xref: /titanic_41/usr/src/uts/sun4v/io/vdc.c (revision 9e86db79b7d1bbc5f2f04e99954cbd5eae0e22bb)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /*
28  * LDoms virtual disk client (vdc) device driver
29  *
30  * This driver runs on a guest logical domain and communicates with the virtual
31  * disk server (vds) driver running on the service domain which is exporting
32  * virtualized "disks" to the guest logical domain.
33  *
34  * The driver can be divided into four sections:
35  *
36  * 1) generic device driver housekeeping
37  *	_init, _fini, attach, detach, ops structures, etc.
38  *
39  * 2) communication channel setup
40  *	Setup the communications link over the LDC channel that vdc uses to
41  *	talk to the vDisk server. Initialise the descriptor ring which
42  *	allows the LDC clients to transfer data via memory mappings.
43  *
44  * 3) Support exported to upper layers (filesystems, etc)
45  *	The upper layers call into vdc via strategy(9E) and DKIO(7I)
46  *	ioctl calls. vdc will copy the data to be written to the descriptor
47  *	ring or maps the buffer to store the data read by the vDisk
48  *	server into the descriptor ring. It then sends a message to the
49  *	vDisk server requesting it to complete the operation.
50  *
51  * 4) Handling responses from vDisk server.
52  *	The vDisk server will ACK some or all of the messages vdc sends to it
53  *	(this is configured during the handshake). Upon receipt of an ACK
54  *	vdc will check the descriptor ring and signal to the upper layer
55  *	code waiting on the IO.
56  */
57 
58 #include <sys/atomic.h>
59 #include <sys/conf.h>
60 #include <sys/disp.h>
61 #include <sys/ddi.h>
62 #include <sys/dkio.h>
63 #include <sys/efi_partition.h>
64 #include <sys/fcntl.h>
65 #include <sys/file.h>
66 #include <sys/kstat.h>
67 #include <sys/mach_descrip.h>
68 #include <sys/modctl.h>
69 #include <sys/mdeg.h>
70 #include <sys/note.h>
71 #include <sys/open.h>
72 #include <sys/sdt.h>
73 #include <sys/stat.h>
74 #include <sys/sunddi.h>
75 #include <sys/types.h>
76 #include <sys/promif.h>
77 #include <sys/var.h>
78 #include <sys/vtoc.h>
79 #include <sys/archsystm.h>
80 #include <sys/sysmacros.h>
81 
82 #include <sys/cdio.h>
83 #include <sys/dktp/fdisk.h>
84 #include <sys/dktp/dadkio.h>
85 #include <sys/mhd.h>
86 #include <sys/scsi/generic/sense.h>
87 #include <sys/scsi/impl/uscsi.h>
88 #include <sys/scsi/impl/services.h>
89 #include <sys/scsi/targets/sddef.h>
90 
91 #include <sys/ldoms.h>
92 #include <sys/ldc.h>
93 #include <sys/vio_common.h>
94 #include <sys/vio_mailbox.h>
95 #include <sys/vio_util.h>
96 #include <sys/vdsk_common.h>
97 #include <sys/vdsk_mailbox.h>
98 #include <sys/vdc.h>
99 
100 #define	VD_OLDVTOC_LIMIT	0x7fffffff
101 
102 /*
103  * function prototypes
104  */
105 
106 /* standard driver functions */
107 static int	vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred);
108 static int	vdc_close(dev_t dev, int flag, int otyp, cred_t *cred);
109 static int	vdc_strategy(struct buf *buf);
110 static int	vdc_print(dev_t dev, char *str);
111 static int	vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk);
112 static int	vdc_read(dev_t dev, struct uio *uio, cred_t *cred);
113 static int	vdc_write(dev_t dev, struct uio *uio, cred_t *cred);
114 static int	vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode,
115 			cred_t *credp, int *rvalp);
116 static int	vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred);
117 static int	vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred);
118 
119 static int	vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,
120 			void *arg, void **resultp);
121 static int	vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd);
122 static int	vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd);
123 static int	vdc_prop_op(dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op,
124 		    int mod_flags, char *name, caddr_t valuep, int *lengthp);
125 
126 /* setup */
127 static void	vdc_min(struct buf *bufp);
128 static int	vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen);
129 static int	vdc_do_ldc_init(vdc_t *vdc, vdc_server_t *srvr);
130 static int	vdc_start_ldc_connection(vdc_t *vdc);
131 static int	vdc_create_device_nodes(vdc_t *vdc);
132 static int	vdc_create_device_nodes_efi(vdc_t *vdc);
133 static int	vdc_create_device_nodes_vtoc(vdc_t *vdc);
134 static void	vdc_create_io_kstats(vdc_t *vdc);
135 static void	vdc_create_err_kstats(vdc_t *vdc);
136 static void	vdc_set_err_kstats(vdc_t *vdc);
137 static int	vdc_get_md_node(dev_info_t *dip, md_t **mdpp,
138 		    mde_cookie_t *vd_nodep);
139 static int	vdc_init_ports(vdc_t *vdc, md_t *mdp, mde_cookie_t vd_nodep);
140 static void	vdc_fini_ports(vdc_t *vdc);
141 static void	vdc_switch_server(vdc_t *vdcp);
142 static int	vdc_do_ldc_up(vdc_t *vdc);
143 static void	vdc_terminate_ldc(vdc_t *vdc, vdc_server_t *srvr);
144 static int	vdc_init_descriptor_ring(vdc_t *vdc);
145 static void	vdc_destroy_descriptor_ring(vdc_t *vdc);
146 static int	vdc_setup_devid(vdc_t *vdc);
147 static void	vdc_store_label_efi(vdc_t *, efi_gpt_t *, efi_gpe_t *);
148 static void	vdc_store_label_vtoc(vdc_t *, struct dk_geom *,
149 		    struct extvtoc *);
150 static void	vdc_store_label_unk(vdc_t *vdc);
151 static boolean_t vdc_is_opened(vdc_t *vdc);
152 static void	vdc_update_size(vdc_t *vdc, size_t, size_t, size_t);
153 static int	vdc_update_vio_bsize(vdc_t *vdc, uint32_t);
154 
155 /* handshake with vds */
156 static int		vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver);
157 static int		vdc_ver_negotiation(vdc_t *vdcp);
158 static int		vdc_init_attr_negotiation(vdc_t *vdc);
159 static int		vdc_attr_negotiation(vdc_t *vdcp);
160 static int		vdc_init_dring_negotiate(vdc_t *vdc);
161 static int		vdc_dring_negotiation(vdc_t *vdcp);
162 static int		vdc_send_rdx(vdc_t *vdcp);
163 static int		vdc_rdx_exchange(vdc_t *vdcp);
164 static boolean_t	vdc_is_supported_version(vio_ver_msg_t *ver_msg);
165 
166 /* processing incoming messages from vDisk server */
167 static void	vdc_process_msg_thread(vdc_t *vdc);
168 static int	vdc_recv(vdc_t *vdc, vio_msg_t *msgp, size_t *nbytesp);
169 
170 static uint_t	vdc_handle_cb(uint64_t event, caddr_t arg);
171 static int	vdc_process_data_msg(vdc_t *vdc, vio_msg_t *msg);
172 static int	vdc_handle_ver_msg(vdc_t *vdc, vio_ver_msg_t *ver_msg);
173 static int	vdc_handle_attr_msg(vdc_t *vdc, vd_attr_msg_t *attr_msg);
174 static int	vdc_handle_dring_reg_msg(vdc_t *vdc, vio_dring_reg_msg_t *msg);
175 static int 	vdc_send_request(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_map_to_shared_dring(vdc_t *vdcp, int idx);
179 static int 	vdc_populate_descriptor(vdc_t *vdcp, int operation,
180 		    caddr_t addr, size_t nbytes, int slice, diskaddr_t offset,
181 		    int cb_type, void *cb_arg, vio_desc_direction_t dir);
182 static int 	vdc_do_sync_op(vdc_t *vdcp, int operation, caddr_t addr,
183 		    size_t nbytes, int slice, diskaddr_t offset, int cb_type,
184 		    void *cb_arg, vio_desc_direction_t dir, boolean_t);
185 
186 static int	vdc_wait_for_response(vdc_t *vdcp, vio_msg_t *msgp);
187 static int	vdc_drain_response(vdc_t *vdcp, vio_cb_type_t cb_type,
188 		    struct buf *buf);
189 static int	vdc_depopulate_descriptor(vdc_t *vdc, uint_t idx);
190 static int	vdc_populate_mem_hdl(vdc_t *vdcp, vdc_local_desc_t *ldep);
191 static int	vdc_verify_seq_num(vdc_t *vdc, vio_dring_msg_t *dring_msg);
192 
193 /* dkio */
194 static int	vd_process_ioctl(dev_t dev, int cmd, caddr_t arg, int mode,
195 		    int *rvalp);
196 static int	vd_process_efi_ioctl(void *vdisk, int cmd, uintptr_t arg);
197 static void	vdc_create_fake_geometry(vdc_t *vdc);
198 static int	vdc_validate_geometry(vdc_t *vdc);
199 static void	vdc_validate(vdc_t *vdc);
200 static void	vdc_validate_task(void *arg);
201 static int	vdc_null_copy_func(vdc_t *vdc, void *from, void *to,
202 		    int mode, int dir);
203 static int	vdc_get_wce_convert(vdc_t *vdc, void *from, void *to,
204 		    int mode, int dir);
205 static int	vdc_set_wce_convert(vdc_t *vdc, void *from, void *to,
206 		    int mode, int dir);
207 static int	vdc_get_vtoc_convert(vdc_t *vdc, void *from, void *to,
208 		    int mode, int dir);
209 static int	vdc_set_vtoc_convert(vdc_t *vdc, void *from, void *to,
210 		    int mode, int dir);
211 static int	vdc_get_extvtoc_convert(vdc_t *vdc, void *from, void *to,
212 		    int mode, int dir);
213 static int	vdc_set_extvtoc_convert(vdc_t *vdc, void *from, void *to,
214 		    int mode, int dir);
215 static int	vdc_get_geom_convert(vdc_t *vdc, void *from, void *to,
216 		    int mode, int dir);
217 static int	vdc_set_geom_convert(vdc_t *vdc, void *from, void *to,
218 		    int mode, int dir);
219 static int	vdc_get_efi_convert(vdc_t *vdc, void *from, void *to,
220 		    int mode, int dir);
221 static int	vdc_set_efi_convert(vdc_t *vdc, void *from, void *to,
222 		    int mode, int dir);
223 
224 static void 	vdc_ownership_update(vdc_t *vdc, int ownership_flags);
225 static int	vdc_access_set(vdc_t *vdc, uint64_t flags, int mode);
226 static vdc_io_t	*vdc_failfast_io_queue(vdc_t *vdc, struct buf *buf);
227 static int	vdc_failfast_check_resv(vdc_t *vdc);
228 
229 /*
230  * Module variables
231  */
232 
233 /*
234  * Tunable variables to control how long vdc waits before timing out on
235  * various operations
236  */
237 static int	vdc_hshake_retries = 3;
238 
239 static int	vdc_timeout = 0; /* units: seconds */
240 static int 	vdc_ldcup_timeout = 1; /* units: seconds */
241 
242 static uint64_t vdc_hz_min_ldc_delay;
243 static uint64_t vdc_min_timeout_ldc = 1 * MILLISEC;
244 static uint64_t vdc_hz_max_ldc_delay;
245 static uint64_t vdc_max_timeout_ldc = 100 * MILLISEC;
246 
247 static uint64_t vdc_ldc_read_init_delay = 1 * MILLISEC;
248 static uint64_t vdc_ldc_read_max_delay = 100 * MILLISEC;
249 
250 /* values for dumping - need to run in a tighter loop */
251 static uint64_t	vdc_usec_timeout_dump = 100 * MILLISEC;	/* 0.1s units: ns */
252 static int	vdc_dump_retries = 100;
253 
254 static uint16_t	vdc_scsi_timeout = 60;	/* 60s units: seconds  */
255 
256 static uint64_t vdc_ownership_delay = 6 * MICROSEC; /* 6s units: usec */
257 
258 /* Count of the number of vdc instances attached */
259 static volatile uint32_t	vdc_instance_count = 0;
260 
261 /* Tunable to log all SCSI errors */
262 static boolean_t vdc_scsi_log_error = B_FALSE;
263 
264 /* Soft state pointer */
265 static void	*vdc_state;
266 
267 /*
268  * Controlling the verbosity of the error/debug messages
269  *
270  * vdc_msglevel - controls level of messages
271  * vdc_matchinst - 64-bit variable where each bit corresponds
272  *                 to the vdc instance the vdc_msglevel applies.
273  */
274 int		vdc_msglevel = 0x0;
275 uint64_t	vdc_matchinst = 0ull;
276 
277 /*
278  * Supported vDisk protocol version pairs.
279  *
280  * The first array entry is the latest and preferred version.
281  */
282 static const vio_ver_t	vdc_version[] = {{1, 1}};
283 
284 static struct cb_ops vdc_cb_ops = {
285 	vdc_open,	/* cb_open */
286 	vdc_close,	/* cb_close */
287 	vdc_strategy,	/* cb_strategy */
288 	vdc_print,	/* cb_print */
289 	vdc_dump,	/* cb_dump */
290 	vdc_read,	/* cb_read */
291 	vdc_write,	/* cb_write */
292 	vdc_ioctl,	/* cb_ioctl */
293 	nodev,		/* cb_devmap */
294 	nodev,		/* cb_mmap */
295 	nodev,		/* cb_segmap */
296 	nochpoll,	/* cb_chpoll */
297 	vdc_prop_op,	/* cb_prop_op */
298 	NULL,		/* cb_str */
299 	D_MP | D_64BIT,	/* cb_flag */
300 	CB_REV,		/* cb_rev */
301 	vdc_aread,	/* cb_aread */
302 	vdc_awrite	/* cb_awrite */
303 };
304 
305 static struct dev_ops vdc_ops = {
306 	DEVO_REV,	/* devo_rev */
307 	0,		/* devo_refcnt */
308 	vdc_getinfo,	/* devo_getinfo */
309 	nulldev,	/* devo_identify */
310 	nulldev,	/* devo_probe */
311 	vdc_attach,	/* devo_attach */
312 	vdc_detach,	/* devo_detach */
313 	nodev,		/* devo_reset */
314 	&vdc_cb_ops,	/* devo_cb_ops */
315 	NULL,		/* devo_bus_ops */
316 	nulldev,	/* devo_power */
317 	ddi_quiesce_not_needed,	/* devo_quiesce */
318 };
319 
320 static struct modldrv modldrv = {
321 	&mod_driverops,
322 	"virtual disk client",
323 	&vdc_ops,
324 };
325 
326 static struct modlinkage modlinkage = {
327 	MODREV_1,
328 	&modldrv,
329 	NULL
330 };
331 
332 /* -------------------------------------------------------------------------- */
333 
334 /*
335  * Device Driver housekeeping and setup
336  */
337 
338 int
339 _init(void)
340 {
341 	int	status;
342 
343 	if ((status = ddi_soft_state_init(&vdc_state, sizeof (vdc_t), 1)) != 0)
344 		return (status);
345 	if ((status = mod_install(&modlinkage)) != 0)
346 		ddi_soft_state_fini(&vdc_state);
347 	return (status);
348 }
349 
350 int
351 _info(struct modinfo *modinfop)
352 {
353 	return (mod_info(&modlinkage, modinfop));
354 }
355 
356 int
357 _fini(void)
358 {
359 	int	status;
360 
361 	if ((status = mod_remove(&modlinkage)) != 0)
362 		return (status);
363 	ddi_soft_state_fini(&vdc_state);
364 	return (0);
365 }
366 
367 static int
368 vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,  void *arg, void **resultp)
369 {
370 	_NOTE(ARGUNUSED(dip))
371 
372 	int	instance = VDCUNIT((dev_t)arg);
373 	vdc_t	*vdc = NULL;
374 
375 	switch (cmd) {
376 	case DDI_INFO_DEVT2DEVINFO:
377 		if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
378 			*resultp = NULL;
379 			return (DDI_FAILURE);
380 		}
381 		*resultp = vdc->dip;
382 		return (DDI_SUCCESS);
383 	case DDI_INFO_DEVT2INSTANCE:
384 		*resultp = (void *)(uintptr_t)instance;
385 		return (DDI_SUCCESS);
386 	default:
387 		*resultp = NULL;
388 		return (DDI_FAILURE);
389 	}
390 }
391 
392 static int
393 vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
394 {
395 	kt_did_t failfast_tid, ownership_tid;
396 	int	instance;
397 	int	rv;
398 	vdc_server_t *srvr;
399 	vdc_t	*vdc = NULL;
400 
401 	switch (cmd) {
402 	case DDI_DETACH:
403 		/* the real work happens below */
404 		break;
405 	case DDI_SUSPEND:
406 		/* nothing to do for this non-device */
407 		return (DDI_SUCCESS);
408 	default:
409 		return (DDI_FAILURE);
410 	}
411 
412 	ASSERT(cmd == DDI_DETACH);
413 	instance = ddi_get_instance(dip);
414 	DMSGX(1, "[%d] Entered\n", instance);
415 
416 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
417 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
418 		return (DDI_FAILURE);
419 	}
420 
421 	/*
422 	 * This function is called when vdc is detached or if it has failed to
423 	 * attach. In that case, the attach may have fail before the vdisk type
424 	 * has been set so we can't call vdc_is_opened(). However as the attach
425 	 * has failed, we know that the vdisk is not opened and we can safely
426 	 * detach.
427 	 */
428 	if (vdc->vdisk_type != VD_DISK_TYPE_UNK && vdc_is_opened(vdc)) {
429 		DMSG(vdc, 0, "[%d] Cannot detach: device is open", instance);
430 		return (DDI_FAILURE);
431 	}
432 
433 	if (vdc->dkio_flush_pending) {
434 		DMSG(vdc, 0,
435 		    "[%d] Cannot detach: %d outstanding DKIO flushes\n",
436 		    instance, vdc->dkio_flush_pending);
437 		return (DDI_FAILURE);
438 	}
439 
440 	if (vdc->validate_pending) {
441 		DMSG(vdc, 0,
442 		    "[%d] Cannot detach: %d outstanding validate request\n",
443 		    instance, vdc->validate_pending);
444 		return (DDI_FAILURE);
445 	}
446 
447 	DMSG(vdc, 0, "[%d] proceeding...\n", instance);
448 
449 	/* If we took ownership, release ownership */
450 	mutex_enter(&vdc->ownership_lock);
451 	if (vdc->ownership & VDC_OWNERSHIP_GRANTED) {
452 		rv = vdc_access_set(vdc, VD_ACCESS_SET_CLEAR, FKIOCTL);
453 		if (rv == 0) {
454 			vdc_ownership_update(vdc, VDC_OWNERSHIP_NONE);
455 		}
456 	}
457 	mutex_exit(&vdc->ownership_lock);
458 
459 	/* mark instance as detaching */
460 	vdc->lifecycle	= VDC_LC_DETACHING;
461 
462 	/*
463 	 * Try and disable callbacks to prevent another handshake. We have to
464 	 * disable callbacks for all servers.
465 	 */
466 	for (srvr = vdc->server_list; srvr != NULL; srvr = srvr->next) {
467 		rv = ldc_set_cb_mode(srvr->ldc_handle, LDC_CB_DISABLE);
468 		DMSG(vdc, 0, "callback disabled (ldc=%lu, rv=%d)\n",
469 		    srvr->ldc_id, rv);
470 	}
471 
472 	if (vdc->initialized & VDC_THREAD) {
473 		mutex_enter(&vdc->read_lock);
474 		if ((vdc->read_state == VDC_READ_WAITING) ||
475 		    (vdc->read_state == VDC_READ_RESET)) {
476 			vdc->read_state = VDC_READ_RESET;
477 			cv_signal(&vdc->read_cv);
478 		}
479 
480 		mutex_exit(&vdc->read_lock);
481 
482 		/* wake up any thread waiting for connection to come online */
483 		mutex_enter(&vdc->lock);
484 		if (vdc->state == VDC_STATE_INIT_WAITING) {
485 			DMSG(vdc, 0,
486 			    "[%d] write reset - move to resetting state...\n",
487 			    instance);
488 			vdc->state = VDC_STATE_RESETTING;
489 			cv_signal(&vdc->initwait_cv);
490 		}
491 		mutex_exit(&vdc->lock);
492 
493 		/* now wait until state transitions to VDC_STATE_DETACH */
494 		thread_join(vdc->msg_proc_thr->t_did);
495 		ASSERT(vdc->state == VDC_STATE_DETACH);
496 		DMSG(vdc, 0, "[%d] Reset thread exit and join ..\n",
497 		    vdc->instance);
498 	}
499 
500 	mutex_enter(&vdc->lock);
501 
502 	if (vdc->initialized & VDC_DRING)
503 		vdc_destroy_descriptor_ring(vdc);
504 
505 	vdc_fini_ports(vdc);
506 
507 	if (vdc->failfast_thread) {
508 		failfast_tid = vdc->failfast_thread->t_did;
509 		vdc->failfast_interval = 0;
510 		cv_signal(&vdc->failfast_cv);
511 	} else {
512 		failfast_tid = 0;
513 	}
514 
515 	if (vdc->ownership & VDC_OWNERSHIP_WANTED) {
516 		ownership_tid = vdc->ownership_thread->t_did;
517 		vdc->ownership = VDC_OWNERSHIP_NONE;
518 		cv_signal(&vdc->ownership_cv);
519 	} else {
520 		ownership_tid = 0;
521 	}
522 
523 	mutex_exit(&vdc->lock);
524 
525 	if (failfast_tid != 0)
526 		thread_join(failfast_tid);
527 
528 	if (ownership_tid != 0)
529 		thread_join(ownership_tid);
530 
531 	if (vdc->initialized & VDC_MINOR)
532 		ddi_remove_minor_node(dip, NULL);
533 
534 	if (vdc->io_stats) {
535 		kstat_delete(vdc->io_stats);
536 		vdc->io_stats = NULL;
537 	}
538 
539 	if (vdc->err_stats) {
540 		kstat_delete(vdc->err_stats);
541 		vdc->err_stats = NULL;
542 	}
543 
544 	if (vdc->initialized & VDC_LOCKS) {
545 		mutex_destroy(&vdc->lock);
546 		mutex_destroy(&vdc->read_lock);
547 		mutex_destroy(&vdc->ownership_lock);
548 		cv_destroy(&vdc->initwait_cv);
549 		cv_destroy(&vdc->dring_free_cv);
550 		cv_destroy(&vdc->membind_cv);
551 		cv_destroy(&vdc->sync_pending_cv);
552 		cv_destroy(&vdc->sync_blocked_cv);
553 		cv_destroy(&vdc->read_cv);
554 		cv_destroy(&vdc->running_cv);
555 		cv_destroy(&vdc->ownership_cv);
556 		cv_destroy(&vdc->failfast_cv);
557 		cv_destroy(&vdc->failfast_io_cv);
558 	}
559 
560 	if (vdc->minfo)
561 		kmem_free(vdc->minfo, sizeof (struct dk_minfo));
562 
563 	if (vdc->cinfo)
564 		kmem_free(vdc->cinfo, sizeof (struct dk_cinfo));
565 
566 	if (vdc->vtoc)
567 		kmem_free(vdc->vtoc, sizeof (struct extvtoc));
568 
569 	if (vdc->geom)
570 		kmem_free(vdc->geom, sizeof (struct dk_geom));
571 
572 	if (vdc->devid) {
573 		ddi_devid_unregister(dip);
574 		ddi_devid_free(vdc->devid);
575 	}
576 
577 	if (vdc->initialized & VDC_SOFT_STATE)
578 		ddi_soft_state_free(vdc_state, instance);
579 
580 	DMSG(vdc, 0, "[%d] End %p\n", instance, (void *)vdc);
581 
582 	return (DDI_SUCCESS);
583 }
584 
585 
586 static int
587 vdc_do_attach(dev_info_t *dip)
588 {
589 	int		instance;
590 	vdc_t		*vdc = NULL;
591 	int		status;
592 	md_t		*mdp;
593 	mde_cookie_t	vd_node;
594 
595 	ASSERT(dip != NULL);
596 
597 	instance = ddi_get_instance(dip);
598 	if (ddi_soft_state_zalloc(vdc_state, instance) != DDI_SUCCESS) {
599 		cmn_err(CE_NOTE, "[%d] Couldn't alloc state structure",
600 		    instance);
601 		return (DDI_FAILURE);
602 	}
603 
604 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
605 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
606 		return (DDI_FAILURE);
607 	}
608 
609 	/*
610 	 * We assign the value to initialized in this case to zero out the
611 	 * variable and then set bits in it to indicate what has been done
612 	 */
613 	vdc->initialized = VDC_SOFT_STATE;
614 
615 	vdc_hz_min_ldc_delay = drv_usectohz(vdc_min_timeout_ldc);
616 	vdc_hz_max_ldc_delay = drv_usectohz(vdc_max_timeout_ldc);
617 
618 	vdc->dip	= dip;
619 	vdc->instance	= instance;
620 	vdc->vdisk_type	= VD_DISK_TYPE_UNK;
621 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
622 	vdc->state	= VDC_STATE_INIT;
623 	vdc->lifecycle	= VDC_LC_ATTACHING;
624 	vdc->session_id = 0;
625 	vdc->vdisk_bsize = DEV_BSIZE;
626 	vdc->vio_bmask = 0;
627 	vdc->vio_bshift = 0;
628 	vdc->max_xfer_sz = maxphys / vdc->vdisk_bsize;
629 
630 	/*
631 	 * We assume, for now, that the vDisk server will export 'read'
632 	 * operations to us at a minimum (this is needed because of checks
633 	 * in vdc for supported operations early in the handshake process).
634 	 * The vDisk server will return ENOTSUP if this is not the case.
635 	 * The value will be overwritten during the attribute exchange with
636 	 * the bitmask of operations exported by server.
637 	 */
638 	vdc->operations = VD_OP_MASK_READ;
639 
640 	vdc->vtoc = NULL;
641 	vdc->geom = NULL;
642 	vdc->cinfo = NULL;
643 	vdc->minfo = NULL;
644 
645 	mutex_init(&vdc->lock, NULL, MUTEX_DRIVER, NULL);
646 	cv_init(&vdc->initwait_cv, NULL, CV_DRIVER, NULL);
647 	cv_init(&vdc->dring_free_cv, NULL, CV_DRIVER, NULL);
648 	cv_init(&vdc->membind_cv, NULL, CV_DRIVER, NULL);
649 	cv_init(&vdc->running_cv, NULL, CV_DRIVER, NULL);
650 
651 	vdc->threads_pending = 0;
652 	vdc->sync_op_pending = B_FALSE;
653 	vdc->sync_op_blocked = B_FALSE;
654 	cv_init(&vdc->sync_pending_cv, NULL, CV_DRIVER, NULL);
655 	cv_init(&vdc->sync_blocked_cv, NULL, CV_DRIVER, NULL);
656 
657 	mutex_init(&vdc->ownership_lock, NULL, MUTEX_DRIVER, NULL);
658 	cv_init(&vdc->ownership_cv, NULL, CV_DRIVER, NULL);
659 	cv_init(&vdc->failfast_cv, NULL, CV_DRIVER, NULL);
660 	cv_init(&vdc->failfast_io_cv, NULL, CV_DRIVER, NULL);
661 
662 	/* init blocking msg read functionality */
663 	mutex_init(&vdc->read_lock, NULL, MUTEX_DRIVER, NULL);
664 	cv_init(&vdc->read_cv, NULL, CV_DRIVER, NULL);
665 	vdc->read_state = VDC_READ_IDLE;
666 
667 	vdc->initialized |= VDC_LOCKS;
668 
669 	/* get device and port MD node for this disk instance */
670 	if (vdc_get_md_node(dip, &mdp, &vd_node) != 0) {
671 		cmn_err(CE_NOTE, "[%d] Could not get machine description node",
672 		    instance);
673 		return (DDI_FAILURE);
674 	}
675 
676 	if (vdc_init_ports(vdc, mdp, vd_node) != 0) {
677 		cmn_err(CE_NOTE, "[%d] Error initialising ports", instance);
678 		return (DDI_FAILURE);
679 	}
680 
681 	(void) md_fini_handle(mdp);
682 
683 	/* Create the kstats for saving the I/O statistics used by iostat(1M) */
684 	vdc_create_io_kstats(vdc);
685 	vdc_create_err_kstats(vdc);
686 
687 	/* Initialize remaining structures before starting the msg thread */
688 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
689 	vdc->vtoc = kmem_zalloc(sizeof (struct extvtoc), KM_SLEEP);
690 	vdc->geom = kmem_zalloc(sizeof (struct dk_geom), KM_SLEEP);
691 	vdc->minfo = kmem_zalloc(sizeof (struct dk_minfo), KM_SLEEP);
692 
693 	/* initialize the thread responsible for managing state with server */
694 	vdc->msg_proc_thr = thread_create(NULL, 0, vdc_process_msg_thread,
695 	    vdc, 0, &p0, TS_RUN, minclsyspri);
696 	if (vdc->msg_proc_thr == NULL) {
697 		cmn_err(CE_NOTE, "[%d] Failed to create msg processing thread",
698 		    instance);
699 		return (DDI_FAILURE);
700 	}
701 
702 	vdc->initialized |= VDC_THREAD;
703 
704 	atomic_inc_32(&vdc_instance_count);
705 
706 	/*
707 	 * Check the disk label. This will send requests and do the handshake.
708 	 * We don't really care about the disk label now. What we really need is
709 	 * the handshake do be done so that we know the type of the disk (slice
710 	 * or full disk) and the appropriate device nodes can be created.
711 	 */
712 
713 	mutex_enter(&vdc->lock);
714 	(void) vdc_validate_geometry(vdc);
715 	mutex_exit(&vdc->lock);
716 
717 	/*
718 	 * Now that we have the device info we can create the device nodes
719 	 */
720 	status = vdc_create_device_nodes(vdc);
721 	if (status) {
722 		DMSG(vdc, 0, "[%d] Failed to create device nodes",
723 		    instance);
724 		goto return_status;
725 	}
726 
727 	/*
728 	 * Setup devid
729 	 */
730 	if (vdc_setup_devid(vdc)) {
731 		DMSG(vdc, 0, "[%d] No device id available\n", instance);
732 	}
733 
734 	/*
735 	 * Fill in the fields of the error statistics kstat that were not
736 	 * available when creating the kstat
737 	 */
738 	vdc_set_err_kstats(vdc);
739 
740 	ddi_report_dev(dip);
741 	vdc->lifecycle	= VDC_LC_ONLINE;
742 	DMSG(vdc, 0, "[%d] Attach tasks successful\n", instance);
743 
744 return_status:
745 	DMSG(vdc, 0, "[%d] Attach completed\n", instance);
746 	return (status);
747 }
748 
749 static int
750 vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
751 {
752 	int	status;
753 
754 	switch (cmd) {
755 	case DDI_ATTACH:
756 		if ((status = vdc_do_attach(dip)) != 0)
757 			(void) vdc_detach(dip, DDI_DETACH);
758 		return (status);
759 	case DDI_RESUME:
760 		/* nothing to do for this non-device */
761 		return (DDI_SUCCESS);
762 	default:
763 		return (DDI_FAILURE);
764 	}
765 }
766 
767 static int
768 vdc_do_ldc_init(vdc_t *vdc, vdc_server_t *srvr)
769 {
770 	int			status = 0;
771 	ldc_status_t		ldc_state;
772 	ldc_attr_t		ldc_attr;
773 
774 	ASSERT(vdc != NULL);
775 	ASSERT(srvr != NULL);
776 
777 	ldc_attr.devclass = LDC_DEV_BLK;
778 	ldc_attr.instance = vdc->instance;
779 	ldc_attr.mode = LDC_MODE_UNRELIABLE;	/* unreliable transport */
780 	ldc_attr.mtu = VD_LDC_MTU;
781 
782 	if ((srvr->state & VDC_LDC_INIT) == 0) {
783 		status = ldc_init(srvr->ldc_id, &ldc_attr,
784 		    &srvr->ldc_handle);
785 		if (status != 0) {
786 			DMSG(vdc, 0, "[%d] ldc_init(chan %ld) returned %d",
787 			    vdc->instance, srvr->ldc_id, status);
788 			return (status);
789 		}
790 		srvr->state |= VDC_LDC_INIT;
791 	}
792 	status = ldc_status(srvr->ldc_handle, &ldc_state);
793 	if (status != 0) {
794 		DMSG(vdc, 0, "[%d] Cannot discover LDC status [err=%d]",
795 		    vdc->instance, status);
796 		goto init_exit;
797 	}
798 	srvr->ldc_state = ldc_state;
799 
800 	if ((srvr->state & VDC_LDC_CB) == 0) {
801 		status = ldc_reg_callback(srvr->ldc_handle, vdc_handle_cb,
802 		    (caddr_t)srvr);
803 		if (status != 0) {
804 			DMSG(vdc, 0, "[%d] LDC callback reg. failed (%d)",
805 			    vdc->instance, status);
806 			goto init_exit;
807 		}
808 		srvr->state |= VDC_LDC_CB;
809 	}
810 
811 	/*
812 	 * At this stage we have initialised LDC, we will now try and open
813 	 * the connection.
814 	 */
815 	if (srvr->ldc_state == LDC_INIT) {
816 		status = ldc_open(srvr->ldc_handle);
817 		if (status != 0) {
818 			DMSG(vdc, 0, "[%d] ldc_open(chan %ld) returned %d",
819 			    vdc->instance, srvr->ldc_id, status);
820 			goto init_exit;
821 		}
822 		srvr->state |= VDC_LDC_OPEN;
823 	}
824 
825 init_exit:
826 	if (status) {
827 		vdc_terminate_ldc(vdc, srvr);
828 	}
829 
830 	return (status);
831 }
832 
833 static int
834 vdc_start_ldc_connection(vdc_t *vdc)
835 {
836 	int		status = 0;
837 
838 	ASSERT(vdc != NULL);
839 
840 	ASSERT(MUTEX_HELD(&vdc->lock));
841 
842 	status = vdc_do_ldc_up(vdc);
843 
844 	DMSG(vdc, 0, "[%d] Finished bringing up LDC\n", vdc->instance);
845 
846 	return (status);
847 }
848 
849 static int
850 vdc_stop_ldc_connection(vdc_t *vdcp)
851 {
852 	int	status;
853 
854 	ASSERT(vdcp != NULL);
855 
856 	ASSERT(MUTEX_HELD(&vdcp->lock));
857 
858 	DMSG(vdcp, 0, ": Resetting connection to vDisk server : state %d\n",
859 	    vdcp->state);
860 
861 	status = ldc_down(vdcp->curr_server->ldc_handle);
862 	DMSG(vdcp, 0, "ldc_down() = %d\n", status);
863 
864 	vdcp->initialized &= ~VDC_HANDSHAKE;
865 	DMSG(vdcp, 0, "initialized=%x\n", vdcp->initialized);
866 
867 	return (status);
868 }
869 
870 static void
871 vdc_create_io_kstats(vdc_t *vdc)
872 {
873 	if (vdc->io_stats != NULL) {
874 		DMSG(vdc, 0, "[%d] I/O kstat already exists\n", vdc->instance);
875 		return;
876 	}
877 
878 	vdc->io_stats = kstat_create(VDC_DRIVER_NAME, vdc->instance, NULL,
879 	    "disk", KSTAT_TYPE_IO, 1, KSTAT_FLAG_PERSISTENT);
880 	if (vdc->io_stats != NULL) {
881 		vdc->io_stats->ks_lock = &vdc->lock;
882 		kstat_install(vdc->io_stats);
883 	} else {
884 		cmn_err(CE_NOTE, "[%d] Failed to create kstat: I/O statistics"
885 		    " will not be gathered", vdc->instance);
886 	}
887 }
888 
889 static void
890 vdc_create_err_kstats(vdc_t *vdc)
891 {
892 	vd_err_stats_t	*stp;
893 	char	kstatmodule_err[KSTAT_STRLEN];
894 	char	kstatname[KSTAT_STRLEN];
895 	int	ndata = (sizeof (vd_err_stats_t) / sizeof (kstat_named_t));
896 	int	instance = vdc->instance;
897 
898 	if (vdc->err_stats != NULL) {
899 		DMSG(vdc, 0, "[%d] ERR kstat already exists\n", vdc->instance);
900 		return;
901 	}
902 
903 	(void) snprintf(kstatmodule_err, sizeof (kstatmodule_err),
904 	    "%serr", VDC_DRIVER_NAME);
905 	(void) snprintf(kstatname, sizeof (kstatname),
906 	    "%s%d,err", VDC_DRIVER_NAME, instance);
907 
908 	vdc->err_stats = kstat_create(kstatmodule_err, instance, kstatname,
909 	    "device_error", KSTAT_TYPE_NAMED, ndata, KSTAT_FLAG_PERSISTENT);
910 
911 	if (vdc->err_stats == NULL) {
912 		cmn_err(CE_NOTE, "[%d] Failed to create kstat: Error statistics"
913 		    " will not be gathered", instance);
914 		return;
915 	}
916 
917 	stp = (vd_err_stats_t *)vdc->err_stats->ks_data;
918 	kstat_named_init(&stp->vd_softerrs,	"Soft Errors",
919 	    KSTAT_DATA_UINT32);
920 	kstat_named_init(&stp->vd_transerrs,	"Transport Errors",
921 	    KSTAT_DATA_UINT32);
922 	kstat_named_init(&stp->vd_protoerrs,	"Protocol Errors",
923 	    KSTAT_DATA_UINT32);
924 	kstat_named_init(&stp->vd_vid,		"Vendor",
925 	    KSTAT_DATA_CHAR);
926 	kstat_named_init(&stp->vd_pid,		"Product",
927 	    KSTAT_DATA_CHAR);
928 	kstat_named_init(&stp->vd_capacity,	"Size",
929 	    KSTAT_DATA_ULONGLONG);
930 
931 	vdc->err_stats->ks_update  = nulldev;
932 
933 	kstat_install(vdc->err_stats);
934 }
935 
936 static void
937 vdc_set_err_kstats(vdc_t *vdc)
938 {
939 	vd_err_stats_t  *stp;
940 
941 	if (vdc->err_stats == NULL)
942 		return;
943 
944 	mutex_enter(&vdc->lock);
945 
946 	stp = (vd_err_stats_t *)vdc->err_stats->ks_data;
947 	ASSERT(stp != NULL);
948 
949 	stp->vd_capacity.value.ui64 = vdc->vdisk_size * vdc->vdisk_bsize;
950 	(void) strcpy(stp->vd_vid.value.c, "SUN");
951 	(void) strcpy(stp->vd_pid.value.c, "VDSK");
952 
953 	mutex_exit(&vdc->lock);
954 }
955 
956 static int
957 vdc_create_device_nodes_efi(vdc_t *vdc)
958 {
959 	ddi_remove_minor_node(vdc->dip, "h");
960 	ddi_remove_minor_node(vdc->dip, "h,raw");
961 
962 	if (ddi_create_minor_node(vdc->dip, "wd", S_IFBLK,
963 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
964 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
965 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'wd'",
966 		    vdc->instance);
967 		return (EIO);
968 	}
969 
970 	/* if any device node is created we set this flag */
971 	vdc->initialized |= VDC_MINOR;
972 
973 	if (ddi_create_minor_node(vdc->dip, "wd,raw", S_IFCHR,
974 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
975 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
976 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'wd,raw'",
977 		    vdc->instance);
978 		return (EIO);
979 	}
980 
981 	return (0);
982 }
983 
984 static int
985 vdc_create_device_nodes_vtoc(vdc_t *vdc)
986 {
987 	ddi_remove_minor_node(vdc->dip, "wd");
988 	ddi_remove_minor_node(vdc->dip, "wd,raw");
989 
990 	if (ddi_create_minor_node(vdc->dip, "h", S_IFBLK,
991 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
992 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
993 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'h'",
994 		    vdc->instance);
995 		return (EIO);
996 	}
997 
998 	/* if any device node is created we set this flag */
999 	vdc->initialized |= VDC_MINOR;
1000 
1001 	if (ddi_create_minor_node(vdc->dip, "h,raw", S_IFCHR,
1002 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
1003 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
1004 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'h,raw'",
1005 		    vdc->instance);
1006 		return (EIO);
1007 	}
1008 
1009 	return (0);
1010 }
1011 
1012 /*
1013  * Function:
1014  *	vdc_create_device_nodes
1015  *
1016  * Description:
1017  *	This function creates the block and character device nodes under
1018  *	/devices. It is called as part of the attach(9E) of the instance
1019  *	during the handshake with vds after vds has sent the attributes
1020  *	to vdc.
