFAULT;
5667 				goto done;
5668 			}
5669 		} else {
5670 			uscsi.uscsi_resid = uscsi.uscsi_buflen -
5671 			    vd_scsi->dataout_len;
5672 		}
5673 	}
5674 
5675 	/* copy-out result */
5676 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5677 		uscsi_cmdtouscsi_cmd32((&uscsi), (&uscsi32));
5678 		if (ddi_copyout(&uscsi32, arg, sizeof (struct uscsi_cmd32),
5679 		    mode) != 0) {
5680 			rv = EFAULT;
5681 			goto done;
5682 		}
5683 	} else {
5684 		if (ddi_copyout(&uscsi, arg, sizeof (struct uscsi_cmd),
5685 		    mode) != 0) {
5686 			rv = EFAULT;
5687 			goto done;
5688 		}
5689 	}
5690 
5691 	/* get the return code from the SCSI command status */
5692 	rv = vdc_scsi_status(vdc, vd_scsi,
5693 	    !(uscsi.uscsi_flags & USCSI_SILENT));
5694 
5695 done:
5696 	kmem_free(vd_scsi, vd_scsi_len);
5697 	return (rv);
5698 }
5699 
5700 /*
5701  * Create a VD_OP_SCSICMD buffer for a SCSI PERSISTENT IN command.
5702  *
5703  * Arguments:
5704  *	cmd		- SCSI PERSISTENT IN command
5705  *	len		- length of the SCSI input buffer
5706  *	vd_scsi_len	- return the length of the allocated buffer
5707  *
5708  * Returned Value:
5709  *	a pointer to the allocated VD_OP_SCSICMD buffer.
5710  */
5711 static vd_scsi_t *
5712 vdc_scsi_alloc_persistent_in(uchar_t cmd, int len, int *vd_scsi_len)
5713 {
5714 	int cdb_len, sense_len, datain_len, dataout_len;
5715 	vd_scsi_t *vd_scsi;
5716 	union scsi_cdb *cdb;
5717 
5718 	cdb_len = CDB_GROUP1;
5719 	sense_len = sizeof (struct scsi_extended_sense);
5720 	datain_len = len;
5721 	dataout_len = 0;
5722 
5723 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, datain_len, dataout_len,
5724 	    vd_scsi_len);
5725 
5726 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5727 
5728 	/* set cdb */
5729 	cdb->scc_cmd = SCMD_PERSISTENT_RESERVE_IN;
5730 	cdb->cdb_opaque[1] = cmd;
5731 	FORMG1COUNT(cdb, datain_len);
5732 
5733 	vd_scsi->timeout = vdc_scsi_timeout;
5734 
5735 	return (vd_scsi);
5736 }
5737 
5738 /*
5739  * Create a VD_OP_SCSICMD buffer for a SCSI PERSISTENT OUT command.
5740  *
5741  * Arguments:
5742  *	cmd		- SCSI PERSISTENT OUT command
5743  *	len		- length of the SCSI output buffer
5744  *	vd_scsi_len	- return the length of the allocated buffer
5745  *
5746  * Returned Code:
5747  *	a pointer to the allocated VD_OP_SCSICMD buffer.
5748  */
5749 static vd_scsi_t *
5750 vdc_scsi_alloc_persistent_out(uchar_t cmd, int len, int *vd_scsi_len)
5751 {
5752 	int cdb_len, sense_len, datain_len, dataout_len;
5753 	vd_scsi_t *vd_scsi;
5754 	union scsi_cdb *cdb;
5755 
5756 	cdb_len = CDB_GROUP1;
5757 	sense_len = sizeof (struct scsi_extended_sense);
5758 	datain_len = 0;
5759 	dataout_len = len;
5760 
5761 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, datain_len, dataout_len,
5762 	    vd_scsi_len);
5763 
5764 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5765 
5766 	/* set cdb */
5767 	cdb->scc_cmd = SCMD_PERSISTENT_RESERVE_OUT;
5768 	cdb->cdb_opaque[1] = cmd;
5769 	FORMG1COUNT(cdb, dataout_len);
5770 
5771 	vd_scsi->timeout = vdc_scsi_timeout;
5772 
5773 	return (vd_scsi);
5774 }
5775 
5776 /*
5777  * Implement the MHIOCGRP_INKEYS mhd(7i) ioctl. The ioctl is converted
5778  * to a SCSI PERSISTENT IN READ KEYS command which is sent to the vdisk
5779  * server with a VD_OP_SCSICMD operation.
5780  */
5781 static int
5782 vdc_mhd_inkeys(vdc_t *vdc, caddr_t arg, int mode)
5783 {
5784 	vd_scsi_t *vd_scsi;
5785 	mhioc_inkeys_t inkeys;
5786 	mhioc_key_list_t klist;
5787 	struct mhioc_inkeys32 inkeys32;
5788 	struct mhioc_key_list32 klist32;
5789 	sd_prin_readkeys_t *scsi_keys;
5790 	void *user_keys;
5791 	int vd_scsi_len;
5792 	int listsize, listlen, rv;
5793 
5794 	/* copyin arguments */
5795 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5796 		rv = ddi_copyin(arg, &inkeys32, sizeof (inkeys32), mode);
5797 		if (rv != 0)
5798 			return (EFAULT);
5799 
5800 		rv = ddi_copyin((caddr_t)(uintptr_t)inkeys32.li, &klist32,
5801 		    sizeof (klist32), mode);
5802 		if (rv != 0)
5803 			return (EFAULT);
5804 
5805 		listsize = klist32.listsize;
5806 	} else {
5807 		rv = ddi_copyin(arg, &inkeys, sizeof (inkeys), mode);
5808 		if (rv != 0)
5809 			return (EFAULT);
5810 
5811 		rv = ddi_copyin(inkeys.li, &klist, sizeof (klist), mode);
5812 		if (rv != 0)
5813 			return (EFAULT);
5814 
5815 		listsize = klist.listsize;
5816 	}
5817 
5818 	/* build SCSI VD_OP request */
5819 	vd_scsi = vdc_scsi_alloc_persistent_in(SD_READ_KEYS,
5820 	    sizeof (sd_prin_readkeys_t) - sizeof (caddr_t) +
5821 	    (sizeof (mhioc_resv_key_t) * listsize), &vd_scsi_len);
5822 
5823 	scsi_keys = (sd_prin_readkeys_t *)VD_SCSI_DATA_IN(vd_scsi);
5824 
5825 	/* submit the request */
5826 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5827 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5828 
5829 	if (rv != 0)
5830 		goto done;
5831 
5832 	listlen = scsi_keys->len / MHIOC_RESV_KEY_SIZE;
5833 
5834 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5835 		inkeys32.generation = scsi_keys->generation;
5836 		rv = ddi_copyout(&inkeys32, arg, sizeof (inkeys32), mode);
5837 		if (rv != 0) {
5838 			rv = EFAULT;
5839 			goto done;
5840 		}
5841 
5842 		klist32.listlen = listlen;
5843 		rv = ddi_copyout(&klist32, (caddr_t)(uintptr_t)inkeys32.li,
5844 		    sizeof (klist32), mode);
5845 		if (rv != 0) {
5846 			rv = EFAULT;
5847 			goto done;
5848 		}
5849 
5850 		user_keys = (caddr_t)(uintptr_t)klist32.list;
5851 	} else {
5852 		inkeys.generation = scsi_keys->generation;
5853 		rv = ddi_copyout(&inkeys, arg, sizeof (inkeys), mode);
5854 		if (rv != 0) {
5855 			rv = EFAULT;
5856 			goto done;
5857 		}
5858 
5859 		klist.listlen = listlen;
5860 		rv = ddi_copyout(&klist, inkeys.li, sizeof (klist), mode);
5861 		if (rv != 0) {
5862 			rv = EFAULT;
5863 			goto done;
5864 		}
5865 
5866 		user_keys = klist.list;
5867 	}
5868 
5869 	/* copy out keys */
5870 	if (listlen > 0 && listsize > 0) {
5871 		if (listsize < listlen)
5872 			listlen = listsize;
5873 		rv = ddi_copyout(&scsi_keys->keylist, user_keys,
5874 		    listlen * MHIOC_RESV_KEY_SIZE, mode);
5875 		if (rv != 0)
5876 			rv = EFAULT;
5877 	}
5878 
5879 	if (rv == 0)
5880 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
5881 
5882 done:
5883 	kmem_free(vd_scsi, vd_scsi_len);
5884 
5885 	return (rv);
5886 }
5887 
5888 /*
5889  * Implement the MHIOCGRP_INRESV mhd(7i) ioctl. The ioctl is converted
5890  * to a SCSI PERSISTENT IN READ RESERVATION command which is sent to
5891  * the vdisk server with a VD_OP_SCSICMD operation.
5892  */
5893 static int
5894 vdc_mhd_inresv(vdc_t *vdc, caddr_t arg, int mode)
5895 {
5896 	vd_scsi_t *vd_scsi;
5897 	mhioc_inresvs_t inresv;
5898 	mhioc_resv_desc_list_t rlist;
5899 	struct mhioc_inresvs32 inresv32;
5900 	struct mhioc_resv_desc_list32 rlist32;
5901 	mhioc_resv_desc_t mhd_resv;
5902 	sd_prin_readresv_t *scsi_resv;
5903 	sd_readresv_desc_t *resv;
5904 	mhioc_resv_desc_t *user_resv;
5905 	int vd_scsi_len;
5906 	int listsize, listlen, i, rv;
5907 
5908 	/* copyin arguments */
5909 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5910 		rv = ddi_copyin(arg, &inresv32, sizeof (inresv32), mode);
5911 		if (rv != 0)
5912 			return (EFAULT);
5913 
5914 		rv = ddi_copyin((caddr_t)(uintptr_t)inresv32.li, &rlist32,
5915 		    sizeof (rlist32), mode);
5916 		if (rv != 0)
5917 			return (EFAULT);
5918 
5919 		listsize = rlist32.listsize;
5920 	} else {
5921 		rv = ddi_copyin(arg, &inresv, sizeof (inresv), mode);
5922 		if (rv != 0)
5923 			return (EFAULT);
5924 
5925 		rv = ddi_copyin(inresv.li, &rlist, sizeof (rlist), mode);
5926 		if (rv != 0)
5927 			return (EFAULT);
5928 
5929 		listsize = rlist.listsize;
5930 	}
5931 
5932 	/* build SCSI VD_OP request */
5933 	vd_scsi = vdc_scsi_alloc_persistent_in(SD_READ_RESV,
5934 	    sizeof (sd_prin_readresv_t) - sizeof (caddr_t) +
5935 	    (SCSI3_RESV_DESC_LEN * listsize), &vd_scsi_len);
5936 
5937 	scsi_resv = (sd_prin_readresv_t *)VD_SCSI_DATA_IN(vd_scsi);
5938 
5939 	/* submit the request */
5940 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5941 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5942 
5943 	if (rv != 0)
5944 		goto done;
5945 
5946 	listlen = scsi_resv->len / SCSI3_RESV_DESC_LEN;
5947 
5948 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5949 		inresv32.generation = scsi_resv->generation;
5950 		rv = ddi_copyout(&inresv32, arg, sizeof (inresv32), mode);
5951 		if (rv != 0) {
5952 			rv = EFAULT;
5953 			goto done;
5954 		}
5955 
5956 		rlist32.listlen = listlen;
5957 		rv = ddi_copyout(&rlist32, (caddr_t)(uintptr_t)inresv32.li,
5958 		    sizeof (rlist32), mode);
5959 		if (rv != 0) {
5960 			rv = EFAULT;
5961 			goto done;
5962 		}
5963 
5964 		user_resv = (mhioc_resv_desc_t *)(uintptr_t)rlist32.list;
5965 	} else {
5966 		inresv.generation = scsi_resv->generation;
5967 		rv = ddi_copyout(&inresv, arg, sizeof (inresv), mode);
5968 		if (rv != 0) {
5969 			rv = EFAULT;
5970 			goto done;
5971 		}
5972 
5973 		rlist.listlen = listlen;
5974 		rv = ddi_copyout(&rlist, inresv.li, sizeof (rlist), mode);
5975 		if (rv != 0) {
5976 			rv = EFAULT;
5977 			goto done;
5978 		}
5979 
5980 		user_resv = rlist.list;
5981 	}
5982 
5983 	/* copy out reservations */
5984 	if (listsize > 0 && listlen > 0) {
5985 		if (listsize < listlen)
5986 			listlen = listsize;
5987 		resv = (sd_readresv_desc_t *)&scsi_resv->readresv_desc;
5988 
5989 		for (i = 0; i < listlen; i++) {
5990 			mhd_resv.type = resv->type;
5991 			mhd_resv.scope = resv->scope;
5992 			mhd_resv.scope_specific_addr =
5993 			    BE_32(resv->scope_specific_addr);
5994 			bcopy(&resv->resvkey, &mhd_resv.key,
5995 			    MHIOC_RESV_KEY_SIZE);
5996 
5997 			rv = ddi_copyout(&mhd_resv, user_resv,
5998 			    sizeof (mhd_resv), mode);
5999 			if (rv != 0) {
6000 				rv = EFAULT;
6001 				goto done;
6002 			}
6003 			resv++;
6004 			user_resv++;
6005 		}
6006 	}
6007 
6008 	if (rv == 0)
6009 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6010 
6011 done:
6012 	kmem_free(vd_scsi, vd_scsi_len);
6013 	return (rv);
6014 }
6015 
6016 /*
6017  * Implement the MHIOCGRP_REGISTER mhd(7i) ioctl. The ioctl is converted
6018  * to a SCSI PERSISTENT OUT REGISTER command which is sent to the vdisk
6019  * server with a VD_OP_SCSICMD operation.
6020  */
6021 static int
6022 vdc_mhd_register(vdc_t *vdc, caddr_t arg, int mode)
6023 {
6024 	vd_scsi_t *vd_scsi;
6025 	sd_prout_t *scsi_prout;
6026 	mhioc_register_t mhd_reg;
6027 	int vd_scsi_len, rv;
6028 
6029 	/* copyin arguments */
6030 	rv = ddi_copyin(arg, &mhd_reg, sizeof (mhd_reg), mode);
6031 	if (rv != 0)
6032 		return (EFAULT);
6033 
6034 	/* build SCSI VD_OP request */
6035 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_REGISTER,
6036 	    sizeof (sd_prout_t), &vd_scsi_len);
6037 
6038 	/* set parameters */
6039 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
6040 	bcopy(mhd_reg.oldkey.key, scsi_prout->res_key, MHIOC_RESV_KEY_SIZE);
6041 	bcopy(mhd_reg.newkey.key, scsi_prout->service_key, MHIOC_RESV_KEY_SIZE);
6042 	scsi_prout->aptpl = (uchar_t)mhd_reg.aptpl;
6043 
6044 	/* submit the request */
6045 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6046 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
6047 
6048 	if (rv == 0)
6049 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6050 
6051 	kmem_free(vd_scsi, vd_scsi_len);
6052 	return (rv);
6053 }
6054 
6055 /*
6056  * Implement the MHIOCGRP_RESERVE mhd(7i) ioctl. The ioctl is converted
6057  * to a SCSI PERSISTENT OUT RESERVE command which is sent to the vdisk
6058  * server with a VD_OP_SCSICMD operation.
6059  */
6060 static int
6061 vdc_mhd_reserve(vdc_t *vdc, caddr_t arg, int mode)
6062 {
6063 	union scsi_cdb *cdb;
6064 	vd_scsi_t *vd_scsi;
6065 	sd_prout_t *scsi_prout;
6066 	mhioc_resv_desc_t mhd_resv;
6067 	int vd_scsi_len, rv;
6068 
6069 	/* copyin arguments */
6070 	rv = ddi_copyin(arg, &mhd_resv, sizeof (mhd_resv), mode);
6071 	if (rv != 0)
6072 		return (EFAULT);
6073 
6074 	/* build SCSI VD_OP request */
6075 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_RESERVE,
6076 	    sizeof (sd_prout_t), &vd_scsi_len);
6077 
6078 	/* set parameters */
6079 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
6080 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
6081 	bcopy(mhd_resv.key.key, scsi_prout->res_key, MHIOC_RESV_KEY_SIZE);
6082 	scsi_prout->scope_address = mhd_resv.scope_specific_addr;
6083 	cdb->cdb_opaque[2] = mhd_resv.type;
6084 
6085 	/* submit the request */
6086 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6087 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
6088 
6089 	if (rv == 0)
6090 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6091 
6092 	kmem_free(vd_scsi, vd_scsi_len);
6093 	return (rv);
6094 }
6095 
6096 /*
6097  * Implement the MHIOCGRP_PREEMPTANDABORT mhd(7i) ioctl. The ioctl is
6098  * converted to a SCSI PERSISTENT OUT PREEMPT AND ABORT command which
6099  * is sent to the vdisk server with a VD_OP_SCSICMD operation.
6100  */
6101 static int
6102 vdc_mhd_preemptabort(vdc_t *vdc, caddr_t arg, int mode)
6103 {
6104 	union scsi_cdb *cdb;
6105 	vd_scsi_t *vd_scsi;
6106 	sd_prout_t *scsi_prout;
6107 	mhioc_preemptandabort_t mhd_preempt;
6108 	int vd_scsi_len, rv;
6109 
6110 	/* copyin arguments */
6111 	rv = ddi_copyin(arg, &mhd_preempt, sizeof (mhd_preempt), mode);
6112 	if (rv != 0)
6113 		return (EFAULT);
6114 
6115 	/* build SCSI VD_OP request */
6116 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_PREEMPTANDABORT,
6117 	    sizeof (sd_prout_t), &vd_scsi_len);
6118 
6119 	/* set parameters */
6120 	vd_scsi->task_attribute = VD_SCSI_TASK_ACA;
6121 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
6122 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
6123 	bcopy(mhd_preempt.resvdesc.key.key, scsi_prout->res_key,
6124 	    MHIOC_RESV_KEY_SIZE);
6125 	bcopy(mhd_preempt.victim_key.key, scsi_prout->service_key,
6126 	    MHIOC_RESV_KEY_SIZE);
6127 	scsi_prout->scope_address = mhd_preempt.resvdesc.scope_specific_addr;
6128 	cdb->cdb_opaque[2] = mhd_preempt.resvdesc.type;
6129 
6130 	/* submit the request */
6131 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6132 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
6133 
6134 	if (rv == 0)
6135 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6136 
6137 	kmem_free(vd_scsi, vd_scsi_len);
6138 	return (rv);
6139 }
6140 
6141 /*
6142  * Implement the MHIOCGRP_REGISTERANDIGNOREKEY mhd(7i) ioctl. The ioctl
6143  * is converted to a SCSI PERSISTENT OUT REGISTER AND IGNORE EXISTING KEY
6144  * command which is sent to the vdisk server with a VD_OP_SCSICMD operation.
6145  */
6146 static int
6147 vdc_mhd_registerignore(vdc_t *vdc, caddr_t arg, int mode)
6148 {
6149 	vd_scsi_t *vd_scsi;
6150 	sd_prout_t *scsi_prout;
6151 	mhioc_registerandignorekey_t mhd_regi;
6152 	int vd_scsi_len, rv;
6153 
6154 	/* copyin arguments */
6155 	rv = ddi_copyin(arg, &mhd_regi, sizeof (mhd_regi), mode);
6156 	if (rv != 0)
6157 		return (EFAULT);
6158 
6159 	/* build SCSI VD_OP request */
6160 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_REGISTERANDIGNOREKEY,
6161 	    sizeof (sd_prout_t), &vd_scsi_len);
6162 
6163 	/* set parameters */
6164 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
6165 	bcopy(mhd_regi.newkey.key, scsi_prout->service_key,
6166 	    MHIOC_RESV_KEY_SIZE);
6167 	scsi_prout->aptpl = (uchar_t)mhd_regi.aptpl;
6168 
6169 	/* submit the request */
6170 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6171 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
6172 
6173 	if (rv == 0)
6174 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6175 
6176 	kmem_free(vd_scsi, vd_scsi_len);
6177 	return (rv);
6178 }
6179 
6180 /*
6181  * This function is used by the failfast mechanism to send a SCSI command
6182  * to check for reservation conflict.
6183  */
6184 static int
6185 vdc_failfast_scsi_cmd(vdc_t *vdc, uchar_t scmd)
6186 {
6187 	int cdb_len, sense_len, vd_scsi_len;
6188 	vd_scsi_t *vd_scsi;
6189 	union scsi_cdb *cdb;
6190 	int rv;
6191 
6192 	ASSERT(scmd == SCMD_TEST_UNIT_READY || scmd == SCMD_WRITE_G1);
6193 
6194 	if (scmd == SCMD_WRITE_G1)
6195 		cdb_len = CDB_GROUP1;
6196 	else
6197 		cdb_len = CDB_GROUP0;
6198 
6199 	sense_len = sizeof (struct scsi_extended_sense);
6200 
6201 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, 0, 0, &vd_scsi_len);
6202 
6203 	/* set cdb */
6204 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
6205 	cdb->scc_cmd = scmd;
6206 
6207 	vd_scsi->timeout = vdc_scsi_timeout;
6208 
6209 	/*
6210 	 * Submit the request. The last argument has to be B_FALSE so that
6211 	 * vdc_do_sync_op does not loop checking for reservation conflict if
6212 	 * the operation returns an error.
6213 	 */
6214 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6215 	    0, 0, CB_SYNC, (void *)(uint64_t)FKIOCTL, VIO_both_dir, B_FALSE);
6216 
6217 	if (rv == 0)
6218 		(void) vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6219 
6220 	kmem_free(vd_scsi, vd_scsi_len);
6221 	return (rv);
6222 }
6223 
6224 /*
6225  * This function is used by the failfast mechanism to check for reservation
6226  * conflict. It sends some SCSI commands which will fail with a reservation
6227  * conflict error if the system does not have access to the disk and this
6228  * will panic the system.
6229  *
6230  * Returned Code:
6231  *	0	- disk is accessible without reservation conflict error
6232  *	!= 0	- unable to check if disk is accessible
6233  */
6234 int
6235 vdc_failfast_check_resv(vdc_t *vdc)
6236 {
6237 	int failure = 0;
6238 
6239 	/*
6240 	 * Send a TEST UNIT READY command. The command will panic
6241 	 * the system if it fails with a reservation conflict.
6242 	 */
6243 	if (vdc_failfast_scsi_cmd(vdc, SCMD_TEST_UNIT_READY) != 0)
6244 		failure++;
6245 
6246 	/*
6247 	 * With SPC-3 compliant devices TEST UNIT READY will succeed on
6248 	 * a reserved device, so we also do a WRITE(10) of zero byte in
6249 	 * order to provoke a Reservation Conflict status on those newer
6250 	 * devices.
6251 	 */
6252 	if (vdc_failfast_scsi_cmd(vdc, SCMD_WRITE_G1) != 0)
6253 		failure++;
6254 
6255 	return (failure);
6256 }
6257 
6258 /*
6259  * Add a pending I/O to the failfast I/O queue. An I/O is added to this
6260  * queue when it has failed and failfast is enabled. Then we have to check
6261  * if it has failed because of a reservation conflict in which case we have
6262  * to panic the system.
6263  *
6264  * Async I/O should be queued with their block I/O data transfer structure
6265  * (buf). Sync I/O should be queued with buf = NULL.