1021  *
1022  *	If the device is of type VD_DISK_TYPE_SLICE then the minor node
1023  *	of 2 is used in keeping with the Solaris convention that slice 2
1024  *	refers to a whole disk. Slices start at 'a'
1025  *
1026  * Parameters:
1027  *	vdc 		- soft state pointer
1028  *
1029  * Return Values
1030  *	0		- Success
1031  *	EIO		- Failed to create node
1032  *	EINVAL		- Unknown type of disk exported
1033  */
1034 static int
1035 vdc_create_device_nodes(vdc_t *vdc)
1036 {
1037 	char		name[sizeof ("s,raw")];
1038 	dev_info_t	*dip = NULL;
1039 	int		instance, status;
1040 	int		num_slices = 1;
1041 	int		i;
1042 
1043 	ASSERT(vdc != NULL);
1044 
1045 	instance = vdc->instance;
1046 	dip = vdc->dip;
1047 
1048 	switch (vdc->vdisk_type) {
1049 	case VD_DISK_TYPE_DISK:
1050 		num_slices = V_NUMPAR;
1051 		break;
1052 	case VD_DISK_TYPE_SLICE:
1053 		num_slices = 1;
1054 		break;
1055 	case VD_DISK_TYPE_UNK:
1056 	default:
1057 		return (EINVAL);
1058 	}
1059 
1060 	/*
1061 	 * Minor nodes are different for EFI disks: EFI disks do not have
1062 	 * a minor node 'g' for the minor number corresponding to slice
1063 	 * VD_EFI_WD_SLICE (slice 7) instead they have a minor node 'wd'
1064 	 * representing the whole disk.
1065 	 */
1066 	for (i = 0; i < num_slices; i++) {
1067 
1068 		if (i == VD_EFI_WD_SLICE) {
1069 			if (vdc->vdisk_label == VD_DISK_LABEL_EFI)
1070 				status = vdc_create_device_nodes_efi(vdc);
1071 			else
1072 				status = vdc_create_device_nodes_vtoc(vdc);
1073 			if (status != 0)
1074 				return (status);
1075 			continue;
1076 		}
1077 
1078 		(void) snprintf(name, sizeof (name), "%c", 'a' + i);
1079 		if (ddi_create_minor_node(dip, name, S_IFBLK,
1080 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
1081 			cmn_err(CE_NOTE, "[%d] Couldn't add block node '%s'",
1082 			    instance, name);
1083 			return (EIO);
1084 		}
1085 
1086 		/* if any device node is created we set this flag */
1087 		vdc->initialized |= VDC_MINOR;
1088 
1089 		(void) snprintf(name, sizeof (name), "%c%s", 'a' + i, ",raw");
1090 
1091 		if (ddi_create_minor_node(dip, name, S_IFCHR,
1092 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
1093 			cmn_err(CE_NOTE, "[%d] Couldn't add raw node '%s'",
1094 			    instance, name);
1095 			return (EIO);
1096 		}
1097 	}
1098 
1099 	return (0);
1100 }
1101 
1102 /*
1103  * Driver prop_op(9e) entry point function. Return the number of blocks for
1104  * the partition in question or forward the request to the property facilities.
1105  */
1106 static int
1107 vdc_prop_op(dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op, int mod_flags,
1108     char *name, caddr_t valuep, int *lengthp)
1109 {
1110 	int instance = ddi_get_instance(dip);
1111 	vdc_t *vdc;
1112 	uint64_t nblocks;
1113 	uint_t blksize;
1114 
1115 	vdc = ddi_get_soft_state(vdc_state, instance);
1116 
1117 	if (dev == DDI_DEV_T_ANY || vdc == NULL) {
1118 		return (ddi_prop_op(dev, dip, prop_op, mod_flags,
1119 		    name, valuep, lengthp));
1120 	}
1121 
1122 	mutex_enter(&vdc->lock);
1123 	(void) vdc_validate_geometry(vdc);
1124 	if (vdc->vdisk_label == VD_DISK_LABEL_UNK) {
1125 		mutex_exit(&vdc->lock);
1126 		return (ddi_prop_op(dev, dip, prop_op, mod_flags,
1127 		    name, valuep, lengthp));
1128 	}
1129 	nblocks = vdc->slice[VDCPART(dev)].nblocks;
1130 	blksize = vdc->vdisk_bsize;
1131 	mutex_exit(&vdc->lock);
1132 
1133 	return (ddi_prop_op_nblocks_blksize(dev, dip, prop_op, mod_flags,
1134 	    name, valuep, lengthp, nblocks, blksize));
1135 }
1136 
1137 /*
1138  * Function:
1139  *	vdc_is_opened
1140  *
1141  * Description:
1142  *	This function checks if any slice of a given virtual disk is
1143  *	currently opened.
1144  *
1145  * Parameters:
1146  *	vdc 		- soft state pointer
1147  *
1148  * Return Values
1149  *	B_TRUE		- at least one slice is opened.
1150  *	B_FALSE		- no slice is opened.
1151  */
1152 static boolean_t
1153 vdc_is_opened(vdc_t *vdc)
1154 {
1155 	int i, nslices;
1156 
1157 	switch (vdc->vdisk_type) {
1158 	case VD_DISK_TYPE_DISK:
1159 		nslices = V_NUMPAR;
1160 		break;
1161 	case VD_DISK_TYPE_SLICE:
1162 		nslices = 1;
1163 		break;
1164 	case VD_DISK_TYPE_UNK:
1165 	default:
1166 		ASSERT(0);
1167 	}
1168 
1169 	/* check if there's any layered open */
1170 	for (i = 0; i < nslices; i++) {
1171 		if (vdc->open_lyr[i] > 0)
1172 			return (B_TRUE);
1173 	}
1174 
1175 	/* check if there is any other kind of open */
1176 	for (i = 0; i < OTYPCNT; i++) {
1177 		if (vdc->open[i] != 0)
1178 			return (B_TRUE);
1179 	}
1180 
1181 	return (B_FALSE);
1182 }
1183 
1184 static int
1185 vdc_mark_opened(vdc_t *vdc, int slice, int flag, int otyp)
1186 {
1187 	uint8_t slicemask;
1188 	int i;
1189 
1190 	ASSERT(otyp < OTYPCNT);
1191 	ASSERT(slice < V_NUMPAR);
1192 	ASSERT(MUTEX_HELD(&vdc->lock));
1193 
1194 	slicemask = 1 << slice;
1195 
1196 	/* check if slice is already exclusively opened */
1197 	if (vdc->open_excl & slicemask)
1198 		return (EBUSY);
1199 
1200 	/* if open exclusive, check if slice is already opened */
1201 	if (flag & FEXCL) {
1202 		if (vdc->open_lyr[slice] > 0)
1203 			return (EBUSY);
1204 		for (i = 0; i < OTYPCNT; i++) {
1205 			if (vdc->open[i] & slicemask)
1206 				return (EBUSY);
1207 		}
1208 		vdc->open_excl |= slicemask;
1209 	}
1210 
1211 	/* mark slice as opened */
1212 	if (otyp == OTYP_LYR) {
1213 		vdc->open_lyr[slice]++;
1214 	} else {
1215 		vdc->open[otyp] |= slicemask;
1216 	}
1217 
1218 	return (0);
1219 }
1220 
1221 static void
1222 vdc_mark_closed(vdc_t *vdc, int slice, int flag, int otyp)
1223 {
1224 	uint8_t slicemask;
1225 
1226 	ASSERT(otyp < OTYPCNT);
1227 	ASSERT(slice < V_NUMPAR);
1228 	ASSERT(MUTEX_HELD(&vdc->lock));
1229 
1230 	slicemask = 1 << slice;
1231 
1232 	if (otyp == OTYP_LYR) {
1233 		ASSERT(vdc->open_lyr[slice] > 0);
1234 		vdc->open_lyr[slice]--;
1235 	} else {
1236 		vdc->open[otyp] &= ~slicemask;
1237 	}
1238 
1239 	if (flag & FEXCL)
1240 		vdc->open_excl &= ~slicemask;
1241 }
1242 
1243 static int
1244 vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred)
1245 {
1246 	_NOTE(ARGUNUSED(cred))
1247 
1248 	int	instance, nodelay;
1249 	int	slice, status = 0;
1250 	vdc_t	*vdc;
1251 
1252 	ASSERT(dev != NULL);
1253 	instance = VDCUNIT(*dev);
1254 
1255 	if (otyp >= OTYPCNT)
1256 		return (EINVAL);
1257 
1258 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1259 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1260 		return (ENXIO);
1261 	}
1262 
1263 	DMSG(vdc, 0, "minor = %d flag = %x, otyp = %x\n",
1264 	    getminor(*dev), flag, otyp);
1265 
1266 	slice = VDCPART(*dev);
1267 
1268 	nodelay = flag & (FNDELAY | FNONBLOCK);
1269 
1270 	if ((flag & FWRITE) && (!nodelay) &&
1271 	    !(VD_OP_SUPPORTED(vdc->operations, VD_OP_BWRITE))) {
1272 		return (EROFS);
1273 	}
1274 
1275 	mutex_enter(&vdc->lock);
1276 
1277 	status = vdc_mark_opened(vdc, slice, flag, otyp);
1278 
1279 	if (status != 0) {
1280 		mutex_exit(&vdc->lock);
1281 		return (status);
1282 	}
1283 
1284 	if (nodelay) {
1285 
1286 		/* don't resubmit a validate request if there's already one */
1287 		if (vdc->validate_pending > 0) {
1288 			mutex_exit(&vdc->lock);
1289 			return (0);
1290 		}
1291 
1292 		/* call vdc_validate() asynchronously to avoid blocking */
1293 		if (taskq_dispatch(system_taskq, vdc_validate_task,
1294 		    (void *)vdc, TQ_NOSLEEP) == NULL) {
1295 			vdc_mark_closed(vdc, slice, flag, otyp);
1296 			mutex_exit(&vdc->lock);
1297 			return (ENXIO);
1298 		}
1299 
1300 		vdc->validate_pending++;
1301 		mutex_exit(&vdc->lock);
1302 		return (0);
1303 	}
1304 
1305 	mutex_exit(&vdc->lock);
1306 
1307 	vdc_validate(vdc);
1308 
1309 	mutex_enter(&vdc->lock);
1310 
1311 	if (vdc->vdisk_label == VD_DISK_LABEL_UNK ||
1312 	    vdc->slice[slice].nblocks == 0) {
1313 		vdc_mark_closed(vdc, slice, flag, otyp);
1314 		status = EIO;
1315 	}
1316 
1317 	mutex_exit(&vdc->lock);
1318 
1319 	return (status);
1320 }
1321 
1322 static int
1323 vdc_close(dev_t dev, int flag, int otyp, cred_t *cred)
1324 {
1325 	_NOTE(ARGUNUSED(cred))
1326 
1327 	int	instance;
1328 	int	slice;
1329 	int	rv, rval;
1330 	vdc_t	*vdc;
1331 
1332 	instance = VDCUNIT(dev);
1333 
1334 	if (otyp >= OTYPCNT)
1335 		return (EINVAL);
1336 
1337 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1338 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1339 		return (ENXIO);
1340 	}
1341 
1342 	DMSG(vdc, 0, "[%d] flag = %x, otyp = %x\n", instance, flag, otyp);
1343 
1344 	slice = VDCPART(dev);
1345 
1346 	/*
1347 	 * Attempt to flush the W$ on a close operation. If this is
1348 	 * not a supported IOCTL command or the backing device is read-only
1349 	 * do not fail the close operation.
1350 	 */
1351 	rv = vd_process_ioctl(dev, DKIOCFLUSHWRITECACHE, NULL, FKIOCTL, &rval);
1352 
1353 	if (rv != 0 && rv != ENOTSUP && rv != ENOTTY && rv != EROFS) {
1354 		DMSG(vdc, 0, "[%d] flush failed with error %d on close\n",
1355 		    instance, rv);
1356 		return (EIO);
1357 	}
1358 
1359 	mutex_enter(&vdc->lock);
1360 	vdc_mark_closed(vdc, slice, flag, otyp);
1361 	mutex_exit(&vdc->lock);
1362 
1363 	return (0);
1364 }
1365 
1366 static int
1367 vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
1368 {
1369 	_NOTE(ARGUNUSED(credp))
1370 
1371 	return (vd_process_ioctl(dev, cmd, (caddr_t)arg, mode, rvalp));
1372 }
1373 
1374 static int
1375 vdc_print(dev_t dev, char *str)
1376 {
1377 	cmn_err(CE_NOTE, "vdc%d:  %s", VDCUNIT(dev), str);
1378 	return (0);
1379 }
1380 
1381 static int
1382 vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk)
1383 {
1384 	int	rv;
1385 	size_t	nbytes = nblk * DEV_BSIZE;
1386 	int	instance = VDCUNIT(dev);
1387 	vdc_t	*vdc = NULL;
1388 	diskaddr_t vio_blkno;
1389 
1390 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1391 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1392 		return (ENXIO);
1393 	}
1394 
1395 	DMSG(vdc, 2, "[%d] dump %ld bytes at block 0x%lx : addr=0x%p\n",
1396 	    instance, nbytes, blkno, (void *)addr);
1397 
1398 	/* convert logical block to vio block */
1399 	if ((blkno & vdc->vio_bmask) != 0) {
1400 		DMSG(vdc, 0, "Misaligned block number (%lu)\n", blkno);
1401 		return (EINVAL);
1402 	}
1403 	vio_blkno = blkno >> vdc->vio_bshift;
1404 
1405 	rv = vdc_send_request(vdc, VD_OP_BWRITE, addr, nbytes,
1406 	    VDCPART(dev), vio_blkno, CB_STRATEGY, 0, VIO_write_dir);
1407 	if (rv) {
1408 		DMSG(vdc, 0, "Failed to do a disk dump (err=%d)\n", rv);
1409 		return (rv);
1410 	}
1411 
1412 	if (ddi_in_panic())
1413 		(void) vdc_drain_response(vdc, CB_STRATEGY, NULL);
1414 
1415 	DMSG(vdc, 0, "[%d] End\n", instance);
1416 
1417 	return (0);
1418 }
1419 
1420 /* -------------------------------------------------------------------------- */
1421 
1422 /*
1423  * Disk access routines
1424  *
1425  */
1426 
1427 /*
1428  * vdc_strategy()
1429  *
1430  * Return Value:
1431  *	0:	As per strategy(9E), the strategy() function must return 0
1432  *		[ bioerror(9f) sets b_flags to the proper error code ]
1433  */
1434 static int
1435 vdc_strategy(struct buf *buf)
1436 {
1437 	diskaddr_t vio_blkno;
1438 	int	rv = -1;
1439 	vdc_t	*vdc = NULL;
1440 	int	instance = VDCUNIT(buf->b_edev);
1441 	int	op = (buf->b_flags & B_READ) ? VD_OP_BREAD : VD_OP_BWRITE;
1442 	int	slice;
1443 
1444 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1445 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1446 		bioerror(buf, ENXIO);
1447 		biodone(buf);
1448 		return (0);
1449 	}
1450 
1451 	DMSG(vdc, 2, "[%d] %s %ld bytes at block %llx : b_addr=0x%p\n",
1452 	    instance, (buf->b_flags & B_READ) ? "Read" : "Write",
1453 	    buf->b_bcount, buf->b_lblkno, (void *)buf->b_un.b_addr);
1454 
1455 	bp_mapin(buf);
1456 
1457 	if ((long)buf->b_private == VD_SLICE_NONE) {
1458 		/* I/O using an absolute disk offset */
1459 		slice = VD_SLICE_NONE;
1460 	} else {
1461 		slice = VDCPART(buf->b_edev);
1462 	}
1463 
1464 	/*
1465 	 * In the buf structure, b_lblkno represents a logical block number
1466 	 * using a block size of 512 bytes. For the VIO request, this block
1467 	 * number has to be converted to be represented with the block size
1468 	 * used by the VIO protocol.
1469 	 */
1470 	if ((buf->b_lblkno & vdc->vio_bmask) != 0) {
1471 		bioerror(buf, EINVAL);
1472 		biodone(buf);
1473 		return (0);
1474 	}
1475 	vio_blkno = buf->b_lblkno >> vdc->vio_bshift;
1476 
1477 	rv = vdc_send_request(vdc, op, (caddr_t)buf->b_un.b_addr,
1478 	    buf->b_bcount, slice, vio_blkno,
1479 	    CB_STRATEGY, buf, (op == VD_OP_BREAD) ? VIO_read_dir :
1480 	    VIO_write_dir);
1481 
1482 	/*
1483 	 * If the request was successfully sent, the strategy call returns and
1484 	 * the ACK handler calls the bioxxx functions when the vDisk server is
1485 	 * done otherwise we handle the error here.
1486 	 */
1487 	if (rv) {
1488 		DMSG(vdc, 0, "Failed to read/write (err=%d)\n", rv);
1489 		bioerror(buf, rv);
1490 		biodone(buf);
1491 	} else if (ddi_in_panic()) {
1492 		rv = vdc_drain_response(vdc, CB_STRATEGY, buf);
1493 		if (rv != 0) {
1494 			bioerror(buf, EIO);
1495 			biodone(buf);
1496 		}
1497 	}
1498 
1499 	return (0);
1500 }
1501 
1502 /*
1503  * Function:
1504  *	vdc_min
1505  *
1506  * Description:
1507  *	Routine to limit the size of a data transfer. Used in
1508  *	conjunction with physio(9F).
1509  *
1510  * Arguments:
1511  *	bp - pointer to the indicated buf(9S) struct.
1512  *
1513  */
1514 static void
1515 vdc_min(struct buf *bufp)
1516 {
1517 	vdc_t	*vdc = NULL;
1518 	int	instance = VDCUNIT(bufp->b_edev);
1519 
1520 	vdc = ddi_get_soft_state(vdc_state, instance);
1521 	VERIFY(vdc != NULL);
1522 
1523 	if (bufp->b_bcount > (vdc->max_xfer_sz * vdc->vdisk_bsize)) {
1524 		bufp->b_bcount = vdc->max_xfer_sz * vdc->vdisk_bsize;
1525 	}
1526 }
1527 
1528 static int
1529 vdc_read(dev_t dev, struct uio *uio, cred_t *cred)
1530 {
1531 	_NOTE(ARGUNUSED(cred))
1532 
1533 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1534 	return (physio(vdc_strategy, NULL, dev, B_READ, vdc_min, uio));
1535 }
1536 
1537 static int
1538 vdc_write(dev_t dev, struct uio *uio, cred_t *cred)
1539 {
1540 	_NOTE(ARGUNUSED(cred))
1541 
1542 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1543 	return (physio(vdc_strategy, NULL, dev, B_WRITE, vdc_min, uio));
1544 }
1545 
1546 static int
1547 vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred)
1548 {
1549 	_NOTE(ARGUNUSED(cred))
1550 
1551 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1552 	return (aphysio(vdc_strategy, anocancel, dev, B_READ, vdc_min, aio));
1553 }
1554 
1555 static int
1556 vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred)
1557 {
1558 	_NOTE(ARGUNUSED(cred))
1559 
1560 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1561 	return (aphysio(vdc_strategy, anocancel, dev, B_WRITE, vdc_min, aio));
1562 }
1563 
1564 
1565 /* -------------------------------------------------------------------------- */
1566 
1567 /*
1568  * Handshake support
1569  */
1570 
1571 
1572 /*
1573  * Function:
1574  *	vdc_init_ver_negotiation()
1575  *
1576  * Description:
1577  *
1578  * Arguments:
1579  *	vdc	- soft state pointer for this instance of the device driver.
1580  *
1581  * Return Code:
1582  *	0	- Success
1583  */
1584 static int
1585 vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver)
1586 {
1587 	vio_ver_msg_t	pkt;
1588 	size_t		msglen = sizeof (pkt);
1589 	int		status = -1;
1590 
1591 	ASSERT(vdc != NULL);
1592 	ASSERT(mutex_owned(&vdc->lock));
1593 
1594 	DMSG(vdc, 0, "[%d] Entered.\n", vdc->instance);
1595 
1596 	/*
1597 	 * set the Session ID to a unique value
1598 	 * (the lower 32 bits of the clock tick)
1599 	 */
1600 	vdc->session_id = ((uint32_t)gettick() & 0xffffffff);
1601 	DMSG(vdc, 0, "[%d] Set SID to 0x%lx\n", vdc->instance, vdc->session_id);
1602 
1603 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1604 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1605 	pkt.tag.vio_subtype_env = VIO_VER_INFO;
1606 	pkt.tag.vio_sid = vdc->session_id;
1607 	pkt.dev_class = VDEV_DISK;
1608 	pkt.ver_major = ver.major;
1609 	pkt.ver_minor = ver.minor;
1610 
1611 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1612 	DMSG(vdc, 0, "[%d] Ver info sent (status = %d)\n",
1613 	    vdc->instance, status);
1614 	if ((status != 0) || (msglen != sizeof (vio_ver_msg_t))) {
1615 		DMSG(vdc, 0, "[%d] Failed to send Ver negotiation info: "
1616 		    "id(%lx) rv(%d) size(%ld)", vdc->instance,
1617 		    vdc->curr_server->ldc_handle, status, msglen);
1618 		if (msglen != sizeof (vio_ver_msg_t))
1619 			status = ENOMSG;
1620 	}
1621 
1622 	return (status);
1623 }
1624 
1625 /*
1626  * Function:
1627  *	vdc_ver_negotiation()
1628  *
1629  * Description:
1630  *
1631  * Arguments:
1632  *	vdcp	- soft state pointer for this instance of the device driver.
1633  *
1634  * Return Code:
1635  *	0	- Success
1636  */
1637 static int
1638 vdc_ver_negotiation(vdc_t *vdcp)
1639 {
1640 	vio_msg_t vio_msg;
1641 	int status;
1642 
1643 	if (status = vdc_init_ver_negotiation(vdcp, vdc_version[0]))
1644 		return (status);
1645 
1646 	/* release lock and wait for response */
1647 	mutex_exit(&vdcp->lock);
1648 	status = vdc_wait_for_response(vdcp, &vio_msg);
1649 	mutex_enter(&vdcp->lock);
1650 	if (status) {
1651 		DMSG(vdcp, 0,
1652 		    "[%d] Failed waiting for Ver negotiation response, rv(%d)",
1653 		    vdcp->instance, status);
1654 		return (status);
1655 	}
1656 
1657 	/* check type and sub_type ... */
1658 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1659 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1660 		DMSG(vdcp, 0, "[%d] Invalid ver negotiation response\n",
1661 		    vdcp->instance);
1662 		return (EPROTO);
1663 	}
1664 
1665 	return (vdc_handle_ver_msg(vdcp, (vio_ver_msg_t *)&vio_msg));
1666 }
1667 
1668 /*
1669  * Function:
1670  *	vdc_init_attr_negotiation()
1671  *
1672  * Description:
1673  *
1674  * Arguments:
1675  *	vdc	- soft state pointer for this instance of the device driver.
1676  *
1677  * Return Code:
1678  *	0	- Success
1679  */
1680 static int
1681 vdc_init_attr_negotiation(vdc_t *vdc)
1682 {
1683 	vd_attr_msg_t	pkt;
1684 	size_t		msglen = sizeof (pkt);
1685 	int		status;
1686 
1687 	ASSERT(vdc != NULL);
1688 	ASSERT(mutex_owned(&vdc->lock));
1689 
1690 	DMSG(vdc, 0, "[%d] entered\n", vdc->instance);
1691 
1692 	/* fill in tag */
1693 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1694 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1695 	pkt.tag.vio_subtype_env = VIO_ATTR_INFO;
1696 	pkt.tag.vio_sid = vdc->session_id;
1697 	/* fill in payload */
1698 	pkt.max_xfer_sz = vdc->max_xfer_sz;
1699 	pkt.vdisk_block_size = vdc->vdisk_bsize;
1700 	pkt.xfer_mode = VIO_DRING_MODE_V1_0;
1701 	pkt.operations = 0;	/* server will set bits of valid operations */
1702 	pkt.vdisk_type = 0;	/* server will set to valid device type */
1703 	pkt.vdisk_media = 0;	/* server will set to valid media type */
1704 	pkt.vdisk_size = 0;	/* server will set to valid size */
1705 
1706 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1707 	DMSG(vdc, 0, "Attr info sent (status = %d)\n", status);
1708 
1709 	if ((status != 0) || (msglen != sizeof (vd_attr_msg_t))) {
1710 		DMSG(vdc, 0, "[%d] Failed to send Attr negotiation info: "
1711 		    "id(%lx) rv(%d) size(%ld)", vdc->instance,
1712 		    vdc->curr_server->ldc_handle, status, msglen);
1713 		if (msglen != sizeof (vd_attr_msg_t))
1714 			status = ENOMSG;
1715 	}
1716 
1717 	return (status);
1718 }
1719 
1720 /*
1721  * Function:
1722  *	vdc_attr_negotiation()
1723  *
1724  * Description:
1725  *
1726  * Arguments:
1727  *	vdc	- soft state pointer for this instance of the device driver.
1728  *
1729  * Return Code:
1730  *	0	- Success
1731  */
1732 static int
1733 vdc_attr_negotiation(vdc_t *vdcp)
1734 {
1735 	int status;
1736 	vio_msg_t vio_msg;
1737 
1738 	if (status = vdc_init_attr_negotiation(vdcp))
1739 		return (status);
1740 
1741 	/* release lock and wait for response */
1742 	mutex_exit(&vdcp->lock);
1743 	status = vdc_wait_for_response(vdcp, &vio_msg);
1744 	mutex_enter(&vdcp->lock);
1745 	if (status) {
1746 		DMSG(vdcp, 0,
1747 		    "[%d] Failed waiting for Attr negotiation response, rv(%d)",
1748 		    vdcp->instance, status);
1749 		return (status);
1750 	}
1751 
1752 	/* check type and sub_type ... */
1753 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1754 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1755 		DMSG(vdcp, 0, "[%d] Invalid attr negotiation response\n",
1756 		    vdcp->instance);
1757 		return (EPROTO);
1758 	}
1759 
1760 	return (vdc_handle_attr_msg(vdcp, (vd_attr_msg_t *)&vio_msg));
1761 }
1762 
1763 
1764 /*
1765  * Function:
1766  *	vdc_init_dring_negotiate()
1767  *
1768  * Description:
1769  *
1770  * Arguments:
1771  *	vdc	- soft state pointer for this instance of the device driver.
1772  *
1773  * Return Code:
1774  *	0	- Success
1775  */
1776 static int
1777 vdc_init_dring_negotiate(vdc_t *vdc)
1778 {
1779 	vio_dring_reg_msg_t	pkt;
1780 	size_t			msglen = sizeof (pkt);
1781 	int			status = -1;
1782 	int			retry;
1783 	int			nretries = 10;
1784 
1785 	ASSERT(vdc != NULL);
1786 	ASSERT(mutex_owned(&vdc->lock));
1787 
1788 	for (retry = 0; retry < nretries; retry++) {
1789 		status = vdc_init_descriptor_ring(vdc);
1790 		if (status != EAGAIN)
1791 			break;
1792 		drv_usecwait(vdc_min_timeout_ldc);
1793 	}
1794 
1795 	if (status != 0) {
1796 		DMSG(vdc, 0, "[%d] Failed to init DRing (status = %d)\n",
1797 		    vdc->instance, status);
1798 		return (status);
1799 	}
1800 
1801 	DMSG(vdc, 0, "[%d] Init of descriptor ring completed (status = %d)\n",
1802 	    vdc->instance, status);
1803 
1804 	/* fill in tag */
1805 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1806 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1807 	pkt.tag.vio_subtype_env = VIO_DRING_REG;
1808 	pkt.tag.vio_sid = vdc->session_id;
1809 	/* fill in payload */
1810 	pkt.dring_ident = 0;
1811 	pkt.num_descriptors = vdc->dring_len;
1812 	pkt.descriptor_size = vdc->dring_entry_size;
1813 	pkt.options = (VIO_TX_DRING | VIO_RX_DRING);
1814 	pkt.ncookies = vdc->dring_cookie_count;
1815 	pkt.cookie[0] = vdc->dring_cookie[0];	/* for now just one cookie */
1816 
1817 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1818 	if (status != 0) {
1819 		DMSG(vdc, 0, "[%d] Failed to register DRing (err = %d)",
1820 		    vdc->instance, status);
1821 	}
1822 
1823 	return (status);
1824 }
1825 
1826 
1827 /*
1828  * Function:
1829  *	vdc_dring_negotiation()
1830  *
1831  * Description:
1832  *
1833  * Arguments:
1834  *	vdc	- soft state pointer for this instance of the device driver.
1835  *
1836  * Return Code:
1837  *	0	- Success
1838  */
1839 static int
1840 vdc_dring_negotiation(vdc_t *vdcp)
1841 {
1842 	int status;
1843 	vio_msg_t vio_msg;
1844 
1845 	if (status = vdc_init_dring_negotiate(vdcp))
1846 		return (status);
1847 
1848 	/* release lock and wait for response */
1849 	mutex_exit(&vdcp->lock);
1850 	status = vdc_wait_for_response(vdcp, &vio_msg);
1851 	mutex_enter(&vdcp->lock);
1852 	if (status) {
1853 		DMSG(vdcp, 0,
1854 		    "[%d] Failed waiting for Dring negotiation response,"
1855 		    " rv(%d)", vdcp->instance, status);
1856 		return (status);
1857 	}
1858 
1859 	/* check type and sub_type ... */
1860 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1861 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1862 		DMSG(vdcp, 0, "[%d] Invalid Dring negotiation response\n",
1863 		    vdcp->instance);
1864 		return (EPROTO);
1865 	}
1866 
1867 	return (vdc_handle_dring_reg_msg(vdcp,
1868 	    (vio_dring_reg_msg_t *)&vio_msg));
1869 }
1870 
1871 
1872 /*
1873  * Function:
1874  *	vdc_send_rdx()
1875  *
1876  * Description:
1877  *
1878  * Arguments:
1879  *	vdc	- soft state pointer for this instance of the device driver.
1880  *
1881  * Return Code:
1882  *	0	- Success
1883  */
1884 static int
1885 vdc_send_rdx(vdc_t *vdcp)
1886 {
1887 	vio_msg_t	msg;
1888 	size_t		msglen = sizeof (vio_msg_t);
1889 	int		status;
1890 
1891 	/*
1892 	 * Send an RDX message to vds to indicate we are ready
1893 	 * to send data
1894 	 */
1895 	msg.tag.vio_msgtype = VIO_TYPE_CTRL;
1896 	msg.tag.vio_subtype = VIO_SUBTYPE_INFO;
1897 	msg.tag.vio_subtype_env = VIO_RDX;
1898 	msg.tag.vio_sid = vdcp->session_id;
1899 	status = vdc_send(vdcp, (caddr_t)&msg, &msglen);
1900 	if (status != 0) {
1901 		DMSG(vdcp, 0, "[%d] Failed to send RDX message (%d)",
1902 		    vdcp->instance, status);
1903 	}
1904 
1905 	return (status);
1906 }
1907 
1908 /*
1909  * Function:
1910  *	vdc_handle_rdx()
1911  *
1912  * Description:
1913  *
1914  * Arguments:
1915  *	vdc	- soft state pointer for this instance of the device driver.
1916  *	msgp	- received msg
1917  *
1918  * Return Code:
1919  *	0	- Success
1920  */
1921 static int
1922 vdc_handle_rdx(vdc_t *vdcp, vio_rdx_msg_t *msgp)
1923 {
1924 	_NOTE(ARGUNUSED(vdcp))
1925 	_NOTE(ARGUNUSED(msgp))
1926 
1927 	ASSERT(msgp->tag.vio_msgtype == VIO_TYPE_CTRL);
1928 	ASSERT(msgp->tag.vio_subtype == VIO_SUBTYPE_ACK);
1929 	ASSERT(msgp->tag.vio_subtype_env == VIO_RDX);
1930 
1931 	DMSG(vdcp, 1, "[%d] Got an RDX msg", vdcp->instance);
1932 
1933 	return (0);
1934 }
1935 
1936 /*
1937  * Function:
1938  *	vdc_rdx_exchange()
1939  *
1940  * Description:
1941  *
1942  * Arguments:
1943  *	vdc	- soft state pointer for this instance of the device driver.
1944  *
1945  * Return Code:
1946  *	0	- Success
1947  */
1948 static int
1949 vdc_rdx_exchange(vdc_t *vdcp)
1950 {
1951 	int status;
1952 	vio_msg_t vio_msg;
1953 
1954 	if (status = vdc_send_rdx(vdcp))
1955 		return (status);
1956 
1957 	/* release lock and wait for response */
1958 	mutex_exit(&vdcp->lock);
1959 	status = vdc_wait_for_response(vdcp, &vio_msg);
1960 	mutex_enter(&vdcp->lock);
1961 	if (status) {
1962 		DMSG(vdcp, 0, "[%d] Failed waiting for RDX response, rv(%d)",
1963 		    vdcp->instance, status);
1964 		return (status);
1965 	}
1966 
1967 	/* check type and sub_type ... */
1968 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1969 	    vio_msg.tag.vio_subtype != VIO_SUBTYPE_ACK) {
1970 		DMSG(vdcp, 0, "[%d] Invalid RDX response\n", vdcp->instance);
1971 		return (EPROTO);
1972 	}
1973 
1974 	return (vdc_handle_rdx(vdcp, (vio_rdx_msg_t *)&vio_msg));
1975 }
1976 
1977 
1978 /* -------------------------------------------------------------------------- */
1979 
1980 /*
1981  * LDC helper routines
1982  */
1983 
1984 static int
1985 vdc_recv(vdc_t *vdc, vio_msg_t *msgp, size_t *nbytesp)
1986 {
1987 	int		status;
1988 	uint64_t	delay_time;
1989 	size_t		len;
1990 
1991 	/*
1992 	 * Until we get a blocking ldc read we have to retry until the entire
1993 	 * LDC message has arrived before ldc_read() will return that message.
1994 	 * If ldc_read() succeed but returns a zero length message then that
1995 	 * means that the LDC queue is empty and we have to wait for a
1996 	 * notification from the LDC callback which will set the read_state to
1997 	 * VDC_READ_PENDING. Note we also bail out if the channel is reset or
1998 	 * goes away.
1999 	 */
2000 	delay_time = vdc_ldc_read_init_delay;
2001 
2002 	for (;;) {
2003 
2004 		len = *nbytesp;
2005 		/*
2006 		 * vdc->curr_server is protected by vdc->lock but to avoid
2007 		 * contentions we don't take the lock here. We can do this
2008 		 * safely because vdc_recv() is only called from thread
2009 		 * process_msg_thread() which is also the only thread that
2010 		 * can change vdc->curr_server.
2011 		 */
2012 		status = ldc_read(vdc->curr_server->ldc_handle,
2013 		    (caddr_t)msgp, &len);
2014 
2015 		if (status == EAGAIN) {
2016 			delay_time *= 2;
2017 			if (delay_time >= vdc_ldc_read_max_delay)
2018 				delay_time = vdc_ldc_read_max_delay;
2019 			delay(delay_time);
2020 			continue;
2021 		}
2022 
2023 		if (status != 0) {
2024 			DMSG(vdc, 0, "ldc_read returned %d\n", status);
2025 			break;
2026 		}
2027 
2028 		if (len != 0) {
2029 			*nbytesp = len;
2030 			break;
2031 		}
2032 
2033 		mutex_enter(&vdc->read_lock);
2034 
2035 		while (vdc->read_state != VDC_READ_PENDING) {
2036 
2037 			/* detect if the connection has been reset */
2038 			if (vdc->read_state == VDC_READ_RESET) {
2039 				mutex_exit(&vdc->read_lock);
2040 				return (ECONNRESET);
2041 			}
2042 
2043 			vdc->read_state = VDC_READ_WAITING;
2044 			cv_wait(&vdc->read_cv, &vdc->read_lock);
2045 		}
2046 
2047 		vdc->read_state = VDC_READ_IDLE;
2048 		mutex_exit(&vdc->read_lock);
2049 
2050 		delay_time = vdc_ldc_read_init_delay;
2051 	}
2052 
2053 	return (status);
2054 }
2055 
2056 
2057 
2058 #ifdef DEBUG
2059 void
2060 vdc_decode_tag(vdc_t *vdcp, vio_msg_t *msg)
2061 {
2062 	char *ms, *ss, *ses;
2063 	switch (msg->tag.vio_msgtype) {
2064 #define	Q(_s)	case _s : ms = #_s; break;
2065 	Q(VIO_TYPE_CTRL)
2066 	Q(VIO_TYPE_DATA)
2067 	Q(VIO_TYPE_ERR)
2068 #undef Q
2069 	default: ms = "unknown"; break;
2070 	}
2071 
2072 	switch (msg->tag.vio_subtype) {
2073 #define	Q(_s)	case _s : ss = #_s; break;
2074 	Q(VIO_SUBTYPE_INFO)
2075 	Q(VIO_SUBTYPE_ACK)
2076 	Q(VIO_SUBTYPE_NACK)
2077 #undef Q
2078 	default: ss = "unknown"; break;
2079 	}
2080 
2081 	switch (msg->tag.vio_subtype_env) {
2082 #define	Q(_s)	case _s : ses = #_s; break;
2083 	Q(VIO_VER_INFO)
2084 	Q(VIO_ATTR_INFO)
2085 	Q(VIO_DRING_REG)
2086 	Q(VIO_DRING_UNREG)
2087 	Q(VIO_RDX)
2088 	Q(VIO_PKT_DATA)
2089 	Q(VIO_DESC_DATA)
2090 	Q(VIO_DRING_DATA)
2091 #undef Q
2092 	default: ses = "unknown"; break;
2093 	}
2094 
2095 	DMSG(vdcp, 3, "(%x/%x/%x) message : (%s/%s/%s)\n",
2096 	    msg->tag.vio_msgtype, msg->tag.vio_subtype,
2097 	    msg->tag.vio_subtype_env, ms, ss, ses);
2098 }
2099 #endif
2100 
2101 /*
2102  * Function:
2103  *	vdc_send()
2104  *
2105  * Description:
2106  *	The function encapsulates the call to write a message using LDC.
2107  *	If LDC indicates that the call failed due to the queue being full,
2108  *	we retry the ldc_write(), otherwise we return the error returned by LDC.
2109  *
2110  * Arguments:
2111  *	ldc_handle	- LDC handle for the channel this instance of vdc uses
2112  *	pkt		- address of LDC message to be sent
2113  *	msglen		- the size of the message being sent. When the function
2114  *			  returns, this contains the number of bytes written.
2115  *
2116  * Return Code:
2117  *	0		- Success.
2118  *	EINVAL		- pkt or msglen were NULL
2119  *	ECONNRESET	- The connection was not up.
2120  *	EWOULDBLOCK	- LDC queue is full
2121  *	xxx		- other error codes returned by ldc_write
2122  */
2123 static int
2124 vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen)
2125 {
2126 	size_t	size = 0;
2127 	int	status = 0;
2128 	clock_t delay_ticks;
2129 
2130 	ASSERT(vdc != NULL);
2131 	ASSERT(mutex_owned(&vdc->lock));
2132 	ASSERT(msglen != NULL);
2133 	ASSERT(*msglen != 0);
2134 
2135 #ifdef DEBUG
2136 	vdc_decode_tag(vdc, (vio_msg_t *)(uintptr_t)pkt);
2137 #endif
2138 	/*
2139 	 * Wait indefinitely to send if channel
2140 	 * is busy, but bail out if we succeed or
2141 	 * if the channel closes or is reset.
2142 	 */
2143 	delay_ticks = vdc_hz_min_ldc_delay;
2144 	do {
2145 		size = *msglen;
2146 		status = ldc_write(vdc->curr_server->ldc_handle, pkt, &size);
2147 		if (status == EWOULDBLOCK) {
2148 			delay(delay_ticks);
2149 			/* geometric backoff */
2150 			delay_ticks *= 2;
2151 			if (delay_ticks > vdc_hz_max_ldc_delay)
2152 				delay_ticks = vdc_hz_max_ldc_delay;
2153 		}
2154 	} while (status == EWOULDBLOCK);
2155 
2156 	/* if LDC had serious issues --- reset vdc state */
2157 	if (status == EIO || status == ECONNRESET) {
2158 		/* LDC had serious issues --- reset vdc state */
2159 		mutex_enter(&vdc->read_lock);
2160 		if ((vdc->read_state == VDC_READ_WAITING) ||
2161 		    (vdc->read_state == VDC_READ_RESET))
2162 			cv_signal(&vdc->read_cv);
2163 		vdc->read_state = VDC_READ_RESET;
2164 		mutex_exit(&vdc->read_lock);
2165 
2166 		/* wake up any waiters in the reset thread */
2167 		if (vdc->state == VDC_STATE_INIT_WAITING) {
2168 			DMSG(vdc, 0, "[%d] write reset - "
2169 			    "vdc is resetting ..\n", vdc->instance);
2170 			vdc->state = VDC_STATE_RESETTING;
2171 			cv_signal(&vdc->initwait_cv);
2172 		}
2173 
2174 		return (ECONNRESET);
2175 	}
2176 
2177 	/* return the last size written */
2178 	*msglen = size;
2179 
2180 	return (status);
2181 }
2182 
2183 /*
2184  * Function:
2185  *	vdc_get_md_node
2186  *
2187  * Description:
2188  *	Get the MD, the device node for the given disk instance. The
2189  *	caller is responsible for cleaning up the reference to the
2190  *	returned MD (mdpp) by calling md_fini_handle().
2191  *
2192  * Arguments:
2193  *	dip	- dev info pointer for this instance of the device driver.
2194  *	mdpp	- the returned MD.
2195  *	vd_nodep - the returned device node.
2196  *
2197  * Return Code:
2198  *	0	- Success.
2199  *	ENOENT	- Expected node or property did not exist.
2200  *	ENXIO	- Unexpected error communicating with MD framework
2201  */
2202 static int
2203 vdc_get_md_node(dev_info_t *dip, md_t **mdpp, mde_cookie_t *vd_nodep)
2204 {
2205 	int		status = ENOENT;
2206 	char		*node_name = NULL;
2207 	md_t		*mdp = NULL;
2208 	int		num_nodes;
2209 	int		num_vdevs;
2210 	mde_cookie_t	rootnode;
2211 	mde_cookie_t	*listp = NULL;
2212 	boolean_t	found_inst = B_FALSE;
2213 	int		listsz;
2214 	int		idx;
2215 	uint64_t	md_inst;
2216 	int		obp_inst;
2217 	int		instance = ddi_get_instance(dip);
2218 
2219 	/*
2220 	 * Get the OBP instance number for comparison with the MD instance
2221 	 *
2222 	 * The "cfg-handle" property of a vdc node in an MD contains the MD's
2223 	 * notion of "instance", or unique identifier, for that node; OBP
2224 	 * stores the value of the "cfg-handle" MD property as the value of
2225 	 * the "reg" property on the node in the device tree it builds from
2226 	 * the MD and passes to Solaris.  Thus, we look up the devinfo node's
2227 	 * "reg" property value to uniquely identify this device instance.
2228 	 * If the "reg" property cannot be found, the device tree state is
2229 	 * presumably so broken that there is no point in continuing.
2230 	 */
2231 	if (!ddi_prop_exists(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS, OBP_REG)) {
2232 		cmn_err(CE_WARN, "'%s' property does not exist", OBP_REG);
2233 		return (ENOENT);
2234 	}
2235 	obp_inst = ddi_prop_get_int(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
2236 	    OBP_REG, -1);
2237 	DMSGX(1, "[%d] OBP inst=%d\n", instance, obp_inst);
2238 
2239 	/*
2240 	 * We now walk the MD nodes to find the node for this vdisk.
2241 	 */
2242 	if ((mdp = md_get_handle()) == NULL) {
2243 		cmn_err(CE_WARN, "unable to init machine description");
2244 		return (ENXIO);
2245 	}
2246 
2247 	num_nodes = md_node_count(mdp);
2248 	ASSERT(num_nodes > 0);
2249 
2250 	listsz = num_nodes * sizeof (mde_cookie_t);
2251 
2252 	/* allocate memory for nodes */
2253 	listp = kmem_zalloc(listsz, KM_SLEEP);
2254 
2255 	rootnode = md_root_node(mdp);
2256 	ASSERT(rootnode != MDE_INVAL_ELEM_COOKIE);
2257 
2258 	/*
2259 	 * Search for all the virtual devices, we will then check to see which
2260 	 * ones are disk nodes.
2261 	 */
2262 	num_vdevs = md_scan_dag(mdp, rootnode,
2263 	    md_find_name(mdp, VDC_MD_VDEV_NAME),
2264 	    md_find_name(mdp, "fwd"), listp);
2265 
2266 	if (num_vdevs <= 0) {
2267 		cmn_err(CE_NOTE, "No '%s' node found", VDC_MD_VDEV_NAME);
2268 		status = ENOENT;
2269 		goto done;
2270 	}
2271 
2272 	DMSGX(1, "[%d] num_vdevs=%d\n", instance, num_vdevs);
2273 	for (idx = 0; idx < num_vdevs; idx++) {
2274 		status = md_get_prop_str(mdp, listp[idx], "name", &node_name);
2275 		if ((status != 0) || (node_name == NULL)) {
2276 			cmn_err(CE_NOTE, "Unable to get name of node type '%s'"
2277 			    ": err %d", VDC_MD_VDEV_NAME, status);
2278 			continue;
2279 		}
2280 
2281 		DMSGX(1, "[%d] Found node '%s'\n", instance, node_name);
2282 		if (strcmp(VDC_MD_DISK_NAME, node_name) == 0) {
2283 			status = md_get_prop_val(mdp, listp[idx],
2284 			    VDC_MD_CFG_HDL, &md_inst);
2285 			DMSGX(1, "[%d] vdc inst in MD=%lx\n",
2286 			    instance, md_inst);
2287 			if ((status == 0) && (md_inst == obp_inst)) {
2288 				found_inst = B_TRUE;
2289 				break;
2290 			}
2291 		}
2292 	}
2293 
2294 	if (!found_inst) {
2295 		DMSGX(0, "Unable to find correct '%s' node", VDC_MD_DISK_NAME);
2296 		status = ENOENT;
2297 		goto done;
2298 	}
2299 	DMSGX(0, "[%d] MD inst=%lx\n", instance, md_inst);
2300 
2301 	*vd_nodep = listp[idx];
2302 	*mdpp = mdp;
2303 done:
2304 	kmem_free(listp, listsz);
2305 	return (status);
2306 }
2307 
2308 /*
2309  * Function:
2310  *	vdc_init_ports
2311  *
2312  * Description:
2313  *	Initialize all the ports for this vdisk instance.
2314  *
2315  * Arguments:
2316  *	vdc	- soft state pointer for this instance of the device driver.
2317  *	mdp	- md pointer
2318  *	vd_nodep - device md node.
2319  *
2320  * Return Code:
2321  *	0	- Success.
2322  *	ENOENT	- Expected node or property did not exist.
2323  */
2324 static int
2325 vdc_init_ports(vdc_t *vdc, md_t *mdp, mde_cookie_t vd_nodep)
2326 {
2327 	int		status = 0;
2328 	int		idx;
2329 	int		num_nodes;
2330 	int		num_vports;
2331 	int		num_chans;
2332 	int		listsz;
2333 	mde_cookie_t	vd_port;
2334 	mde_cookie_t	*chanp = NULL;
2335 	mde_cookie_t	*portp = NULL;
2336 	vdc_server_t	*srvr;
2337 	vdc_server_t	*prev_srvr = NULL;
2338 
2339 	/*
2340 	 * We now walk the MD nodes to find the port nodes for this vdisk.
2341 	 */
2342 	num_nodes = md_node_count(mdp);
2343 	ASSERT(num_nodes > 0);
2344 
2345 	listsz = num_nodes * sizeof (mde_cookie_t);
2346 
2347 	/* allocate memory for nodes */
2348 	portp = kmem_zalloc(listsz, KM_SLEEP);
2349 	chanp = kmem_zalloc(listsz, KM_SLEEP);
2350 
2351 	num_vports = md_scan_dag(mdp, vd_nodep,
2352 	    md_find_name(mdp, VDC_MD_PORT_NAME),
2353 	    md_find_name(mdp, "fwd"), portp);
2354 	if (num_vports == 0) {
2355 		DMSGX(0, "Found no '%s' node for '%s' port\n",
2356 		    VDC_MD_PORT_NAME, VDC_MD_VDEV_NAME);
2357 		status = ENOENT;
2358 		goto done;
2359 	}
2360 
2361 	DMSGX(1, "Found %d '%s' node(s) for '%s' port\n",
2362 	    num_vports, VDC_MD_PORT_NAME, VDC_MD_VDEV_NAME);
2363 
2364 	vdc->num_servers = 0;
2365 	for (idx = 0; idx < num_vports; idx++) {
2366 
2367 		/* initialize this port */
2368 		vd_port = portp[idx];
2369 		srvr = kmem_zalloc(sizeof (vdc_server_t), KM_SLEEP);
2370 		srvr->vdcp = vdc;
2371 
2372 		/* get port id */
2373 		if (md_get_prop_val(mdp, vd_port, VDC_MD_ID, &srvr->id) != 0) {
2374 			cmn_err(CE_NOTE, "vDisk port '%s' property not found",
2375 			    VDC_MD_ID);
2376 			kmem_free(srvr, sizeof (vdc_server_t));
2377 			continue;
2378 		}
2379 
2380 		/* set the connection timeout */
2381 		if (md_get_prop_val(mdp, vd_port, VDC_MD_TIMEOUT,
2382 		    &srvr->ctimeout) != 0) {
2383 			srvr->ctimeout = 0;
2384 		}
2385 
2386 		/* get the ldc id */
2387 		num_chans = md_scan_dag(mdp, vd_port,
2388 		    md_find_name(mdp, VDC_MD_CHAN_NAME),
2389 		    md_find_name(mdp, "fwd"), chanp);
2390 
2391 		/* expecting at least one channel */
2392 		if (num_chans <= 0) {
2393 			cmn_err(CE_NOTE, "No '%s' node for '%s' port",
2394 			    VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME);
2395 			kmem_free(srvr, sizeof (vdc_server_t));
2396 			continue;
2397 		} else if (num_chans != 1) {
2398 			DMSGX(0, "Expected 1 '%s' node for '%s' port, "
2399 			    "found %d\n", VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME,
2400 			    num_chans);
2401 		}
2402 
2403 		/*
2404 		 * We use the first channel found (index 0), irrespective of how
2405 		 * many are there in total.
2406 		 */
2407 		if (md_get_prop_val(mdp, chanp[0], VDC_MD_ID,
2408 		    &srvr->ldc_id) != 0) {
2409 			cmn_err(CE_NOTE, "Channel '%s' property not found",
2410 			    VDC_MD_ID);
2411 			kmem_free(srvr, sizeof (vdc_server_t));
2412 			continue;
2413 		}
2414 
2415 		/*
2416 		 * now initialise LDC channel which will be used to
2417 		 * communicate with this server
2418 		 */
2419 		if (vdc_do_ldc_init(vdc, srvr) != 0) {
2420 			kmem_free(srvr, sizeof (vdc_server_t));
2421 			continue;
2422 		}
2423 
2424 		/* add server to list */
2425 		if (prev_srvr)
2426 			prev_srvr->next = srvr;
2427 		else
2428 			vdc->server_list = srvr;
2429 
2430 		prev_srvr = srvr;
2431 
2432 		/* inc numbers of servers */
2433 		vdc->num_servers++;
2434 	}
2435 
2436 	/*
2437 	 * Adjust the max number of handshake retries to match
2438 	 * the number of vdisk servers.
2439 	 */
2440 	if (vdc_hshake_retries < vdc->num_servers)
2441 		vdc_hshake_retries = vdc->num_servers;
2442 
2443 	/* pick first server as current server */
2444 	if (vdc->server_list != NULL) {
2445 		vdc->curr_server = vdc->server_list;
2446 		status = 0;
2447 	} else {
2448 		status = ENOENT;
2449 	}
2450 
2451 done:
2452 	kmem_free(chanp, listsz);
2453 	kmem_free(portp, listsz);
2454 	return (status);
2455 }
2456 
2457 
2458 /*
2459  * Function:
2460  *	vdc_do_ldc_up
2461  *
2462  * Description:
2463  *	Bring the channel for the current server up.
2464  *
2465  * Arguments:
2466  *	vdc	- soft state pointer for this instance of the device driver.
2467  *
2468  * Return Code:
2469  *	0		- Success.
2470  *	EINVAL		- Driver is detaching / LDC error
2471  *	ECONNREFUSED	- Other end is not listening
2472  */
2473 static int
2474 vdc_do_ldc_up(vdc_t *vdc)
2475 {
2476 	int		status;
2477 	ldc_status_t	ldc_state;
2478 
2479 	ASSERT(MUTEX_HELD(&vdc->lock));
2480 
2481 	DMSG(vdc, 0, "[%d] Bringing up channel %lx\n",
2482 	    vdc->instance, vdc->curr_server->ldc_id);
2483 
2484 	if (vdc->lifecycle == VDC_LC_DETACHING)
2485 		return (EINVAL);
2486 
2487 	if ((status = ldc_up(vdc->curr_server->ldc_handle)) != 0) {
2488 		switch (status) {
2489 		case ECONNREFUSED:	/* listener not ready at other end */
2490 			DMSG(vdc, 0, "[%d] ldc_up(%lx,...) return %d\n",
2491 			    vdc->instance, vdc->curr_server->ldc_id, status);
2492 			status = 0;
2493 			break;
2494 		default:
2495 			DMSG(vdc, 0, "[%d] Failed to bring up LDC: "
2496 			    "channel=%ld, err=%d", vdc->instance,
2497 			    vdc->curr_server->ldc_id, status);
2498 			break;
2499 		}
2500 	}
2501 
2502 	if (ldc_status(vdc->curr_server->ldc_handle, &ldc_state) == 0) {
2503 		vdc->curr_server->ldc_state = ldc_state;
2504 		if (ldc_state == LDC_UP) {
2505 			DMSG(vdc, 0, "[%d] LDC channel already up\n",
2506 			    vdc->instance);
2507 			vdc->seq_num = 1;
2508 			vdc->seq_num_reply = 0;
2509 		}
2510 	}
2511 
2512 	return (status);
2513 }
2514 
2515 /*
2516  * Function:
2517  *	vdc_terminate_ldc()
2518  *
2519  * Description:
2520  *
2521  * Arguments:
2522  *	vdc	- soft state pointer for this instance of the device driver.
2523  *	srvr	- vdc per-server info structure
2524  *
2525  * Return Code:
2526  *	None
2527  */
2528 static void
2529 vdc_terminate_ldc(vdc_t *vdc, vdc_server_t *srvr)
2530 {
2531 	int	instance = ddi_get_instance(vdc->dip);
2532 
2533 	if (srvr->state & VDC_LDC_OPEN) {
2534 		DMSG(vdc, 0, "[%d] ldc_close()\n", instance);
2535 		(void) ldc_close(srvr->ldc_handle);
2536 	}
2537 	if (srvr->state & VDC_LDC_CB) {
2538 		DMSG(vdc, 0, "[%d] ldc_unreg_callback()\n", instance);
2539 		(void) ldc_unreg_callback(srvr->ldc_handle);
2540 	}
2541 	if (srvr->state & VDC_LDC_INIT) {
2542 		DMSG(vdc, 0, "[%d] ldc_fini()\n", instance);
2543 		(void) ldc_fini(srvr->ldc_handle);
2544 		srvr->ldc_handle = NULL;
2545 	}
2546 
2547 	srvr->state &= ~(VDC_LDC_INIT | VDC_LDC_CB | VDC_LDC_OPEN);
2548 }
2549 
2550 /*
2551  * Function:
2552  *	vdc_fini_ports()
2553  *
2554  * Description:
2555  *	Finalize all ports by closing the channel associated with each
2556  *	port and also freeing the server structure.
2557  *
2558  * Arguments:
2559  *	vdc	- soft state pointer for this instance of the device driver.
2560  *
2561  * Return Code:
2562  *	None
2563  */
2564 static void
2565 vdc_fini_ports(vdc_t *vdc)
2566 {
2567 	int		instance = ddi_get_instance(vdc->dip);
2568 	vdc_server_t	*srvr, *prev_srvr;
2569 
2570 	ASSERT(vdc != NULL);
2571 	ASSERT(mutex_owned(&vdc->lock));
2572 
2573 	DMSG(vdc, 0, "[%d] initialized=%x\n", instance, vdc->initialized);
2574 
2575 	srvr = vdc->server_list;
2576 
2577 	while (srvr) {
2578 
2579 		vdc_terminate_ldc(vdc, srvr);
2580 
2581 		/* next server */
2582 		prev_srvr = srvr;
2583 		srvr = srvr->next;
2584 
2585 		/* free server */
2586 		kmem_free(prev_srvr, sizeof (vdc_server_t));
2587 	}
2588 
2589 	vdc->server_list = NULL;
2590 }
2591 
2592 /* -------------------------------------------------------------------------- */
2593 
2594 /*
2595  * Descriptor Ring helper routines
2596  */
2597 
2598 /*
2599  * Function:
2600  *	vdc_init_descriptor_ring()
2601  *
2602  * Description:
2603  *
2604  * Arguments:
2605  *	vdc	- soft state pointer for this instance of the device driver.
2606  *
2607  * Return Code:
2608  *	0	- Success
2609  */
2610 static int
2611 vdc_init_descriptor_ring(vdc_t *vdc)
2612 {
2613 	vd_dring_entry_t	*dep = NULL;	/* DRing Entry pointer */
2614 	int	status = 0;
2615 	int	i;
2616 
2617 	DMSG(vdc, 0, "[%d] initialized=%x\n", vdc->instance, vdc->initialized);
2618 
2619 	ASSERT(vdc != NULL);
2620 	ASSERT(mutex_owned(&vdc->lock));
2621 
2622 	/* ensure we have enough room to store max sized block */
2623 	ASSERT(maxphys <= VD_MAX_BLOCK_SIZE);
2624 
2625 	if ((vdc->initialized & VDC_DRING_INIT) == 0) {
2626 		DMSG(vdc, 0, "[%d] ldc_mem_dring_create\n", vdc->instance);
2627 		/*
2628 		 * Calculate the maximum block size we can transmit using one
2629 		 * Descriptor Ring entry from the attributes returned by the
2630 		 * vDisk server. This is subject to a minimum of 'maxphys'
2631 		 * as we do not have the capability to split requests over
2632 		 * multiple DRing entries.
2633 		 */
2634 		if ((vdc->max_xfer_sz * vdc->vdisk_bsize) < maxphys) {
2635 			DMSG(vdc, 0, "[%d] using minimum DRing size\n",
2636 			    vdc->instance);
2637 			vdc->dring_max_cookies = maxphys / PAGESIZE;
2638 		} else {
2639 			vdc->dring_max_cookies =
2640 			    (vdc->max_xfer_sz * vdc->vdisk_bsize) / PAGESIZE;
2641 		}
2642 		vdc->dring_entry_size = (sizeof (vd_dring_entry_t) +
2643 		    (sizeof (ldc_mem_cookie_t) *
2644 		    (vdc->dring_max_cookies - 1)));
2645 		vdc->dring_len = VD_DRING_LEN;
2646 
2647 		status = ldc_mem_dring_create(vdc->dring_len,
2648 		    vdc->dring_entry_size, &vdc->dring_hdl);
2649 		if ((vdc->dring_hdl == NULL) || (status != 0)) {
2650 			DMSG(vdc, 0, "[%d] Descriptor ring creation failed",
2651 			    vdc->instance);
2652 			return (status);
2653 		}
2654 		vdc->initialized |= VDC_DRING_INIT;
2655 	}
2656 
2657 	if ((vdc->initialized & VDC_DRING_BOUND) == 0) {
2658 		DMSG(vdc, 0, "[%d] ldc_mem_dring_bind\n", vdc->instance);
2659 		vdc->dring_cookie =
2660 		    kmem_zalloc(sizeof (ldc_mem_cookie_t), KM_SLEEP);
2661 
2662 		status = ldc_mem_dring_bind(vdc->curr_server->ldc_handle,
2663 		    vdc->dring_hdl,
2664 		    LDC_SHADOW_MAP|LDC_DIRECT_MAP, LDC_MEM_RW,
2665 		    &vdc->dring_cookie[0],
2666 		    &vdc->dring_cookie_count);
2667 		if (status != 0) {
2668 			DMSG(vdc, 0, "[%d] Failed to bind descriptor ring "
2669 			    "(%lx) to channel (%lx) status=%d\n",
2670 			    vdc->instance, vdc->dring_hdl,
2671 			    vdc->curr_server->ldc_handle, status);
2672 			return (status);
2673 		}
2674 		ASSERT(vdc->dring_cookie_count == 1);
2675 		vdc->initialized |= VDC_DRING_BOUND;
2676 	}
2677 
2678 	status = ldc_mem_dring_info(vdc->dring_hdl, &vdc->dring_mem_info);
2679 	if (status != 0) {
2680 		DMSG(vdc, 0,
2681 		    "[%d] Failed to get info for descriptor ring (%lx)\n",
2682 		    vdc->instance, vdc->dring_hdl);
2683 		return (status);
2684 	}
2685 
2686 	if ((vdc->initialized & VDC_DRING_LOCAL) == 0) {
2687 		DMSG(vdc, 0, "[%d] local dring\n", vdc->instance);
2688 
2689 		/* Allocate the local copy of this dring */
2690 		vdc->local_dring =
2691 		    kmem_zalloc(vdc->dring_len * sizeof (vdc_local_desc_t),
2692 		    KM_SLEEP);
2693 		vdc->initialized |= VDC_DRING_LOCAL;
2694 	}
2695 
2696 	/*
2697 	 * Mark all DRing entries as free and initialize the private
2698 	 * descriptor's memory handles. If any entry is initialized,
2699 	 * we need to free it later so we set the bit in 'initialized'
2700 	 * at the start.
2701 	 */
2702 	vdc->initialized |= VDC_DRING_ENTRY;
2703 	for (i = 0; i < vdc->dring_len; i++) {
2704 		dep = VDC_GET_DRING_ENTRY_PTR(vdc, i);
2705 		dep->hdr.dstate = VIO_DESC_FREE;
2706 
2707 		status = ldc_mem_alloc_handle(vdc->curr_server->ldc_handle,
2708 		    &vdc->local_dring[i].desc_mhdl);
2709 		if (status != 0) {
2710 			DMSG(vdc, 0, "![%d] Failed to alloc mem handle for"
2711 			    " descriptor %d", vdc->instance, i);
2712 			return (status);
2713 		}
2714 		vdc->local_dring[i].is_free = B_TRUE;
2715 		vdc->local_dring[i].dep = dep;
2716 	}
2717 
2718 	/* Initialize the starting index */
2719 	vdc->dring_curr_idx = 0;
2720 
2721 	return (status);
2722 }
2723 
2724 /*
2725  * Function:
2726  *	vdc_destroy_descriptor_ring()
2727  *
2728  * Description:
2729  *
2730  * Arguments:
2731  *	vdc	- soft state pointer for this instance of the device driver.
2732  *
2733  * Return Code:
2734  *	None
2735  */
2736 static void
2737 vdc_destroy_descriptor_ring(vdc_t *vdc)
2738 {
2739 	vdc_local_desc_t	*ldep = NULL;	/* Local Dring Entry Pointer */
2740 	ldc_mem_handle_t	mhdl = NULL;
2741 	ldc_mem_info_t		minfo;
2742 	int			status = -1;
2743 	int			i;	/* loop */
2744 
2745 	ASSERT(vdc != NULL);
2746 	ASSERT(mutex_owned(&vdc->lock));
2747 
2748 	DMSG(vdc, 0, "[%d] Entered\n", vdc->instance);
2749 
2750 	if (vdc->initialized & VDC_DRING_ENTRY) {
2751 		DMSG(vdc, 0,
2752 		    "[%d] Removing Local DRing entries\n", vdc->instance);
2753 		for (i = 0; i < vdc->dring_len; i++) {
2754 			ldep = &vdc->local_dring[i];
2755 			mhdl = ldep->desc_mhdl;
2756 
2757 			if (mhdl == NULL)
2758 				continue;
2759 
2760 			if ((status = ldc_mem_info(mhdl, &minfo)) != 0) {
2761 				DMSG(vdc, 0,
2762 				    "ldc_mem_info returned an error: %d\n",
2763 				    status);
2764 
2765 				/*
2766 				 * This must mean that the mem handle
2767 				 * is not valid. Clear it out so that
2768 				 * no one tries to use it.
2769 				 */
2770 				ldep->desc_mhdl = NULL;
2771 				continue;
2772 			}
2773 
2774 			if (minfo.status == LDC_BOUND) {
2775 				(void) ldc_mem_unbind_handle(mhdl);
2776 			}
2777 
2778 			(void) ldc_mem_free_handle(mhdl);
2779 
2780 			ldep->desc_mhdl = NULL;
2781 		}
2782 		vdc->initialized &= ~VDC_DRING_ENTRY;
2783 	}
2784 
2785 	if (vdc->initialized & VDC_DRING_LOCAL) {
2786 		DMSG(vdc, 0, "[%d] Freeing Local DRing\n", vdc->instance);
2787 		kmem_free(vdc->local_dring,
2788 		    vdc->dring_len * sizeof (vdc_local_desc_t));
2789 		vdc->initialized &= ~VDC_DRING_LOCAL;
2790 	}
2791 
2792 	if (vdc->initialized & VDC_DRING_BOUND) {
2793 		DMSG(vdc, 0, "[%d] Unbinding DRing\n", vdc->instance);
2794 		status = ldc_mem_dring_unbind(vdc->dring_hdl);
2795 		if (status == 0) {
2796 			vdc->initialized &= ~VDC_DRING_BOUND;
2797 		} else {
2798 			DMSG(vdc, 0, "[%d] Error %d unbinding DRing %lx",
2799 			    vdc->instance, status, vdc->dring_hdl);
2800 		}
2801 		kmem_free(vdc->dring_cookie, sizeof (ldc_mem_cookie_t));
2802 	}
2803 
2804 	if (vdc->initialized & VDC_DRING_INIT) {
2805 		DMSG(vdc, 0, "[%d] Destroying DRing\n", vdc->instance);
2806 		status = ldc_mem_dring_destroy(vdc->dring_hdl);
2807 		if (status == 0) {
2808 			vdc->dring_hdl = NULL;
2809 			bzero(&vdc->dring_mem_info, sizeof (ldc_mem_info_t));
2810 			vdc->initialized &= ~VDC_DRING_INIT;
2811 		} else {
2812 			DMSG(vdc, 0, "[%d] Error %d destroying DRing (%lx)",
2813 			    vdc->instance, status, vdc->dring_hdl);
2814 		}
2815 	}
2816 }
2817 
2818 /*
2819  * Function:
2820  *	vdc_map_to_shared_dring()
2821  *
2822  * Description:
2823  *	Copy contents of the local descriptor to the shared
2824  *	memory descriptor.
2825  *
2826  * Arguments:
2827  *	vdcp	- soft state pointer for this instance of the device driver.
2828  *	idx	- descriptor ring index
2829  *
2830  * Return Code:
2831  *	None
2832  */
2833 static int
2834 vdc_map_to_shared_dring(vdc_t *vdcp, int idx)
2835 {
2836 	vdc_local_desc_t	*ldep;
2837 	vd_dring_entry_t	*dep;
2838 	int			rv;
2839 
2840 	ldep = &(vdcp->local_dring[idx]);
2841 
2842 	/* for now leave in the old pop_mem_hdl stuff */
2843 	if (ldep->nbytes > 0) {
2844 		rv = vdc_populate_mem_hdl(vdcp, ldep);
2845 		if (rv) {
2846 			DMSG(vdcp, 0, "[%d] Cannot populate mem handle\n",
2847 			    vdcp->instance);
2848 			return (rv);
2849 		}
2850 	}
2851 
2852 	/*
2853 	 * fill in the data details into the DRing
2854 	 */
2855 	dep = ldep->dep;
2856 	ASSERT(dep != NULL);
2857 
2858 	dep->payload.req_id = VDC_GET_NEXT_REQ_ID(vdcp);
2859 	dep->payload.operation = ldep->operation;
2860 	dep->payload.addr = ldep->offset;
2861 	dep->payload.nbytes = ldep->nbytes;
2862 	dep->payload.status = (uint32_t)-1;	/* vds will set valid value */
2863 	dep->payload.slice = ldep->slice;
2864 	dep->hdr.dstate = VIO_DESC_READY;
2865 	dep->hdr.ack = 1;		/* request an ACK for every message */
2866 
2867 	return (0);
2868 }
2869 
2870 /*
2871  * Function:
2872  *	vdc_send_request
2873  *
2874  * Description:
2875  *	This routine writes the data to be transmitted to vds into the
2876  *	descriptor, notifies vds that the ring has been updated and
2877  *	then waits for the request to be processed.
2878  *
2879  * Arguments:
2880  *	vdcp	  - the soft state pointer
2881  *	operation - operation we want vds to perform (VD_OP_XXX)
2882  *	addr	  - address of data buf to be read/written.
2883  *	nbytes	  - number of bytes to read/write
2884  *	slice	  - the disk slice this request is for
2885  *	offset	  - relative disk offset
2886  *	cb_type   - type of call - STRATEGY or SYNC
2887  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
2888  *			. mode for ioctl(9e)
2889  *			. LP64 diskaddr_t (block I/O)
2890  *	dir	  - direction of operation (READ/WRITE/BOTH)
2891  *
2892  * Return Codes:
2893  *	0
2894  *	ENXIO
2895  */
2896 static int
2897 vdc_send_request(vdc_t *vdcp, int operation, caddr_t addr,
2898     size_t nbytes, int slice, diskaddr_t offset, int cb_type,
2899     void *cb_arg, vio_desc_direction_t dir)
2900 {
2901 	int	rv = 0;
2902 
2903 	ASSERT(vdcp != NULL);
2904 	ASSERT(slice == VD_SLICE_NONE || slice < V_NUMPAR);
2905 
2906 	mutex_enter(&vdcp->lock);
2907 
2908 	/*
2909 	 * If this is a block read/write operation we update the I/O statistics
2910 	 * to indicate that the request is being put on the waitq to be
2911 	 * serviced.
2912 	 *
2913 	 * We do it here (a common routine for both synchronous and strategy
2914 	 * calls) for performance reasons - we are already holding vdc->lock
2915 	 * so there is no extra locking overhead. We would have to explicitly
2916 	 * grab the 'lock' mutex to update the stats if we were to do this
2917 	 * higher up the stack in vdc_strategy() et. al.
2918 	 */
2919 	if ((operation == VD_OP_BREAD) || (operation == VD_OP_BWRITE)) {
2920 		DTRACE_IO1(start, buf_t *, cb_arg);
2921 		VD_KSTAT_WAITQ_ENTER(vdcp);
2922 	}
2923 
2924 	do {
2925 		while (vdcp->state != VDC_STATE_RUNNING) {
2926 
2927 			/* return error if detaching */
2928 			if (vdcp->state == VDC_STATE_DETACH) {
2929 				rv = ENXIO;
2930 				goto done;
2931 			}
2932 
2933 			/* fail request if connection timeout is reached */
2934 			if (vdcp->ctimeout_reached) {
2935 				rv = EIO;
2936 				goto done;
2937 			}
2938 
2939 			/*
2940 			 * If we are panicking and the disk is not ready then
2941 			 * we can't send any request because we can't complete
2942 			 * the handshake now.
2943 			 */
2944 			if (ddi_in_panic()) {
2945 				rv = EIO;
2946 				goto done;
2947 			}
2948 
2949 			cv_wait(&vdcp->running_cv, &vdcp->lock);
2950 		}
2951 
2952 	} while (vdc_populate_descriptor(vdcp, operation, addr,
2953 	    nbytes, slice, offset, cb_type, cb_arg, dir));
2954 
2955 done:
2956 	/*
2957 	 * If this is a block read/write we update the I/O statistics kstat
2958 	 * to indicate that this request has been placed on the queue for
2959 	 * processing (i.e sent to the vDisk server) - iostat(1M) will
2960 	 * report the time waiting for the vDisk server under the %b column
2961 	 * In the case of an error we simply take it off the wait queue.
2962 	 */
2963 	if ((operation == VD_OP_BREAD) || (operation == VD_OP_BWRITE)) {
2964 		if (rv == 0) {
2965 			VD_KSTAT_WAITQ_TO_RUNQ(vdcp);
2966 			DTRACE_PROBE1(send, buf_t *, cb_arg);
2967 		} else {
2968 			VD_UPDATE_ERR_STATS(vdcp, vd_transerrs);
2969 			VD_KSTAT_WAITQ_EXIT(vdcp);
2970 			DTRACE_IO1(done, buf_t *, cb_arg);
2971 		}
2972 	}
2973 
2974 	mutex_exit(&vdcp->lock);
2975 
2976 	return (rv);
2977 }
2978 
2979 
2980 /*
2981  * Function:
2982  *	vdc_populate_descriptor
2983  *
2984  * Description:
2985  *	This routine writes the data to be transmitted to vds into the
2986  *	descriptor, notifies vds that the ring has been updated and
2987  *	then waits for the request to be processed.
2988  *
2989  * Arguments:
2990  *	vdcp	  - the soft state pointer
2991  *	operation - operation we want vds to perform (VD_OP_XXX)
2992  *	addr	  - address of data buf to be read/written.
2993  *	nbytes	  - number of bytes to read/write
2994  *	slice	  - the disk slice this request is for
2995  *	offset	  - relative disk offset
2996  *	cb_type   - type of call - STRATEGY or SYNC
2997  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
2998  *			. mode for ioctl(9e)
2999  *			. LP64 diskaddr_t (block I/O)
3000  *	dir	  - direction of operation (READ/WRITE/BOTH)
3001  *
3002  * Return Codes:
3003  *	0
3004  *	EAGAIN
3005  *	ECONNRESET
3006  *	ENXIO
3007  */
3008 static int
3009 vdc_populate_descriptor(vdc_t *vdcp, int operation, caddr_t addr,
3010     size_t nbytes, int slice, diskaddr_t offset, int cb_type,
3011     void *cb_arg, vio_desc_direction_t dir)
3012 {
3013 	vdc_local_desc_t	*local_dep = NULL; /* Local Dring Pointer */
3014 	int			idx;		/* Index of DRing entry used */
3015 	int			next_idx;
3016 	vio_dring_msg_t		dmsg;
3017 	size_t			msglen;
3018 	int			rv;
3019 
3020 	ASSERT(MUTEX_HELD(&vdcp->lock));
3021 	vdcp->threads_pending++;
3022 loop:
3023 	DMSG(vdcp, 2, ": dring_curr_idx = %d\n", vdcp->dring_curr_idx);
3024 
3025 	/* Get next available D-Ring entry */
3026 	idx = vdcp->dring_curr_idx;
3027 	local_dep = &(vdcp->local_dring[idx]);
3028 
3029 	if (!local_dep->is_free) {
3030 		DMSG(vdcp, 2, "[%d]: dring full - waiting for space\n",
3031 		    vdcp->instance);
3032 		cv_wait(&vdcp->dring_free_cv, &vdcp->lock);
3033 		if (vdcp->state == VDC_STATE_RUNNING ||
3034 		    vdcp->state == VDC_STATE_HANDLE_PENDING) {
3035 			goto loop;
3036 		}
3037 		vdcp->threads_pending--;
3038 		return (ECONNRESET);
3039 	}
3040 
3041 	next_idx = idx + 1;
3042 	if (next_idx >= vdcp->dring_len)
3043 		next_idx = 0;
3044 	vdcp->dring_curr_idx = next_idx;
3045 
3046 	ASSERT(local_dep->is_free);
3047 
3048 	local_dep->operation = operation;
3049 	local_dep->addr = addr;
3050 	local_dep->nbytes = nbytes;
3051 	local_dep->slice = slice;
3052 	local_dep->offset = offset;
3053 	local_dep->cb_type = cb_type;
3054 	local_dep->cb_arg = cb_arg;
3055 	local_dep->dir = dir;
3056 
3057 	local_dep->is_free = B_FALSE;
3058 
3059 	rv = vdc_map_to_shared_dring(vdcp, idx);
3060 	if (rv) {
3061 		DMSG(vdcp, 0, "[%d]: cannot bind memory - waiting ..\n",
3062 		    vdcp->instance);
3063 		/* free the descriptor */
3064 		local_dep->is_free = B_TRUE;
3065 		vdcp->dring_curr_idx = idx;
3066 		cv_wait(&vdcp->membind_cv, &vdcp->lock);
3067 		if (vdcp->state == VDC_STATE_RUNNING ||
3068 		    vdcp->state == VDC_STATE_HANDLE_PENDING) {
3069 			goto loop;
3070 		}
3071 		vdcp->threads_pending--;
3072 		return (ECONNRESET);
3073 	}
3074 
3075 	/*
3076 	 * Send a msg with the DRing details to vds
3077 	 */
3078 	VIO_INIT_DRING_DATA_TAG(dmsg);
3079 	VDC_INIT_DRING_DATA_MSG_IDS(dmsg, vdcp);
3080 	dmsg.dring_ident = vdcp->dring_ident;
3081 	dmsg.start_idx = idx;
3082 	dmsg.end_idx = idx;
3083 	vdcp->seq_num++;
3084 
3085 	DTRACE_PROBE2(populate, int, vdcp->instance,
3086 	    vdc_local_desc_t *, local_dep);
3087 	DMSG(vdcp, 2, "ident=0x%lx, st=%u, end=%u, seq=%ld\n",
3088 	    vdcp->dring_ident, dmsg.start_idx, dmsg.end_idx, dmsg.seq_num);
3089 
3090 	/*
3091 	 * note we're still holding the lock here to
3092 	 * make sure the message goes out in order !!!...
3093 	 */
3094 	msglen = sizeof (dmsg);
3095 	rv = vdc_send(vdcp, (caddr_t)&dmsg, &msglen);
3096 	switch (rv) {
3097 	case ECONNRESET:
3098 		/*
3099 		 * vdc_send initiates the reset on failure.
3100 		 * Since the transaction has already been put
3101 		 * on the local dring, it will automatically get
3102 		 * retried when the channel is reset. Given that,
3103 		 * it is ok to just return success even though the
3104 		 * send failed.
3105 		 */
3106 		rv = 0;
3107 		break;
3108 
3109 	case 0: /* EOK */
3110 		DMSG(vdcp, 1, "sent via LDC: rv=%d\n", rv);
3111 		break;
3112 
3113 	default:
3114 		goto cleanup_and_exit;
3115 	}
3116 
3117 	vdcp->threads_pending--;
3118 	return (rv);
3119 
3120 cleanup_and_exit:
3121 	DMSG(vdcp, 0, "unexpected error, rv=%d\n", rv);
3122 	return (ENXIO);
3123 }
3124 
3125 /*
3126  * Function:
3127  *	vdc_do_sync_op
3128  *
3129  * Description:
3130  * 	Wrapper around vdc_populate_descriptor that blocks until the
3131  * 	response to the message is available.
3132  *
3133  * Arguments:
3134  *	vdcp	  - the soft state pointer
3135  *	operation - operation we want vds to perform (VD_OP_XXX)
3136  *	addr	  - address of data buf to be read/written.
3137  *	nbytes	  - number of bytes to read/write
3138  *	slice	  - the disk slice this request is for
3139  *	offset	  - relative disk offset
3140  *	cb_type   - type of call - STRATEGY or SYNC
3141  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
3142  *			. mode for ioctl(9e)
3143  *			. LP64 diskaddr_t (block I/O)
3144  *	dir	  - direction of operation (READ/WRITE/BOTH)
3145  *	rconflict - check for reservation conflict in case of failure
3146  *
3147  * rconflict should be set to B_TRUE by most callers. Callers invoking the
3148  * VD_OP_SCSICMD operation can set rconflict to B_FALSE if they check the
3149  * result of a successful operation with vd_scsi_status().
3150  *
3151  * Return Codes:
3152  *	0
3153  *	EAGAIN
3154  *	EFAULT
3155  *	ENXIO
3156  *	EIO
3157  */
3158 static int
3159 vdc_do_sync_op(vdc_t *vdcp, int operation, caddr_t addr, size_t nbytes,
3160     int slice, diskaddr_t offset, int cb_type, void *cb_arg,
3161     vio_desc_direction_t dir, boolean_t rconflict)
3162 {
3163 	int status;
3164 	vdc_io_t *vio;
3165 	boolean_t check_resv_conflict = B_FALSE;
3166 
3167 	ASSERT(cb_type == CB_SYNC);
3168 
3169 	/*
3170 	 * Grab the lock, if blocked wait until the server
3171 	 * response causes us to wake up again.