6266  */
6267 static vdc_io_t *
6268 vdc_failfast_io_queue(vdc_t *vdc, struct buf *buf)
6269 {
6270 	vdc_io_t *vio;
6271 
6272 	ASSERT(MUTEX_HELD(&vdc->lock));
6273 
6274 	vio = kmem_alloc(sizeof (vdc_io_t), KM_SLEEP);
6275 	vio->vio_next = vdc->failfast_io_queue;
6276 	vio->vio_buf = buf;
6277 	vio->vio_qtime = ddi_get_lbolt();
6278 
6279 	vdc->failfast_io_queue = vio;
6280 
6281 	/* notify the failfast thread that a new I/O is queued */
6282 	cv_signal(&vdc->failfast_cv);
6283 
6284 	return (vio);
6285 }
6286 
6287 /*
6288  * Remove and complete I/O in the failfast I/O queue which have been
6289  * added after the indicated deadline. A deadline of 0 means that all
6290  * I/O have to be unqueued and marked as completed.
6291  */
6292 static void
6293 vdc_failfast_io_unqueue(vdc_t *vdc, clock_t deadline)
6294 {
6295 	vdc_io_t *vio, *vio_tmp;
6296 
6297 	ASSERT(MUTEX_HELD(&vdc->lock));
6298 
6299 	vio_tmp = NULL;
6300 	vio = vdc->failfast_io_queue;
6301 
6302 	if (deadline != 0) {
6303 		/*
6304 		 * Skip any io queued after the deadline. The failfast
6305 		 * I/O queue is ordered starting with the last I/O added
6306 		 * to the queue.
6307 		 */
6308 		while (vio != NULL && vio->vio_qtime > deadline) {
6309 			vio_tmp = vio;
6310 			vio = vio->vio_next;
6311 		}
6312 	}
6313 
6314 	if (vio == NULL)
6315 		/* nothing to unqueue */
6316 		return;
6317 
6318 	/* update the queue */
6319 	if (vio_tmp == NULL)
6320 		vdc->failfast_io_queue = NULL;
6321 	else
6322 		vio_tmp->vio_next = NULL;
6323 
6324 	/*
6325 	 * Complete unqueued I/O. Async I/O have a block I/O data transfer
6326 	 * structure (buf) and they are completed by calling biodone(). Sync
6327 	 * I/O do not have a buf and they are completed by setting the
6328 	 * vio_qtime to zero and signaling failfast_io_cv. In that case, the
6329 	 * thread waiting for the I/O to complete is responsible for freeing
6330 	 * the vio structure.
6331 	 */
6332 	while (vio != NULL) {
6333 		vio_tmp = vio->vio_next;
6334 		if (vio->vio_buf != NULL) {
6335 			VD_KSTAT_RUNQ_EXIT(vdc);
6336 			DTRACE_IO1(done, buf_t *, vio->vio_buf);
6337 			biodone(vio->vio_buf);
6338 			kmem_free(vio, sizeof (vdc_io_t));
6339 		} else {
6340 			vio->vio_qtime = 0;
6341 		}
6342 		vio = vio_tmp;
6343 	}
6344 
6345 	cv_broadcast(&vdc->failfast_io_cv);
6346 }
6347 
6348 /*
6349  * Failfast Thread.
6350  *
6351  * While failfast is enabled, the failfast thread sends a TEST UNIT READY
6352  * and a zero size WRITE(10) SCSI commands on a regular basis to check that
6353  * we still have access to the disk. If a command fails with a RESERVATION
6354  * CONFLICT error then the system will immediatly panic.
6355  *
6356  * The failfast thread is also woken up when an I/O has failed. It then check
6357  * the access to the disk to ensure that the I/O failure was not due to a
6358  * reservation conflict.
6359  *
6360  * There is one failfast thread for each virtual disk for which failfast is
6361  * enabled. We could have only one thread sending requests for all disks but
6362  * this would need vdc to send asynchronous requests and to have callbacks to
6363  * process replies.
6364  */
6365 static void
6366 vdc_failfast_thread(void *arg)
6367 {
6368 	int status;
6369 	vdc_t *vdc = (vdc_t *)arg;
6370 	clock_t timeout, starttime;
6371 
6372 	mutex_enter(&vdc->lock);
6373 
6374 	while (vdc->failfast_interval != 0) {
6375 
6376 		starttime = ddi_get_lbolt();
6377 
6378 		mutex_exit(&vdc->lock);
6379 
6380 		/* check for reservation conflict */
6381 		status = vdc_failfast_check_resv(vdc);
6382 
6383 		mutex_enter(&vdc->lock);
6384 		/*
6385 		 * We have dropped the lock to send the SCSI command so we have
6386 		 * to check that failfast is still enabled.
6387 		 */
6388 		if (vdc->failfast_interval == 0)
6389 			break;
6390 
6391 		/*
6392 		 * If we have successfully check the disk access and there was
6393 		 * no reservation conflict then we can complete any I/O queued
6394 		 * before the last check.
6395 		 */
6396 		if (status == 0)
6397 			vdc_failfast_io_unqueue(vdc, starttime);
6398 
6399 		/* proceed again if some I/O are still in the queue */
6400 		if (vdc->failfast_io_queue != NULL)
6401 			continue;
6402 
6403 		timeout = ddi_get_lbolt() +
6404 		    drv_usectohz(vdc->failfast_interval);
6405 		(void) cv_timedwait(&vdc->failfast_cv, &vdc->lock, timeout);
6406 	}
6407 
6408 	/*
6409 	 * Failfast is being stop so we can complete any queued I/O.
6410 	 */
6411 	vdc_failfast_io_unqueue(vdc, 0);
6412 	vdc->failfast_thread = NULL;
6413 	mutex_exit(&vdc->lock);
6414 	thread_exit();
6415 }
6416 
6417 /*
6418  * Implement the MHIOCENFAILFAST mhd(7i) ioctl.
6419  */
6420 static int
6421 vdc_failfast(vdc_t *vdc, caddr_t arg, int mode)
6422 {
6423 	unsigned int mh_time;
6424 
6425 	if (ddi_copyin((void *)arg, &mh_time, sizeof (int), mode))
6426 		return (EFAULT);
6427 
6428 	mutex_enter(&vdc->lock);
6429 	if (mh_time != 0 && vdc->failfast_thread == NULL) {
6430 		vdc->failfast_thread = thread_create(NULL, 0,
6431 		    vdc_failfast_thread, vdc, 0, &p0, TS_RUN,
6432 		    v.v_maxsyspri - 2);
6433 	}
6434 
6435 	vdc->failfast_interval = mh_time * 1000;
6436 	cv_signal(&vdc->failfast_cv);
6437 	mutex_exit(&vdc->lock);
6438 
6439 	return (0);
6440 }
6441 
6442 /*
6443  * Implement the MHIOCTKOWN and MHIOCRELEASE mhd(7i) ioctls. These ioctls are
6444  * converted to VD_OP_SET_ACCESS operations.
6445  */
6446 static int
6447 vdc_access_set(vdc_t *vdc, uint64_t flags, int mode)
6448 {
6449 	int rv;
6450 
6451 	/* submit owership command request */
6452 	rv = vdc_do_sync_op(vdc, VD_OP_SET_ACCESS, (caddr_t)&flags,
6453 	    sizeof (uint64_t), 0, 0, CB_SYNC, (void *)(uint64_t)mode,
6454 	    VIO_both_dir, B_TRUE);
6455 
6456 	return (rv);
6457 }
6458 
6459 /*
6460  * Implement the MHIOCSTATUS mhd(7i) ioctl. This ioctl is converted to a
6461  * VD_OP_GET_ACCESS operation.
6462  */
6463 static int
6464 vdc_access_get(vdc_t *vdc, uint64_t *status, int mode)
6465 {
6466 	int rv;
6467 
6468 	/* submit owership command request */
6469 	rv = vdc_do_sync_op(vdc, VD_OP_GET_ACCESS, (caddr_t)status,
6470 	    sizeof (uint64_t), 0, 0, CB_SYNC, (void *)(uint64_t)mode,
6471 	    VIO_both_dir, B_TRUE);
6472 
6473 	return (rv);
6474 }
6475 
6476 /*
6477  * Disk Ownership Thread.
6478  *
6479  * When we have taken the ownership of a disk, this thread waits to be
6480  * notified when the LDC channel is reset so that it can recover the
6481  * ownership.
6482  *
6483  * Note that the thread handling the LDC reset (vdc_process_msg_thread())
6484  * can not be used to do the ownership recovery because it has to be
6485  * running to handle the reply message to the ownership operation.
6486  */
6487 static void
6488 vdc_ownership_thread(void *arg)
6489 {
6490 	vdc_t *vdc = (vdc_t *)arg;
6491 	clock_t timeout;
6492 	uint64_t status;
6493 
6494 	mutex_enter(&vdc->ownership_lock);
6495 	mutex_enter(&vdc->lock);
6496 
6497 	while (vdc->ownership & VDC_OWNERSHIP_WANTED) {
6498 
6499 		if ((vdc->ownership & VDC_OWNERSHIP_RESET) ||
6500 		    !(vdc->ownership & VDC_OWNERSHIP_GRANTED)) {
6501 			/*
6502 			 * There was a reset so the ownership has been lost,
6503 			 * try to recover. We do this without using the preempt
6504 			 * option so that we don't steal the ownership from
6505 			 * someone who has preempted us.
6506 			 */
6507 			DMSG(vdc, 0, "[%d] Ownership lost, recovering",
6508 			    vdc->instance);
6509 
6510 			vdc->ownership &= ~(VDC_OWNERSHIP_RESET |
6511 			    VDC_OWNERSHIP_GRANTED);
6512 
6513 			mutex_exit(&vdc->lock);
6514 
6515 			status = vdc_access_set(vdc, VD_ACCESS_SET_EXCLUSIVE |
6516 			    VD_ACCESS_SET_PRESERVE, FKIOCTL);
6517 
6518 			mutex_enter(&vdc->lock);
6519 
6520 			if (status == 0) {
6521 				DMSG(vdc, 0, "[%d] Ownership recovered",
6522 				    vdc->instance);
6523 				vdc->ownership |= VDC_OWNERSHIP_GRANTED;
6524 			} else {
6525 				DMSG(vdc, 0, "[%d] Fail to recover ownership",
6526 				    vdc->instance);
6527 			}
6528 
6529 		}
6530 
6531 		/*
6532 		 * If we have the ownership then we just wait for an event
6533 		 * to happen (LDC reset), otherwise we will retry to recover
6534 		 * after a delay.
6535 		 */
6536 		if (vdc->ownership & VDC_OWNERSHIP_GRANTED)
6537 			timeout = 0;
6538 		else
6539 			timeout = ddi_get_lbolt() +
6540 			    drv_usectohz(vdc_ownership_delay);
6541 
6542 		/* Release the ownership_lock and wait on the vdc lock */
6543 		mutex_exit(&vdc->ownership_lock);
6544 
6545 		if (timeout == 0)
6546 			(void) cv_wait(&vdc->ownership_cv, &vdc->lock);
6547 		else
6548 			(void) cv_timedwait(&vdc->ownership_cv,
6549 			    &vdc->lock, timeout);
6550 
6551 		mutex_exit(&vdc->lock);
6552 
6553 		mutex_enter(&vdc->ownership_lock);
6554 		mutex_enter(&vdc->lock);
6555 	}
6556 
6557 	vdc->ownership_thread = NULL;
6558 	mutex_exit(&vdc->lock);
6559 	mutex_exit(&vdc->ownership_lock);
6560 
6561 	thread_exit();
6562 }
6563 
6564 static void
6565 vdc_ownership_update(vdc_t *vdc, int ownership_flags)
6566 {
6567 	ASSERT(MUTEX_HELD(&vdc->ownership_lock));
6568 
6569 	mutex_enter(&vdc->lock);
6570 	vdc->ownership = ownership_flags;
6571 	if ((vdc->ownership & VDC_OWNERSHIP_WANTED) &&
6572 	    vdc->ownership_thread == NULL) {
6573 		/* start ownership thread */
6574 		vdc->ownership_thread = thread_create(NULL, 0,
6575 		    vdc_ownership_thread, vdc, 0, &p0, TS_RUN,
6576 		    v.v_maxsyspri - 2);
6577 	} else {
6578 		/* notify the ownership thread */
6579 		cv_signal(&vdc->ownership_cv);
6580 	}
6581 	mutex_exit(&vdc->lock);
6582 }
6583 
6584 /*
6585  * Get the size and the block size of a virtual disk from the vdisk server.