3172 	 */
3173 	mutex_enter(&vdcp->lock);
3174 	vdcp->sync_op_cnt++;
3175 	while (vdcp->sync_op_blocked && vdcp->state != VDC_STATE_DETACH) {
3176 		if (ddi_in_panic()) {
3177 			/* don't block if we are panicking */
3178 			vdcp->sync_op_cnt--;
3179 			mutex_exit(&vdcp->lock);
3180 			return (EIO);
3181 		} else {
3182 			cv_wait(&vdcp->sync_blocked_cv, &vdcp->lock);
3183 		}
3184 	}
3185 
3186 	if (vdcp->state == VDC_STATE_DETACH) {
3187 		cv_broadcast(&vdcp->sync_blocked_cv);
3188 		vdcp->sync_op_cnt--;
3189 		mutex_exit(&vdcp->lock);
3190 		return (ENXIO);
3191 	}
3192 
3193 	/* now block anyone other thread entering after us */
3194 	vdcp->sync_op_blocked = B_TRUE;
3195 	vdcp->sync_op_pending = B_TRUE;
3196 	mutex_exit(&vdcp->lock);
3197 
3198 	status = vdc_send_request(vdcp, operation, addr,
3199 	    nbytes, slice, offset, cb_type, cb_arg, dir);
3200 
3201 	mutex_enter(&vdcp->lock);
3202 
3203 	if (status != 0) {
3204 		vdcp->sync_op_pending = B_FALSE;
3205 	} else if (ddi_in_panic()) {
3206 		if (vdc_drain_response(vdcp, CB_SYNC, NULL) == 0) {
3207 			status = vdcp->sync_op_status;
3208 		} else {
3209 			vdcp->sync_op_pending = B_FALSE;
3210 			status = EIO;
3211 		}
3212 	} else {
3213 		/*
3214 		 * block until our transaction completes.
3215 		 * Also anyone else waiting also gets to go next.
3216 		 */
3217 		while (vdcp->sync_op_pending && vdcp->state != VDC_STATE_DETACH)
3218 			cv_wait(&vdcp->sync_pending_cv, &vdcp->lock);
3219 
3220 		DMSG(vdcp, 2, ": operation returned %d\n",
3221 		    vdcp->sync_op_status);
3222 		if (vdcp->state == VDC_STATE_DETACH) {
3223 			vdcp->sync_op_pending = B_FALSE;
3224 			status = ENXIO;
3225 		} else {
3226 			status = vdcp->sync_op_status;
3227 			if (status != 0 && vdcp->failfast_interval != 0) {
3228 				/*
3229 				 * Operation has failed and failfast is enabled.
3230 				 * We need to check if the failure is due to a
3231 				 * reservation conflict if this was requested.
3232 				 */
3233 				check_resv_conflict = rconflict;
3234 			}
3235 
3236 		}
3237 	}
3238 
3239 	vdcp->sync_op_status = 0;
3240 	vdcp->sync_op_blocked = B_FALSE;
3241 	vdcp->sync_op_cnt--;
3242 
3243 	/* signal the next waiting thread */
3244 	cv_signal(&vdcp->sync_blocked_cv);
3245 
3246 	/*
3247 	 * We have to check for reservation conflict after unblocking sync
3248 	 * operations because some sync operations will be used to do this
3249 	 * check.
3250 	 */
3251 	if (check_resv_conflict) {
3252 		vio = vdc_failfast_io_queue(vdcp, NULL);
3253 		while (vio->vio_qtime != 0)
3254 			cv_wait(&vdcp->failfast_io_cv, &vdcp->lock);
3255 		kmem_free(vio, sizeof (vdc_io_t));
3256 	}
3257 
3258 	mutex_exit(&vdcp->lock);
3259 
3260 	return (status);
3261 }
3262 
3263 
3264 /*
3265  * Function:
3266  *	vdc_drain_response()
3267  *
3268  * Description:
3269  * 	When a guest is panicking, the completion of requests needs to be
3270  * 	handled differently because interrupts are disabled and vdc
3271  * 	will not get messages. We have to poll for the messages instead.
3272  *
3273  *	Note: since we are panicking we don't implement	the io:::done
3274  *	DTrace probe or update the I/O statistics kstats.
3275  *
3276  * Arguments:
3277  *	vdc	- soft state pointer for this instance of the device driver.
3278  *	cb_type	- the type of request we want to drain. If type is CB_SYNC
3279  *		  then we drain all responses until we find a CB_SYNC request.
3280  *		  If the type is CB_STRATEGY then the behavior depends on the
3281  *		  value of the buf argument.
3282  *	buf	- if the cb_type argument is CB_SYNC then the buf argument
3283  *		  must be NULL. If the cb_type argument is CB_STRATEGY and
3284  *		  if buf is NULL then we drain all responses, otherwise we
3285  *		  poll until we receive a ACK/NACK for the specific I/O
3286  *		  described by buf.
3287  *
3288  * Return Code:
3289  *	0	- Success. If we were expecting a response to a particular
3290  *		  CB_SYNC or CB_STRATEGY request then this means that a
3291  *		  response has been received.
3292  */
3293 static int
3294 vdc_drain_response(vdc_t *vdc, vio_cb_type_t cb_type, struct buf *buf)
3295 {
3296 	int 			rv, idx, retries;
3297 	size_t			msglen;
3298 	vdc_local_desc_t 	*ldep = NULL;	/* Local Dring Entry Pointer */
3299 	vio_dring_msg_t		dmsg;
3300 	struct buf		*mbuf;
3301 	boolean_t		ack;
3302 
3303 	ASSERT(cb_type == CB_STRATEGY || cb_type == CB_SYNC);
3304 
3305 	mutex_enter(&vdc->lock);
3306 
3307 	retries = 0;
3308 	for (;;) {
3309 		msglen = sizeof (dmsg);
3310 		rv = ldc_read(vdc->curr_server->ldc_handle, (caddr_t)&dmsg,
3311 		    &msglen);
3312 		if (rv) {
3313 			rv = EINVAL;
3314 			break;
3315 		}
3316 
3317 		/*
3318 		 * if there are no packets wait and check again
3319 		 */
3320 		if ((rv == 0) && (msglen == 0)) {
3321 			if (retries++ > vdc_dump_retries) {
3322 				rv = EAGAIN;
3323 				break;
3324 			}
3325 
3326 			drv_usecwait(vdc_usec_timeout_dump);
3327 			continue;
3328 		}
3329 
3330 		/*
3331 		 * Ignore all messages that are not ACKs/NACKs to
3332 		 * DRing requests.
3333 		 */
3334 		if ((dmsg.tag.vio_msgtype != VIO_TYPE_DATA) ||
3335 		    (dmsg.tag.vio_subtype_env != VIO_DRING_DATA)) {
3336 			DMSG(vdc, 0, "discard pkt: type=%d sub=%d env=%d\n",
3337 			    dmsg.tag.vio_msgtype,
3338 			    dmsg.tag.vio_subtype,
3339 			    dmsg.tag.vio_subtype_env);
3340 			continue;
3341 		}
3342 
3343 		/*
3344 		 * Record if the packet was ACK'ed or not. If the packet was not
3345 		 * ACK'ed then we will just mark the request as failed; we don't
3346 		 * want to reset the connection at this point.
3347 		 */
3348 		switch (dmsg.tag.vio_subtype) {
3349 		case VIO_SUBTYPE_ACK:
3350 			ack = B_TRUE;
3351 			break;
3352 		case VIO_SUBTYPE_NACK:
3353 			ack = B_FALSE;
3354 			break;
3355 		default:
3356 			continue;
3357 		}
3358 
3359 		idx = dmsg.start_idx;
3360 		if (idx >= vdc->dring_len) {
3361 			DMSG(vdc, 0, "[%d] Bogus ack data : start %d\n",
3362 			    vdc->instance, idx);
3363 			continue;
3364 		}
3365 		ldep = &vdc->local_dring[idx];
3366 		if (ldep->dep->hdr.dstate != VIO_DESC_DONE) {
3367 			DMSG(vdc, 0, "[%d] Entry @ %d - state !DONE %d\n",
3368 			    vdc->instance, idx, ldep->dep->hdr.dstate);
3369 			continue;
3370 		}
3371 
3372 		switch (ldep->cb_type) {
3373 
3374 		case CB_STRATEGY:
3375 			mbuf = ldep->cb_arg;
3376 			if (mbuf != NULL) {
3377 				mbuf->b_resid = mbuf->b_bcount -
3378 				    ldep->dep->payload.nbytes;
3379 				bioerror(mbuf,
3380 				    ack ? ldep->dep->payload.status : EIO);
3381 				biodone(mbuf);
3382 			}
3383 			rv = vdc_depopulate_descriptor(vdc, idx);
3384 			if (buf != NULL && buf == mbuf) {
3385 				rv = 0;
3386 				goto done;
3387 			}
3388 			break;
3389 
3390 		case CB_SYNC:
3391 			rv = vdc_depopulate_descriptor(vdc, idx);
3392 			vdc->sync_op_status = ack ? rv : EIO;
3393 			vdc->sync_op_pending = B_FALSE;
3394 			cv_signal(&vdc->sync_pending_cv);
3395 			if (cb_type == CB_SYNC) {
3396 				rv = 0;
3397 				goto done;
3398 			}
3399 			break;
3400 		}
3401 
3402 		/* if this is the last descriptor - break out of loop */
3403 		if ((idx + 1) % vdc->dring_len == vdc->dring_curr_idx) {
3404 			/*
3405 			 * If we were expecting a response for a particular
3406 			 * request then we return with an error otherwise we
3407 			 * have successfully completed the drain.
3408 			 */
3409 			rv = (buf != NULL || cb_type == CB_SYNC)? ESRCH: 0;
3410 			break;
3411 		}
3412 	}
3413 
3414 done:
3415 	mutex_exit(&vdc->lock);
3416 	DMSG(vdc, 0, "End idx=%d\n", idx);
3417 
3418 	return (rv);
3419 }
3420 
3421 
3422 /*
3423  * Function:
3424  *	vdc_depopulate_descriptor()
3425  *
3426  * Description:
3427  *
3428  * Arguments:
3429  *	vdc	- soft state pointer for this instance of the device driver.
3430  *	idx	- Index of the Descriptor Ring entry being modified
3431  *
3432  * Return Code:
3433  *	0	- Success
3434  */
3435 static int
3436 vdc_depopulate_descriptor(vdc_t *vdc, uint_t idx)
3437 {
3438 	vd_dring_entry_t *dep = NULL;		/* Dring Entry Pointer */
3439 	vdc_local_desc_t *ldep = NULL;		/* Local Dring Entry Pointer */
3440 	int		status = ENXIO;
3441 	int		rv = 0;
3442 
3443 	ASSERT(vdc != NULL);
3444 	ASSERT(idx < vdc->dring_len);
3445 	ldep = &vdc->local_dring[idx];
3446 	ASSERT(ldep != NULL);
3447 	ASSERT(MUTEX_HELD(&vdc->lock));
3448 
3449 	DTRACE_PROBE2(depopulate, int, vdc->instance, vdc_local_desc_t *, ldep);
3450 	DMSG(vdc, 2, ": idx = %d\n", idx);
3451 
3452 	dep = ldep->dep;
3453 	ASSERT(dep != NULL);
3454 	ASSERT((dep->hdr.dstate == VIO_DESC_DONE) ||
3455 	    (dep->payload.status == ECANCELED));
3456 
3457 	VDC_MARK_DRING_ENTRY_FREE(vdc, idx);
3458 
3459 	ldep->is_free = B_TRUE;
3460 	status = dep->payload.status;
3461 	DMSG(vdc, 2, ": is_free = %d : status = %d\n", ldep->is_free, status);
3462 
3463 	/*
3464 	 * If no buffers were used to transfer information to the server when
3465 	 * populating the descriptor then no memory handles need to be unbound
3466 	 * and we can return now.
3467 	 */
3468 	if (ldep->nbytes == 0) {
3469 		cv_signal(&vdc->dring_free_cv);
3470 		return (status);
3471 	}
3472 
3473 	/*
3474 	 * If the upper layer passed in a misaligned address we copied the
3475 	 * data into an aligned buffer before sending it to LDC - we now
3476 	 * copy it back to the original buffer.
3477 	 */
3478 	if (ldep->align_addr) {
3479 		ASSERT(ldep->addr != NULL);
3480 
3481 		if (dep->payload.nbytes > 0)
3482 			bcopy(ldep->align_addr, ldep->addr,
3483 			    dep->payload.nbytes);
3484 		kmem_free(ldep->align_addr,
3485 		    sizeof (caddr_t) * P2ROUNDUP(ldep->nbytes, 8));
3486 		ldep->align_addr = NULL;
3487 	}
3488 
3489 	rv = ldc_mem_unbind_handle(ldep->desc_mhdl);
3490 	if (rv != 0) {
3491 		DMSG(vdc, 0, "?[%d] unbind mhdl 0x%lx @ idx %d failed (%d)",
3492 		    vdc->instance, ldep->desc_mhdl, idx, rv);
3493 		/*
3494 		 * The error returned by the vDisk server is more informative
3495 		 * and thus has a higher priority but if it isn't set we ensure
3496 		 * that this function returns an error.
3497 		 */
3498 		if (status == 0)
3499 			status = EINVAL;
3500 	}
3501 
3502 	cv_signal(&vdc->membind_cv);
3503 	cv_signal(&vdc->dring_free_cv);
3504 
3505 	return (status);
3506 }
3507 
3508 /*
3509  * Function:
3510  *	vdc_populate_mem_hdl()
3511  *
3512  * Description:
3513  *
3514  * Arguments:
3515  *	vdc	- soft state pointer for this instance of the device driver.
3516  *	idx	- Index of the Descriptor Ring entry being modified
3517  *	addr	- virtual address being mapped in
3518  *	nybtes	- number of bytes in 'addr'
3519  *	operation - the vDisk operation being performed (VD_OP_xxx)
3520  *
3521  * Return Code:
3522  *	0	- Success
3523  */
3524 static int
3525 vdc_populate_mem_hdl(vdc_t *vdcp, vdc_local_desc_t *ldep)
3526 {
3527 	vd_dring_entry_t	*dep = NULL;
3528 	ldc_mem_handle_t	mhdl;
3529 	caddr_t			vaddr;
3530 	size_t			nbytes;
3531 	uint8_t			perm = LDC_MEM_RW;
3532 	uint8_t			maptype;
3533 	int			rv = 0;
3534 	int			i;
3535 
3536 	ASSERT(vdcp != NULL);
3537 
3538 	dep = ldep->dep;
3539 	mhdl = ldep->desc_mhdl;
3540 
3541 	switch (ldep->dir) {
3542 	case VIO_read_dir:
3543 		perm = LDC_MEM_W;
3544 		break;
3545 
3546 	case VIO_write_dir:
3547 		perm = LDC_MEM_R;
3548 		break;
3549 
3550 	case VIO_both_dir:
3551 		perm = LDC_MEM_RW;
3552 		break;
3553 
3554 	default:
3555 		ASSERT(0);	/* catch bad programming in vdc */
3556 	}
3557 
3558 	/*
3559 	 * LDC expects any addresses passed in to be 8-byte aligned. We need
3560 	 * to copy the contents of any misaligned buffers to a newly allocated
3561 	 * buffer and bind it instead (and copy the the contents back to the
3562 	 * original buffer passed in when depopulating the descriptor)
3563 	 */
3564 	vaddr = ldep->addr;
3565 	nbytes = ldep->nbytes;
3566 	if (((uint64_t)vaddr & 0x7) != 0) {
3567 		ASSERT(ldep->align_addr == NULL);
3568 		ldep->align_addr =
3569 		    kmem_alloc(sizeof (caddr_t) *
3570 		    P2ROUNDUP(nbytes, 8), KM_SLEEP);
3571 		DMSG(vdcp, 0, "[%d] Misaligned address %p reallocating "
3572 		    "(buf=%p nb=%ld op=%d)\n",
3573 		    vdcp->instance, (void *)vaddr, (void *)ldep->align_addr,
3574 		    nbytes, ldep->operation);
3575 		if (perm != LDC_MEM_W)
3576 			bcopy(vaddr, ldep->align_addr, nbytes);
3577 		vaddr = ldep->align_addr;
3578 	}
3579 
3580 	maptype = LDC_IO_MAP|LDC_SHADOW_MAP|LDC_DIRECT_MAP;
3581 	rv = ldc_mem_bind_handle(mhdl, vaddr, P2ROUNDUP(nbytes, 8),
3582 	    maptype, perm, &dep->payload.cookie[0], &dep->payload.ncookies);
3583 	DMSG(vdcp, 2, "[%d] bound mem handle; ncookies=%d\n",
3584 	    vdcp->instance, dep->payload.ncookies);
3585 	if (rv != 0) {
3586 		DMSG(vdcp, 0, "[%d] Failed to bind LDC memory handle "
3587 		    "(mhdl=%p, buf=%p, err=%d)\n",
3588 		    vdcp->instance, (void *)mhdl, (void *)vaddr, rv);
3589 		if (ldep->align_addr) {
3590 			kmem_free(ldep->align_addr,
3591 			    sizeof (caddr_t) * P2ROUNDUP(nbytes, 8));
3592 			ldep->align_addr = NULL;
3593 		}
3594 		return (EAGAIN);
3595 	}
3596 
3597 	/*
3598 	 * Get the other cookies (if any).
3599 	 */
3600 	for (i = 1; i < dep->payload.ncookies; i++) {
3601 		rv = ldc_mem_nextcookie(mhdl, &dep->payload.cookie[i]);
3602 		if (rv != 0) {
3603 			(void) ldc_mem_unbind_handle(mhdl);
3604 			DMSG(vdcp, 0, "?[%d] Failed to get next cookie "
3605 			    "(mhdl=%lx cnum=%d), err=%d",
3606 			    vdcp->instance, mhdl, i, rv);
3607 			if (ldep->align_addr) {
3608 				kmem_free(ldep->align_addr,
3609 				    sizeof (caddr_t) * ldep->nbytes);
3610 				ldep->align_addr = NULL;
3611 			}
3612 			return (EAGAIN);
3613 		}
3614 	}
3615 
3616 	return (rv);
3617 }
3618 
3619 /*
3620  * Interrupt handlers for messages from LDC
3621  */
3622 
3623 /*
3624  * Function:
3625  *	vdc_handle_cb()
3626  *
3627  * Description:
3628  *
3629  * Arguments:
3630  *	event	- Type of event (LDC_EVT_xxx) that triggered the callback
3631  *	arg	- soft state pointer for this instance of the device driver.
3632  *
3633  * Return Code:
3634  *	0	- Success
3635  */
3636 static uint_t
3637 vdc_handle_cb(uint64_t event, caddr_t arg)
3638 {
3639 	ldc_status_t	ldc_state;
3640 	int		rv = 0;
3641 	vdc_server_t	*srvr = (vdc_server_t *)(void *)arg;
3642 	vdc_t		*vdc = srvr->vdcp;
3643 
3644 	ASSERT(vdc != NULL);
3645 
3646 	DMSG(vdc, 1, "evt=%lx seqID=%ld\n", event, vdc->seq_num);
3647 
3648 	/* If callback is not for the current server, ignore it */
3649 	mutex_enter(&vdc->lock);
3650 
3651 	if (vdc->curr_server != srvr) {
3652 		DMSG(vdc, 0, "[%d] Ignoring event 0x%lx for port@%ld\n",
3653 		    vdc->instance, event, srvr->id);
3654 		mutex_exit(&vdc->lock);
3655 		return (LDC_SUCCESS);
3656 	}
3657 
3658 	/*
3659 	 * Depending on the type of event that triggered this callback,
3660 	 * we modify the handshake state or read the data.
3661 	 *
3662 	 * NOTE: not done as a switch() as event could be triggered by
3663 	 * a state change and a read request. Also the ordering	of the
3664 	 * check for the event types is deliberate.
3665 	 */
3666 	if (event & LDC_EVT_UP) {
3667 		DMSG(vdc, 0, "[%d] Received LDC_EVT_UP\n", vdc->instance);
3668 
3669 		/* get LDC state */
3670 		rv = ldc_status(srvr->ldc_handle, &ldc_state);
3671 		if (rv != 0) {
3672 			DMSG(vdc, 0, "[%d] Couldn't get LDC status %d",
3673 			    vdc->instance, rv);
3674 			mutex_exit(&vdc->lock);
3675 			return (LDC_SUCCESS);
3676 		}
3677 		if (srvr->ldc_state != LDC_UP &&
3678 		    ldc_state == LDC_UP) {
3679 			/*
3680 			 * Reset the transaction sequence numbers when
3681 			 * LDC comes up. We then kick off the handshake
3682 			 * negotiation with the vDisk server.
3683 			 */
3684 			vdc->seq_num = 1;
3685 			vdc->seq_num_reply = 0;
3686 			srvr->ldc_state = ldc_state;
3687 			cv_signal(&vdc->initwait_cv);
3688 		}
3689 	}
3690 
3691 	if (event & LDC_EVT_READ) {
3692 		DMSG(vdc, 1, "[%d] Received LDC_EVT_READ\n", vdc->instance);
3693 		mutex_enter(&vdc->read_lock);
3694 		cv_signal(&vdc->read_cv);
3695 		vdc->read_state = VDC_READ_PENDING;
3696 		mutex_exit(&vdc->read_lock);
3697 		mutex_exit(&vdc->lock);
3698 
3699 		/* that's all we have to do - no need to handle DOWN/RESET */
3700 		return (LDC_SUCCESS);
3701 	}
3702 
3703 	if (event & (LDC_EVT_RESET|LDC_EVT_DOWN)) {
3704 
3705 		DMSG(vdc, 0, "[%d] Received LDC RESET event\n", vdc->instance);
3706 
3707 		/*
3708 		 * Need to wake up any readers so they will
3709 		 * detect that a reset has occurred.
3710 		 */
3711 		mutex_enter(&vdc->read_lock);
3712 		if ((vdc->read_state == VDC_READ_WAITING) ||
3713 		    (vdc->read_state == VDC_READ_RESET))
3714 			cv_signal(&vdc->read_cv);
3715 		vdc->read_state = VDC_READ_RESET;
3716 		mutex_exit(&vdc->read_lock);
3717 
3718 		/* wake up any threads waiting for connection to come up */
3719 		if (vdc->state == VDC_STATE_INIT_WAITING) {
3720 			vdc->state = VDC_STATE_RESETTING;
3721 			cv_signal(&vdc->initwait_cv);
3722 		}
3723 
3724 	}
3725 
3726 	mutex_exit(&vdc->lock);
3727 
3728 	if (event & ~(LDC_EVT_UP | LDC_EVT_RESET | LDC_EVT_DOWN | LDC_EVT_READ))
3729 		DMSG(vdc, 0, "![%d] Unexpected LDC event (%lx) received",
3730 		    vdc->instance, event);
3731 
3732 	return (LDC_SUCCESS);
3733 }
3734 
3735 /*
3736  * Function:
3737  *	vdc_wait_for_response()
3738  *
3739  * Description:
3740  *	Block waiting for a response from the server. If there is
3741  *	no data the thread block on the read_cv that is signalled
3742  *	by the callback when an EVT_READ occurs.
3743  *
3744  * Arguments:
3745  *	vdcp	- soft state pointer for this instance of the device driver.
3746  *
3747  * Return Code:
3748  *	0	- Success
3749  */
3750 static int
3751 vdc_wait_for_response(vdc_t *vdcp, vio_msg_t *msgp)
3752 {
3753 	size_t		nbytes = sizeof (*msgp);
3754 	int		status;
3755 
3756 	ASSERT(vdcp != NULL);
3757 
3758 	DMSG(vdcp, 1, "[%d] Entered\n", vdcp->instance);
3759 
3760 	status = vdc_recv(vdcp, msgp, &nbytes);
3761 	DMSG(vdcp, 3, "vdc_read() done.. status=0x%x size=0x%x\n",
3762 	    status, (int)nbytes);
3763 	if (status) {
3764 		DMSG(vdcp, 0, "?[%d] Error %d reading LDC msg\n",
3765 		    vdcp->instance, status);
3766 		return (status);
3767 	}
3768 
3769 	if (nbytes < sizeof (vio_msg_tag_t)) {
3770 		DMSG(vdcp, 0, "?[%d] Expect %lu bytes; recv'd %lu\n",
3771 		    vdcp->instance, sizeof (vio_msg_tag_t), nbytes);
3772 		return (ENOMSG);
3773 	}
3774 
3775 	DMSG(vdcp, 2, "[%d] (%x/%x/%x)\n", vdcp->instance,
3776 	    msgp->tag.vio_msgtype,
3777 	    msgp->tag.vio_subtype,
3778 	    msgp->tag.vio_subtype_env);
3779 
3780 	/*
3781 	 * Verify the Session ID of the message
3782 	 *
3783 	 * Every message after the Version has been negotiated should
3784 	 * have the correct session ID set.
3785 	 */
3786 	if ((msgp->tag.vio_sid != vdcp->session_id) &&
3787 	    (msgp->tag.vio_subtype_env != VIO_VER_INFO)) {
3788 		DMSG(vdcp, 0, "[%d] Invalid SID: received 0x%x, "
3789 		    "expected 0x%lx [seq num %lx @ %d]",
3790 		    vdcp->instance, msgp->tag.vio_sid,
3791 		    vdcp->session_id,
3792 		    ((vio_dring_msg_t *)msgp)->seq_num,
3793 		    ((vio_dring_msg_t *)msgp)->start_idx);
3794 		return (ENOMSG);
3795 	}
3796 	return (0);
3797 }
3798 
3799 
3800 /*
3801  * Function:
3802  *	vdc_resubmit_backup_dring()
3803  *
3804  * Description:
3805  *	Resubmit each descriptor in the backed up dring to
3806  * 	vDisk server. The Dring was backed up during connection
3807  *	reset.
3808  *
3809  * Arguments:
3810  *	vdcp	- soft state pointer for this instance of the device driver.
3811  *
3812  * Return Code:
3813  *	0	- Success
3814  */
3815 static int
3816 vdc_resubmit_backup_dring(vdc_t *vdcp)
3817 {
3818 	int		processed = 0;
3819 	int		count;
3820 	int		b_idx;
3821 	int		rv = 0;
3822 	int		dring_size;
3823 	int		op;
3824 	vio_msg_t	vio_msg;
3825 	vdc_local_desc_t	*curr_ldep;
3826 
3827 	ASSERT(MUTEX_NOT_HELD(&vdcp->lock));
3828 	ASSERT(vdcp->state == VDC_STATE_HANDLE_PENDING);
3829 
3830 	if (vdcp->local_dring_backup == NULL) {
3831 		/* the pending requests have already been processed */
3832 		return (0);
3833 	}
3834 
3835 	DMSG(vdcp, 1, "restoring pending dring entries (len=%d, tail=%d)\n",
3836 	    vdcp->local_dring_backup_len, vdcp->local_dring_backup_tail);
3837 
3838 	/*
3839 	 * Walk the backup copy of the local descriptor ring and
3840 	 * resubmit all the outstanding transactions.
3841 	 */
3842 	b_idx = vdcp->local_dring_backup_tail;
3843 	for (count = 0; count < vdcp->local_dring_backup_len; count++) {
3844 
3845 		curr_ldep = &(vdcp->local_dring_backup[b_idx]);
3846 
3847 		/* only resubmit outstanding transactions */
3848 		if (!curr_ldep->is_free) {
3849 			/*
3850 			 * If we are retrying a block read/write operation we
3851 			 * need to update the I/O statistics to indicate that
3852 			 * the request is being put back on the waitq to be
3853 			 * serviced (it will have been taken off after the
3854 			 * error was reported).
3855 			 */
3856 			mutex_enter(&vdcp->lock);
3857 			op = curr_ldep->operation;
3858 			if ((op == VD_OP_BREAD) || (op == VD_OP_BWRITE)) {
3859 				DTRACE_IO1(start, buf_t *, curr_ldep->cb_arg);
3860 				VD_KSTAT_WAITQ_ENTER(vdcp);
3861 			}
3862 
3863 			DMSG(vdcp, 1, "resubmitting entry idx=%x\n", b_idx);
3864 			rv = vdc_populate_descriptor(vdcp, op,
3865 			    curr_ldep->addr, curr_ldep->nbytes,
3866 			    curr_ldep->slice, curr_ldep->offset,
3867 			    curr_ldep->cb_type, curr_ldep->cb_arg,
3868 			    curr_ldep->dir);
3869 
3870 			if (rv) {
3871 				if (op == VD_OP_BREAD || op == VD_OP_BWRITE) {
3872 					VD_UPDATE_ERR_STATS(vdcp, vd_transerrs);
3873 					VD_KSTAT_WAITQ_EXIT(vdcp);
3874 					DTRACE_IO1(done, buf_t *,
3875 					    curr_ldep->cb_arg);
3876 				}
3877 				DMSG(vdcp, 1, "[%d] cannot resubmit entry %d\n",
3878 				    vdcp->instance, b_idx);
3879 				mutex_exit(&vdcp->lock);
3880 				goto done;
3881 			}
3882 
3883 			/*
3884 			 * If this is a block read/write we update the I/O
3885 			 * statistics kstat to indicate that the request
3886 			 * has been sent back to the vDisk server and should
3887 			 * now be put on the run queue.
3888 			 */
3889 			if ((op == VD_OP_BREAD) || (op == VD_OP_BWRITE)) {
3890 				DTRACE_PROBE1(send, buf_t *, curr_ldep->cb_arg);
3891 				VD_KSTAT_WAITQ_TO_RUNQ(vdcp);
3892 			}
3893 			mutex_exit(&vdcp->lock);
3894 
3895 			/* Wait for the response message. */
3896 			DMSG(vdcp, 1, "waiting for response to idx=%x\n",
3897 			    b_idx);
3898 			rv = vdc_wait_for_response(vdcp, &vio_msg);
3899 			if (rv) {
3900 				/*
3901 				 * If this is a block read/write we update
3902 				 * the I/O statistics kstat to take it
3903 				 * off the run queue.
3904 				 */
3905 				mutex_enter(&vdcp->lock);
3906 				if (op == VD_OP_BREAD || op == VD_OP_BWRITE) {
3907 					VD_UPDATE_ERR_STATS(vdcp, vd_transerrs);
3908 					VD_KSTAT_RUNQ_EXIT(vdcp);
3909 					DTRACE_IO1(done, buf_t *,
3910 					    curr_ldep->cb_arg);
3911 				}
3912 				DMSG(vdcp, 1, "[%d] wait_for_response "
3913 				    "returned err=%d\n", vdcp->instance,
3914 				    rv);
3915 				mutex_exit(&vdcp->lock);
3916 				goto done;
3917 			}
3918 
3919 			DMSG(vdcp, 1, "processing msg for idx=%x\n", b_idx);
3920 			rv = vdc_process_data_msg(vdcp, &vio_msg);
3921 			if (rv) {
3922 				DMSG(vdcp, 1, "[%d] process_data_msg "
3923 				    "returned err=%d\n", vdcp->instance,
3924 				    rv);
3925 				goto done;
3926 			}
3927 			/*
3928 			 * Mark this entry as free so that we will not resubmit
3929 			 * this "done" request again, if we were to use the same
3930 			 * backup_dring again in future. This could happen when
3931 			 * a reset happens while processing the backup_dring.
3932 			 */
3933 			curr_ldep->is_free = B_TRUE;
3934 			processed++;
3935 		}
3936 
3937 		/* get the next element to submit */
3938 		if (++b_idx >= vdcp->local_dring_backup_len)
3939 			b_idx = 0;
3940 	}
3941 
3942 	/* all done - now clear up pending dring copy */
3943 	dring_size = vdcp->local_dring_backup_len *
3944 	    sizeof (vdcp->local_dring_backup[0]);
3945 
3946 	(void) kmem_free(vdcp->local_dring_backup, dring_size);
3947 
3948 	vdcp->local_dring_backup = NULL;
3949 
3950 done:
3951 	DTRACE_PROBE2(processed, int, processed, vdc_t *, vdcp);
3952 
3953 	return (rv);
3954 }
3955 
3956 /*
3957  * Function:
3958  *	vdc_cancel_backup_dring
3959  *
3960  * Description:
3961  *	Cancel each descriptor in the backed up dring to vDisk server.
3962  *	The Dring was backed up during connection reset.
3963  *
3964  * Arguments:
3965  *	vdcp	- soft state pointer for this instance of the device driver.
3966  *
3967  * Return Code:
3968  *	None
3969  */
3970 void
3971 vdc_cancel_backup_dring(vdc_t *vdcp)
3972 {
3973 	vdc_local_desc_t *ldep;
3974 	struct buf 	*bufp;
3975 	int		count;
3976 	int		b_idx;
3977 	int		dring_size;
3978 	int		cancelled = 0;
3979 
3980 	ASSERT(MUTEX_HELD(&vdcp->lock));
3981 	ASSERT(vdcp->state == VDC_STATE_INIT ||
3982 	    vdcp->state == VDC_STATE_INIT_WAITING ||
3983 	    vdcp->state == VDC_STATE_NEGOTIATE ||
3984 	    vdcp->state == VDC_STATE_RESETTING);
3985 
3986 	if (vdcp->local_dring_backup == NULL) {
3987 		/* the pending requests have already been processed */
3988 		return;
3989 	}
3990 
3991 	DMSG(vdcp, 1, "cancelling pending dring entries (len=%d, tail=%d)\n",
3992 	    vdcp->local_dring_backup_len, vdcp->local_dring_backup_tail);
3993 
3994 	/*
3995 	 * Walk the backup copy of the local descriptor ring and
3996 	 * cancel all the outstanding transactions.
3997 	 */
3998 	b_idx = vdcp->local_dring_backup_tail;
3999 	for (count = 0; count < vdcp->local_dring_backup_len; count++) {
4000 
4001 		ldep = &(vdcp->local_dring_backup[b_idx]);
4002 
4003 		/* only cancel outstanding transactions */
4004 		if (!ldep->is_free) {
4005 
4006 			DMSG(vdcp, 1, "cancelling entry idx=%x\n", b_idx);
4007 			cancelled++;
4008 
4009 			/*
4010 			 * All requests have already been cleared from the
4011 			 * local descriptor ring and the LDC channel has been
4012 			 * reset so we will never get any reply for these
4013 			 * requests. Now we just have to notify threads waiting
4014 			 * for replies that the request has failed.
4015 			 */
4016 			switch (ldep->cb_type) {
4017 			case CB_SYNC:
4018 				ASSERT(vdcp->sync_op_pending);
4019 				vdcp->sync_op_status = EIO;
4020 				vdcp->sync_op_pending = B_FALSE;
4021 				cv_signal(&vdcp->sync_pending_cv);
4022 				break;
4023 
4024 			case CB_STRATEGY:
4025 				bufp = ldep->cb_arg;
4026 				ASSERT(bufp != NULL);
4027 				bufp->b_resid = bufp->b_bcount;
4028 				VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4029 				VD_KSTAT_RUNQ_EXIT(vdcp);
4030 				DTRACE_IO1(done, buf_t *, bufp);
4031 				bioerror(bufp, EIO);
4032 				biodone(bufp);
4033 				break;
4034 
4035 			default:
4036 				ASSERT(0);
4037 			}
4038 
4039 		}
4040 
4041 		/* get the next element to cancel */
4042 		if (++b_idx >= vdcp->local_dring_backup_len)
4043 			b_idx = 0;
4044 	}
4045 
4046 	/* all done - now clear up pending dring copy */
4047 	dring_size = vdcp->local_dring_backup_len *
4048 	    sizeof (vdcp->local_dring_backup[0]);
4049 
4050 	(void) kmem_free(vdcp->local_dring_backup, dring_size);
4051 
4052 	vdcp->local_dring_backup = NULL;
4053 
4054 	DTRACE_PROBE2(cancelled, int, cancelled, vdc_t *, vdcp);
4055 }
4056 
4057 /*
4058  * Function:
4059  *	vdc_connection_timeout
4060  *
4061  * Description:
4062  *	This function is invoked if the timeout set to establish the connection
4063  *	with vds expires. This will happen if we spend too much time in the
4064  *	VDC_STATE_INIT_WAITING or VDC_STATE_NEGOTIATE states. Then we will
4065  *	cancel any pending request and mark them as failed.
4066  *
4067  *	If the timeout does not expire, it will be cancelled when we reach the
4068  *	VDC_STATE_HANDLE_PENDING or VDC_STATE_RESETTING state. This function can
4069  *	be invoked while we are in the VDC_STATE_HANDLE_PENDING or
4070  *	VDC_STATE_RESETTING state in which case we do nothing because the
4071  *	timeout is being cancelled.
4072  *
4073  * Arguments:
4074  *	arg	- argument of the timeout function actually a soft state
4075  *		  pointer for the instance of the device driver.
4076  *
4077  * Return Code:
4078  *	None
4079  */
4080 void
4081 vdc_connection_timeout(void *arg)
4082 {
4083 	vdc_t 		*vdcp = (vdc_t *)arg;
4084 
4085 	mutex_enter(&vdcp->lock);
4086 
4087 	if (vdcp->state == VDC_STATE_HANDLE_PENDING ||
4088 	    vdcp->state == VDC_STATE_DETACH) {
4089 		/*
4090 		 * The connection has just been re-established or
4091 		 * we are detaching.
4092 		 */
4093 		vdcp->ctimeout_reached = B_FALSE;
4094 		mutex_exit(&vdcp->lock);
4095 		return;
4096 	}
4097 
4098 	vdcp->ctimeout_reached = B_TRUE;
4099 
4100 	/* notify requests waiting for sending */
4101 	cv_broadcast(&vdcp->running_cv);
4102 
4103 	/* cancel requests waiting for a result */
4104 	vdc_cancel_backup_dring(vdcp);
4105 
4106 	mutex_exit(&vdcp->lock);
4107 
4108 	cmn_err(CE_NOTE, "[%d] connection to service domain timeout",
4109 	    vdcp->instance);
4110 }
4111 
4112 /*
4113  * Function:
4114  *	vdc_backup_local_dring()
4115  *
4116  * Description:
4117  *	Backup the current dring in the event of a reset. The Dring
4118  *	transactions will be resubmitted to the server when the
4119  *	connection is restored.
4120  *
4121  * Arguments:
4122  *	vdcp	- soft state pointer for this instance of the device driver.
4123  *
4124  * Return Code:
4125  *	NONE
4126  */
4127 static void
4128 vdc_backup_local_dring(vdc_t *vdcp)
4129 {
4130 	int dring_size;
4131 
4132 	ASSERT(MUTEX_HELD(&vdcp->lock));
4133 	ASSERT(vdcp->state == VDC_STATE_RESETTING);
4134 
4135 	/*
4136 	 * If the backup dring is stil around, it means
4137 	 * that the last restore did not complete. However,
4138 	 * since we never got back into the running state,
4139 	 * the backup copy we have is still valid.
4140 	 */
4141 	if (vdcp->local_dring_backup != NULL) {
4142 		DMSG(vdcp, 1, "reusing local descriptor ring backup "
4143 		    "(len=%d, tail=%d)\n", vdcp->local_dring_backup_len,
4144 		    vdcp->local_dring_backup_tail);
4145 		return;
4146 	}
4147 
4148 	/*
4149 	 * The backup dring can be NULL and the local dring may not be
4150 	 * initialized. This can happen if we had a reset while establishing
4151 	 * a new connection but after the connection has timed out. In that
4152 	 * case the backup dring is NULL because the requests have been
4153 	 * cancelled and the request occured before the local dring is
4154 	 * initialized.