6586  * We need to use this operation when the vdisk_size attribute was not
6587  * available during the handshake with the vdisk server.
6588  */
6589 static int
6590 vdc_check_capacity(vdc_t *vdc)
6591 {
6592 	int rv = 0;
6593 	size_t alloc_len;
6594 	vd_capacity_t *vd_cap;
6595 
6596 	if (vdc->vdisk_size != 0)
6597 		return (0);
6598 
6599 	alloc_len = P2ROUNDUP(sizeof (vd_capacity_t), sizeof (uint64_t));
6600 
6601 	vd_cap = kmem_zalloc(alloc_len, KM_SLEEP);
6602 
6603 	rv = vdc_do_sync_op(vdc, VD_OP_GET_CAPACITY, (caddr_t)vd_cap, alloc_len,
6604 	    0, 0, CB_SYNC, (void *)(uint64_t)FKIOCTL, VIO_both_dir, B_TRUE);
6605 
6606 	if (rv == 0) {
6607 		if (vd_cap->vdisk_block_size != vdc->block_size ||
6608 		    vd_cap->vdisk_size == VD_SIZE_UNKNOWN ||
6609 		    vd_cap->vdisk_size == 0)
6610 			rv = EINVAL;
6611 		else
6612 			vdc->vdisk_size = vd_cap->vdisk_size;
6613 	}
6614 
6615 	kmem_free(vd_cap, alloc_len);
6616 	return (rv);
6617 }
6618 
6619 /*
6620  * This structure is used in the DKIO(7I) array below.
6621  */
6622 typedef struct vdc_dk_ioctl {
6623 	uint8_t		op;		/* VD_OP_XXX value */
6624 	int		cmd;		/* Solaris ioctl operation number */
6625 	size_t		nbytes;		/* size of structure to be copied */
6626 
6627 	/* function to convert between vDisk and Solaris structure formats */
6628 	int	(*convert)(vdc_t *vdc, void *vd_buf, void *ioctl_arg,
6629 	    int mode, int dir);
6630 } vdc_dk_ioctl_t;
6631 
6632 /*
6633  * Subset of DKIO(7I) operations currently supported
6634  */
6635 static vdc_dk_ioctl_t	dk_ioctl[] = {
6636 	{VD_OP_FLUSH,		DKIOCFLUSHWRITECACHE,	0,
6637 		vdc_null_copy_func},
6638 	{VD_OP_GET_WCE,		DKIOCGETWCE,		sizeof (int),
6639 		vdc_get_wce_convert},
6640 	{VD_OP_SET_WCE,		DKIOCSETWCE,		sizeof (int),
6641 		vdc_set_wce_convert},
6642 	{VD_OP_GET_VTOC,	DKIOCGVTOC,		sizeof (vd_vtoc_t),
6643 		vdc_get_vtoc_convert},
6644 	{VD_OP_SET_VTOC,	DKIOCSVTOC,		sizeof (vd_vtoc_t),
6645 		vdc_set_vtoc_convert},
6646 	{VD_OP_GET_DISKGEOM,	DKIOCGGEOM,		sizeof (vd_geom_t),
6647 		vdc_get_geom_convert},
6648 	{VD_OP_GET_DISKGEOM,	DKIOCG_PHYGEOM,		sizeof (vd_geom_t),
6649 		vdc_get_geom_convert},
6650 	{VD_OP_GET_DISKGEOM, 	DKIOCG_VIRTGEOM,	sizeof (vd_geom_t),
6651 		vdc_get_geom_convert},
6652 	{VD_OP_SET_DISKGEOM,	DKIOCSGEOM,		sizeof (vd_geom_t),
6653 		vdc_set_geom_convert},
6654 	{VD_OP_GET_EFI,		DKIOCGETEFI,		0,
6655 		vdc_get_efi_convert},
6656 	{VD_OP_SET_EFI,		DKIOCSETEFI,		0,
6657 		vdc_set_efi_convert},
6658 
6659 	/* DIOCTL_RWCMD is converted to a read or a write */
6660 	{0, DIOCTL_RWCMD,  sizeof (struct dadkio_rwcmd), NULL},
6661 
6662 	/* mhd(7I) non-shared multihost disks ioctls */
6663 	{0, MHIOCTKOWN,				0, vdc_null_copy_func},
6664 	{0, MHIOCRELEASE,			0, vdc_null_copy_func},
6665 	{0, MHIOCSTATUS,			0, vdc_null_copy_func},
6666 	{0, MHIOCQRESERVE,			0, vdc_null_copy_func},
6667 
6668 	/* mhd(7I) shared multihost disks ioctls */
6669 	{0, MHIOCGRP_INKEYS,			0, vdc_null_copy_func},
6670 	{0, MHIOCGRP_INRESV,			0, vdc_null_copy_func},
6671 	{0, MHIOCGRP_REGISTER,			0, vdc_null_copy_func},
6672 	{0, MHIOCGRP_RESERVE, 			0, vdc_null_copy_func},
6673 	{0, MHIOCGRP_PREEMPTANDABORT,		0, vdc_null_copy_func},
6674 	{0, MHIOCGRP_REGISTERANDIGNOREKEY,	0, vdc_null_copy_func},
6675 
6676 	/* mhd(7I) failfast ioctl */
6677 	{0, MHIOCENFAILFAST,			0, vdc_null_copy_func},
6678 
6679 	/*
6680 	 * These particular ioctls are not sent to the server - vdc fakes up
6681 	 * the necessary info.
6682 	 */
6683 	{0, DKIOCINFO, sizeof (struct dk_cinfo), vdc_null_copy_func},
6684 	{0, DKIOCGMEDIAINFO, sizeof (struct dk_minfo), vdc_null_copy_func},
6685 	{0, USCSICMD,	sizeof (struct uscsi_cmd), vdc_null_copy_func},
6686 	{0, DKIOCPARTITION, 0, vdc_null_copy_func },
6687 	{0, DKIOCGAPART, 0, vdc_null_copy_func },
6688 	{0, DKIOCREMOVABLE, 0, vdc_null_copy_func},
6689 	{0, CDROMREADOFFSET, 0, vdc_null_copy_func}
6690 };
6691 
6692 /*
6693  * This function handles ioctl requests from the vd_efi_alloc_and_read()
6694  * function and forward them to the vdisk.
6695  */
6696 static int
6697 vd_process_efi_ioctl(void *vdisk, int cmd, uintptr_t arg)
6698 {
6699 	vdc_t *vdc = (vdc_t *)vdisk;
6700 	dev_t dev;
6701 	int rval;
6702 
6703 	dev = makedevice(ddi_driver_major(vdc->dip),
6704 	    VD_MAKE_DEV(vdc->instance, 0));
6705 
6706 	return (vd_process_ioctl(dev, cmd, (caddr_t)arg, FKIOCTL, &rval));
6707 }
6708 
6709 /*
6710  * Function:
6711  *	vd_process_ioctl()
6712  *
6713  * Description:
6714  *	This routine processes disk specific ioctl calls
6715  *
6716  * Arguments:
6717  *	dev	- the device number
6718  *	cmd	- the operation [dkio(7I)] to be processed
6719  *	arg	- pointer to user provided structure
6720  *		  (contains data to be set or reference parameter for get)
6721  *	mode	- bit flag, indicating open settings, 32/64 bit type, etc
6722  *	rvalp	- pointer to return value for calling process.
6723  *
6724  * Return Code:
6725  *	0
6726  *	EFAULT
6727  *	ENXIO
6728  *	EIO
6729  *	ENOTSUP
6730  */
6731 static int
6732 vd_process_ioctl(dev_t dev, int cmd, caddr_t arg, int mode, int *rvalp)
6733 {
6734 	int		instance = VDCUNIT(dev);
6735 	vdc_t		*vdc = NULL;
6736 	int		rv = -1;
6737 	int		idx = 0;		/* index into dk_ioctl[] */
6738 	size_t		len = 0;		/* #bytes to send to vds */
6739 	size_t		alloc_len = 0;		/* #bytes to allocate mem for */
6740 	caddr_t		mem_p = NULL;
6741 	size_t		nioctls = (sizeof (dk_ioctl)) / (sizeof (dk_ioctl[0]));
6742 	vdc_dk_ioctl_t	*iop;
6743 
6744 	vdc = ddi_get_soft_state(vdc_state, instance);
6745 	if (vdc == NULL) {
6746 		cmn_err(CE_NOTE, "![%d] Could not get soft state structure",
6747 		    instance);
6748 		return (ENXIO);
6749 	}
6750 
6751 	DMSG(vdc, 0, "[%d] Processing ioctl(%x) for dev %lx : model %x\n",
6752 	    instance, cmd, dev, ddi_model_convert_from(mode & FMODELS));
6753 
6754 	if (rvalp != NULL) {
6755 		/* the return value of the ioctl is 0 by default */
6756 		*rvalp = 0;
6757 	}
6758 
6759 	/*
6760 	 * Validate the ioctl operation to be performed.
6761 	 *
6762 	 * If we have looped through the array without finding a match then we
6763 	 * don't support this ioctl.
6764 	 */
6765 	for (idx = 0; idx < nioctls; idx++) {
6766 		if (cmd == dk_ioctl[idx].cmd)
6767 			break;
6768 	}
6769 
6770 	if (idx >= nioctls) {
6771 		DMSG(vdc, 0, "[%d] Unsupported ioctl (0x%x)\n",
6772 		    vdc->instance, cmd);
6773 		return (ENOTSUP);
6774 	}
6775 
6776 	iop = &(dk_ioctl[idx]);
6777 
6778 	if (cmd == DKIOCGETEFI || cmd == DKIOCSETEFI) {
6779 		/* size is not fixed for EFI ioctls, it depends on ioctl arg */
6780 		dk_efi_t	dk_efi;
6781 
6782 		rv = ddi_copyin(arg, &dk_efi, sizeof (dk_efi_t), mode);
6783 		if (rv != 0)
6784 			return (EFAULT);
6785 
6786 		len = sizeof (vd_efi_t) - 1 + dk_efi.dki_length;
6787 	} else {
6788 		len = iop->nbytes;
6789 	}
6790 
6791 	/* check if the ioctl is applicable */
6792 	switch (cmd) {
6793 	case CDROMREADOFFSET:
6794 	case DKIOCREMOVABLE:
6795 		return (ENOTTY);
6796 
6797 	case USCSICMD:
6798 	case MHIOCTKOWN:
6799 	case MHIOCSTATUS:
6800 	case MHIOCQRESERVE:
6801 	case MHIOCRELEASE:
6802 	case MHIOCGRP_INKEYS:
6803 	case MHIOCGRP_INRESV:
6804 	case MHIOCGRP_REGISTER:
6805 	case MHIOCGRP_RESERVE:
6806 	case MHIOCGRP_PREEMPTANDABORT:
6807 	case MHIOCGRP_REGISTERANDIGNOREKEY:
6808 	case MHIOCENFAILFAST:
6809 		if (vdc->cinfo == NULL)
6810 			return (ENXIO);
6811 		if (vdc->cinfo->dki_ctype != DKC_SCSI_CCS)
6812 			return (ENOTTY);
6813 		break;
6814 
6815 	case DIOCTL_RWCMD:
6816 		if (vdc->cinfo == NULL)
6817 			return (ENXIO);
6818 		if (vdc->cinfo->dki_ctype != DKC_DIRECT)
6819 			return (ENOTTY);
6820 		break;
6821 
6822 	case DKIOCINFO:
6823 		if (vdc->cinfo == NULL)
6824 			return (ENXIO);
6825 		break;
6826 
6827 	case DKIOCGMEDIAINFO:
6828 		if (vdc->minfo == NULL)
6829 			return (ENXIO);
6830 		if (vdc_check_capacity(vdc) != 0)
6831 			/* disk capacity is not available */
6832 			return (EIO);
6833 		break;
6834 	}
6835 
6836 	/*
6837 	 * Deal with ioctls which require a processing different than
6838 	 * converting ioctl arguments and sending a corresponding
6839 	 * VD operation.