4155 	 */
4156 	if (!(vdcp->initialized & VDC_DRING_LOCAL))
4157 		return;
4158 
4159 	DMSG(vdcp, 1, "backing up the local descriptor ring (len=%d, "
4160 	    "tail=%d)\n", vdcp->dring_len, vdcp->dring_curr_idx);
4161 
4162 	dring_size = vdcp->dring_len * sizeof (vdcp->local_dring[0]);
4163 
4164 	vdcp->local_dring_backup = kmem_alloc(dring_size, KM_SLEEP);
4165 	bcopy(vdcp->local_dring, vdcp->local_dring_backup, dring_size);
4166 
4167 	vdcp->local_dring_backup_tail = vdcp->dring_curr_idx;
4168 	vdcp->local_dring_backup_len = vdcp->dring_len;
4169 }
4170 
4171 static void
4172 vdc_switch_server(vdc_t *vdcp)
4173 {
4174 	int		rv;
4175 	vdc_server_t 	*curr_server, *new_server;
4176 
4177 	ASSERT(MUTEX_HELD(&vdcp->lock));
4178 
4179 	/* if there is only one server return back */
4180 	if (vdcp->num_servers == 1) {
4181 		return;
4182 	}
4183 
4184 	/* Get current and next server */
4185 	curr_server = vdcp->curr_server;
4186 	new_server =
4187 	    (curr_server->next) ? curr_server->next : vdcp->server_list;
4188 	ASSERT(curr_server != new_server);
4189 
4190 	/* bring current server's channel down */
4191 	rv = ldc_down(curr_server->ldc_handle);
4192 	if (rv) {
4193 		DMSG(vdcp, 0, "[%d] Cannot bring channel down, port %ld\n",
4194 		    vdcp->instance, curr_server->id);
4195 		return;
4196 	}
4197 
4198 	/* switch the server */
4199 	vdcp->curr_server = new_server;
4200 
4201 	DMSG(vdcp, 0, "[%d] Switched to next vdisk server, port@%ld, ldc@%ld\n",
4202 	    vdcp->instance, vdcp->curr_server->id, vdcp->curr_server->ldc_id);
4203 }
4204 
4205 /* -------------------------------------------------------------------------- */
4206 
4207 /*
4208  * The following functions process the incoming messages from vds
4209  */
4210 
4211 /*
4212  * Function:
4213  *      vdc_process_msg_thread()
4214  *
4215  * Description:
4216  *
4217  *	Main VDC message processing thread. Each vDisk instance
4218  * 	consists of a copy of this thread. This thread triggers
4219  * 	all the handshakes and data exchange with the server. It
4220  * 	also handles all channel resets
4221  *
4222  * Arguments:
4223  *      vdc     - soft state pointer for this instance of the device driver.
4224  *
4225  * Return Code:
4226  *      None
4227  */
4228 static void
4229 vdc_process_msg_thread(vdc_t *vdcp)
4230 {
4231 	int		status;
4232 	int		ctimeout;
4233 	timeout_id_t	tmid = 0;
4234 	clock_t		ldcup_timeout = 0;
4235 
4236 	mutex_enter(&vdcp->lock);
4237 
4238 	for (;;) {
4239 
4240 #define	Q(_s)	(vdcp->state == _s) ? #_s :
4241 		DMSG(vdcp, 3, "state = %d (%s)\n", vdcp->state,
4242 		    Q(VDC_STATE_INIT)
4243 		    Q(VDC_STATE_INIT_WAITING)
4244 		    Q(VDC_STATE_NEGOTIATE)
4245 		    Q(VDC_STATE_HANDLE_PENDING)
4246 		    Q(VDC_STATE_RUNNING)
4247 		    Q(VDC_STATE_RESETTING)
4248 		    Q(VDC_STATE_DETACH)
4249 		    "UNKNOWN");
4250 
4251 		switch (vdcp->state) {
4252 		case VDC_STATE_INIT:
4253 
4254 			/*
4255 			 * If requested, start a timeout to check if the
4256 			 * connection with vds is established in the
4257 			 * specified delay. If the timeout expires, we
4258 			 * will cancel any pending request.
4259 			 *
4260 			 * If some reset have occurred while establishing
4261 			 * the connection, we already have a timeout armed
4262 			 * and in that case we don't need to arm a new one.
4263 			 *
4264 			 * The same rule applies when there are multiple vds'.
4265 			 * If either a connection cannot be established or
4266 			 * the handshake times out, the connection thread will
4267 			 * try another server. The 'ctimeout' will report
4268 			 * back an error after it expires irrespective of
4269 			 * whether the vdisk is trying to connect to just
4270 			 * one or multiple servers.
4271 			 */
4272 			ctimeout = (vdc_timeout != 0)?
4273 			    vdc_timeout : vdcp->curr_server->ctimeout;
4274 
4275 			if (ctimeout != 0 && tmid == 0) {
4276 				tmid = timeout(vdc_connection_timeout, vdcp,
4277 				    ctimeout * drv_usectohz(MICROSEC));
4278 			}
4279 
4280 			/* Check if we are re-initializing repeatedly */
4281 			if (vdcp->hshake_cnt > vdc_hshake_retries &&
4282 			    vdcp->lifecycle != VDC_LC_ONLINE) {
4283 
4284 				DMSG(vdcp, 0, "[%d] too many handshakes,cnt=%d",
4285 				    vdcp->instance, vdcp->hshake_cnt);
4286 				cmn_err(CE_NOTE, "[%d] disk access failed.\n",
4287 				    vdcp->instance);
4288 				vdcp->state = VDC_STATE_DETACH;
4289 				break;
4290 			}
4291 
4292 			/* Switch to STATE_DETACH if drv is detaching */
4293 			if (vdcp->lifecycle == VDC_LC_DETACHING) {
4294 				vdcp->state = VDC_STATE_DETACH;
4295 				break;
4296 			}
4297 
4298 			/* Switch server */
4299 			if (vdcp->hshake_cnt > 0)
4300 				vdc_switch_server(vdcp);
4301 			vdcp->hshake_cnt++;
4302 
4303 			/* Bring up connection with vds via LDC */
4304 			status = vdc_start_ldc_connection(vdcp);
4305 			if (status != EINVAL) {
4306 				vdcp->state = VDC_STATE_INIT_WAITING;
4307 			}
4308 			break;
4309 
4310 		case VDC_STATE_INIT_WAITING:
4311 
4312 			/* if channel is UP, start negotiation */
4313 			if (vdcp->curr_server->ldc_state == LDC_UP) {
4314 				vdcp->state = VDC_STATE_NEGOTIATE;
4315 				break;
4316 			}
4317 
4318 			/* check if only one server exists */
4319 			if (vdcp->num_servers == 1) {
4320 				cv_wait(&vdcp->initwait_cv, &vdcp->lock);
4321 			} else {
4322 				/*
4323 				 * wait for LDC_UP, if it times out, switch
4324 				 * to another server.
4325 				 */
4326 				ldcup_timeout = ddi_get_lbolt() +
4327 				    (vdc_ldcup_timeout *
4328 				    drv_usectohz(MICROSEC));
4329 				status = cv_timedwait(&vdcp->initwait_cv,
4330 				    &vdcp->lock, ldcup_timeout);
4331 				if (status == -1 &&
4332 				    vdcp->state == VDC_STATE_INIT_WAITING &&
4333 				    vdcp->curr_server->ldc_state != LDC_UP) {
4334 					/* timed out & still waiting */
4335 					vdcp->state = VDC_STATE_INIT;
4336 					break;
4337 				}
4338 			}
4339 
4340 			if (vdcp->state != VDC_STATE_INIT_WAITING) {
4341 				DMSG(vdcp, 0,
4342 				    "state moved to %d out from under us...\n",
4343 				    vdcp->state);
4344 			}
4345 			break;
4346 
4347 		case VDC_STATE_NEGOTIATE:
4348 			switch (status = vdc_ver_negotiation(vdcp)) {
4349 			case 0:
4350 				break;
4351 			default:
4352 				DMSG(vdcp, 0, "ver negotiate failed (%d)..\n",
4353 				    status);
4354 				goto reset;
4355 			}
4356 
4357 			switch (status = vdc_attr_negotiation(vdcp)) {
4358 			case 0:
4359 				break;
4360 			default:
4361 				DMSG(vdcp, 0, "attr negotiate failed (%d)..\n",
4362 				    status);
4363 				goto reset;
4364 			}
4365 
4366 			switch (status = vdc_dring_negotiation(vdcp)) {
4367 			case 0:
4368 				break;
4369 			default:
4370 				DMSG(vdcp, 0, "dring negotiate failed (%d)..\n",
4371 				    status);
4372 				goto reset;
4373 			}
4374 
4375 			switch (status = vdc_rdx_exchange(vdcp)) {
4376 			case 0:
4377 				vdcp->state = VDC_STATE_HANDLE_PENDING;
4378 				goto done;
4379 			default:
4380 				DMSG(vdcp, 0, "RDX xchg failed ..(%d)\n",
4381 				    status);
4382 				goto reset;
4383 			}
4384 reset:
4385 			DMSG(vdcp, 0, "negotiation failed: resetting (%d)\n",
4386 			    status);
4387 			vdcp->state = VDC_STATE_RESETTING;
4388 			vdcp->self_reset = B_TRUE;
4389 done:
4390 			DMSG(vdcp, 0, "negotiation complete (state=0x%x)...\n",
4391 			    vdcp->state);
4392 			break;
4393 
4394 		case VDC_STATE_HANDLE_PENDING:
4395 
4396 			if (vdcp->ctimeout_reached) {
4397 				/*
4398 				 * The connection timeout had been reached so
4399 				 * pending requests have been cancelled. Now
4400 				 * that the connection is back we can reset
4401 				 * the timeout.
4402 				 */
4403 				ASSERT(vdcp->local_dring_backup == NULL);
4404 				ASSERT(tmid != 0);
4405 				tmid = 0;
4406 				vdcp->ctimeout_reached = B_FALSE;
4407 				vdcp->state = VDC_STATE_RUNNING;
4408 				DMSG(vdcp, 0, "[%d] connection to service "
4409 				    "domain is up", vdcp->instance);
4410 				break;
4411 			}
4412 
4413 			mutex_exit(&vdcp->lock);
4414 			if (tmid != 0) {
4415 				(void) untimeout(tmid);
4416 				tmid = 0;
4417 			}
4418 			status = vdc_resubmit_backup_dring(vdcp);
4419 			mutex_enter(&vdcp->lock);
4420 
4421 			if (status)
4422 				vdcp->state = VDC_STATE_RESETTING;
4423 			else
4424 				vdcp->state = VDC_STATE_RUNNING;
4425 
4426 			break;
4427 
4428 		/* enter running state */
4429 		case VDC_STATE_RUNNING:
4430 			/*
4431 			 * Signal anyone waiting for the connection
4432 			 * to come on line.
4433 			 */
4434 			vdcp->hshake_cnt = 0;
4435 			cv_broadcast(&vdcp->running_cv);
4436 
4437 			/* failfast has to been checked after reset */
4438 			cv_signal(&vdcp->failfast_cv);
4439 
4440 			/* ownership is lost during reset */
4441 			if (vdcp->ownership & VDC_OWNERSHIP_WANTED)
4442 				vdcp->ownership |= VDC_OWNERSHIP_RESET;
4443 			cv_signal(&vdcp->ownership_cv);
4444 
4445 			cmn_err(CE_CONT, "?vdisk@%d is online using "
4446 			    "ldc@%ld,%ld\n", vdcp->instance,
4447 			    vdcp->curr_server->ldc_id, vdcp->curr_server->id);
4448 
4449 			mutex_exit(&vdcp->lock);
4450 
4451 			for (;;) {
4452 				vio_msg_t msg;
4453 				status = vdc_wait_for_response(vdcp, &msg);
4454 				if (status) break;
4455 
4456 				DMSG(vdcp, 1, "[%d] new pkt(s) available\n",
4457 				    vdcp->instance);
4458 				status = vdc_process_data_msg(vdcp, &msg);
4459 				if (status) {
4460 					DMSG(vdcp, 1, "[%d] process_data_msg "
4461 					    "returned err=%d\n", vdcp->instance,
4462 					    status);
4463 					break;
4464 				}
4465 
4466 			}
4467 
4468 			mutex_enter(&vdcp->lock);
4469 
4470 			cmn_err(CE_CONT, "?vdisk@%d is offline\n",
4471 			    vdcp->instance);
4472 
4473 			vdcp->state = VDC_STATE_RESETTING;
4474 			vdcp->self_reset = B_TRUE;
4475 			break;
4476 
4477 		case VDC_STATE_RESETTING:
4478 			/*
4479 			 * When we reach this state, we either come from the
4480 			 * VDC_STATE_RUNNING state and we can have pending
4481 			 * request but no timeout is armed; or we come from
4482 			 * the VDC_STATE_INIT_WAITING, VDC_NEGOTIATE or
4483 			 * VDC_HANDLE_PENDING state and there is no pending
4484 			 * request or pending requests have already been copied
4485 			 * into the backup dring. So we can safely keep the
4486 			 * connection timeout armed while we are in this state.
4487 			 */
4488 
4489 			DMSG(vdcp, 0, "Initiating channel reset "
4490 			    "(pending = %d)\n", (int)vdcp->threads_pending);
4491 
4492 			if (vdcp->self_reset) {
4493 				DMSG(vdcp, 0,
4494 				    "[%d] calling stop_ldc_connection.\n",
4495 				    vdcp->instance);
4496 				status = vdc_stop_ldc_connection(vdcp);
4497 				vdcp->self_reset = B_FALSE;
4498 			}
4499 
4500 			/*
4501 			 * Wait for all threads currently waiting
4502 			 * for a free dring entry to use.
4503 			 */
4504 			while (vdcp->threads_pending) {
4505 				cv_broadcast(&vdcp->membind_cv);
4506 				cv_broadcast(&vdcp->dring_free_cv);
4507 				mutex_exit(&vdcp->lock);
4508 				/* give the waiters enough time to wake up */
4509 				delay(vdc_hz_min_ldc_delay);
4510 				mutex_enter(&vdcp->lock);
4511 			}
4512 
4513 			ASSERT(vdcp->threads_pending == 0);
4514 
4515 			/* Sanity check that no thread is receiving */
4516 			ASSERT(vdcp->read_state != VDC_READ_WAITING);
4517 
4518 			vdcp->read_state = VDC_READ_IDLE;
4519 
4520 			vdc_backup_local_dring(vdcp);
4521 
4522 			/* cleanup the old d-ring */
4523 			vdc_destroy_descriptor_ring(vdcp);
4524 
4525 			/* go and start again */
4526 			vdcp->state = VDC_STATE_INIT;
4527 
4528 			break;
4529 
4530 		case VDC_STATE_DETACH:
4531 			DMSG(vdcp, 0, "[%d] Reset thread exit cleanup ..\n",
4532 			    vdcp->instance);
4533 
4534 			/* cancel any pending timeout */
4535 			mutex_exit(&vdcp->lock);
4536 			if (tmid != 0) {
4537 				(void) untimeout(tmid);
4538 				tmid = 0;
4539 			}
4540 			mutex_enter(&vdcp->lock);
4541 
4542 			/*
4543 			 * Signal anyone waiting for connection
4544 			 * to come online
4545 			 */
4546 			cv_broadcast(&vdcp->running_cv);
4547 
4548 			while (vdcp->sync_op_pending) {
4549 				cv_signal(&vdcp->sync_pending_cv);
4550 				cv_signal(&vdcp->sync_blocked_cv);
4551 				mutex_exit(&vdcp->lock);
4552 				/* give the waiters enough time to wake up */
4553 				delay(vdc_hz_min_ldc_delay);
4554 				mutex_enter(&vdcp->lock);
4555 			}
4556 
4557 			mutex_exit(&vdcp->lock);
4558 
4559 			DMSG(vdcp, 0, "[%d] Msg processing thread exiting ..\n",
4560 			    vdcp->instance);
4561 			thread_exit();
4562 			break;
4563 		}
4564 	}
4565 }
4566 
4567 
4568 /*
4569  * Function:
4570  *	vdc_process_data_msg()
4571  *
4572  * Description:
4573  *	This function is called by the message processing thread each time
4574  *	a message with a msgtype of VIO_TYPE_DATA is received. It will either
4575  *	be an ACK or NACK from vds[1] which vdc handles as follows.
4576  *		ACK	- wake up the waiting thread
4577  *		NACK	- resend any messages necessary
4578  *
4579  *	[1] Although the message format allows it, vds should not send a
4580  *	    VIO_SUBTYPE_INFO message to vdc asking it to read data; if for
4581  *	    some bizarre reason it does, vdc will reset the connection.
4582  *
4583  * Arguments:
4584  *	vdc	- soft state pointer for this instance of the device driver.
4585  *	msg	- the LDC message sent by vds
4586  *
4587  * Return Code:
4588  *	0	- Success.
4589  *	> 0	- error value returned by LDC
4590  */
4591 static int
4592 vdc_process_data_msg(vdc_t *vdcp, vio_msg_t *msg)
4593 {
4594 	int			status = 0;
4595 	vio_dring_msg_t		*dring_msg;
4596 	vdc_local_desc_t	*ldep = NULL;
4597 	int			start, end;
4598 	int			idx;
4599 	int			op;
4600 
4601 	dring_msg = (vio_dring_msg_t *)msg;
4602 
4603 	ASSERT(msg->tag.vio_msgtype == VIO_TYPE_DATA);
4604 	ASSERT(vdcp != NULL);
4605 
4606 	mutex_enter(&vdcp->lock);
4607 
4608 	/*
4609 	 * Check to see if the message has bogus data
4610 	 */
4611 	idx = start = dring_msg->start_idx;
4612 	end = dring_msg->end_idx;
4613 	if ((start >= vdcp->dring_len) ||
4614 	    (end >= vdcp->dring_len) || (end < -1)) {
4615 		/*
4616 		 * Update the I/O statistics to indicate that an error ocurred.
4617 		 * No need to update the wait/run queues as no specific read or
4618 		 * write request is being completed in response to this 'msg'.
4619 		 */
4620 		VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4621 		DMSG(vdcp, 0, "[%d] Bogus ACK data : start %d, end %d\n",
4622 		    vdcp->instance, start, end);
4623 		mutex_exit(&vdcp->lock);
4624 		return (EINVAL);
4625 	}
4626 
4627 	/*
4628 	 * Verify that the sequence number is what vdc expects.
4629 	 */
4630 	switch (vdc_verify_seq_num(vdcp, dring_msg)) {
4631 	case VDC_SEQ_NUM_TODO:
4632 		break;	/* keep processing this message */
4633 	case VDC_SEQ_NUM_SKIP:
4634 		mutex_exit(&vdcp->lock);
4635 		return (0);
4636 	case VDC_SEQ_NUM_INVALID:
4637 		/*
4638 		 * Update the I/O statistics to indicate that an error ocurred.
4639 		 * No need to update the wait/run queues as no specific read or
4640 		 * write request is being completed in response to this 'msg'.
4641 		 */
4642 		VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4643 		DMSG(vdcp, 0, "[%d] invalid seqno\n", vdcp->instance);
4644 		mutex_exit(&vdcp->lock);
4645 		return (ENXIO);
4646 	}
4647 
4648 	if (msg->tag.vio_subtype == VIO_SUBTYPE_NACK) {
4649 		/*
4650 		 * Update the I/O statistics to indicate that an error ocurred.
4651 		 *
4652 		 * We need to update the run queue if a read or write request
4653 		 * is being NACKed - otherwise there will appear to be an
4654 		 * indefinite outstanding request and statistics reported by
4655 		 * iostat(1M) will be incorrect. The transaction will be
4656 		 * resubmitted from the backup DRing following the reset
4657 		 * and the wait/run queues will be entered again.
4658 		 */
4659 		ldep = &vdcp->local_dring[idx];
4660 		op = ldep->operation;
4661 		if ((op == VD_OP_BREAD) || (op == VD_OP_BWRITE)) {
4662 			DTRACE_IO1(done, buf_t *, ldep->cb_arg);
4663 			VD_KSTAT_RUNQ_EXIT(vdcp);
4664 		}
4665 		VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4666 		VDC_DUMP_DRING_MSG(dring_msg);
4667 		DMSG(vdcp, 0, "[%d] DATA NACK\n", vdcp->instance);
4668 		mutex_exit(&vdcp->lock);
4669 		return (EIO);
4670 
4671 	} else if (msg->tag.vio_subtype == VIO_SUBTYPE_INFO) {
4672 		/*
4673 		 * Update the I/O statistics to indicate that an error occurred.
4674 		 * No need to update the wait/run queues as no specific read or
4675 		 * write request is being completed in response to this 'msg'.
4676 		 */
4677 		VD_UPDATE_ERR_STATS(vdcp, vd_protoerrs);
4678 		mutex_exit(&vdcp->lock);
4679 		return (EPROTO);
4680 	}
4681 
4682 	DMSG(vdcp, 1, ": start %d end %d\n", start, end);
4683 	ASSERT(start == end);
4684 
4685 	ldep = &vdcp->local_dring[idx];
4686 
4687 	DMSG(vdcp, 1, ": state 0x%x - cb_type 0x%x\n",
4688 	    ldep->dep->hdr.dstate, ldep->cb_type);
4689 
4690 	if (ldep->dep->hdr.dstate == VIO_DESC_DONE) {
4691 		struct buf *bufp;
4692 
4693 		switch (ldep->cb_type) {
4694 		case CB_SYNC:
4695 			ASSERT(vdcp->sync_op_pending);
4696 
4697 			status = vdc_depopulate_descriptor(vdcp, idx);
4698 			vdcp->sync_op_status = status;
4699 			vdcp->sync_op_pending = B_FALSE;
4700 			cv_signal(&vdcp->sync_pending_cv);
4701 			break;
4702 
4703 		case CB_STRATEGY:
4704 			bufp = ldep->cb_arg;
4705 			ASSERT(bufp != NULL);
4706 			bufp->b_resid =
4707 			    bufp->b_bcount - ldep->dep->payload.nbytes;
4708 			status = ldep->dep->payload.status; /* Future:ntoh */
4709 			if (status != 0) {
4710 				DMSG(vdcp, 1, "strategy status=%d\n", status);
4711 				VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4712 				bioerror(bufp, status);
4713 			}
4714 
4715 			(void) vdc_depopulate_descriptor(vdcp, idx);
4716 
4717 			DMSG(vdcp, 1,
4718 			    "strategy complete req=%ld bytes resp=%ld bytes\n",
4719 			    bufp->b_bcount, ldep->dep->payload.nbytes);
4720 
4721 			if (status != 0 && vdcp->failfast_interval != 0) {
4722 				/*
4723 				 * The I/O has failed and failfast is enabled.
4724 				 * We need the failfast thread to check if the
4725 				 * failure is due to a reservation conflict.
4726 				 */
4727 				(void) vdc_failfast_io_queue(vdcp, bufp);
4728 			} else {
4729 				if (status == 0) {
4730 					op = (bufp->b_flags & B_READ) ?
4731 					    VD_OP_BREAD : VD_OP_BWRITE;
4732 					VD_UPDATE_IO_STATS(vdcp, op,
4733 					    ldep->dep->payload.nbytes);
4734 				}
4735 				VD_KSTAT_RUNQ_EXIT(vdcp);
4736 				DTRACE_IO1(done, buf_t *, bufp);
4737 				biodone(bufp);
4738 			}
4739 			break;
4740 
4741 		default:
4742 			ASSERT(0);
4743 		}
4744 	}
4745 
4746 	/* let the arrival signal propogate */
4747 	mutex_exit(&vdcp->lock);
4748 
4749 	/* probe gives the count of how many entries were processed */
4750 	DTRACE_PROBE2(processed, int, 1, vdc_t *, vdcp);
4751 
4752 	return (0);
4753 }
4754 
4755 
4756 /*
4757  * Function:
4758  *	vdc_handle_ver_msg()
4759  *
4760  * Description:
4761  *
4762  * Arguments:
4763  *	vdc	- soft state pointer for this instance of the device driver.
4764  *	ver_msg	- LDC message sent by vDisk server
4765  *
4766  * Return Code:
4767  *	0	- Success
4768  */
4769 static int
4770 vdc_handle_ver_msg(vdc_t *vdc, vio_ver_msg_t *ver_msg)
4771 {
4772 	int status = 0;
4773 
4774 	ASSERT(vdc != NULL);
4775 	ASSERT(mutex_owned(&vdc->lock));
4776 
4777 	if (ver_msg->tag.vio_subtype_env != VIO_VER_INFO) {
4778 		return (EPROTO);
4779 	}
4780 
4781 	if (ver_msg->dev_class != VDEV_DISK_SERVER) {
4782 		return (EINVAL);
4783 	}
4784 
4785 	switch (ver_msg->tag.vio_subtype) {
4786 	case VIO_SUBTYPE_ACK:
4787 		/*
4788 		 * We check to see if the version returned is indeed supported
4789 		 * (The server may have also adjusted the minor number downwards
4790 		 * and if so 'ver_msg' will contain the actual version agreed)
4791 		 */
4792 		if (vdc_is_supported_version(ver_msg)) {
4793 			vdc->ver.major = ver_msg->ver_major;
4794 			vdc->ver.minor = ver_msg->ver_minor;
4795 			ASSERT(vdc->ver.major > 0);
4796 		} else {
4797 			status = EPROTO;
4798 		}
4799 		break;
4800 
4801 	case VIO_SUBTYPE_NACK:
4802 		/*
4803 		 * call vdc_is_supported_version() which will return the next
4804 		 * supported version (if any) in 'ver_msg'
4805 		 */
4806 		(void) vdc_is_supported_version(ver_msg);
4807 		if (ver_msg->ver_major > 0) {
4808 			size_t len = sizeof (*ver_msg);
4809 
4810 			ASSERT(vdc->ver.major > 0);
4811 
4812 			/* reset the necessary fields and resend */
4813 			ver_msg->tag.vio_subtype = VIO_SUBTYPE_INFO;
4814 			ver_msg->dev_class = VDEV_DISK;
4815 
4816 			status = vdc_send(vdc, (caddr_t)ver_msg, &len);
4817 			DMSG(vdc, 0, "[%d] Resend VER info (LDC status = %d)\n",
4818 			    vdc->instance, status);
4819 			if (len != sizeof (*ver_msg))
4820 				status = EBADMSG;
4821 		} else {
4822 			DMSG(vdc, 0, "[%d] No common version with vDisk server",
4823 			    vdc->instance);
4824 			status = ENOTSUP;
4825 		}
4826 
4827 		break;
4828 	case VIO_SUBTYPE_INFO:
4829 		/*
4830 		 * Handle the case where vds starts handshake
4831 		 * (for now only vdc is the instigator)
4832 		 */
4833 		status = ENOTSUP;
4834 		break;
4835 
4836 	default:
4837 		status = EINVAL;
4838 		break;
4839 	}
4840 
4841 	return (status);
4842 }
4843 
4844 /*
4845  * Function:
4846  *	vdc_handle_attr_msg()
4847  *
4848  * Description:
4849  *
4850  * Arguments:
4851  *	vdc	- soft state pointer for this instance of the device driver.
4852  *	attr_msg	- LDC message sent by vDisk server
4853  *
4854  * Return Code:
4855  *	0	- Success
4856  */
4857 static int
4858 vdc_handle_attr_msg(vdc_t *vdc, vd_attr_msg_t *attr_msg)
4859 {
4860 	int status = 0;
4861 
4862 	ASSERT(vdc != NULL);
4863 	ASSERT(mutex_owned(&vdc->lock));
4864 
4865 	if (attr_msg->tag.vio_subtype_env != VIO_ATTR_INFO) {
4866 		return (EPROTO);
4867 	}
4868 
4869 	switch (attr_msg->tag.vio_subtype) {
4870 	case VIO_SUBTYPE_ACK:
4871 		/*
4872 		 * We now verify the attributes sent by vds.
4873 		 */
4874 		if (attr_msg->vdisk_size == 0) {
4875 			DMSG(vdc, 0, "[%d] Invalid disk size from vds",
4876 			    vdc->instance);
4877 			status = EINVAL;
4878 			break;
4879 		}
4880 
4881 		if (attr_msg->max_xfer_sz == 0) {
4882 			DMSG(vdc, 0, "[%d] Invalid transfer size from vds",
4883 			    vdc->instance);
4884 			status = EINVAL;
4885 			break;
4886 		}
4887 
4888 		if (attr_msg->vdisk_size == VD_SIZE_UNKNOWN) {
4889 			DMSG(vdc, 0, "[%d] Unknown disk size from vds",
4890 			    vdc->instance);
4891 			attr_msg->vdisk_size = 0;
4892 		}
4893 
4894 		/* update the VIO block size */
4895 		if (attr_msg->vdisk_block_size > 0 &&
4896 		    vdc_update_vio_bsize(vdc,
4897 		    attr_msg->vdisk_block_size) != 0) {
4898 			DMSG(vdc, 0, "[%d] Invalid block size (%u) from vds",
4899 			    vdc->instance, attr_msg->vdisk_block_size);
4900 			status = EINVAL;
4901 			break;
4902 		}
4903 
4904 		/* update disk, block and transfer sizes */
4905 		vdc_update_size(vdc, attr_msg->vdisk_size,
4906 		    attr_msg->vdisk_block_size, attr_msg->max_xfer_sz);
4907 		vdc->vdisk_type = attr_msg->vdisk_type;
4908 		vdc->operations = attr_msg->operations;
4909 		if (vio_ver_is_supported(vdc->ver, 1, 1))
4910 			vdc->vdisk_media = attr_msg->vdisk_media;
4911 		else
4912 			vdc->vdisk_media = 0;
4913 
4914 		DMSG(vdc, 0, "[%d] max_xfer_sz: sent %lx acked %lx\n",
4915 		    vdc->instance, vdc->max_xfer_sz, attr_msg->max_xfer_sz);
4916 		DMSG(vdc, 0, "[%d] vdisk_block_size: sent %lx acked %x\n",
4917 		    vdc->instance, vdc->vdisk_bsize,
4918 		    attr_msg->vdisk_block_size);
4919 
4920 		if ((attr_msg->xfer_mode != VIO_DRING_MODE_V1_0) ||
4921 		    (attr_msg->vdisk_size > INT64_MAX) ||
4922 		    (attr_msg->operations == 0) ||
4923 		    (attr_msg->vdisk_type > VD_DISK_TYPE_DISK)) {
4924 			DMSG(vdc, 0, "[%d] Invalid attributes from vds",
4925 			    vdc->instance);
4926 			status = EINVAL;
4927 			break;
4928 		}
4929 
4930 		/*
4931 		 * Now that we have received all attributes we can create a
4932 		 * fake geometry for the disk.
4933 		 */
4934 		vdc_create_fake_geometry(vdc);
4935 		break;
4936 
4937 	case VIO_SUBTYPE_NACK:
4938 		/*
4939 		 * vds could not handle the attributes we sent so we
4940 		 * stop negotiating.
4941 		 */
4942 		status = EPROTO;
4943 		break;
4944 
4945 	case VIO_SUBTYPE_INFO:
4946 		/*
4947 		 * Handle the case where vds starts the handshake
4948 		 * (for now; vdc is the only supported instigatior)
4949 		 */
4950 		status = ENOTSUP;
4951 		break;
4952 
4953 	default:
4954 		status = ENOTSUP;
4955 		break;
4956 	}
4957 
4958 	return (status);
4959 }
4960 
4961 /*
4962  * Function:
4963  *	vdc_handle_dring_reg_msg()
4964  *
4965  * Description:
4966  *
4967  * Arguments:
4968  *	vdc		- soft state pointer for this instance of the driver.
4969  *	dring_msg	- LDC message sent by vDisk server
4970  *
4971  * Return Code:
4972  *	0	- Success
4973  */
4974 static int
4975 vdc_handle_dring_reg_msg(vdc_t *vdc, vio_dring_reg_msg_t *dring_msg)
4976 {
4977 	int		status = 0;
4978 
4979 	ASSERT(vdc != NULL);
4980 	ASSERT(mutex_owned(&vdc->lock));
4981 
4982 	if (dring_msg->tag.vio_subtype_env != VIO_DRING_REG) {
4983 		return (EPROTO);
4984 	}
4985 
4986 	switch (dring_msg->tag.vio_subtype) {
4987 	case VIO_SUBTYPE_ACK:
4988 		/* save the received dring_ident */
4989 		vdc->dring_ident = dring_msg->dring_ident;
4990 		DMSG(vdc, 0, "[%d] Received dring ident=0x%lx\n",
4991 		    vdc->instance, vdc->dring_ident);
4992 		break;
4993 
4994 	case VIO_SUBTYPE_NACK:
4995 		/*
4996 		 * vds could not handle the DRing info we sent so we
4997 		 * stop negotiating.
4998 		 */
4999 		DMSG(vdc, 0, "[%d] server could not register DRing\n",
5000 		    vdc->instance);
5001 		status = EPROTO;
5002 		break;
5003 
5004 	case VIO_SUBTYPE_INFO:
5005 		/*
5006 		 * Handle the case where vds starts handshake
5007 		 * (for now only vdc is the instigatior)
5008 		 */
5009 		status = ENOTSUP;
5010 		break;
5011 	default:
5012 		status = ENOTSUP;
5013 	}
5014 
5015 	return (status);
5016 }
5017 
5018 /*
5019  * Function:
5020  *	vdc_verify_seq_num()
5021  *
5022  * Description:
5023  *	This functions verifies that the sequence number sent back by the vDisk
5024  *	server with the latest message is what is expected (i.e. it is greater
5025  *	than the last seq num sent by the vDisk server and less than or equal
5026  *	to the last seq num generated by vdc).
5027  *
5028  *	It then checks the request ID to see if any requests need processing
5029  *	in the DRing.
5030  *
5031  * Arguments:
5032  *	vdc		- soft state pointer for this instance of the driver.
5033  *	dring_msg	- pointer to the LDC message sent by vds
5034  *
5035  * Return Code:
5036  *	VDC_SEQ_NUM_TODO	- Message needs to be processed
5037  *	VDC_SEQ_NUM_SKIP	- Message has already been processed
5038  *	VDC_SEQ_NUM_INVALID	- The seq numbers are so out of sync,
5039  *				  vdc cannot deal with them
5040  */
5041 static int
5042 vdc_verify_seq_num(vdc_t *vdc, vio_dring_msg_t *dring_msg)
5043 {
5044 	ASSERT(vdc != NULL);
5045 	ASSERT(dring_msg != NULL);
5046 	ASSERT(mutex_owned(&vdc->lock));
5047 
5048 	/*
5049 	 * Check to see if the messages were responded to in the correct
5050 	 * order by vds.
5051 	 */
5052 	if ((dring_msg->seq_num <= vdc->seq_num_reply) ||
5053 	    (dring_msg->seq_num > vdc->seq_num)) {
5054 		DMSG(vdc, 0, "?[%d] Bogus sequence_number %lu: "
5055 		    "%lu > expected <= %lu (last proc req %lu sent %lu)\n",
5056 		    vdc->instance, dring_msg->seq_num,
5057 		    vdc->seq_num_reply, vdc->seq_num,
5058 		    vdc->req_id_proc, vdc->req_id);
5059 		return (VDC_SEQ_NUM_INVALID);
5060 	}
5061 	vdc->seq_num_reply = dring_msg->seq_num;
5062 
5063 	if (vdc->req_id_proc < vdc->req_id)
5064 		return (VDC_SEQ_NUM_TODO);
5065 	else
5066 		return (VDC_SEQ_NUM_SKIP);
5067 }
5068 
5069 
5070 /*
5071  * Function:
5072  *	vdc_is_supported_version()
5073  *
5074  * Description:
5075  *	This routine checks if the major/minor version numbers specified in
5076  *	'ver_msg' are supported. If not it finds the next version that is
5077  *	in the supported version list 'vdc_version[]' and sets the fields in
5078  *	'ver_msg' to those values
5079  *
5080  * Arguments:
5081  *	ver_msg	- LDC message sent by vDisk server
5082  *
5083  * Return Code:
5084  *	B_TRUE	- Success
5085  *	B_FALSE	- Version not supported
5086  */
5087 static boolean_t
5088 vdc_is_supported_version(vio_ver_msg_t *ver_msg)
5089 {
5090 	int vdc_num_versions = sizeof (vdc_version) / sizeof (vdc_version[0]);
5091 
5092 	for (int i = 0; i < vdc_num_versions; i++) {
5093 		ASSERT(vdc_version[i].major > 0);
5094 		ASSERT((i == 0) ||
5095 		    (vdc_version[i].major < vdc_version[i-1].major));
5096 
5097 		/*
5098 		 * If the major versions match, adjust the minor version, if
5099 		 * necessary, down to the highest value supported by this
5100 		 * client. The server should support all minor versions lower
5101 		 * than the value it sent
5102 		 */
5103 		if (ver_msg->ver_major == vdc_version[i].major) {
5104 			if (ver_msg->ver_minor > vdc_version[i].minor) {
5105 				DMSGX(0,
5106 				    "Adjusting minor version from %u to %u",
5107 				    ver_msg->ver_minor, vdc_version[i].minor);
5108 				ver_msg->ver_minor = vdc_version[i].minor;
5109 			}
5110 			return (B_TRUE);
5111 		}
5112 
5113 		/*
5114 		 * If the message contains a higher major version number, set
5115 		 * the message's major/minor versions to the current values
5116 		 * and return false, so this message will get resent with
5117 		 * these values, and the server will potentially try again
5118 		 * with the same or a lower version
5119 		 */
5120 		if (ver_msg->ver_major > vdc_version[i].major) {
5121 			ver_msg->ver_major = vdc_version[i].major;
5122 			ver_msg->ver_minor = vdc_version[i].minor;
5123 			DMSGX(0, "Suggesting major/minor (0x%x/0x%x)\n",
5124 			    ver_msg->ver_major, ver_msg->ver_minor);
5125 
5126 			return (B_FALSE);
5127 		}
5128 
5129 		/*
5130 		 * Otherwise, the message's major version is less than the
5131 		 * current major version, so continue the loop to the next
5132 		 * (lower) supported version
5133 		 */
5134 	}
5135 
5136 	/*
5137 	 * No common version was found; "ground" the version pair in the
5138 	 * message to terminate negotiation
5139 	 */
5140 	ver_msg->ver_major = 0;
5141 	ver_msg->ver_minor = 0;
5142 
5143 	return (B_FALSE);
5144 }
5145 /* -------------------------------------------------------------------------- */
5146 
5147 /*
5148  * DKIO(7) support
5149  */
5150 
5151 typedef struct vdc_dk_arg {
5152 	struct dk_callback	dkc;
5153 	int			mode;
5154 	dev_t			dev;
5155 	vdc_t			*vdc;
5156 } vdc_dk_arg_t;
5157 
5158 /*
5159  * Function:
5160  * 	vdc_dkio_flush_cb()
5161  *
5162  * Description:
5163  *	This routine is a callback for DKIOCFLUSHWRITECACHE which can be called
5164  *	by kernel code.
5165  *
5166  * Arguments:
5167  *	arg	- a pointer to a vdc_dk_arg_t structure.
5168  */
5169 void
5170 vdc_dkio_flush_cb(void *arg)
5171 {
5172 	struct vdc_dk_arg	*dk_arg = (struct vdc_dk_arg *)arg;
5173 	struct dk_callback	*dkc = NULL;
5174 	vdc_t			*vdc = NULL;
5175 	int			rv;
5176 
5177 	if (dk_arg == NULL) {
5178 		cmn_err(CE_NOTE, "?[Unk] DKIOCFLUSHWRITECACHE arg is NULL\n");
5179 		return;
5180 	}
5181 	dkc = &dk_arg->dkc;
5182 	vdc = dk_arg->vdc;
5183 	ASSERT(vdc != NULL);
5184 
5185 	rv = vdc_do_sync_op(vdc, VD_OP_FLUSH, NULL, 0,
5186 	    VDCPART(dk_arg->dev), 0, CB_SYNC, 0, VIO_both_dir, B_TRUE);
5187 	if (rv != 0) {
5188 		DMSG(vdc, 0, "[%d] DKIOCFLUSHWRITECACHE failed %d : model %x\n",
5189 		    vdc->instance, rv,
5190 		    ddi_model_convert_from(dk_arg->mode & FMODELS));
5191 	}
5192 
5193 	/*
5194 	 * Trigger the call back to notify the caller the the ioctl call has
5195 	 * been completed.
5196 	 */
5197 	if ((dk_arg->mode & FKIOCTL) &&
5198 	    (dkc != NULL) &&
5199 	    (dkc->dkc_callback != NULL)) {
5200 		ASSERT(dkc->dkc_cookie != NULL);
5201 		(*dkc->dkc_callback)(dkc->dkc_cookie, rv);
5202 	}
5203 
5204 	/* Indicate that one less DKIO write flush is outstanding */
5205 	mutex_enter(&vdc->lock);
5206 	vdc->dkio_flush_pending--;
5207 	ASSERT(vdc->dkio_flush_pending >= 0);
5208 	mutex_exit(&vdc->lock);
5209 
5210 	/* free the mem that was allocated when the callback was dispatched */
5211 	kmem_free(arg, sizeof (vdc_dk_arg_t));
5212 }
5213 
5214 /*
5215  * Function:
5216  * 	vdc_dkio_gapart()
5217  *
5218  * Description:
5219  *	This function implements the DKIOCGAPART ioctl.
5220  *
5221  * Arguments:
5222  *	vdc	- soft state pointer
5223  *	arg	- a pointer to a dk_map[NDKMAP] or dk_map32[NDKMAP] structure
5224  *	flag	- ioctl flags
5225  */
5226 static int
5227 vdc_dkio_gapart(vdc_t *vdc, caddr_t arg, int flag)
5228 {
5229 	struct dk_geom *geom;
5230 	struct extvtoc *vtoc;
5231 	union {
5232 		struct dk_map map[NDKMAP];
5233 		struct dk_map32 map32[NDKMAP];
5234 	} data;
5235 	int i, rv, size;
5236 
5237 	mutex_enter(&vdc->lock);
5238 
5239 	if ((rv = vdc_validate_geometry(vdc)) != 0) {
5240 		mutex_exit(&vdc->lock);
5241 		return (rv);
5242 	}
5243 
5244 	if (vdc->vdisk_size > VD_OLDVTOC_LIMIT) {
5245 		mutex_exit(&vdc->lock);
5246 		return (EOVERFLOW);
5247 	}
5248 
5249 	vtoc = vdc->vtoc;
5250 	geom = vdc->geom;
5251 
5252 	if (ddi_model_convert_from(flag & FMODELS) == DDI_MODEL_ILP32) {
5253 
5254 		for (i = 0; i < vtoc->v_nparts; i++) {
5255 			data.map32[i].dkl_cylno = vtoc->v_part[i].p_start /
5256 			    (geom->dkg_nhead * geom->dkg_nsect);
5257 			data.map32[i].dkl_nblk = vtoc->v_part[i].p_size;
5258 		}
5259 		size = NDKMAP * sizeof (struct dk_map32);
5260 
5261 	} else {
5262 
5263 		for (i = 0; i < vtoc->v_nparts; i++) {
5264 			data.map[i].dkl_cylno = vtoc->v_part[i].p_start /
5265 			    (geom->dkg_nhead * geom->dkg_nsect);
5266 			data.map[i].dkl_nblk = vtoc->v_part[i].p_size;
5267 		}
5268 		size = NDKMAP * sizeof (struct dk_map);
5269 
5270 	}
5271 
5272 	mutex_exit(&vdc->lock);
5273 
5274 	if (ddi_copyout(&data, arg, size, flag) != 0)
5275 		return (EFAULT);
5276 
5277 	return (0);
5278 }
5279 
5280 /*
5281  * Function:
5282  * 	vdc_dkio_partition()
5283  *
5284  * Description:
5285  *	This function implements the DKIOCPARTITION ioctl.
5286  *
5287  * Arguments:
5288  *	vdc	- soft state pointer
5289  *	arg	- a pointer to a struct partition64 structure
5290  *	flag	- ioctl flags
5291  */
5292 static int
5293 vdc_dkio_partition(vdc_t *vdc, caddr_t arg, int flag)
5294 {
5295 	struct partition64 p64;
5296 	efi_gpt_t *gpt;
5297 	efi_gpe_t *gpe;
5298 	vd_efi_dev_t edev;
5299 	uint_t partno;
5300 	int rv;
5301 
5302 	if (ddi_copyin(arg, &p64, sizeof (struct partition64), flag)) {
5303 		return (EFAULT);
5304 	}
5305 
5306 	VDC_EFI_DEV_SET(edev, vdc, vd_process_efi_ioctl);
5307 
5308 	if ((rv = vd_efi_alloc_and_read(&edev, &gpt, &gpe)) != 0) {
5309 		return (rv);
5310 	}
5311 
5312 	partno = p64.p_partno;
5313 
5314 	if (partno >= gpt->efi_gpt_NumberOfPartitionEntries) {
5315 		vd_efi_free(&edev, gpt, gpe);
5316 		return (ESRCH);
5317 	}
5318 
5319 	bcopy(&gpe[partno].efi_gpe_PartitionTypeGUID, &p64.p_type,
5320 	    sizeof (struct uuid));
5321 	p64.p_start = gpe[partno].efi_gpe_StartingLBA;
5322 	p64.p_size = gpe[partno].efi_gpe_EndingLBA - p64.p_start + 1;
5323 
5324 	if (ddi_copyout(&p64, arg, sizeof (struct partition64), flag)) {
5325 		vd_efi_free(&edev, gpt, gpe);
5326 		return (EFAULT);
5327 	}
5328 
5329 	vd_efi_free(&edev, gpt, gpe);
5330 	return (0);
5331 }
5332 
5333 /*
5334  * Function:
5335  * 	vdc_dioctl_rwcmd()
5336  *
5337  * Description:
5338  *	This function implements the DIOCTL_RWCMD ioctl. This ioctl is used
5339  *	for DKC_DIRECT disks to read or write at an absolute disk offset.
5340  *
5341  * Arguments:
5342  *	dev	- device
5343  *	arg	- a pointer to a dadkio_rwcmd or dadkio_rwcmd32 structure
5344  *	flag	- ioctl flags
5345  */
5346 static int
5347 vdc_dioctl_rwcmd(vdc_t *vdc, caddr_t arg, int flag)
5348 {
5349 	struct dadkio_rwcmd32 rwcmd32;
5350 	struct dadkio_rwcmd rwcmd;
5351 	struct iovec aiov;
5352 	struct uio auio;
5353 	int rw, status;
5354 	struct buf *buf;
5355 
5356 	if (ddi_model_convert_from(flag & FMODELS) == DDI_MODEL_ILP32) {
5357 		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd32,
5358 		    sizeof (struct dadkio_rwcmd32), flag)) {
5359 			return (EFAULT);
5360 		}
5361 		rwcmd.cmd = rwcmd32.cmd;
5362 		rwcmd.flags = rwcmd32.flags;
5363 		rwcmd.blkaddr = (daddr_t)rwcmd32.blkaddr;
5364 		rwcmd.buflen = rwcmd32.buflen;
5365 		rwcmd.bufaddr = (caddr_t)(uintptr_t)rwcmd32.bufaddr;
5366 	} else {
5367 		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd,
5368 		    sizeof (struct dadkio_rwcmd), flag)) {
5369 			return (EFAULT);
5370 		}
5371 	}
5372 
5373 	switch (rwcmd.cmd) {
5374 	case DADKIO_RWCMD_READ:
5375 		rw = B_READ;
5376 		break;
5377 	case DADKIO_RWCMD_WRITE:
5378 		rw = B_WRITE;
5379 		break;
5380 	default:
5381 		return (EINVAL);
5382 	}
5383 
5384 	bzero((caddr_t)&aiov, sizeof (struct iovec));
5385 	aiov.iov_base   = rwcmd.bufaddr;
5386 	aiov.iov_len    = rwcmd.buflen;
5387 
5388 	bzero((caddr_t)&auio, sizeof (struct uio));
5389 	auio.uio_iov    = &aiov;
5390 	auio.uio_iovcnt = 1;
5391 	auio.uio_loffset = rwcmd.blkaddr * vdc->vdisk_bsize;
5392 	auio.uio_resid  = rwcmd.buflen;
5393 	auio.uio_segflg = flag & FKIOCTL ? UIO_SYSSPACE : UIO_USERSPACE;
5394 
5395 	buf = kmem_alloc(sizeof (buf_t), KM_SLEEP);
5396 	bioinit(buf);
5397 	/*
5398 	 * We use the private field of buf to specify that this is an
5399 	 * I/O using an absolute offset.
5400 	 */
5401 	buf->b_private = (void *)VD_SLICE_NONE;
5402 
5403 	status = physio(vdc_strategy, buf, VD_MAKE_DEV(vdc->instance, 0),
5404 	    rw, vdc_min, &auio);
5405 
5406 	biofini(buf);
5407 	kmem_free(buf, sizeof (buf_t));
5408 
5409 	return (status);
5410 }
5411 
5412 /*
5413  * Allocate a buffer for a VD_OP_SCSICMD operation. The size of the allocated
5414  * buffer is returned in alloc_len.
5415  */
5416 static vd_scsi_t *
5417 vdc_scsi_alloc(int cdb_len, int sense_len, int datain_len, int dataout_len,
5418     int *alloc_len)
5419 {
5420 	vd_scsi_t *vd_scsi;
5421 	int vd_scsi_len = VD_SCSI_SIZE;
5422 
5423 	vd_scsi_len += P2ROUNDUP(cdb_len, sizeof (uint64_t));
5424 	vd_scsi_len += P2ROUNDUP(sense_len, sizeof (uint64_t));
5425 	vd_scsi_len += P2ROUNDUP(datain_len, sizeof (uint64_t));
5426 	vd_scsi_len += P2ROUNDUP(dataout_len, sizeof (uint64_t));
5427 
5428 	ASSERT(vd_scsi_len % sizeof (uint64_t) == 0);
5429 
5430 	vd_scsi = kmem_zalloc(vd_scsi_len, KM_SLEEP);
5431 
5432 	vd_scsi->cdb_len = cdb_len;
5433 	vd_scsi->sense_len = sense_len;
5434 	vd_scsi->datain_len = datain_len;
5435 	vd_scsi->dataout_len = dataout_len;
5436 
5437 	*alloc_len = vd_scsi_len;
5438 
5439 	return (vd_scsi);
5440 }
5441 
5442 /*
5443  * Convert the status of a SCSI command to a Solaris return code.
5444  *
5445  * Arguments:
5446  *	vd_scsi		- The SCSI operation buffer.
5447  *	log_error	- indicate if an error message should be logged.
5448  *
5449  * Note that our SCSI error messages are rather primitive for the moment
5450  * and could be improved by decoding some data like the SCSI command and
5451  * the sense key.
5452  *
5453  * Return value:
5454  *	0		- Status is good.
5455  *	EACCES		- Status reports a reservation conflict.
5456  *	ENOTSUP		- Status reports a check condition and sense key
5457  *			  reports an illegal request.
5458  *	EIO		- Any other status.
5459  */
5460 static int
5461 vdc_scsi_status(vdc_t *vdc, vd_scsi_t *vd_scsi, boolean_t log_error)
5462 {
5463 	int rv;
5464 	char path_str[MAXPATHLEN];
5465 	char panic_str[VDC_RESV_CONFLICT_FMT_LEN + MAXPATHLEN];
5466 	union scsi_cdb *cdb;
5467 	struct scsi_extended_sense *sense;
5468 
5469 	if (vd_scsi->cmd_status == STATUS_GOOD)
5470 		/* no error */
5471 		return (0);
5472 
5473 	/* when the tunable vdc_scsi_log_error is true we log all errors */
5474 	if (vdc_scsi_log_error)
5475 		log_error = B_TRUE;
5476 
5477 	if (log_error) {
5478 		cmn_err(CE_WARN, "%s (vdc%d):\tError for Command: 0x%x)\n",
5479 		    ddi_pathname(vdc->dip, path_str), vdc->instance,
5480 		    GETCMD(VD_SCSI_DATA_CDB(vd_scsi)));
5481 	}
5482 
5483 	/* default returned value */
5484 	rv = EIO;
5485 
5486 	switch (vd_scsi->cmd_status) {
5487 
5488 	case STATUS_CHECK:
5489 	case STATUS_TERMINATED:
5490 		if (log_error)
5491 			cmn_err(CE_CONT, "\tCheck Condition Error\n");
5492 
5493 		/* check sense buffer */
5494 		if (vd_scsi->sense_len == 0 ||
5495 		    vd_scsi->sense_status != STATUS_GOOD) {
5496 			if (log_error)
5497 				cmn_err(CE_CONT, "\tNo Sense Data Available\n");
5498 			break;
5499 		}
5500 
5501 		sense = VD_SCSI_DATA_SENSE(vd_scsi);
5502 
5503 		if (log_error) {
5504 			cmn_err(CE_CONT, "\tSense Key:  0x%x\n"
5505 			    "\tASC: 0x%x, ASCQ: 0x%x\n",
5506 			    scsi_sense_key((uint8_t *)sense),
5507 			    scsi_sense_asc((uint8_t *)sense),
5508 			    scsi_sense_ascq((uint8_t *)sense));
5509 		}
5510 
5511 		if (scsi_sense_key((uint8_t *)sense) == KEY_ILLEGAL_REQUEST)
5512 			rv = ENOTSUP;
5513 		break;
5514 
5515 	case STATUS_BUSY:
5516 		if (log_error)
5517 			cmn_err(CE_NOTE, "\tDevice Busy\n");
5518 		break;
5519 
5520 	case STATUS_RESERVATION_CONFLICT:
5521 		/*
5522 		 * If the command was PERSISTENT_RESERVATION_[IN|OUT] then
5523 		 * reservation conflict could be due to various reasons like
5524 		 * incorrect keys, not registered or not reserved etc. So,
5525 		 * we should not panic in that case.
5526 		 */
5527 		cdb = VD_SCSI_DATA_CDB(vd_scsi);
5528 		if (vdc->failfast_interval != 0 &&
5529 		    cdb->scc_cmd != SCMD_PERSISTENT_RESERVE_IN &&
5530 		    cdb->scc_cmd != SCMD_PERSISTENT_RESERVE_OUT) {
5531 			/* failfast is enabled so we have to panic */
5532 			(void) snprintf(panic_str, sizeof (panic_str),
5533 			    VDC_RESV_CONFLICT_FMT_STR "%s",
5534 			    ddi_pathname(vdc->dip, path_str));
5535 			panic(panic_str);
5536 		}
5537 		if (log_error)
5538 			cmn_err(CE_NOTE, "\tReservation Conflict\n");
5539 		rv = EACCES;
5540 		break;
5541 
5542 	case STATUS_QFULL:
5543 		if (log_error)
5544 			cmn_err(CE_NOTE, "\tQueue Full\n");
5545 		break;
5546 
5547 	case STATUS_MET:
5548 	case STATUS_INTERMEDIATE:
5549 	case STATUS_SCSI2:
5550 	case STATUS_INTERMEDIATE_MET:
5551 	case STATUS_ACA_ACTIVE:
5552 		if (log_error)
5553 			cmn_err(CE_CONT,
5554 			    "\tUnexpected SCSI status received: 0x%x\n",
5555 			    vd_scsi->cmd_status);
5556 		break;
5557 
5558 	default:
5559 		if (log_error)
5560 			cmn_err(CE_CONT,
5561 			    "\tInvalid SCSI status received: 0x%x\n",
5562 			    vd_scsi->cmd_status);
5563 		break;
5564 	}
5565 
5566 	return (rv);
5567 }
5568 
5569 /*
5570  * Implemented the USCSICMD uscsi(7I) ioctl. This ioctl is converted to
5571  * a VD_OP_SCSICMD operation which is sent to the vdisk server. If a SCSI
5572  * reset is requested (i.e. a flag USCSI_RESET* is set) then the ioctl is
5573  * converted to a VD_OP_RESET operation.
5574  */
5575 static int
5576 vdc_uscsi_cmd(vdc_t *vdc, caddr_t arg, int mode)
5577 {
5578 	struct uscsi_cmd 	uscsi;
5579 	struct uscsi_cmd32	uscsi32;
5580 	vd_scsi_t 		*vd_scsi;
5581 	int 			vd_scsi_len;
5582 	union scsi_cdb		*cdb;
5583 	struct scsi_extended_sense *sense;
5584 	char 			*datain, *dataout;
5585 	size_t			cdb_len, datain_len, dataout_len, sense_len;
5586 	int 			rv;
5587 
5588 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5589 		if (ddi_copyin(arg, &uscsi32, sizeof (struct uscsi_cmd32),
5590 		    mode) != 0)
5591 			return (EFAULT);
5592 		uscsi_cmd32touscsi_cmd((&uscsi32), (&uscsi));
5593 	} else {
5594 		if (ddi_copyin(arg, &uscsi, sizeof (struct uscsi_cmd),
5595 		    mode) != 0)
5596 			return (EFAULT);
5597 	}
5598 
5599 	/* a uscsi reset is converted to a VD_OP_RESET operation */
5600 	if (uscsi.uscsi_flags & (USCSI_RESET | USCSI_RESET_LUN |
5601 	    USCSI_RESET_ALL)) {
5602 		rv = vdc_do_sync_op(vdc, VD_OP_RESET, NULL, 0, 0, 0, CB_SYNC,
5603 		    (void *)(uint64_t)mode, VIO_both_dir, B_TRUE);
5604 		return (rv);
5605 	}
5606 
5607 	/* cdb buffer length */
5608 	cdb_len = uscsi.uscsi_cdblen;
5609 
5610 	/* data in and out buffers length */
5611 	if (uscsi.uscsi_flags & USCSI_READ) {
5612 		datain_len = uscsi.uscsi_buflen;
5613 		dataout_len = 0;
5614 	} else {
5615 		datain_len = 0;
5616 		dataout_len = uscsi.uscsi_buflen;
5617 	}
5618 
5619 	/* sense buffer length */
5620 	if (uscsi.uscsi_flags & USCSI_RQENABLE)
5621 		sense_len = uscsi.uscsi_rqlen;
5622 	else
5623 		sense_len = 0;
5624 
5625 	/* allocate buffer for the VD_SCSICMD_OP operation */
5626 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, datain_len, dataout_len,
5627 	    &vd_scsi_len);
5628 
5629 	/*
5630 	 * The documentation of USCSI_ISOLATE and USCSI_DIAGNOSE is very vague,
5631 	 * but basically they prevent a SCSI command from being retried in case
5632 	 * of an error.
5633 	 */
5634 	if ((uscsi.uscsi_flags & USCSI_ISOLATE) ||
5635 	    (uscsi.uscsi_flags & USCSI_DIAGNOSE))
5636 		vd_scsi->options |= VD_SCSI_OPT_NORETRY;
5637 
5638 	/* set task attribute */
5639 	if (uscsi.uscsi_flags & USCSI_NOTAG) {
5640 		vd_scsi->task_attribute = 0;
5641 	} else {
5642 		if (uscsi.uscsi_flags & USCSI_HEAD)
5643 			vd_scsi->task_attribute = VD_SCSI_TASK_ACA;
5644 		else if (uscsi.uscsi_flags & USCSI_HTAG)
5645 			vd_scsi->task_attribute = VD_SCSI_TASK_HQUEUE;
5646 		else if (uscsi.uscsi_flags & USCSI_OTAG)
5647 			vd_scsi->task_attribute = VD_SCSI_TASK_ORDERED;
5648 		else
5649 			vd_scsi->task_attribute = 0;
5650 	}
5651 
5652 	/* set timeout */
5653 	vd_scsi->timeout = uscsi.uscsi_timeout;
5654 
5655 	/* copy-in cdb data */
5656 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5657 	if (ddi_copyin(uscsi.uscsi_cdb, cdb, cdb_len, mode) != 0) {
5658 		rv = EFAULT;
5659 		goto done;
5660 	}
5661 
5662 	/* keep a pointer to the sense buffer */
5663 	sense = VD_SCSI_DATA_SENSE(vd_scsi);
5664 
5665 	/* keep a pointer to the data-in buffer */
5666 	datain = (char *)VD_SCSI_DATA_IN(vd_scsi);
5667 
5668 	/* copy-in request data to the data-out buffer */
5669 	dataout = (char *)VD_SCSI_DATA_OUT(vd_scsi);
5670 	if (!(uscsi.uscsi_flags & USCSI_READ)) {
5671 		if (ddi_copyin(uscsi.uscsi_bufaddr, dataout, dataout_len,
5672 		    mode)) {
5673 			rv = EFAULT;
5674 			goto done;
5675 		}
5676 	}
5677 
5678 	/* submit the request */
5679 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5680 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5681 
5682 	if (rv != 0)
5683 		goto done;
5684 
5685 	/* update scsi status */
5686 	uscsi.uscsi_status = vd_scsi->cmd_status;
5687 
5688 	/* update sense data */
5689 	if ((uscsi.uscsi_flags & USCSI_RQENABLE) &&
5690 	    (uscsi.uscsi_status == STATUS_CHECK ||
5691 	    uscsi.uscsi_status == STATUS_TERMINATED)) {
5692 
5693 		uscsi.uscsi_rqstatus = vd_scsi->sense_status;
5694 
5695 		if (uscsi.uscsi_rqstatus == STATUS_GOOD) {
5696 			uscsi.uscsi_rqresid = uscsi.uscsi_rqlen -
5697 			    vd_scsi->sense_len;
5698 			if (ddi_copyout(sense, uscsi.uscsi_rqbuf,
5699 			    vd_scsi->sense_len, mode) != 0) {
5700 				rv = EFAULT;
5701 				goto done;
5702 			}
5703 		}
5704 	}
5705 
5706 	/* update request data */
5707 	if (uscsi.uscsi_status == STATUS_GOOD) {
5708 		if (uscsi.uscsi_flags & USCSI_READ) {
5709 			uscsi.uscsi_resid = uscsi.uscsi_buflen -
5710 			    vd_scsi->datain_len;
5711 			if (ddi_copyout(datain, uscsi.uscsi_bufaddr,
5712 			    vd_scsi->datain_len, mode) != 0) {
5713 				rv = EFAULT;
5714 				goto done;
5715 			}
5716 		} else {
5717 			uscsi.uscsi_resid = uscsi.uscsi_buflen -
5718 			    vd_scsi->dataout_len;
5719 		}
5720 	}
5721 
5722 	/* copy-out result */
5723 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5724 		uscsi_cmdtouscsi_cmd32((&uscsi), (&uscsi32));
5725 		if (ddi_copyout(&uscsi32, arg, sizeof (struct uscsi_cmd32),
5726 		    mode) != 0) {
5727 			rv = EFAULT;
5728 			goto done;
5729 		}
5730 	} else {
5731 		if (ddi_copyout(&uscsi, arg, sizeof (struct uscsi_cmd),
5732 		    mode) != 0) {
5733 			rv = EFAULT;
5734 			goto done;
5735 		}
5736 	}
5737 
5738 	/* get the return code from the SCSI command status */
5739 	rv = vdc_scsi_status(vdc, vd_scsi,
5740 	    !(uscsi.uscsi_flags & USCSI_SILENT));
5741 
5742 done:
5743 	kmem_free(vd_scsi, vd_scsi_len);
5744 	return (rv);
5745 }
5746 
5747 /*
5748  * Create a VD_OP_SCSICMD buffer for a SCSI PERSISTENT IN command.
5749  *
5750  * Arguments:
5751  *	cmd		- SCSI PERSISTENT IN command
5752  *	len		- length of the SCSI input buffer
5753  *	vd_scsi_len	- return the length of the allocated buffer
5754  *
5755  * Returned Value:
5756  *	a pointer to the allocated VD_OP_SCSICMD buffer.
5757  */
5758 static vd_scsi_t *
5759 vdc_scsi_alloc_persistent_in(uchar_t cmd, int len, int *vd_scsi_len)
5760 {
5761 	int cdb_len, sense_len, datain_len, dataout_len;
5762 	vd_scsi_t *vd_scsi;
5763 	union scsi_cdb *cdb;
5764 
5765 	cdb_len = CDB_GROUP1;
5766 	sense_len = sizeof (struct scsi_extended_sense);
5767 	datain_len = len;
5768 	dataout_len = 0;
5769 
5770 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, datain_len, dataout_len,
5771 	    vd_scsi_len);
5772 
5773 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5774 
5775 	/* set cdb */
5776 	cdb->scc_cmd = SCMD_PERSISTENT_RESERVE_IN;
5777 	cdb->cdb_opaque[1] = cmd;
5778 	FORMG1COUNT(cdb, datain_len);
5779 
5780 	vd_scsi->timeout = vdc_scsi_timeout;
5781 
5782 	return (vd_scsi);
5783 }
5784 
5785 /*
5786  * Create a VD_OP_SCSICMD buffer for a SCSI PERSISTENT OUT command.
5787  *
5788  * Arguments:
5789  *	cmd		- SCSI PERSISTENT OUT command
5790  *	len		- length of the SCSI output buffer
5791  *	vd_scsi_len	- return the length of the allocated buffer
5792  *
5793  * Returned Code:
5794  *	a pointer to the allocated VD_OP_SCSICMD buffer.
5795  */
5796 static vd_scsi_t *
5797 vdc_scsi_alloc_persistent_out(uchar_t cmd, int len, int *vd_scsi_len)
5798 {
5799 	int cdb_len, sense_len, datain_len, dataout_len;
5800 	vd_scsi_t *vd_scsi;
5801 	union scsi_cdb *cdb;
5802 
5803 	cdb_len = CDB_GROUP1;
5804 	sense_len = sizeof (struct scsi_extended_sense);
5805 	datain_len = 0;
5806 	dataout_len = len;
5807 
5808 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, datain_len, dataout_len,
5809 	    vd_scsi_len);
5810 
5811 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5812 
5813 	/* set cdb */
5814 	cdb->scc_cmd = SCMD_PERSISTENT_RESERVE_OUT;
5815 	cdb->cdb_opaque[1] = cmd;
5816 	FORMG1COUNT(cdb, dataout_len);
5817 
5818 	vd_scsi->timeout = vdc_scsi_timeout;
5819 
5820 	return (vd_scsi);
5821 }
5822 
5823 /*
5824  * Implement the MHIOCGRP_INKEYS mhd(7i) ioctl. The ioctl is converted
5825  * to a SCSI PERSISTENT IN READ KEYS command which is sent to the vdisk
5826  * server with a VD_OP_SCSICMD operation.
5827  */
5828 static int
5829 vdc_mhd_inkeys(vdc_t *vdc, caddr_t arg, int mode)
5830 {
5831 	vd_scsi_t *vd_scsi;
5832 	mhioc_inkeys_t inkeys;
5833 	mhioc_key_list_t klist;
5834 	struct mhioc_inkeys32 inkeys32;
5835 	struct mhioc_key_list32 klist32;
5836 	sd_prin_readkeys_t *scsi_keys;
5837 	void *user_keys;
5838 	int vd_scsi_len;
5839 	int listsize, listlen, rv;
5840 
5841 	/* copyin arguments */
5842 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5843 		rv = ddi_copyin(arg, &inkeys32, sizeof (inkeys32), mode);
5844 		if (rv != 0)
5845 			return (EFAULT);
5846 
5847 		rv = ddi_copyin((caddr_t)(uintptr_t)inkeys32.li, &klist32,
5848 		    sizeof (klist32), mode);
5849 		if (rv != 0)
5850 			return (EFAULT);
5851 
5852 		listsize = klist32.listsize;
5853 	} else {
5854 		rv = ddi_copyin(arg, &inkeys, sizeof (inkeys), mode);
5855 		if (rv != 0)
5856 			return (EFAULT);
5857 
5858 		rv = ddi_copyin(inkeys.li, &klist, sizeof (klist), mode);
5859 		if (rv != 0)
5860 			return (EFAULT);
5861 
5862 		listsize = klist.listsize;
5863 	}
5864 
5865 	/* build SCSI VD_OP request */
5866 	vd_scsi = vdc_scsi_alloc_persistent_in(SD_READ_KEYS,
5867 	    sizeof (sd_prin_readkeys_t) - sizeof (caddr_t) +
5868 	    (sizeof (mhioc_resv_key_t) * listsize), &vd_scsi_len);
5869 
5870 	scsi_keys = (sd_prin_readkeys_t *)VD_SCSI_DATA_IN(vd_scsi);
5871 
5872 	/* submit the request */
5873 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5874 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5875 
5876 	if (rv != 0)
5877 		goto done;
5878 
5879 	listlen = scsi_keys->len / MHIOC_RESV_KEY_SIZE;
5880 
5881 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5882 		inkeys32.generation = scsi_keys->generation;
5883 		rv = ddi_copyout(&inkeys32, arg, sizeof (inkeys32), mode);
5884 		if (rv != 0) {
5885 			rv = EFAULT;
5886 			goto done;
5887 		}
5888 
5889 		klist32.listlen = listlen;
5890 		rv = ddi_copyout(&klist32, (caddr_t)(uintptr_t)inkeys32.li,
5891 		    sizeof (klist32), mode);
5892 		if (rv != 0) {
5893 			rv = EFAULT;
5894 			goto done;
5895 		}
5896 
5897 		user_keys = (caddr_t)(uintptr_t)klist32.list;
5898 	} else {
5899 		inkeys.generation = scsi_keys->generation;
5900 		rv = ddi_copyout(&inkeys, arg, sizeof (inkeys), mode);
5901 		if (rv != 0) {
5902 			rv = EFAULT;
5903 			goto done;
5904 		}
5905 
5906 		klist.listlen = listlen;
5907 		rv = ddi_copyout(&klist, inkeys.li, sizeof (klist), mode);
5908 		if (rv != 0) {
5909 			rv = EFAULT;
5910 			goto done;
5911 		}
5912 
5913 		user_keys = klist.list;
5914 	}
5915 
5916 	/* copy out keys */
5917 	if (listlen > 0 && listsize > 0) {
5918 		if (listsize < listlen)
5919 			listlen = listsize;
5920 		rv = ddi_copyout(&scsi_keys->keylist, user_keys,
5921 		    listlen * MHIOC_RESV_KEY_SIZE, mode);
5922 		if (rv != 0)
5923 			rv = EFAULT;
5924 	}
5925 
5926 	if (rv == 0)
5927 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
5928 
5929 done:
5930 	kmem_free(vd_scsi, vd_scsi_len);
5931 
5932 	return (rv);
5933 }
5934 
5935 /*
5936  * Implement the MHIOCGRP_INRESV mhd(7i) ioctl. The ioctl is converted
5937  * to a SCSI PERSISTENT IN READ RESERVATION command which is sent to
5938  * the vdisk server with a VD_OP_SCSICMD operation.
5939  */
5940 static int
5941 vdc_mhd_inresv(vdc_t *vdc, caddr_t arg, int mode)
5942 {
5943 	vd_scsi_t *vd_scsi;
5944 	mhioc_inresvs_t inresv;
5945 	mhioc_resv_desc_list_t rlist;
5946 	struct mhioc_inresvs32 inresv32;
5947 	struct mhioc_resv_desc_list32 rlist32;
5948 	mhioc_resv_desc_t mhd_resv;
5949 	sd_prin_readresv_t *scsi_resv;
5950 	sd_readresv_desc_t *resv;
5951 	mhioc_resv_desc_t *user_resv;
5952 	int vd_scsi_len;
5953 	int listsize, listlen, i, rv;
5954 
5955 	/* copyin arguments */
5956 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5957 		rv = ddi_copyin(arg, &inresv32, sizeof (inresv32), mode);
5958 		if (rv != 0)
5959 			return (EFAULT);
5960 
5961 		rv = ddi_copyin((caddr_t)(uintptr_t)inresv32.li, &rlist32,
5962 		    sizeof (rlist32), mode);
5963 		if (rv != 0)
5964 			return (EFAULT);
5965 
5966 		listsize = rlist32.listsize;
5967 	} else {
5968 		rv = ddi_copyin(arg, &inresv, sizeof (inresv), mode);
5969 		if (rv != 0)
5970 			return (EFAULT);
5971 
5972 		rv = ddi_copyin(inresv.li, &rlist, sizeof (rlist), mode);
5973 		if (rv != 0)
5974 			return (EFAULT);
5975 
5976 		listsize = rlist.listsize;
5977 	}
5978 
5979 	/* build SCSI VD_OP request */
5980 	vd_scsi = vdc_scsi_alloc_persistent_in(SD_READ_RESV,
5981 	    sizeof (sd_prin_readresv_t) - sizeof (caddr_t) +
5982 	    (SCSI3_RESV_DESC_LEN * listsize), &vd_scsi_len);
5983 
5984 	scsi_resv = (sd_prin_readresv_t *)VD_SCSI_DATA_IN(vd_scsi);
5985 
5986 	/* submit the request */
5987 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5988 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5989 
5990 	if (rv != 0)
5991 		goto done;
5992 
5993 	listlen = scsi_resv->len / SCSI3_RESV_DESC_LEN;
5994 
5995 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5996 		inresv32.generation = scsi_resv->generation;
5997 		rv = ddi_copyout(&inresv32, arg, sizeof (inresv32), mode);
5998 		if (rv != 0) {
5999 			rv = EFAULT;
6000 			goto done;
6001 		}
6002 
6003 		rlist32.listlen = listlen;
6004 		rv = ddi_copyout(&rlist32, (caddr_t)(uintptr_t)inresv32.li,
6005 		    sizeof (rlist32), mode);
6006 		if (rv != 0) {
6007 			rv = EFAULT;
6008 			goto done;
6009 		}
6010 
6011 		user_resv = (mhioc_resv_desc_t *)(uintptr_t)rlist32.list;
6012 	} else {
6013 		inresv.generation = scsi_resv->generation;
6014 		rv = ddi_copyout(&inresv, arg, sizeof (inresv), mode);
6015 		if (rv != 0) {
6016 			rv = EFAULT;
6017 			goto done;
6018 		}
6019 
6020 		rlist.listlen = listlen;
6021 		rv = ddi_copyout(&rlist, inresv.li, sizeof (rlist), mode);
6022 		if (rv != 0) {
6023 			rv = EFAULT;
6024 			goto done;
6025 		}
6026 
6027 		user_resv = rlist.list;
6028 	}
6029 
6030 	/* copy out reservations */
6031 	if (listsize > 0 && listlen > 0) {
6032 		if (listsize < listlen)
6033 			listlen = listsize;
6034 		resv = (sd_readresv_desc_t *)&scsi_resv->readresv_desc;
6035 
6036 		for (i = 0; i < listlen; i++) {
6037 			mhd_resv.type = resv->type;
6038 			mhd_resv.scope = resv->scope;
6039 			mhd_resv.scope_specific_addr =
6040 			    BE_32(resv->scope_specific_addr);
6041 			bcopy(&resv->resvkey, &mhd_resv.key,
6042 			    MHIOC_RESV_KEY_SIZE);
6043 
6044 			rv = ddi_copyout(&mhd_resv, user_resv,
6045 			    sizeof (mhd_resv), mode);
6046 			if (rv != 0) {
6047 				rv = EFAULT;
6048 				goto done;
6049 			}
6050 			resv++;
6051 			user_resv++;
6052 		}
6053 	}
6054 
6055 	if (rv == 0)
6056 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6057 
6058 done:
6059 	kmem_free(vd_scsi, vd_scsi_len);
6060 	return (rv);
6061 }
6062 
6063 /*
6064  * Implement the MHIOCGRP_REGISTER mhd(7i) ioctl. The ioctl is converted
6065  * to a SCSI PERSISTENT OUT REGISTER command which is sent to the vdisk
6066  * server with a VD_OP_SCSICMD operation.
6067  */
6068 static int
6069 vdc_mhd_register(vdc_t *vdc, caddr_t arg, int mode)
6070 {
6071 	vd_scsi_t *vd_scsi;
6072 	sd_prout_t *scsi_prout;
6073 	mhioc_register_t mhd_reg;
6074 	int vd_scsi_len, rv;
6075 
6076 	/* copyin arguments */
6077 	rv = ddi_copyin(arg, &mhd_reg, sizeof (mhd_reg), mode);
6078 	if (rv != 0)
6079 		return (EFAULT);
6080 
6081 	/* build SCSI VD_OP request */
6082 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_REGISTER,
6083 	    sizeof (sd_prout_t), &vd_scsi_len);
6084 
6085 	/* set parameters */
6086 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
6087 	bcopy(mhd_reg.oldkey.key, scsi_prout->res_key, MHIOC_RESV_KEY_SIZE);
6088 	bcopy(mhd_reg.newkey.key, scsi_prout->service_key, MHIOC_RESV_KEY_SIZE);
6089 	scsi_prout->aptpl = (uchar_t)mhd_reg.aptpl;
6090 
6091 	/* submit the request */
6092 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6093 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
6094 
6095 	if (rv == 0)
6096 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6097 
6098 	kmem_free(vd_scsi, vd_scsi_len);
6099 	return (rv);
6100 }
6101 
6102 /*
6103  * Implement the MHIOCGRP_RESERVE mhd(7i) ioctl. The ioctl is converted
6104  * to a SCSI PERSISTENT OUT RESERVE command which is sent to the vdisk
6105  * server with a VD_OP_SCSICMD operation.
6106  */
6107 static int
6108 vdc_mhd_reserve(vdc_t *vdc, caddr_t arg, int mode)
6109 {
6110 	union scsi_cdb *cdb;
6111 	vd_scsi_t *vd_scsi;
6112 	sd_prout_t *scsi_prout;
6113 	mhioc_resv_desc_t mhd_resv;
6114 	int vd_scsi_len, rv;
6115 
6116 	/* copyin arguments */
6117 	rv = ddi_copyin(arg, &mhd_resv, sizeof (mhd_resv), mode);
6118 	if (rv != 0)
6119 		return (EFAULT);
6120 
6121 	/* build SCSI VD_OP request */
6122 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_RESERVE,
6123 	    sizeof (sd_prout_t), &vd_scsi_len);
6124 
6125 	/* set parameters */
6126 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
6127 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
6128 	bcopy(mhd_resv.key.key, scsi_prout->res_key, MHIOC_RESV_KEY_SIZE);
6129 	scsi_prout->scope_address = mhd_resv.scope_specific_addr;
6130 	cdb->cdb_opaque[2] = mhd_resv.type;
6131 
6132 	/* submit the request */
6133 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6134 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
6135 
6136 	if (rv == 0)
6137 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6138 
6139 	kmem_free(vd_scsi, vd_scsi_len);
6140 	return (rv);
6141 }
6142 
6143 /*
6144  * Implement the MHIOCGRP_PREEMPTANDABORT mhd(7i) ioctl. The ioctl is
6145  * converted to a SCSI PERSISTENT OUT PREEMPT AND ABORT command which
6146  * is sent to the vdisk server with a VD_OP_SCSICMD operation.
6147  */
6148 static int
6149 vdc_mhd_preemptabort(vdc_t *vdc, caddr_t arg, int mode)
6150 {
6151 	union scsi_cdb *cdb;
6152 	vd_scsi_t *vd_scsi;
6153 	sd_prout_t *scsi_prout;
6154 	mhioc_preemptandabort_t mhd_preempt;
6155 	int vd_scsi_len, rv;
6156 
6157 	/* copyin arguments */
6158 	rv = ddi_copyin(arg, &mhd_preempt, sizeof (mhd_preempt), mode);
6159 	if (rv != 0)
6160 		return (EFAULT);
6161 
6162 	/* build SCSI VD_OP request */
6163 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_PREEMPTANDABORT,
6164 	    sizeof (sd_prout_t), &vd_scsi_len);
6165 
6166 	/* set parameters */
6167 	vd_scsi->task_attribute = VD_SCSI_TASK_ACA;
6168 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
6169 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
6170 	bcopy(mhd_preempt.resvdesc.key.key, scsi_prout->res_key,
6171 	    MHIOC_RESV_KEY_SIZE);
6172 	bcopy(mhd_preempt.victim_key.key, scsi_prout->service_key,
6173 	    MHIOC_RESV_KEY_SIZE);
6174 	scsi_prout->scope_address = mhd_preempt.resvdesc.scope_specific_addr;
6175 	cdb->cdb_opaque[2] = mhd_preempt.resvdesc.type;
6176 
6177 	/* submit the request */
6178 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6179 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
6180 
6181 	if (rv == 0)
6182 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6183 
6184 	kmem_free(vd_scsi, vd_scsi_len);
6185 	return (rv);
6186 }
6187 
6188 /*
6189  * Implement the MHIOCGRP_REGISTERANDIGNOREKEY mhd(7i) ioctl. The ioctl
6190  * is converted to a SCSI PERSISTENT OUT REGISTER AND IGNORE EXISTING KEY
6191  * command which is sent to the vdisk server with a VD_OP_SCSICMD operation.
6192  */
6193 static int
6194 vdc_mhd_registerignore(vdc_t *vdc, caddr_t arg, int mode)
6195 {
6196 	vd_scsi_t *vd_scsi;
6197 	sd_prout_t *scsi_prout;
6198 	mhioc_registerandignorekey_t mhd_regi;
6199 	int vd_scsi_len, rv;
6200 
6201 	/* copyin arguments */
6202 	rv = ddi_copyin(arg, &mhd_regi, sizeof (mhd_regi), mode);
6203 	if (rv != 0)
6204 		return (EFAULT);
6205 
6206 	/* build SCSI VD_OP request */
6207 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_REGISTERANDIGNOREKEY,
6208 	    sizeof (sd_prout_t), &vd_scsi_len);
6209 
6210 	/* set parameters */
6211 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
6212 	bcopy(mhd_regi.newkey.key, scsi_prout->service_key,
6213 	    MHIOC_RESV_KEY_SIZE);
6214 	scsi_prout->aptpl = (uchar_t)mhd_regi.aptpl;
6215 
6216 	/* submit the request */
6217 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6218 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
6219 
6220 	if (rv == 0)
6221 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6222 
6223 	kmem_free(vd_scsi, vd_scsi_len);
6224 	return (rv);
6225 }
6226 
6227 /*
6228  * This function is used by the failfast mechanism to send a SCSI command
6229  * to check for reservation conflict.