6840 	 */
6841 	switch (cmd) {
6842 
6843 	case USCSICMD:
6844 	{
6845 		return (vdc_uscsi_cmd(vdc, arg, mode));
6846 	}
6847 
6848 	case MHIOCTKOWN:
6849 	{
6850 		mutex_enter(&vdc->ownership_lock);
6851 		/*
6852 		 * We have to set VDC_OWNERSHIP_WANTED now so that the ownership
6853 		 * can be flagged with VDC_OWNERSHIP_RESET if the LDC is reset
6854 		 * while we are processing the ioctl.
6855 		 */
6856 		vdc_ownership_update(vdc, VDC_OWNERSHIP_WANTED);
6857 
6858 		rv = vdc_access_set(vdc, VD_ACCESS_SET_EXCLUSIVE |
6859 		    VD_ACCESS_SET_PREEMPT | VD_ACCESS_SET_PRESERVE, mode);
6860 		if (rv == 0) {
6861 			vdc_ownership_update(vdc, VDC_OWNERSHIP_WANTED |
6862 			    VDC_OWNERSHIP_GRANTED);
6863 		} else {
6864 			vdc_ownership_update(vdc, VDC_OWNERSHIP_NONE);
6865 		}
6866 		mutex_exit(&vdc->ownership_lock);
6867 		return (rv);
6868 	}
6869 
6870 	case MHIOCRELEASE:
6871 	{
6872 		mutex_enter(&vdc->ownership_lock);
6873 		rv = vdc_access_set(vdc, VD_ACCESS_SET_CLEAR, mode);
6874 		if (rv == 0) {
6875 			vdc_ownership_update(vdc, VDC_OWNERSHIP_NONE);
6876 		}
6877 		mutex_exit(&vdc->ownership_lock);
6878 		return (rv);
6879 	}
6880 
6881 	case MHIOCSTATUS:
6882 	{
6883 		uint64_t status;
6884 
6885 		rv = vdc_access_get(vdc, &status, mode);
6886 		if (rv == 0 && rvalp != NULL)
6887 			*rvalp = (status & VD_ACCESS_ALLOWED)? 0 : 1;
6888 		return (rv);
6889 	}
6890 
6891 	case MHIOCQRESERVE:
6892 	{
6893 		rv = vdc_access_set(vdc, VD_ACCESS_SET_EXCLUSIVE, mode);
6894 		return (rv);
6895 	}
6896 
6897 	case MHIOCGRP_INKEYS:
6898 	{
6899 		return (vdc_mhd_inkeys(vdc, arg, mode));
6900 	}
6901 
6902 	case MHIOCGRP_INRESV:
6903 	{
6904 		return (vdc_mhd_inresv(vdc, arg, mode));
6905 	}
6906 
6907 	case MHIOCGRP_REGISTER:
6908 	{
6909 		return (vdc_mhd_register(vdc, arg, mode));
6910 	}
6911 
6912 	case MHIOCGRP_RESERVE:
6913 	{
6914 		return (vdc_mhd_reserve(vdc, arg, mode));
6915 	}
6916 
6917 	case MHIOCGRP_PREEMPTANDABORT:
6918 	{
6919 		return (vdc_mhd_preemptabort(vdc, arg, mode));
6920 	}
6921 
6922 	case MHIOCGRP_REGISTERANDIGNOREKEY:
6923 	{
6924 		return (vdc_mhd_registerignore(vdc, arg, mode));
6925 	}
6926 
6927 	case MHIOCENFAILFAST:
6928 	{
6929 		rv = vdc_failfast(vdc, arg, mode);
6930 		return (rv);
6931 	}
6932 
6933 	case DIOCTL_RWCMD:
6934 	{
6935 		return (vdc_dioctl_rwcmd(dev, arg, mode));
6936 	}
6937 
6938 	case DKIOCGAPART:
6939 	{
6940 		return (vdc_dkio_gapart(vdc, arg, mode));
6941 	}
6942 
6943 	case DKIOCPARTITION:
6944 	{
6945 		return (vdc_dkio_partition(vdc, arg, mode));
6946 	}
6947 
6948 	case DKIOCINFO:
6949 	{
6950 		struct dk_cinfo	cinfo;
6951 
6952 		bcopy(vdc->cinfo, &cinfo, sizeof (struct dk_cinfo));
6953 		cinfo.dki_partition = VDCPART(dev);
6954 
6955 		rv = ddi_copyout(&cinfo, (void *)arg,
6956 		    sizeof (struct dk_cinfo), mode);
6957 		if (rv != 0)
6958 			return (EFAULT);
6959 
6960 		return (0);
6961 	}
6962 
6963 	case DKIOCGMEDIAINFO:
6964 	{
6965 		ASSERT(vdc->vdisk_size != 0);
6966 		if (vdc->minfo->dki_capacity == 0)
6967 			vdc->minfo->dki_capacity = vdc->vdisk_size;
6968 		rv = ddi_copyout(vdc->minfo, (void *)arg,
6969 		    sizeof (struct dk_minfo), mode);
6970 		if (rv != 0)
6971 			return (EFAULT);
6972 
6973 		return (0);
6974 	}
6975 
6976 	case DKIOCFLUSHWRITECACHE:
6977 		{
6978 			struct dk_callback *dkc =
6979 			    (struct dk_callback *)(uintptr_t)arg;
6980 			vdc_dk_arg_t	*dkarg = NULL;
6981 
6982 			DMSG(vdc, 1, "[%d] Flush W$: mode %x\n",
6983 			    instance, mode);
6984 
6985 			/*
6986 			 * If arg is NULL, then there is no callback function
6987 			 * registered and the call operates synchronously; we
6988 			 * break and continue with the rest of the function and
6989 			 * wait for vds to return (i.e. after the request to
6990 			 * vds returns successfully, all writes completed prior
6991 			 * to the ioctl will have been flushed from the disk
6992 			 * write cache to persistent media.
6993 			 *
6994 			 * If a callback function is registered, we dispatch
6995 			 * the request on a task queue and return immediately.
6996 			 * The callback will deal with informing the calling
6997 			 * thread that the flush request is completed.
6998 			 */
6999 			if (dkc == NULL)
7000 				break;
7001 
7002 			/*
7003 			 * the asynchronous callback is only supported if
7004 			 * invoked from within the kernel
7005 			 */
7006 			if ((mode & FKIOCTL) == 0)
7007 				return (ENOTSUP);
7008 
7009 			dkarg = kmem_zalloc(sizeof (vdc_dk_arg_t), KM_SLEEP);
7010 
7011 			dkarg->mode = mode;
7012 			dkarg->dev = dev;
7013 			bcopy(dkc, &dkarg->dkc, sizeof (*dkc));
7014 
7015 			mutex_enter(&vdc->lock);
7016 			vdc->dkio_flush_pending++;
7017 			dkarg->vdc = vdc;
7018 			mutex_exit(&vdc->lock);
7019 
7020 			/* put the request on a task queue */
7021 			rv = taskq_dispatch(system_taskq, vdc_dkio_flush_cb,
7022 			    (void *)dkarg, DDI_SLEEP);
7023 			if (rv == NULL) {
7024 				/* clean up if dispatch fails */
7025 				mutex_enter(&vdc->lock);
7026 				vdc->dkio_flush_pending--;
7027 				mutex_exit(&vdc->lock);
7028 				kmem_free(dkarg, sizeof (vdc_dk_arg_t));
7029 			}
7030 
7031 			return (rv == NULL ? ENOMEM : 0);
7032 		}
7033 	}
7034 
7035 	/* catch programming error in vdc - should be a VD_OP_XXX ioctl */
7036 	ASSERT(iop->op != 0);
7037 
7038 	/* check if the vDisk server handles the operation for this vDisk */
7039 	if (VD_OP_SUPPORTED(vdc->operations, iop->op) == B_FALSE) {
7040 		DMSG(vdc, 0, "[%d] Unsupported VD_OP operation (0x%x)\n",
7041 		    vdc->instance, iop->op);
7042 		return (ENOTSUP);
7043 	}
7044 
7045 	/* LDC requires that the memory being mapped is 8-byte aligned */
7046 	alloc_len = P2ROUNDUP(len, sizeof (uint64_t));
7047 	DMSG(vdc, 1, "[%d] struct size %ld alloc %ld\n",
7048 	    instance, len, alloc_len);
7049 
7050 	if (alloc_len > 0)
7051 		mem_p = kmem_zalloc(alloc_len, KM_SLEEP);
7052 
7053 	/*
7054 	 * Call the conversion function for this ioctl which, if necessary,
7055 	 * converts from the Solaris format to the format ARC'ed
7056 	 * as part of the vDisk protocol (FWARC 2006/195)
7057 	 */
7058 	ASSERT(iop->convert != NULL);
7059 	rv = (iop->convert)(vdc, arg, mem_p, mode, VD_COPYIN);
7060 	if (rv != 0) {
7061 		DMSG(vdc, 0, "[%d] convert func returned %d for ioctl 0x%x\n",
7062 		    instance, rv, cmd);
7063 		if (mem_p != NULL)
7064 			kmem_free(mem_p, alloc_len);
7065 		return (rv);
7066 	}
7067 
7068 	/*
7069 	 * send request to vds to service the ioctl.
7070 	 */
7071 	rv = vdc_do_sync_op(vdc, iop->op, mem_p, alloc_len,
7072 	    VDCPART(dev), 0, CB_SYNC, (void *)(uint64_t)mode,
7073 	    VIO_both_dir, B_TRUE);
7074 
7075 	if (rv != 0) {
7076 		/*
7077 		 * This is not necessarily an error. The ioctl could
7078 		 * be returning a value such as ENOTTY to indicate
7079 		 * that the ioctl is not applicable.
7080 		 */
7081 		DMSG(vdc, 0, "[%d] vds returned %d for ioctl 0x%x\n",
7082 		    instance, rv, cmd);
7083 		if (mem_p != NULL)
7084 			kmem_free(mem_p, alloc_len);
7085 
7086 		return (rv);
7087 	}
7088 
7089 	/*
7090 	 * Call the conversion function (if it exists) for this ioctl
7091 	 * which converts from the format ARC'ed as part of the vDisk
7092 	 * protocol (FWARC 2006/195) back to a format understood by
7093 	 * the rest of Solaris.