6230  */
6231 static int
6232 vdc_failfast_scsi_cmd(vdc_t *vdc, uchar_t scmd)
6233 {
6234 	int cdb_len, sense_len, vd_scsi_len;
6235 	vd_scsi_t *vd_scsi;
6236 	union scsi_cdb *cdb;
6237 	int rv;
6238 
6239 	ASSERT(scmd == SCMD_TEST_UNIT_READY || scmd == SCMD_WRITE_G1);
6240 
6241 	if (scmd == SCMD_WRITE_G1)
6242 		cdb_len = CDB_GROUP1;
6243 	else
6244 		cdb_len = CDB_GROUP0;
6245 
6246 	sense_len = sizeof (struct scsi_extended_sense);
6247 
6248 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, 0, 0, &vd_scsi_len);
6249 
6250 	/* set cdb */
6251 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
6252 	cdb->scc_cmd = scmd;
6253 
6254 	vd_scsi->timeout = vdc_scsi_timeout;
6255 
6256 	/*
6257 	 * Submit the request. The last argument has to be B_FALSE so that
6258 	 * vdc_do_sync_op does not loop checking for reservation conflict if
6259 	 * the operation returns an error.
6260 	 */
6261 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6262 	    0, 0, CB_SYNC, (void *)(uint64_t)FKIOCTL, VIO_both_dir, B_FALSE);
6263 
6264 	if (rv == 0)
6265 		(void) vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6266 
6267 	kmem_free(vd_scsi, vd_scsi_len);
6268 	return (rv);
6269 }
6270 
6271 /*
6272  * This function is used by the failfast mechanism to check for reservation
6273  * conflict. It sends some SCSI commands which will fail with a reservation
6274  * conflict error if the system does not have access to the disk and this
6275  * will panic the system.
6276  *
6277  * Returned Code:
6278  *	0	- disk is accessible without reservation conflict error
6279  *	!= 0	- unable to check if disk is accessible
6280  */
6281 int
6282 vdc_failfast_check_resv(vdc_t *vdc)
6283 {
6284 	int failure = 0;
6285 
6286 	/*
6287 	 * Send a TEST UNIT READY command. The command will panic
6288 	 * the system if it fails with a reservation conflict.
6289 	 */
6290 	if (vdc_failfast_scsi_cmd(vdc, SCMD_TEST_UNIT_READY) != 0)
6291 		failure++;
6292 
6293 	/*
6294 	 * With SPC-3 compliant devices TEST UNIT READY will succeed on
6295 	 * a reserved device, so we also do a WRITE(10) of zero byte in
6296 	 * order to provoke a Reservation Conflict status on those newer
6297 	 * devices.
6298 	 */
6299 	if (vdc_failfast_scsi_cmd(vdc, SCMD_WRITE_G1) != 0)
6300 		failure++;
6301 
6302 	return (failure);
6303 }
6304 
6305 /*
6306  * Add a pending I/O to the failfast I/O queue. An I/O is added to this
6307  * queue when it has failed and failfast is enabled. Then we have to check
6308  * if it has failed because of a reservation conflict in which case we have
6309  * to panic the system.
6310  *
6311  * Async I/O should be queued with their block I/O data transfer structure
6312  * (buf). Sync I/O should be queued with buf = NULL.
6313  */
6314 static vdc_io_t *
6315 vdc_failfast_io_queue(vdc_t *vdc, struct buf *buf)
6316 {
6317 	vdc_io_t *vio;
6318 
6319 	ASSERT(MUTEX_HELD(&vdc->lock));
6320 
6321 	vio = kmem_alloc(sizeof (vdc_io_t), KM_SLEEP);
6322 	vio->vio_next = vdc->failfast_io_queue;
6323 	vio->vio_buf = buf;
6324 	vio->vio_qtime = ddi_get_lbolt();
6325 
6326 	vdc->failfast_io_queue = vio;
6327 
6328 	/* notify the failfast thread that a new I/O is queued */
6329 	cv_signal(&vdc->failfast_cv);
6330 
6331 	return (vio);
6332 }
6333 
6334 /*
6335  * Remove and complete I/O in the failfast I/O queue which have been
6336  * added after the indicated deadline. A deadline of 0 means that all
6337  * I/O have to be unqueued and marked as completed.
6338  */
6339 static void
6340 vdc_failfast_io_unqueue(vdc_t *vdc, clock_t deadline)
6341 {
6342 	vdc_io_t *vio, *vio_tmp;
6343 
6344 	ASSERT(MUTEX_HELD(&vdc->lock));
6345 
6346 	vio_tmp = NULL;
6347 	vio = vdc->failfast_io_queue;
6348 
6349 	if (deadline != 0) {
6350 		/*
6351 		 * Skip any io queued after the deadline. The failfast
6352 		 * I/O queue is ordered starting with the last I/O added
6353 		 * to the queue.
6354 		 */
6355 		while (vio != NULL && vio->vio_qtime > deadline) {
6356 			vio_tmp = vio;
6357 			vio = vio->vio_next;
6358 		}
6359 	}
6360 
6361 	if (vio == NULL)
6362 		/* nothing to unqueue */
6363 		return;
6364 
6365 	/* update the queue */
6366 	if (vio_tmp == NULL)
6367 		vdc->failfast_io_queue = NULL;
6368 	else
6369 		vio_tmp->vio_next = NULL;
6370 
6371 	/*
6372 	 * Complete unqueued I/O. Async I/O have a block I/O data transfer
6373 	 * structure (buf) and they are completed by calling biodone(). Sync
6374 	 * I/O do not have a buf and they are completed by setting the
6375 	 * vio_qtime to zero and signaling failfast_io_cv. In that case, the
6376 	 * thread waiting for the I/O to complete is responsible for freeing
6377 	 * the vio structure.
6378 	 */
6379 	while (vio != NULL) {
6380 		vio_tmp = vio->vio_next;
6381 		if (vio->vio_buf != NULL) {
6382 			VD_KSTAT_RUNQ_EXIT(vdc);
6383 			DTRACE_IO1(done, buf_t *, vio->vio_buf);
6384 			biodone(vio->vio_buf);
6385 			kmem_free(vio, sizeof (vdc_io_t));
6386 		} else {
6387 			vio->vio_qtime = 0;
6388 		}
6389 		vio = vio_tmp;
6390 	}
6391 
6392 	cv_broadcast(&vdc->failfast_io_cv);
6393 }
6394 
6395 /*
6396  * Failfast Thread.
6397  *
6398  * While failfast is enabled, the failfast thread sends a TEST UNIT READY
6399  * and a zero size WRITE(10) SCSI commands on a regular basis to check that
6400  * we still have access to the disk. If a command fails with a RESERVATION
6401  * CONFLICT error then the system will immediatly panic.
6402  *
6403  * The failfast thread is also woken up when an I/O has failed. It then check
6404  * the access to the disk to ensure that the I/O failure was not due to a
6405  * reservation conflict.
6406  *
6407  * There is one failfast thread for each virtual disk for which failfast is
6408  * enabled. We could have only one thread sending requests for all disks but
6409  * this would need vdc to send asynchronous requests and to have callbacks to
6410  * process replies.
6411  */
6412 static void
6413 vdc_failfast_thread(void *arg)
6414 {
6415 	int status;
6416 	vdc_t *vdc = (vdc_t *)arg;
6417 	clock_t timeout, starttime;
6418 
6419 	mutex_enter(&vdc->lock);
6420 
6421 	while (vdc->failfast_interval != 0) {
6422 
6423 		starttime = ddi_get_lbolt();
6424 
6425 		mutex_exit(&vdc->lock);
6426 
6427 		/* check for reservation conflict */
6428 		status = vdc_failfast_check_resv(vdc);
6429 
6430 		mutex_enter(&vdc->lock);
6431 		/*
6432 		 * We have dropped the lock to send the SCSI command so we have
6433 		 * to check that failfast is still enabled.
6434 		 */
6435 		if (vdc->failfast_interval == 0)
6436 			break;
6437 
6438 		/*
6439 		 * If we have successfully check the disk access and there was
6440 		 * no reservation conflict then we can complete any I/O queued
6441 		 * before the last check.
6442 		 */
6443 		if (status == 0)
6444 			vdc_failfast_io_unqueue(vdc, starttime);
6445 
6446 		/* proceed again if some I/O are still in the queue */
6447 		if (vdc->failfast_io_queue != NULL)
6448 			continue;
6449 
6450 		timeout = ddi_get_lbolt() +
6451 		    drv_usectohz(vdc->failfast_interval);
6452 		(void) cv_timedwait(&vdc->failfast_cv, &vdc->lock, timeout);
6453 	}
6454 
6455 	/*
6456 	 * Failfast is being stop so we can complete any queued I/O.
6457 	 */
6458 	vdc_failfast_io_unqueue(vdc, 0);
6459 	vdc->failfast_thread = NULL;
6460 	mutex_exit(&vdc->lock);
6461 	thread_exit();
6462 }
6463 
6464 /*
6465  * Implement the MHIOCENFAILFAST mhd(7i) ioctl.
6466  */
6467 static int
6468 vdc_failfast(vdc_t *vdc, caddr_t arg, int mode)
6469 {
6470 	unsigned int mh_time;
6471 
6472 	if (ddi_copyin((void *)arg, &mh_time, sizeof (int), mode))
6473 		return (EFAULT);
6474 
6475 	mutex_enter(&vdc->lock);
6476 	if (mh_time != 0 && vdc->failfast_thread == NULL) {
6477 		vdc->failfast_thread = thread_create(NULL, 0,
6478 		    vdc_failfast_thread, vdc, 0, &p0, TS_RUN,
6479 		    v.v_maxsyspri - 2);
6480 	}
6481 
6482 	vdc->failfast_interval = mh_time * 1000;
6483 	cv_signal(&vdc->failfast_cv);
6484 	mutex_exit(&vdc->lock);
6485 
6486 	return (0);
6487 }
6488 
6489 /*
6490  * Implement the MHIOCTKOWN and MHIOCRELEASE mhd(7i) ioctls. These ioctls are
6491  * converted to VD_OP_SET_ACCESS operations.
6492  */
6493 static int
6494 vdc_access_set(vdc_t *vdc, uint64_t flags, int mode)
6495 {
6496 	int rv;
6497 
6498 	/* submit owership command request */
6499 	rv = vdc_do_sync_op(vdc, VD_OP_SET_ACCESS, (caddr_t)&flags,
6500 	    sizeof (uint64_t), 0, 0, CB_SYNC, (void *)(uint64_t)mode,
6501 	    VIO_both_dir, B_TRUE);
6502 
6503 	return (rv);
6504 }
6505 
6506 /*
6507  * Implement the MHIOCSTATUS mhd(7i) ioctl. This ioctl is converted to a
6508  * VD_OP_GET_ACCESS operation.
6509  */
6510 static int
6511 vdc_access_get(vdc_t *vdc, uint64_t *status, int mode)
6512 {
6513 	int rv;
6514 
6515 	/* submit owership command request */
6516 	rv = vdc_do_sync_op(vdc, VD_OP_GET_ACCESS, (caddr_t)status,
6517 	    sizeof (uint64_t), 0, 0, CB_SYNC, (void *)(uint64_t)mode,
6518 	    VIO_both_dir, B_TRUE);
6519 
6520 	return (rv);
6521 }
6522 
6523 /*
6524  * Disk Ownership Thread.
6525  *
6526  * When we have taken the ownership of a disk, this thread waits to be
6527  * notified when the LDC channel is reset so that it can recover the
6528  * ownership.
6529  *
6530  * Note that the thread handling the LDC reset (vdc_process_msg_thread())
6531  * can not be used to do the ownership recovery because it has to be
6532  * running to handle the reply message to the ownership operation.
6533  */
6534 static void
6535 vdc_ownership_thread(void *arg)
6536 {
6537 	vdc_t *vdc = (vdc_t *)arg;
6538 	clock_t timeout;
6539 	uint64_t status;
6540 
6541 	mutex_enter(&vdc->ownership_lock);
6542 	mutex_enter(&vdc->lock);
6543 
6544 	while (vdc->ownership & VDC_OWNERSHIP_WANTED) {
6545 
6546 		if ((vdc->ownership & VDC_OWNERSHIP_RESET) ||
6547 		    !(vdc->ownership & VDC_OWNERSHIP_GRANTED)) {
6548 			/*
6549 			 * There was a reset so the ownership has been lost,
6550 			 * try to recover. We do this without using the preempt
6551 			 * option so that we don't steal the ownership from
6552 			 * someone who has preempted us.
6553 			 */
6554 			DMSG(vdc, 0, "[%d] Ownership lost, recovering",
6555 			    vdc->instance);
6556 
6557 			vdc->ownership &= ~(VDC_OWNERSHIP_RESET |
6558 			    VDC_OWNERSHIP_GRANTED);
6559 
6560 			mutex_exit(&vdc->lock);
6561 
6562 			status = vdc_access_set(vdc, VD_ACCESS_SET_EXCLUSIVE |
6563 			    VD_ACCESS_SET_PRESERVE, FKIOCTL);
6564 
6565 			mutex_enter(&vdc->lock);
6566 
6567 			if (status == 0) {
6568 				DMSG(vdc, 0, "[%d] Ownership recovered",
6569 				    vdc->instance);
6570 				vdc->ownership |= VDC_OWNERSHIP_GRANTED;
6571 			} else {
6572 				DMSG(vdc, 0, "[%d] Fail to recover ownership",
6573 				    vdc->instance);
6574 			}
6575 
6576 		}
6577 
6578 		/*
6579 		 * If we have the ownership then we just wait for an event
6580 		 * to happen (LDC reset), otherwise we will retry to recover
6581 		 * after a delay.
6582 		 */
6583 		if (vdc->ownership & VDC_OWNERSHIP_GRANTED)
6584 			timeout = 0;
6585 		else
6586 			timeout = ddi_get_lbolt() +
6587 			    drv_usectohz(vdc_ownership_delay);
6588 
6589 		/* Release the ownership_lock and wait on the vdc lock */
6590 		mutex_exit(&vdc->ownership_lock);
6591 
6592 		if (timeout == 0)
6593 			(void) cv_wait(&vdc->ownership_cv, &vdc->lock);
6594 		else
6595 			(void) cv_timedwait(&vdc->ownership_cv,
6596 			    &vdc->lock, timeout);
6597 
6598 		mutex_exit(&vdc->lock);
6599 
6600 		mutex_enter(&vdc->ownership_lock);
6601 		mutex_enter(&vdc->lock);
6602 	}
6603 
6604 	vdc->ownership_thread = NULL;
6605 	mutex_exit(&vdc->lock);
6606 	mutex_exit(&vdc->ownership_lock);
6607 
6608 	thread_exit();
6609 }
6610 
6611 static void
6612 vdc_ownership_update(vdc_t *vdc, int ownership_flags)
6613 {
6614 	ASSERT(MUTEX_HELD(&vdc->ownership_lock));
6615 
6616 	mutex_enter(&vdc->lock);
6617 	vdc->ownership = ownership_flags;
6618 	if ((vdc->ownership & VDC_OWNERSHIP_WANTED) &&
6619 	    vdc->ownership_thread == NULL) {
6620 		/* start ownership thread */
6621 		vdc->ownership_thread = thread_create(NULL, 0,
6622 		    vdc_ownership_thread, vdc, 0, &p0, TS_RUN,
6623 		    v.v_maxsyspri - 2);
6624 	} else {
6625 		/* notify the ownership thread */
6626 		cv_signal(&vdc->ownership_cv);
6627 	}
6628 	mutex_exit(&vdc->lock);
6629 }
6630 
6631 /*
6632  * Get the size and the block size of a virtual disk from the vdisk server.
6633  */
6634 static int
6635 vdc_get_capacity(vdc_t *vdc, size_t *dsk_size, size_t *blk_size)
6636 {
6637 	int rv = 0;
6638 	size_t alloc_len;
6639 	vd_capacity_t *vd_cap;
6640 
6641 	ASSERT(MUTEX_NOT_HELD(&vdc->lock));
6642 
6643 	alloc_len = P2ROUNDUP(sizeof (vd_capacity_t), sizeof (uint64_t));
6644 
6645 	vd_cap = kmem_zalloc(alloc_len, KM_SLEEP);
6646 
6647 	rv = vdc_do_sync_op(vdc, VD_OP_GET_CAPACITY, (caddr_t)vd_cap, alloc_len,
6648 	    0, 0, CB_SYNC, (void *)(uint64_t)FKIOCTL, VIO_both_dir, B_TRUE);
6649 
6650 	*dsk_size = vd_cap->vdisk_size;
6651 	*blk_size = vd_cap->vdisk_block_size;
6652 
6653 	kmem_free(vd_cap, alloc_len);
6654 	return (rv);
6655 }
6656 
6657 /*
6658  * Check the disk capacity. Disk size information is updated if size has
6659  * changed.
6660  *
6661  * Return 0 if the disk capacity is available, or non-zero if it is not.
6662  */
6663 static int
6664 vdc_check_capacity(vdc_t *vdc)
6665 {
6666 	size_t dsk_size, blk_size;
6667 	int rv;
6668 
6669 	/*
6670 	 * If the vdisk does not support the VD_OP_GET_CAPACITY operation
6671 	 * then the disk capacity has been retrieved during the handshake
6672 	 * and there's nothing more to do here.
6673 	 */
6674 	if (!VD_OP_SUPPORTED(vdc->operations, VD_OP_GET_CAPACITY))
6675 		return (0);
6676 
6677 	if ((rv = vdc_get_capacity(vdc, &dsk_size, &blk_size)) != 0)
6678 		return (rv);
6679 
6680 	if (dsk_size == VD_SIZE_UNKNOWN || dsk_size == 0 || blk_size == 0)
6681 		return (EINVAL);
6682 
6683 	mutex_enter(&vdc->lock);
6684 	/*
6685 	 * First try to update the VIO block size (which is the same as the
6686 	 * vdisk block size). If this returns an error then that means that
6687 	 * we can not use that block size so basically the vdisk is unusable
6688 	 * and we return an error.
6689 	 */
6690 	rv = vdc_update_vio_bsize(vdc, blk_size);
6691 	if (rv == 0)
6692 		vdc_update_size(vdc, dsk_size, blk_size, vdc->max_xfer_sz);
6693 
6694 	mutex_exit(&vdc->lock);
6695 
6696 	return (rv);
6697 }
6698 
6699 /*
6700  * This structure is used in the DKIO(7I) array below.
6701  */
6702 typedef struct vdc_dk_ioctl {
6703 	uint8_t		op;		/* VD_OP_XXX value */
6704 	int		cmd;		/* Solaris ioctl operation number */
6705 	size_t		nbytes;		/* size of structure to be copied */
6706 
6707 	/* function to convert between vDisk and Solaris structure formats */
6708 	int	(*convert)(vdc_t *vdc, void *vd_buf, void *ioctl_arg,
6709 	    int mode, int dir);
6710 } vdc_dk_ioctl_t;
6711 
6712 /*
6713  * Subset of DKIO(7I) operations currently supported
6714  */
6715 static vdc_dk_ioctl_t	dk_ioctl[] = {
6716 	{VD_OP_FLUSH,		DKIOCFLUSHWRITECACHE,	0,
6717 		vdc_null_copy_func},
6718 	{VD_OP_GET_WCE,		DKIOCGETWCE,		sizeof (int),
6719 		vdc_get_wce_convert},
6720 	{VD_OP_SET_WCE,		DKIOCSETWCE,		sizeof (int),
6721 		vdc_set_wce_convert},
6722 	{VD_OP_GET_VTOC,	DKIOCGVTOC,		sizeof (vd_vtoc_t),
6723 		vdc_get_vtoc_convert},
6724 	{VD_OP_SET_VTOC,	DKIOCSVTOC,		sizeof (vd_vtoc_t),
6725 		vdc_set_vtoc_convert},
6726 	{VD_OP_GET_VTOC,	DKIOCGEXTVTOC,		sizeof (vd_vtoc_t),
6727 		vdc_get_extvtoc_convert},
6728 	{VD_OP_SET_VTOC,	DKIOCSEXTVTOC,		sizeof (vd_vtoc_t),
6729 		vdc_set_extvtoc_convert},
6730 	{VD_OP_GET_DISKGEOM,	DKIOCGGEOM,		sizeof (vd_geom_t),
6731 		vdc_get_geom_convert},
6732 	{VD_OP_GET_DISKGEOM,	DKIOCG_PHYGEOM,		sizeof (vd_geom_t),
6733 		vdc_get_geom_convert},
6734 	{VD_OP_GET_DISKGEOM, 	DKIOCG_VIRTGEOM,	sizeof (vd_geom_t),
6735 		vdc_get_geom_convert},
6736 	{VD_OP_SET_DISKGEOM,	DKIOCSGEOM,		sizeof (vd_geom_t),
6737 		vdc_set_geom_convert},
6738 	{VD_OP_GET_EFI,		DKIOCGETEFI,		0,
6739 		vdc_get_efi_convert},
6740 	{VD_OP_SET_EFI,		DKIOCSETEFI,		0,
6741 		vdc_set_efi_convert},
6742 
6743 	/* DIOCTL_RWCMD is converted to a read or a write */
6744 	{0, DIOCTL_RWCMD,  sizeof (struct dadkio_rwcmd), NULL},
6745 
6746 	/* mhd(7I) non-shared multihost disks ioctls */
6747 	{0, MHIOCTKOWN,				0, vdc_null_copy_func},
6748 	{0, MHIOCRELEASE,			0, vdc_null_copy_func},
6749 	{0, MHIOCSTATUS,			0, vdc_null_copy_func},
6750 	{0, MHIOCQRESERVE,			0, vdc_null_copy_func},
6751 
6752 	/* mhd(7I) shared multihost disks ioctls */
6753 	{0, MHIOCGRP_INKEYS,			0, vdc_null_copy_func},
6754 	{0, MHIOCGRP_INRESV,			0, vdc_null_copy_func},
6755 	{0, MHIOCGRP_REGISTER,			0, vdc_null_copy_func},
6756 	{0, MHIOCGRP_RESERVE, 			0, vdc_null_copy_func},
6757 	{0, MHIOCGRP_PREEMPTANDABORT,		0, vdc_null_copy_func},
6758 	{0, MHIOCGRP_REGISTERANDIGNOREKEY,	0, vdc_null_copy_func},
6759 
6760 	/* mhd(7I) failfast ioctl */
6761 	{0, MHIOCENFAILFAST,			0, vdc_null_copy_func},
6762 
6763 	/*
6764 	 * These particular ioctls are not sent to the server - vdc fakes up
6765 	 * the necessary info.
6766 	 */
6767 	{0, DKIOCINFO, sizeof (struct dk_cinfo), vdc_null_copy_func},
6768 	{0, DKIOCGMEDIAINFO, sizeof (struct dk_minfo), vdc_null_copy_func},
6769 	{0, USCSICMD,	sizeof (struct uscsi_cmd), vdc_null_copy_func},
6770 	{0, DKIOCPARTITION, 0, vdc_null_copy_func },
6771 	{0, DKIOCGAPART, 0, vdc_null_copy_func },
6772 	{0, DKIOCREMOVABLE, 0, vdc_null_copy_func},
6773 	{0, CDROMREADOFFSET, 0, vdc_null_copy_func}
6774 };
6775 
6776 /*
6777  * This function handles ioctl requests from the vd_efi_alloc_and_read()
6778  * function and forward them to the vdisk.
6779  */
6780 static int
6781 vd_process_efi_ioctl(void *vdisk, int cmd, uintptr_t arg)
6782 {
6783 	vdc_t *vdc = (vdc_t *)vdisk;
6784 	dev_t dev;
6785 	int rval;
6786 
6787 	dev = makedevice(ddi_driver_major(vdc->dip),
6788 	    VD_MAKE_DEV(vdc->instance, 0));
6789 
6790 	return (vd_process_ioctl(dev, cmd, (caddr_t)arg, FKIOCTL, &rval));
6791 }
6792 
6793 /*
6794  * Function:
6795  *	vd_process_ioctl()
6796  *
6797  * Description:
6798  *	This routine processes disk specific ioctl calls
6799  *
6800  * Arguments:
6801  *	dev	- the device number
6802  *	cmd	- the operation [dkio(7I)] to be processed
6803  *	arg	- pointer to user provided structure
6804  *		  (contains data to be set or reference parameter for get)
6805  *	mode	- bit flag, indicating open settings, 32/64 bit type, etc
6806  *	rvalp	- pointer to return value for calling process.
6807  *
6808  * Return Code:
6809  *	0
6810  *	EFAULT
6811  *	ENXIO
6812  *	EIO
6813  *	ENOTSUP
6814  */
6815 static int
6816 vd_process_ioctl(dev_t dev, int cmd, caddr_t arg, int mode, int *rvalp)
6817 {
6818 	int		instance = VDCUNIT(dev);
6819 	vdc_t		*vdc = NULL;
6820 	int		rv = -1;
6821 	int		idx = 0;		/* index into dk_ioctl[] */
6822 	size_t		len = 0;		/* #bytes to send to vds */
6823 	size_t		alloc_len = 0;		/* #bytes to allocate mem for */
6824 	caddr_t		mem_p = NULL;
6825 	size_t		nioctls = (sizeof (dk_ioctl)) / (sizeof (dk_ioctl[0]));
6826 	vdc_dk_ioctl_t	*iop;
6827 
6828 	vdc = ddi_get_soft_state(vdc_state, instance);
6829 	if (vdc == NULL) {
6830 		cmn_err(CE_NOTE, "![%d] Could not get soft state structure",
6831 		    instance);
6832 		return (ENXIO);
6833 	}
6834 
6835 	DMSG(vdc, 0, "[%d] Processing ioctl(%x) for dev %lx : model %x\n",
6836 	    instance, cmd, dev, ddi_model_convert_from(mode & FMODELS));
6837 
6838 	if (rvalp != NULL) {
6839 		/* the return value of the ioctl is 0 by default */
6840 		*rvalp = 0;
6841 	}
6842 
6843 	/*
6844 	 * Validate the ioctl operation to be performed.
6845 	 *
6846 	 * If we have looped through the array without finding a match then we
6847 	 * don't support this ioctl.
6848 	 */
6849 	for (idx = 0; idx < nioctls; idx++) {
6850 		if (cmd == dk_ioctl[idx].cmd)
6851 			break;
6852 	}
6853 
6854 	if (idx >= nioctls) {
6855 		DMSG(vdc, 0, "[%d] Unsupported ioctl (0x%x)\n",
6856 		    vdc->instance, cmd);
6857 		return (ENOTSUP);
6858 	}
6859 
6860 	iop = &(dk_ioctl[idx]);
6861 
6862 	if (cmd == DKIOCGETEFI || cmd == DKIOCSETEFI) {
6863 		/* size is not fixed for EFI ioctls, it depends on ioctl arg */
6864 		dk_efi_t	dk_efi;
6865 
6866 		rv = ddi_copyin(arg, &dk_efi, sizeof (dk_efi_t), mode);
6867 		if (rv != 0)
6868 			return (EFAULT);
6869 
6870 		len = sizeof (vd_efi_t) - 1 + dk_efi.dki_length;
6871 	} else {
6872 		len = iop->nbytes;
6873 	}
6874 
6875 	/* check if the ioctl is applicable */
6876 	switch (cmd) {
6877 	case CDROMREADOFFSET:
6878 	case DKIOCREMOVABLE:
6879 		return (ENOTTY);
6880 
6881 	case USCSICMD:
6882 	case MHIOCTKOWN:
6883 	case MHIOCSTATUS:
6884 	case MHIOCQRESERVE:
6885 	case MHIOCRELEASE:
6886 	case MHIOCGRP_INKEYS:
6887 	case MHIOCGRP_INRESV:
6888 	case MHIOCGRP_REGISTER:
6889 	case MHIOCGRP_RESERVE:
6890 	case MHIOCGRP_PREEMPTANDABORT:
6891 	case MHIOCGRP_REGISTERANDIGNOREKEY:
6892 	case MHIOCENFAILFAST:
6893 		if (vdc->cinfo == NULL)
6894 			return (ENXIO);
6895 		if (vdc->cinfo->dki_ctype != DKC_SCSI_CCS)
6896 			return (ENOTTY);
6897 		break;
6898 
6899 	case DIOCTL_RWCMD:
6900 		if (vdc->cinfo == NULL)
6901 			return (ENXIO);
6902 		if (vdc->cinfo->dki_ctype != DKC_DIRECT)
6903 			return (ENOTTY);
6904 		break;
6905 
6906 	case DKIOCINFO:
6907 		if (vdc->cinfo == NULL)
6908 			return (ENXIO);
6909 		break;
6910 
6911 	case DKIOCGMEDIAINFO:
6912 		if (vdc->minfo == NULL)
6913 			return (ENXIO);
6914 		if (vdc_check_capacity(vdc) != 0)
6915 			/* disk capacity is not available */
6916 			return (EIO);
6917 		break;
6918 	}
6919 
6920 	/*
6921 	 * Deal with ioctls which require a processing different than
6922 	 * converting ioctl arguments and sending a corresponding
6923 	 * VD operation.
6924 	 */
6925 	switch (cmd) {
6926 
6927 	case USCSICMD:
6928 	{
6929 		return (vdc_uscsi_cmd(vdc, arg, mode));
6930 	}
6931 
6932 	case MHIOCTKOWN:
6933 	{
6934 		mutex_enter(&vdc->ownership_lock);
6935 		/*
6936 		 * We have to set VDC_OWNERSHIP_WANTED now so that the ownership
6937 		 * can be flagged with VDC_OWNERSHIP_RESET if the LDC is reset
6938 		 * while we are processing the ioctl.
6939 		 */
6940 		vdc_ownership_update(vdc, VDC_OWNERSHIP_WANTED);
6941 
6942 		rv = vdc_access_set(vdc, VD_ACCESS_SET_EXCLUSIVE |
6943 		    VD_ACCESS_SET_PREEMPT | VD_ACCESS_SET_PRESERVE, mode);
6944 		if (rv == 0) {
6945 			vdc_ownership_update(vdc, VDC_OWNERSHIP_WANTED |
6946 			    VDC_OWNERSHIP_GRANTED);
6947 		} else {
6948 			vdc_ownership_update(vdc, VDC_OWNERSHIP_NONE);
6949 		}
6950 		mutex_exit(&vdc->ownership_lock);
6951 		return (rv);
6952 	}
6953 
6954 	case MHIOCRELEASE:
6955 	{
6956 		mutex_enter(&vdc->ownership_lock);
6957 		rv = vdc_access_set(vdc, VD_ACCESS_SET_CLEAR, mode);
6958 		if (rv == 0) {
6959 			vdc_ownership_update(vdc, VDC_OWNERSHIP_NONE);
6960 		}
6961 		mutex_exit(&vdc->ownership_lock);
6962 		return (rv);
6963 	}
6964 
6965 	case MHIOCSTATUS:
6966 	{
6967 		uint64_t status;
6968 
6969 		rv = vdc_access_get(vdc, &status, mode);
6970 		if (rv == 0 && rvalp != NULL)
6971 			*rvalp = (status & VD_ACCESS_ALLOWED)? 0 : 1;
6972 		return (rv);
6973 	}
6974 
6975 	case MHIOCQRESERVE:
6976 	{
6977 		rv = vdc_access_set(vdc, VD_ACCESS_SET_EXCLUSIVE, mode);
6978 		return (rv);
6979 	}
6980 
6981 	case MHIOCGRP_INKEYS:
6982 	{
6983 		return (vdc_mhd_inkeys(vdc, arg, mode));
6984 	}
6985 
6986 	case MHIOCGRP_INRESV:
6987 	{
6988 		return (vdc_mhd_inresv(vdc, arg, mode));
6989 	}
6990 
6991 	case MHIOCGRP_REGISTER:
6992 	{
6993 		return (vdc_mhd_register(vdc, arg, mode));
6994 	}
6995 
6996 	case MHIOCGRP_RESERVE:
6997 	{
6998 		return (vdc_mhd_reserve(vdc, arg, mode));
6999 	}
7000 
7001 	case MHIOCGRP_PREEMPTANDABORT:
7002 	{
7003 		return (vdc_mhd_preemptabort(vdc, arg, mode));
7004 	}
7005 
7006 	case MHIOCGRP_REGISTERANDIGNOREKEY:
7007 	{
7008 		return (vdc_mhd_registerignore(vdc, arg, mode));
7009 	}
7010 
7011 	case MHIOCENFAILFAST:
7012 	{
7013 		rv = vdc_failfast(vdc, arg, mode);
7014 		return (rv);
7015 	}
7016 
7017 	case DIOCTL_RWCMD:
7018 	{
7019 		return (vdc_dioctl_rwcmd(vdc, arg, mode));
7020 	}
7021 
7022 	case DKIOCGAPART:
7023 	{
7024 		return (vdc_dkio_gapart(vdc, arg, mode));
7025 	}
7026 
7027 	case DKIOCPARTITION:
7028 	{
7029 		return (vdc_dkio_partition(vdc, arg, mode));
7030 	}
7031 
7032 	case DKIOCINFO:
7033 	{
7034 		struct dk_cinfo	cinfo;
7035 
7036 		bcopy(vdc->cinfo, &cinfo, sizeof (struct dk_cinfo));
7037 		cinfo.dki_partition = VDCPART(dev);
7038 
7039 		rv = ddi_copyout(&cinfo, (void *)arg,
7040 		    sizeof (struct dk_cinfo), mode);
7041 		if (rv != 0)
7042 			return (EFAULT);
7043 
7044 		return (0);
7045 	}
7046 
7047 	case DKIOCGMEDIAINFO:
7048 	{
7049 		ASSERT(vdc->vdisk_size != 0);
7050 		ASSERT(vdc->minfo->dki_capacity != 0);
7051 		rv = ddi_copyout(vdc->minfo, (void *)arg,
7052 		    sizeof (struct dk_minfo), mode);
7053 		if (rv != 0)
7054 			return (EFAULT);
7055 
7056 		return (0);
7057 	}
7058 
7059 	case DKIOCFLUSHWRITECACHE:
7060 		{
7061 			struct dk_callback *dkc =
7062 			    (struct dk_callback *)(uintptr_t)arg;
7063 			vdc_dk_arg_t	*dkarg = NULL;
7064 
7065 			DMSG(vdc, 1, "[%d] Flush W$: mode %x\n",
7066 			    instance, mode);
7067 
7068 			/*
7069 			 * If arg is NULL, then there is no callback function
7070 			 * registered and the call operates synchronously; we
7071 			 * break and continue with the rest of the function and
7072 			 * wait for vds to return (i.e. after the request to
7073 			 * vds returns successfully, all writes completed prior
7074 			 * to the ioctl will have been flushed from the disk
7075 			 * write cache to persistent media.
7076 			 *
7077 			 * If a callback function is registered, we dispatch
7078 			 * the request on a task queue and return immediately.
7079 			 * The callback will deal with informing the calling
7080 			 * thread that the flush request is completed.
7081 			 */
7082 			if (dkc == NULL)
7083 				break;
7084 
7085 			/*
7086 			 * the asynchronous callback is only supported if
7087 			 * invoked from within the kernel
7088 			 */
7089 			if ((mode & FKIOCTL) == 0)
7090 				return (ENOTSUP);
7091 
7092 			dkarg = kmem_zalloc(sizeof (vdc_dk_arg_t), KM_SLEEP);
7093 
7094 			dkarg->mode = mode;
7095 			dkarg->dev = dev;
7096 			bcopy(dkc, &dkarg->dkc, sizeof (*dkc));
7097 
7098 			mutex_enter(&vdc->lock);
7099 			vdc->dkio_flush_pending++;
7100 			dkarg->vdc = vdc;
7101 			mutex_exit(&vdc->lock);
7102 
7103 			/* put the request on a task queue */
7104 			rv = taskq_dispatch(system_taskq, vdc_dkio_flush_cb,
7105 			    (void *)dkarg, DDI_SLEEP);
7106 			if (rv == NULL) {
7107 				/* clean up if dispatch fails */
7108 				mutex_enter(&vdc->lock);
7109 				vdc->dkio_flush_pending--;
7110 				mutex_exit(&vdc->lock);
7111 				kmem_free(dkarg, sizeof (vdc_dk_arg_t));
7112 			}
7113 
7114 			return (rv == NULL ? ENOMEM : 0);
7115 		}
7116 	}
7117 
7118 	/* catch programming error in vdc - should be a VD_OP_XXX ioctl */
7119 	ASSERT(iop->op != 0);
7120 
7121 	/* check if the vDisk server handles the operation for this vDisk */
7122 	if (VD_OP_SUPPORTED(vdc->operations, iop->op) == B_FALSE) {
7123 		DMSG(vdc, 0, "[%d] Unsupported VD_OP operation (0x%x)\n",
7124 		    vdc->instance, iop->op);
7125 		return (ENOTSUP);
7126 	}
7127 
7128 	/* LDC requires that the memory being mapped is 8-byte aligned */
7129 	alloc_len = P2ROUNDUP(len, sizeof (uint64_t));
7130 	DMSG(vdc, 1, "[%d] struct size %ld alloc %ld\n",
7131 	    instance, len, alloc_len);
7132 
7133 	if (alloc_len > 0)
7134 		mem_p = kmem_zalloc(alloc_len, KM_SLEEP);
7135 
7136 	/*
7137 	 * Call the conversion function for this ioctl which, if necessary,
7138 	 * converts from the Solaris format to the format ARC'ed
7139 	 * as part of the vDisk protocol (FWARC 2006/195)
7140 	 */
7141 	ASSERT(iop->convert != NULL);
7142 	rv = (iop->convert)(vdc, arg, mem_p, mode, VD_COPYIN);
7143 	if (rv != 0) {
7144 		DMSG(vdc, 0, "[%d] convert func returned %d for ioctl 0x%x\n",
7145 		    instance, rv, cmd);
7146 		if (mem_p != NULL)
7147 			kmem_free(mem_p, alloc_len);
7148 		return (rv);
7149 	}
7150 
7151 	/*
7152 	 * send request to vds to service the ioctl.
7153 	 */
7154 	rv = vdc_do_sync_op(vdc, iop->op, mem_p, alloc_len,
7155 	    VDCPART(dev), 0, CB_SYNC, (void *)(uint64_t)mode,
7156 	    VIO_both_dir, B_TRUE);
7157 
7158 	if (rv != 0) {
7159 		/*
7160 		 * This is not necessarily an error. The ioctl could
7161 		 * be returning a value such as ENOTTY to indicate
7162 		 * that the ioctl is not applicable.
7163 		 */
7164 		DMSG(vdc, 0, "[%d] vds returned %d for ioctl 0x%x\n",
7165 		    instance, rv, cmd);
7166 		if (mem_p != NULL)
7167 			kmem_free(mem_p, alloc_len);
7168 
7169 		return (rv);
7170 	}
7171 
7172 	/*
7173 	 * Call the conversion function (if it exists) for this ioctl
7174 	 * which converts from the format ARC'ed as part of the vDisk
7175 	 * protocol (FWARC 2006/195) back to a format understood by
7176 	 * the rest of Solaris.