7094 	 */
7095 	rv = (iop->convert)(vdc, mem_p, arg, mode, VD_COPYOUT);
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 	if (mem_p != NULL)
7105 		kmem_free(mem_p, alloc_len);
7106 
7107 	return (rv);
7108 }
7109 
7110 /*
7111  * Function:
7112  *
7113  * Description:
7114  *	This is an empty conversion function used by ioctl calls which
7115  *	do not need to convert the data being passed in/out to userland
7116  */
7117 static int
7118 vdc_null_copy_func(vdc_t *vdc, void *from, void *to, int mode, int dir)
7119 {
7120 	_NOTE(ARGUNUSED(vdc))
7121 	_NOTE(ARGUNUSED(from))
7122 	_NOTE(ARGUNUSED(to))
7123 	_NOTE(ARGUNUSED(mode))
7124 	_NOTE(ARGUNUSED(dir))
7125 
7126 	return (0);
7127 }
7128 
7129 static int
7130 vdc_get_wce_convert(vdc_t *vdc, void *from, void *to,
7131     int mode, int dir)
7132 {
7133 	_NOTE(ARGUNUSED(vdc))
7134 
7135 	if (dir == VD_COPYIN)
7136 		return (0);		/* nothing to do */
7137 
7138 	if (ddi_copyout(from, to, sizeof (int), mode) != 0)
7139 		return (EFAULT);
7140 
7141 	return (0);
7142 }
7143 
7144 static int
7145 vdc_set_wce_convert(vdc_t *vdc, void *from, void *to,
7146     int mode, int dir)
7147 {
7148 	_NOTE(ARGUNUSED(vdc))
7149 
7150 	if (dir == VD_COPYOUT)
7151 		return (0);		/* nothing to do */
7152 
7153 	if (ddi_copyin(from, to, sizeof (int), mode) != 0)
7154 		return (EFAULT);
7155 
7156 	return (0);
7157 }
7158 
7159 /*
7160  * Function:
7161  *	vdc_get_vtoc_convert()
7162  *
7163  * Description:
7164  *	This routine performs the necessary convertions from the DKIOCGVTOC
7165  *	Solaris structure to the format defined in FWARC 2006/195.
7166  *
7167  *	In the struct vtoc definition, the timestamp field is marked as not
7168  *	supported so it is not part of vDisk protocol (FWARC 2006/195).
7169  *	However SVM uses that field to check it can write into the VTOC,
7170  *	so we fake up the info of that field.
7171  *
7172  * Arguments:
7173  *	vdc	- the vDisk client
7174  *	from	- the buffer containing the data to be copied from
7175  *	to	- the buffer to be copied to
7176  *	mode	- flags passed to ioctl() call
7177  *	dir	- the "direction" of the copy - VD_COPYIN or VD_COPYOUT
7178  *
7179  * Return Code:
7180  *	0	- Success
7181  *	ENXIO	- incorrect buffer passed in.
7182  *	EFAULT	- ddi_copyout routine encountered an error.
7183  */
7184 static int
7185 vdc_get_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7186 {
7187 	int		i;
7188 	void		*tmp_mem = NULL;
7189 	void		*tmp_memp;
7190 	struct vtoc	vt;
7191 	struct vtoc32	vt32;
7192 	int		copy_len = 0;
7193 	int		rv = 0;
7194 
7195 	if (dir != VD_COPYOUT)
7196 		return (0);	/* nothing to do */
7197 
7198 	if ((from == NULL) || (to == NULL))
7199 		return (ENXIO);
7200 
7201 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32)
7202 		copy_len = sizeof (struct vtoc32);
7203 	else
7204 		copy_len = sizeof (struct vtoc);
7205 
7206 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
7207 
7208 	VD_VTOC2VTOC((vd_vtoc_t *)from, &vt);
7209 
7210 	/* fake the VTOC timestamp field */
7211 	for (i = 0; i < V_NUMPAR; i++) {
7212 		vt.timestamp[i] = vdc->vtoc->timestamp[i];
7213 	}
7214 
7215 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
7216 		/* LINTED E_ASSIGN_NARROW_CONV */
7217 		vtoctovtoc32(vt, vt32);
7218 		tmp_memp = &vt32;
7219 	} else {
7220 		tmp_memp = &vt;
7221 	}
7222 	rv = ddi_copyout(tmp_memp, to, copy_len, mode);
7223 	if (rv != 0)
7224 		rv = EFAULT;
7225 
7226 	kmem_free(tmp_mem, copy_len);
7227 	return (rv);
7228 }
7229 
7230 /*
7231  * Function:
7232  *	vdc_set_vtoc_convert()
7233  *
7234  * Description:
7235  *	This routine performs the necessary convertions from the DKIOCSVTOC
7236  *	Solaris structure to the format defined in FWARC 2006/195.
7237  *
7238  * Arguments:
7239  *	vdc	- the vDisk client
7240  *	from	- Buffer with data
7241  *	to	- Buffer where data is to be copied to
7242  *	mode	- flags passed to ioctl
7243  *	dir	- direction of copy (in or out)
7244  *
7245  * Return Code:
7246  *	0	- Success
7247  *	ENXIO	- Invalid buffer passed in
7248  *	EFAULT	- ddi_copyin of data failed
7249  */
7250 static int
7251 vdc_set_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7252 {
7253 	_NOTE(ARGUNUSED(vdc))
7254 
7255 	void		*tmp_mem = NULL, *uvtoc;
7256 	struct vtoc	vt;
7257 	struct vtoc	*vtp = &vt;
7258 	vd_vtoc_t	vtvd;
7259 	int		copy_len = 0;
7260 	int		i, rv = 0;
7261 
7262 	if ((from == NULL) || (to == NULL))
7263 		return (ENXIO);
7264 
7265 	if (dir == VD_COPYIN)
7266 		uvtoc = from;
7267 	else
7268 		uvtoc = to;
7269 
7270 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32)
7271 		copy_len = sizeof (struct vtoc32);
7272 	else
7273 		copy_len = sizeof (struct vtoc);
7274 
7275 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
7276 
7277 	rv = ddi_copyin(uvtoc, tmp_mem, copy_len, mode);
7278 	if (rv != 0) {
7279 		kmem_free(tmp_mem, copy_len);
7280 		return (EFAULT);
7281 	}
7282 
7283 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
7284 		vtoc32tovtoc((*(struct vtoc32 *)tmp_mem), vt);
7285 	} else {
7286 		vtp = tmp_mem;
7287 	}
7288 
7289 	if (dir == VD_COPYOUT) {
7290 		/*
7291 		 * The disk label may have changed. Revalidate the disk
7292 		 * geometry. This will also update the device nodes and
7293 		 * properties.
7294 		 */
7295 		vdc_validate(vdc);
7296 
7297 		/*
7298 		 * We also need to keep track of the timestamp fields.
7299 		 */
7300 		for (i = 0; i < V_NUMPAR; i++) {
7301 			vdc->vtoc->timestamp[i] = vtp->timestamp[i];
7302 		}
7303 
7304 		return (0);
7305 	}
7306 
7307 	VTOC2VD_VTOC(vtp, &vtvd);
7308 	bcopy(&vtvd, to, sizeof (vd_vtoc_t));
7309 	kmem_free(tmp_mem, copy_len);
7310 
7311 	return (0);
7312 }
7313 
7314 /*
7315  * Function:
7316  *	vdc_get_geom_convert()
7317  *
7318  * Description:
7319  *	This routine performs the necessary convertions from the DKIOCGGEOM,
7320  *	DKIOCG_PHYSGEOM and DKIOG_VIRTGEOM Solaris structures to the format
7321  *	defined in FWARC 2006/195
7322  *
7323  * Arguments:
7324  *	vdc	- the vDisk client
7325  *	from	- Buffer with data
7326  *	to	- Buffer where data is to be copied to
7327  *	mode	- flags passed to ioctl
7328  *	dir	- direction of copy (in or out)
7329  *
7330  * Return Code:
7331  *	0	- Success
7332  *	ENXIO	- Invalid buffer passed in
7333  *	EFAULT	- ddi_copyout of data failed
7334  */
7335 static int
7336 vdc_get_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7337 {
7338 	_NOTE(ARGUNUSED(vdc))
7339 
7340 	struct dk_geom	geom;
7341 	int	copy_len = sizeof (struct dk_geom);
7342 	int	rv = 0;
7343 
7344 	if (dir != VD_COPYOUT)
7345 		return (0);	/* nothing to do */
7346 
7347 	if ((from == NULL) || (to == NULL))
7348 		return (ENXIO);
7349 
7350 	VD_GEOM2DK_GEOM((vd_geom_t *)from, &geom);
7351 	rv = ddi_copyout(&geom, to, copy_len, mode);
7352 	if (rv != 0)
7353 		rv = EFAULT;
7354 
7355 	return (rv);
7356 }
7357 
7358 /*
7359  * Function:
7360  *	vdc_set_geom_convert()
7361  *
7362  * Description:
7363  *	This routine performs the necessary convertions from the DKIOCSGEOM
7364  *	Solaris structure to the format defined in FWARC 2006/195.
7365  *
7366  * Arguments:
7367  *	vdc	- the vDisk client
7368  *	from	- Buffer with data
7369  *	to	- Buffer where data is to be copied to
7370  *	mode	- flags passed to ioctl
7371  *	dir	- direction of copy (in or out)
7372  *
7373  * Return Code:
7374  *	0	- Success
7375  *	ENXIO	- Invalid buffer passed in
7376  *	EFAULT	- ddi_copyin of data failed
7377  */
7378 static int
7379 vdc_set_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7380 {
7381 	_NOTE(ARGUNUSED(vdc))
7382 
7383 	vd_geom_t	vdgeom;
7384 	void		*tmp_mem = NULL;
7385 	int		copy_len = sizeof (struct dk_geom);
7386 	int		rv = 0;
7387 
7388 	if (dir != VD_COPYIN)
7389 		return (0);	/* nothing to do */
7390 
7391 	if ((from == NULL) || (to == NULL))
7392 		return (ENXIO);
7393 
7394 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
7395 
7396 	rv = ddi_copyin(from, tmp_mem, copy_len, mode);
7397 	if (rv != 0) {
7398 		kmem_free(tmp_mem, copy_len);
7399 		return (EFAULT);
7400 	}
7401 	DK_GEOM2VD_GEOM((struct dk_geom *)tmp_mem, &vdgeom);
7402 	bcopy(&vdgeom, to, sizeof (vdgeom));
7403 	kmem_free(tmp_mem, copy_len);
7404 
7405 	return (0);
7406 }
7407 
7408 static int
7409 vdc_get_efi_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7410 {
7411 	_NOTE(ARGUNUSED(vdc))
7412 
7413 	vd_efi_t	*vd_efi;
7414 	dk_efi_t	dk_efi;
7415 	int		rv = 0;
7416 	void		*uaddr;
7417 
7418 	if ((from == NULL) || (to == NULL))
7419 		return (ENXIO);
7420 
7421 	if (dir == VD_COPYIN) {
7422 
7423 		vd_efi = (vd_efi_t *)to;
7424 
7425 		rv = ddi_copyin(from, &dk_efi, sizeof (dk_efi_t), mode);
7426 		if (rv != 0)
7427 			return (EFAULT);
7428 
7429 		vd_efi->lba = dk_efi.dki_lba;
7430 		vd_efi->length = dk_efi.dki_length;
7431 		bzero(vd_efi->data, vd_efi->length);
7432 
7433 	} else {
7434 
7435 		rv = ddi_copyin(to, &dk_efi, sizeof (dk_efi_t), mode);
7436 		if (rv != 0)
7437 			return (EFAULT);
7438 
7439 		uaddr = dk_efi.dki_data;
7440 
7441 		dk_efi.dki_data = kmem_alloc(dk_efi.dki_length, KM_SLEEP);
7442 
7443 		VD_EFI2DK_EFI((vd_efi_t *)from, &dk_efi);
7444 
7445 		rv = ddi_copyout(dk_efi.dki_data, uaddr, dk_efi.dki_length,
7446 		    mode);
7447 		if (rv != 0)
7448 			return (EFAULT);
7449 
7450 		kmem_free(dk_efi.dki_data, dk_efi.dki_length);
7451 	}
7452 
7453 	return (0);
7454 }
7455 
7456 static int
7457 vdc_set_efi_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7458 {
7459 	_NOTE(ARGUNUSED(vdc))
7460 
7461 	dk_efi_t	dk_efi;
7462 	void		*uaddr;
7463 
7464 	if (dir == VD_COPYOUT) {
7465 		/*
7466 		 * The disk label may have changed. Revalidate the disk
7467 		 * geometry. This will also update the device nodes and
7468 		 * properties.