7177 	 */
7178 	rv = (iop->convert)(vdc, mem_p, arg, mode, VD_COPYOUT);
7179 	if (rv != 0) {
7180 		DMSG(vdc, 0, "[%d] convert func returned %d for ioctl 0x%x\n",
7181 		    instance, rv, cmd);
7182 		if (mem_p != NULL)
7183 			kmem_free(mem_p, alloc_len);
7184 		return (rv);
7185 	}
7186 
7187 	if (mem_p != NULL)
7188 		kmem_free(mem_p, alloc_len);
7189 
7190 	return (rv);
7191 }
7192 
7193 /*
7194  * Function:
7195  *
7196  * Description:
7197  *	This is an empty conversion function used by ioctl calls which
7198  *	do not need to convert the data being passed in/out to userland
7199  */
7200 static int
7201 vdc_null_copy_func(vdc_t *vdc, void *from, void *to, int mode, int dir)
7202 {
7203 	_NOTE(ARGUNUSED(vdc))
7204 	_NOTE(ARGUNUSED(from))
7205 	_NOTE(ARGUNUSED(to))
7206 	_NOTE(ARGUNUSED(mode))
7207 	_NOTE(ARGUNUSED(dir))
7208 
7209 	return (0);
7210 }
7211 
7212 static int
7213 vdc_get_wce_convert(vdc_t *vdc, void *from, void *to,
7214     int mode, int dir)
7215 {
7216 	_NOTE(ARGUNUSED(vdc))
7217 
7218 	if (dir == VD_COPYIN)
7219 		return (0);		/* nothing to do */
7220 
7221 	if (ddi_copyout(from, to, sizeof (int), mode) != 0)
7222 		return (EFAULT);
7223 
7224 	return (0);
7225 }
7226 
7227 static int
7228 vdc_set_wce_convert(vdc_t *vdc, void *from, void *to,
7229     int mode, int dir)
7230 {
7231 	_NOTE(ARGUNUSED(vdc))
7232 
7233 	if (dir == VD_COPYOUT)
7234 		return (0);		/* nothing to do */
7235 
7236 	if (ddi_copyin(from, to, sizeof (int), mode) != 0)
7237 		return (EFAULT);
7238 
7239 	return (0);
7240 }
7241 
7242 /*
7243  * Function:
7244  *	vdc_get_vtoc_convert()
7245  *
7246  * Description:
7247  *	This routine performs the necessary convertions from the DKIOCGVTOC
7248  *	Solaris structure to the format defined in FWARC 2006/195.
7249  *
7250  *	In the struct vtoc definition, the timestamp field is marked as not
7251  *	supported so it is not part of vDisk protocol (FWARC 2006/195).
7252  *	However SVM uses that field to check it can write into the VTOC,
7253  *	so we fake up the info of that field.
7254  *
7255  * Arguments:
7256  *	vdc	- the vDisk client
7257  *	from	- the buffer containing the data to be copied from
7258  *	to	- the buffer to be copied to
7259  *	mode	- flags passed to ioctl() call
7260  *	dir	- the "direction" of the copy - VD_COPYIN or VD_COPYOUT
7261  *
7262  * Return Code:
7263  *	0	- Success
7264  *	ENXIO	- incorrect buffer passed in.
7265  *	EFAULT	- ddi_copyout routine encountered an error.
7266  */
7267 static int
7268 vdc_get_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7269 {
7270 	int		i;
7271 	struct vtoc	vtoc;
7272 	struct vtoc32	vtoc32;
7273 	struct extvtoc	evtoc;
7274 	int		rv;
7275 
7276 	if (dir != VD_COPYOUT)
7277 		return (0);	/* nothing to do */
7278 
7279 	if ((from == NULL) || (to == NULL))
7280 		return (ENXIO);
7281 
7282 	if (vdc->vdisk_size > VD_OLDVTOC_LIMIT)
7283 		return (EOVERFLOW);
7284 
7285 	VD_VTOC2VTOC((vd_vtoc_t *)from, &evtoc);
7286 
7287 	/* fake the VTOC timestamp field */
7288 	for (i = 0; i < V_NUMPAR; i++) {
7289 		evtoc.timestamp[i] = vdc->vtoc->timestamp[i];
7290 	}
7291 
7292 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
7293 		/* LINTED E_ASSIGN_NARROW_CONV */
7294 		extvtoctovtoc32(evtoc, vtoc32);
7295 		rv = ddi_copyout(&vtoc32, to, sizeof (vtoc32), mode);
7296 		if (rv != 0)
7297 			rv = EFAULT;
7298 	} else {
7299 		extvtoctovtoc(evtoc, vtoc);
7300 		rv = ddi_copyout(&vtoc, to, sizeof (vtoc), mode);
7301 		if (rv != 0)
7302 			rv = EFAULT;
7303 	}
7304 
7305 	return (rv);
7306 }
7307 
7308 /*
7309  * Function:
7310  *	vdc_set_vtoc_convert()
7311  *
7312  * Description:
7313  *	This routine performs the necessary convertions from the DKIOCSVTOC
7314  *	Solaris structure to the format defined in FWARC 2006/195.
7315  *
7316  * Arguments:
7317  *	vdc	- the vDisk client
7318  *	from	- Buffer with data
7319  *	to	- Buffer where data is to be copied to
7320  *	mode	- flags passed to ioctl
7321  *	dir	- direction of copy (in or out)
7322  *
7323  * Return Code:
7324  *	0	- Success
7325  *	ENXIO	- Invalid buffer passed in
7326  *	EFAULT	- ddi_copyin of data failed
7327  */
7328 static int
7329 vdc_set_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7330 {
7331 	void		*uvtoc;
7332 	struct vtoc	vtoc;
7333 	struct vtoc32	vtoc32;
7334 	struct extvtoc	evtoc;
7335 	int		i, rv;
7336 
7337 	if ((from == NULL) || (to == NULL))
7338 		return (ENXIO);
7339 
7340 	if (vdc->vdisk_size > VD_OLDVTOC_LIMIT)
7341 		return (EOVERFLOW);
7342 
7343 	uvtoc = (dir == VD_COPYIN)? from : to;
7344 
7345 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
7346 		rv = ddi_copyin(uvtoc, &vtoc32, sizeof (vtoc32), mode);
7347 		if (rv != 0)
7348 			return (EFAULT);
7349 		vtoc32toextvtoc(vtoc32, evtoc);
7350 	} else {
7351 		rv = ddi_copyin(uvtoc, &vtoc, sizeof (vtoc), mode);
7352 		if (rv != 0)
7353 			return (EFAULT);
7354 		vtoctoextvtoc(vtoc, evtoc);
7355 	}
7356 
7357 	if (dir == VD_COPYOUT) {
7358 		/*
7359 		 * The disk label may have changed. Revalidate the disk
7360 		 * geometry. This will also update the device nodes.
7361 		 */
7362 		vdc_validate(vdc);
7363 
7364 		/*
7365 		 * We also need to keep track of the timestamp fields.
7366 		 */
7367 		for (i = 0; i < V_NUMPAR; i++) {
7368 			vdc->vtoc->timestamp[i] = evtoc.timestamp[i];
7369 		}
7370 
7371 	} else {
7372 		VTOC2VD_VTOC(&evtoc, (vd_vtoc_t *)to);
7373 	}
7374 
7375 	return (0);
7376 }
7377 
7378 static int
7379 vdc_get_extvtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7380 {
7381 	int		i, rv;
7382 	struct extvtoc	evtoc;
7383 
7384 	if (dir != VD_COPYOUT)
7385 		return (0);	/* nothing to do */
7386 
7387 	if ((from == NULL) || (to == NULL))
7388 		return (ENXIO);
7389 
7390 	VD_VTOC2VTOC((vd_vtoc_t *)from, &evtoc);
7391 
7392 	/* fake the VTOC timestamp field */
7393 	for (i = 0; i < V_NUMPAR; i++) {
7394 		evtoc.timestamp[i] = vdc->vtoc->timestamp[i];
7395 	}
7396 
7397 	rv = ddi_copyout(&evtoc, to, sizeof (struct extvtoc), mode);
7398 	if (rv != 0)
7399 		rv = EFAULT;
7400 
7401 	return (rv);
7402 }
7403 
7404 static int
7405 vdc_set_extvtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7406 {
7407 	void		*uvtoc;
7408 	struct extvtoc	evtoc;
7409 	int		i, rv;
7410 
7411 	if ((from == NULL) || (to == NULL))
7412 		return (ENXIO);
7413 
7414 	uvtoc = (dir == VD_COPYIN)? from : to;
7415 
7416 	rv = ddi_copyin(uvtoc, &evtoc, sizeof (struct extvtoc), mode);
7417 	if (rv != 0)
7418 		return (EFAULT);
7419 
7420 	if (dir == VD_COPYOUT) {
7421 		/*
7422 		 * The disk label may have changed. Revalidate the disk
7423 		 * geometry. This will also update the device nodes.
7424 		 */
7425 		vdc_validate(vdc);
7426 
7427 		/*
7428 		 * We also need to keep track of the timestamp fields.
7429 		 */
7430 		for (i = 0; i < V_NUMPAR; i++) {
7431 			vdc->vtoc->timestamp[i] = evtoc.timestamp[i];
7432 		}
7433 
7434 	} else {
7435 		VTOC2VD_VTOC(&evtoc, (vd_vtoc_t *)to);
7436 	}
7437 
7438 	return (0);
7439 }
7440 
7441 /*
7442  * Function:
7443  *	vdc_get_geom_convert()
7444  *
7445  * Description:
7446  *	This routine performs the necessary convertions from the DKIOCGGEOM,
7447  *	DKIOCG_PHYSGEOM and DKIOG_VIRTGEOM Solaris structures to the format
7448  *	defined in FWARC 2006/195
7449  *
7450  * Arguments:
7451  *	vdc	- the vDisk client
7452  *	from	- Buffer with data
7453  *	to	- Buffer where data is to be copied to
7454  *	mode	- flags passed to ioctl
7455  *	dir	- direction of copy (in or out)
7456  *
7457  * Return Code:
7458  *	0	- Success
7459  *	ENXIO	- Invalid buffer passed in
7460  *	EFAULT	- ddi_copyout of data failed
7461  */
7462 static int
7463 vdc_get_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7464 {
7465 	_NOTE(ARGUNUSED(vdc))
7466 
7467 	struct dk_geom	geom;
7468 	int	copy_len = sizeof (struct dk_geom);
7469 	int	rv = 0;
7470 
7471 	if (dir != VD_COPYOUT)
7472 		return (0);	/* nothing to do */
7473 
7474 	if ((from == NULL) || (to == NULL))
7475 		return (ENXIO);
7476 
7477 	VD_GEOM2DK_GEOM((vd_geom_t *)from, &geom);
7478 	rv = ddi_copyout(&geom, to, copy_len, mode);
7479 	if (rv != 0)
7480 		rv = EFAULT;
7481 
7482 	return (rv);
7483 }
7484 
7485 /*
7486  * Function:
7487  *	vdc_set_geom_convert()
7488  *
7489  * Description:
7490  *	This routine performs the necessary convertions from the DKIOCSGEOM
7491  *	Solaris structure to the format defined in FWARC 2006/195.
7492  *
7493  * Arguments:
7494  *	vdc	- the vDisk client
7495  *	from	- Buffer with data
7496  *	to	- Buffer where data is to be copied to
7497  *	mode	- flags passed to ioctl
7498  *	dir	- direction of copy (in or out)
7499  *
7500  * Return Code:
7501  *	0	- Success
7502  *	ENXIO	- Invalid buffer passed in
7503  *	EFAULT	- ddi_copyin of data failed
7504  */
7505 static int
7506 vdc_set_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7507 {
7508 	_NOTE(ARGUNUSED(vdc))
7509 
7510 	vd_geom_t	vdgeom;
7511 	void		*tmp_mem = NULL;
7512 	int		copy_len = sizeof (struct dk_geom);
7513 	int		rv = 0;
7514 
7515 	if (dir != VD_COPYIN)
7516 		return (0);	/* nothing to do */
7517 
7518 	if ((from == NULL) || (to == NULL))
7519 		return (ENXIO);
7520 
7521 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
7522 
7523 	rv = ddi_copyin(from, tmp_mem, copy_len, mode);
7524 	if (rv != 0) {
7525 		kmem_free(tmp_mem, copy_len);
7526 		return (EFAULT);
7527 	}
7528 	DK_GEOM2VD_GEOM((struct dk_geom *)tmp_mem, &vdgeom);
7529 	bcopy(&vdgeom, to, sizeof (vdgeom));
7530 	kmem_free(tmp_mem, copy_len);
7531 
7532 	return (0);
7533 }
7534 
7535 static int
7536 vdc_get_efi_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7537 {
7538 	_NOTE(ARGUNUSED(vdc))
7539 
7540 	vd_efi_t	*vd_efi;
7541 	dk_efi_t	dk_efi;
7542 	int		rv = 0;
7543 	void		*uaddr;
7544 
7545 	if ((from == NULL) || (to == NULL))
7546 		return (ENXIO);
7547 
7548 	if (dir == VD_COPYIN) {
7549 
7550 		vd_efi = (vd_efi_t *)to;
7551 
7552 		rv = ddi_copyin(from, &dk_efi, sizeof (dk_efi_t), mode);
7553 		if (rv != 0)
7554 			return (EFAULT);
7555 
7556 		vd_efi->lba = dk_efi.dki_lba;
7557 		vd_efi->length = dk_efi.dki_length;
7558 		bzero(vd_efi->data, vd_efi->length);
7559 
7560 	} else {
7561 
7562 		rv = ddi_copyin(to, &dk_efi, sizeof (dk_efi_t), mode);
7563 		if (rv != 0)
7564 			return (EFAULT);
7565 
7566 		uaddr = dk_efi.dki_data;
7567 
7568 		dk_efi.dki_data = kmem_alloc(dk_efi.dki_length, KM_SLEEP);
7569 
7570 		VD_EFI2DK_EFI((vd_efi_t *)from, &dk_efi);
7571 
7572 		rv = ddi_copyout(dk_efi.dki_data, uaddr, dk_efi.dki_length,
7573 		    mode);
7574 		if (rv != 0)
7575 			return (EFAULT);
7576 
7577 		kmem_free(dk_efi.dki_data, dk_efi.dki_length);
7578 	}
7579 
7580 	return (0);
7581 }
7582 
7583 static int
7584 vdc_set_efi_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7585 {
7586 	_NOTE(ARGUNUSED(vdc))
7587 
7588 	dk_efi_t	dk_efi;
7589 	void		*uaddr;
7590 
7591 	if (dir == VD_COPYOUT) {
7592 		/*
7593 		 * The disk label may have changed. Revalidate the disk
7594 		 * geometry. This will also update the device nodes.
7595 		 */
7596 		vdc_validate(vdc);
7597 		return (0);
7598 	}
7599 
7600 	if ((from == NULL) || (to == NULL))
7601 		return (ENXIO);
7602 
7603 	if (ddi_copyin(from, &dk_efi, sizeof (dk_efi_t), mode) != 0)
7604 		return (EFAULT);
7605 
7606 	uaddr = dk_efi.dki_data;
7607 
7608 	dk_efi.dki_data = kmem_alloc(dk_efi.dki_length, KM_SLEEP);
7609 
7610 	if (ddi_copyin(uaddr, dk_efi.dki_data, dk_efi.dki_length, mode) != 0)
7611 		return (EFAULT);
7612 
7613 	DK_EFI2VD_EFI(&dk_efi, (vd_efi_t *)to);
7614 
7615 	kmem_free(dk_efi.dki_data, dk_efi.dki_length);
7616 
7617 	return (0);
7618 }
7619 
7620 
7621 /* -------------------------------------------------------------------------- */
7622 
7623 /*
7624  * Function:
7625  *	vdc_create_fake_geometry()
7626  *
7627  * Description:
7628  *	This routine fakes up the disk info needed for some DKIO ioctls such
7629  *	as DKIOCINFO and DKIOCGMEDIAINFO [just like lofi(7D) and ramdisk(7D) do]
7630  *
7631  *	Note: This function must not be called until the vDisk attributes have
7632  *	been exchanged as part of the handshake with the vDisk server.
7633  *
7634  * Arguments:
7635  *	vdc	- soft state pointer for this instance of the device driver.
7636  *
7637  * Return Code:
7638  *	none.
7639  */
7640 static void
7641 vdc_create_fake_geometry(vdc_t *vdc)
7642 {
7643 	ASSERT(vdc != NULL);
7644 	ASSERT(vdc->max_xfer_sz != 0);
7645 
7646 	/*
7647 	 * DKIOCINFO support
7648 	 */
7649 	if (vdc->cinfo == NULL)
7650 		vdc->cinfo = kmem_zalloc(sizeof (struct dk_cinfo), KM_SLEEP);
7651 
7652 	(void) strcpy(vdc->cinfo->dki_cname, VDC_DRIVER_NAME);
7653 	(void) strcpy(vdc->cinfo->dki_dname, VDC_DRIVER_NAME);
7654 	/* max_xfer_sz is #blocks so we don't need to divide by vdisk_bsize */
7655 	vdc->cinfo->dki_maxtransfer = vdc->max_xfer_sz;
7656 
7657 	/*
7658 	 * We set the controller type to DKC_SCSI_CCS only if the VD_OP_SCSICMD
7659 	 * operation is supported, otherwise the controller type is DKC_DIRECT.
7660 	 * Version 1.0 does not support the VD_OP_SCSICMD operation, so the
7661 	 * controller type is always DKC_DIRECT in that case.
7662 	 *
7663 	 * If the virtual disk is backed by a physical CD/DVD device or
7664 	 * an ISO image, modify the controller type to indicate this
7665 	 */
7666 	switch (vdc->vdisk_media) {
7667 	case VD_MEDIA_CD:
7668 	case VD_MEDIA_DVD:
7669 		vdc->cinfo->dki_ctype = DKC_CDROM;
7670 		break;
7671 	case VD_MEDIA_FIXED:
7672 		if (VD_OP_SUPPORTED(vdc->operations, VD_OP_SCSICMD))
7673 			vdc->cinfo->dki_ctype = DKC_SCSI_CCS;
7674 		else
7675 			vdc->cinfo->dki_ctype = DKC_DIRECT;
7676 		break;
7677 	default:
7678 		/* in the case of v1.0 we default to a fixed disk */
7679 		vdc->cinfo->dki_ctype = DKC_DIRECT;
7680 		break;
7681 	}
7682 	vdc->cinfo->dki_flags = DKI_FMTVOL;
7683 	vdc->cinfo->dki_cnum = 0;
7684 	vdc->cinfo->dki_addr = 0;
7685 	vdc->cinfo->dki_space = 0;
7686 	vdc->cinfo->dki_prio = 0;
7687 	vdc->cinfo->dki_vec = 0;
7688 	vdc->cinfo->dki_unit = vdc->instance;
7689 	vdc->cinfo->dki_slave = 0;
7690 	/*
7691 	 * The partition number will be created on the fly depending on the
7692 	 * actual slice (i.e. minor node) that is used to request the data.
7693 	 */
7694 	vdc->cinfo->dki_partition = 0;
7695 
7696 	/*
7697 	 * DKIOCGMEDIAINFO support
7698 	 */
7699 	if (vdc->minfo == NULL)
7700 		vdc->minfo = kmem_zalloc(sizeof (struct dk_minfo), KM_SLEEP);
7701 
7702 	if (vio_ver_is_supported(vdc->ver, 1, 1)) {
7703 		vdc->minfo->dki_media_type =
7704 		    VD_MEDIATYPE2DK_MEDIATYPE(vdc->vdisk_media);
7705 	} else {
7706 		vdc->minfo->dki_media_type = DK_FIXED_DISK;
7707 	}
7708 
7709 	vdc->minfo->dki_capacity = vdc->vdisk_size;
7710 	vdc->minfo->dki_lbsize = vdc->vdisk_bsize;
7711 }
7712 
7713 static ushort_t
7714 vdc_lbl2cksum(struct dk_label *label)
7715 {
7716 	int	count;
7717 	ushort_t sum, *sp;
7718 
7719 	count =	(sizeof (struct dk_label)) / (sizeof (short)) - 1;
7720 	sp = (ushort_t *)label;
7721 	sum = 0;
7722 	while (count--) {
7723 		sum ^= *sp++;
7724 	}
7725 
7726 	return (sum);
7727 }
7728 
7729 static void
7730 vdc_update_size(vdc_t *vdc, size_t dsk_size, size_t blk_size, size_t xfr_size)
7731 {
7732 	vd_err_stats_t  *stp;
7733 
7734 	ASSERT(MUTEX_HELD(&vdc->lock));
7735 	ASSERT(xfr_size != 0);
7736 
7737 	/*
7738 	 * If the disk size is unknown or sizes are unchanged then don't
7739 	 * update anything.
7740 	 */
7741 	if (dsk_size == VD_SIZE_UNKNOWN || dsk_size == 0 ||
7742 	    (blk_size == vdc->vdisk_bsize && dsk_size == vdc->vdisk_size &&
7743 	    xfr_size == vdc->max_xfer_sz))
7744 		return;
7745 
7746 	/*
7747 	 * We don't know at compile time what the vDisk server will think
7748 	 * are good values but we apply a large (arbitrary) upper bound to
7749 	 * prevent memory exhaustion in vdc if it was allocating a DRing
7750 	 * based of huge values sent by the server. We probably will never
7751 	 * exceed this except if the message was garbage.
7752 	 */
7753 	if ((xfr_size * blk_size) > (PAGESIZE * DEV_BSIZE)) {
7754 		DMSG(vdc, 0, "[%d] vds block transfer size too big;"
7755 		    " using max supported by vdc", vdc->instance);
7756 		xfr_size = maxphys / blk_size;
7757 	}
7758 
7759 	vdc->max_xfer_sz = xfr_size;
7760 	vdc->vdisk_bsize = blk_size;
7761 	vdc->vdisk_size = dsk_size;
7762 
7763 	stp = (vd_err_stats_t *)vdc->err_stats->ks_data;
7764 	stp->vd_capacity.value.ui64 = dsk_size * blk_size;
7765 
7766 	vdc->minfo->dki_capacity = dsk_size;
7767 	vdc->minfo->dki_lbsize = (uint_t)blk_size;
7768 }
7769 
7770 /*
7771  * Update information about the VIO block size. The VIO block size is the
7772  * same as the vdisk block size which is stored in vdc->vdisk_bsize so we
7773  * do not store that information again.
7774  *
7775  * However, buf structures will always use a logical block size of 512 bytes
7776  * (DEV_BSIZE) and we will need to convert logical block numbers to VIO block
7777  * numbers for each read or write operation using vdc_strategy(). To speed up
7778  * this conversion, we expect the VIO block size to be a power of 2 and a
7779  * multiple 512 bytes (DEV_BSIZE), and we cache some useful information.
7780  *
7781  * The function return EINVAL if the new VIO block size (blk_size) is not a
7782  * power of 2 or not a multiple of 512 bytes, otherwise it returns 0.
7783  */
7784 static int
7785 vdc_update_vio_bsize(vdc_t *vdc, uint32_t blk_size)
7786 {
7787 	uint32_t ratio, n;
7788 	int nshift = 0;
7789 
7790 	vdc->vio_bmask = 0;
7791 	vdc->vio_bshift = 0;
7792 
7793 	ASSERT(blk_size > 0);
7794 
7795 	if ((blk_size % DEV_BSIZE) != 0)
7796 		return (EINVAL);
7797 
7798 	ratio = blk_size / DEV_BSIZE;
7799 
7800 	for (n = ratio; n > 1; n >>= 1) {
7801 		if ((n & 0x1) != 0) {
7802 			/* blk_size is not a power of 2 */
7803 			return (EINVAL);
7804 		}
7805 		nshift++;
7806 	}
7807 
7808 	vdc->vio_bshift = nshift;
7809 	vdc->vio_bmask = ratio - 1;
7810 
7811 	return (0);
7812 }
7813 
7814 /*
7815  * Function:
7816  *	vdc_validate_geometry
7817  *
7818  * Description:
7819  *	This routine discovers the label and geometry of the disk. It stores
7820  *	the disk label and related information in the vdc structure. If it
7821  *	fails to validate the geometry or to discover the disk label then
7822  *	the label is marked as unknown (VD_DISK_LABEL_UNK).
7823  *
7824  * Arguments:
7825  *	vdc	- soft state pointer for this instance of the device driver.
7826  *
7827  * Return Code:
7828  *	0	- success.
7829  *	EINVAL	- unknown disk label.
7830  *	ENOTSUP	- geometry not applicable (EFI label).
7831  *	EIO	- error accessing the disk.
7832  */
7833 static int
7834 vdc_validate_geometry(vdc_t *vdc)
7835 {
7836 	buf_t	*buf;	/* BREAD requests need to be in a buf_t structure */
7837 	dev_t	dev;
7838 	int	rv, rval;
7839 	struct dk_label *label;
7840 	struct dk_geom geom;
7841 	struct extvtoc vtoc;
7842 	efi_gpt_t *gpt;
7843 	efi_gpe_t *gpe;
7844 	vd_efi_dev_t edev;
7845 
7846 	ASSERT(vdc != NULL);
7847 	ASSERT(vdc->vtoc != NULL && vdc->geom != NULL);
7848 	ASSERT(MUTEX_HELD(&vdc->lock));
7849 
7850 	mutex_exit(&vdc->lock);
7851 	/*
7852 	 * Check the disk capacity in case it has changed. If that fails then
7853 	 * we proceed and we will be using the disk size we currently have.
7854 	 */
7855 	(void) vdc_check_capacity(vdc);
7856 	dev = makedevice(ddi_driver_major(vdc->dip),
7857 	    VD_MAKE_DEV(vdc->instance, 0));
7858 
7859 	rv = vd_process_ioctl(dev, DKIOCGGEOM, (caddr_t)&geom, FKIOCTL, &rval);
7860 	if (rv == 0)
7861 		rv = vd_process_ioctl(dev, DKIOCGEXTVTOC, (caddr_t)&vtoc,
7862 		    FKIOCTL, &rval);
7863 
7864 	if (rv == ENOTSUP) {
7865 		/*
7866 		 * If the device does not support VTOC then we try
7867 		 * to read an EFI label.
7868 		 *
7869 		 * We need to know the block size and the disk size to
7870 		 * be able to read an EFI label.
7871 		 */
7872 		if (vdc->vdisk_size == 0) {
7873 			mutex_enter(&vdc->lock);
7874 			vdc_store_label_unk(vdc);
7875 			return (EIO);
7876 		}
7877 
7878 		VDC_EFI_DEV_SET(edev, vdc, vd_process_efi_ioctl);
7879 
7880 		rv = vd_efi_alloc_and_read(&edev, &gpt, &gpe);
7881 
7882 		if (rv) {
7883 			DMSG(vdc, 0, "[%d] Failed to get EFI (err=%d)",
7884 			    vdc->instance, rv);
7885 			mutex_enter(&vdc->lock);
7886 			vdc_store_label_unk(vdc);
7887 			return (EIO);
7888 		}
7889 
7890 		mutex_enter(&vdc->lock);
7891 		vdc_store_label_efi(vdc, gpt, gpe);
7892 		vd_efi_free(&edev, gpt, gpe);
7893 		return (ENOTSUP);
7894 	}
7895 
7896 	if (rv != 0) {
7897 		DMSG(vdc, 0, "[%d] Failed to get VTOC (err=%d)",
7898 		    vdc->instance, rv);
7899 		mutex_enter(&vdc->lock);
7900 		vdc_store_label_unk(vdc);
7901 		if (rv != EINVAL)
7902 			rv = EIO;
7903 		return (rv);
7904 	}
7905 
7906 	/* check that geometry and vtoc are valid */
7907 	if (geom.dkg_nhead == 0 || geom.dkg_nsect == 0 ||
7908 	    vtoc.v_sanity != VTOC_SANE) {
7909 		mutex_enter(&vdc->lock);
7910 		vdc_store_label_unk(vdc);
7911 		return (EINVAL);
7912 	}
7913 
7914 	/*
7915 	 * We have a disk and a valid VTOC. However this does not mean
7916 	 * that the disk currently have a VTOC label. The returned VTOC may
7917 	 * be a default VTOC to be used for configuring the disk (this is
7918 	 * what is done for disk image). So we read the label from the
7919 	 * beginning of the disk to ensure we really have a VTOC label.
7920 	 *
7921 	 * FUTURE: This could be the default way for reading the VTOC
7922 	 * from the disk as opposed to sending the VD_OP_GET_VTOC
7923 	 * to the server. This will be the default if vdc is implemented
7924 	 * ontop of cmlb.
7925 	 */
7926 
7927 	/*
7928 	 * Single slice disk does not support read using an absolute disk
7929 	 * offset so we just rely on the DKIOCGVTOC ioctl in that case.
7930 	 */
7931 	if (vdc->vdisk_type == VD_DISK_TYPE_SLICE) {
7932 		mutex_enter(&vdc->lock);
7933 		if (vtoc.v_nparts != 1) {
7934 			vdc_store_label_unk(vdc);
7935 			return (EINVAL);
7936 		}
7937 		vdc_store_label_vtoc(vdc, &geom, &vtoc);
7938 		return (0);
7939 	}
7940 
7941 	if (vtoc.v_nparts != V_NUMPAR) {
7942 		mutex_enter(&vdc->lock);
7943 		vdc_store_label_unk(vdc);
7944 		return (EINVAL);
7945 	}
7946 
7947 	/*
7948 	 * Most CD/DVDs do not have a disk label and the label is
7949 	 * generated by the disk driver.  So the on-disk label check
7950 	 * below may fail and we return now to avoid this problem.
7951 	 */
7952 	if (vdc->vdisk_media == VD_MEDIA_CD ||
7953 	    vdc->vdisk_media == VD_MEDIA_DVD) {
7954 		mutex_enter(&vdc->lock);
7955 		vdc_store_label_vtoc(vdc, &geom, &vtoc);
7956 		return (0);
7957 	}
7958 
7959 	/*
7960 	 * Read disk label from start of disk
7961 	 */
7962 	label = kmem_alloc(vdc->vdisk_bsize, KM_SLEEP);
7963 	buf = kmem_alloc(sizeof (buf_t), KM_SLEEP);
7964 	bioinit(buf);
7965 	buf->b_un.b_addr = (caddr_t)label;
7966 	buf->b_bcount = vdc->vdisk_bsize;
7967 	buf->b_flags = B_BUSY | B_READ;
7968 	buf->b_dev = cmpdev(dev);
7969 	rv = vdc_send_request(vdc, VD_OP_BREAD, (caddr_t)label,
7970 	    vdc->vdisk_bsize, VD_SLICE_NONE, 0, CB_STRATEGY, buf, VIO_read_dir);
7971 	if (rv) {
7972 		DMSG(vdc, 1, "[%d] Failed to read disk block 0\n",
7973 		    vdc->instance);
7974 	} else if (ddi_in_panic()) {
7975 		rv = vdc_drain_response(vdc, CB_STRATEGY, buf);
7976 		if (rv == 0) {
7977 			rv = geterror(buf);
7978 		}
7979 	} else {
7980 		rv = biowait(buf);
7981 	}
7982 	biofini(buf);
7983 	kmem_free(buf, sizeof (buf_t));
7984 
7985 	if (rv != 0 || label->dkl_magic != DKL_MAGIC ||
7986 	    label->dkl_cksum != vdc_lbl2cksum(label)) {
7987 		DMSG(vdc, 1, "[%d] Got VTOC with invalid label\n",
7988 		    vdc->instance);
7989 		kmem_free(label, vdc->vdisk_bsize);
7990 		mutex_enter(&vdc->lock);
7991 		vdc_store_label_unk(vdc);
7992 		return (EINVAL);
7993 	}
7994 
7995 	kmem_free(label, vdc->vdisk_bsize);
7996 	mutex_enter(&vdc->lock);
7997 	vdc_store_label_vtoc(vdc, &geom, &vtoc);
7998 	return (0);
7999 }
8000 
8001 /*
8002  * Function:
8003  *	vdc_validate
8004  *
8005  * Description:
8006  *	This routine discovers the label of the disk and create the
8007  *	appropriate device nodes if the label has changed.
8008  *
8009  * Arguments:
8010  *	vdc	- soft state pointer for this instance of the device driver.
8011  *
8012  * Return Code:
8013  *	none.
8014  */
8015 static void
8016 vdc_validate(vdc_t *vdc)
8017 {
8018 	vd_disk_label_t old_label;
8019 	vd_slice_t old_slice[V_NUMPAR];
8020 	int rv;
8021 
8022 	ASSERT(!MUTEX_HELD(&vdc->lock));
8023 
8024 	mutex_enter(&vdc->lock);
8025 
8026 	/* save the current label and vtoc */
8027 	old_label = vdc->vdisk_label;
8028 	bcopy(vdc->slice, &old_slice, sizeof (vd_slice_t) * V_NUMPAR);
8029 
8030 	/* check the geometry */
8031 	(void) vdc_validate_geometry(vdc);
8032 
8033 	/* if the disk label has changed, update device nodes */
8034 	if (vdc->vdisk_label != old_label) {
8035 
8036 		if (vdc->vdisk_label == VD_DISK_LABEL_EFI)
8037 			rv = vdc_create_device_nodes_efi(vdc);
8038 		else
8039 			rv = vdc_create_device_nodes_vtoc(vdc);
8040 
8041 		if (rv != 0) {
8042 			DMSG(vdc, 0, "![%d] Failed to update device nodes",
8043 			    vdc->instance);
8044 		}
8045 	}
8046 
8047 	mutex_exit(&vdc->lock);
8048 }
8049 
8050 static void
8051 vdc_validate_task(void *arg)
8052 {
8053 	vdc_t *vdc = (vdc_t *)arg;
8054 
8055 	vdc_validate(vdc);
8056 
8057 	mutex_enter(&vdc->lock);
8058 	ASSERT(vdc->validate_pending > 0);
8059 	vdc->validate_pending--;
8060 	mutex_exit(&vdc->lock);
8061 }
8062 
8063 /*
8064  * Function:
8065  *	vdc_setup_devid()
8066  *
8067  * Description:
8068  *	This routine discovers the devid of a vDisk. It requests the devid of
8069  *	the underlying device from the vDisk server, builds an encapsulated
8070  *	devid based on the retrieved devid and registers that new devid to
8071  *	the vDisk.
8072  *
8073  * Arguments:
8074  *	vdc	- soft state pointer for this instance of the device driver.
8075  *
8076  * Return Code:
8077  *	0	- A devid was succesfully registered for the vDisk
8078  */
8079 static int
8080 vdc_setup_devid(vdc_t *vdc)
8081 {
8082 	int rv;
8083 	vd_devid_t *vd_devid;
8084 	size_t bufsize, bufid_len;
8085 
8086 	/*
8087 	 * At first sight, we don't know the size of the devid that the
8088 	 * server will return but this size will be encoded into the
8089 	 * reply. So we do a first request using a default size then we
8090 	 * check if this size was large enough. If not then we do a second
8091 	 * request with the correct size returned by the server. Note that
8092 	 * ldc requires size to be 8-byte aligned.
8093 	 */
8094 	bufsize = P2ROUNDUP(VD_DEVID_SIZE(VD_DEVID_DEFAULT_LEN),
8095 	    sizeof (uint64_t));
8096 	vd_devid = kmem_zalloc(bufsize, KM_SLEEP);
8097 	bufid_len = bufsize - sizeof (vd_efi_t) - 1;
8098 
8099 	rv = vdc_do_sync_op(vdc, VD_OP_GET_DEVID, (caddr_t)vd_devid,
8100 	    bufsize, 0, 0, CB_SYNC, 0, VIO_both_dir, B_TRUE);
8101 
8102 	DMSG(vdc, 2, "sync_op returned %d\n", rv);
8103 
8104 	if (rv) {
8105 		kmem_free(vd_devid, bufsize);
8106 		return (rv);
8107 	}
8108 
8109 	if (vd_devid->length > bufid_len) {
8110 		/*
8111 		 * The returned devid is larger than the buffer used. Try again
8112 		 * with a buffer with the right size.
8113 		 */
8114 		kmem_free(vd_devid, bufsize);
8115 		bufsize = P2ROUNDUP(VD_DEVID_SIZE(vd_devid->length),
8116 		    sizeof (uint64_t));
8117 		vd_devid = kmem_zalloc(bufsize, KM_SLEEP);
8118 		bufid_len = bufsize - sizeof (vd_efi_t) - 1;
8119 
8120 		rv = vdc_do_sync_op(vdc, VD_OP_GET_DEVID,
8121 		    (caddr_t)vd_devid, bufsize, 0, 0, CB_SYNC, 0,
8122 		    VIO_both_dir, B_TRUE);
8123 
8124 		if (rv) {
8125 			kmem_free(vd_devid, bufsize);
8126 			return (rv);
8127 		}
8128 	}
8129 
8130 	/*
8131 	 * The virtual disk should have the same device id as the one associated
8132 	 * with the physical disk it is mapped on, otherwise sharing a disk
8133 	 * between a LDom and a non-LDom may not work (for example for a shared
8134 	 * SVM disk set).
8135 	 *
8136 	 * The DDI framework does not allow creating a device id with any
8137 	 * type so we first create a device id of type DEVID_ENCAP and then
8138 	 * we restore the orignal type of the physical device.
8139 	 */
8140 
8141 	DMSG(vdc, 2, ": devid length = %d\n", vd_devid->length);
8142 
8143 	/* build an encapsulated devid based on the returned devid */
8144 	if (ddi_devid_init(vdc->dip, DEVID_ENCAP, vd_devid->length,
8145 	    vd_devid->id, &vdc->devid) != DDI_SUCCESS) {
8146 		DMSG(vdc, 1, "[%d] Fail to created devid\n", vdc->instance);
8147 		kmem_free(vd_devid, bufsize);
8148 		return (1);
8149 	}
8150 
8151 	DEVID_FORMTYPE((impl_devid_t *)vdc->devid, vd_devid->type);
8152 
8153 	ASSERT(ddi_devid_valid(vdc->devid) == DDI_SUCCESS);
8154 
8155 	kmem_free(vd_devid, bufsize);
8156 
8157 	if (ddi_devid_register(vdc->dip, vdc->devid) != DDI_SUCCESS) {
8158 		DMSG(vdc, 1, "[%d] Fail to register devid\n", vdc->instance);
8159 		return (1);
8160 	}
8161 
8162 	return (0);
8163 }
8164 
8165 static void
8166 vdc_store_label_efi(vdc_t *vdc, efi_gpt_t *gpt, efi_gpe_t *gpe)
8167 {
8168 	int i, nparts;
8169 
8170 	ASSERT(MUTEX_HELD(&vdc->lock));
8171 
8172 	vdc->vdisk_label = VD_DISK_LABEL_EFI;
8173 	bzero(vdc->vtoc, sizeof (struct extvtoc));
8174 	bzero(vdc->geom, sizeof (struct dk_geom));
8175 	bzero(vdc->slice, sizeof (vd_slice_t) * V_NUMPAR);
8176 
8177 	nparts = gpt->efi_gpt_NumberOfPartitionEntries;
8178 
8179 	for (i = 0; i < nparts && i < VD_EFI_WD_SLICE; i++) {
8180 
8181 		if (gpe[i].efi_gpe_StartingLBA == 0 &&
8182 		    gpe[i].efi_gpe_EndingLBA == 0) {
8183 			continue;
8184 		}
8185 
8186 		vdc->slice[i].start = gpe[i].efi_gpe_StartingLBA;
8187 		vdc->slice[i].nblocks = gpe[i].efi_gpe_EndingLBA -
8188 		    gpe[i].efi_gpe_StartingLBA + 1;
8189 	}
8190 
8191 	ASSERT(vdc->vdisk_size != 0);
8192 	vdc->slice[VD_EFI_WD_SLICE].start = 0;
8193 	vdc->slice[VD_EFI_WD_SLICE].nblocks = vdc->vdisk_size;
8194 
8195 }
8196 
8197 static void
8198 vdc_store_label_vtoc(vdc_t *vdc, struct dk_geom *geom, struct extvtoc *vtoc)
8199 {
8200 	int i;
8201 
8202 	ASSERT(MUTEX_HELD(&vdc->lock));
8203 	ASSERT(vdc->vdisk_bsize == vtoc->v_sectorsz);
8204 
8205 	vdc->vdisk_label = VD_DISK_LABEL_VTOC;
8206 	bcopy(vtoc, vdc->vtoc, sizeof (struct extvtoc));
8207 	bcopy(geom, vdc->geom, sizeof (struct dk_geom));
8208 	bzero(vdc->slice, sizeof (vd_slice_t) * V_NUMPAR);
8209 
8210 	for (i = 0; i < vtoc->v_nparts; i++) {
8211 		vdc->slice[i].start = vtoc->v_part[i].p_start;
8212 		vdc->slice[i].nblocks = vtoc->v_part[i].p_size;
8213 	}
8214 }
8215 
8216 static void
8217 vdc_store_label_unk(vdc_t *vdc)
8218 {
8219 	ASSERT(MUTEX_HELD(&vdc->lock));
8220 
8221 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
8222 	bzero(vdc->vtoc, sizeof (struct extvtoc));
8223 	bzero(vdc->geom, sizeof (struct dk_geom));
8224 	bzero(vdc->slice, sizeof (vd_slice_t) * V_NUMPAR);
8225 }
8226