7469 		 */
7470 		vdc_validate(vdc);
7471 		return (0);
7472 	}
7473 
7474 	if ((from == NULL) || (to == NULL))
7475 		return (ENXIO);
7476 
7477 	if (ddi_copyin(from, &dk_efi, sizeof (dk_efi_t), mode) != 0)
7478 		return (EFAULT);
7479 
7480 	uaddr = dk_efi.dki_data;
7481 
7482 	dk_efi.dki_data = kmem_alloc(dk_efi.dki_length, KM_SLEEP);
7483 
7484 	if (ddi_copyin(uaddr, dk_efi.dki_data, dk_efi.dki_length, mode) != 0)
7485 		return (EFAULT);
7486 
7487 	DK_EFI2VD_EFI(&dk_efi, (vd_efi_t *)to);
7488 
7489 	kmem_free(dk_efi.dki_data, dk_efi.dki_length);
7490 
7491 	return (0);
7492 }
7493 
7494 
7495 /* -------------------------------------------------------------------------- */
7496 
7497 /*
7498  * Function:
7499  *	vdc_create_fake_geometry()
7500  *
7501  * Description:
7502  *	This routine fakes up the disk info needed for some DKIO ioctls such
7503  *	as DKIOCINFO and DKIOCGMEDIAINFO [just like lofi(7D) and ramdisk(7D) do]
7504  *
7505  *	Note: This function must not be called until the vDisk attributes have
7506  *	been exchanged as part of the handshake with the vDisk server.
7507  *
7508  * Arguments:
7509  *	vdc	- soft state pointer for this instance of the device driver.
7510  *
7511  * Return Code:
7512  *	none.
7513  */
7514 static void
7515 vdc_create_fake_geometry(vdc_t *vdc)
7516 {
7517 	ASSERT(vdc != NULL);
7518 	ASSERT(vdc->max_xfer_sz != 0);
7519 
7520 	/*
7521 	 * DKIOCINFO support
7522 	 */
7523 	if (vdc->cinfo == NULL)
7524 		vdc->cinfo = kmem_zalloc(sizeof (struct dk_cinfo), KM_SLEEP);
7525 
7526 	(void) strcpy(vdc->cinfo->dki_cname, VDC_DRIVER_NAME);
7527 	(void) strcpy(vdc->cinfo->dki_dname, VDC_DRIVER_NAME);
7528 	/* max_xfer_sz is #blocks so we don't need to divide by DEV_BSIZE */
7529 	vdc->cinfo->dki_maxtransfer = vdc->max_xfer_sz;
7530 
7531 	/*
7532 	 * We set the controller type to DKC_SCSI_CCS only if the VD_OP_SCSICMD
7533 	 * operation is supported, otherwise the controller type is DKC_DIRECT.
7534 	 * Version 1.0 does not support the VD_OP_SCSICMD operation, so the
7535 	 * controller type is always DKC_DIRECT in that case.
7536 	 *
7537 	 * If the virtual disk is backed by a physical CD/DVD device or
7538 	 * an ISO image, modify the controller type to indicate this
7539 	 */
7540 	switch (vdc->vdisk_media) {
7541 	case VD_MEDIA_CD:
7542 	case VD_MEDIA_DVD:
7543 		vdc->cinfo->dki_ctype = DKC_CDROM;
7544 		break;
7545 	case VD_MEDIA_FIXED:
7546 		if (VD_OP_SUPPORTED(vdc->operations, VD_OP_SCSICMD))
7547 			vdc->cinfo->dki_ctype = DKC_SCSI_CCS;
7548 		else
7549 			vdc->cinfo->dki_ctype = DKC_DIRECT;
7550 		break;
7551 	default:
7552 		/* in the case of v1.0 we default to a fixed disk */
7553 		vdc->cinfo->dki_ctype = DKC_DIRECT;
7554 		break;
7555 	}
7556 	vdc->cinfo->dki_flags = DKI_FMTVOL;
7557 	vdc->cinfo->dki_cnum = 0;
7558 	vdc->cinfo->dki_addr = 0;
7559 	vdc->cinfo->dki_space = 0;
7560 	vdc->cinfo->dki_prio = 0;
7561 	vdc->cinfo->dki_vec = 0;
7562 	vdc->cinfo->dki_unit = vdc->instance;
7563 	vdc->cinfo->dki_slave = 0;
7564 	/*
7565 	 * The partition number will be created on the fly depending on the
7566 	 * actual slice (i.e. minor node) that is used to request the data.
7567 	 */
7568 	vdc->cinfo->dki_partition = 0;
7569 
7570 	/*
7571 	 * DKIOCGMEDIAINFO support
7572 	 */
7573 	if (vdc->minfo == NULL)
7574 		vdc->minfo = kmem_zalloc(sizeof (struct dk_minfo), KM_SLEEP);
7575 
7576 	if (vio_ver_is_supported(vdc->ver, 1, 1)) {
7577 		vdc->minfo->dki_media_type =
7578 		    VD_MEDIATYPE2DK_MEDIATYPE(vdc->vdisk_media);
7579 	} else {
7580 		vdc->minfo->dki_media_type = DK_FIXED_DISK;
7581 	}
7582 
7583 	vdc->minfo->dki_capacity = vdc->vdisk_size;
7584 	vdc->minfo->dki_lbsize = vdc->block_size;
7585 }
7586 
7587 static ushort_t
7588 vdc_lbl2cksum(struct dk_label *label)
7589 {
7590 	int	count;
7591 	ushort_t sum, *sp;
7592 
7593 	count =	(sizeof (struct dk_label)) / (sizeof (short)) - 1;
7594 	sp = (ushort_t *)label;
7595 	sum = 0;
7596 	while (count--) {
7597 		sum ^= *sp++;
7598 	}
7599 
7600 	return (sum);
7601 }
7602 
7603 /*
7604  * Function:
7605  *	vdc_validate_geometry
7606  *
7607  * Description:
7608  *	This routine discovers the label and geometry of the disk. It stores
7609  *	the disk label and related information in the vdc structure. If it
7610  *	fails to validate the geometry or to discover the disk label then
7611  *	the label is marked as unknown (VD_DISK_LABEL_UNK).
7612  *
7613  * Arguments:
7614  *	vdc	- soft state pointer for this instance of the device driver.
7615  *
7616  * Return Code:
7617  *	0	- success.
7618  *	EINVAL	- unknown disk label.
7619  *	ENOTSUP	- geometry not applicable (EFI label).
7620  *	EIO	- error accessing the disk.
7621  */
7622 static int
7623 vdc_validate_geometry(vdc_t *vdc)
7624 {
7625 	buf_t	*buf;	/* BREAD requests need to be in a buf_t structure */
7626 	dev_t	dev;
7627 	int	rv, rval;
7628 	struct dk_label label;
7629 	struct dk_geom geom;
7630 	struct vtoc vtoc;
7631 	efi_gpt_t *gpt;
7632 	efi_gpe_t *gpe;
7633 	vd_efi_dev_t edev;
7634 
7635 	ASSERT(vdc != NULL);
7636 	ASSERT(vdc->vtoc != NULL && vdc->geom != NULL);
7637 	ASSERT(MUTEX_HELD(&vdc->lock));
7638 
7639 	mutex_exit(&vdc->lock);
7640 
7641 	dev = makedevice(ddi_driver_major(vdc->dip),
7642 	    VD_MAKE_DEV(vdc->instance, 0));
7643 
7644 	rv = vd_process_ioctl(dev, DKIOCGGEOM, (caddr_t)&geom, FKIOCTL, &rval);
7645 	if (rv == 0)
7646 		rv = vd_process_ioctl(dev, DKIOCGVTOC, (caddr_t)&vtoc,
7647 		    FKIOCTL, &rval);
7648 
7649 	if (rv == ENOTSUP) {
7650 		/*
7651 		 * If the device does not support VTOC then we try
7652 		 * to read an EFI label.
7653 		 *
7654 		 * We need to know the block size and the disk size to
7655 		 * be able to read an EFI label.
7656 		 */
7657 		if (vdc->vdisk_size == 0) {
7658 			if ((rv = vdc_check_capacity(vdc)) != 0) {
7659 				mutex_enter(&vdc->lock);
7660 				vdc_store_label_unk(vdc);
7661 				return (rv);
7662 			}
7663 		}
7664 
7665 		VD_EFI_DEV_SET(edev, vdc, vd_process_efi_ioctl);
7666 
7667 		rv = vd_efi_alloc_and_read(&edev, &gpt, &gpe);
7668 
7669 		if (rv) {
7670 			DMSG(vdc, 0, "[%d] Failed to get EFI (err=%d)",
7671 			    vdc->instance, rv);
7672 			mutex_enter(&vdc->lock);
7673 			vdc_store_label_unk(vdc);
7674 			return (EIO);
7675 		}
7676 
7677 		mutex_enter(&vdc->lock);
7678 		vdc_store_label_efi(vdc, gpt, gpe);
7679 		vd_efi_free(&edev, gpt, gpe);
7680 		return (ENOTSUP);
7681 	}
7682 
7683 	if (rv != 0) {
7684 		DMSG(vdc, 0, "[%d] Failed to get VTOC (err=%d)",
7685 		    vdc->instance, rv);
7686 		mutex_enter(&vdc->lock);
7687 		vdc_store_label_unk(vdc);
7688 		if (rv != EINVAL)
7689 			rv = EIO;
7690 		return (rv);
7691 	}
7692 
7693 	/* check that geometry and vtoc are valid */
7694 	if (geom.dkg_nhead == 0 || geom.dkg_nsect == 0 ||
7695 	    vtoc.v_sanity != VTOC_SANE) {
7696 		mutex_enter(&vdc->lock);
7697 		vdc_store_label_unk(vdc);
7698 		return (EINVAL);
7699 	}
7700 
7701 	/*
7702 	 * We have a disk and a valid VTOC. However this does not mean
7703 	 * that the disk currently have a VTOC label. The returned VTOC may
7704 	 * be a default VTOC to be used for configuring the disk (this is
7705 	 * what is done for disk image). So we read the label from the
7706 	 * beginning of the disk to ensure we really have a VTOC label.
7707 	 *
7708 	 * FUTURE: This could be the default way for reading the VTOC
7709 	 * from the disk as opposed to sending the VD_OP_GET_VTOC
7710 	 * to the server. This will be the default if vdc is implemented
7711 	 * ontop of cmlb.
7712 	 */
7713 
7714 	/*
7715 	 * Single slice disk does not support read using an absolute disk
7716 	 * offset so we just rely on the DKIOCGVTOC ioctl in that case.
7717 	 */
7718 	if (vdc->vdisk_type == VD_DISK_TYPE_SLICE) {
7719 		mutex_enter(&vdc->lock);
7720 		if (vtoc.v_nparts != 1) {
7721 			vdc_store_label_unk(vdc);
7722 			return (EINVAL);
7723 		}
7724 		vdc_store_label_vtoc(vdc, &geom, &vtoc);
7725 		return (0);
7726 	}
7727 
7728 	if (vtoc.v_nparts != V_NUMPAR) {
7729 		mutex_enter(&vdc->lock);
7730 		vdc_store_label_unk(vdc);
7731 		return (EINVAL);
7732 	}
7733 
7734 	/*
7735 	 * Read disk label from start of disk
7736 	 */
7737 	buf = kmem_alloc(sizeof (buf_t), KM_SLEEP);
7738 	bioinit(buf);
7739 	buf->b_un.b_addr = (caddr_t)&label;
7740 	buf->b_bcount = DK_LABEL_SIZE;
7741 	buf->b_flags = B_BUSY | B_READ;
7742 	buf->b_dev = cmpdev(dev);
7743 	rv = vdc_send_request(vdc, VD_OP_BREAD, (caddr_t)&label,
7744 	    DK_LABEL_SIZE, VD_SLICE_NONE, 0, CB_STRATEGY, buf, VIO_read_dir);
7745 	if (rv) {
7746 		DMSG(vdc, 1, "[%d] Failed to read disk block 0\n",
7747 		    vdc->instance);
7748 	} else {
7749 		rv = biowait(buf);
7750 		biofini(buf);
7751 	}
7752 	kmem_free(buf, sizeof (buf_t));
7753 
7754 	if (rv != 0 || label.dkl_magic != DKL_MAGIC ||
7755 	    label.dkl_cksum != vdc_lbl2cksum(&label)) {
7756 		DMSG(vdc, 1, "[%d] Got VTOC with invalid label\n",
7757 		    vdc->instance);
7758 		mutex_enter(&vdc->lock);
7759 		vdc_store_label_unk(vdc);
7760 		return (EINVAL);
7761 	}
7762 
7763 	mutex_enter(&vdc->lock);
7764 	vdc_store_label_vtoc(vdc, &geom, &vtoc);
7765 	return (0);
7766 }
7767 
7768 /*
7769  * Function:
7770  *	vdc_validate
7771  *
7772  * Description:
7773  *	This routine discovers the label of the disk and create the
7774  *	appropriate device nodes if the label has changed.
7775  *
7776  * Arguments:
7777  *	vdc	- soft state pointer for this instance of the device driver.
7778  *
7779  * Return Code:
7780  *	none.
7781  */
7782 static void
7783 vdc_validate(vdc_t *vdc)
7784 {
7785 	vd_disk_label_t old_label;
7786 	vd_slice_t old_slice[V_NUMPAR];
7787 	int rv;
7788 
7789 	ASSERT(!MUTEX_HELD(&vdc->lock));
7790 
7791 	mutex_enter(&vdc->lock);
7792 
7793 	/* save the current label and vtoc */
7794 	old_label = vdc->vdisk_label;
7795 	bcopy(vdc->slice, &old_slice, sizeof (vd_slice_t) * V_NUMPAR);
7796 
7797 	/* check the geometry */
7798 	(void) vdc_validate_geometry(vdc);
7799 
7800 	/* if the disk label has changed, update device nodes */
7801 	if (vdc->vdisk_label != old_label) {
7802 
7803 		if (vdc->vdisk_label == VD_DISK_LABEL_EFI)
7804 			rv = vdc_create_device_nodes_efi(vdc);
7805 		else
7806 			rv = vdc_create_device_nodes_vtoc(vdc);
7807 
7808 		if (rv != 0) {
7809 			DMSG(vdc, 0, "![%d] Failed to update device nodes",
7810 			    vdc->instance);
7811 		}
7812 	}
7813 
7814 	/* if the vtoc has changed, update device nodes properties */
7815 	if (bcmp(vdc->slice, &old_slice, sizeof (vd_slice_t) * V_NUMPAR) != 0) {
7816 
7817 		if (vdc_create_device_nodes_props(vdc) != 0) {
7818 			DMSG(vdc, 0, "![%d] Failed to update device nodes"
7819 			    " properties", vdc->instance);
7820 		}
7821 	}
7822 
7823 	mutex_exit(&vdc->lock);
7824 }
7825 
7826 static void
7827 vdc_validate_task(void *arg)
7828 {
7829 	vdc_t *vdc = (vdc_t *)arg;
7830 
7831 	vdc_validate(vdc);
7832 
7833 	mutex_enter(&vdc->lock);
7834 	ASSERT(vdc->validate_pending > 0);
7835 	vdc->validate_pending--;
7836 	mutex_exit(&vdc->lock);
7837 }
7838 
7839 /*
7840  * Function:
7841  *	vdc_setup_devid()
7842  *
7843  * Description:
7844  *	This routine discovers the devid of a vDisk. It requests the devid of
7845  *	the underlying device from the vDisk server, builds an encapsulated
7846  *	devid based on the retrieved devid and registers that new devid to
7847  *	the vDisk.
7848  *
7849  * Arguments:
7850  *	vdc	- soft state pointer for this instance of the device driver.
7851  *
7852  * Return Code:
7853  *	0	- A devid was succesfully registered for the vDisk
7854  */
7855 static int
7856 vdc_setup_devid(vdc_t *vdc)
7857 {
7858 	int rv;
7859 	vd_devid_t *vd_devid;
7860 	size_t bufsize, bufid_len;
7861 
7862 	/*
7863 	 * At first sight, we don't know the size of the devid that the
7864 	 * server will return but this size will be encoded into the
7865 	 * reply. So we do a first request using a default size then we
7866 	 * check if this size was large enough. If not then we do a second
7867 	 * request with the correct size returned by the server. Note that
7868 	 * ldc requires size to be 8-byte aligned.
7869 	 */
7870 	bufsize = P2ROUNDUP(VD_DEVID_SIZE(VD_DEVID_DEFAULT_LEN),
7871 	    sizeof (uint64_t));
7872 	vd_devid = kmem_zalloc(bufsize, KM_SLEEP);
7873 	bufid_len = bufsize - sizeof (vd_efi_t) - 1;
7874 
7875 	rv = vdc_do_sync_op(vdc, VD_OP_GET_DEVID, (caddr_t)vd_devid,
7876 	    bufsize, 0, 0, CB_SYNC, 0, VIO_both_dir, B_TRUE);
7877 
7878 	DMSG(vdc, 2, "sync_op returned %d\n", rv);
7879 
7880 	if (rv) {
7881 		kmem_free(vd_devid, bufsize);
7882 		return (rv);
7883 	}
7884 
7885 	if (vd_devid->length > bufid_len) {
7886 		/*
7887 		 * The returned devid is larger than the buffer used. Try again
7888 		 * with a buffer with the right size.
7889 		 */
7890 		kmem_free(vd_devid, bufsize);
7891 		bufsize = P2ROUNDUP(VD_DEVID_SIZE(vd_devid->length),
7892 		    sizeof (uint64_t));
7893 		vd_devid = kmem_zalloc(bufsize, KM_SLEEP);
7894 		bufid_len = bufsize - sizeof (vd_efi_t) - 1;
7895 
7896 		rv = vdc_do_sync_op(vdc, VD_OP_GET_DEVID,
7897 		    (caddr_t)vd_devid, bufsize, 0, 0, CB_SYNC, 0,
7898 		    VIO_both_dir, B_TRUE);
7899 
7900 		if (rv) {
7901 			kmem_free(vd_devid, bufsize);
7902 			return (rv);
7903 		}
7904 	}
7905 
7906 	/*
7907 	 * The virtual disk should have the same device id as the one associated
7908 	 * with the physical disk it is mapped on, otherwise sharing a disk
7909 	 * between a LDom and a non-LDom may not work (for example for a shared
7910 	 * SVM disk set).
7911 	 *
7912 	 * The DDI framework does not allow creating a device id with any
7913 	 * type so we first create a device id of type DEVID_ENCAP and then
7914 	 * we restore the orignal type of the physical device.
7915 	 */
7916 
7917 	DMSG(vdc, 2, ": devid length = %d\n", vd_devid->length);
7918 
7919 	/* build an encapsulated devid based on the returned devid */
7920 	if (ddi_devid_init(vdc->dip, DEVID_ENCAP, vd_devid->length,
7921 	    vd_devid->id, &vdc->devid) != DDI_SUCCESS) {
7922 		DMSG(vdc, 1, "[%d] Fail to created devid\n", vdc->instance);
7923 		kmem_free(vd_devid, bufsize);
7924 		return (1);
7925 	}
7926 
7927 	DEVID_FORMTYPE((impl_devid_t *)vdc->devid, vd_devid->type);
7928 
7929 	ASSERT(ddi_devid_valid(vdc->devid) == DDI_SUCCESS);
7930 
7931 	kmem_free(vd_devid, bufsize);
7932 
7933 	if (ddi_devid_register(vdc->dip, vdc->devid) != DDI_SUCCESS) {
7934 		DMSG(vdc, 1, "[%d] Fail to register devid\n", vdc->instance);
7935 		return (1);
7936 	}
7937 
7938 	return (0);
7939 }
7940 
7941 static void
7942 vdc_store_label_efi(vdc_t *vdc, efi_gpt_t *gpt, efi_gpe_t *gpe)
7943 {
7944 	int i, nparts;
7945 
7946 	ASSERT(MUTEX_HELD(&vdc->lock));
7947 
7948 	vdc->vdisk_label = VD_DISK_LABEL_EFI;
7949 	bzero(vdc->vtoc, sizeof (struct vtoc));
7950 	bzero(vdc->geom, sizeof (struct dk_geom));
7951 	bzero(vdc->slice, sizeof (vd_slice_t) * V_NUMPAR);
7952 
7953 	nparts = gpt->efi_gpt_NumberOfPartitionEntries;
7954 
7955 	for (i = 0; i < nparts && i < VD_EFI_WD_SLICE; i++) {
7956 
7957 		if (gpe[i].efi_gpe_StartingLBA == 0 ||
7958 		    gpe[i].efi_gpe_EndingLBA == 0) {
7959 			continue;
7960 		}
7961 
7962 		vdc->slice[i].start = gpe[i].efi_gpe_StartingLBA;
7963 		vdc->slice[i].nblocks = gpe[i].efi_gpe_EndingLBA -
7964 		    gpe[i].efi_gpe_StartingLBA + 1;
7965 	}
7966 
7967 	ASSERT(vdc->vdisk_size != 0);
7968 	vdc->slice[VD_EFI_WD_SLICE].start = 0;
7969 	vdc->slice[VD_EFI_WD_SLICE].nblocks = vdc->vdisk_size;
7970 
7971 }
7972 
7973 static void
7974 vdc_store_label_vtoc(vdc_t *vdc, struct dk_geom *geom, struct vtoc *vtoc)
7975 {
7976 	int i;
7977 
7978 	ASSERT(MUTEX_HELD(&vdc->lock));
7979 	ASSERT(vdc->block_size == vtoc->v_sectorsz);
7980 
7981 	vdc->vdisk_label = VD_DISK_LABEL_VTOC;
7982 	bcopy(vtoc, vdc->vtoc, sizeof (struct vtoc));
7983 	bcopy(geom, vdc->geom, sizeof (struct dk_geom));
7984 	bzero(vdc->slice, sizeof (vd_slice_t) * V_NUMPAR);
7985 
7986 	for (i = 0; i < vtoc->v_nparts; i++) {
7987 		vdc->slice[i].start = vtoc->v_part[i].p_start;
7988 		vdc->slice[i].nblocks = vtoc->v_part[i].p_size;
7989 	}
7990 }
7991 
7992 static void
7993 vdc_store_label_unk(vdc_t *vdc)
7994 {
7995 	ASSERT(MUTEX_HELD(&vdc->lock));
7996 
7997 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
7998 	bzero(vdc->vtoc, sizeof (struct vtoc));
7999 	bzero(vdc->geom, sizeof (struct dk_geom));
8000 	bzero(vdc->slice, sizeof (vd_slice_t) * V_NUMPAR);
8001 }
8002